update ChangeLog

Change-Id: I32a22786f99d0c239761a362485a4a91c783851b

.gitattributes

@@ -2,3 +2,4 @@
 .gitignore export-ignore
 .mailmap export-ignore
 *.pdf -text -diff
+*.ppm -text -diff

.gitignore

@@ -1,8 +1,10 @@
 *.l[ao]
-*.o
+*.[ao]
+*.pc
 .deps
 .libs
 /aclocal.m4
+/ar-lib
 /autom4te.cache
 /compile
 /config.*
@@ -13,11 +15,18 @@
 /libtool
 /ltmain.sh
 /missing
+/mkinstalldirs
 /stamp-h1
 Makefile
 Makefile.in
-examples/[cd]webp
+examples/[cdv]webp
+examples/gif2webp
+examples/webpmux
+src/webp/config.h*
+src/webp/stamp-h1
 /output
 /doc/output
 *.idb
 *.pdb
+/iosbuild
+/WebP.framework

.mailmap

@@ -1,2 +1,8 @@
+<johann.koenig@duck.com> <johannkoenig@google.com>
 Mikołaj Zalewski <mikolajz@google.com>
 Pascal Massimino <pascal.massimino@gmail.com>
+<pascal.massimino@gmail.com> <skal@google.com>
+Vikas Arora <vikasa@google.com>
+<vikasa@google.com> <vikasa@gmail.com>
+<vikasa@google.com> <vikaas.arora@gmail.com>
+<slobodan.prijic@imgtec.com> <Slobodan.Prijic@imgtec.com>

AUTHORS

@@ -1,15 +1,25 @@
 Contributors:
+- Charles Munger (clm at google dot com)
+- Christian Duvivier (cduvivier at google dot com)
+- Djordje Pesut (djordje dot pesut at imgtec dot com)
 - James Zern (jzern at google dot com)
 - Jan Engelhardt (jengelh at medozas dot de)
-- Johann (johannkoenig at google dot com)
+- Johann (johann dot koenig at duck dot com)
+- Jovan Zelincevic (jovan dot zelincevic at imgtec dot com)
 - Jyrki Alakuijala (jyrki at google dot com)
+- levytamar82 (tamar dot levy at intel dot com)
 - Lou Quillio (louquillio at google dot com)
+- Mans Rullgard (mans at mansr dot com)
 - Martin Olsson (mnemo at minimum dot se)
 - Mikołaj Zalewski (mikolajz at google dot com)
 - Noel Chromium (noel at chromium dot org)
 - Pascal Massimino (pascal dot massimino at gmail dot com)
+- Paweł Hajdan, Jr (phajdan dot jr at chromium dot org)
 - Pierre Joye (pierre dot php at gmail dot com)
 - Scott LaVarnway (slavarnway at google dot com)
+- Scott Talbot (s at chikachow dot org)
+- Slobodan Prijic (slobodan dot prijic at imgtec dot com)
 - Somnath Banerjee (somnath dot banerjee at gmail dot com)
+- Timothy Gu (timothygu99 at gmail dot com)
 - Urvang Joshi (urvang at google dot com)
 - Vikas Arora (vikasa at google dot com)

Android.mk

@@ -1,27 +1,60 @@
-LOCAL_PATH:= $(call my-dir)
+LOCAL_PATH := $(call my-dir)
+
+WEBP_CFLAGS := -Wall -DANDROID -DHAVE_MALLOC_H -DHAVE_PTHREAD -DWEBP_USE_THREAD
+
+ifeq ($(APP_OPTIM),release)
+  WEBP_CFLAGS += -finline-functions -ffast-math \
+                 -ffunction-sections -fdata-sections
+  ifeq ($(findstring clang,$(NDK_TOOLCHAIN_VERSION)),)
+    WEBP_CFLAGS += -frename-registers -s
+  endif
+endif
 
 include $(CLEAR_VARS)
+
+ifneq ($(findstring armeabi-v7a, $(TARGET_ARCH_ABI)),)
+  # Setting LOCAL_ARM_NEON will enable -mfpu=neon which may cause illegal
+  # instructions to be generated for armv7a code. Instead target the neon code
+  # specifically.
+  NEON := c.neon
+else
+  NEON := c
+endif
+
 LOCAL_SRC_FILES := \
     src/dec/alpha.c \
     src/dec/buffer.c \
     src/dec/frame.c \
     src/dec/idec.c \
     src/dec/io.c \
-    src/dec/layer.c \
     src/dec/quant.c \
     src/dec/tree.c \
     src/dec/vp8.c \
     src/dec/vp8l.c \
     src/dec/webp.c \
+    src/dsp/alpha_processing.c \
+    src/dsp/alpha_processing_sse2.c \
     src/dsp/cpu.c \
     src/dsp/dec.c \
+    src/dsp/dec_clip_tables.c \
+    src/dsp/dec_mips32.c \
+    src/dsp/dec_neon.$(NEON) \
     src/dsp/dec_sse2.c \
     src/dsp/enc.c \
+    src/dsp/enc_avx2.c \
+    src/dsp/enc_mips32.c \
+    src/dsp/enc_neon.$(NEON) \
     src/dsp/enc_sse2.c \
     src/dsp/lossless.c \
+    src/dsp/lossless_mips32.c \
+    src/dsp/lossless_neon.$(NEON) \
+    src/dsp/lossless_sse2.c \
     src/dsp/upsampling.c \
+    src/dsp/upsampling_neon.$(NEON) \
     src/dsp/upsampling_sse2.c \
     src/dsp/yuv.c \
+    src/dsp/yuv_mips32.c \
+    src/dsp/yuv_sse2.c \
     src/enc/alpha.c \
     src/enc/analysis.c \
     src/enc/backward_references.c \
@@ -31,10 +64,14 @@ LOCAL_SRC_FILES := \
     src/enc/frame.c \
     src/enc/histogram.c \
     src/enc/iterator.c \
-    src/enc/layer.c \
     src/enc/picture.c \
+    src/enc/picture_csp.c \
+    src/enc/picture_psnr.c \
+    src/enc/picture_rescale.c \
+    src/enc/picture_tools.c \
     src/enc/quant.c \
     src/enc/syntax.c \
+    src/enc/token.c \
     src/enc/tree.c \
     src/enc/vp8l.c \
     src/enc/webpenc.c \
@@ -45,27 +82,28 @@ LOCAL_SRC_FILES := \
     src/utils/huffman.c \
     src/utils/huffman_encode.c \
     src/utils/quant_levels.c \
+    src/utils/quant_levels_dec.c \
+    src/utils/random.c \
     src/utils/rescaler.c \
     src/utils/thread.c \
     src/utils/utils.c \
 
-LOCAL_CFLAGS := -Wall -DANDROID -DHAVE_MALLOC_H -DHAVE_PTHREAD \
-                -DWEBP_USE_THREAD \
-                -finline-functions -frename-registers -ffast-math \
-                -s -fomit-frame-pointer -Isrc/webp
-
+LOCAL_CFLAGS := $(WEBP_CFLAGS)
 LOCAL_C_INCLUDES += $(LOCAL_PATH)/src
 
-ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)
-  # Setting LOCAL_ARM_NEON will enable -mfpu=neon which may cause illegal
-  # instructions to be generated for armv7a code. Instead target the neon code
-  # specifically.
-  LOCAL_SRC_FILES += src/dsp/dec_neon.c.neon
-endif
+# prefer arm over thumb mode for performance gains
+LOCAL_ARM_MODE := arm
+
 LOCAL_STATIC_LIBRARIES := cpufeatures
 
-LOCAL_MODULE:= webp
+LOCAL_MODULE := webp
 
-include $(BUILD_STATIC_LIBRARY)
+ifeq ($(ENABLE_SHARED),1)
+  include $(BUILD_SHARED_LIBRARY)
+else
+  include $(BUILD_STATIC_LIBRARY)
+endif
+
+include $(LOCAL_PATH)/examples/Android.mk
 
 $(call import-module,android/cpufeatures)

Makefile.vc

@@ -1,8 +1,10 @@
 #
 # Stem for static libs and DLLs
 #
+LIBWEBPDECODER_BASENAME = libwebpdecoder
 LIBWEBP_BASENAME = libwebp
 LIBWEBPMUX_BASENAME = libwebpmux
+LIBWEBPDEMUX_BASENAME = libwebpdemux
 
 !IFNDEF ARCH
 !IF ! [ cl 2>&1 | find "x86" > NUL ]
@@ -22,18 +24,18 @@ PLATFORM_LDFLAGS = /SAFESEH
 #############################################################
 ## Nothing more to do below this line!
 
-MT         = mt.exe
-CCNODBG    = cl.exe /nologo /O2 /DNDEBUG
-CCDEBUG    = cl.exe /nologo /Od /Gm /Zi /D_DEBUG /RTC1
-CFLAGS     = /Isrc /nologo /W3 /EHsc /FD /c /GS
+NOLOGO     = /nologo
+CCNODBG    = cl.exe $(NOLOGO) /O2 /DNDEBUG
+CCDEBUG    = cl.exe $(NOLOGO) /Od /Gm /Zi /D_DEBUG /RTC1
+CFLAGS     = /Isrc $(NOLOGO) /W3 /EHsc /c /GS
 CFLAGS     = $(CFLAGS) /DWIN32 /D_CRT_SECURE_NO_WARNINGS /DWIN32_LEAN_AND_MEAN
 CFLAGS     = $(CFLAGS) /DHAVE_WINCODEC_H /DWEBP_USE_THREAD
 LDFLAGS    = /LARGEADDRESSAWARE /MANIFEST /NXCOMPAT /DYNAMICBASE
 LDFLAGS    = $(LDFLAGS) $(PLATFORM_LDFLAGS)
-LNKDLL     = link.exe /DLL
-LNKLIB     = link.exe /lib
-LNKEXE     = link.exe
-LFLAGS     = /nologo /machine:$(ARCH)
+LNKDLL     = link.exe /DLL $(NOLOGO)
+LNKEXE     = link.exe $(NOLOGO)
+LNKLIB     = lib.exe $(NOLOGO)
+MT         = mt.exe $(NOLOGO)
 
 CFGSET     = FALSE
 !IF "$(OBJDIR)" == ""
@@ -42,6 +44,11 @@ OUTDIR = ..\obj\
 OUTDIR = $(OBJDIR)
 !ENDIF
 
+!IF "$(HAVE_AVX2)" == "1"
+CFLAGS = $(CFLAGS) /DWEBP_HAVE_AVX2
+AVX2_FLAGS = /arch:AVX2
+!ENDIF
+
 ##############################################################
 # Runtime library configuration
 !IF "$(RTLIBCFG)" == "static"
@@ -59,6 +66,7 @@ DIRBIN = $(DIRBASE)\bin
 LIBWEBP_PDBNAME = $(DIROBJ)\$(LIBWEBP_BASENAME).pdb
 OUTPUT_DIRS = $(DIRBIN) $(DIRINC) $(DIRLIB) \
               $(DIROBJ)\dec \
+              $(DIROBJ)\demux \
               $(DIROBJ)\dsp \
               $(DIROBJ)\enc \
               $(DIROBJ)\examples \
@@ -73,8 +81,10 @@ STATICLIBBUILD = TRUE
 CC             = $(CCDEBUG)
 RTLIB          = $(RTLIBD)
 STATICLIBBUILD = TRUE
+LIBWEBPDECODER_BASENAME = $(LIBWEBPDECODER_BASENAME)_debug
 LIBWEBP_BASENAME = $(LIBWEBP_BASENAME)_debug
 LIBWEBPMUX_BASENAME = $(LIBWEBPMUX_BASENAME)_debug
+LIBWEBPDEMUX_BASENAME = $(LIBWEBPDEMUX_BASENAME)_debug
 !ELSE IF "$(CFG)" == "release-dynamic"
 CC        = $(CCNODBG)
 DLLBUILD  = TRUE
@@ -82,22 +92,28 @@ DLLBUILD  = TRUE
 CC        = $(CCDEBUG)
 RTLIB     = $(RTLIBD)
 DLLBUILD  = TRUE
+LIBWEBPDECODER_BASENAME = $(LIBWEBPDECODER_BASENAME)_debug
 LIBWEBP_BASENAME = $(LIBWEBP_BASENAME)_debug
 LIBWEBPMUX_BASENAME = $(LIBWEBPMUX_BASENAME)_debug
+LIBWEBPDEMUX_BASENAME = $(LIBWEBPDEMUX_BASENAME)_debug
 !ENDIF
 
 !IF "$(STATICLIBBUILD)" == "TRUE"
 CC     = $(CC) $(RTLIB)
 CFGSET = TRUE
+LIBWEBPDECODER = $(DIRLIB)\$(LIBWEBPDECODER_BASENAME).lib
 LIBWEBP = $(DIRLIB)\$(LIBWEBP_BASENAME).lib
 LIBWEBPMUX = $(DIRLIB)\$(LIBWEBPMUX_BASENAME).lib
+LIBWEBPDEMUX = $(DIRLIB)\$(LIBWEBPDEMUX_BASENAME).lib
 !ELSE IF "$(DLLBUILD)" == "TRUE"
 DLLC   = webp_dll.c
 DLLINC = webp_dll.h
+DLL_OBJS = $(DIROBJ)\$(DLLC:.c=.obj)
 CC     = $(CC) /I$(DIROBJ) /FI$(DLLINC) $(RTLIB) /DWEBP_DLL
+LIBWEBPDECODER = $(DIRLIB)\$(LIBWEBPDECODER_BASENAME)_dll.lib
 LIBWEBP = $(DIRLIB)\$(LIBWEBP_BASENAME)_dll.lib
 LIBWEBPMUX = $(DIRLIB)\$(LIBWEBPMUX_BASENAME)_dll.lib
-LIBWEBP_OBJS = $(DIROBJ)\$(DLLC:.c=.obj)
+LIBWEBPDEMUX = $(DIRLIB)\$(LIBWEBPDEMUX_BASENAME)_dll.lib
 LIBWEBP_PDBNAME = $(DIROBJ)\$(LIBWEBP_BASENAME)_dll.pdb
 CFGSET = TRUE
 !ENDIF
@@ -119,7 +135,8 @@ CFGSET = TRUE
 !MESSAGE -  clean                         - perform a clean for CFG
 !MESSAGE -  experimental                  - build CFG with experimental
 !MESSAGE .                                  features enabled.
-!MESSAGE - (empty) or all                 - build all targets for CFG
+!MESSAGE - (empty)                        - build libwebp-based targets for CFG
+!MESSAGE - all                            - build (de)mux-based targets for CFG
 !MESSAGE
 !MESSAGE RTLIBCFG controls the runtime library linkage - 'static' or 'dynamic'.
 !MESSAGE OBJDIR is the path where you like to build (obj, bins, etc.),
@@ -144,23 +161,49 @@ DEC_OBJS = \
     $(DIROBJ)\dec\frame.obj \
     $(DIROBJ)\dec\idec.obj \
     $(DIROBJ)\dec\io.obj \
-    $(DIROBJ)\dec\layer.obj \
     $(DIROBJ)\dec\quant.obj \
     $(DIROBJ)\dec\tree.obj \
     $(DIROBJ)\dec\vp8.obj \
     $(DIROBJ)\dec\vp8l.obj \
     $(DIROBJ)\dec\webp.obj \
 
-DSP_OBJS = \
+DEMUX_OBJS = \
+    $(DIROBJ)\demux\demux.obj \
+
+DSP_DEC_OBJS = \
+    $(DIROBJ)\dsp\alpha_processing.obj \
+    $(DIROBJ)\dsp\alpha_processing_sse2.obj \
     $(DIROBJ)\dsp\cpu.obj \
     $(DIROBJ)\dsp\dec.obj \
+    $(DIROBJ)\dsp\dec_clip_tables.obj \
+    $(DIROBJ)\dsp\dec_mips32.obj \
+    $(DIROBJ)\dsp\dec_neon.obj \
     $(DIROBJ)\dsp\dec_sse2.obj \
-    $(DIROBJ)\dsp\enc.obj \
-    $(DIROBJ)\dsp\enc_sse2.obj \
     $(DIROBJ)\dsp\lossless.obj \
+    $(DIROBJ)\dsp\lossless_mips32.obj \
+    $(DIROBJ)\dsp\lossless_neon.obj \
+    $(DIROBJ)\dsp\lossless_sse2.obj \
     $(DIROBJ)\dsp\upsampling.obj \
+    $(DIROBJ)\dsp\upsampling_neon.obj \
     $(DIROBJ)\dsp\upsampling_sse2.obj \
     $(DIROBJ)\dsp\yuv.obj \
+    $(DIROBJ)\dsp\yuv_mips32.obj \
+    $(DIROBJ)\dsp\yuv_sse2.obj \
+
+DSP_ENC_OBJS = \
+    $(DIROBJ)\dsp\enc.obj \
+    $(DIROBJ)\dsp\enc_avx2.obj \
+    $(DIROBJ)\dsp\enc_mips32.obj \
+    $(DIROBJ)\dsp\enc_neon.obj \
+    $(DIROBJ)\dsp\enc_sse2.obj \
+
+EX_FORMAT_DEC_OBJS = \
+    $(DIROBJ)\examples\jpegdec.obj \
+    $(DIROBJ)\examples\metadata.obj \
+    $(DIROBJ)\examples\pngdec.obj \
+    $(DIROBJ)\examples\tiffdec.obj \
+    $(DIROBJ)\examples\webpdec.obj \
+    $(DIROBJ)\examples\wicdec.obj \
 
 EX_UTIL_OBJS = \
     $(DIROBJ)\examples\example_util.obj \
@@ -175,72 +218,91 @@ ENC_OBJS = \
     $(DIROBJ)\enc\frame.obj \
     $(DIROBJ)\enc\histogram.obj \
     $(DIROBJ)\enc\iterator.obj \
-    $(DIROBJ)\enc\layer.obj \
     $(DIROBJ)\enc\picture.obj \
+    $(DIROBJ)\enc\picture_csp.obj \
+    $(DIROBJ)\enc\picture_psnr.obj \
+    $(DIROBJ)\enc\picture_rescale.obj \
+    $(DIROBJ)\enc\picture_tools.obj \
     $(DIROBJ)\enc\quant.obj \
     $(DIROBJ)\enc\syntax.obj \
+    $(DIROBJ)\enc\token.obj \
     $(DIROBJ)\enc\tree.obj \
     $(DIROBJ)\enc\vp8l.obj \
     $(DIROBJ)\enc\webpenc.obj \
 
 MUX_OBJS = \
-    $(DIROBJ)\mux\demux.obj \
     $(DIROBJ)\mux\muxedit.obj \
     $(DIROBJ)\mux\muxinternal.obj \
     $(DIROBJ)\mux\muxread.obj \
 
-UTILS_OBJS = \
+UTILS_DEC_OBJS = \
     $(DIROBJ)\utils\bit_reader.obj \
-    $(DIROBJ)\utils\bit_writer.obj \
     $(DIROBJ)\utils\color_cache.obj \
     $(DIROBJ)\utils\filters.obj \
     $(DIROBJ)\utils\huffman.obj \
-    $(DIROBJ)\utils\huffman_encode.obj \
-    $(DIROBJ)\utils\quant_levels.obj \
+    $(DIROBJ)\utils\quant_levels_dec.obj \
     $(DIROBJ)\utils\rescaler.obj \
+    $(DIROBJ)\utils\random.obj \
     $(DIROBJ)\utils\thread.obj \
     $(DIROBJ)\utils\utils.obj \
 
-LIBWEBP_OBJS = $(DEC_OBJS) $(DSP_OBJS) $(ENC_OBJS) $(UTILS_OBJS) $(LIBWEBP_OBJS)
+UTILS_ENC_OBJS = \
+    $(DIROBJ)\utils\bit_writer.obj \
+    $(DIROBJ)\utils\huffman_encode.obj \
+    $(DIROBJ)\utils\quant_levels.obj \
+
+LIBWEBPDECODER_OBJS = $(DEC_OBJS) $(DSP_DEC_OBJS) $(UTILS_DEC_OBJS)
+LIBWEBP_OBJS = $(LIBWEBPDECODER_OBJS) $(ENC_OBJS) $(DSP_ENC_OBJS) \
+               $(UTILS_ENC_OBJS) $(DLL_OBJS)
 LIBWEBPMUX_OBJS = $(MUX_OBJS) $(LIBWEBPMUX_OBJS)
+LIBWEBPDEMUX_OBJS = $(DEMUX_OBJS) $(LIBWEBPDEMUX_OBJS)
 
-OUT_LIBS = $(LIBWEBP)
+OUT_LIBS = $(LIBWEBPDECODER) $(LIBWEBP)
 OUT_EXAMPLES = $(DIRBIN)\cwebp.exe $(DIRBIN)\dwebp.exe
+EXTRA_EXAMPLES = $(DIRBIN)\vwebp.exe $(DIRBIN)\webpmux.exe
 
-all: $(OUT_LIBS) $(OUT_EXAMPLES)
-$(DIRBIN)\cwebp.exe: $(DIROBJ)\examples\cwebp.obj
+ex: $(OUT_LIBS) $(OUT_EXAMPLES)
+all: ex $(EXTRA_EXAMPLES)
+$(DIRBIN)\cwebp.exe: $(DIROBJ)\examples\cwebp.obj $(EX_FORMAT_DEC_OBJS)
 $(DIRBIN)\dwebp.exe: $(DIROBJ)\examples\dwebp.obj
+$(DIRBIN)\vwebp.exe: $(DIROBJ)\examples\vwebp.obj
+$(DIRBIN)\vwebp.exe: $(EX_UTIL_OBJS) $(LIBWEBPDEMUX) $(LIBWEBP)
 $(DIRBIN)\webpmux.exe: $(DIROBJ)\examples\webpmux.obj $(LIBWEBPMUX)
 $(DIRBIN)\webpmux.exe: $(EX_UTIL_OBJS) $(LIBWEBP)
 $(OUT_EXAMPLES): $(EX_UTIL_OBJS) $(LIBWEBP)
+$(EX_UTIL_OBJS) $(EX_FORMAT_DEC_OBJS): $(OUTPUT_DIRS)
 
 experimental:
 	$(MAKE) /f Makefile.vc \
 	    CFG=$(CFG) \
 	    CFLAGS="$(CFLAGS) /DWEBP_EXPERIMENTAL_FEATURES" /$(MAKEFLAGS)
 
+$(LIBWEBPDECODER): $(LIBWEBPDECODER_OBJS)
 $(LIBWEBP): $(LIBWEBP_OBJS)
 $(LIBWEBPMUX): $(LIBWEBPMUX_OBJS)
+$(LIBWEBPDEMUX): $(LIBWEBPDEMUX_OBJS)
 
-$(LIBWEBP_OBJS) $(LIBWEBPMUX_OBJS): $(OUTPUT_DIRS)
+$(LIBWEBP_OBJS) $(LIBWEBPMUX_OBJS) $(LIBWEBPDEMUX_OBJS): $(OUTPUT_DIRS)
 
 !IF "$(DLLBUILD)" == "TRUE"
-$(LIBWEBP_OBJS) $(LIBWEBPMUX_OBJS): $(DIROBJ)\$(DLLINC) $(DIROBJ)\$(DLLC)
+$(LIBWEBP_OBJS) $(LIBWEBPMUX_OBJS) $(LIBWEBPDEMUX_OBJS): \
+    $(DIROBJ)\$(DLLINC) $(DIROBJ)\$(DLLC)
 
 {$(DIROBJ)}.c{$(DIROBJ)}.obj:
 	$(CC) $(CFLAGS) /Fd$(LIBWEBP_PDBNAME) /Fo$@  $<
 
 $(LIBWEBPMUX): $(LIBWEBP)
+$(LIBWEBPDEMUX): $(LIBWEBP)
 
-$(LIBWEBP) $(LIBWEBPMUX):
+$(LIBWEBPDECODER) $(LIBWEBP) $(LIBWEBPMUX) $(LIBWEBPDEMUX):
 	$(LNKDLL) /out:$(DIRBIN)\$(@B:_dll=.dll) /implib:$@ $(LFLAGS) $**
 	-xcopy $(DIROBJ)\*.pdb $(DIRLIB) /y
 
 clean::
 	@-erase /s $(DIROBJ)\$(DLLC) $(DIROBJ)\$(DLLINC) 2> NUL
 !ELSE
-$(LIBWEBP) $(LIBWEBPMUX):
-	$(LNKLIB) /out:$@ $(LFLAGS) $**
+$(LIBWEBPDECODER) $(LIBWEBP) $(LIBWEBPMUX) $(LIBWEBPDEMUX):
+	$(LNKLIB) /out:$@ $**
 	-xcopy $(DIROBJ)\*.pdb $(DIRLIB) /y
 !ENDIF
 
@@ -265,10 +327,18 @@ $(DIROBJ)\$(DLLC): $(DIROBJ)\$(DLLINC)
 	@echo } >> $@
 
 .SUFFIXES: .c .obj .res .exe
+# File-specific flag builds. Note batch rules take precedence over wildcards,
+# so for now name each file individually.
+$(DIROBJ)\dsp\enc_avx2.obj: src\dsp\enc_avx2.c
+	$(CC) $(CFLAGS) $(AVX2_FLAGS) /Fd$(LIBWEBP_PDBNAME) /Fo$(DIROBJ)\dsp\ \
+	  src\dsp\$(@B).c
+# Batch rules
 {examples}.c{$(DIROBJ)\examples}.obj::
-	$(CC) $(CFLAGS) /Fd$(DIROBJ)\examples\ /Fo$(DIROBJ)\examples\  $<
+	$(CC) $(CFLAGS) /Fd$(DIROBJ)\examples\ /Fo$(DIROBJ)\examples\ $<
 {src\dec}.c{$(DIROBJ)\dec}.obj::
 	$(CC) $(CFLAGS) /Fd$(LIBWEBP_PDBNAME) /Fo$(DIROBJ)\dec\ $<
+{src\demux}.c{$(DIROBJ)\demux}.obj::
+	$(CC) $(CFLAGS) /Fd$(LIBWEBP_PDBNAME) /Fo$(DIROBJ)\demux\ $<
 {src\dsp}.c{$(DIROBJ)\dsp}.obj::
 	$(CC) $(CFLAGS) /Fd$(LIBWEBP_PDBNAME) /Fo$(DIROBJ)\dsp\ $<
 {src\enc}.c{$(DIROBJ)\enc}.obj::

NEWS

@@ -1,3 +1,58 @@
+- 10/13/14: version 0.4.2
+  This is a binary compatible release.
+  * Android / gcc build fixes
+  * (Windows) fix reading from stdin and writing to stdout
+  * gif2webp: miscellaneous fixes
+  * fix 'alpha-leak' with lossy compression (issue #220)
+  * the lossless bitstream spec has been amended to reflect the current code
+
+- 7/24/14: version 0.4.1
+  This is a binary compatible release.
+  * AArch64 (arm64) & MIPS support/optimizations
+  * NEON assembly additions:
+    - ~25% faster lossy decode / encode (-m 4)
+    - ~10% faster lossless decode
+    - ~5-10% faster lossless encode (-m 3/4)
+  * dwebp/vwebp can read from stdin
+  * cwebp/gif2webp can write to stdout
+  * cwebp can read webp files; useful if storing sources as webp lossless
+
+- 12/19/13: version 0.4.0
+  * improved gif2webp tool
+  * numerous fixes, compression improvement and speed-up
+  * dither option added to decoder (dwebp -dither 50 ...)
+  * improved multi-threaded modes (-mt option)
+  * improved filtering strength determination
+  * New function: WebPMuxGetCanvasSize
+  * BMP and TIFF format output added to 'dwebp'
+  * Significant memory reduction for decoding lossy images with alpha.
+  * Intertwined decoding of RGB and alpha for a shorter
+    time-to-first-decoded-pixel.
+  * WebPIterator has a new member 'has_alpha' denoting whether the frame
+    contains transparency.
+  * Container spec amended with new 'blending method' for animation.
+
+- 6/13/13: version 0.3.1
+  This is a binary compatible release.
+  * Add incremental decoding support for images containing ALPH and ICCP chunks.
+  * Python bindings via swig for the simple encode/decode interfaces similar to
+    Java.
+
+- 3/20/13: version 0.3.0
+  This is a binary compatible release.
+  * WebPINewRGB/WebPINewYUVA accept being passed a NULL output buffer
+    and will perform auto-allocation.
+  * default filter option is now '-strong -f 60'
+  * encoding speed-up for lossy methods 3 to 6
+  * alpha encoding can be done in parallel to lossy using 'cwebp -mt ...'
+  * color profile, metadata (XMP/EXIF) and animation support finalized in the
+    container.
+  * various NEON assembly additions
+  Tool updates / additions:
+    * gif2webp added
+    * vwebp given color profile & animation support
+    * cwebp can preserve color profile / metadata with '-metadata'
+
 - 10/30/12: version 0.2.1
   * Various security related fixes
   * cwebp.exe: fix import errors on Windows XP

PATENTS

@@ -1,22 +1,23 @@
 Additional IP Rights Grant (Patents)
+------------------------------------
 
-"This implementation" means the copyrightable works distributed by
-Google as part of the WebM Project.
+"These implementations" means the copyrightable works that implement the WebM
+codecs distributed by Google as part of the WebM Project.
 
-Google hereby grants to you a perpetual, worldwide, non-exclusive,
-no-charge, royalty-free, irrevocable (except as stated in this section)
-patent license to make, have made, use, offer to sell, sell, import,
-transfer, and otherwise run, modify and propagate the contents of this
-implementation of VP8, where such license applies only to those patent
-claims, both currently owned by Google and acquired in the future,
-licensable by Google that are necessarily infringed by this
-implementation of VP8. This grant does not include claims that would be
-infringed only as a consequence of further modification of this
-implementation. If you or your agent or exclusive licensee institute or
-order or agree to the institution of patent litigation against any
-entity (including a cross-claim or counterclaim in a lawsuit) alleging
-that this implementation of VP8 or any code incorporated within this
-implementation of VP8 constitutes direct or contributory patent
-infringement, or inducement of patent infringement, then any patent
-rights granted to you under this License for this implementation of VP8
-shall terminate as of the date such litigation is filed.
+Google hereby grants to you a perpetual, worldwide, non-exclusive, no-charge,
+royalty-free, irrevocable (except as stated in this section) patent license to
+make, have made, use, offer to sell, sell, import, transfer, and otherwise
+run, modify and propagate the contents of these implementations of WebM, where
+such license applies only to those patent claims, both currently owned by
+Google and acquired in the future, licensable by Google that are necessarily
+infringed by these implementations of WebM. This grant does not include claims
+that would be infringed only as a consequence of further modification of these
+implementations. If you or your agent or exclusive licensee institute or order
+or agree to the institution of patent litigation or any other patent
+enforcement activity against any entity (including a cross-claim or
+counterclaim in a lawsuit) alleging that any of these implementations of WebM
+or any code incorporated within any of these implementations of WebM
+constitutes direct or contributory patent infringement, or inducement of
+patent infringement, then any patent rights granted to you under this License
+for these implementations of WebM shall terminate as of the date such
+litigation is filed.

README

@@ -4,7 +4,7 @@
           \__\__/\____/\_____/__/ ____  ___
                 / _/ /    \    \ /  _ \/ _/
                /  \_/   / /   \ \   __/  \__
-               \____/____/\_____/_____/____/v0.2.1
+               \____/____/\_____/_____/____/v0.4.2
 
 Description:
 ============
@@ -49,7 +49,7 @@ will build the binaries examples/cwebp and examples/dwebp, along
 with the static library src/libwebp.a. No system-wide installation
 is supplied, as this is a simple alternative to the full installation
 system based on the autoconf tools (see below).
-Please refer to the makefile.unix for additional details and customizations.
+Please refer to makefile.unix for additional details and customizations.
 
 Using autoconf tools:
 ---------------------
@@ -71,16 +71,17 @@ should be all you need to have the following files
 
 installed.
 
-Note: The encoding and decoding libraries are compiled separately
-(as src/dec/libwebpdecode.* and src/dec/libwebpencode.*). They
-can be installed independently using a minor modification in the
-corresponding Makefile.am configure files (see comments there).
+Note: A decode-only library, libwebpdecoder, is available using the
+'--enable-libwebpdecoder' flag. The encode library is built separately and can
+be installed independently using a minor modification in the corresponding
+Makefile.am configure files (see comments there). See './configure --help' for
+more options.
 
 SWIG bindings:
 --------------
 
-To generate language bindings from swig/libwebp.i swig-1.3
-(http://www.swig.org) is required. 2.0 may work, but has not been tested.
+To generate language bindings from swig/libwebp.swig at least swig-1.3
+(http://www.swig.org) is required.
 
 Currently the following functions are mapped:
 Decode:
@@ -103,12 +104,20 @@ Encode:
   WebPEncodeLosslessRGB
   WebPEncodeLosslessBGR
 
+See swig/README for more detailed build instructions.
+
 Java bindings:
 
 To build the swig-generated JNI wrapper code at least JDK-1.5 (or equivalent)
 is necessary for enum support. The output is intended to be a shared object /
 DLL that can be loaded via System.loadLibrary("webp_jni").
 
+Python bindings:
+
+To build the swig-generated Python extension code at least Python 2.6 is
+required. Python < 2.6 may build with some minor changes to libwebp.swig or the
+generated code, but is untested.
+
 Encoding tool:
 ==============
 
@@ -131,57 +140,71 @@ A longer list of options is available using the -longhelp command line flag:
 Usage:
  cwebp [-preset <...>] [options] in_file [-o out_file]
 
-If input size (-s) for an image is not specified, it is assumed to be a PNG,
-JPEG or TIFF file.
-options:
+If input size (-s) for an image is not specified, it is
+assumed to be a PNG, JPEG, TIFF or WebP file.
+
+Options:
   -h / -help  ............ short help
   -H / -longhelp  ........ long help
   -q <float> ............. quality factor (0:small..100:big)
-  -alpha_q <int> ......... Transparency-compression quality (0..100).
-  -preset <string> ....... Preset setting, one of:
+  -alpha_q <int> ......... transparency-compression quality (0..100)
+  -preset <string> ....... preset setting, one of:
                             default, photo, picture,
                             drawing, icon, text
-     -preset must come first, as it overwrites other parameters.
+     -preset must come first, as it overwrites other parameters
+
   -m <int> ............... compression method (0=fast, 6=slowest)
   -segments <int> ........ number of segments to use (1..4)
-  -size <int> ............ Target size (in bytes)
-  -psnr <float> .......... Target PSNR (in dB. typically: 42)
+  -size <int> ............ target size (in bytes)
+  -psnr <float> .......... target PSNR (in dB. typically: 42)
 
-  -s <int> <int> ......... Input size (width x height) for YUV
-  -sns <int> ............. Spatial Noise Shaping (0:off, 100:max)
+  -s <int> <int> ......... input size (width x height) for YUV
+  -sns <int> ............. spatial noise shaping (0:off, 100:max)
   -f <int> ............... filter strength (0=off..100)
   -sharpness <int> ....... filter sharpness (0:most .. 7:least sharp)
-  -strong ................ use strong filter instead of simple.
+  -strong ................ use strong filter instead of simple (default)
+  -nostrong .............. use simple filter instead of strong
   -partition_limit <int> . limit quality to fit the 512k limit on
                            the first partition (0=no degradation ... 100=full)
   -pass <int> ............ analysis pass number (1..10)
   -crop <x> <y> <w> <h> .. crop picture with the given rectangle
   -resize <w> <h> ........ resize picture (after any cropping)
-  -map <int> ............. print map of extra info.
-  -print_ssim ............ prints averaged SSIM distortion.
-  -print_psnr ............ prints averaged PSNR distortion.
-  -d <file.pgm> .......... dump the compressed output (PGM file).
-  -alpha_method <int> .... Transparency-compression method (0..1)
-  -alpha_filter <string> . predictive filtering for alpha plane.
-                           One of: none, fast (default) or best.
-  -alpha_cleanup ......... Clean RGB values in transparent area.
-  -noalpha ............... discard any transparency information.
-  -lossless .............. Encode image losslessly.
-  -hint <string> ......... Specify image characteristics hint.
-                           One of: photo, picture or graph
+  -mt .................... use multi-threading if available
+  -low_memory ............ reduce memory usage (slower encoding)
+  -map <int> ............. print map of extra info
+  -print_psnr ............ prints averaged PSNR distortion
+  -print_ssim ............ prints averaged SSIM distortion
+  -print_lsim ............ prints local-similarity distortion
+  -d <file.pgm> .......... dump the compressed output (PGM file)
+  -alpha_method <int> .... transparency-compression method (0..1)
+  -alpha_filter <string> . predictive filtering for alpha plane,
+                           one of: none, fast (default) or best
+  -alpha_cleanup ......... clean RGB values in transparent area
+  -blend_alpha <hex> ..... blend colors against background color
+                           expressed as RGB values written in
+                           hexadecimal, e.g. 0xc0e0d0 for red=0xc0
+                           green=0xe0 and blue=0xd0
+  -noalpha ............... discard any transparency information
+  -lossless .............. encode image losslessly
+  -hint <string> ......... specify image characteristics hint,
+                           one of: photo, picture or graph
+
+  -metadata <string> ..... comma separated list of metadata to
+                           copy from the input to the output if present.
+                           Valid values: all, none (default), exif, icc, xmp
 
   -short ................. condense printed message
-  -quiet ................. don't print anything.
-  -version ............... print version number and exit.
-  -noasm ................. disable all assembly optimizations.
+  -quiet ................. don't print anything
+  -version ............... print version number and exit
+  -noasm ................. disable all assembly optimizations
   -v ..................... verbose, e.g. print encoding/decoding times
   -progress .............. report encoding progress
 
 Experimental Options:
-  -af .................... auto-adjust filter strength.
+  -jpeg_like ............. roughly match expected JPEG size
+  -af .................... auto-adjust filter strength
   -pre <int> ............. pre-processing filter
 
-
 The main options you might want to try in order to further tune the
 visual quality are:
  -preset
@@ -205,8 +228,8 @@ Namely:
      in-loop processing. The higher the value, the smoother the
      highly-compressed area will look. This is particularly useful when aiming
      at very small files. Typical values are around 20-30. Note that using the
-     option -strong will change the type of filtering. Use "-f 0" to turn
-     filtering off.
+     option -strong/-nostrong will change the type of filtering. Use "-f 0" to
+     turn filtering off.
   * 'm' controls the trade-off between encoding speed and quality. Default is 4.
      You can try -m 5 or -m 6 to explore more (time-consuming) encoding
      possibilities. A lower value will result in faster encoding at the expense
@@ -233,31 +256,115 @@ Decodes the WebP image file to PNG format [Default]
 Use following options to convert into alternate image formats:
   -pam ......... save the raw RGBA samples as a color PAM
   -ppm ......... save the raw RGB samples as a color PPM
+  -bmp ......... save as uncompressed BMP format
+  -tiff ........ save as uncompressed TIFF format
   -pgm ......... save the raw YUV samples as a grayscale PGM
-                 file with IMC4 layout.
+                 file with IMC4 layout
+  -yuv ......... save the raw YUV samples in flat layout
+
  Other options are:
-  -version  .... print version number and exit.
-  -nofancy ..... don't use the fancy YUV420 upscaler.
-  -nofilter .... disable in-loop filtering.
+  -version  .... print version number and exit
+  -nofancy ..... don't use the fancy YUV420 upscaler
+  -nofilter .... disable in-loop filtering
+  -nodither .... disable dithering
+  -dither <d> .. dithering strength (in 0..100)
   -mt .......... use multi-threading
   -crop <x> <y> <w> <h> ... crop output with the given rectangle
   -scale <w> <h> .......... scale the output (*after* any cropping)
-  -alpha ....... only save the alpha plane.
-  -h     ....... this help message.
+  -alpha ....... only save the alpha plane
+  -incremental . use incremental decoding (useful for tests)
+  -h     ....... this help message
   -v     ....... verbose (e.g. print encoding/decoding times)
-  -noasm ....... disable all assembly optimizations.
+  -noasm ....... disable all assembly optimizations
 
 Visualization tool:
 ===================
 
 There's a little self-serve visualization tool called 'vwebp' under the
 examples/ directory. It uses OpenGL to open a simple drawing window and show
-a decoded WebP file. It's not yet integrated in the automake or makefile.unix
-build system, but you can try to manually compile it using the recommendations
-at the top of the source file.
+a decoded WebP file. It's not yet integrated in the automake build system, but
+you can try to manually compile it using the recommendations below.
 
-Usage: 'vwebp my_picture.webp'
+Usage: vwebp in_file [options]
 
+Decodes the WebP image file and visualize it using OpenGL
+Options are:
+  -version  .... print version number and exit
+  -noicc ....... don't use the icc profile if present
+  -nofancy ..... don't use the fancy YUV420 upscaler
+  -nofilter .... disable in-loop filtering
+  -dither <int>  dithering strength (0..100), default=50
+  -mt .......... use multi-threading
+  -info ........ print info
+  -h     ....... this help message
+
+Keyboard shortcuts:
+  'c' ................ toggle use of color profile
+  'i' ................ overlay file information
+  'q' / 'Q' / ESC .... quit
+
+Building:
+---------
+
+Prerequisites:
+1) OpenGL & OpenGL Utility Toolkit (GLUT)
+  Linux:
+    $ sudo apt-get install freeglut3-dev mesa-common-dev
+  Mac + XCode:
+    - These libraries should be available in the OpenGL / GLUT frameworks.
+  Windows:
+    http://freeglut.sourceforge.net/index.php#download
+
+2) (Optional) qcms (Quick Color Management System)
+  i. Download qcms from Mozilla / Chromium:
+    http://hg.mozilla.org/mozilla-central/file/0e7639e3bdfb/gfx/qcms
+    http://src.chromium.org/viewvc/chrome/trunk/src/third_party/qcms
+  ii. Build and archive the source files as libqcms.a / qcms.lib
+  iii. Update makefile.unix / Makefile.vc
+    a) Define WEBP_HAVE_QCMS
+    b) Update include / library paths to reference the qcms directory.
+
+Build using makefile.unix / Makefile.vc:
+$ make -f makefile.unix examples/vwebp
+> nmake /f Makefile.vc CFG=release-static \
+    ../obj/x64/release-static/bin/vwebp.exe
+
+Animated GIF conversion:
+========================
+Animated GIF files can be converted to WebP files with animation using the
+gif2webp utility available under examples/. The files can then be viewed using
+vwebp.
+
+Usage:
+ gif2webp [options] gif_file -o webp_file
+Options:
+  -h / -help  ............ this help
+  -lossy ................. encode image using lossy compression
+  -mixed ................. for each frame in the image, pick lossy
+                           or lossless compression heuristically
+  -q <float> ............. quality factor (0:small..100:big)
+  -m <int> ............... compression method (0=fast, 6=slowest)
+  -kmin <int> ............ min distance between key frames
+  -kmax <int> ............ max distance between key frames
+  -f <int> ............... filter strength (0=off..100)
+  -metadata <string> ..... comma separated list of metadata to
+                           copy from the input to the output if present
+                           Valid values: all, none, icc, xmp (default)
+  -mt .................... use multi-threading if available
+
+  -version ............... print version number and exit
+  -v ..................... verbose
+  -quiet ................. don't print anything
+
+Building:
+---------
+With the libgif development files installed, gif2webp can be built using
+makefile.unix:
+$ make -f makefile.unix examples/gif2webp
+
+or using autoconf:
+$ ./configure --enable-everything
+$ make
 
 Encoding API:
 =============
@@ -336,15 +443,20 @@ The encoding flow looks like:
 
   // Set up a byte-output write method. WebPMemoryWriter, for instance.
   WebPMemoryWriter wrt;
+  WebPMemoryWriterInit(&wrt);     // initialize 'wrt'
+
   pic.writer = MyFileWriter;
   pic.custom_ptr = my_opaque_structure_to_make_MyFileWriter_work;
-  // initialize 'wrt' here...
 
   // Compress!
   int ok = WebPEncode(&config, &pic);   // ok = 0 => error occurred!
   WebPPictureFree(&pic);  // must be called independently of the 'ok' result.
 
   // output data should have been handled by the writer at that point.
+  // -> compressed data is the memory buffer described by wrt.mem / wrt.size
+
+  // deallocate the memory used by compressed data
+  WebPMemoryWriterClear(&wrt);
 
 -------------------------------------- END PSEUDO EXAMPLE
 

README.mux

@@ -1,23 +1,24 @@
           __   __  ____  ____  ____  __ __  _     __ __
          /  \\/  \/  _ \/  _ \/  _ \/  \  \/ \___/_ / _\
          \       /   __/  _  \   __/      /  /  (_/  /__
-          \__\__/\_____/_____/__/  \__//_/\_____/__/___/
+          \__\__/\_____/_____/__/  \__//_/\_____/__/___/v0.2.2
 
 
 Description:
 ============
 
-WebP Mux: library to create a WebP container object for features like
-color profile, XMP metadata, animation & tiling. A reference command line
-tool 'webpmux' and WebP container specification 'doc/webp-container-spec.txt'
-are also provided in this package.
+WebPMux: set of two libraries 'Mux' and 'Demux' for creation, extraction and
+manipulation of an extended format WebP file, which can have features like
+color profile, metadata and animation. Reference command-line tools 'webpmux'
+and 'vwebp' as well as the WebP container specification
+'doc/webp-container-spec.txt' are also provided in this package.
 
 WebP Mux tool:
 ==============
 
 The examples/ directory contains a tool (webpmux) for manipulating WebP
-files. The webpmux tool can be used to create a WebP container file and to
-extract or strip relevant data from the container file.
+files. The webpmux tool can be used to create an extended format WebP file and
+also to extract or strip relevant data from such a file.
 
 A list of options is available using the -help command line flag:
 
@@ -25,82 +26,150 @@ A list of options is available using the -help command line flag:
 Usage: webpmux -get GET_OPTIONS INPUT -o OUTPUT
        webpmux -set SET_OPTIONS INPUT -o OUTPUT
        webpmux -strip STRIP_OPTIONS INPUT -o OUTPUT
-       webpmux -tile TILE_OPTIONS [-tile...] -o OUTPUT
-       webpmux -frame FRAME_OPTIONS [-frame...] -loop LOOP_COUNT -o OUTPUT
+       webpmux -frame FRAME_OPTIONS [-frame...] [-loop LOOP_COUNT]
+               [-bgcolor BACKGROUND_COLOR] -o OUTPUT
        webpmux -info INPUT
        webpmux [-h|-help]
+       webpmux -version
 
 GET_OPTIONS:
- Extract relevant data.
-   icc       Get ICCP Color profile.
-   xmp       Get XMP metadata.
-   tile n    Get nth tile.
-   frame n   Get nth frame.
+ Extract relevant data:
+   icc       get ICC profile
+   exif      get EXIF metadata
+   xmp       get XMP metadata
+   frame n   get nth frame
 
 SET_OPTIONS:
- Set color profile/metadata.
-   icc       Set ICC Color profile.
-   xmp       Set XMP metadata.
+ Set color profile/metadata:
+   icc  file.icc     set ICC profile
+   exif file.exif    set EXIF metadata
+   xmp  file.xmp     set XMP metadata
+   where:    'file.icc' contains the ICC profile to be set,
+             'file.exif' contains the EXIF metadata to be set
+             'file.xmp' contains the XMP metadata to be set
 
 STRIP_OPTIONS:
- Strip color profile/metadata.
-   icc       Strip ICCP color profile.
-   xmp       Strip XMP metadata.
-
-TILE_OPTIONS(i):
- Create tiled image.
-   file_i +xi+yi
-   where:    'file_i' is the i'th tile (webp format),
-             'xi','yi' specify the image offset for this tile.
+ Strip color profile/metadata:
+   icc       strip ICC profile
+   exif      strip EXIF metadata
+   xmp       strip XMP metadata
 
 FRAME_OPTIONS(i):
- Create animation.
-   file_i +xi+yi+di
-   where:    'file_i' is the i'th animation frame (webp format),
-             'xi','yi' specify the image offset for this frame.
-             'di' is the pause duration before next frame.
+ Create animation:
+   file_i +di+[xi+yi[+mi[bi]]]
+   where:    'file_i' is the i'th animation frame (WebP format),
+             'di' is the pause duration before next frame,
+             'xi','yi' specify the image offset for this frame,
+             'mi' is the dispose method for this frame (0 or 1),
+             'bi' is the blending method for this frame (+b or -b)
 
-INPUT & OUTPUT are in webp format.
+LOOP_COUNT:
+ Number of times to repeat the animation.
+ Valid range is 0 to 65535 [Default: 0 (infinite)].
 
-WebP Mux API:
-==============
-The WebP Mux API contains methods for adding data to and reading data from
-WebPMux (a WebP container object). This API currently supports XMP metadata,
-color profile, animation & tiling. Other features will be added in subsequent
-releases.
+BACKGROUND_COLOR:
+ Background color of the canvas.
+  A,R,G,B
+  where:    'A', 'R', 'G' and 'B' are integers in the range 0 to 255 specifying
+            the Alpha, Red, Green and Blue component values respectively
+            [Default: 255,255,255,255]
+
+INPUT & OUTPUT are in WebP format.
+
+Note: The nature of EXIF, XMP and ICC data is not checked and is assumed to be
+valid.
+
+Visualization tool:
+===================
+
+The examples/ directory also contains a tool (vwebp) for viewing WebP files.
+It decodes the image and visualizes it using OpenGL. See the libwebp README
+for details on building and running this program.
+
+Mux API:
+========
+The Mux API contains methods for adding data to and reading data from WebP
+files. This API currently supports XMP/EXIF metadata, ICC profile and animation.
+Other features may be added in subsequent releases.
 
 Example#1 (pseudo code): Creating a WebPMux object with image data, color
-profile & XMP metadata.
+profile and XMP metadata.
 
   int copy_data = 0;
   WebPMux* mux = WebPMuxNew();
   // ... (Prepare image data).
   WebPMuxSetImage(mux, &image, copy_data);
-  // ... (Prepare ICCP color profile data).
-  WebPMuxSetColorProfile(mux, &icc_profile, copy_data);
+  // ... (Prepare ICC profile data).
+  WebPMuxSetChunk(mux, "ICCP", &icc_profile, copy_data);
   // ... (Prepare XMP metadata).
-  WebPMuxSetMetadata(mux, &xmp, copy_data);
+  WebPMuxSetChunk(mux, "XMP ", &xmp, copy_data);
   // Get data from mux in WebP RIFF format.
   WebPMuxAssemble(mux, &output_data);
   WebPMuxDelete(mux);
-  // ... (Consume output_data; e.g. write output_data.bytes_ to file).
+  // ... (Consume output_data; e.g. write output_data.bytes to file).
   WebPDataClear(&output_data);
 
 
-Example#2 (pseudo code): Get image & color profile data from a WebP file.
+Example#2 (pseudo code): Get image and color profile data from a WebP file.
 
   int copy_data = 0;
   // ... (Read data from file).
   WebPMux* mux = WebPMuxCreate(&data, copy_data);
-  WebPMuxGetImage(mux, &image);
+  WebPMuxGetFrame(mux, 1, &image);
   // ... (Consume image; e.g. call WebPDecode() to decode the data).
-  WebPMuxGetColorProfile(mux, &icc_profile);
+  WebPMuxGetChunk(mux, "ICCP", &icc_profile);
   // ... (Consume icc_profile).
   WebPMuxDelete(mux);
   free(data);
 
 
-For detailed Mux API reference, please refer to the header file (src/webp/mux.h)
+For a detailed Mux API reference, please refer to the header file
+(src/webp/mux.h).
+
+Demux API:
+==========
+The Demux API enables extraction of images and extended format data from
+WebP files. This API currently supports reading of XMP/EXIF metadata, ICC
+profile and animated images. Other features may be added in subsequent
+releases.
+
+Code Example: Demuxing WebP data to extract all the frames, ICC profile
+and EXIF/XMP metadata.
+
+  WebPDemuxer* demux = WebPDemux(&webp_data);
+  uint32_t width = WebPDemuxGetI(demux, WEBP_FF_CANVAS_WIDTH);
+  uint32_t height = WebPDemuxGetI(demux, WEBP_FF_CANVAS_HEIGHT);
+  // ... (Get information about the features present in the WebP file).
+  uint32_t flags = WebPDemuxGetI(demux, WEBP_FF_FORMAT_FLAGS);
+
+  // ... (Iterate over all frames).
+  WebPIterator iter;
+  if (WebPDemuxGetFrame(demux, 1, &iter)) {
+    do {
+      // ... (Consume 'iter'; e.g. Decode 'iter.fragment' with WebPDecode(),
+      // ... and get other frame properties like width, height, offsets etc.
+      // ... see 'struct WebPIterator' below for more info).
+    } while (WebPDemuxNextFrame(&iter));
+    WebPDemuxReleaseIterator(&iter);
+  }
+
+  // ... (Extract metadata).
+  WebPChunkIterator chunk_iter;
+  if (flags & ICCP_FLAG) WebPDemuxGetChunk(demux, "ICCP", 1, &chunk_iter);
+  // ... (Consume the ICC profile in 'chunk_iter.chunk').
+  WebPDemuxReleaseChunkIterator(&chunk_iter);
+  if (flags & EXIF_FLAG) WebPDemuxGetChunk(demux, "EXIF", 1, &chunk_iter);
+  // ... (Consume the EXIF metadata in 'chunk_iter.chunk').
+  WebPDemuxReleaseChunkIterator(&chunk_iter);
+  if (flags & XMP_FLAG) WebPDemuxGetChunk(demux, "XMP ", 1, &chunk_iter);
+  // ... (Consume the XMP metadata in 'chunk_iter.chunk').
+  WebPDemuxReleaseChunkIterator(&chunk_iter);
+  WebPDemuxDelete(demux);
+
+
+For a detailed Demux API reference, please refer to the header file
+(src/webp/demux.h).
+
 
 Bugs:
 =====

configure.ac

@@ -1,21 +1,115 @@
-AC_INIT([libwebp], [0.2.1],
+AC_INIT([libwebp], [0.4.2],
         [http://code.google.com/p/webp/issues],,
         [http://developers.google.com/speed/webp])
-AC_CANONICAL_TARGET
+AC_CANONICAL_HOST
+AC_PREREQ([2.60])
 AM_INIT_AUTOMAKE([-Wall foreign subdir-objects])
+
+dnl === automake >= 1.12 requires this for 'unusual archivers' support.
+dnl === it must occur before LT_INIT (AC_PROG_LIBTOOL).
+m4_ifdef([AM_PROG_AR], [AM_PROG_AR])
+
 AC_PROG_LIBTOOL
 AM_PROG_CC_C_O
 
 dnl === Enable less verbose output when building.
-dnl === If an older aclocal exits with an error comment these lines out.
-m4_define_default([AM_SILENT_RULES], [])
-AM_SILENT_RULES
+m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
+
+dnl == test endianness
+AC_C_BIGENDIAN
+
+dnl === SET_IF_UNSET(shell_var, value)
+dnl ===   Set the shell variable 'shell_var' to 'value' if it is unset.
+AC_DEFUN([SET_IF_UNSET], [test "${$1+set}" = "set" || $1=$2])
+
+AC_ARG_ENABLE([everything],
+              AS_HELP_STRING([--enable-everything],
+                             [Enable all optional targets. These can still be
+                              disabled with --disable-target]),
+              [SET_IF_UNSET([enable_libwebpdecoder], [$enableval])
+               SET_IF_UNSET([enable_libwebpdemux], [$enableval])
+               SET_IF_UNSET([enable_libwebpmux], [$enableval])])
 
 AC_ARG_WITH([pkgconfigdir], AS_HELP_STRING([--with-pkgconfigdir=DIR],
             [Path to the pkgconfig directory @<:@LIBDIR/pkgconfig@:>@]),
             [pkgconfigdir="$withval"], [pkgconfigdir='${libdir}/pkgconfig'])
 AC_SUBST([pkgconfigdir])
 
+dnl === TEST_AND_ADD_CFLAGS(var, flag)
+dnl ===   Checks whether $CC supports 'flag' and adds it to 'var'
+dnl ===   on success.
+AC_DEFUN([TEST_AND_ADD_CFLAGS],
+         [SAVED_CFLAGS="$CFLAGS"
+          CFLAGS="-Werror $2"
+          AC_MSG_CHECKING([whether $CC supports $2])
+          dnl Note AC_LANG_PROGRAM([]) uses an old-style main definition.
+          AC_COMPILE_IFELSE([AC_LANG_SOURCE([int main(void) { return 0; }])],
+                            [AC_MSG_RESULT([yes])]
+                            dnl Simply append the variable avoiding a
+                            dnl compatibility ifdef for AS_VAR_APPEND as this
+                            dnl variable shouldn't grow all that large.
+                            [$1="${$1} $2"],
+                            [AC_MSG_RESULT([no])])
+          CFLAGS="$SAVED_CFLAGS"])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wall])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wdeclaration-after-statement])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wextra])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wformat-security])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wmissing-declarations])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wmissing-prototypes])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wold-style-definition])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wshadow])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wunused-but-set-variable])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wunused])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wvla])
+# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62040
+# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61622
+AS_IF([test "$GCC" = "yes" ], [
+       gcc_version=`$CC -dumpversion`
+       gcc_wht_bug=""
+       case "$host_cpu" in
+         aarch64|arm64)
+          case "$gcc_version" in
+            4.9|4.9.0|4.9.1) gcc_wht_bug=yes ;;
+          esac
+       esac
+       AS_IF([test "$gcc_wht_bug" = "yes"], [
+              TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-frename-registers])])])
+AC_SUBST([AM_CFLAGS])
+
+dnl === Check for machine specific flags
+TEST_AND_ADD_CFLAGS([AVX2_FLAGS], [-mavx2])
+AS_IF([test -n "$AVX2_FLAGS"], [
+  SAVED_CFLAGS=$CFLAGS
+  CFLAGS="$CFLAGS $AVX2_FLAGS"
+  AC_CHECK_HEADER([immintrin.h],
+                  [AC_DEFINE(WEBP_HAVE_AVX2, [1],
+                   [Set to 1 if AVX2 is supported])],
+                  [AVX2_FLAGS=""],
+                  dnl it's illegal to directly include avx2intrin.h, but it's
+                  dnl included conditionally in immintrin.h, tricky!
+                  [#ifndef __AVX2__
+                   #error avx2 is not enabled
+                   #endif
+                  ])
+  CFLAGS=$SAVED_CFLAGS])
+AC_SUBST([AVX2_FLAGS])
+
+TEST_AND_ADD_CFLAGS([SSE2_FLAGS], [-msse2])
+AS_IF([test -n "$SSE2_FLAGS"], [
+  SAVED_CFLAGS=$CFLAGS
+  CFLAGS="$CFLAGS $SSE2_FLAGS"
+  AC_CHECK_HEADER([emmintrin.h],
+                  [AC_DEFINE(WEBP_HAVE_SSE2, [1],
+                   [Set to 1 if SSE2 is supported])],
+                  [SSE2_FLAGS=""])
+  CFLAGS=$SAVED_CFLAGS])
+AC_SUBST([SSE2_FLAGS])
+
+dnl === CLEAR_LIBVARS([var_pfx])
+dnl ===   Clears <var_pfx>_{INCLUDES,LIBS}.
+AC_DEFUN([CLEAR_LIBVARS], [$1_INCLUDES=""; $1_LIBS=""])
+
 dnl === WITHLIB_OPTION([opt_pfx], [outvar_pfx])
 dnl ===   Defines --with-<opt_pfx>{include,lib}dir options which set
 dnl ===   the variables <outvar_pfx>_{INCLUDES,LIBS}.
@@ -29,6 +123,44 @@ AC_DEFUN([WITHLIB_OPTION],
                               [use $2 libraries from DIR]),
                [$2_LIBS="-L$withval"])])
 
+dnl === LIBCHECK_PROLOGUE([var_pfx])
+dnl ===   Caches the current values of CPPFLAGS/LIBS in SAVED_* then
+dnl ===   prepends the current values with <var_pfx>_{INCLUDES,LIBS}.
+AC_DEFUN([LIBCHECK_PROLOGUE],
+         [SAVED_CPPFLAGS=$CPPFLAGS
+          SAVED_LIBS=$LIBS
+          CPPFLAGS="$$1_INCLUDES $CPPFLAGS"
+          LIBS="$$1_LIBS $LIBS"])
+
+dnl === LIBCHECK_EPILOGUE([var_pfx])
+dnl ===   Restores the values of CPPFLAGS/LIBS from SAVED_* and exports
+dnl ===   <var_pfx>_{INCLUDES,LIBS} with AC_SUBST.
+AC_DEFUN([LIBCHECK_EPILOGUE],
+         [AC_SUBST($1_LIBS)
+          AC_SUBST($1_INCLUDES)
+          CPPFLAGS=$SAVED_CPPFLAGS
+          LIBS=$SAVED_LIBS])
+
+dnl === Check for gcc builtins
+
+dnl === CHECK_FOR_BUILTIN([builtin], [param], [define])
+dnl ===   links a C AC_LANG_PROGRAM, with <builtin>(<param>)
+dnl ===   AC_DEFINE'ing <define> if successful.
+AC_DEFUN([CHECK_FOR_BUILTIN],
+         [AC_LANG_PUSH([C])
+          AC_MSG_CHECKING([for $1])
+          AC_LINK_IFELSE([AC_LANG_PROGRAM([], [$1($2)])],
+                         [AC_MSG_RESULT([yes])
+                          AC_DEFINE([$3], [1],
+                                    [Set to 1 if $1 is available])],
+                         [AC_MSG_RESULT([no])]),
+          AC_LANG_POP])
+
+dnl AC_CHECK_FUNC doesn't work with builtin's.
+CHECK_FOR_BUILTIN([__builtin_bswap16], [1u << 15], [HAVE_BUILTIN_BSWAP16])
+CHECK_FOR_BUILTIN([__builtin_bswap32], [1u << 31], [HAVE_BUILTIN_BSWAP32])
+CHECK_FOR_BUILTIN([__builtin_bswap64], [1ull << 63], [HAVE_BUILTIN_BSWAP64])
+
 dnl === Check for pthread support
 AC_ARG_ENABLE([threading],
               AS_HELP_STRING([--disable-threading],
@@ -42,115 +174,254 @@ if test "$enable_threading" = "yes"; then
               CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
               CC="$PTHREAD_CC"
              ],
-             [enable_threading=no])
+             [AC_CHECK_FUNC([_beginthreadex],
+                            [AC_DEFINE([WEBP_USE_THREAD], [1],
+                                       [Undefine this to disable thread
+                                        support.])],
+                            [enable_threading=no])])
 fi
 AC_MSG_NOTICE([checking if threading is enabled... ${enable_threading-no}])
 
+dnl === check for OpenGL/GLUT support ===
+
+AC_ARG_ENABLE([gl], AS_HELP_STRING([--disable-gl],
+                                   [Disable detection of OpenGL support
+                                    @<:@default=auto@:>@]))
+AS_IF([test "x$enable_gl" != "xno"], [
+  CLEAR_LIBVARS([GL])
+  WITHLIB_OPTION([gl], [GL])
+
+  LIBCHECK_PROLOGUE([GL])
+
+  glut_cflags="none"
+  glut_ldflags="none"
+  case $host_os in
+    darwin*)
+      # Special case for OSX builds. Append these to give the user a chance to
+      # override with --with-gl*
+      glut_cflags="$glut_cflags|-framework GLUT -framework OpenGL"
+      glut_ldflags="$glut_ldflags|-framework GLUT -framework OpenGL"
+      ;;
+  esac
+
+  GLUT_SAVED_CPPFLAGS="$CPPFLAGS"
+  SAVED_IFS="$IFS"
+  IFS="|"
+  for flag in $glut_cflags; do
+    # restore IFS immediately as the autoconf macros may need the default.
+    IFS="$SAVED_IFS"
+    unset ac_cv_header_GL_glut_h
+    unset ac_cv_header_OpenGL_glut_h
+
+    case $flag in
+      none) ;;
+      *) CPPFLAGS="$flag $CPPFLAGS";;
+    esac
+    AC_CHECK_HEADERS([GL/glut.h GLUT/glut.h OpenGL/glut.h],
+                     [glut_headers=yes;
+                      test "$flag" = "none" || GL_INCLUDES="$CPPFLAGS";
+                      break])
+    CPPFLAGS="$GLUT_SAVED_CPPFLAGS"
+    test "$glut_headers" = "yes" && break
+  done
+  IFS="$SAVED_IFS"
+
+  if test "$glut_headers" = "yes"; then
+    AC_LANG_PUSH([C])
+    GLUT_SAVED_LDFLAGS="$LDFLAGS"
+    SAVED_IFS="$IFS"
+    IFS="|"
+    for flag in $glut_ldflags; do
+      # restore IFS immediately as the autoconf macros may need the default.
+      IFS="$SAVED_IFS"
+      unset ac_cv_search_glBegin
+
+      case $flag in
+        none) ;;
+        *) LDFLAGS="$flag $LDFLAGS";;
+      esac
+
+      # find libGL
+      GL_SAVED_LIBS="$LIBS"
+      AC_SEARCH_LIBS([glBegin], [GL OpenGL opengl32])
+      LIBS="$GL_SAVED_LIBS"
+
+      # A direct link to libGL may not be necessary on e.g., linux.
+      GLUT_SAVED_LIBS="$LIBS"
+      for lib in "" "-lglut" "-lglut $ac_cv_search_glBegin"; do
+        LIBS="$lib"
+        AC_LINK_IFELSE(
+          [AC_LANG_PROGRAM([
+             #ifdef __cplusplus
+             # define EXTERN_C extern "C"
+             #else
+             # define EXTERN_C
+             #endif
+             EXTERN_C char glOrtho();
+             EXTERN_C char glutMainLoop();
+            ],[
+             glOrtho();
+             glutMainLoop();
+            ])
+          ],
+          AC_DEFINE(WEBP_HAVE_GL, [1],
+                    [Set to 1 if OpenGL is supported])
+          [glut_support=yes], []
+        )
+        if test "$glut_support" = "yes"; then
+          GL_LIBS="$LDFLAGS $lib"
+          break
+        fi
+      done
+      LIBS="$GLUT_SAVED_LIBS"
+      LDFLAGS="$GLUT_SAVED_LDFLAGS"
+      test "$glut_support" = "yes" && break
+    done
+    IFS="$SAVED_IFS"
+    AC_LANG_POP
+  fi
+
+  LIBCHECK_EPILOGUE([GL])
+
+  if test "$glut_support" = "yes" -a "$enable_libwebpdemux" = "yes"; then
+    build_vwebp=yes
+  fi
+])
+AM_CONDITIONAL([BUILD_VWEBP], [test "$build_vwebp" = "yes"])
+
 dnl === check for PNG support ===
 
-PNG_INCLUDES=""
-PNG_LIBS=""
-AC_PATH_PROGS(LIBPNG_CONFIG,
-              [libpng-config libpng15-config libpng14-config libpng12-config])
-if test -n "$LIBPNG_CONFIG"; then
-  PNG_INCLUDES=`$LIBPNG_CONFIG --cflags`
-  PNG_PREFIX=`$LIBPNG_CONFIG --prefix`
-  if test "${PNG_PREFIX}/lib" != "/usr/lib" ; then
-    PNG_LIBS="-L${PNG_PREFIX}/lib"
+AC_ARG_ENABLE([png], AS_HELP_STRING([--disable-png],
+                                    [Disable detection of PNG format support
+                                     @<:@default=auto@:>@]))
+AS_IF([test "x$enable_png" != "xno"], [
+  CLEAR_LIBVARS([PNG])
+  AC_PATH_PROGS([LIBPNG_CONFIG],
+                [libpng-config libpng16-config libpng15-config libpng14-config \
+                 libpng12-config])
+  if test -n "$LIBPNG_CONFIG"; then
+    PNG_INCLUDES=`$LIBPNG_CONFIG --cflags`
+    PNG_LIBS="`$LIBPNG_CONFIG --ldflags`"
   fi
-  PNG_LIBS="$PNG_LIBS `$LIBPNG_CONFIG --libs`"
-fi
 
-WITHLIB_OPTION([png], [PNG])
+  WITHLIB_OPTION([png], [PNG])
 
-SAVED_CPPFLAGS=$CPPFLAGS
-SAVED_LIBS=$LIBS
-CPPFLAGS="$PNG_INCLUDES $CPPFLAGS"
-LIBS="$PNG_LIBS $LIBS"
-
-AC_CHECK_HEADER(png.h,
-  AC_SEARCH_LIBS(png_get_libpng_ver, [png],
-                 [test "$ac_cv_search_png_get_libpng_ver" = "none required" \
-                    || PNG_LIBS="$PNG_LIBS $ac_cv_search_png_get_libpng_ver"
-                  PNG_INCLUDES="$PNG_INCLUDES -DWEBP_HAVE_PNG"
-                  AC_DEFINE(WEBP_HAVE_PNG, [1],
-                            [Set to 1 if PNG library is installed])
-                 ],
-                 [AC_MSG_WARN(Optional png library not found)
-                  PNG_LIBS=""
-                  PNG_INCLUDES=""
-                 ],
-                 [$MATH_LIBS]),
-  [AC_MSG_WARN(png library not available - no png.h)
-   PNG_LIBS=""
-   PNG_INCLUDES=""
-  ],
-)
-AC_SUBST(PNG_LIBS)
-AC_SUBST(PNG_INCLUDES)
-
-CPPFLAGS=$SAVED_CPPFLAGS
-LIBS=$SAVED_LIBS
+  LIBCHECK_PROLOGUE([PNG])
+  AC_CHECK_HEADER(png.h,
+    AC_SEARCH_LIBS(png_get_libpng_ver, [png],
+                   [test "$ac_cv_search_png_get_libpng_ver" = "none required" \
+                      || PNG_LIBS="$PNG_LIBS $ac_cv_search_png_get_libpng_ver"
+                    PNG_INCLUDES="$PNG_INCLUDES -DWEBP_HAVE_PNG"
+                    AC_DEFINE(WEBP_HAVE_PNG, [1],
+                              [Set to 1 if PNG library is installed])
+                    png_support=yes
+                   ],
+                   [AC_MSG_WARN(Optional png library not found)
+                    PNG_LIBS=""
+                    PNG_INCLUDES=""
+                   ],
+                   [$MATH_LIBS]),
+    [AC_MSG_WARN(png library not available - no png.h)
+     PNG_LIBS=""
+     PNG_INCLUDES=""
+    ],
+  )
+  LIBCHECK_EPILOGUE([PNG])
+])
 
 dnl === check for JPEG support ===
 
-JPEG_INCLUDES=""
-JPEG_LIBS=""
-WITHLIB_OPTION([jpeg], [JPEG])
+AC_ARG_ENABLE([jpeg],
+              AS_HELP_STRING([--disable-jpeg],
+                             [Disable detection of JPEG format support
+                              @<:@default=auto@:>@]))
+AS_IF([test "x$enable_jpeg" != "xno"], [
+  CLEAR_LIBVARS([JPEG])
+  WITHLIB_OPTION([jpeg], [JPEG])
 
-SAVED_CPPFLAGS=$CPPFLAGS
-SAVED_LIBS=$LIBS
-CPPFLAGS="$JPEG_INCLUDES $CPPFLAGS"
-LIBS="$JPEG_LIBS $LIBS"
-
-AC_CHECK_HEADER(jpeglib.h,
-  AC_CHECK_LIB(jpeg, jpeg_set_defaults,
-               [JPEG_LIBS="$JPEG_LIBS -ljpeg"
-                JPEG_INCLUDES="$JPEG_INCLUDES -DWEBP_HAVE_JPEG"
-                AC_DEFINE(WEBP_HAVE_JPEG, [1],
-                          [Set to 1 if JPEG library is installed])
-               ],
-               AC_MSG_WARN(Optional jpeg library not found),
-               [$MATH_LIBS]),
-  AC_MSG_WARN(jpeg library not available - no jpeglib.h)
-)
-AC_SUBST(JPEG_LIBS)
-AC_SUBST(JPEG_INCLUDES)
-
-CPPFLAGS=$SAVED_CPPFLAGS
-LIBS=$SAVED_LIBS
+  LIBCHECK_PROLOGUE([JPEG])
+  AC_CHECK_HEADER(jpeglib.h,
+    AC_CHECK_LIB(jpeg, jpeg_set_defaults,
+                 [JPEG_LIBS="$JPEG_LIBS -ljpeg"
+                  JPEG_INCLUDES="$JPEG_INCLUDES -DWEBP_HAVE_JPEG"
+                  AC_DEFINE(WEBP_HAVE_JPEG, [1],
+                            [Set to 1 if JPEG library is installed])
+                  jpeg_support=yes
+                 ],
+                 AC_MSG_WARN(Optional jpeg library not found),
+                 [$MATH_LIBS]),
+    AC_MSG_WARN(jpeg library not available - no jpeglib.h)
+  )
+  LIBCHECK_EPILOGUE([JPEG])
+])
 
 dnl === check for TIFF support ===
 
-TIFF_INCLUDES=""
-TIFF_LIBS=""
-WITHLIB_OPTION([tiff], [TIFF])
+AC_ARG_ENABLE([tiff],
+              AS_HELP_STRING([--disable-tiff],
+                             [Disable detection of TIFF format support
+                              @<:@default=auto@:>@]))
+AS_IF([test "x$enable_tiff" != "xno"], [
+  CLEAR_LIBVARS([TIFF])
+  WITHLIB_OPTION([tiff], [TIFF])
 
-SAVED_CPPFLAGS=$CPPFLAGS
-SAVED_LIBS=$LIBS
-CPPFLAGS="$TIFF_INCLUDES $CPPFLAGS"
-LIBS="$TIFF_LIBS $LIBS"
+  LIBCHECK_PROLOGUE([TIFF])
+  AC_CHECK_HEADER(tiffio.h,
+    AC_CHECK_LIB(tiff, TIFFGetVersion,
+                 [TIFF_LIBS="$TIFF_LIBS -ltiff"
+                  TIFF_INCLUDES="$TIFF_INCLUDES -DWEBP_HAVE_TIFF"
+                  AC_DEFINE(WEBP_HAVE_TIFF, [1],
+                            [Set to 1 if TIFF library is installed])
+                  tiff_support=yes
+                 ],
+                 AC_MSG_WARN(Optional tiff library not found),
+                 [$MATH_LIBS]),
+    AC_MSG_WARN(tiff library not available - no tiffio.h)
+  )
+  LIBCHECK_EPILOGUE([TIFF])
+])
 
-AC_CHECK_HEADER(tiffio.h,
-  AC_CHECK_LIB(tiff, TIFFGetVersion,
-               [TIFF_LIBS="$TIFF_LIBS -ltiff"
-                TIFF_INCLUDES="$TIFF_INCLUDES -DWEBP_HAVE_TIFF"
-                AC_DEFINE(WEBP_HAVE_TIFF, [1],
-                          [Set to 1 if TIFF library is installed])
-               ],
-               AC_MSG_WARN(Optional tiff library not found),
-               [$MATH_LIBS]),
-  AC_MSG_WARN(tiff library not available - no tiffio.h)
-)
-AC_SUBST(TIFF_LIBS)
-AC_SUBST(TIFF_INCLUDES)
+dnl === check for GIF support ===
 
-CPPFLAGS=$SAVED_CPPFLAGS
-LIBS=$SAVED_LIBS
+AC_ARG_ENABLE([gif], AS_HELP_STRING([--disable-gif],
+                                    [Disable detection of GIF format support
+                                     @<:@default=auto@:>@]))
+AS_IF([test "x$enable_gif" != "xno"], [
+  CLEAR_LIBVARS([GIF])
+  WITHLIB_OPTION([gif], [GIF])
+
+  LIBCHECK_PROLOGUE([GIF])
+  AC_CHECK_HEADER(gif_lib.h,
+    AC_CHECK_LIB([gif], [DGifOpenFileHandle],
+                 [GIF_LIBS="$GIF_LIBS -lgif"
+                  AC_DEFINE(WEBP_HAVE_GIF, [1],
+                            [Set to 1 if GIF library is installed])
+                  gif_support=yes
+                 ],
+                 AC_MSG_WARN(Optional gif library not found),
+                 [$MATH_LIBS]),
+    AC_MSG_WARN(gif library not available - no gif_lib.h)
+  )
+  LIBCHECK_EPILOGUE([GIF])
+
+  if test "$gif_support" = "yes" -a \
+          "$enable_libwebpmux" = "yes"; then
+    build_gif2webp=yes
+  fi
+])
+AM_CONDITIONAL([BUILD_GIF2WEBP], [test "${build_gif2webp}" = "yes"])
 
 dnl === check for WIC support ===
 
-if test "$target_os" = "mingw32"; then
+AC_ARG_ENABLE([wic],
+              AS_HELP_STRING([--disable-wic],
+                             [Disable Windows Imaging Component (WIC) detection.
+                              @<:@default=auto@:>@]),,
+              [enable_wic=yes])
+
+case $host_os in
+mingw*)
+if test "$enable_wic" = "yes"; then
   AC_CHECK_HEADERS([wincodec.h shlwapi.h windows.h])
   if test "$ac_cv_header_wincodec_h" = "yes"; then
     AC_MSG_CHECKING(for Windows Imaging Component support)
@@ -190,6 +461,33 @@ if test "$target_os" = "mingw32"; then
     AC_MSG_RESULT(${wic_support-no})
   fi
 fi
+esac
+
+dnl === If --enable-aligned is defined, define WEBP_FORCE_ALIGNED
+
+AC_MSG_CHECKING(if --enable-aligned option is specified)
+AC_ARG_ENABLE([aligned],
+              AS_HELP_STRING([--enable-aligned],
+                             [Force aligned memory operations in non-dsp code
+                              (may be slower)]))
+if test "$enable_aligned" = "yes"; then
+  AC_DEFINE(WEBP_FORCE_ALIGNED, [1],
+            [Define to 1 to force aligned memory operations])
+fi
+AC_MSG_RESULT(${enable_aligned-no})
+
+dnl === If --enable-swap-16bit-csp is defined, add -DWEBP_SWAP_16BIT_CSP
+
+USE_SWAP_16BIT_CSP=""
+AC_MSG_CHECKING(if --enable-swap-16bit-csp option is specified)
+AC_ARG_ENABLE([swap-16bit-csp],
+              AS_HELP_STRING([--enable-swap-16bit-csp],
+                             [Enable byte swap for 16 bit colorspaces]))
+if test "$enable_swap_16bit_csp" = "yes"; then
+  USE_SWAP_16BIT_CSP="-DWEBP_SWAP_16BIT_CSP"
+fi
+AC_MSG_RESULT(${enable_swap_16bit_csp-no})
+AC_SUBST(USE_SWAP_16BIT_CSP)
 
 dnl === If --enable-experimental is defined, add -DWEBP_EXPERIMENTAL_FEATURES
 
@@ -198,30 +496,78 @@ AC_MSG_CHECKING(if --enable-experimental option is specified)
 AC_ARG_ENABLE([experimental], AS_HELP_STRING([--enable-experimental],
                                              [Activate experimental features]))
 if test "$enable_experimental" = "yes"; then
-        AC_DEFINE(EXPERIMENTAL,,[Enable experimental code])
-        USE_EXPERIMENTAL_CODE="-DWEBP_EXPERIMENTAL_FEATURES"
+  AC_DEFINE(WEBP_EXPERIMENTAL_FEATURES, [1], [Enable experimental code])
+  USE_EXPERIMENTAL_CODE="-DWEBP_EXPERIMENTAL_FEATURES"
 fi
 AC_MSG_RESULT(${enable_experimental-no})
 AC_SUBST(USE_EXPERIMENTAL_CODE)
 
 dnl === Check whether libwebpmux should be built
 AC_MSG_CHECKING(whether libwebpmux is to be built)
-AC_ARG_ENABLE([experimental-libwebpmux],
-              AS_HELP_STRING([--enable-experimental-libwebpmux],
+AC_ARG_ENABLE([libwebpmux],
+              AS_HELP_STRING([--enable-libwebpmux],
                              [Build libwebpmux @<:@default=no@:>@]))
-AC_MSG_RESULT(${enable_experimental_libwebpmux-no})
-AM_CONDITIONAL([WANT_MUX], [test "$enable_experimental_libwebpmux" = "yes"])
+AC_MSG_RESULT(${enable_libwebpmux-no})
+AM_CONDITIONAL([WANT_MUX], [test "$enable_libwebpmux" = "yes"])
+
+dnl === Check whether libwebpdemux should be built
+AC_MSG_CHECKING(whether libwebpdemux is to be built)
+AC_ARG_ENABLE([libwebpdemux],
+              AS_HELP_STRING([--enable-libwebpdemux],
+                             [Build libwebpdemux @<:@default=no@:>@]))
+AC_MSG_RESULT(${enable_libwebpdemux-no})
+AM_CONDITIONAL([WANT_DEMUX], [test "$enable_libwebpdemux" = "yes"])
+
+dnl === Check whether decoder library should be built.
+AC_MSG_CHECKING(whether decoder library is to be built)
+AC_ARG_ENABLE([libwebpdecoder],
+              AS_HELP_STRING([--enable-libwebpdecoder],
+                             [Build libwebpdecoder @<:@default=no@:>@]))
+AC_MSG_RESULT(${enable_libwebpdecoder-no})
+AM_CONDITIONAL([BUILD_LIBWEBPDECODER], [test "$enable_libwebpdecoder" = "yes"])
 
 dnl =========================
 
 AC_CONFIG_MACRO_DIR([m4])
-AC_CONFIG_HEADERS([config.h])
+AC_CONFIG_HEADERS([src/webp/config.h])
 AC_CONFIG_FILES([Makefile src/Makefile man/Makefile \
                  examples/Makefile src/dec/Makefile \
                  src/enc/Makefile src/dsp/Makefile \
+                 src/demux/Makefile src/mux/Makefile \
                  src/utils/Makefile \
-                 src/mux/Makefile \
-                 src/libwebp.pc])
+                 src/libwebp.pc src/libwebpdecoder.pc \
+                 src/demux/libwebpdemux.pc src/mux/libwebpmux.pc])
 
 
 AC_OUTPUT
+
+AC_MSG_NOTICE([
+WebP Configuration Summary
+--------------------------
+
+Shared libraries: ${enable_shared}
+Static libraries: ${enable_static}
+Threading support: ${enable_threading-no}
+libwebp: yes
+libwebpdecoder: ${enable_libwebpdecoder-no}
+libwebpdemux: ${enable_libwebpdemux-no}
+libwebpmux: ${enable_libwebpmux-no}
+
+Tools:
+cwebp : yes
+  Input format support
+  ====================
+  JPEG : ${jpeg_support-no}
+  PNG  : ${png_support-no}
+  TIFF : ${tiff_support-no}
+  WIC  : ${wic_support-no}
+dwebp : yes
+  Output format support
+  =====================
+  PNG  : ${png_support-no}
+  WIC  : ${wic_support-no}
+GIF support : ${gif_support-no}
+gif2webp    : ${build_gif2webp-no}
+webpmux     : ${enable_libwebpmux-no}
+vwebp       : ${build_vwebp-no}
+])

doc/webp-container-spec.txt

@@ -13,9 +13,6 @@ end of this file.
 WebP Container Specification
 ============================
 
-_Working Draft, v0.5, 20120713_
-
-
 * TOC placeholder
 {:toc}
 
@@ -27,8 +24,9 @@ WebP is an image format that uses either (i) the VP8 key frame encoding
 to compress image data in a lossy way, or (ii) the WebP lossless encoding
 (and possibly other encodings in the future). These encoding schemes should
 make it more efficient than currently used formats. It is optimized for fast
-image transfer over the network (e.g., for websites). This document describes
-the structure of a WebP file.
+image transfer over the network (e.g., for websites). The WebP format has
+feature parity (color profile, metadata, animation etc) with other formats as
+well. This document describes the structure of a WebP file.
 
 The WebP container (i.e., RIFF container for WebP) allows feature support over
 and above the basic use case of WebP (i.e., a file containing a single image
@@ -38,24 +36,50 @@ for:
   * **Lossless compression.** An image can be losslessly compressed, using the
     WebP Lossless Format.
 
+  * **Metadata.** An image may have metadata stored in EXIF or XMP formats.
+
   * **Transparency.** An image may have transparency, i.e., an alpha channel.
 
+  * **Color Profile.** An image may have an embedded ICC profile as described
+    by the [International Color Consortium][iccspec].
+
+  * **Animation.** An image may have multiple frames with pauses between them,
+    making it an animation.
+
+  * **Image Fragmentation.** A single bitstream in WebP has an inherent
+    limitation for width or height of 2^14 pixels, and, when using VP8, a 512
+    KiB limit on the size of the first compressed partition. To support larger
+    images, the format supports images that are composed of multiple fragments,
+    each encoded as a separate bitstream. All fragments logically form a single
+    image: they have common metadata, color profile, etc. Image fragmentation
+    may also improve efficiency for larger images, e.g., grass can be encoded
+    differently than sky.
+
 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
 document are to be interpreted as described in [RFC 2119][].
 
+**Note:** Out of the features mentioned above, lossy compression, lossless
+compression, transparency, metadata, color profile and animation are finalized
+and are to be considered stable. On the other hand, image fragmentation is
+experimental as of now, and is open to discussion, feedback and comments.
+The same is indicated using annotation "_status: experimental_" in the relevant
+sections of this document.
 
 Terminology & Basics
 ------------------------
 
-A WebP file contains a still image (i.e., an encoded matrix of pixels) and,
-optionally, transparency information. In case we need to refer only to the
+A WebP file contains either a still image (i.e., an encoded matrix of pixels)
+or an [animation](#animation). Optionally, it can also contain transparency
+information, color profile and metadata. In case we need to refer only to the
 matrix of pixels, we will call it the _canvas_ of the image.
 
 Below are additional terms used throughout this document:
 
-Code that reads WebP files is referred to as a _reader_, while
-code that writes them is referred to as a _writer_.
+_Reader/Writer_
+
+: Code that reads WebP files is referred to as a _reader_, while code that
+writes them is referred to as a _writer_.
 
 _uint16_
 
@@ -69,10 +93,21 @@ _uint32_
 
 : A 32-bit, little-endian, unsigned integer.
 
+_FourCC_
+
+: A _FourCC_ (four-character code) is a _uint32_ created by concatenating four
+  ASCII characters in little-endian order.
+
 _1-based_
+
 : An unsigned integer field storing values offset by `-1`. e.g., Such a field
 would store value _25_ as _24_.
 
+RIFF file format
+----------------
+The WebP file format is based on the RIFF (resource interchange file format)
+document format.
+
 The basic element of a RIFF file is a _chunk_. It consists of:
 
      0                   1                   2                   3
@@ -87,57 +122,27 @@ The basic element of a RIFF file is a _chunk_. It consists of:
 
 Chunk FourCC: 32 bits
 
-: ASCII four character code or _chunk tag_ used for chunk identification.
+: ASCII four-character code used for chunk identification.
 
 Chunk Size: 32 bits (_uint32_)
 
-: The size of the chunk (_ckSize_) not including this field, the chunk
-  identifier and padding.
+: The size of the chunk not including this field, the chunk identifier or
+  padding.
 
 Chunk Payload: _Chunk Size_ bytes
 
-: The data payload. If _Chunk Size_ is odd a single padding byte that
-  SHOULD be `0` is added.
+: The data payload. If _Chunk Size_ is odd, a single padding byte -- that
+  SHOULD be `0` -- is added.
 
 _ChunkHeader('ABCD')_
 
-: This is used to describe the fourcc and size header of individual
-  chunks, where 'ABCD' is the fourcc for the chunk. This element's
+: This is used to describe the _FourCC_ and _Chunk Size_ header of individual
+  chunks, where 'ABCD' is the FourCC for the chunk. This element's
   size is 8 bytes.
 
-: Note that, in this specification, all chunk tag characters are in
-  file order, not in byte order of a uint32 of any particular
-  architecture.
-
-_list of chunks_
-
-: A concatenation of multiple chunks.
-
-: We will refer to the first chunk as having _position_ 0, the second
-  as position 1, etc. By _chunk with index 0 among "ABCD"_ we mean
-  the first chunk among the chunks of type "ABCD" in the list, the
-  _chunk with index 1 among "ABCD"_ is the second such chunk, etc.
-
-A WebP file MUST begin with a single chunk with a tag 'RIFF'. All
-other defined chunks are contained within this chunk. The file SHOULD
-NOT contain anything after it.
-
-The maximum size of RIFF's _ckSize_ is 2^32 minus 10 bytes. The size
-of the whole file is at most 4GiB minus 2 bytes.
-
-**Note:** some RIFF libraries are said to have bugs when handling files
-larger than 1GiB or 2GiB. If you are using an existing library, check
-that it handles large files correctly.
-
-The first four bytes of the RIFF chunk contents (i.e., bytes 8-11 of the file)
-MUST be the ASCII string "WEBP". They are followed by a list of chunks. As the
-size of any chunk is even, the size of the RIFF chunk is also even.  The
-contents of the chunks in that list will be described in the following sections.
-
-**Note:** RIFF has a convention that all-uppercase chunks are standard
-chunks that apply to any RIFF file format, while chunks specific to a
-file format are all lowercase. WebP does not follow this convention.
-
+**Note:** RIFF has a convention that all-uppercase chunk FourCCs are standard
+chunks that apply to any RIFF file format, while FourCCs specific to a file
+format are all lowercase. WebP does not follow this convention.
 
 WebP file header
 ----------------
@@ -158,12 +163,20 @@ WebP file header
 
 File Size: 32 bits (_uint32_)
 
-: The size of the file in bytes starting at offset 8.
+: The size of the file in bytes starting at offset 8. The maximum value of
+this field is 2^32 minus 10 bytes and thus the size of the whole file is at
+most 4GiB minus 2 bytes.
 
 'WEBP': 32 bits
 
 : The ASCII characters 'W' 'E' 'B' 'P'.
 
+A WebP file MUST begin with a RIFF header with the FourCC 'WEBP'. The file size
+in the header is the total size of the chunks that follow plus `4` bytes for
+the 'WEBP' FourCC. The file SHOULD NOT contain anything after it. As the size
+of any chunk is even, the size given by the RIFF header is also even. The
+contents of individual chunks will be described in the following sections.
+
 Simple file format (lossy)
 --------------------------
 
@@ -249,9 +262,25 @@ An extended format file consists of:
 
   * A 'VP8X' chunk with information about features used in the file.
 
-  * An optional 'ALPH' chunk with transparency information.
+  * An optional 'ICCP' chunk with color profile.
 
-  * The image bitstream contained in either a 'VP8 ' or 'VP8L' chunk.
+  * An optional 'ANIM' chunk with animation control data.
+
+  * Image data.
+
+  * An optional 'EXIF' chunk with EXIF metadata.
+
+  * An optional 'XMP ' chunk with XMP metadata.
+
+  * An optional list of [unknown chunks](#unknown-chunks). _\[status: experimental\]_
+
+For a _still image_, the _image data_ consists of a single frame, whereas for
+an _animated image_, it consists of multiple frames. More details about frames
+can be found in the [Animation](#animation) section.
+
+Moreover, each frame can be fragmented or non-fragmented, as will be described
+in the [Extended WebP file header](#extended_header) section. More details about
+fragments can be found in the [Fragments](#fragments) section.
 
 All chunks SHOULD be placed in the same order as listed above. If a chunk
 appears in the wrong place, the file is invalid, but readers MAY parse the
@@ -264,6 +293,7 @@ ignoring late chunks should make programs that need to do a full search
 give the same results as the ones stopping early.
 
 Extended WebP file header:
+{:#extended_header}
 
      0                   1                   2                   3
      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
@@ -272,25 +302,42 @@ Extended WebP file header:
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |                      ChunkHeader('VP8X')                      |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
-    | Rsv |L|  Rsv  |                   Reserved                    |
+    |Rsv|I|L|E|X|A|F|                   Reserved                    |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |          Canvas Width Minus One               |             ...
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     ...  Canvas Height Minus One    |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 
-Reserved (Rsv): 4 bits
+Reserved (Rsv): 2 bits
 
 : SHOULD be `0`.
 
+ICC profile (I): 1 bit
+
+: Set if the file contains an ICC profile.
+
 Alpha (L): 1 bit
 
-: Set if the file contains some (or all) images with transparency information
+: Set if any of the frames of the image contain transparency information
 ("alpha").
 
-Reserved (Rsv): 3 bits
+EXIF metadata (E): 1 bit
 
-: SHOULD be `0`.
+: Set if the file contains EXIF metadata.
+
+XMP metadata (X): 1 bit
+
+: Set if the file contains XMP metadata.
+
+Animation (A): 1 bit
+
+: Set if this is an animated image. Data in 'ANIM' and 'ANMF' chunks should be
+used to control the animation.
+
+Image Fragmentation (F): 1 bit _\[status: experimental\]_
+
+: Set if any of the frames in the image are represented by fragments.
 
 Reserved: 24 bits
 
@@ -312,6 +359,209 @@ Future specifications MAY add more fields.
 
 ### Chunks
 
+#### Animation
+
+An animation is controlled by ANIM and ANMF chunks.
+
+ANIM Chunk:
+{:#anim_chunk}
+
+For an animated image, this chunk contains the _global parameters_ of the
+animation.
+
+     0                   1                   2                   3
+     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                      ChunkHeader('ANIM')                      |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                       Background Color                        |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |          Loop Count           |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+Background Color: 32 bits (_uint32_)
+
+: The default background color of the canvas in \[Blue, Green, Red, Alpha\]
+byte order. This color MAY be used to fill the unused space on the canvas around
+the frames, as well as the transparent pixels of the first frame. Background
+color is also used when disposal method is `1`.
+
+**Note**:
+
+  * Background color MAY contain a transparency value (alpha), even if the
+    _Alpha_ flag in [VP8X chunk](#extended_header) is unset.
+
+  * Viewer applications SHOULD treat the background color value as a hint, and
+    are not required to use it.
+
+Loop Count: 16 bits (_uint16_)
+
+: The number of times to loop the animation. `0` means infinitely.
+
+This chunk MUST appear if the _Animation_ flag in the VP8X chunk is set.
+If the _Animation_ flag is not set and this chunk is present, it
+SHOULD be ignored.
+
+
+ANMF chunk:
+
+For animated images, this chunk contains information about a _single_ frame.
+If the _Animation flag_ is not set, then this chunk SHOULD NOT be present.
+
+     0                   1                   2                   3
+     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                      ChunkHeader('ANMF')                      |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                        Frame X                |             ...
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    ...          Frame Y            |   Frame Width Minus One     ...
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    ...             |           Frame Height Minus One              |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                 Frame Duration                |  Reserved |B|D|
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                         Frame Data                            |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+Frame X: 24 bits (_uint24_)
+
+: The X coordinate of the upper left corner of the frame is `Frame X * 2`
+
+Frame Y: 24 bits (_uint24_)
+
+: The Y coordinate of the upper left corner of the frame is `Frame Y * 2`
+
+Frame Width Minus One: 24 bits (_uint24_)
+
+: The _1-based_ width of the frame.
+  The frame width is `1 + Frame Width Minus One`
+
+Frame Height Minus One: 24 bits (_uint24_)
+
+: The _1-based_ height of the frame.
+  The frame height is `1 + Frame Height Minus One`
+
+Frame Duration: 24 bits (_uint24_)
+
+: The time to wait before displaying the next frame, in 1 millisecond units.
+In particular, frame duration of 0 is useful when one wants to update multiple
+areas of the canvas at once during the animation.
+
+Reserved: 6 bits
+
+: SHOULD be 0.
+
+Blending method (B): 1 bit
+
+: Indicates how transparent pixels of _the current frame_ are to be blended with
+corresponding pixels of the previous canvas:
+
+  * `0`: Use alpha blending. After disposing of the previous frame, render the
+    current frame on the canvas using [alpha-blending](#alpha-blending). If the
+    current frame does not have an alpha channel, assume alpha value of 255,
+    effectively replacing the rectangle.
+
+  * `1`: Do not blend. After disposing of the previous frame, render the
+    current frame on the canvas by overwriting the rectangle covered by the
+    current frame.
+
+Disposal method (D): 1 bit
+
+: Indicates how _the current frame_ is to be treated after it has been displayed
+(before rendering the next frame) on the canvas:
+
+  * `0`: Do not dispose. Leave the canvas as is.
+
+  * `1`: Dispose to background color. Fill the _rectangle_ on the canvas covered
+    by the _current frame_ with background color specified in the
+    [ANIM chunk](#anim_chunk).
+
+**Notes**:
+
+  * The frame disposal only applies to the _frame rectangle_, that is, the
+    rectangle defined by _Frame X_, _Frame Y_, _frame width_ and _frame height_.
+    It may or may not cover the whole canvas.
+
+{:#alpha-blending}
+  * **Alpha-blending**:
+
+    Given that each of the R, G, B and A channels is 8-bit, and the RGB
+    channels are _not premultiplied_ by alpha, the formula for blending
+    'dst' onto 'src' is:
+
+~~~~~
+    blend.A = src.A + dst.A * (1 - src.A / 255)
+    if blend.A = 0 then
+      blend.RGB = 0
+    else
+      blend.RGB = (src.RGB * src.A +
+                   dst.RGB * dst.A * (1 - src.A / 255)) / blend.A
+~~~~~
+
+  * Alpha-blending SHOULD be done in linear color space, by taking into account
+    the [color profile](#color-profile) of the image. If the color profile is
+    not present, sRGB is to be assumed. (Note that sRGB also needs to be
+    linearized due to a gamma of ~2.2).
+
+Frame Data: _Chunk Size_ - `16` bytes
+
+: For a fragmented frame, it consists of multiple [fragment chunks](#fragments).
+
+: For a non-fragmented frame, it consists of:
+
+  * An optional [alpha subchunk](#alpha) for the frame.
+
+  * A [bitstream subchunk](#bitstream-vp8vp8l) for the frame.
+
+  * An optional list of [unknown chunks](#unknown-chunks).
+
+**Note**: The 'ANMF' payload, _Frame Data_ above, consists of individual
+_padded_ chunks as described by the [RIFF file format](#riff-file-format).
+
+#### Fragments _\[status: experimental\]_
+
+For images that are represented by fragments, this chunk contains data for
+a single fragment. If the _Image Fragmentation Flag_ is not set, then this chunk
+SHOULD NOT be present.
+
+     0                   1                   2                   3
+     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                      ChunkHeader('FRGM')                      |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                  Fragment X                   |             ...
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    ...       Fragment Y            |         Fragment Data         |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+Fragment X: 24 bits (_uint24_)
+
+: The X coordinate of the upper left corner of the fragment is `Fragment X * 2`
+
+Fragment Y: 24 bits (_uint24_)
+
+: The Y coordinate of the upper left corner of the fragment is `Fragment Y * 2`
+
+Fragment Data: _Chunk Size_ - `6` bytes
+
+: It contains:
+
+  * An optional [alpha subchunk](#alpha) for the fragment.
+  * The [bitstream subchunk](#bitstream-vp8vp8l) for the fragment.
+  * An optional list of [unknown chunks](#unknown-chunks).
+
+Note: The width and height of the fragment is obtained from the bitstream
+subchunk.
+
+The fragments of a frame SHOULD have the following properties:
+
+  * They collectively cover the whole frame.
+
+  * No pair of fragments have any overlapping region on the frame.
+
+  * No portion of any fragment should be located outside of the canvas.
+
 #### Alpha
 
      0                   1                   2                   3
@@ -322,12 +572,18 @@ Future specifications MAY add more fields.
     |Rsv| P | F | C |     Alpha Bitstream...                        |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 
-Compression method (C): 2 bits
+Reserved (Rsv): 2 bits
 
-: The compression method used:
+: SHOULD be `0`.
 
-  * `0`: No compression.
-  * `1`: Compressed using the WebP lossless format.
+Pre-processing (P): 2 bits
+
+: These INFORMATIVE bits are used to signal the pre-processing that has
+been performed during compression. The decoder can use this information to
+e.g. dither the values or smooth the gradients prior to display.
+
+  * `0`: no pre-processing
+  * `1`: level reduction
 
 Filtering method (F): 2 bits
 
@@ -360,8 +616,8 @@ where `clip(v)` is equal to:
   * v    otherwise
 
 The final value is derived by adding the decompressed value `X` to the
-predictor and using modulo-256 arithmetic to wrap the [256-511] range
-into the [0-255] one:
+predictor and using modulo-256 arithmetic to wrap the \[256-511\] range
+into the \[0-255\] one:
 
 `alpha = (predictor + X) % 256`
 
@@ -374,30 +630,24 @@ There are special cases for left-most and top-most pixel positions:
     location (x, 0) are predicted using the location (x-1, 0) on the left.
 
 
-Pre-processing (P): 2 bits
-
-: These INFORMATIVE bits are used to signal the pre-processing that has
-been performed during compression. The decoder can use this information to
-e.g. dither the values or smooth the gradients prior to display.
-
-  * `0`: no pre-processing
-  * `1`: level reduction
-
 Decoders are not required to use this information in any specified way.
 
-Reserved (Rsv): 2 bits
+Compression method (C): 2 bits
 
-: SHOULD be `0`.
+: The compression method used:
+
+  * `0`: No compression.
+  * `1`: Compressed using the WebP lossless format.
 
 Alpha bitstream: _Chunk Size_ - `1` bytes
 
 : Encoded alpha bitstream.
 
-This optional chunk contains encoded alpha data for the image. An image
-containing a 'VP8L' chunk SHOULD NOT contain this chunk.
+This optional chunk contains encoded alpha data for this frame/fragment. A
+frame/fragment containing a 'VP8L' chunk SHOULD NOT contain this chunk.
 
-**Rationale**: The transparency information of the image is already part
-of the 'VP8L' chunk.
+**Rationale**: The transparency information is already part of the 'VP8L'
+chunk.
 
 The alpha channel data is stored as uncompressed raw data (when
 compression method is '0') or compressed using the lossless format
@@ -425,7 +675,7 @@ compression method is '0') or compressed using the lossless format
 
 #### Bitstream (VP8/VP8L)
 
-This chunk contains compressed image data.
+This chunk contains compressed bitstream data for a single frame/fragment.
 
 A bitstream chunk may be either (i) a VP8 chunk, using "VP8 " (note the
 significant fourth-character space) as its tag _or_ (ii) a VP8L chunk, using
@@ -435,10 +685,166 @@ The formats of VP8 and VP8L chunks are as described in sections
 [Simple file format (lossy)](#simple-file-format-lossy)
 and [Simple file format (lossless)](#simple-file-format-lossless) respectively.
 
-#### Unknown Chunks
+#### Color profile
 
-A file MAY contain other unknown chunks. Readers SHOULD ignore these chunks.
-Writers SHOULD preserve them in their original order.
+     0                   1                   2                   3
+     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                      ChunkHeader('ICCP')                      |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                       Color Profile                           |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+Color Profile: _Chunk Size_ bytes
+
+: ICC profile.
+
+This chunk MUST appear before the image data.
+
+There SHOULD be at most one such chunk. If there are more such chunks, readers
+MAY ignore all except the first one.
+See the [ICC Specification][iccspec] for details.
+
+If this chunk is not present, sRGB SHOULD be assumed.
+
+#### Metadata
+
+Metadata can be stored in 'EXIF' or 'XMP ' chunks.
+
+There SHOULD be at most one chunk of each type ('EXIF' and 'XMP '). If there
+are more such chunks, readers MAY ignore all except the first one. Also, a file
+may possibly contain both 'EXIF' and 'XMP ' chunks.
+
+The chunks are defined as follows:
+
+EXIF chunk:
+
+     0                   1                   2                   3
+     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                      ChunkHeader('EXIF')                      |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                        EXIF Metadata                          |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+EXIF Metadata: _Chunk Size_ bytes
+
+: image metadata in EXIF format.
+
+
+XMP chunk:
+
+     0                   1                   2                   3
+     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                      ChunkHeader('XMP ')                      |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    |                        XMP Metadata                           |
+    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+XMP Metadata: _Chunk Size_ bytes
+
+: image metadata in XMP format.
+
+Additional guidance about handling metadata can be found in the
+Metadata Working Group's [Guidelines for Handling Metadata][metadata].
+
+#### Unknown Chunks _\[status: experimental\]_
+
+A RIFF chunk (described in [this](#terminology-amp-basics) section) whose _chunk
+tag_ is different from any of the chunks described in this document, is
+considered an _unknown chunk_.
+
+**Rationale**: Allowing unknown chunks gives a provision for future extension
+of the format, and also allows storage of any application-specific data.
+
+A file MAY contain unknown chunks:
+
+  * At the end of the file as described in [Extended WebP file
+    header](#extended_header) section.
+  * At the end of FRGM and ANMF chunks as described in [Fragments](#fragments)
+    and [Animation](#animation) sections.
+
+Readers SHOULD ignore these chunks. Writers SHOULD preserve them in their
+original order (unless they specifically intend to modify these chunks).
+
+### Assembling the Canvas from fragments/frames
+
+Here we provide an overview of how a reader should assemble a canvas in case
+of a fragmented-image and in case of an animated image. The notation
+_VP8X.field_ means the field in the 'VP8X' chunk with the same description.
+
+Displaying a _fragmented image_ canvas MUST be equivalent to the following
+pseudocode: _\[status: experimental\]_
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+assert VP8X.flags.hasFragments
+canvas ← new black image of size VP8X.canvasWidth x VP8X.canvasHeight.
+frgm_params ← nil
+for chunk in image_data:
+    assert chunk.tag is "FRGM"
+    frgm_params.fragmentX = Fragment X
+    frgm_params.fragmentY = Fragment Y
+    for subchunk in 'Fragment Data':
+        if subchunk.tag == "ALPH":
+            assert alpha subchunks not found in 'Fragment Data' earlier
+            frgm_params.alpha = alpha_data
+        else if subchunk.tag == "VP8 " OR subchunk.tag == "VP8L":
+            assert bitstream subchunks not found in 'Fragment Data' earlier
+            frgm_params.bitstream = bitstream_data
+    frgm_params.fragmentWidth = Width extracted from bitstream subchunk
+    frgm_params.fragmentHeight = Height extracted from bitstream subchunk
+    assert VP8X.canvasWidth >=
+        frgm_params.fragmentX + frgm_params.fragmentWidth
+    assert VP8X.canvasHeight >=
+        frgm_params.fragmentY + frgm_params.fragmentHeight
+    assert fragment has the properties mentioned in "Image Fragments" section.
+    render fragment with frame_params.alpha and frame_params.bitstream on canvas
+    with top-left corner in (frgm_params.fragmentX, frgm_params.fragmentY).
+canvas contains the decoded canvas.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Displaying an _animated image_ canvas MUST be equivalent to the following
+pseudocode:
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+assert VP8X.flags.hasAnimation
+canvas ← new image of size VP8X.canvasWidth x VP8X.canvasHeight with
+background color ANIM.background_color.
+loop_count ← ANIM.loopCount
+dispose_method ← ANIM.disposeMethod
+if loop_count == 0:
+    loop_count = ∞
+frame_params ← nil
+for loop = 0, ..., loop_count - 1
+    assert next chunk in image_data is ANMF
+    frame_params.frameX = Frame X
+    frame_params.frameY = Frame Y
+    frame_params.frameWidth = Frame Width Minus One + 1
+    frame_params.frameHeight = Frame Height Minus One + 1
+    frame_params.frameDuration = Frame Duration
+    assert VP8X.canvasWidth >= frame_params.frameX + frame_params.frameWidth
+    assert VP8X.canvasHeight >= frame_params.frameY + frame_params.frameHeight
+    if VP8X.flags.hasFragments and first subchunk in 'Frame Data' is FRGM
+        // Fragmented frame.
+        frame_params.{bitstream,alpha} = canvas decoded from subchunks in
+                                         'Frame Data' as per the pseudocode for
+                                         _fragmented image_ above.
+    else
+        // Non-fragmented frame.
+        for subchunk in 'Frame Data':
+            if subchunk.tag == "ALPH":
+                assert alpha subchunks not found in 'Frame Data' earlier
+                frame_params.alpha = alpha_data
+            else if subchunk.tag == "VP8 " OR subchunk.tag == "VP8L":
+                assert bitstream subchunks not found in 'Frame Data' earlier
+                frame_params.bitstream = bitstream_data
+    render frame with frame_params.alpha and frame_params.bitstream on canvas
+    with top-left corner in (frame_params.frameX, frame_params.frameY), using
+    dispose method dispose_method.
+    Show the contents of the image for frame_params.frameDuration * 1ms.
+canvas contains the decoded canvas.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 Example file layouts
 --------------------
@@ -461,7 +867,43 @@ RIFF/WEBP
 +- VP8L (lossless bitstream)
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+A lossless image with ICC profile and XMP metadata may
+look as follows:
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+RIFF/WEBP
++- VP8X (descriptions of features used)
++- ICCP (color profile)
++- VP8L (lossless bitstream)
++- XMP  (metadata)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+A fragmented image may look as follows:
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+RIFF/WEBP
++- VP8X (descriptions of features used)
++- FRGM (fragment1 parameters + data)
++- FRGM (fragment2 parameters + data)
++- FRGM (fragment3 parameters + data)
++- FRGM (fragment4 parameters + data)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+An animated image with EXIF metadata may look as follows:
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+RIFF/WEBP
++- VP8X (descriptions of features used)
++- ANIM (global animation parameters)
++- ANMF (frame1 parameters + data)
++- ANMF (frame2 parameters + data)
++- ANMF (frame3 parameters + data)
++- ANMF (frame4 parameters + data)
++- EXIF (metadata)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
 [vp8spec]:  http://tools.ietf.org/html/rfc6386
 [webpllspec]: https://gerrit.chromium.org/gerrit/gitweb?p=webm/libwebp.git;a=blob;f=doc/webp-lossless-bitstream-spec.txt;hb=master
+[iccspec]: http://www.color.org/icc_specs2.xalter
 [metadata]: http://www.metadataworkinggroup.org/pdf/mwg_guidance.pdf
 [rfc 2119]: http://tools.ietf.org/html/rfc2119

doc/webp-lossless-bitstream-spec.txt

@@ -14,6 +14,7 @@ Specification for WebP Lossless Bitstream
 
 _Jyrki Alakuijala, Ph.D., Google, Inc., 2012-06-19_
 
+Paragraphs marked as \[AMENDED\] were amended on 2014-09-16.
 
 Abstract
 --------
@@ -172,8 +173,8 @@ It should be set to 0 when all alpha values are 255 in the picture, and
 int alpha_is_used = ReadBits(1);
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-The version_number is a 3 bit code that must be discarded by the decoder
-at this time. Complying encoders write a 3-bit value 0.
+The version_number is a 3 bit code that must be set to 0. Any other value
+should be treated as an error. \[AMENDED\]
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 int version_number = ReadBits(3);
@@ -236,7 +237,7 @@ predicted) is encoded. The _prediction mode_ determines the type of
 prediction to use. We divide the image into squares and all the pixels
 in a square use same prediction mode.
 
-The first 4 bits of prediction data define the block width and height in
+The first 3 bits of prediction data define the block width and height in
 number of bits. The number of block columns, `block_xsize`, is used in
 indexing two-dimensionally.
 
@@ -330,7 +331,7 @@ uint32 Select(uint32 L, uint32 T, uint32 TL) {
            abs(pGreen - GREEN(T)) + abs(pBlue - BLUE(T));
 
   // Return either left or top, the one closer to the prediction.
-  if (pL <= pT) {
+  if (pL < pT) {     // \[AMENDED\]
     return L;
   } else {
     return T;
@@ -361,14 +362,14 @@ int ClampAddSubtractHalf(int a, int b) {
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 There are special handling rules for some border pixels. If there is a
-prediction transform, regardless of the mode [0..13] for these pixels,
+prediction transform, regardless of the mode \[0..13\] for these pixels,
 the predicted value for the left-topmost pixel of the image is
 0xff000000, L-pixel for all pixels on the top row, and T-pixel for all
 pixels on the leftmost column.
 
 Addressing the TR-pixel for pixels on the rightmost column is
 exceptional. The pixels on the rightmost column are predicted by using
-the modes [0..13] just like pixels not on border, but by using the
+the modes \[0..13\] just like pixels not on border, but by using the
 leftmost pixel on the same row as the current TR-pixel. The TR-pixel
 offset in memory is the same for border and non-border pixels.
 
@@ -420,7 +421,7 @@ void ColorTransform(uint8 red, uint8 blue, uint8 green,
 
 `ColorTransformDelta` is computed using a signed 8-bit integer
 representing a 3.5-fixed-point number, and a signed 8-bit RGB color
-channel (c) [-128..127] and is defined as follows:
+channel (c) \[-128..127\] and is defined as follows:
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 int8 ColorTransformDelta(int8 t, int8 c) {
@@ -436,12 +437,12 @@ consistent with each other.
 
 Now we describe the contents of color transform data so that decoding
 can apply the inverse color transform and recover the original red and
-blue values. The first 4 bits of the color transform data contain the
+blue values. The first 3 bits of the color transform data contain the
 width and height of the image block in number of bits, just like the
 predictor transform:
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-int size_bits = ReadStream(3) + 2;
+int size_bits = ReadBits(3) + 2;
 int block_width = 1 << size_bits;
 int block_height = 1 << size_bits;
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -518,7 +519,7 @@ follows:
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 // 8 bit value for color table size
-int color_table_size = ReadStream(8) + 1;
+int color_table_size = ReadBits(8) + 1;
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 The color table is stored using the image storage format itself. The
@@ -542,6 +543,9 @@ color.
 argb = color_table[GREEN(argb)];
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+If the index is equal or larger than color_table_size, the argb color value
+should be set to 0x00000000 (transparent black).  \[AMENDED\]
+
 When the color table is small (equal to or less than 16 colors), several
 pixels are bundled into a single pixel. The pixel bundling packs several
 (2, 4, or 8) pixels into a single pixel, reducing the image width
@@ -567,10 +571,10 @@ if (color_table_size <= 2) {
 
 `width_bits` has a value of 0, 1, 2 or 3. A value of 0 indicates no
 pixel bundling to be done for the image. A value of 1 indicates that two
-pixels are combined together, and each pixel has a range of [0..15]. A
+pixels are combined together, and each pixel has a range of \[0..15\]. A
 value of 2 indicates that four pixels are combined together, and each
-pixel has a range of [0..3]. A value of 3 indicates that eight pixels
-are combined together and each pixel has a range of [0..1], i.e., a
+pixel has a range of \[0..3\]. A value of 3 indicates that eight pixels
+are combined together and each pixel has a range of \[0..1\], i.e., a
 binary value.
 
 The values are packed into the green component as follows:
@@ -592,80 +596,107 @@ The values are packed into the green component as follows:
 4 Image Data
 ------------
 
-Image data is an array of pixel values in scan-line order. We use image
-data in five different roles: The main role, an auxiliary role related
-to entropy coding, and three further roles related to transforms.
+Image data is an array of pixel values in scan-line order.
 
-  1. ARGB image.
-  2. Entropy image. The red and green components define the meta Huffman
-     code used in a particular area of the image.
-  3. Predictor image. The green component defines which of the 14 values
-     is used within a particular square of the image.
-  4. Color indexing image. An array of up to 256 ARGB colors is used for
-     transforming a green-only image, using the green value as an index
-     to this one-dimensional array.
-  5. Color transformation image. Defines signed 3.5 fixed-point
-     multipliers that are used to predict the red, green, and blue
-     components, to reduce entropy.
+### 4.1 Roles of Image Data
 
-To divide the image into multiple regions, the image is first divided
-into a set of fixed-size blocks (typically 16x16 blocks). Each of these
-blocks can be modeled using an entropy code, in a way where several
-blocks can share the same entropy code. There is a cost in transmitting
-an entropy code, and in order to minimize this cost, statistically
-similar blocks can share an entropy code. The blocks sharing an entropy
-code can be found by clustering their statistical properties, or by
-repeatedly joining two randomly selected clusters when it reduces the
-overall amount of bits needed to encode the image. See the section
-[Decoding of Meta Huffman Codes](#decoding-of-meta-huffman-codes) in
-[Chapter 5](#entropy-code) for an explanation of how this entropy image
-is stored.
+We use image data in five different roles:
 
-Each pixel is encoded using one of three possible methods:
+  1. ARGB image: Stores the actual pixels of the image.
+  1. Entropy image: Stores the
+     [meta Huffman codes](#decoding-of-meta-huffman-codes). The red and green
+     components of a pixel define the meta Huffman code used in a particular
+     block of the ARGB image.
+  1. Predictor image: Stores the metadata for [Predictor
+     Transform](#predictor-transform). The green component of a pixel defines
+     which of the 14 predictors is used within a particular block of the
+     ARGB image.
+  1. Color transform image. It is created by `ColorTransformElement` values
+     (defined in [Color Transform](#color-transform)) for different blocks of
+     the image. Each `ColorTransformElement` `'cte'` is treated as a pixel whose
+     alpha component is `255`, red component is `cte.red_to_blue`, green
+     component is `cte.green_to_blue` and blue component is `cte.green_to_red`.
+  1. Color indexing image: An array of of size `color_table_size` (up to 256
+     ARGB values) storing the metadata for the
+     [Color Indexing Transform](#color-indexing-transform). This is stored as an
+     image of width `color_table_size` and height `1`.
 
-  1. Huffman coded literals, where each channel (green, alpha, red,
-     blue) is entropy-coded independently;
-  2. LZ77, a sequence of pixels in scan-line order copied from elsewhere
+### 4.2 Encoding of Image data
+
+The encoding of image data is independent of its role.
+
+The image is first divided into a set of fixed-size blocks (typically 16x16
+blocks). Each of these blocks are modeled using their own entropy codes. Also,
+several blocks may share the same entropy codes.
+
+**Rationale:** Storing an entropy code incurs a cost. This cost can be minimized
+if statistically similar blocks share an entropy code, thereby storing that code
+only once. For example, an encoder can find similar blocks by clustering them
+using their statistical properties, or by repeatedly joining a pair of randomly
+selected clusters when it reduces the overall amount of bits needed to encode
+the image.
+
+Each pixel is encoded using one of the three possible methods:
+
+  1. Huffman coded literal: each channel (green, red, blue and alpha) is
+     entropy-coded independently;
+  2. LZ77 backward reference: a sequence of pixels are copied from elsewhere
      in the image; or
-  3. Color cache, using a short multiplicative hash code (color cache
+  3. Color cache code: using a short multiplicative hash code (color cache
      index) of a recently seen color.
 
-In the following sections we introduce the main concepts in LZ77 prefix
-coding, LZ77 entropy coding, LZ77 distance mapping, and color cache
-codes. The actual details of the entropy code are described in more
-detail in [Chapter 5](#entropy-code).
+The following sub-sections describe each of these in detail.
 
+#### 4.2.1 Huffman Coded Literals
 
-### LZ77 Prefix Coding
+The pixel is stored as Huffman coded values of green, red, blue and alpha (in
+that order). See [this section](#decoding-entropy-coded-image-data) for details.
 
-Prefix coding divides large integer values into two parts: the prefix
-code and the extra bits. The benefit of this approach is that entropy
-coding is later used only for the prefix code, reducing the resources
-needed by the entropy code. The extra bits are stored as they are,
-without an entropy code.
+#### 4.2.2 LZ77 Backward Reference
 
-This prefix code is used for coding backward reference lengths and
-distances. The extra bits form an integer that is added to the lower
-value of the range. Hence the LZ77 lengths and distances are divided
-into prefix codes and extra bits. Performing the Huffman coding only on
-the prefixes reduces the size of the Huffman codes to tens of values
-instead of a million (distance) or several thousands (length).
+Backward references are tuples of _length_ and _distance code_:
 
-| Prefix code | Value range     | Extra bits |
-| ----------- | --------------- | ---------- |
-| 0           | 1               | 0          |
-| 1           | 2               | 0          |
-| 2           | 3               | 0          |
-| 3           | 4               | 0          |
-| 4           | 5..6            | 1          |
-| 5           | 7..8            | 1          |
-| 6           | 9..12           | 2          |
-| 7           | 13..16          | 2          |
-| ...         | ...             | ...        |
-| 38          | 262145..524288  | 18         |
-| 39          | 524289..1048576 | 18         |
+  * Length indicates how many pixels in scan-line order are to be copied.
+  * Distance code is a number indicating the position of a previously seen
+    pixel, from which the pixels are to be copied. The exact mapping is
+    described [below](#distance-mapping).
 
-The code to obtain a value from the prefix code is as follows:
+The length and distance values are stored using **LZ77 prefix coding**.
+
+LZ77 prefix coding divides large integer values into two parts: the _prefix
+code_ and the _extra bits_: the prefix code is stored using an entropy code,
+while the extra bits are stored as they are (without an entropy code).
+
+**Rationale**: This approach reduces the storage requirement for the entropy
+code. Also, large values are usually rare, and so extra bits would be used for
+very few values in the image. Thus, this approach results in a better
+compression overall.
+
+The following table denotes the prefix codes and extra bits used for storing
+different range of values.
+
+Note: The maximum backward reference length is limited to 4096. Hence, only the
+first 24 prefix codes (with the respective extra bits) are meaningful for length
+values. For distance values, however, all the 40 prefix codes are valid.
+
+| Value range     | Prefix code | Extra bits |
+| --------------- | ----------- | ---------- |
+| 1               | 0           | 0          |
+| 2               | 1           | 0          |
+| 3               | 2           | 0          |
+| 4               | 3           | 0          |
+| 5..6            | 4           | 1          |
+| 7..8            | 5           | 1          |
+| 9..12           | 6           | 2          |
+| 13..16          | 7           | 2          |
+| ...             | ...         | ...        |
+| 3072..4096      | 23          | 10         |
+| ...             | ...         | ...        |
+| 524289..786432  | 38          | 18         |
+| 786433..1048576 | 39          | 18         |
+
+The pseudocode to obtain a (length or distance) value from the prefix code is
+as follows:
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 if (prefix_code < 4) {
@@ -676,26 +707,28 @@ int offset = (2 + (prefix_code & 1)) << extra_bits;
 return offset + ReadBits(extra_bits) + 1;
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+**Distance Mapping:**
+{:#distance-mapping}
 
-### LZ77 Backward Reference Entropy Coding
+As noted previously, distance code is a number indicating the position of a
+previously seen pixel, from which the pixels are to be copied. This sub-section
+defines the mapping between a distance code and the position of a previous
+pixel.
 
-Backward references are tuples of length and distance. Length indicates
-how many pixels in scan-line order are to be copied. The length is
-codified in two steps: prefix and extra bits. Only the first 24 prefix
-codes with their respective extra bits are used for length codes,
-limiting the maximum length to 4096. For distances, all 40 prefix codes
-are used.
+The distance codes larger than 120 denote the pixel-distance in scan-line
+order, offset by 120.
 
+The smallest distance codes \[1..120\] are special, and are reserved for a close
+neighborhood of the current pixel. This neighborhood consists of 120 pixels:
 
-### LZ77 Distance Mapping
+  * Pixels that are 1 to 7 rows above the current pixel, and are up to 8 columns
+    to the left or up to 7 columns to the right of the current pixel. \[Total
+    such pixels = `7 * (8 + 1 + 7) = 112`\].
+  * Pixels that are in same row as the current pixel, and are up to 8 columns to
+    the left of the current pixel. \[`8` such pixels\].
 
-120 smallest distance codes [1..120] are reserved for a close
-neighborhood within the current pixel. The rest are pure distance codes
-in scan-line order, just offset by 120. The smallest codes are coded
-into x and y offsets by the following table. Each tuple shows the x and
-the y coordinates in 2D offsets -- for example the first tuple (0, 1)
-means 0 for no difference in x, and 1 pixel difference in y (indicating
-previous row).
+The mapping between distance code `i` and the neighboring pixel offset
+`(xi, yi)` is as follows:
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 (0, 1),  (1, 0),  (1, 1),  (-1, 1), (0, 2),  (2, 0),  (1, 2),  (-1, 2),
@@ -715,38 +748,51 @@ previous row).
 (-6, 7), (7, 6),  (-7, 6), (8, 5),  (7, 7),  (-7, 7), (8, 6),  (8, 7)
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-The distances codes that map into these tuples are changes into
-scan-line order distances using the following formula:
-_dist = x + y * xsize_, where _xsize_ is the width of the image in
-pixels. If a decoder detects a computed _dist_ value smaller than 1,
-the value of 1 is used instead.
+For example, distance code `1` indicates offset of `(0, 1)` for the neighboring
+pixel, that is, the pixel above the current pixel (0-pixel difference in
+X-direction and 1 pixel difference in Y-direction). Similarly, distance code
+`3` indicates left-top pixel.
+
+The decoder can convert a distances code 'i' to a scan-line order distance
+'dist' as follows:
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+(xi, yi) = distance_map[i]
+dist = x + y * xsize
+if (dist < 1) {
+  dist = 1
+}
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+where 'distance_map' is the mapping noted above and `xsize` is the width of the
+image in pixels.
 
 
-### Color Cache Code
+#### 4.2.3 Color Cache Coding
 
-Color cache stores a set of colors that have been recently used in the
-image. Using the color cache code, the color cache colors can be
-referred to more efficiently than emitting the respective ARGB values
-independently or sending them as backward references with a length of
-one pixel.
+Color cache stores a set of colors that have been recently used in the image.
 
-Color cache codes are coded as follows. First, there is a bit that
-indicates if the color cache is used or not. If this bit is 0, no color
-cache codes exist, and they are not transmitted in the Huffman code that
-decodes the green symbols and the length prefix codes. However, if this
-bit is 1, the color cache size is read:
+**Rationale:** This way, the recently used colors can sometimes be referred to
+more efficiently than emitting them using other two methods (described in
+[4.2.1](#huffman-coded-literals) and [4.2.2](#lz77-backward-reference)).
+
+Color cache codes are stored as follows. First, there is a 1-bit value that
+indicates if the color cache is used. If this bit is 0, no color cache codes
+exist, and they are not transmitted in the Huffman code that decodes the green
+symbols and the length prefix codes. However, if this bit is 1, the color cache
+size is read next:
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-int color_cache_code_bits = ReadBits(br, 4);
+int color_cache_code_bits = ReadBits(4);
 int color_cache_size = 1 << color_cache_code_bits;
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 `color_cache_code_bits` defines the size of the color_cache by (1 <<
 `color_cache_code_bits`). The range of allowed values for
-`color_cache_code_bits` is [1..11]. Compliant decoders must indicate a
+`color_cache_code_bits` is \[1..11\]. Compliant decoders must indicate a
 corrupted bitstream for other values.
 
-A color cache is an array of the size `color_cache_size`. Each entry
+A color cache is an array of size `color_cache_size`. Each entry
 stores one ARGB color. Colors are looked up by indexing them by
 (0x1e35a7bd * `color`) >> (32 - `color_cache_code_bits`). Only one
 lookup is done in a color cache; there is no conflict resolution.
@@ -761,91 +807,188 @@ literals, into the cache in the order they appear in the stream.
 5 Entropy Code
 --------------
 
-### Huffman Coding
+### 5.1 Overview
 
-Most of the data is coded using a canonical Huffman code. This includes
-the following:
+Most of the data is coded using [canonical Huffman code][canonical_huff]. Hence,
+the codes are transmitted by sending the _Huffman code lengths_, as opposed to
+the actual _Huffman codes_.
 
-  * a combined code that defines either the value of the green
-    component, a color cache code, or a prefix of the length codes;
-  * the data for alpha, red and blue components; and
-  * prefixes of the distance codes.
+In particular, the format uses **spatially-variant Huffman coding**. In other
+words, different blocks of the image can potentially use different entropy
+codes.
 
-The Huffman codes are transmitted by sending the code lengths; the
-actual symbols are implicit and done in order for each length. The
-Huffman code lengths are run-length-encoded using three different
-prefixes, and the result of this coding is further Huffman coded.
+**Rationale**: Different areas of the image may have different characteristics. So, allowing them to use different entropy codes provides more flexibility and
+potentially a better compression.
 
+### 5.2 Details
 
-### Spatially-variant Huffman Coding
+The encoded image data consists of two parts:
 
-For every pixel (x, y) in the image, there is a definition of which
-entropy code to use. First, there is an integer called 'meta Huffman
-code' that can be obtained from the entropy image. This
-meta Huffman code identifies a set of five Huffman codes, one for green
-(along with length codes and color cache codes), one for each of red,
-blue and alpha, and one for distance. The Huffman codes are identified
-by their position in a table by an integer.
+  1. Meta Huffman codes
+  1. Entropy-coded image data
 
+#### 5.2.1 Decoding of Meta Huffman Codes
 
-### Decoding Flow of Image Data
+As noted earlier, the format allows the use of different Huffman codes for
+different blocks of the image. _Meta Huffman codes_ are indexes identifying
+which Huffman codes to use in different parts of the image.
 
-Read next symbol S
+Meta Huffman codes may be used _only_ when the image is being used in the
+[role](#roles-of-image-data) of an _ARGB image_.
 
-  1. S < 256
-     1. Use S as green component
-     2. read alpha
-     3. read red
-     4. read blue
-  2. S < 256 + 24
+There are two possibilities for the meta Huffman codes, indicated by a 1-bit
+value:
+
+  * If this bit is zero, there is only one meta Huffman code used everywhere in
+    the image. No more data is stored.
+  * If this bit is one, the image uses multiple meta Huffman codes. These meta
+    Huffman codes are stored as an _entropy image_ (described below).
+
+**Entropy image:**
+
+The entropy image defines which Huffman codes are used in different parts of the
+image, as described below.
+
+The first 3-bits contain the `huffman_bits` value. The dimensions of the entropy
+image are derived from 'huffman_bits'.
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+int huffman_bits = ReadBits(3) + 2;
+int huffman_xsize = DIV_ROUND_UP(xsize, 1 << huffman_bits);
+int huffman_ysize = DIV_ROUND_UP(ysize, 1 << huffman_bits);
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+where `DIV_ROUND_UP` is as defined [earlier](#predictor-transform).
+
+Next bits contain an entropy image of width `huffman_xsize` and height
+`huffman_ysize`.
+
+**Interpretation of Meta Huffman Codes:**
+
+For any given pixel (x, y), there is a set of five Huffman codes associated with
+it. These codes are (in bitstream order):
+
+  * **Huffman code #1**: used for green channel, backward-reference length and
+    color cache
+  * **Huffman code #2, #3 and #4**: used for red, blue and alpha channels
+    respectively.
+  * **Huffman code #5**: used for backward-reference distance.
+
+From here on, we refer to this set as a **Huffman code group**.
+
+The number of Huffman code groups in the ARGB image can be obtained by finding
+the _largest meta Huffman code_ from the entropy image:
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+int num_huff_groups = max(entropy image) + 1;
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+where `max(entropy image)` indicates the largest Huffman code stored in the
+entropy image.
+
+As each Huffman code groups contains five Huffman codes, the total number of
+Huffman codes is:
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+int num_huff_codes = 5 * num_huff_groups;
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Given a pixel (x, y) in the ARGB image, we can obtain the corresponding Huffman
+codes to be used as follows:
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+int position = (y >> huffman_bits) * huffman_xsize + (x >> huffman_bits);
+int meta_huff_code = (entropy_image[pos] >> 8) & 0xffff;
+HuffmanCodeGroup huff_group = huffman_code_groups[meta_huff_code];
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+where, we have assumed the existence of `HuffmanCodeGroup` structure, which
+represents a set of five Huffman codes. Also, `huffman_code_groups` is an array
+of `HuffmanCodeGroup` (of size `num_huff_groups`).
+
+The decoder then uses Huffman code group `huff_group` to decode the pixel
+(x, y) as explained in the [next section](#decoding-entropy-coded-image-data).
+
+#### 5.2.2 Decoding Entropy-coded Image Data
+
+For the current position (x, y) in the image, the decoder first identifies the
+corresponding Huffman code group (as explained in the last section). Given the
+Huffman code group, the pixel is read and decoded as follows:
+
+Read next symbol S from the bitstream using Huffman code #1. \[See
+[next section](#decoding-the-code-lengths) for details on decoding the Huffman
+code lengths\]. Note that S is any integer in the range `0` to
+`(256 + 24 + ` [`color_cache_size`](#color-cache-code)`- 1)`.
+
+The interpretation of S depends on its value:
+
+  1. if S < 256
+     1. Use S as the green component
+     1. Read red from the bitstream using Huffman code #2
+     1. Read blue from the bitstream using Huffman code #3
+     1. Read alpha from the bitstream using Huffman code #4
+  1. if S < 256 + 24
      1. Use S - 256 as a length prefix code
-     2. read length extra bits
-     3. read distance prefix code
-     4. read distance extra bits
-  3. S >= 256 + 24
-     1. Use ARGB color from the color cache, at index S - 256 + 24
+     1. Read extra bits for length from the bitstream
+     1. Determine backward-reference length L from length prefix code and the
+        extra bits read.
+     1. Read distance prefix code from the bitstream using Huffman code #5
+     1. Read extra bits for distance from the bitstream
+     1. Determine backward-reference distance D from distance prefix code and
+        the extra bits read.
+     1. Copy the L pixels (in scan-line order) from the sequence of pixels
+        prior to them by D pixels.
+  1. if S >= 256 + 24
+     1. Use S - (256 + 24) as the index into the color cache.
+     1. Get ARGB color from the color cache at that index.
 
 
-### Decoding the Code Lengths
+**Decoding the Code Lengths:**
+{:#decoding-the-code-lengths}
 
-There are two different ways to encode the code lengths of a Huffman
-code, indicated by the first bit of the code: _simple code length code_
-(1), and _normal code length code_ (0).
+This section describes the details about reading a symbol from the bitstream by
+decoding the Huffman code length.
 
+The Huffman code lengths can be coded in two ways. The method used is specified
+by a 1-bit value.
 
-#### Simple Code Length Code
+  * If this bit is 1, it is a _simple code length code_, and
+  * If this bit is 0, it is a _normal code length code_.
 
-This variant can codify 1 or 2 non-zero length codes in the range of [0,
-255]. All other code lengths are implicitly zeros.
+**(i) Simple Code Length Code:**
 
-The first bit indicates the number of codes:
+This variant is used in the special case when only 1 or 2 Huffman code lengths
+are non-zero, and are in the range of \[0, 255\]. All other Huffman code lengths
+are implicitly zeros.
+
+The first bit indicates the number of non-zero code lengths:
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-int num_symbols = ReadBits(1) + 1;
+int num_code_lengths = ReadBits(1) + 1;
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-The first symbol is stored either using a 1-bit code for values of 0 and
-1, or using a 8-bit code for values in range [0, 255]. The second
-symbol, when present, is coded as an 8-bit code.
+The first code length is stored either using a 1-bit code for values of 0 and 1,
+or using an 8-bit code for values in range \[0, 255\]. The second code length,
+when present, is coded as an 8-bit code.
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-int first_symbol_len_code = VP8LReadBits(br, 1);
-symbols[0] = ReadBits(1 + 7 * first_symbol_len_code);
-if (num_symbols == 2) {
-  symbols[1] = ReadBits(8);
+int is_first_8bits = ReadBits(1);
+code_lengths[0] = ReadBits(1 + 7 * is_first_8bits);
+if (num_code_lengths == 2) {
+  code_lengths[1] = ReadBits(8);
 }
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-Empty trees can be coded as trees that contain one 0 symbol, and can be
-codified using four bits. For example, a distance tree can be empty if
-there are no backward references. Similarly, alpha, red, and blue trees
-can be empty if all pixels within the same meta Huffman code are
-produced using the color cache.
+**Note:** Another special case is when _all_ Huffman code lengths are _zeros_
+(an empty Huffman code). For example, a Huffman code for distance can be empty
+if there are no backward references. Similarly, Huffman codes for alpha, red,
+and blue can be empty if all pixels within the same meta Huffman code are
+produced using the color cache. However, this case doesn't need a special
+handling, as empty Huffman codes can be coded as those containing a single
+symbol `0`.
 
+**(ii) Normal Code Length Code:**
 
-#### Normal Code Length Code
-
-The code lengths of a Huffman code are read as follows: `num_codes`
+The code lengths of a Huffman code are read as follows: `num_code_lengths`
 specifies the number of code lengths; the rest of the code lengths
 (according to the order in `kCodeLengthCodeOrder`) are zeros.
 
@@ -854,91 +997,23 @@ int kCodeLengthCodes = 19;
 int kCodeLengthCodeOrder[kCodeLengthCodes] = {
   17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
 };
-int num_codes = 4 + ReadStream(4);
-for (i = 0; i < num_codes; ++i) {
+int code_lengths[kCodeLengthCodes] = { 0 };  // All zeros.
+int num_code_lengths = 4 + ReadBits(4);
+for (i = 0; i < num_code_lengths; ++i) {
   code_lengths[kCodeLengthCodeOrder[i]] = ReadBits(3);
 }
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-  * Code length code [0..15] indicates literal code lengths.
+  * Code length code \[0..15\] indicates literal code lengths.
     * Value 0 means no symbols have been coded.
-    * Values [1..15] indicate the bit length of the respective code.
-  * Code 16 repeats the previous non-zero value [3..6] times, i.e.,
-    3 + `ReadStream(2)` times.  If code 16 is used before a non-zero
+    * Values \[1..15\] indicate the bit length of the respective code.
+  * Code 16 repeats the previous non-zero value \[3..6\] times, i.e.,
+    3 + `ReadBits(2)` times.  If code 16 is used before a non-zero
     value has been emitted, a value of 8 is repeated.
-  * Code 17 emits a streak of zeros [3..10], i.e., 3 + `ReadStream(3)`
+  * Code 17 emits a streak of zeros \[3..10\], i.e., 3 + `ReadBits(3)`
     times.
-  * Code 18 emits a streak of zeros of length [11..138], i.e.,
-    11 + `ReadStream(7)` times.
-
-The entropy codes for alpha, red and blue have a total of 256 symbols.
-The entropy code for distance prefix codes has 40 symbols. The entropy
-code for green has 256 + 24 + `color_cache_size`, 256 symbols for
-different green symbols, 24 length code prefix symbols, and symbols for
-the color cache.
-
-The meta Huffman code, specified in the next section, defines how many
-Huffman codes there are. There are always 5 times the number of Huffman
-codes to the number of meta Huffman codes.
-
-
-### Decoding of Meta Huffman Codes
-
-There are two ways to code the meta Huffman codes, indicated by one bit
-for the ARGB image and is an implicit zero, i.e., not present in the
-stream for all transform images and the entropy image itself.
-
-If this bit is zero, there is only one meta Huffman code, using Huffman
-codes 0, 1, 2, 3 and 4 for green, alpha, red, blue and distance,
-respectively. This meta Huffman code is used everywhere in the image.
-
-If this bit is one, the meta Huffman codes are controlled by the entropy
-image, where the index of the meta Huffman code is codified in the red
-and green components. The index can be obtained from the uint32 value by
-_((pixel >> 8) & 0xffff)_, thus there can be up to 65536 unique meta
-Huffman codes. When decoding a Huffman encoded symbol at a pixel x, y,
-one chooses the meta Huffman code respective to these coordinates.
-However, not all bits of the coordinates are used for choosing the meta
-Huffman code, i.e., the entropy image is of subresolution to the real
-image.
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-int huffman_bits = ReadBits(3) + 2;
-int huffman_xsize = DIV_ROUND_UP(xsize, 1 << huffman_bits);
-int huffman_ysize = DIV_ROUND_UP(ysize, 1 << huffman_bits);
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-`huffman_bits` gives the amount of subsampling in the entropy image.
-
-After reading the `huffman_bits`, an entropy image stream of size
-`huffman_xsize`, `huffman_ysize` is read.
-
-The meta Huffman code, identifying the five Huffman codes per meta
-Huffman code, is coded only by the number of codes:
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-int num_meta_codes = max(entropy_image) + 1;
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-Now, we can obtain the five Huffman codes for green, alpha, red, blue
-and distance for a given (x, y) by the following expression:
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-meta_codes[(entropy_image[(y >> huffman_bits) * huffman_xsize +
-                          (x >> huffman_bits)] >> 8) & 0xffff]
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The `huffman_code[5 * meta_code + k]`, codes with _k_ == 0 are for the
-green & length code, _k_ == 4 for the distance code, and the codes at
-_k_ == 1, 2, and 3, are for codes of length 256 for red, blue and alpha,
-respectively.
-
-The value of _k_ for the reference position in `meta_code` determines the
-length of the Huffman code:
-
-  * k = 0; length = 256 + 24 + cache_size
-  * k = 1, 2, or 3;  length = 256
-  * k = 4, length = 40.
+  * Code 18 emits a streak of zeros of length \[11..138\], i.e.,
+    11 + `ReadBits(7)` times.
 
 
 6 Overall Structure of the Format
@@ -953,21 +1028,21 @@ of pixels (xsize * ysize).
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 <format> ::= <RIFF header><image size><image stream>
-<image stream> ::= (<optional-transform><image stream>);
-                    <spatially-coded image>
+<image stream> ::= <optional-transform><spatially-coded image>
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 
 #### Structure of Transforms
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-<optional-transform> ::= 1-bit <transform> <optional-transform> | 0-bit
+<optional-transform> ::= (1-bit value 1; <transform> <optional-transform>) |
+                         1-bit value 0
 <transform> ::= <predictor-tx> | <color-tx> | <subtract-green-tx> |
                 <color-indexing-tx>
 <predictor-tx> ::= 2-bit value 0; <predictor image>
-<predictor image> ::= 3-bit sub-pixel code | <entropy-coded image>
+<predictor image> ::= 3-bit sub-pixel code ; <entropy-coded image>
 <color-tx> ::= 2-bit value 1; <color image>
-<color image> ::= 3-bit sub-pixel code | <entropy-coded image>
+<color image> ::= 3-bit sub-pixel code ; <entropy-coded image>
 <subtract-green-tx> ::= 2-bit value 2
 <color-indexing-tx> ::= 2-bit value 3; <color-indexing image>
 <color-indexing image> ::= 8-bit color count; <entropy-coded image>
@@ -984,13 +1059,18 @@ of pixels (xsize * ysize).
 <entropy image> ::= 3-bit subsample value; <entropy-coded image>
 <color cache info> ::= 1 bit value 0 |
                        (1-bit value 1; 4-bit value for color cache size)
-<huffman codes> ::= <huffman code> | <huffman code><huffman codes>
+<huffman codes> ::= <huffman code group> | <huffman code group><huffman codes>
+<huffman code group> ::= <huffman code><huffman code><huffman code>
+                         <huffman code><huffman code>
+                         See "Interpretation of Meta Huffman codes" to
+                         understand what each of these five Huffman codes are
+                         for.
 <huffman code> ::= <simple huffman code> | <normal huffman code>
 <simple huffman code> ::= see "Simple code length code" for details
 <normal huffman code> ::= <code length code>; encoded code lengths
 <code length code> ::= see section "Normal code length code"
-<lz77-coded image> ::= (<argb-pixel> | <color-cache-code> | <lz77-copy>) |
-                       (<lz77-coded image> | "")
+<lz77-coded image> ::= ((<argb-pixel> | <lz77-copy> | <color-cache-code>)
+                       <lz77-coded image>) | ""
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 A possible example sequence:
@@ -1001,3 +1081,5 @@ A possible example sequence:
 <color cache info><huffman codes>
 <lz77-coded image>
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+[canonical_huff]: http://en.wikipedia.org/wiki/Canonical_Huffman_code

examples/Android.mk

@@ -0,0 +1,46 @@
+LOCAL_PATH := $(call my-dir)
+
+include $(CLEAR_VARS)
+
+LOCAL_SRC_FILES := \
+    example_util.c \
+
+LOCAL_CFLAGS := $(WEBP_CFLAGS)
+LOCAL_C_INCLUDES := $(LOCAL_PATH)/../src
+
+LOCAL_MODULE := example_util
+
+include $(BUILD_STATIC_LIBRARY)
+
+include $(CLEAR_VARS)
+
+# Note: to enable jpeg/png encoding the sources from AOSP can be used with
+# minor modification to their Android.mk files.
+LOCAL_SRC_FILES := \
+    cwebp.c \
+    jpegdec.c \
+    metadata.c \
+    pngdec.c \
+    tiffdec.c \
+    webpdec.c \
+
+LOCAL_CFLAGS := $(WEBP_CFLAGS)
+LOCAL_C_INCLUDES := $(LOCAL_PATH)/../src
+LOCAL_STATIC_LIBRARIES := example_util webp
+
+LOCAL_MODULE := cwebp
+
+include $(BUILD_EXECUTABLE)
+
+include $(CLEAR_VARS)
+
+LOCAL_SRC_FILES := \
+    dwebp.c \
+
+LOCAL_CFLAGS := $(WEBP_CFLAGS)
+LOCAL_C_INCLUDES := $(LOCAL_PATH)/../src
+LOCAL_STATIC_LIBRARIES := example_util webp
+
+LOCAL_MODULE := dwebp
+
+include $(BUILD_EXECUTABLE)

examples/Makefile.am

@@ -1,25 +1,54 @@
-AM_CPPFLAGS = -I$(top_srcdir)/src
+AM_CPPFLAGS = -I$(top_builddir)/src -I$(top_srcdir)/src
 
 bin_PROGRAMS = dwebp cwebp
+if BUILD_VWEBP
+  bin_PROGRAMS += vwebp
+endif
 if WANT_MUX
   bin_PROGRAMS += webpmux
 endif
+
+if BUILD_GIF2WEBP
+  bin_PROGRAMS += gif2webp
+endif
+
 noinst_LTLIBRARIES = libexampleutil.la
 
-libexampleutil_la_SOURCES = example_util.c
-libexampleutilinclude_HEADERS = example_util.h
-libexampleutilincludedir =
+libexampleutil_la_SOURCES = example_util.c example_util.h stopwatch.h
 
 dwebp_SOURCES = dwebp.c stopwatch.h
 dwebp_CPPFLAGS  = $(AM_CPPFLAGS) $(USE_EXPERIMENTAL_CODE)
 dwebp_CPPFLAGS += $(JPEG_INCLUDES) $(PNG_INCLUDES)
-dwebp_LDADD = libexampleutil.la ../src/libwebp.la $(PNG_LIBS) $(JPEG_LIBS)
+dwebp_LDADD = libexampleutil.la $(PNG_LIBS) $(JPEG_LIBS)
 
-cwebp_SOURCES = cwebp.c stopwatch.h
+cwebp_SOURCES  = cwebp.c metadata.c metadata.h stopwatch.h
+cwebp_SOURCES += jpegdec.c jpegdec.h
+cwebp_SOURCES += pngdec.c pngdec.h
+cwebp_SOURCES += tiffdec.c tiffdec.h
+cwebp_SOURCES += webpdec.c webpdec.h
+cwebp_SOURCES += wicdec.c wicdec.h
 cwebp_CPPFLAGS  = $(AM_CPPFLAGS) $(USE_EXPERIMENTAL_CODE)
 cwebp_CPPFLAGS += $(JPEG_INCLUDES) $(PNG_INCLUDES) $(TIFF_INCLUDES)
-cwebp_LDADD = ../src/libwebp.la $(JPEG_LIBS) $(PNG_LIBS) $(TIFF_LIBS)
+cwebp_LDADD  = libexampleutil.la ../src/libwebp.la
+cwebp_LDADD += $(JPEG_LIBS) $(PNG_LIBS) $(TIFF_LIBS)
+
+gif2webp_SOURCES = gif2webp.c gif2webp_util.c
+gif2webp_CPPFLAGS = $(AM_CPPFLAGS) $(USE_EXPERIMENTAL_CODE) $(GIF_INCLUDES)
+gif2webp_LDADD  = libexampleutil.la ../src/mux/libwebpmux.la ../src/libwebp.la
+gif2webp_LDADD += $(GIF_LIBS)
 
 webpmux_SOURCES = webpmux.c
 webpmux_CPPFLAGS = $(AM_CPPFLAGS) $(USE_EXPERIMENTAL_CODE)
 webpmux_LDADD = libexampleutil.la ../src/mux/libwebpmux.la ../src/libwebp.la
+
+vwebp_SOURCES = vwebp.c
+vwebp_CPPFLAGS = $(AM_CPPFLAGS) $(USE_EXPERIMENTAL_CODE) $(GL_INCLUDES)
+vwebp_LDADD = libexampleutil.la ../src/demux/libwebpdemux.la $(GL_LIBS)
+
+if BUILD_LIBWEBPDECODER
+  dwebp_LDADD += ../src/libwebpdecoder.la
+  vwebp_LDADD += ../src/libwebpdecoder.la
+else
+  dwebp_LDADD += ../src/libwebp.la
+  vwebp_LDADD += ../src/libwebp.la
+endif

examples/dwebp.c

@@ -1,13 +1,13 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
-//  Command-line tool for decoding a WebP image
-//
-//  Compile with:     gcc -o dwebp dwebp.c -lwebpdecode
+//  Command-line tool for decoding a WebP image.
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
@@ -17,7 +17,7 @@
 #include <string.h>
 
 #ifdef HAVE_CONFIG_H
-#include "config.h"
+#include "webp/config.h"
 #endif
 
 #ifdef WEBP_HAVE_PNG
@@ -32,6 +32,7 @@
 #define COBJMACROS
 #define _WIN32_IE 0x500  // Workaround bug in shlwapi.h when compiling C++
                          // code with COBJMACROS.
+#include <ole2.h>  // CreateStreamOnHGlobal()
 #include <shlwapi.h>
 #include <windows.h>
 #include <wincodec.h>
@@ -43,13 +44,13 @@
 
 static int verbose = 0;
 #ifndef WEBP_DLL
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 extern "C" {
 #endif
 
 extern void* VP8GetCPUInfo;   // opaque forward declaration.
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 #endif  // WEBP_DLL
@@ -62,6 +63,9 @@ typedef enum {
   PAM,
   PPM,
   PGM,
+  BMP,
+  TIFF,
+  YUV,
   ALPHA_PLANE_ONLY  // this is for experimenting only
 } OutputFileFormat;
 
@@ -69,10 +73,10 @@ typedef enum {
 
 #define IFS(fn)                                                     \
   do {                                                              \
-     if (SUCCEEDED(hr)) {                                           \
-        hr = (fn);                                                  \
-        if (FAILED(hr)) fprintf(stderr, #fn " failed %08x\n", hr);  \
-     }                                                              \
+    if (SUCCEEDED(hr)) {                                            \
+      hr = (fn);                                                    \
+      if (FAILED(hr)) fprintf(stderr, #fn " failed %08lx\n", hr);   \
+    }                                                               \
   } while (0)
 
 #ifdef __cplusplus
@@ -82,71 +86,102 @@ typedef enum {
 #endif
 
 static HRESULT CreateOutputStream(const char* out_file_name,
-                                  IStream** ppStream) {
+                                  int write_to_mem, IStream** stream) {
   HRESULT hr = S_OK;
-  IFS(SHCreateStreamOnFileA(out_file_name, STGM_WRITE | STGM_CREATE, ppStream));
-  if (FAILED(hr))
-    fprintf(stderr, "Error opening output file %s (%08x)\n", out_file_name, hr);
+  if (write_to_mem) {
+    // Output to a memory buffer. This is freed when 'stream' is released.
+    IFS(CreateStreamOnHGlobal(NULL, TRUE, stream));
+  } else {
+    IFS(SHCreateStreamOnFileA(out_file_name, STGM_WRITE | STGM_CREATE, stream));
+  }
+  if (FAILED(hr)) {
+    fprintf(stderr, "Error opening output file %s (%08lx)\n",
+            out_file_name, hr);
+  }
   return hr;
 }
 
-static HRESULT WriteUsingWIC(const char* out_file_name, REFGUID container_guid,
-                             unsigned char* rgb, int stride,
+static HRESULT WriteUsingWIC(const char* out_file_name, int use_stdout,
+                             REFGUID container_guid,
+                             uint8_t* rgb, int stride,
                              uint32_t width, uint32_t height, int has_alpha) {
   HRESULT hr = S_OK;
-  IWICImagingFactory* pFactory = NULL;
-  IWICBitmapFrameEncode* pFrame = NULL;
-  IWICBitmapEncoder* pEncoder = NULL;
-  IStream* pStream = NULL;
+  IWICImagingFactory* factory = NULL;
+  IWICBitmapFrameEncode* frame = NULL;
+  IWICBitmapEncoder* encoder = NULL;
+  IStream* stream = NULL;
   WICPixelFormatGUID pixel_format = has_alpha ? GUID_WICPixelFormat32bppBGRA
                                               : GUID_WICPixelFormat24bppBGR;
 
   IFS(CoInitialize(NULL));
   IFS(CoCreateInstance(MAKE_REFGUID(CLSID_WICImagingFactory), NULL,
-          CLSCTX_INPROC_SERVER, MAKE_REFGUID(IID_IWICImagingFactory),
-          (LPVOID*)&pFactory));
+                       CLSCTX_INPROC_SERVER,
+                       MAKE_REFGUID(IID_IWICImagingFactory),
+                       (LPVOID*)&factory));
   if (hr == REGDB_E_CLASSNOTREG) {
     fprintf(stderr,
             "Couldn't access Windows Imaging Component (are you running "
             "Windows XP SP3 or newer?). PNG support not available. "
             "Use -ppm or -pgm for available PPM and PGM formats.\n");
   }
-  IFS(CreateOutputStream(out_file_name, &pStream));
-  IFS(IWICImagingFactory_CreateEncoder(pFactory, container_guid, NULL,
-          &pEncoder));
-  IFS(IWICBitmapEncoder_Initialize(pEncoder, pStream,
+  IFS(CreateOutputStream(out_file_name, use_stdout, &stream));
+  IFS(IWICImagingFactory_CreateEncoder(factory, container_guid, NULL,
+                                       &encoder));
+  IFS(IWICBitmapEncoder_Initialize(encoder, stream,
                                    WICBitmapEncoderNoCache));
-  IFS(IWICBitmapEncoder_CreateNewFrame(pEncoder, &pFrame, NULL));
-  IFS(IWICBitmapFrameEncode_Initialize(pFrame, NULL));
-  IFS(IWICBitmapFrameEncode_SetSize(pFrame, width, height));
-  IFS(IWICBitmapFrameEncode_SetPixelFormat(pFrame, &pixel_format));
-  IFS(IWICBitmapFrameEncode_WritePixels(pFrame, height, stride,
-          height * stride, rgb));
-  IFS(IWICBitmapFrameEncode_Commit(pFrame));
-  IFS(IWICBitmapEncoder_Commit(pEncoder));
+  IFS(IWICBitmapEncoder_CreateNewFrame(encoder, &frame, NULL));
+  IFS(IWICBitmapFrameEncode_Initialize(frame, NULL));
+  IFS(IWICBitmapFrameEncode_SetSize(frame, width, height));
+  IFS(IWICBitmapFrameEncode_SetPixelFormat(frame, &pixel_format));
+  IFS(IWICBitmapFrameEncode_WritePixels(frame, height, stride,
+                                        height * stride, rgb));
+  IFS(IWICBitmapFrameEncode_Commit(frame));
+  IFS(IWICBitmapEncoder_Commit(encoder));
 
-  if (pFrame != NULL) IUnknown_Release(pFrame);
-  if (pEncoder != NULL) IUnknown_Release(pEncoder);
-  if (pFactory != NULL) IUnknown_Release(pFactory);
-  if (pStream != NULL) IUnknown_Release(pStream);
+  if (SUCCEEDED(hr) && use_stdout) {
+    HGLOBAL image;
+    IFS(GetHGlobalFromStream(stream, &image));
+    if (SUCCEEDED(hr)) {
+      HANDLE std_output = GetStdHandle(STD_OUTPUT_HANDLE);
+      DWORD mode;
+      const BOOL update_mode = GetConsoleMode(std_output, &mode);
+      const void* const image_mem = GlobalLock(image);
+      DWORD bytes_written = 0;
+
+      // Clear output processing if necessary, then output the image.
+      if (update_mode) SetConsoleMode(std_output, 0);
+      if (!WriteFile(std_output, image_mem, (DWORD)GlobalSize(image),
+                     &bytes_written, NULL) ||
+          bytes_written != GlobalSize(image)) {
+        hr = E_FAIL;
+      }
+      if (update_mode) SetConsoleMode(std_output, mode);
+      GlobalUnlock(image);
+    }
+  }
+
+  if (frame != NULL) IUnknown_Release(frame);
+  if (encoder != NULL) IUnknown_Release(encoder);
+  if (factory != NULL) IUnknown_Release(factory);
+  if (stream != NULL) IUnknown_Release(stream);
   return hr;
 }
 
-static int WritePNG(const char* out_file_name,
+static int WritePNG(const char* out_file_name, int use_stdout,
                     const WebPDecBuffer* const buffer) {
   const uint32_t width = buffer->width;
   const uint32_t height = buffer->height;
-  unsigned char* const rgb = buffer->u.RGBA.rgba;
+  uint8_t* const rgb = buffer->u.RGBA.rgba;
   const int stride = buffer->u.RGBA.stride;
   const int has_alpha = (buffer->colorspace == MODE_BGRA);
 
-  return SUCCEEDED(WriteUsingWIC(out_file_name,
-             MAKE_REFGUID(GUID_ContainerFormatPng), rgb, stride, width,
-             height, has_alpha));
+  return SUCCEEDED(WriteUsingWIC(out_file_name, use_stdout,
+                                 MAKE_REFGUID(GUID_ContainerFormatPng),
+                                 rgb, stride, width, height, has_alpha));
 }
 
 #elif defined(WEBP_HAVE_PNG)    // !HAVE_WINCODEC_H
-static void PNGAPI error_function(png_structp png, png_const_charp dummy) {
+static void PNGAPI PNGErrorFunction(png_structp png, png_const_charp dummy) {
   (void)dummy;  // remove variable-unused warning
   longjmp(png_jmpbuf(png), 1);
 }
@@ -154,7 +189,7 @@ static void PNGAPI error_function(png_structp png, png_const_charp dummy) {
 static int WritePNG(FILE* out_file, const WebPDecBuffer* const buffer) {
   const uint32_t width = buffer->width;
   const uint32_t height = buffer->height;
-  unsigned char* const rgb = buffer->u.RGBA.rgba;
+  uint8_t* const rgb = buffer->u.RGBA.rgba;
   const int stride = buffer->u.RGBA.stride;
   const int has_alpha = (buffer->colorspace == MODE_RGBA);
   png_structp png;
@@ -162,7 +197,7 @@ static int WritePNG(FILE* out_file, const WebPDecBuffer* const buffer) {
   png_uint_32 y;
 
   png = png_create_write_struct(PNG_LIBPNG_VER_STRING,
-                                NULL, error_function, NULL);
+                                NULL, PNGErrorFunction, NULL);
   if (png == NULL) {
     return 0;
   }
@@ -203,7 +238,7 @@ static int WritePNG(FILE* out_file, const WebPDecBuffer* const buffer) {
 static int WritePPM(FILE* fout, const WebPDecBuffer* const buffer, int alpha) {
   const uint32_t width = buffer->width;
   const uint32_t height = buffer->height;
-  const unsigned char* const rgb = buffer->u.RGBA.rgba;
+  const uint8_t* const rgb = buffer->u.RGBA.rgba;
   const int stride = buffer->u.RGBA.stride;
   const size_t bytes_per_px = alpha ? 4 : 3;
   uint32_t y;
@@ -222,10 +257,150 @@ static int WritePPM(FILE* fout, const WebPDecBuffer* const buffer, int alpha) {
   return 1;
 }
 
+static void PutLE16(uint8_t* const dst, uint32_t value) {
+  dst[0] = (value >> 0) & 0xff;
+  dst[1] = (value >> 8) & 0xff;
+}
+
+static void PutLE32(uint8_t* const dst, uint32_t value) {
+  PutLE16(dst + 0, (value >>  0) & 0xffff);
+  PutLE16(dst + 2, (value >> 16) & 0xffff);
+}
+
+#define BMP_HEADER_SIZE 54
+static int WriteBMP(FILE* fout, const WebPDecBuffer* const buffer) {
+  const int has_alpha = (buffer->colorspace != MODE_BGR);
+  const uint32_t width = buffer->width;
+  const uint32_t height = buffer->height;
+  const uint8_t* const rgba = buffer->u.RGBA.rgba;
+  const int stride = buffer->u.RGBA.stride;
+  const uint32_t bytes_per_px = has_alpha ? 4 : 3;
+  uint32_t y;
+  const uint32_t line_size = bytes_per_px * width;
+  const uint32_t bmp_stride = (line_size + 3) & ~3;   // pad to 4
+  const uint32_t total_size = bmp_stride * height + BMP_HEADER_SIZE;
+  uint8_t bmp_header[BMP_HEADER_SIZE] = { 0 };
+
+  // bitmap file header
+  PutLE16(bmp_header + 0, 0x4d42);                // signature 'BM'
+  PutLE32(bmp_header + 2, total_size);            // size including header
+  PutLE32(bmp_header + 6, 0);                     // reserved
+  PutLE32(bmp_header + 10, BMP_HEADER_SIZE);      // offset to pixel array
+  // bitmap info header
+  PutLE32(bmp_header + 14, 40);                   // DIB header size
+  PutLE32(bmp_header + 18, width);                // dimensions
+  PutLE32(bmp_header + 22, -(int)height);         // vertical flip!
+  PutLE16(bmp_header + 26, 1);                    // number of planes
+  PutLE16(bmp_header + 28, bytes_per_px * 8);     // bits per pixel
+  PutLE32(bmp_header + 30, 0);                    // no compression (BI_RGB)
+  PutLE32(bmp_header + 34, 0);                    // image size (dummy)
+  PutLE32(bmp_header + 38, 2400);                 // x pixels/meter
+  PutLE32(bmp_header + 42, 2400);                 // y pixels/meter
+  PutLE32(bmp_header + 46, 0);                    // number of palette colors
+  PutLE32(bmp_header + 50, 0);                    // important color count
+
+  // TODO(skal): color profile
+
+  // write header
+  if (fwrite(bmp_header, sizeof(bmp_header), 1, fout) != 1) {
+    return 0;
+  }
+
+  // write pixel array
+  for (y = 0; y < height; ++y) {
+    if (fwrite(rgba + y * stride, line_size, 1, fout) != 1) {
+      return 0;
+    }
+    // write padding zeroes
+    if (bmp_stride != line_size) {
+      const uint8_t zeroes[3] = { 0 };
+      if (fwrite(zeroes, bmp_stride - line_size, 1, fout) != 1) {
+        return 0;
+      }
+    }
+  }
+  return 1;
+}
+#undef BMP_HEADER_SIZE
+
+#define NUM_IFD_ENTRIES 15
+#define EXTRA_DATA_SIZE 16
+// 10b for signature/header + n * 12b entries + 4b for IFD terminator:
+#define EXTRA_DATA_OFFSET (10 + 12 * NUM_IFD_ENTRIES + 4)
+#define TIFF_HEADER_SIZE (EXTRA_DATA_OFFSET + EXTRA_DATA_SIZE)
+
+static int WriteTIFF(FILE* fout, const WebPDecBuffer* const buffer) {
+  const int has_alpha = (buffer->colorspace != MODE_RGB);
+  const uint32_t width = buffer->width;
+  const uint32_t height = buffer->height;
+  const uint8_t* const rgba = buffer->u.RGBA.rgba;
+  const int stride = buffer->u.RGBA.stride;
+  const uint8_t bytes_per_px = has_alpha ? 4 : 3;
+  // For non-alpha case, we omit tag 0x152 (ExtraSamples).
+  const uint8_t num_ifd_entries = has_alpha ? NUM_IFD_ENTRIES
+                                            : NUM_IFD_ENTRIES - 1;
+  uint8_t tiff_header[TIFF_HEADER_SIZE] = {
+    0x49, 0x49, 0x2a, 0x00,   // little endian signature
+    8, 0, 0, 0,               // offset to the unique IFD that follows
+    // IFD (offset = 8). Entries must be written in increasing tag order.
+    num_ifd_entries, 0,       // Number of entries in the IFD (12 bytes each).
+    0x00, 0x01, 3, 0, 1, 0, 0, 0, 0, 0, 0, 0,    //  10: Width  (TBD)
+    0x01, 0x01, 3, 0, 1, 0, 0, 0, 0, 0, 0, 0,    //  22: Height (TBD)
+    0x02, 0x01, 3, 0, bytes_per_px, 0, 0, 0,     //  34: BitsPerSample: 8888
+        EXTRA_DATA_OFFSET + 0, 0, 0, 0,
+    0x03, 0x01, 3, 0, 1, 0, 0, 0, 1, 0, 0, 0,    //  46: Compression: none
+    0x06, 0x01, 3, 0, 1, 0, 0, 0, 2, 0, 0, 0,    //  58: Photometric: RGB
+    0x11, 0x01, 4, 0, 1, 0, 0, 0,                //  70: Strips offset:
+        TIFF_HEADER_SIZE, 0, 0, 0,               //      data follows header
+    0x12, 0x01, 3, 0, 1, 0, 0, 0, 1, 0, 0, 0,    //  82: Orientation: topleft
+    0x15, 0x01, 3, 0, 1, 0, 0, 0,                //  94: SamplesPerPixels
+        bytes_per_px, 0, 0, 0,
+    0x16, 0x01, 3, 0, 1, 0, 0, 0, 0, 0, 0, 0,    // 106: Rows per strip (TBD)
+    0x17, 0x01, 4, 0, 1, 0, 0, 0, 0, 0, 0, 0,    // 118: StripByteCount (TBD)
+    0x1a, 0x01, 5, 0, 1, 0, 0, 0,                // 130: X-resolution
+        EXTRA_DATA_OFFSET + 8, 0, 0, 0,
+    0x1b, 0x01, 5, 0, 1, 0, 0, 0,                // 142: Y-resolution
+        EXTRA_DATA_OFFSET + 8, 0, 0, 0,
+    0x1c, 0x01, 3, 0, 1, 0, 0, 0, 1, 0, 0, 0,    // 154: PlanarConfiguration
+    0x28, 0x01, 3, 0, 1, 0, 0, 0, 2, 0, 0, 0,    // 166: ResolutionUnit (inch)
+    0x52, 0x01, 3, 0, 1, 0, 0, 0, 1, 0, 0, 0,    // 178: ExtraSamples: rgbA
+    0, 0, 0, 0,                                  // 190: IFD terminator
+    // EXTRA_DATA_OFFSET:
+    8, 0, 8, 0, 8, 0, 8, 0,      // BitsPerSample
+    72, 0, 0, 0, 1, 0, 0, 0      // 72 pixels/inch, for X/Y-resolution
+  };
+  uint32_t y;
+
+  // Fill placeholders in IFD:
+  PutLE32(tiff_header + 10 + 8, width);
+  PutLE32(tiff_header + 22 + 8, height);
+  PutLE32(tiff_header + 106 + 8, height);
+  PutLE32(tiff_header + 118 + 8, width * bytes_per_px * height);
+  if (!has_alpha) PutLE32(tiff_header + 178, 0);  // IFD terminator
+
+  // write header
+  if (fwrite(tiff_header, sizeof(tiff_header), 1, fout) != 1) {
+    return 0;
+  }
+  // write pixel values
+  for (y = 0; y < height; ++y) {
+    if (fwrite(rgba + y * stride, bytes_per_px, width, fout) != width) {
+      return 0;
+    }
+  }
+
+  return 1;
+}
+
+#undef TIFF_HEADER_SIZE
+#undef EXTRA_DATA_OFFSET
+#undef EXTRA_DATA_SIZE
+#undef NUM_IFD_ENTRIES
+
 static int WriteAlphaPlane(FILE* fout, const WebPDecBuffer* const buffer) {
   const uint32_t width = buffer->width;
   const uint32_t height = buffer->height;
-  const unsigned char* const a = buffer->u.YUVA.a;
+  const uint8_t* const a = buffer->u.YUVA.a;
   const int a_stride = buffer->u.YUVA.a_stride;
   uint32_t y;
   assert(a != NULL);
@@ -238,60 +413,81 @@ static int WriteAlphaPlane(FILE* fout, const WebPDecBuffer* const buffer) {
   return 1;
 }
 
-static int WritePGM(FILE* fout, const WebPDecBuffer* const buffer) {
+// format=PGM: save a grayscale PGM file using the IMC4 layout
+// (http://www.fourcc.org/yuv.php#IMC4). This is a very convenient format for
+// viewing the samples, esp. for odd dimensions.
+// format=YUV: just save the Y/U/V/A planes sequentially without header.
+static int WritePGMOrYUV(FILE* fout, const WebPDecBuffer* const buffer,
+                         OutputFileFormat format) {
   const int width = buffer->width;
   const int height = buffer->height;
   const WebPYUVABuffer* const yuv = &buffer->u.YUVA;
-  // Save a grayscale PGM file using the IMC4 layout
-  // (http://www.fourcc.org/yuv.php#IMC4). This is a very
-  // convenient format for viewing the samples, esp. for
-  // odd dimensions.
   int ok = 1;
   int y;
+  const int pad = (format == YUV) ? 0 : 1;
   const int uv_width = (width + 1) / 2;
   const int uv_height = (height + 1) / 2;
-  const int out_stride = (width + 1) & ~1;
+  const int out_stride = (width + pad) & ~pad;
   const int a_height = yuv->a ? height : 0;
-  fprintf(fout, "P5\n%d %d\n255\n", out_stride, height + uv_height + a_height);
+  if (format == PGM) {
+    fprintf(fout, "P5\n%d %d\n255\n",
+            out_stride, height + uv_height + a_height);
+  }
   for (y = 0; ok && y < height; ++y) {
     ok &= (fwrite(yuv->y + y * yuv->y_stride, width, 1, fout) == 1);
-    if (width & 1) fputc(0, fout);    // padding byte
+    if (format == PGM) {
+      if (width & 1) fputc(0, fout);    // padding byte
+    }
   }
-  for (y = 0; ok && y < uv_height; ++y) {
-    ok &= (fwrite(yuv->u + y * yuv->u_stride, uv_width, 1, fout) == 1);
-    ok &= (fwrite(yuv->v + y * yuv->v_stride, uv_width, 1, fout) == 1);
+  if (format == PGM) {   // IMC4 layout
+    for (y = 0; ok && y < uv_height; ++y) {
+      ok &= (fwrite(yuv->u + y * yuv->u_stride, uv_width, 1, fout) == 1);
+      ok &= (fwrite(yuv->v + y * yuv->v_stride, uv_width, 1, fout) == 1);
+    }
+  } else {
+    for (y = 0; ok && y < uv_height; ++y) {
+      ok &= (fwrite(yuv->u + y * yuv->u_stride, uv_width, 1, fout) == 1);
+    }
+    for (y = 0; ok && y < uv_height; ++y) {
+      ok &= (fwrite(yuv->v + y * yuv->v_stride, uv_width, 1, fout) == 1);
+    }
   }
   for (y = 0; ok && y < a_height; ++y) {
     ok &= (fwrite(yuv->a + y * yuv->a_stride, width, 1, fout) == 1);
-    if (width & 1) fputc(0, fout);    // padding byte
+    if (format == PGM) {
+      if (width & 1) fputc(0, fout);    // padding byte
+    }
   }
   return ok;
 }
 
-static void SaveOutput(const WebPDecBuffer* const buffer,
-                       OutputFileFormat format, const char* const out_file) {
+static int SaveOutput(const WebPDecBuffer* const buffer,
+                      OutputFileFormat format, const char* const out_file) {
   FILE* fout = NULL;
   int needs_open_file = 1;
+  const int use_stdout = !strcmp(out_file, "-");
   int ok = 1;
   Stopwatch stop_watch;
 
-  if (verbose)
-    StopwatchReadAndReset(&stop_watch);
+  if (verbose) {
+    StopwatchReset(&stop_watch);
+  }
 
 #ifdef HAVE_WINCODEC_H
   needs_open_file = (format != PNG);
 #endif
+
   if (needs_open_file) {
-    fout = fopen(out_file, "wb");
-    if (!fout) {
+    fout = use_stdout ? ExUtilSetBinaryMode(stdout) : fopen(out_file, "wb");
+    if (fout == NULL) {
       fprintf(stderr, "Error opening output file %s\n", out_file);
-      return;
+      return 0;
     }
   }
 
   if (format == PNG) {
 #ifdef HAVE_WINCODEC_H
-    ok &= WritePNG(out_file, buffer);
+    ok &= WritePNG(out_file, use_stdout, buffer);
 #else
     ok &= WritePNG(fout, buffer);
 #endif
@@ -299,23 +495,36 @@ static void SaveOutput(const WebPDecBuffer* const buffer,
     ok &= WritePPM(fout, buffer, 1);
   } else if (format == PPM) {
     ok &= WritePPM(fout, buffer, 0);
-  } else if (format == PGM) {
-    ok &= WritePGM(fout, buffer);
+  } else if (format == BMP) {
+    ok &= WriteBMP(fout, buffer);
+  } else if (format == TIFF) {
+    ok &= WriteTIFF(fout, buffer);
+  } else if (format == PGM || format == YUV) {
+    ok &= WritePGMOrYUV(fout, buffer, format);
   } else if (format == ALPHA_PLANE_ONLY) {
     ok &= WriteAlphaPlane(fout, buffer);
   }
-  if (fout) {
+  if (fout != NULL && fout != stdout) {
     fclose(fout);
   }
   if (ok) {
-    printf("Saved file %s\n", out_file);
+    if (use_stdout) {
+      fprintf(stderr, "Saved to stdout\n");
+    } else {
+      fprintf(stderr, "Saved file %s\n", out_file);
+    }
     if (verbose) {
-      const double time = StopwatchReadAndReset(&stop_watch);
-      printf("Time to write output: %.3fs\n", time);
+      const double write_time = StopwatchReadAndReset(&stop_watch);
+      fprintf(stderr, "Time to write output: %.3fs\n", write_time);
     }
   } else {
-    fprintf(stderr, "Error writing file %s !!\n", out_file);
+    if (use_stdout) {
+      fprintf(stderr, "Error writing to stdout !!\n");
+    } else {
+      fprintf(stderr, "Error writing file %s !!\n", out_file);
+    }
   }
+  return ok;
 }
 
 static void Help(void) {
@@ -324,30 +533,43 @@ static void Help(void) {
          "Use following options to convert into alternate image formats:\n"
          "  -pam ......... save the raw RGBA samples as a color PAM\n"
          "  -ppm ......... save the raw RGB samples as a color PPM\n"
+         "  -bmp ......... save as uncompressed BMP format\n"
+         "  -tiff ........ save as uncompressed TIFF format\n"
          "  -pgm ......... save the raw YUV samples as a grayscale PGM\n"
-         "                 file with IMC4 layout.\n"
+         "                 file with IMC4 layout\n"
+         "  -yuv ......... save the raw YUV samples in flat layout\n"
+         "\n"
          " Other options are:\n"
-         "  -version  .... print version number and exit.\n"
-         "  -nofancy ..... don't use the fancy YUV420 upscaler.\n"
-         "  -nofilter .... disable in-loop filtering.\n"
+         "  -version  .... print version number and exit\n"
+         "  -nofancy ..... don't use the fancy YUV420 upscaler\n"
+         "  -nofilter .... disable in-loop filtering\n"
+         "  -nodither .... disable dithering\n"
+         "  -dither <d> .. dithering strength (in 0..100)\n"
+#if WEBP_DECODER_ABI_VERSION > 0x0204
+         "  -alpha_dither  use alpha-plane dithering if needed\n"
+#endif
          "  -mt .......... use multi-threading\n"
          "  -crop <x> <y> <w> <h> ... crop output with the given rectangle\n"
          "  -scale <w> <h> .......... scale the output (*after* any cropping)\n"
-         "  -alpha ....... only save the alpha plane.\n"
-         "  -h     ....... this help message.\n"
+#if WEBP_DECODER_ABI_VERSION > 0x0203
+         "  -flip ........ flip the output vertically\n"
+#endif
+         "  -alpha ....... only save the alpha plane\n"
+         "  -incremental . use incremental decoding (useful for tests)\n"
+         "  -h     ....... this help message\n"
          "  -v     ....... verbose (e.g. print encoding/decoding times)\n"
 #ifndef WEBP_DLL
-         "  -noasm ....... disable all assembly optimizations.\n"
+         "  -noasm ....... disable all assembly optimizations\n"
 #endif
         );
 }
 
-static const char* const kStatusMessages[] = {
-  "OK", "OUT_OF_MEMORY", "INVALID_PARAM", "BITSTREAM_ERROR",
-  "UNSUPPORTED_FEATURE", "SUSPENDED", "USER_ABORT", "NOT_ENOUGH_DATA"
+static const char* const kFormatType[] = {
+  "unspecified", "lossy", "lossless"
 };
 
 int main(int argc, const char *argv[]) {
+  int ok = 0;
   const char *in_file = NULL;
   const char *out_file = NULL;
 
@@ -355,6 +577,7 @@ int main(int argc, const char *argv[]) {
   WebPDecBuffer* const output_buffer = &config.output;
   WebPBitstreamFeatures* const bitstream = &config.input;
   OutputFileFormat format = PNG;
+  int incremental = 0;
   int c;
 
   if (!WebPInitDecoderConfig(&config)) {
@@ -363,6 +586,7 @@ int main(int argc, const char *argv[]) {
   }
 
   for (c = 1; c < argc; ++c) {
+    int parse_error = 0;
     if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       Help();
       return 0;
@@ -378,6 +602,10 @@ int main(int argc, const char *argv[]) {
       format = PAM;
     } else if (!strcmp(argv[c], "-ppm")) {
       format = PPM;
+    } else if (!strcmp(argv[c], "-bmp")) {
+      format = BMP;
+    } else if (!strcmp(argv[c], "-tiff")) {
+      format = TIFF;
     } else if (!strcmp(argv[c], "-version")) {
       const int version = WebPGetDecoderVersion();
       printf("%d.%d.%d\n",
@@ -385,24 +613,44 @@ int main(int argc, const char *argv[]) {
       return 0;
     } else if (!strcmp(argv[c], "-pgm")) {
       format = PGM;
+    } else if (!strcmp(argv[c], "-yuv")) {
+      format = YUV;
     } else if (!strcmp(argv[c], "-mt")) {
       config.options.use_threads = 1;
+#if WEBP_DECODER_ABI_VERSION > 0x0204
+    } else if (!strcmp(argv[c], "-alpha_dither")) {
+      config.options.alpha_dithering_strength = 100;
+#endif
+    } else if (!strcmp(argv[c], "-nodither")) {
+      config.options.dithering_strength = 0;
+    } else if (!strcmp(argv[c], "-dither") && c < argc - 1) {
+      config.options.dithering_strength =
+          ExUtilGetInt(argv[++c], 0, &parse_error);
     } else if (!strcmp(argv[c], "-crop") && c < argc - 4) {
       config.options.use_cropping = 1;
-      config.options.crop_left   = strtol(argv[++c], NULL, 0);
-      config.options.crop_top    = strtol(argv[++c], NULL, 0);
-      config.options.crop_width  = strtol(argv[++c], NULL, 0);
-      config.options.crop_height = strtol(argv[++c], NULL, 0);
+      config.options.crop_left   = ExUtilGetInt(argv[++c], 0, &parse_error);
+      config.options.crop_top    = ExUtilGetInt(argv[++c], 0, &parse_error);
+      config.options.crop_width  = ExUtilGetInt(argv[++c], 0, &parse_error);
+      config.options.crop_height = ExUtilGetInt(argv[++c], 0, &parse_error);
     } else if (!strcmp(argv[c], "-scale") && c < argc - 2) {
       config.options.use_scaling = 1;
-      config.options.scaled_width  = strtol(argv[++c], NULL, 0);
-      config.options.scaled_height = strtol(argv[++c], NULL, 0);
+      config.options.scaled_width  = ExUtilGetInt(argv[++c], 0, &parse_error);
+      config.options.scaled_height = ExUtilGetInt(argv[++c], 0, &parse_error);
+#if WEBP_DECODER_ABI_VERSION > 0x0203
+    } else if (!strcmp(argv[c], "-flip")) {
+      config.options.flip = 1;
+#endif
     } else if (!strcmp(argv[c], "-v")) {
       verbose = 1;
 #ifndef WEBP_DLL
     } else if (!strcmp(argv[c], "-noasm")) {
       VP8GetCPUInfo = NULL;
 #endif
+    } else if (!strcmp(argv[c], "-incremental")) {
+      incremental = 1;
+    } else if (!strcmp(argv[c], "--")) {
+      if (c < argc - 1) in_file = argv[++c];
+      break;
     } else if (argv[c][0] == '-') {
       fprintf(stderr, "Unknown option '%s'\n", argv[c]);
       Help();
@@ -410,6 +658,11 @@ int main(int argc, const char *argv[]) {
     } else {
       in_file = argv[c];
     }
+
+    if (parse_error) {
+      Help();
+      return -1;
+    }
   }
 
   if (in_file == NULL) {
@@ -419,20 +672,11 @@ int main(int argc, const char *argv[]) {
   }
 
   {
-    Stopwatch stop_watch;
     VP8StatusCode status = VP8_STATUS_OK;
-    int ok;
     size_t data_size = 0;
     const uint8_t* data = NULL;
-
-    if (!ExUtilReadFile(in_file, &data, &data_size)) return -1;
-
-    if (verbose)
-      StopwatchReadAndReset(&stop_watch);
-
-    status = WebPGetFeatures(data, data_size, bitstream);
-    if (status != VP8_STATUS_OK) {
-      goto end;
+    if (!ExUtilLoadWebP(in_file, &data, &data_size, bitstream)) {
+      return -1;
     }
 
     switch (format) {
@@ -449,7 +693,15 @@ int main(int argc, const char *argv[]) {
       case PPM:
         output_buffer->colorspace = MODE_RGB;  // drops alpha for PPM
         break;
+      case BMP:
+        output_buffer->colorspace = bitstream->has_alpha ? MODE_BGRA : MODE_BGR;
+        break;
+      case TIFF:    // note: force pre-multiplied alpha
+        output_buffer->colorspace =
+            bitstream->has_alpha ? MODE_rgbA : MODE_RGB;
+        break;
       case PGM:
+      case YUV:
         output_buffer->colorspace = bitstream->has_alpha ? MODE_YUVA : MODE_YUV;
         break;
       case ALPHA_PLANE_ONLY:
@@ -459,36 +711,40 @@ int main(int argc, const char *argv[]) {
         free((void*)data);
         return -1;
     }
-    status = WebPDecode(data, data_size, &config);
 
-    if (verbose) {
-      const double time = StopwatchReadAndReset(&stop_watch);
-      printf("Time to decode picture: %.3fs\n", time);
+    if (incremental) {
+      status = ExUtilDecodeWebPIncremental(data, data_size, verbose, &config);
+    } else {
+      status = ExUtilDecodeWebP(data, data_size, verbose, &config);
     }
- end:
+
     free((void*)data);
     ok = (status == VP8_STATUS_OK);
     if (!ok) {
-      fprintf(stderr, "Decoding of %s failed.\n", in_file);
-      fprintf(stderr, "Status: %d (%s)\n", status, kStatusMessages[status]);
-      return -1;
+      ExUtilPrintWebPError(in_file, status);
+      goto Exit;
     }
   }
 
-  if (out_file) {
-    printf("Decoded %s. Dimensions: %d x %d%s. Now saving...\n", in_file,
-           output_buffer->width, output_buffer->height,
-           bitstream->has_alpha ? " (with alpha)" : "");
-    SaveOutput(output_buffer, format, out_file);
+  if (out_file != NULL) {
+    fprintf(stderr, "Decoded %s. Dimensions: %d x %d %s. Format: %s. "
+                    "Now saving...\n",
+            in_file, output_buffer->width, output_buffer->height,
+            bitstream->has_alpha ? " (with alpha)" : "",
+            kFormatType[bitstream->format]);
+    ok = SaveOutput(output_buffer, format, out_file);
   } else {
-    printf("File %s can be decoded (dimensions: %d x %d)%s.\n",
-           in_file, output_buffer->width, output_buffer->height,
-           bitstream->has_alpha ? " (with alpha)" : "");
-    printf("Nothing written; use -o flag to save the result as e.g. PNG.\n");
+    fprintf(stderr, "File %s can be decoded "
+                    "(dimensions: %d x %d %s. Format: %s).\n",
+            in_file, output_buffer->width, output_buffer->height,
+            bitstream->has_alpha ? " (with alpha)" : "",
+            kFormatType[bitstream->format]);
+    fprintf(stderr, "Nothing written; "
+                    "use -o flag to save the result as e.g. PNG.\n");
   }
+ Exit:
   WebPFreeDecBuffer(output_buffer);
-
-  return 0;
+  return ok ? 0 : -1;
 }
 
 //------------------------------------------------------------------------------

examples/example_util.c

@@ -1,32 +1,114 @@
 // Copyright 2012 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 //  Utility functions used by the example programs.
 //
 
 #include "./example_util.h"
+
+#if defined(_WIN32)
+#include <fcntl.h>   // for _O_BINARY
+#include <io.h>      // for _setmode()
+#endif
 #include <stdio.h>
 #include <stdlib.h>
+#include <string.h>
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
+#include "webp/decode.h"
+#include "./stopwatch.h"
+
+//------------------------------------------------------------------------------
+// String parsing
+
+uint32_t ExUtilGetUInt(const char* const v, int base, int* const error) {
+  char* end = NULL;
+  const uint32_t n = (v != NULL) ? (uint32_t)strtoul(v, &end, base) : 0u;
+  if (end == v && error != NULL && !*error) {
+    *error = 1;
+    fprintf(stderr, "Error! '%s' is not an integer.\n",
+            (v != NULL) ? v : "(null)");
+  }
+  return n;
+}
+
+int ExUtilGetInt(const char* const v, int base, int* const error) {
+  return (int)ExUtilGetUInt(v, base, error);
+}
+
+float ExUtilGetFloat(const char* const v, int* const error) {
+  char* end = NULL;
+  const float f = (v != NULL) ? (float)strtod(v, &end) : 0.f;
+  if (end == v && error != NULL && !*error) {
+    *error = 1;
+    fprintf(stderr, "Error! '%s' is not a floating point number.\n",
+            (v != NULL) ? v : "(null)");
+  }
+  return f;
+}
 
 // -----------------------------------------------------------------------------
 // File I/O
 
+FILE* ExUtilSetBinaryMode(FILE* file) {
+#if defined(_WIN32)
+  if (_setmode(_fileno(file), _O_BINARY) == -1) {
+    fprintf(stderr, "Failed to reopen file in O_BINARY mode.\n");
+    return NULL;
+  }
+#endif
+  return file;
+}
+
+int ExUtilReadFromStdin(const uint8_t** data, size_t* data_size) {
+  static const size_t kBlockSize = 16384;  // default initial size
+  size_t max_size = 0;
+  size_t size = 0;
+  uint8_t* input = NULL;
+
+  if (data == NULL || data_size == NULL) return 0;
+  *data = NULL;
+  *data_size = 0;
+
+  if (!ExUtilSetBinaryMode(stdin)) return 0;
+
+  while (!feof(stdin)) {
+    // We double the buffer size each time and read as much as possible.
+    const size_t extra_size = (max_size == 0) ? kBlockSize : max_size;
+    void* const new_data = realloc(input, max_size + extra_size);
+    if (new_data == NULL) goto Error;
+    input = (uint8_t*)new_data;
+    max_size += extra_size;
+    size += fread(input + size, 1, extra_size, stdin);
+    if (size < max_size) break;
+  }
+  if (ferror(stdin)) goto Error;
+  *data = input;
+  *data_size = size;
+  return 1;
+
+ Error:
+  free(input);
+  fprintf(stderr, "Could not read from stdin\n");
+  return 0;
+}
+
 int ExUtilReadFile(const char* const file_name,
                    const uint8_t** data, size_t* data_size) {
   int ok;
   void* file_data;
   size_t file_size;
   FILE* in;
+  const int from_stdin = (file_name == NULL) || !strcmp(file_name, "-");
 
-  if (file_name == NULL || data == NULL || data_size == NULL) return 0;
+  if (from_stdin) return ExUtilReadFromStdin(data, data_size);
+
+  if (data == NULL || data_size == NULL) return 0;
   *data = NULL;
   *data_size = 0;
 
@@ -44,8 +126,8 @@ int ExUtilReadFile(const char* const file_name,
   fclose(in);
 
   if (!ok) {
-    fprintf(stderr, "Could not read %zu bytes of data from file %s\n",
-            file_size, file_name);
+    fprintf(stderr, "Could not read %d bytes of data from file %s\n",
+            (int)file_size, file_name);
     free(file_data);
     return 0;
   }
@@ -54,6 +136,123 @@ int ExUtilReadFile(const char* const file_name,
   return 1;
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif
+int ExUtilWriteFile(const char* const file_name,
+                    const uint8_t* data, size_t data_size) {
+  int ok;
+  FILE* out;
+  const int to_stdout = (file_name == NULL) || !strcmp(file_name, "-");
+
+  if (data == NULL) {
+    return 0;
+  }
+  out = to_stdout ? stdout : fopen(file_name, "wb");
+  if (out == NULL) {
+    fprintf(stderr, "Error! Cannot open output file '%s'\n", file_name);
+    return 0;
+  }
+  ok = (fwrite(data, data_size, 1, out) == 1);
+  if (out != stdout) fclose(out);
+  return ok;
+}
+
+//------------------------------------------------------------------------------
+// WebP decoding
+
+static const char* const kStatusMessages[VP8_STATUS_NOT_ENOUGH_DATA + 1] = {
+  "OK", "OUT_OF_MEMORY", "INVALID_PARAM", "BITSTREAM_ERROR",
+  "UNSUPPORTED_FEATURE", "SUSPENDED", "USER_ABORT", "NOT_ENOUGH_DATA"
+};
+
+static void PrintAnimationWarning(const WebPDecoderConfig* const config) {
+  if (config->input.has_animation) {
+    fprintf(stderr,
+            "Error! Decoding of an animated WebP file is not supported.\n"
+            "       Use webpmux to extract the individual frames or\n"
+            "       vwebp to view this image.\n");
+  }
+}
+
+void ExUtilPrintWebPError(const char* const in_file, int status) {
+  fprintf(stderr, "Decoding of %s failed.\n", in_file);
+  fprintf(stderr, "Status: %d", status);
+  if (status >= VP8_STATUS_OK && status <= VP8_STATUS_NOT_ENOUGH_DATA) {
+    fprintf(stderr, "(%s)", kStatusMessages[status]);
+  }
+  fprintf(stderr, "\n");
+}
+
+int ExUtilLoadWebP(const char* const in_file,
+                   const uint8_t** data, size_t* data_size,
+                   WebPBitstreamFeatures* bitstream) {
+  VP8StatusCode status;
+  WebPBitstreamFeatures local_features;
+  if (!ExUtilReadFile(in_file, data, data_size)) return 0;
+
+  if (bitstream == NULL) {
+    bitstream = &local_features;
+  }
+
+  status = WebPGetFeatures(*data, *data_size, bitstream);
+  if (status != VP8_STATUS_OK) {
+    free((void*)*data);
+    *data = NULL;
+    *data_size = 0;
+    ExUtilPrintWebPError(in_file, status);
+    return 0;
+  }
+  return 1;
+}
+
+//------------------------------------------------------------------------------
+
+VP8StatusCode ExUtilDecodeWebP(const uint8_t* const data, size_t data_size,
+                               int verbose, WebPDecoderConfig* const config) {
+  Stopwatch stop_watch;
+  VP8StatusCode status = VP8_STATUS_OK;
+  if (config == NULL) return VP8_STATUS_INVALID_PARAM;
+
+  PrintAnimationWarning(config);
+
+  StopwatchReset(&stop_watch);
+
+  // Decoding call.
+  status = WebPDecode(data, data_size, config);
+
+  if (verbose) {
+    const double decode_time = StopwatchReadAndReset(&stop_watch);
+    fprintf(stderr, "Time to decode picture: %.3fs\n", decode_time);
+  }
+  return status;
+}
+
+VP8StatusCode ExUtilDecodeWebPIncremental(
+    const uint8_t* const data, size_t data_size,
+    int verbose, WebPDecoderConfig* const config) {
+  Stopwatch stop_watch;
+  VP8StatusCode status = VP8_STATUS_OK;
+  if (config == NULL) return VP8_STATUS_INVALID_PARAM;
+
+  PrintAnimationWarning(config);
+
+  StopwatchReset(&stop_watch);
+
+  // Decoding call.
+  {
+    WebPIDecoder* const idec = WebPIDecode(data, data_size, config);
+    if (idec == NULL) {
+      fprintf(stderr, "Failed during WebPINewDecoder().\n");
+      return VP8_STATUS_OUT_OF_MEMORY;
+    } else {
+      status = WebPIUpdate(idec, data, data_size);
+      WebPIDelete(idec);
+    }
+  }
+
+  if (verbose) {
+    const double decode_time = StopwatchReadAndReset(&stop_watch);
+    fprintf(stderr, "Time to decode picture: %.3fs\n", decode_time);
+  }
+  return status;
+}
+
+// -----------------------------------------------------------------------------

examples/example_util.h

@@ -1,8 +1,10 @@
 // Copyright 2012 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 //  Utility functions used by the example programs.
@@ -11,19 +13,76 @@
 #ifndef WEBP_EXAMPLES_EXAMPLE_UTIL_H_
 #define WEBP_EXAMPLES_EXAMPLE_UTIL_H_
 
-#include "webp/types.h"
+#include <stdio.h>
+#include "webp/decode.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 extern "C" {
 #endif
 
+//------------------------------------------------------------------------------
+// String parsing
+
+// Parses 'v' using strto(ul|l|d)(). If error is non-NULL, '*error' is set to
+// true on failure while on success it is left unmodified to allow chaining of
+// calls. An error is only printed on the first occurrence.
+uint32_t ExUtilGetUInt(const char* const v, int base, int* const error);
+int ExUtilGetInt(const char* const v, int base, int* const error);
+float ExUtilGetFloat(const char* const v, int* const error);
+
+//------------------------------------------------------------------------------
+// File I/O
+
+// Reopen file in binary (O_BINARY) mode.
+// Returns 'file' on success, NULL otherwise.
+FILE* ExUtilSetBinaryMode(FILE* file);
+
 // Allocates storage for entire file 'file_name' and returns contents and size
 // in 'data' and 'data_size'. Returns 1 on success, 0 otherwise. '*data' should
 // be deleted using free().
+// If 'file_name' is NULL or equal to "-", input is read from stdin by calling
+// the function ExUtilReadFromStdin().
 int ExUtilReadFile(const char* const file_name,
                    const uint8_t** data, size_t* data_size);
 
-#if defined(__cplusplus) || defined(c_plusplus)
+// Same as ExUtilReadFile(), but reads until EOF from stdin instead.
+int ExUtilReadFromStdin(const uint8_t** data, size_t* data_size);
+
+// Write a data segment into a file named 'file_name'. Returns true if ok.
+// If 'file_name' is NULL or equal to "-", output is written to stdout.
+int ExUtilWriteFile(const char* const file_name,
+                    const uint8_t* data, size_t data_size);
+
+//------------------------------------------------------------------------------
+// WebP decoding
+
+// Prints an informative error message regarding decode failure of 'in_file'.
+// 'status' is treated as a VP8StatusCode and if valid will be printed as a
+// text string.
+void ExUtilPrintWebPError(const char* const in_file, int status);
+
+// Reads a WebP from 'in_file', returning the contents and size in 'data' and
+// 'data_size'. If not NULL, 'bitstream' is populated using WebPGetFeatures().
+// Returns true on success.
+int ExUtilLoadWebP(const char* const in_file,
+                   const uint8_t** data, size_t* data_size,
+                   WebPBitstreamFeatures* bitstream);
+
+// Decodes the WebP contained in 'data'.
+// 'config' is a structure previously initialized by WebPInitDecoderConfig().
+// 'config->output' should have the desired colorspace selected. 'verbose' will
+// cause decode timing to be reported.
+// Returns the decoder status. On success 'config->output' will contain the
+// decoded picture.
+VP8StatusCode ExUtilDecodeWebP(const uint8_t* const data, size_t data_size,
+                               int verbose, WebPDecoderConfig* const config);
+
+// Same as ExUtilDecodeWebP(), but using the incremental decoder.
+VP8StatusCode ExUtilDecodeWebPIncremental(
+    const uint8_t* const data, size_t data_size,
+    int verbose, WebPDecoderConfig* const config);
+
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 

examples/gif2webp.c

@@ -0,0 +1,732 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//  simple tool to convert animated GIFs to WebP
+//
+// Authors: Skal (pascal.massimino@gmail.com)
+//          Urvang (urvang@google.com)
+
+#include <assert.h>
+#include <stdio.h>
+#include <string.h>
+
+#ifdef HAVE_CONFIG_H
+#include "webp/config.h"
+#endif
+
+#ifdef WEBP_HAVE_GIF
+
+#include <gif_lib.h>
+#include "webp/encode.h"
+#include "webp/mux.h"
+#include "./example_util.h"
+#include "./gif2webp_util.h"
+
+// GIFLIB_MAJOR is only defined in libgif >= 4.2.0.
+#if defined(GIFLIB_MAJOR) && defined(GIFLIB_MINOR)
+# define LOCAL_GIF_VERSION ((GIFLIB_MAJOR << 8) | GIFLIB_MINOR)
+# define LOCAL_GIF_PREREQ(maj, min) \
+    (LOCAL_GIF_VERSION >= (((maj) << 8) | (min)))
+#else
+# define LOCAL_GIF_VERSION 0
+# define LOCAL_GIF_PREREQ(maj, min) 0
+#endif
+
+#define GIF_TRANSPARENT_MASK 0x01
+#define GIF_DISPOSE_MASK     0x07
+#define GIF_DISPOSE_SHIFT    2
+#define WHITE_COLOR          0xffffffff
+#define MAX_CACHE_SIZE       30
+
+//------------------------------------------------------------------------------
+
+static int transparent_index;  // Index of transparent color in the map.
+
+static void ResetFrameInfo(WebPMuxFrameInfo* const info) {
+  WebPDataInit(&info->bitstream);
+  info->x_offset = 0;
+  info->y_offset = 0;
+  info->duration = 0;
+  info->id = WEBP_CHUNK_ANMF;
+  info->dispose_method = WEBP_MUX_DISPOSE_NONE;
+  info->blend_method = WEBP_MUX_BLEND;
+  transparent_index = -1;  // Opaque frame by default.
+}
+
+static void SanitizeKeyFrameIntervals(size_t* const kmin_ptr,
+                                      size_t* const kmax_ptr) {
+  size_t kmin = *kmin_ptr;
+  size_t kmax = *kmax_ptr;
+  int print_warning = 1;
+
+  if (kmin == 0) {  // Disable keyframe insertion.
+    kmax = ~0;
+    kmin = kmax - 1;
+    print_warning = 0;
+  }
+  if (kmax == 0) {
+    kmax = ~0;
+    print_warning = 0;
+  }
+
+  if (kmin >= kmax) {
+    kmin = kmax - 1;
+    if (print_warning) {
+      fprintf(stderr,
+              "WARNING: Setting kmin = %d, so that kmin < kmax.\n", (int)kmin);
+    }
+  } else if (kmin < (kmax / 2 + 1)) {
+    // This ensures that cache.keyframe + kmin >= kmax is always true. So, we
+    // can flush all the frames in the ‘count_since_key_frame == kmax’ case.
+    kmin = (kmax / 2 + 1);
+    if (print_warning) {
+      fprintf(stderr,
+              "WARNING: Setting kmin = %d, so that kmin >= kmax / 2 + 1.\n",
+              (int)kmin);
+    }
+  }
+  // Limit the max number of frames that are allocated.
+  if (kmax - kmin > MAX_CACHE_SIZE) {
+    kmin = kmax - MAX_CACHE_SIZE;
+    if (print_warning) {
+      fprintf(stderr,
+              "WARNING: Setting kmin = %d, so that kmax - kmin <= 30.\n",
+              (int)kmin);
+    }
+  }
+  *kmin_ptr = kmin;
+  *kmax_ptr = kmax;
+}
+
+static void Remap(const uint8_t* const src, const GifFileType* const gif,
+                  uint32_t* dst, int len) {
+  int i;
+  const GifColorType* colors;
+  const ColorMapObject* const cmap =
+      gif->Image.ColorMap ? gif->Image.ColorMap : gif->SColorMap;
+  if (cmap == NULL) return;
+  colors = cmap->Colors;
+
+  for (i = 0; i < len; ++i) {
+    const GifColorType c = colors[src[i]];
+    dst[i] = (src[i] == transparent_index) ? WEBP_UTIL_TRANSPARENT_COLOR
+           : c.Blue | (c.Green << 8) | (c.Red << 16) | (0xff << 24);
+  }
+}
+
+// Read the GIF image frame.
+static int ReadFrame(GifFileType* const gif, WebPFrameRect* const gif_rect,
+                     WebPPicture* const webp_frame) {
+  WebPPicture sub_image;
+  const GifImageDesc* const image_desc = &gif->Image;
+  uint32_t* dst = NULL;
+  uint8_t* tmp = NULL;
+  int ok = 0;
+  WebPFrameRect rect = {
+      image_desc->Left, image_desc->Top, image_desc->Width, image_desc->Height
+  };
+  *gif_rect = rect;
+
+  // Use a view for the sub-picture:
+  if (!WebPPictureView(webp_frame, rect.x_offset, rect.y_offset,
+                       rect.width, rect.height, &sub_image)) {
+    fprintf(stderr, "Sub-image %dx%d at position %d,%d is invalid!\n",
+            rect.width, rect.height, rect.x_offset, rect.y_offset);
+    return 0;
+  }
+  dst = sub_image.argb;
+
+  tmp = (uint8_t*)malloc(rect.width * sizeof(*tmp));
+  if (tmp == NULL) goto End;
+
+  if (image_desc->Interlace) {  // Interlaced image.
+    // We need 4 passes, with the following offsets and jumps.
+    const int interlace_offsets[] = { 0, 4, 2, 1 };
+    const int interlace_jumps[]   = { 8, 8, 4, 2 };
+    int pass;
+    for (pass = 0; pass < 4; ++pass) {
+      int y;
+      for (y = interlace_offsets[pass]; y < rect.height;
+           y += interlace_jumps[pass]) {
+        if (DGifGetLine(gif, tmp, rect.width) == GIF_ERROR) goto End;
+        Remap(tmp, gif, dst + y * sub_image.argb_stride, rect.width);
+      }
+    }
+  } else {  // Non-interlaced image.
+    int y;
+    for (y = 0; y < rect.height; ++y) {
+      if (DGifGetLine(gif, tmp, rect.width) == GIF_ERROR) goto End;
+      Remap(tmp, gif, dst + y * sub_image.argb_stride, rect.width);
+    }
+  }
+  ok = 1;
+
+ End:
+  if (!ok) webp_frame->error_code = sub_image.error_code;
+  WebPPictureFree(&sub_image);
+  free(tmp);
+  return ok;
+}
+
+static void GetBackgroundColor(const ColorMapObject* const color_map,
+                               int bgcolor_idx, uint32_t* const bgcolor) {
+  if (transparent_index != -1 && bgcolor_idx == transparent_index) {
+    *bgcolor = WEBP_UTIL_TRANSPARENT_COLOR;  // Special case.
+  } else if (color_map == NULL || color_map->Colors == NULL
+             || bgcolor_idx >= color_map->ColorCount) {
+    *bgcolor = WHITE_COLOR;
+    fprintf(stderr,
+            "GIF decode warning: invalid background color index. Assuming "
+            "white background.\n");
+  } else {
+    const GifColorType color = color_map->Colors[bgcolor_idx];
+    *bgcolor = (0xff        << 24)
+             | (color.Red   << 16)
+             | (color.Green <<  8)
+             | (color.Blue  <<  0);
+  }
+}
+
+static void DisplayGifError(const GifFileType* const gif, int gif_error) {
+  // libgif 4.2.0 has retired PrintGifError() and added GifErrorString().
+#if LOCAL_GIF_PREREQ(4,2)
+#if LOCAL_GIF_PREREQ(5,0)
+  // Static string actually, hence the const char* cast.
+  const char* error_str = (const char*)GifErrorString(
+      (gif == NULL) ? gif_error : gif->Error);
+#else
+  const char* error_str = (const char*)GifErrorString();
+  (void)gif;
+#endif
+  if (error_str == NULL) error_str = "Unknown error";
+  fprintf(stderr, "GIFLib Error %d: %s\n", gif_error, error_str);
+#else
+  (void)gif;
+  fprintf(stderr, "GIFLib Error %d: ", gif_error);
+  PrintGifError();
+  fprintf(stderr, "\n");
+#endif
+}
+
+static const char* const kErrorMessages[-WEBP_MUX_NOT_ENOUGH_DATA + 1] = {
+  "WEBP_MUX_NOT_FOUND", "WEBP_MUX_INVALID_ARGUMENT", "WEBP_MUX_BAD_DATA",
+  "WEBP_MUX_MEMORY_ERROR", "WEBP_MUX_NOT_ENOUGH_DATA"
+};
+
+static const char* ErrorString(WebPMuxError err) {
+  assert(err <= WEBP_MUX_NOT_FOUND && err >= WEBP_MUX_NOT_ENOUGH_DATA);
+  return kErrorMessages[-err];
+}
+
+enum {
+  METADATA_ICC  = (1 << 0),
+  METADATA_XMP  = (1 << 1),
+  METADATA_ALL  = METADATA_ICC | METADATA_XMP
+};
+
+//------------------------------------------------------------------------------
+
+static void Help(void) {
+  printf("Usage:\n");
+  printf(" gif2webp [options] gif_file -o webp_file\n");
+  printf("Options:\n");
+  printf("  -h / -help  ............ this help\n");
+  printf("  -lossy ................. encode image using lossy compression\n");
+  printf("  -mixed ................. for each frame in the image, pick lossy\n"
+         "                           or lossless compression heuristically\n");
+  printf("  -q <float> ............. quality factor (0:small..100:big)\n");
+  printf("  -m <int> ............... compression method (0=fast, 6=slowest)\n");
+  printf("  -kmin <int> ............ min distance between key frames\n");
+  printf("  -kmax <int> ............ max distance between key frames\n");
+  printf("  -f <int> ............... filter strength (0=off..100)\n");
+  printf("  -metadata <string> ..... comma separated list of metadata to\n");
+  printf("                           ");
+  printf("copy from the input to the output if present\n");
+  printf("                           "
+         "Valid values: all, none, icc, xmp (default)\n");
+  printf("  -mt .................... use multi-threading if available\n");
+  printf("\n");
+  printf("  -version ............... print version number and exit\n");
+  printf("  -v ..................... verbose\n");
+  printf("  -quiet ................. don't print anything\n");
+  printf("\n");
+}
+
+//------------------------------------------------------------------------------
+
+int main(int argc, const char *argv[]) {
+  int verbose = 0;
+  int gif_error = GIF_ERROR;
+  WebPMuxError err = WEBP_MUX_OK;
+  int ok = 0;
+  const char *in_file = NULL, *out_file = NULL;
+  FILE* out = NULL;
+  GifFileType* gif = NULL;
+  WebPConfig config;
+  WebPPicture frame;
+  WebPMuxFrameInfo info;
+  WebPMuxAnimParams anim = { WHITE_COLOR, 0 };
+  WebPFrameCache* cache = NULL;
+
+  int is_first_frame = 1;     // Whether we are processing the first frame.
+  int done;
+  int c;
+  int quiet = 0;
+  WebPMux* mux = NULL;
+  WebPData webp_data = { NULL, 0 };
+  int keep_metadata = METADATA_XMP;  // ICC not output by default.
+  int stored_icc = 0;  // Whether we have already stored an ICC profile.
+  int stored_xmp = 0;
+
+  int default_kmin = 1;  // Whether to use default kmin value.
+  int default_kmax = 1;
+  size_t kmin = 0;
+  size_t kmax = 0;
+  int allow_mixed = 0;   // If true, each frame can be lossy or lossless.
+
+  ResetFrameInfo(&info);
+
+  if (!WebPConfigInit(&config) || !WebPPictureInit(&frame)) {
+    fprintf(stderr, "Error! Version mismatch!\n");
+    return -1;
+  }
+  config.lossless = 1;  // Use lossless compression by default.
+  config.image_hint = WEBP_HINT_GRAPH;   // always low-color
+
+  if (argc == 1) {
+    Help();
+    return 0;
+  }
+
+  for (c = 1; c < argc; ++c) {
+    int parse_error = 0;
+    if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
+      Help();
+      return 0;
+    } else if (!strcmp(argv[c], "-o") && c < argc - 1) {
+      out_file = argv[++c];
+    } else if (!strcmp(argv[c], "-lossy")) {
+      config.lossless = 0;
+    } else if (!strcmp(argv[c], "-mixed")) {
+      allow_mixed = 1;
+      config.lossless = 0;
+    } else if (!strcmp(argv[c], "-q") && c < argc - 1) {
+      config.quality = ExUtilGetFloat(argv[++c], &parse_error);
+    } else if (!strcmp(argv[c], "-m") && c < argc - 1) {
+      config.method = ExUtilGetInt(argv[++c], 0, &parse_error);
+    } else if (!strcmp(argv[c], "-kmax") && c < argc - 1) {
+      kmax = ExUtilGetUInt(argv[++c], 0, &parse_error);
+      default_kmax = 0;
+    } else if (!strcmp(argv[c], "-kmin") && c < argc - 1) {
+      kmin = ExUtilGetUInt(argv[++c], 0, &parse_error);
+      default_kmin = 0;
+    } else if (!strcmp(argv[c], "-f") && c < argc - 1) {
+      config.filter_strength = ExUtilGetInt(argv[++c], 0, &parse_error);
+    } else if (!strcmp(argv[c], "-metadata") && c < argc - 1) {
+      static const struct {
+        const char* option;
+        int flag;
+      } kTokens[] = {
+        { "all",  METADATA_ALL },
+        { "none", 0 },
+        { "icc",  METADATA_ICC },
+        { "xmp",  METADATA_XMP },
+      };
+      const size_t kNumTokens = sizeof(kTokens) / sizeof(*kTokens);
+      const char* start = argv[++c];
+      const char* const end = start + strlen(start);
+
+      keep_metadata = 0;
+      while (start < end) {
+        size_t i;
+        const char* token = strchr(start, ',');
+        if (token == NULL) token = end;
+
+        for (i = 0; i < kNumTokens; ++i) {
+          if ((size_t)(token - start) == strlen(kTokens[i].option) &&
+              !strncmp(start, kTokens[i].option, strlen(kTokens[i].option))) {
+            if (kTokens[i].flag != 0) {
+              keep_metadata |= kTokens[i].flag;
+            } else {
+              keep_metadata = 0;
+            }
+            break;
+          }
+        }
+        if (i == kNumTokens) {
+          fprintf(stderr, "Error! Unknown metadata type '%.*s'\n",
+                  (int)(token - start), start);
+          Help();
+          return -1;
+        }
+        start = token + 1;
+      }
+    } else if (!strcmp(argv[c], "-mt")) {
+      ++config.thread_level;
+    } else if (!strcmp(argv[c], "-version")) {
+      const int enc_version = WebPGetEncoderVersion();
+      const int mux_version = WebPGetMuxVersion();
+      printf("WebP Encoder version: %d.%d.%d\nWebP Mux version: %d.%d.%d\n",
+             (enc_version >> 16) & 0xff, (enc_version >> 8) & 0xff,
+             enc_version & 0xff, (mux_version >> 16) & 0xff,
+             (mux_version >> 8) & 0xff, mux_version & 0xff);
+      return 0;
+    } else if (!strcmp(argv[c], "-quiet")) {
+      quiet = 1;
+    } else if (!strcmp(argv[c], "-v")) {
+      verbose = 1;
+    } else if (!strcmp(argv[c], "--")) {
+      if (c < argc - 1) in_file = argv[++c];
+      break;
+    } else if (argv[c][0] == '-') {
+      fprintf(stderr, "Error! Unknown option '%s'\n", argv[c]);
+      Help();
+      return -1;
+    } else {
+      in_file = argv[c];
+    }
+
+    if (parse_error) {
+      Help();
+      return -1;
+    }
+  }
+
+  // Appropriate default kmin, kmax values for lossy and lossless.
+  if (default_kmin) {
+    kmin = config.lossless ? 9 : 3;
+  }
+  if (default_kmax) {
+    kmax = config.lossless ? 17 : 5;
+  }
+  SanitizeKeyFrameIntervals(&kmin, &kmax);
+
+  if (!WebPValidateConfig(&config)) {
+    fprintf(stderr, "Error! Invalid configuration.\n");
+    goto End;
+  }
+
+  if (in_file == NULL) {
+    fprintf(stderr, "No input file specified!\n");
+    Help();
+    goto End;
+  }
+
+  // Start the decoder object
+#if LOCAL_GIF_PREREQ(5,0)
+  gif = DGifOpenFileName(in_file, &gif_error);
+#else
+  gif = DGifOpenFileName(in_file);
+#endif
+  if (gif == NULL) goto End;
+
+  mux = WebPMuxNew();
+  if (mux == NULL) {
+    fprintf(stderr, "ERROR: could not create a mux object.\n");
+    goto End;
+  }
+
+  // Loop over GIF images
+  done = 0;
+  do {
+    GifRecordType type;
+    if (DGifGetRecordType(gif, &type) == GIF_ERROR) goto End;
+
+    switch (type) {
+      case IMAGE_DESC_RECORD_TYPE: {
+        WebPFrameRect gif_rect;
+        GifImageDesc* const image_desc = &gif->Image;
+
+        if (!DGifGetImageDesc(gif)) goto End;
+
+        // Fix some broken GIF global headers that report
+        // 0 x 0 screen dimension.
+        if (is_first_frame) {
+          if (verbose) {
+            printf("Canvas screen: %d x %d\n", gif->SWidth, gif->SHeight);
+          }
+          if (gif->SWidth == 0 || gif->SHeight == 0) {
+            image_desc->Left = 0;
+            image_desc->Top = 0;
+            gif->SWidth = image_desc->Width;
+            gif->SHeight = image_desc->Height;
+            if (gif->SWidth <= 0 || gif->SHeight <= 0) {
+              goto End;
+            }
+            if (verbose) {
+              printf("Fixed canvas screen dimension to: %d x %d\n",
+                     gif->SWidth, gif->SHeight);
+            }
+          }
+#if WEBP_MUX_ABI_VERSION > 0x0101
+          // Set definitive canvas size.
+          err = WebPMuxSetCanvasSize(mux, gif->SWidth, gif->SHeight);
+          if (err != WEBP_MUX_OK) {
+            fprintf(stderr, "Invalid canvas size %d x %d\n",
+                    gif->SWidth, gif->SHeight);
+            goto End;
+          }
+#endif
+          // Allocate current buffer.
+          frame.width = gif->SWidth;
+          frame.height = gif->SHeight;
+          frame.use_argb = 1;
+          if (!WebPPictureAlloc(&frame)) goto End;
+          WebPUtilClearPic(&frame, NULL);
+
+          // Initialize cache.
+          cache = WebPFrameCacheNew(frame.width, frame.height,
+                                    kmin, kmax, allow_mixed);
+          if (cache == NULL) goto End;
+
+          // Background color.
+          GetBackgroundColor(gif->SColorMap, gif->SBackGroundColor,
+                             &anim.bgcolor);
+        }
+        // Some even more broken GIF can have sub-rect with zero width/height.
+        if (image_desc->Width == 0 || image_desc->Height == 0) {
+          image_desc->Width = gif->SWidth;
+          image_desc->Height = gif->SHeight;
+        }
+
+        if (!ReadFrame(gif, &gif_rect, &frame)) {
+          goto End;
+        }
+
+        if (!WebPFrameCacheAddFrame(cache, &config, &gif_rect, &frame, &info)) {
+          fprintf(stderr, "Error! Cannot encode frame as WebP\n");
+          fprintf(stderr, "Error code: %d\n", frame.error_code);
+        }
+
+        err = WebPFrameCacheFlush(cache, verbose, mux);
+        if (err != WEBP_MUX_OK) {
+          fprintf(stderr, "ERROR (%s): Could not add animation frame.\n",
+                  ErrorString(err));
+          goto End;
+        }
+        is_first_frame = 0;
+
+        // In GIF, graphic control extensions are optional for a frame, so we
+        // may not get one before reading the next frame. To handle this case,
+        // we reset frame properties to reasonable defaults for the next frame.
+        ResetFrameInfo(&info);
+        break;
+      }
+      case EXTENSION_RECORD_TYPE: {
+        int extension;
+        GifByteType *data = NULL;
+        if (DGifGetExtension(gif, &extension, &data) == GIF_ERROR) {
+          goto End;
+        }
+        switch (extension) {
+          case COMMENT_EXT_FUNC_CODE: {
+            break;  // Do nothing for now.
+          }
+          case GRAPHICS_EXT_FUNC_CODE: {
+            const int flags = data[1];
+            const int dispose = (flags >> GIF_DISPOSE_SHIFT) & GIF_DISPOSE_MASK;
+            const int delay = data[2] | (data[3] << 8);  // In 10 ms units.
+            if (data[0] != 4) goto End;
+            info.duration = delay * 10;  // Duration is in 1 ms units for WebP.
+            if (dispose == 3) {
+              static int warning_printed = 0;
+              if (!warning_printed) {
+                fprintf(stderr, "WARNING: GIF_DISPOSE_RESTORE unsupported.\n");
+                warning_printed = 1;
+              }
+              // failsafe. TODO(urvang): emulate the correct behaviour by
+              // recoding the whole frame.
+              info.dispose_method = WEBP_MUX_DISPOSE_BACKGROUND;
+            } else {
+              info.dispose_method =
+                  (dispose == 2) ? WEBP_MUX_DISPOSE_BACKGROUND
+                                 : WEBP_MUX_DISPOSE_NONE;
+            }
+            transparent_index = (flags & GIF_TRANSPARENT_MASK) ? data[4] : -1;
+            break;
+          }
+          case PLAINTEXT_EXT_FUNC_CODE: {
+            break;
+          }
+          case APPLICATION_EXT_FUNC_CODE: {
+            if (data[0] != 11) break;    // Chunk is too short
+            if (!memcmp(data + 1, "NETSCAPE2.0", 11) ||
+                !memcmp(data + 1, "ANIMEXTS1.0", 11)) {
+              // Recognize and parse Netscape2.0 NAB extension for loop count.
+              if (DGifGetExtensionNext(gif, &data) == GIF_ERROR) goto End;
+              if (data == NULL) goto End;  // Loop count sub-block missing.
+              if (data[0] < 3 || data[1] != 1) break;   // wrong size/marker
+              anim.loop_count = data[2] | (data[3] << 8);
+              if (verbose) {
+                fprintf(stderr, "Loop count: %d\n", anim.loop_count);
+              }
+            } else {  // An extension containing metadata.
+              // We only store the first encountered chunk of each type, and
+              // only if requested by the user.
+              const int is_xmp = (keep_metadata & METADATA_XMP) &&
+                                 !stored_xmp &&
+                                 !memcmp(data + 1, "XMP DataXMP", 11);
+              const int is_icc = (keep_metadata & METADATA_ICC) &&
+                                 !stored_icc &&
+                                 !memcmp(data + 1, "ICCRGBG1012", 11);
+              if (is_xmp || is_icc) {
+                const char* const fourccs[2] = { "XMP " , "ICCP" };
+                const char* const features[2] = { "XMP" , "ICC" };
+                WebPData metadata = { NULL, 0 };
+                // Construct metadata from sub-blocks.
+                // Usual case (including ICC profile): In each sub-block, the
+                // first byte specifies its size in bytes (0 to 255) and the
+                // rest of the bytes contain the data.
+                // Special case for XMP data: In each sub-block, the first byte
+                // is also part of the XMP payload. XMP in GIF also has a 257
+                // byte padding data. See the XMP specification for details.
+                while (1) {
+                  WebPData prev_metadata = metadata;
+                  WebPData subblock;
+                  if (DGifGetExtensionNext(gif, &data) == GIF_ERROR) {
+                    WebPDataClear(&metadata);
+                    goto End;
+                  }
+                  if (data == NULL) break;  // Finished.
+                  subblock.size = is_xmp ? data[0] + 1 : data[0];
+                  assert(subblock.size > 0);
+                  subblock.bytes = is_xmp ? data : data + 1;
+                  metadata.bytes =
+                      (uint8_t*)realloc((void*)metadata.bytes,
+                                        prev_metadata.size + subblock.size);
+                  if (metadata.bytes == NULL) {
+                    WebPDataClear(&prev_metadata);
+                    goto End;
+                  }
+                  metadata.size += subblock.size;
+                  memcpy((void*)(metadata.bytes + prev_metadata.size),
+                         subblock.bytes, subblock.size);
+                }
+                if (is_xmp) {
+                  // XMP padding data is 0x01, 0xff, 0xfe ... 0x01, 0x00.
+                  const size_t xmp_pading_size = 257;
+                  if (metadata.size > xmp_pading_size) {
+                    metadata.size -= xmp_pading_size;
+                  }
+                }
+
+                // Add metadata chunk.
+                err = WebPMuxSetChunk(mux, fourccs[is_icc], &metadata, 1);
+                if (verbose) {
+                  fprintf(stderr, "%s size: %d\n",
+                          features[is_icc], (int)metadata.size);
+                }
+                WebPDataClear(&metadata);
+                if (err != WEBP_MUX_OK) {
+                  fprintf(stderr, "ERROR (%s): Could not set %s chunk.\n",
+                          ErrorString(err), features[is_icc]);
+                  goto End;
+                }
+                if (is_icc) {
+                  stored_icc = 1;
+                } else if (is_xmp) {
+                  stored_xmp = 1;
+                }
+              }
+            }
+            break;
+          }
+          default: {
+            break;  // skip
+          }
+        }
+        while (data != NULL) {
+          if (DGifGetExtensionNext(gif, &data) == GIF_ERROR) goto End;
+        }
+        break;
+      }
+      case TERMINATE_RECORD_TYPE: {
+        done = 1;
+        break;
+      }
+      default: {
+        if (verbose) {
+          fprintf(stderr, "Skipping over unknown record type %d\n", type);
+        }
+        break;
+      }
+    }
+  } while (!done);
+
+  // Flush any pending frames.
+  err = WebPFrameCacheFlushAll(cache, verbose, mux);
+  if (err != WEBP_MUX_OK) {
+    fprintf(stderr, "ERROR (%s): Could not add animation frame.\n",
+            ErrorString(err));
+    goto End;
+  }
+
+  // Finish muxing
+  err = WebPMuxSetAnimationParams(mux, &anim);
+  if (err != WEBP_MUX_OK) {
+    fprintf(stderr, "ERROR (%s): Could not set animation parameters.\n",
+            ErrorString(err));
+    goto End;
+  }
+
+  err = WebPMuxAssemble(mux, &webp_data);
+  if (err != WEBP_MUX_OK) {
+    fprintf(stderr, "ERROR (%s) assembling the WebP file.\n", ErrorString(err));
+    goto End;
+  }
+  if (out_file != NULL) {
+    if (!ExUtilWriteFile(out_file, webp_data.bytes, webp_data.size)) {
+      fprintf(stderr, "Error writing output file: %s\n", out_file);
+      goto End;
+    }
+    if (!quiet) {
+      fprintf(stderr, "Saved output file: %s\n", out_file);
+    }
+  } else {
+    if (!quiet) {
+      fprintf(stderr, "Nothing written; use -o flag to save the result.\n");
+    }
+  }
+
+  // All OK.
+  ok = 1;
+  gif_error = GIF_OK;
+
+ End:
+  WebPDataClear(&webp_data);
+  WebPMuxDelete(mux);
+  WebPPictureFree(&frame);
+  WebPFrameCacheDelete(cache);
+  if (out != NULL && out_file != NULL) fclose(out);
+
+  if (gif_error != GIF_OK) {
+    DisplayGifError(gif, gif_error);
+  }
+  if (gif != NULL) {
+#if LOCAL_GIF_PREREQ(5,1)
+    DGifCloseFile(gif, &gif_error);
+#else
+    DGifCloseFile(gif);
+#endif
+  }
+
+  return !ok;
+}
+
+#else  // !WEBP_HAVE_GIF
+
+int main(int argc, const char *argv[]) {
+  fprintf(stderr, "GIF support not enabled in %s.\n", argv[0]);
+  (void)argc;
+  return 0;
+}
+
+#endif
+
+//------------------------------------------------------------------------------

examples/gif2webp_util.c

@@ -0,0 +1,708 @@
+// Copyright 2013 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//  Helper structs and methods for gif2webp tool.
+//
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "webp/encode.h"
+#include "./gif2webp_util.h"
+
+#define DELTA_INFINITY      1ULL << 32
+#define KEYFRAME_NONE       -1
+
+//------------------------------------------------------------------------------
+// Helper utilities.
+
+static void ClearRectangle(WebPPicture* const picture,
+                           int left, int top, int width, int height) {
+  int j;
+  for (j = top; j < top + height; ++j) {
+    uint32_t* const dst = picture->argb + j * picture->argb_stride;
+    int i;
+    for (i = left; i < left + width; ++i) {
+      dst[i] = WEBP_UTIL_TRANSPARENT_COLOR;
+    }
+  }
+}
+
+void WebPUtilClearPic(WebPPicture* const picture,
+                      const WebPFrameRect* const rect) {
+  if (rect != NULL) {
+    ClearRectangle(picture, rect->x_offset, rect->y_offset,
+                   rect->width, rect->height);
+  } else {
+    ClearRectangle(picture, 0, 0, picture->width, picture->height);
+  }
+}
+
+// TODO: Also used in picture.c. Move to a common location?
+// Copy width x height pixels from 'src' to 'dst' honoring the strides.
+static void CopyPlane(const uint8_t* src, int src_stride,
+                      uint8_t* dst, int dst_stride, int width, int height) {
+  while (height-- > 0) {
+    memcpy(dst, src, width);
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+// Copy pixels from 'src' to 'dst' honoring strides. 'src' and 'dst' are assumed
+// to be already allocated.
+static void CopyPixels(const WebPPicture* const src, WebPPicture* const dst) {
+  assert(src->width == dst->width && src->height == dst->height);
+  CopyPlane((uint8_t*)src->argb, 4 * src->argb_stride, (uint8_t*)dst->argb,
+            4 * dst->argb_stride, 4 * src->width, src->height);
+}
+
+// Given 'src' picture and its frame rectangle 'rect', blend it into 'dst'.
+static void BlendPixels(const WebPPicture* const src,
+                        const WebPFrameRect* const rect,
+                        WebPPicture* const dst) {
+  int j;
+  assert(src->width == dst->width && src->height == dst->height);
+  for (j = rect->y_offset; j < rect->y_offset + rect->height; ++j) {
+    int i;
+    for (i = rect->x_offset; i < rect->x_offset + rect->width; ++i) {
+      const uint32_t src_pixel = src->argb[j * src->argb_stride + i];
+      const int src_alpha = src_pixel >> 24;
+      if (src_alpha != 0) {
+        dst->argb[j * dst->argb_stride + i] = src_pixel;
+      }
+    }
+  }
+}
+
+// Replace transparent pixels within 'dst_rect' of 'dst' by those in the 'src'.
+static void ReduceTransparency(const WebPPicture* const src,
+                               const WebPFrameRect* const rect,
+                               WebPPicture* const dst) {
+  int i, j;
+  assert(src != NULL && dst != NULL && rect != NULL);
+  assert(src->width == dst->width && src->height == dst->height);
+  for (j = rect->y_offset; j < rect->y_offset + rect->height; ++j) {
+    for (i = rect->x_offset; i < rect->x_offset + rect->width; ++i) {
+      const uint32_t src_pixel = src->argb[j * src->argb_stride + i];
+      const int src_alpha = src_pixel >> 24;
+      const uint32_t dst_pixel = dst->argb[j * dst->argb_stride + i];
+      const int dst_alpha = dst_pixel >> 24;
+      if (dst_alpha == 0 && src_alpha == 0xff) {
+        dst->argb[j * dst->argb_stride + i] = src_pixel;
+      }
+    }
+  }
+}
+
+// Replace similar blocks of pixels by a 'see-through' transparent block
+// with uniform average color.
+static void FlattenSimilarBlocks(const WebPPicture* const src,
+                                 const WebPFrameRect* const rect,
+                                 WebPPicture* const dst) {
+  int i, j;
+  const int block_size = 8;
+  const int y_start = (rect->y_offset + block_size) & ~(block_size - 1);
+  const int y_end = (rect->y_offset + rect->height) & ~(block_size - 1);
+  const int x_start = (rect->x_offset + block_size) & ~(block_size - 1);
+  const int x_end = (rect->x_offset + rect->width) & ~(block_size - 1);
+  assert(src != NULL && dst != NULL && rect != NULL);
+  assert(src->width == dst->width && src->height == dst->height);
+  assert((block_size & (block_size - 1)) == 0);  // must be a power of 2
+  // Iterate over each block and count similar pixels.
+  for (j = y_start; j < y_end; j += block_size) {
+    for (i = x_start; i < x_end; i += block_size) {
+      int cnt = 0;
+      int avg_r = 0, avg_g = 0, avg_b = 0;
+      int x, y;
+      const uint32_t* const psrc = src->argb + j * src->argb_stride + i;
+      uint32_t* const pdst = dst->argb + j * dst->argb_stride + i;
+      for (y = 0; y < block_size; ++y) {
+        for (x = 0; x < block_size; ++x) {
+          const uint32_t src_pixel = psrc[x + y * src->argb_stride];
+          const int alpha = src_pixel >> 24;
+          if (alpha == 0xff &&
+              src_pixel == pdst[x + y * dst->argb_stride]) {
+              ++cnt;
+              avg_r += (src_pixel >> 16) & 0xff;
+              avg_g += (src_pixel >>  8) & 0xff;
+              avg_b += (src_pixel >>  0) & 0xff;
+          }
+        }
+      }
+      // If we have a fully similar block, we replace it with an
+      // average transparent block. This compresses better in lossy mode.
+      if (cnt == block_size * block_size) {
+        const uint32_t color = (0x00          << 24) |
+                               ((avg_r / cnt) << 16) |
+                               ((avg_g / cnt) <<  8) |
+                               ((avg_b / cnt) <<  0);
+        for (y = 0; y < block_size; ++y) {
+          for (x = 0; x < block_size; ++x) {
+            pdst[x + y * dst->argb_stride] = color;
+          }
+        }
+      }
+    }
+  }
+}
+
+//------------------------------------------------------------------------------
+// Key frame related utilities.
+
+// Returns true if 'curr' frame with frame rectangle 'curr_rect' is a key frame,
+// that is, it can be decoded independently of 'prev' canvas.
+static int IsKeyFrame(const WebPPicture* const curr,
+                      const WebPFrameRect* const curr_rect,
+                      const WebPPicture* const prev) {
+  int i, j;
+  int is_key_frame = 1;
+
+  // If previous canvas (with previous frame disposed) is all transparent,
+  // current frame is a key frame.
+  for (j = 0; j < prev->height; ++j) {
+    const uint32_t* const row = &prev->argb[j * prev->argb_stride];
+    for (i = 0; i < prev->width; ++i) {
+      if (row[i] & 0xff000000u) {   // has alpha?
+        is_key_frame = 0;
+        break;
+      }
+    }
+    if (!is_key_frame) break;
+  }
+  if (is_key_frame) return 1;
+
+  // If current frame covers the whole canvas and does not contain any
+  // transparent pixels that depend on previous canvas, then current frame is
+  // a key frame.
+  if (curr_rect->width == curr->width && curr_rect->height == curr->height) {
+    assert(curr_rect->x_offset == 0 && curr_rect->y_offset == 0);
+    is_key_frame = 1;
+    for (j = 0; j < prev->height; ++j) {
+      for (i = 0; i < prev->width; ++i) {
+        const uint32_t prev_alpha =
+            (prev->argb[j * prev->argb_stride + i]) >> 24;
+        const uint32_t curr_alpha =
+            (curr->argb[j * curr->argb_stride + i]) >> 24;
+        if (curr_alpha != 0xff && prev_alpha != 0) {
+          is_key_frame = 0;
+          break;
+        }
+      }
+      if (!is_key_frame) break;
+    }
+    if (is_key_frame) return 1;
+  }
+
+  return 0;
+}
+
+// Given 'prev' frame and current frame rectangle 'rect', convert 'curr' frame
+// to a key frame.
+static void ConvertToKeyFrame(const WebPPicture* const prev,
+                              WebPFrameRect* const rect,
+                              WebPPicture* const curr) {
+  int j;
+  assert(curr->width == prev->width && curr->height == prev->height);
+
+  // Replace transparent pixels of current canvas with those from previous
+  // canvas (with previous frame disposed).
+  for (j = 0; j < curr->height; ++j) {
+    int i;
+    for (i = 0; i < curr->width; ++i) {
+      uint32_t* const curr_pixel = curr->argb + j * curr->argb_stride + i;
+      const int curr_alpha = *curr_pixel >> 24;
+      if (curr_alpha == 0) {
+        *curr_pixel = prev->argb[j * prev->argb_stride + i];
+      }
+    }
+  }
+
+  // Frame rectangle now covers the whole canvas.
+  rect->x_offset = 0;
+  rect->y_offset = 0;
+  rect->width = curr->width;
+  rect->height = curr->height;
+}
+
+//------------------------------------------------------------------------------
+// Encoded frame.
+
+// Used to store two candidates of encoded data for an animation frame. One of
+// the two will be chosen later.
+typedef struct {
+  WebPMuxFrameInfo sub_frame;  // Encoded frame rectangle.
+  WebPMuxFrameInfo key_frame;  // Encoded frame if it was converted to keyframe.
+} EncodedFrame;
+
+// Release the data contained by 'encoded_frame'.
+static void FrameRelease(EncodedFrame* const encoded_frame) {
+  if (encoded_frame != NULL) {
+    WebPDataClear(&encoded_frame->sub_frame.bitstream);
+    WebPDataClear(&encoded_frame->key_frame.bitstream);
+    memset(encoded_frame, 0, sizeof(*encoded_frame));
+  }
+}
+
+//------------------------------------------------------------------------------
+// Frame cache.
+
+// Used to store encoded frames that haven't been output yet.
+struct WebPFrameCache {
+  EncodedFrame* encoded_frames;  // Array of encoded frames.
+  size_t size;               // Number of allocated data elements.
+  size_t start;              // Start index.
+  size_t count;              // Number of valid data elements.
+  int flush_count;           // If >0, ‘flush_count’ frames starting from
+                             // 'start' are ready to be added to mux.
+  int64_t best_delta;        // min(canvas size - frame size) over the frames.
+                             // Can be negative in certain cases due to
+                             // transparent pixels in a frame.
+  int keyframe;              // Index of selected keyframe relative to 'start'.
+
+  size_t kmin;                   // Min distance between key frames.
+  size_t kmax;                   // Max distance between key frames.
+  size_t count_since_key_frame;  // Frames seen since the last key frame.
+  int allow_mixed;           // If true, each frame can be lossy or lossless.
+  WebPPicture prev_canvas;   // Previous canvas (properly disposed).
+  WebPPicture curr_canvas;   // Current canvas (temporary buffer).
+  int is_first_frame;        // True if no frames have been added to the cache
+                             // since WebPFrameCacheNew().
+};
+
+// Reset the counters in the cache struct. Doesn't touch 'cache->encoded_frames'
+// and 'cache->size'.
+static void CacheReset(WebPFrameCache* const cache) {
+  cache->start = 0;
+  cache->count = 0;
+  cache->flush_count = 0;
+  cache->best_delta = DELTA_INFINITY;
+  cache->keyframe = KEYFRAME_NONE;
+}
+
+WebPFrameCache* WebPFrameCacheNew(int width, int height,
+                                  size_t kmin, size_t kmax, int allow_mixed) {
+  WebPFrameCache* cache = (WebPFrameCache*)malloc(sizeof(*cache));
+  if (cache == NULL) return NULL;
+  CacheReset(cache);
+  // sanity init, so we can call WebPFrameCacheDelete():
+  cache->encoded_frames = NULL;
+
+  cache->is_first_frame = 1;
+
+  // Picture buffers.
+  if (!WebPPictureInit(&cache->prev_canvas) ||
+      !WebPPictureInit(&cache->curr_canvas)) {
+    return NULL;
+  }
+  cache->prev_canvas.width = width;
+  cache->prev_canvas.height = height;
+  cache->prev_canvas.use_argb = 1;
+  if (!WebPPictureAlloc(&cache->prev_canvas) ||
+      !WebPPictureCopy(&cache->prev_canvas, &cache->curr_canvas)) {
+    goto Err;
+  }
+  WebPUtilClearPic(&cache->prev_canvas, NULL);
+
+  // Cache data.
+  cache->allow_mixed = allow_mixed;
+  cache->kmin = kmin;
+  cache->kmax = kmax;
+  cache->count_since_key_frame = 0;
+  assert(kmax > kmin);
+  cache->size = kmax - kmin;
+  cache->encoded_frames =
+      (EncodedFrame*)calloc(cache->size, sizeof(*cache->encoded_frames));
+  if (cache->encoded_frames == NULL) goto Err;
+
+  return cache;  // All OK.
+
+ Err:
+  WebPFrameCacheDelete(cache);
+  return NULL;
+}
+
+void WebPFrameCacheDelete(WebPFrameCache* const cache) {
+  if (cache != NULL) {
+    if (cache->encoded_frames != NULL) {
+      size_t i;
+      for (i = 0; i < cache->size; ++i) {
+        FrameRelease(&cache->encoded_frames[i]);
+      }
+      free(cache->encoded_frames);
+    }
+    WebPPictureFree(&cache->prev_canvas);
+    WebPPictureFree(&cache->curr_canvas);
+    free(cache);
+  }
+}
+
+static int EncodeFrame(const WebPConfig* const config, WebPPicture* const pic,
+                       WebPMemoryWriter* const memory) {
+  pic->use_argb = 1;
+  pic->writer = WebPMemoryWrite;
+  pic->custom_ptr = memory;
+  if (!WebPEncode(config, pic)) {
+    return 0;
+  }
+  return 1;
+}
+
+static void GetEncodedData(const WebPMemoryWriter* const memory,
+                           WebPData* const encoded_data) {
+  encoded_data->bytes = memory->mem;
+  encoded_data->size  = memory->size;
+}
+
+#define MIN_COLORS_LOSSY     31  // Don't try lossy below this threshold.
+#define MAX_COLORS_LOSSLESS 194  // Don't try lossless above this threshold.
+#define MAX_COLOR_COUNT     256  // Power of 2 greater than MAX_COLORS_LOSSLESS.
+#define HASH_SIZE (MAX_COLOR_COUNT * 4)
+#define HASH_RIGHT_SHIFT     22  // 32 - log2(HASH_SIZE).
+
+// TODO(urvang): Also used in enc/vp8l.c. Move to utils.
+// If the number of colors in the 'pic' is at least MAX_COLOR_COUNT, return
+// MAX_COLOR_COUNT. Otherwise, return the exact number of colors in the 'pic'.
+static int GetColorCount(const WebPPicture* const pic) {
+  int x, y;
+  int num_colors = 0;
+  uint8_t in_use[HASH_SIZE] = { 0 };
+  uint32_t colors[HASH_SIZE];
+  static const uint32_t kHashMul = 0x1e35a7bd;
+  const uint32_t* argb = pic->argb;
+  const int width = pic->width;
+  const int height = pic->height;
+  uint32_t last_pix = ~argb[0];   // so we're sure that last_pix != argb[0]
+
+  for (y = 0; y < height; ++y) {
+    for (x = 0; x < width; ++x) {
+      int key;
+      if (argb[x] == last_pix) {
+        continue;
+      }
+      last_pix = argb[x];
+      key = (kHashMul * last_pix) >> HASH_RIGHT_SHIFT;
+      while (1) {
+        if (!in_use[key]) {
+          colors[key] = last_pix;
+          in_use[key] = 1;
+          ++num_colors;
+          if (num_colors >= MAX_COLOR_COUNT) {
+            return MAX_COLOR_COUNT;  // Exact count not needed.
+          }
+          break;
+        } else if (colors[key] == last_pix) {
+          break;  // The color is already there.
+        } else {
+          // Some other color sits here, so do linear conflict resolution.
+          ++key;
+          key &= (HASH_SIZE - 1);  // Key mask.
+        }
+      }
+    }
+    argb += pic->argb_stride;
+  }
+  return num_colors;
+}
+
+#undef MAX_COLOR_COUNT
+#undef HASH_SIZE
+#undef HASH_RIGHT_SHIFT
+
+static WebPEncodingError SetFrame(const WebPConfig* const config,
+                                  int allow_mixed, int is_key_frame,
+                                  const WebPPicture* const prev_canvas,
+                                  WebPPicture* const frame,
+                                  const WebPFrameRect* const rect,
+                                  const WebPMuxFrameInfo* const info,
+                                  WebPPicture* const sub_frame,
+                                  EncodedFrame* encoded_frame) {
+  WebPEncodingError error_code = VP8_ENC_OK;
+  int try_lossless;
+  int try_lossy;
+  int try_both;
+  WebPMemoryWriter mem1, mem2;
+  WebPData* encoded_data;
+  WebPMuxFrameInfo* const dst =
+      is_key_frame ? &encoded_frame->key_frame : &encoded_frame->sub_frame;
+  *dst = *info;
+  encoded_data = &dst->bitstream;
+  WebPMemoryWriterInit(&mem1);
+  WebPMemoryWriterInit(&mem2);
+
+  if (!allow_mixed) {
+    try_lossless = config->lossless;
+    try_lossy = !try_lossless;
+  } else {  // Use a heuristic for trying lossless and/or lossy compression.
+    const int num_colors = GetColorCount(sub_frame);
+    try_lossless = (num_colors < MAX_COLORS_LOSSLESS);
+    try_lossy = (num_colors >= MIN_COLORS_LOSSY);
+  }
+  try_both = try_lossless && try_lossy;
+
+  if (try_lossless) {
+    WebPConfig config_ll = *config;
+    config_ll.lossless = 1;
+    if (!EncodeFrame(&config_ll, sub_frame, &mem1)) {
+      error_code = sub_frame->error_code;
+      goto Err;
+    }
+  }
+
+  if (try_lossy) {
+    WebPConfig config_lossy = *config;
+    config_lossy.lossless = 0;
+    if (!is_key_frame) {
+      // For lossy compression of a frame, it's better to replace transparent
+      // pixels of 'curr' with actual RGB values, whenever possible.
+      ReduceTransparency(prev_canvas, rect, frame);
+      // TODO(later): Investigate if this helps lossless compression as well.
+      FlattenSimilarBlocks(prev_canvas, rect, frame);
+    }
+    if (!EncodeFrame(&config_lossy, sub_frame, &mem2)) {
+      error_code = sub_frame->error_code;
+      goto Err;
+    }
+  }
+
+  if (try_both) {  // Pick the encoding with smallest size.
+    // TODO(later): Perhaps a rough SSIM/PSNR produced by the encoder should
+    // also be a criteria, in addition to sizes.
+    if (mem1.size <= mem2.size) {
+#if WEBP_ENCODER_ABI_VERSION > 0x0203
+      WebPMemoryWriterClear(&mem2);
+#else
+      free(mem2.mem);
+      memset(&mem2, 0, sizeof(mem2));
+#endif
+      GetEncodedData(&mem1, encoded_data);
+    } else {
+#if WEBP_ENCODER_ABI_VERSION > 0x0203
+      WebPMemoryWriterClear(&mem1);
+#else
+      free(mem1.mem);
+      memset(&mem1, 0, sizeof(mem1));
+#endif
+      GetEncodedData(&mem2, encoded_data);
+    }
+  } else {
+    GetEncodedData(try_lossless ? &mem1 : &mem2, encoded_data);
+  }
+  return error_code;
+
+ Err:
+#if WEBP_ENCODER_ABI_VERSION > 0x0203
+  WebPMemoryWriterClear(&mem1);
+  WebPMemoryWriterClear(&mem2);
+#else
+  free(mem1.mem);
+  free(mem2.mem);
+#endif
+  return error_code;
+}
+
+#undef MIN_COLORS_LOSSY
+#undef MAX_COLORS_LOSSLESS
+
+// Returns cached frame at given 'position' index.
+static EncodedFrame* CacheGetFrame(const WebPFrameCache* const cache,
+                                   size_t position) {
+  assert(cache->start + position < cache->size);
+  return &cache->encoded_frames[cache->start + position];
+}
+
+// Calculate the penalty incurred if we encode given frame as a key frame
+// instead of a sub-frame.
+static int64_t KeyFramePenalty(const EncodedFrame* const encoded_frame) {
+  return ((int64_t)encoded_frame->key_frame.bitstream.size -
+          encoded_frame->sub_frame.bitstream.size);
+}
+
+static void DisposeFrame(WebPMuxAnimDispose dispose_method,
+                         const WebPFrameRect* const gif_rect,
+                         WebPPicture* const frame, WebPPicture* const canvas) {
+  if (dispose_method == WEBP_MUX_DISPOSE_BACKGROUND) {
+    WebPUtilClearPic(frame, NULL);
+    WebPUtilClearPic(canvas, gif_rect);
+  }
+}
+
+int WebPFrameCacheAddFrame(WebPFrameCache* const cache,
+                           const WebPConfig* const config,
+                           const WebPFrameRect* const orig_rect_ptr,
+                           WebPPicture* const frame,
+                           WebPMuxFrameInfo* const info) {
+  int ok = 0;
+  WebPEncodingError error_code = VP8_ENC_OK;
+  WebPFrameRect rect;
+  WebPPicture sub_image;  // View extracted from 'frame' with rectangle 'rect'.
+  WebPPicture* const prev_canvas = &cache->prev_canvas;
+  const size_t position = cache->count;
+  const int allow_mixed = cache->allow_mixed;
+  EncodedFrame* const encoded_frame = CacheGetFrame(cache, position);
+  WebPFrameRect orig_rect;
+  assert(position < cache->size);
+
+  if (frame == NULL || info == NULL) {
+    return 0;
+  }
+
+  if (orig_rect_ptr == NULL) {
+    orig_rect.width = frame->width;
+    orig_rect.height = frame->height;
+    orig_rect.x_offset = 0;
+    orig_rect.y_offset = 0;
+  } else {
+    orig_rect = *orig_rect_ptr;
+  }
+
+  // Snap to even offsets (and adjust dimensions if needed).
+  rect = orig_rect;
+  rect.width += (rect.x_offset & 1);
+  rect.height += (rect.y_offset & 1);
+  rect.x_offset &= ~1;
+  rect.y_offset &= ~1;
+
+  if (!WebPPictureView(frame, rect.x_offset, rect.y_offset,
+                       rect.width, rect.height, &sub_image)) {
+    return 0;
+  }
+  info->x_offset = rect.x_offset;
+  info->y_offset = rect.y_offset;
+
+  ++cache->count;
+
+  if (cache->is_first_frame || IsKeyFrame(frame, &rect, prev_canvas)) {
+    // Add this as a key frame.
+    error_code = SetFrame(config, allow_mixed, 1, NULL, NULL, NULL,
+                          info, &sub_image, encoded_frame);
+    if (error_code != VP8_ENC_OK) {
+      goto End;
+    }
+    cache->keyframe = position;
+    cache->flush_count = cache->count;
+    cache->count_since_key_frame = 0;
+    // Update prev_canvas by simply copying from 'curr'.
+    CopyPixels(frame, prev_canvas);
+  } else {
+    ++cache->count_since_key_frame;
+    if (cache->count_since_key_frame <= cache->kmin) {
+      // Add this as a frame rectangle.
+      error_code = SetFrame(config, allow_mixed, 0, prev_canvas, frame,
+                            &rect, info, &sub_image, encoded_frame);
+      if (error_code != VP8_ENC_OK) {
+        goto End;
+      }
+      cache->flush_count = cache->count;
+      // Update prev_canvas by blending 'curr' into it.
+      BlendPixels(frame, &orig_rect, prev_canvas);
+    } else {
+      WebPPicture full_image;
+      WebPMuxFrameInfo full_image_info;
+      int64_t curr_delta;
+
+      // Add frame rectangle to cache.
+      error_code = SetFrame(config, allow_mixed, 0, prev_canvas, frame, &rect,
+                            info, &sub_image, encoded_frame);
+      if (error_code != VP8_ENC_OK) {
+        goto End;
+      }
+
+      // Convert to a key frame.
+      CopyPixels(frame, &cache->curr_canvas);
+      ConvertToKeyFrame(prev_canvas, &rect, &cache->curr_canvas);
+      if (!WebPPictureView(&cache->curr_canvas, rect.x_offset, rect.y_offset,
+                           rect.width, rect.height, &full_image)) {
+        goto End;
+      }
+      full_image_info = *info;
+      full_image_info.x_offset = rect.x_offset;
+      full_image_info.y_offset = rect.y_offset;
+
+      // Add key frame to cache, too.
+      error_code = SetFrame(config, allow_mixed, 1, NULL, NULL, NULL,
+                            &full_image_info, &full_image, encoded_frame);
+      WebPPictureFree(&full_image);
+      if (error_code != VP8_ENC_OK) goto End;
+
+      // Analyze size difference of the two variants.
+      curr_delta = KeyFramePenalty(encoded_frame);
+      if (curr_delta <= cache->best_delta) {  // Pick this as keyframe.
+        cache->keyframe = position;
+        cache->best_delta = curr_delta;
+        cache->flush_count = cache->count - 1;  // We can flush previous frames.
+      }
+      if (cache->count_since_key_frame == cache->kmax) {
+        cache->flush_count = cache->count;
+        cache->count_since_key_frame = 0;
+      }
+
+      // Update prev_canvas by simply copying from 'curr_canvas'.
+      CopyPixels(&cache->curr_canvas, prev_canvas);
+    }
+  }
+
+  DisposeFrame(info->dispose_method, &orig_rect, frame, prev_canvas);
+
+  cache->is_first_frame = 0;
+  ok = 1;
+
+ End:
+  WebPPictureFree(&sub_image);
+  if (!ok) {
+    FrameRelease(encoded_frame);
+    --cache->count;  // We reset the count, as the frame addition failed.
+  }
+  frame->error_code = error_code;   // report error_code
+  assert(ok || error_code != VP8_ENC_OK);
+  return ok;
+}
+
+WebPMuxError WebPFrameCacheFlush(WebPFrameCache* const cache, int verbose,
+                                 WebPMux* const mux) {
+  while (cache->flush_count > 0) {
+    WebPMuxFrameInfo* info;
+    WebPMuxError err;
+    EncodedFrame* const curr = CacheGetFrame(cache, 0);
+    // Pick frame or full canvas.
+    if (cache->keyframe == 0) {
+      info = &curr->key_frame;
+      info->blend_method = WEBP_MUX_NO_BLEND;
+      cache->keyframe = KEYFRAME_NONE;
+      cache->best_delta = DELTA_INFINITY;
+    } else {
+      info = &curr->sub_frame;
+      info->blend_method = WEBP_MUX_BLEND;
+    }
+    // Add to mux.
+    err = WebPMuxPushFrame(mux, info, 1);
+    if (err != WEBP_MUX_OK) return err;
+    if (verbose) {
+      printf("Added frame. offset:%d,%d duration:%d dispose:%d blend:%d\n",
+             info->x_offset, info->y_offset, info->duration,
+             info->dispose_method, info->blend_method);
+    }
+    FrameRelease(curr);
+    ++cache->start;
+    --cache->flush_count;
+    --cache->count;
+    if (cache->keyframe != KEYFRAME_NONE) --cache->keyframe;
+  }
+
+  if (cache->count == 0) CacheReset(cache);
+  return WEBP_MUX_OK;
+}
+
+WebPMuxError WebPFrameCacheFlushAll(WebPFrameCache* const cache, int verbose,
+                                    WebPMux* const mux) {
+  cache->flush_count = cache->count;  // Force flushing of all frames.
+  return WebPFrameCacheFlush(cache, verbose, mux);
+}
+
+//------------------------------------------------------------------------------

examples/gif2webp_util.h

@@ -0,0 +1,81 @@
+// Copyright 2013 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//  Helper structs and methods for gif2webp tool.
+//
+// Author: Urvang (urvang@google.com)
+
+#ifndef WEBP_EXAMPLES_GIF2WEBP_UTIL_H_
+#define WEBP_EXAMPLES_GIF2WEBP_UTIL_H_
+
+#include <stdlib.h>
+
+#include "webp/mux.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+//------------------------------------------------------------------------------
+// Helper utilities.
+
+#define WEBP_UTIL_TRANSPARENT_COLOR 0x00ffffff
+
+struct WebPPicture;
+
+typedef struct {
+  int x_offset, y_offset, width, height;
+} WebPFrameRect;
+
+// Clear pixels in 'picture' within given 'rect' to transparent color.
+void WebPUtilClearPic(struct WebPPicture* const picture,
+                      const WebPFrameRect* const rect);
+
+//------------------------------------------------------------------------------
+// Frame cache.
+
+typedef struct WebPFrameCache WebPFrameCache;
+
+// Given the minimum distance between key frames 'kmin' and maximum distance
+// between key frames 'kmax', returns an appropriately allocated cache object.
+// If 'allow_mixed' is true, the subsequent calls to WebPFrameCacheAddFrame()
+// will heuristically pick lossy or lossless compression for each frame.
+// Use WebPFrameCacheDelete() to deallocate the 'cache'.
+WebPFrameCache* WebPFrameCacheNew(int width, int height,
+                                  size_t kmin, size_t kmax, int allow_mixed);
+
+// Release all the frame data from 'cache' and free 'cache'.
+void WebPFrameCacheDelete(WebPFrameCache* const cache);
+
+// Given an image described by 'frame', 'info' and 'orig_rect', optimize it for
+// WebP, encode it and add it to 'cache'. 'orig_rect' can be NULL.
+// This takes care of frame disposal too, according to 'info->dispose_method'.
+// Returns false in case of error (and sets frame->error_code accordingly).
+int WebPFrameCacheAddFrame(WebPFrameCache* const cache,
+                           const WebPConfig* const config,
+                           const WebPFrameRect* const orig_rect,
+                           WebPPicture* const frame,
+                           WebPMuxFrameInfo* const info);
+
+// Flush the *ready* frames from cache and add them to 'mux'. If 'verbose' is
+// true, prints the information about these frames.
+WebPMuxError WebPFrameCacheFlush(WebPFrameCache* const cache, int verbose,
+                                 WebPMux* const mux);
+
+// Similar to 'WebPFrameCacheFlushFrames()', but flushes *all* the frames.
+WebPMuxError WebPFrameCacheFlushAll(WebPFrameCache* const cache, int verbose,
+                                    WebPMux* const mux);
+
+//------------------------------------------------------------------------------
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_EXAMPLES_GIF2WEBP_UTIL_H_

examples/jpegdec.c

@@ -0,0 +1,295 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// JPEG decode.
+
+#include "./jpegdec.h"
+
+#ifdef HAVE_CONFIG_H
+#include "webp/config.h"
+#endif
+
+#include <stdio.h>
+
+#ifdef WEBP_HAVE_JPEG
+#include <jpeglib.h>
+#include <setjmp.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "webp/encode.h"
+#include "./metadata.h"
+
+// -----------------------------------------------------------------------------
+// Metadata processing
+
+#ifndef JPEG_APP1
+# define JPEG_APP1 (JPEG_APP0 + 1)
+#endif
+#ifndef JPEG_APP2
+# define JPEG_APP2 (JPEG_APP0 + 2)
+#endif
+
+typedef struct {
+  const uint8_t* data;
+  size_t data_length;
+  int seq;  // this segment's sequence number [1, 255] for use in reassembly.
+} ICCPSegment;
+
+static void SaveMetadataMarkers(j_decompress_ptr dinfo) {
+  const unsigned int max_marker_length = 0xffff;
+  jpeg_save_markers(dinfo, JPEG_APP1, max_marker_length);  // Exif/XMP
+  jpeg_save_markers(dinfo, JPEG_APP2, max_marker_length);  // ICC profile
+}
+
+static int CompareICCPSegments(const void* a, const void* b) {
+  const ICCPSegment* s1 = (const ICCPSegment*)a;
+  const ICCPSegment* s2 = (const ICCPSegment*)b;
+  return s1->seq - s2->seq;
+}
+
+// Extract ICC profile segments from the marker list in 'dinfo', reassembling
+// and storing them in 'iccp'.
+// Returns true on success and false for memory errors and corrupt profiles.
+static int StoreICCP(j_decompress_ptr dinfo, MetadataPayload* const iccp) {
+  // ICC.1:2010-12 (4.3.0.0) Annex B.4 Embedding ICC Profiles in JPEG files
+  static const char kICCPSignature[] = "ICC_PROFILE";
+  static const size_t kICCPSignatureLength = 12;  // signature includes '\0'
+  static const size_t kICCPSkipLength = 14;  // signature + seq & count
+  int expected_count = 0;
+  int actual_count = 0;
+  int seq_max = 0;
+  size_t total_size = 0;
+  ICCPSegment iccp_segments[255];
+  jpeg_saved_marker_ptr marker;
+
+  memset(iccp_segments, 0, sizeof(iccp_segments));
+  for (marker = dinfo->marker_list; marker != NULL; marker = marker->next) {
+    if (marker->marker == JPEG_APP2 &&
+        marker->data_length > kICCPSkipLength &&
+        !memcmp(marker->data, kICCPSignature, kICCPSignatureLength)) {
+      // ICC_PROFILE\0<seq><count>; 'seq' starts at 1.
+      const int seq = marker->data[kICCPSignatureLength];
+      const int count = marker->data[kICCPSignatureLength + 1];
+      const size_t segment_size = marker->data_length - kICCPSkipLength;
+      ICCPSegment* segment;
+
+      if (segment_size == 0 || count == 0 || seq == 0) {
+        fprintf(stderr, "[ICCP] size (%d) / count (%d) / sequence number (%d)"
+                        " cannot be 0!\n",
+                (int)segment_size, seq, count);
+        return 0;
+      }
+
+      if (expected_count == 0) {
+        expected_count = count;
+      } else if (expected_count != count) {
+        fprintf(stderr, "[ICCP] Inconsistent segment count (%d / %d)!\n",
+                expected_count, count);
+        return 0;
+      }
+
+      segment = iccp_segments + seq - 1;
+      if (segment->data_length != 0) {
+        fprintf(stderr, "[ICCP] Duplicate segment number (%d)!\n" , seq);
+        return 0;
+      }
+
+      segment->data = marker->data + kICCPSkipLength;
+      segment->data_length = segment_size;
+      segment->seq = seq;
+      total_size += segment_size;
+      if (seq > seq_max) seq_max = seq;
+      ++actual_count;
+    }
+  }
+
+  if (actual_count == 0) return 1;
+  if (seq_max != actual_count) {
+    fprintf(stderr, "[ICCP] Discontinuous segments, expected: %d actual: %d!\n",
+            actual_count, seq_max);
+    return 0;
+  }
+  if (expected_count != actual_count) {
+    fprintf(stderr, "[ICCP] Segment count: %d does not match expected: %d!\n",
+            actual_count, expected_count);
+    return 0;
+  }
+
+  // The segments may appear out of order in the file, sort them based on
+  // sequence number before assembling the payload.
+  qsort(iccp_segments, actual_count, sizeof(*iccp_segments),
+        CompareICCPSegments);
+
+  iccp->bytes = (uint8_t*)malloc(total_size);
+  if (iccp->bytes == NULL) return 0;
+  iccp->size = total_size;
+
+  {
+    int i;
+    size_t offset = 0;
+    for (i = 0; i < seq_max; ++i) {
+      memcpy(iccp->bytes + offset,
+             iccp_segments[i].data, iccp_segments[i].data_length);
+      offset += iccp_segments[i].data_length;
+    }
+  }
+  return 1;
+}
+
+// Returns true on success and false for memory errors and corrupt profiles.
+// The caller must use MetadataFree() on 'metadata' in all cases.
+static int ExtractMetadataFromJPEG(j_decompress_ptr dinfo,
+                                   Metadata* const metadata) {
+  static const struct {
+    int marker;
+    const char* signature;
+    size_t signature_length;
+    size_t storage_offset;
+  } kJPEGMetadataMap[] = {
+    // Exif 2.2 Section 4.7.2 Interoperability Structure of APP1 ...
+    { JPEG_APP1, "Exif\0",                        6, METADATA_OFFSET(exif) },
+    // XMP Specification Part 3 Section 3 Embedding XMP Metadata ... #JPEG
+    // TODO(jzern) Add support for 'ExtendedXMP'
+    { JPEG_APP1, "http://ns.adobe.com/xap/1.0/", 29, METADATA_OFFSET(xmp) },
+    { 0, NULL, 0, 0 },
+  };
+  jpeg_saved_marker_ptr marker;
+  // Treat ICC profiles separately as they may be segmented and out of order.
+  if (!StoreICCP(dinfo, &metadata->iccp)) return 0;
+
+  for (marker = dinfo->marker_list; marker != NULL; marker = marker->next) {
+    int i;
+    for (i = 0; kJPEGMetadataMap[i].marker != 0; ++i) {
+      if (marker->marker == kJPEGMetadataMap[i].marker &&
+          marker->data_length > kJPEGMetadataMap[i].signature_length &&
+          !memcmp(marker->data, kJPEGMetadataMap[i].signature,
+                  kJPEGMetadataMap[i].signature_length)) {
+        MetadataPayload* const payload =
+            (MetadataPayload*)((uint8_t*)metadata +
+                               kJPEGMetadataMap[i].storage_offset);
+
+        if (payload->bytes == NULL) {
+          const char* marker_data = (const char*)marker->data +
+                                    kJPEGMetadataMap[i].signature_length;
+          const size_t marker_data_length =
+              marker->data_length - kJPEGMetadataMap[i].signature_length;
+          if (!MetadataCopy(marker_data, marker_data_length, payload)) return 0;
+        } else {
+          fprintf(stderr, "Ignoring additional '%s' marker\n",
+                  kJPEGMetadataMap[i].signature);
+        }
+      }
+    }
+  }
+  return 1;
+}
+
+#undef JPEG_APP1
+#undef JPEG_APP2
+
+// -----------------------------------------------------------------------------
+// JPEG decoding
+
+struct my_error_mgr {
+  struct jpeg_error_mgr pub;
+  jmp_buf setjmp_buffer;
+};
+
+static void my_error_exit(j_common_ptr dinfo) {
+  struct my_error_mgr* myerr = (struct my_error_mgr*)dinfo->err;
+  dinfo->err->output_message(dinfo);
+  longjmp(myerr->setjmp_buffer, 1);
+}
+
+int ReadJPEG(FILE* in_file, WebPPicture* const pic, Metadata* const metadata) {
+  int ok = 0;
+  int stride, width, height;
+  volatile struct jpeg_decompress_struct dinfo;
+  struct my_error_mgr jerr;
+  uint8_t* volatile rgb = NULL;
+  JSAMPROW buffer[1];
+
+  memset((j_decompress_ptr)&dinfo, 0, sizeof(dinfo));   // for setjmp sanity
+  dinfo.err = jpeg_std_error(&jerr.pub);
+  jerr.pub.error_exit = my_error_exit;
+
+  if (setjmp(jerr.setjmp_buffer)) {
+ Error:
+    MetadataFree(metadata);
+    jpeg_destroy_decompress((j_decompress_ptr)&dinfo);
+    goto End;
+  }
+
+  jpeg_create_decompress((j_decompress_ptr)&dinfo);
+  jpeg_stdio_src((j_decompress_ptr)&dinfo, in_file);
+  if (metadata != NULL) SaveMetadataMarkers((j_decompress_ptr)&dinfo);
+  jpeg_read_header((j_decompress_ptr)&dinfo, TRUE);
+
+  dinfo.out_color_space = JCS_RGB;
+  dinfo.do_fancy_upsampling = TRUE;
+
+  jpeg_start_decompress((j_decompress_ptr)&dinfo);
+
+  if (dinfo.output_components != 3) {
+    goto Error;
+  }
+
+  width = dinfo.output_width;
+  height = dinfo.output_height;
+  stride = dinfo.output_width * dinfo.output_components * sizeof(*rgb);
+
+  rgb = (uint8_t*)malloc(stride * height);
+  if (rgb == NULL) {
+    goto End;
+  }
+  buffer[0] = (JSAMPLE*)rgb;
+
+  while (dinfo.output_scanline < dinfo.output_height) {
+    if (jpeg_read_scanlines((j_decompress_ptr)&dinfo, buffer, 1) != 1) {
+      goto End;
+    }
+    buffer[0] += stride;
+  }
+
+  if (metadata != NULL) {
+    ok = ExtractMetadataFromJPEG((j_decompress_ptr)&dinfo, metadata);
+    if (!ok) {
+      fprintf(stderr, "Error extracting JPEG metadata!\n");
+      goto Error;
+    }
+  }
+
+  jpeg_finish_decompress((j_decompress_ptr)&dinfo);
+  jpeg_destroy_decompress((j_decompress_ptr)&dinfo);
+
+  // WebP conversion.
+  pic->width = width;
+  pic->height = height;
+  pic->use_argb = 1;      // store raw RGB samples
+  ok = WebPPictureImportRGB(pic, rgb, stride);
+  if (!ok) goto Error;
+
+ End:
+  free(rgb);
+  return ok;
+}
+#else  // !WEBP_HAVE_JPEG
+int ReadJPEG(FILE* in_file, struct WebPPicture* const pic,
+             struct Metadata* const metadata) {
+  (void)in_file;
+  (void)pic;
+  (void)metadata;
+  fprintf(stderr, "JPEG support not compiled. Please install the libjpeg "
+          "development package before building.\n");
+  return 0;
+}
+#endif  // WEBP_HAVE_JPEG
+
+// -----------------------------------------------------------------------------

examples/jpegdec.h

@@ -0,0 +1,35 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// JPEG decode.
+
+#ifndef WEBP_EXAMPLES_JPEGDEC_H_
+#define WEBP_EXAMPLES_JPEGDEC_H_
+
+#include <stdio.h>
+#include "webp/types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct Metadata;
+struct WebPPicture;
+
+// Reads a JPEG from 'in_file', returning the decoded output in 'pic'.
+// The output is RGB.
+// Returns true on success.
+int ReadJPEG(FILE* in_file, struct WebPPicture* const pic,
+             struct Metadata* const metadata);
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_EXAMPLES_JPEGDEC_H_

examples/metadata.c

@@ -0,0 +1,49 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//  Metadata types and functions.
+//
+
+#include "./metadata.h"
+
+#include <stdlib.h>
+#include <string.h>
+
+#include "webp/types.h"
+
+void MetadataInit(Metadata* const metadata) {
+  if (metadata == NULL) return;
+  memset(metadata, 0, sizeof(*metadata));
+}
+
+void MetadataPayloadDelete(MetadataPayload* const payload) {
+  if (payload == NULL) return;
+  free(payload->bytes);
+  payload->bytes = NULL;
+  payload->size = 0;
+}
+
+void MetadataFree(Metadata* const metadata) {
+  if (metadata == NULL) return;
+  MetadataPayloadDelete(&metadata->exif);
+  MetadataPayloadDelete(&metadata->iccp);
+  MetadataPayloadDelete(&metadata->xmp);
+}
+
+int MetadataCopy(const char* metadata, size_t metadata_len,
+                 MetadataPayload* const payload) {
+  if (metadata == NULL || metadata_len == 0 || payload == NULL) return 0;
+  payload->bytes = (uint8_t*)malloc(metadata_len);
+  if (payload->bytes == NULL) return 0;
+  payload->size = metadata_len;
+  memcpy(payload->bytes, metadata, metadata_len);
+  return 1;
+}
+
+// -----------------------------------------------------------------------------

examples/metadata.h

@@ -0,0 +1,47 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//  Metadata types and functions.
+//
+
+#ifndef WEBP_EXAMPLES_METADATA_H_
+#define WEBP_EXAMPLES_METADATA_H_
+
+#include "webp/types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct MetadataPayload {
+  uint8_t* bytes;
+  size_t size;
+} MetadataPayload;
+
+typedef struct Metadata {
+  MetadataPayload exif;
+  MetadataPayload iccp;
+  MetadataPayload xmp;
+} Metadata;
+
+#define METADATA_OFFSET(x) offsetof(Metadata, x)
+
+void MetadataInit(Metadata* const metadata);
+void MetadataPayloadDelete(MetadataPayload* const payload);
+void MetadataFree(Metadata* const metadata);
+
+// Stores 'metadata' to 'payload->bytes', returns false on allocation error.
+int MetadataCopy(const char* metadata, size_t metadata_len,
+                 MetadataPayload* const payload);
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_EXAMPLES_METADATA_H_

examples/pngdec.c

@@ -0,0 +1,302 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// PNG decode.
+
+#include "./pngdec.h"
+
+#ifdef HAVE_CONFIG_H
+#include "webp/config.h"
+#endif
+
+#include <stdio.h>
+
+#ifdef WEBP_HAVE_PNG
+#include <png.h>
+#include <setjmp.h>   // note: this must be included *after* png.h
+#include <stdlib.h>
+#include <string.h>
+
+#include "webp/encode.h"
+#include "./metadata.h"
+
+static void PNGAPI error_function(png_structp png, png_const_charp error) {
+  if (error != NULL) fprintf(stderr, "libpng error: %s\n", error);
+  longjmp(png_jmpbuf(png), 1);
+}
+
+// Converts the NULL terminated 'hexstring' which contains 2-byte character
+// representations of hex values to raw data.
+// 'hexstring' may contain values consisting of [A-F][a-f][0-9] in pairs,
+// e.g., 7af2..., separated by any number of newlines.
+// 'expected_length' is the anticipated processed size.
+// On success the raw buffer is returned with its length equivalent to
+// 'expected_length'. NULL is returned if the processed length is less than
+// 'expected_length' or any character aside from those above is encountered.
+// The returned buffer must be freed by the caller.
+static uint8_t* HexStringToBytes(const char* hexstring,
+                                 size_t expected_length) {
+  const char* src = hexstring;
+  size_t actual_length = 0;
+  uint8_t* const raw_data = (uint8_t*)malloc(expected_length);
+  uint8_t* dst;
+
+  if (raw_data == NULL) return NULL;
+
+  for (dst = raw_data; actual_length < expected_length && *src != '\0'; ++src) {
+    char* end;
+    char val[3];
+    if (*src == '\n') continue;
+    val[0] = *src++;
+    val[1] = *src;
+    val[2] = '\0';
+    *dst++ = (uint8_t)strtol(val, &end, 16);
+    if (end != val + 2) break;
+    ++actual_length;
+  }
+
+  if (actual_length != expected_length) {
+    free(raw_data);
+    return NULL;
+  }
+  return raw_data;
+}
+
+static int ProcessRawProfile(const char* profile, size_t profile_len,
+                             MetadataPayload* const payload) {
+  const char* src = profile;
+  char* end;
+  int expected_length;
+
+  if (profile == NULL || profile_len == 0) return 0;
+
+  // ImageMagick formats 'raw profiles' as
+  // '\n<name>\n<length>(%8lu)\n<hex payload>\n'.
+  if (*src != '\n') {
+    fprintf(stderr, "Malformed raw profile, expected '\\n' got '\\x%.2X'\n",
+            *src);
+    return 0;
+  }
+  ++src;
+  // skip the profile name and extract the length.
+  while (*src != '\0' && *src++ != '\n') {}
+  expected_length = (int)strtol(src, &end, 10);
+  if (*end != '\n') {
+    fprintf(stderr, "Malformed raw profile, expected '\\n' got '\\x%.2X'\n",
+            *end);
+    return 0;
+  }
+  ++end;
+
+  // 'end' now points to the profile payload.
+  payload->bytes = HexStringToBytes(end, expected_length);
+  if (payload->bytes == NULL) return 0;
+  payload->size = expected_length;
+  return 1;
+}
+
+static const struct {
+  const char* name;
+  int (*process)(const char* profile, size_t profile_len,
+                 MetadataPayload* const payload);
+  size_t storage_offset;
+} kPNGMetadataMap[] = {
+  // http://www.sno.phy.queensu.ca/~phil/exiftool/TagNames/PNG.html#TextualData
+  // See also: ExifTool on CPAN.
+  { "Raw profile type exif", ProcessRawProfile, METADATA_OFFSET(exif) },
+  { "Raw profile type xmp",  ProcessRawProfile, METADATA_OFFSET(xmp) },
+  // Exiftool puts exif data in APP1 chunk, too.
+  { "Raw profile type APP1", ProcessRawProfile, METADATA_OFFSET(exif) },
+  // XMP Specification Part 3, Section 3 #PNG
+  { "XML:com.adobe.xmp",     MetadataCopy,      METADATA_OFFSET(xmp) },
+  { NULL, NULL, 0 },
+};
+
+// Looks for metadata at both the beginning and end of the PNG file, giving
+// preference to the head.
+// Returns true on success. The caller must use MetadataFree() on 'metadata' in
+// all cases.
+static int ExtractMetadataFromPNG(png_structp png,
+                                  png_infop const head_info,
+                                  png_infop const end_info,
+                                  Metadata* const metadata) {
+  int p;
+
+  for (p = 0; p < 2; ++p)  {
+    png_infop const info = (p == 0) ? head_info : end_info;
+    png_textp text = NULL;
+    const int num = png_get_text(png, info, &text, NULL);
+    int i;
+    // Look for EXIF / XMP metadata.
+    for (i = 0; i < num; ++i, ++text) {
+      int j;
+      for (j = 0; kPNGMetadataMap[j].name != NULL; ++j) {
+        if (!strcmp(text->key, kPNGMetadataMap[j].name)) {
+          MetadataPayload* const payload =
+              (MetadataPayload*)((uint8_t*)metadata +
+                                 kPNGMetadataMap[j].storage_offset);
+          png_size_t text_length;
+          switch (text->compression) {
+#ifdef PNG_iTXt_SUPPORTED
+            case PNG_ITXT_COMPRESSION_NONE:
+            case PNG_ITXT_COMPRESSION_zTXt:
+              text_length = text->itxt_length;
+              break;
+#endif
+            case PNG_TEXT_COMPRESSION_NONE:
+            case PNG_TEXT_COMPRESSION_zTXt:
+            default:
+              text_length = text->text_length;
+              break;
+          }
+          if (payload->bytes != NULL) {
+            fprintf(stderr, "Ignoring additional '%s'\n", text->key);
+          } else if (!kPNGMetadataMap[j].process(text->text, text_length,
+                                                 payload)) {
+            fprintf(stderr, "Failed to process: '%s'\n", text->key);
+            return 0;
+          }
+          break;
+        }
+      }
+    }
+    // Look for an ICC profile.
+    {
+      png_charp name;
+      int comp_type;
+#if ((PNG_LIBPNG_VER_MAJOR << 8) | PNG_LIBPNG_VER_MINOR << 0) < \
+    ((1 << 8) | (5 << 0))
+      png_charp profile;
+#else  // >= libpng 1.5.0
+      png_bytep profile;
+#endif
+      png_uint_32 len;
+
+      if (png_get_iCCP(png, info,
+                       &name, &comp_type, &profile, &len) == PNG_INFO_iCCP) {
+        if (!MetadataCopy((const char*)profile, len, &metadata->iccp)) return 0;
+      }
+    }
+  }
+
+  return 1;
+}
+
+int ReadPNG(FILE* in_file, WebPPicture* const pic, int keep_alpha,
+            Metadata* const metadata) {
+  volatile png_structp png;
+  volatile png_infop info = NULL;
+  volatile png_infop end_info = NULL;
+  int color_type, bit_depth, interlaced;
+  int has_alpha;
+  int num_passes;
+  int p;
+  int ok = 0;
+  png_uint_32 width, height, y;
+  int stride;
+  uint8_t* volatile rgb = NULL;
+
+  png = png_create_read_struct(PNG_LIBPNG_VER_STRING, 0, 0, 0);
+  if (png == NULL) {
+    goto End;
+  }
+
+  png_set_error_fn(png, 0, error_function, NULL);
+  if (setjmp(png_jmpbuf(png))) {
+ Error:
+    MetadataFree(metadata);
+    goto End;
+  }
+
+  info = png_create_info_struct(png);
+  if (info == NULL) goto Error;
+  end_info = png_create_info_struct(png);
+  if (end_info == NULL) goto Error;
+
+  png_init_io(png, in_file);
+  png_read_info(png, info);
+  if (!png_get_IHDR(png, info,
+                    &width, &height, &bit_depth, &color_type, &interlaced,
+                    NULL, NULL)) goto Error;
+
+  png_set_strip_16(png);
+  png_set_packing(png);
+  if (color_type == PNG_COLOR_TYPE_PALETTE) {
+    png_set_palette_to_rgb(png);
+  }
+  if (color_type == PNG_COLOR_TYPE_GRAY ||
+      color_type == PNG_COLOR_TYPE_GRAY_ALPHA) {
+    if (bit_depth < 8) {
+      png_set_expand_gray_1_2_4_to_8(png);
+    }
+    png_set_gray_to_rgb(png);
+  }
+  if (png_get_valid(png, info, PNG_INFO_tRNS)) {
+    png_set_tRNS_to_alpha(png);
+    has_alpha = 1;
+  } else {
+    has_alpha = !!(color_type & PNG_COLOR_MASK_ALPHA);
+  }
+
+  if (!keep_alpha) {
+    png_set_strip_alpha(png);
+    has_alpha = 0;
+  }
+
+  num_passes = png_set_interlace_handling(png);
+  png_read_update_info(png, info);
+  stride = (has_alpha ? 4 : 3) * width * sizeof(*rgb);
+  rgb = (uint8_t*)malloc(stride * height);
+  if (rgb == NULL) goto Error;
+  for (p = 0; p < num_passes; ++p) {
+    for (y = 0; y < height; ++y) {
+      png_bytep row = (png_bytep)(rgb + y * stride);
+      png_read_rows(png, &row, NULL, 1);
+    }
+  }
+  png_read_end(png, end_info);
+
+  if (metadata != NULL &&
+      !ExtractMetadataFromPNG(png, info, end_info, metadata)) {
+    fprintf(stderr, "Error extracting PNG metadata!\n");
+    goto Error;
+  }
+
+  pic->width = width;
+  pic->height = height;
+  pic->use_argb = 1;
+  ok = has_alpha ? WebPPictureImportRGBA(pic, rgb, stride)
+                 : WebPPictureImportRGB(pic, rgb, stride);
+
+  if (!ok) {
+    goto Error;
+  }
+
+ End:
+  if (png != NULL) {
+    png_destroy_read_struct((png_structpp)&png,
+                            (png_infopp)&info, (png_infopp)&end_info);
+  }
+  free(rgb);
+  return ok;
+}
+#else  // !WEBP_HAVE_PNG
+int ReadPNG(FILE* in_file, struct WebPPicture* const pic, int keep_alpha,
+            struct Metadata* const metadata) {
+  (void)in_file;
+  (void)pic;
+  (void)keep_alpha;
+  (void)metadata;
+  fprintf(stderr, "PNG support not compiled. Please install the libpng "
+          "development package before building.\n");
+  return 0;
+}
+#endif  // WEBP_HAVE_PNG
+
+// -----------------------------------------------------------------------------

examples/pngdec.h

@@ -0,0 +1,35 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// PNG decode.
+
+#ifndef WEBP_EXAMPLES_PNGDEC_H_
+#define WEBP_EXAMPLES_PNGDEC_H_
+
+#include <stdio.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct Metadata;
+struct WebPPicture;
+
+// Reads a PNG from 'in_file', returning the decoded output in 'pic'.
+// If 'keep_alpha' is true and the PNG has an alpha channel, the output is RGBA
+// otherwise it will be RGB.
+// Returns true on success.
+int ReadPNG(FILE* in_file, struct WebPPicture* const pic, int keep_alpha,
+            struct Metadata* const metadata);
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_EXAMPLES_PNGDEC_H_

examples/stopwatch.h

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 //  Helper functions to measure elapsed time.
@@ -17,6 +19,10 @@
 
 typedef LARGE_INTEGER Stopwatch;
 
+static WEBP_INLINE void StopwatchReset(Stopwatch* watch) {
+  QueryPerformanceCounter(watch);
+}
+
 static WEBP_INLINE double StopwatchReadAndReset(Stopwatch* watch) {
   const LARGE_INTEGER old_value = *watch;
   LARGE_INTEGER freq;
@@ -35,11 +41,18 @@ static WEBP_INLINE double StopwatchReadAndReset(Stopwatch* watch) {
 
 typedef struct timeval Stopwatch;
 
-static WEBP_INLINE double StopwatchReadAndReset(Stopwatch* watch) {
-  const struct timeval old_value = *watch;
+static WEBP_INLINE void StopwatchReset(Stopwatch* watch) {
   gettimeofday(watch, NULL);
-  return watch->tv_sec - old_value.tv_sec +
-      (watch->tv_usec - old_value.tv_usec) / 1000000.0;
+}
+
+static WEBP_INLINE double StopwatchReadAndReset(Stopwatch* watch) {
+  struct timeval old_value;
+  double delta_sec, delta_usec;
+  memcpy(&old_value, watch, sizeof(old_value));
+  gettimeofday(watch, NULL);
+  delta_sec = (double)watch->tv_sec - old_value.tv_sec;
+  delta_usec = (double)watch->tv_usec - old_value.tv_usec;
+  return delta_sec + delta_usec / 1000000.0;
 }
 
 #endif   /* _WIN32 */

examples/tiffdec.c

@@ -0,0 +1,141 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// TIFF decode.
+
+#include "./tiffdec.h"
+
+#ifdef HAVE_CONFIG_H
+#include "webp/config.h"
+#endif
+
+#include <stdio.h>
+
+#ifdef WEBP_HAVE_TIFF
+#include <tiffio.h>
+
+#include "webp/encode.h"
+#include "./metadata.h"
+
+static const struct {
+  ttag_t tag;
+  size_t storage_offset;
+} kTIFFMetadataMap[] = {
+  { TIFFTAG_ICCPROFILE, METADATA_OFFSET(iccp) },
+  { TIFFTAG_XMLPACKET,  METADATA_OFFSET(xmp) },
+  { 0, 0 },
+};
+
+// Returns true on success. The caller must use MetadataFree() on 'metadata' in
+// all cases.
+static int ExtractMetadataFromTIFF(TIFF* const tif, Metadata* const metadata) {
+  int i;
+  toff_t exif_ifd_offset;
+
+  for (i = 0; kTIFFMetadataMap[i].tag != 0; ++i) {
+    MetadataPayload* const payload =
+        (MetadataPayload*)((uint8_t*)metadata +
+                           kTIFFMetadataMap[i].storage_offset);
+    void* tag_data;
+    uint32 tag_data_len;
+
+    if (TIFFGetField(tif, kTIFFMetadataMap[i].tag, &tag_data_len, &tag_data) &&
+        !MetadataCopy((const char*)tag_data, tag_data_len, payload)) {
+      return 0;
+    }
+  }
+
+  // TODO(jzern): To extract the raw EXIF directory some parsing of it would be
+  // necessary to determine the overall size. In addition, value offsets in
+  // individual directory entries may need to be updated as, depending on the
+  // type, they are file based.
+  // Exif 2.2 Section 4.6.2 Tag Structure
+  // TIFF Revision 6.0 Part 1 Section 2 TIFF Structure #Image File Directory
+  if (TIFFGetField(tif, TIFFTAG_EXIFIFD, &exif_ifd_offset)) {
+    fprintf(stderr, "Warning: EXIF extraction from TIFF is unsupported.\n");
+  }
+  return 1;
+}
+
+int ReadTIFF(const char* const filename,
+             WebPPicture* const pic, int keep_alpha,
+             Metadata* const metadata) {
+  TIFF* const tif = TIFFOpen(filename, "r");
+  uint32 width, height;
+  uint32* raster;
+  int ok = 0;
+  tdir_t dircount;
+
+  if (tif == NULL) {
+    fprintf(stderr, "Error! Cannot open TIFF file '%s'\n", filename);
+    return 0;
+  }
+
+  dircount = TIFFNumberOfDirectories(tif);
+  if (dircount > 1) {
+    fprintf(stderr, "Warning: multi-directory TIFF files are not supported.\n"
+                    "Only the first will be used, %d will be ignored.\n",
+                    dircount - 1);
+  }
+
+  if (!(TIFFGetField(tif, TIFFTAG_IMAGEWIDTH, &width) &&
+        TIFFGetField(tif, TIFFTAG_IMAGELENGTH, &height))) {
+    fprintf(stderr, "Error! Cannot retrieve TIFF image dimensions.\n");
+    return 0;
+  }
+  raster = (uint32*)_TIFFmalloc(width * height * sizeof(*raster));
+  if (raster != NULL) {
+    if (TIFFReadRGBAImageOriented(tif, width, height, raster,
+                                  ORIENTATION_TOPLEFT, 1)) {
+      const int stride = width * sizeof(*raster);
+      pic->width = width;
+      pic->height = height;
+      // TIFF data is ABGR
+#ifdef WORDS_BIGENDIAN
+      TIFFSwabArrayOfLong(raster, width * height);
+#endif
+      pic->use_argb = 1;
+      ok = keep_alpha
+         ? WebPPictureImportRGBA(pic, (const uint8_t*)raster, stride)
+         : WebPPictureImportRGBX(pic, (const uint8_t*)raster, stride);
+    }
+    _TIFFfree(raster);
+  } else {
+    fprintf(stderr, "Error allocating TIFF RGBA memory!\n");
+  }
+
+  if (ok) {
+    if (metadata != NULL) {
+      ok = ExtractMetadataFromTIFF(tif, metadata);
+      if (!ok) {
+        fprintf(stderr, "Error extracting TIFF metadata!\n");
+        MetadataFree(metadata);
+        WebPPictureFree(pic);
+      }
+    }
+  }
+
+  TIFFClose(tif);
+  return ok;
+}
+#else  // !WEBP_HAVE_TIFF
+int ReadTIFF(const char* const filename,
+             struct WebPPicture* const pic, int keep_alpha,
+             struct Metadata* const metadata) {
+  (void)filename;
+  (void)pic;
+  (void)keep_alpha;
+  (void)metadata;
+  fprintf(stderr, "TIFF support not compiled. Please install the libtiff "
+          "development package before building.\n");
+  return 0;
+}
+#endif  // WEBP_HAVE_TIFF
+
+// -----------------------------------------------------------------------------

examples/tiffdec.h

@@ -0,0 +1,34 @@
+// Copyright 2012 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// TIFF decode.
+
+#ifndef WEBP_EXAMPLES_TIFFDEC_H_
+#define WEBP_EXAMPLES_TIFFDEC_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct Metadata;
+struct WebPPicture;
+
+// Reads a TIFF from 'filename', returning the decoded output in 'pic'.
+// If 'keep_alpha' is true and the TIFF has an alpha channel, the output is RGBA
+// otherwise it will be RGB.
+// Returns true on success.
+int ReadTIFF(const char* const filename,
+             struct WebPPicture* const pic, int keep_alpha,
+             struct Metadata* const metadata);
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_EXAMPLES_TIFFDEC_H_

examples/vwebp.c

@@ -1,28 +1,26 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
-//  Simple WebP file viewer.
-//
-// Compiling on linux:
-//   sudo apt-get install libglut3-dev mesa-common-dev
-//   gcc -o vwebp vwebp.c -O3 -lwebp -lwebpmux -lglut -lGL -lpthread -lm
-// Compiling on Mac + XCode:
-//   gcc -o vwebp vwebp.c -lwebp -lwebpmux -framework GLUT -framework OpenGL
+//  Simple OpenGL-based WebP file viewer.
 //
 // Author: Skal (pascal.massimino@gmail.com)
+#ifdef HAVE_CONFIG_H
+#include "webp/config.h"
+#endif
 
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 
-#include "webp/decode.h"
-#include "webp/mux.h"
+#if defined(WEBP_HAVE_GL)
 
-#ifdef __APPLE__
+#if defined(HAVE_GLUT_GLUT_H)
 #include <GLUT/glut.h>
 #else
 #include <GL/glut.h>
@@ -31,6 +29,13 @@
 #endif
 #endif
 
+#ifdef WEBP_HAVE_QCMS
+#include <qcms.h>
+#endif
+
+#include "webp/decode.h"
+#include "webp/demux.h"
+
 #include "./example_util.h"
 
 #ifdef _MSC_VER
@@ -42,26 +47,25 @@ static void Help(void);
 // Unfortunate global variables. Gathered into a struct for comfort.
 static struct {
   int has_animation;
+  int has_color_profile;
   int done;
   int decoding_error;
   int print_info;
+  int use_color_profile;
 
-  uint32_t flags;
+  int canvas_width, canvas_height;
   int loop_count;
-  int frame_num;
-  int frame_max;
+  uint32_t bg_color;
 
   const char* file_name;
   WebPData data;
-  WebPMux* mux;
-  WebPDecoderConfig* config;
+  WebPDecoderConfig config;
   const WebPDecBuffer* pic;
-} kParams = {
-  0, 0, 0, 0,         // has_animation, ...
-  0, 1, 1, 0,         // flags, ...
-  NULL, { NULL, 0 },  // file_name, ...
-  NULL, NULL, NULL    // mux, ...
-};
+  WebPDemuxer* dmux;
+  WebPIterator curr_frame;
+  WebPIterator prev_frame;
+  WebPChunkIterator iccp;
+} kParams;
 
 static void ClearPreviousPic(void) {
   WebPFreeDecBuffer((WebPDecBuffer*)kParams.pic);
@@ -71,8 +75,128 @@ static void ClearPreviousPic(void) {
 static void ClearParams(void) {
   ClearPreviousPic();
   WebPDataClear(&kParams.data);
-  WebPMuxDelete(kParams.mux);
-  kParams.mux = NULL;
+  WebPDemuxReleaseIterator(&kParams.curr_frame);
+  WebPDemuxReleaseIterator(&kParams.prev_frame);
+  WebPDemuxReleaseChunkIterator(&kParams.iccp);
+  WebPDemuxDelete(kParams.dmux);
+  kParams.dmux = NULL;
+}
+
+// -----------------------------------------------------------------------------
+// Color profile handling
+static int ApplyColorProfile(const WebPData* const profile,
+                             WebPDecBuffer* const rgba) {
+#ifdef WEBP_HAVE_QCMS
+  int i, ok = 0;
+  uint8_t* line;
+  uint8_t major_revision;
+  qcms_profile* input_profile = NULL;
+  qcms_profile* output_profile = NULL;
+  qcms_transform* transform = NULL;
+  const qcms_data_type input_type = QCMS_DATA_RGBA_8;
+  const qcms_data_type output_type = QCMS_DATA_RGBA_8;
+  const qcms_intent intent = QCMS_INTENT_DEFAULT;
+
+  if (profile == NULL || rgba == NULL) return 0;
+  if (profile->bytes == NULL || profile->size < 10) return 1;
+  major_revision = profile->bytes[8];
+
+  qcms_enable_iccv4();
+  input_profile = qcms_profile_from_memory(profile->bytes, profile->size);
+  // qcms_profile_is_bogus() is broken with ICCv4.
+  if (input_profile == NULL ||
+      (major_revision < 4 && qcms_profile_is_bogus(input_profile))) {
+    fprintf(stderr, "Color profile is bogus!\n");
+    goto Error;
+  }
+
+  output_profile = qcms_profile_sRGB();
+  if (output_profile == NULL) {
+    fprintf(stderr, "Error creating output color profile!\n");
+    goto Error;
+  }
+
+  qcms_profile_precache_output_transform(output_profile);
+  transform = qcms_transform_create(input_profile, input_type,
+                                    output_profile, output_type,
+                                    intent);
+  if (transform == NULL) {
+    fprintf(stderr, "Error creating color transform!\n");
+    goto Error;
+  }
+
+  line = rgba->u.RGBA.rgba;
+  for (i = 0; i < rgba->height; ++i, line += rgba->u.RGBA.stride) {
+    qcms_transform_data(transform, line, line, rgba->width);
+  }
+  ok = 1;
+
+ Error:
+  if (input_profile != NULL) qcms_profile_release(input_profile);
+  if (output_profile != NULL) qcms_profile_release(output_profile);
+  if (transform != NULL) qcms_transform_release(transform);
+  return ok;
+#else
+  (void)profile;
+  (void)rgba;
+  return 1;
+#endif  // WEBP_HAVE_QCMS
+}
+
+//------------------------------------------------------------------------------
+// File decoding
+
+static int Decode(void) {   // Fills kParams.curr_frame
+  const WebPIterator* const curr = &kParams.curr_frame;
+  WebPDecoderConfig* const config = &kParams.config;
+  WebPDecBuffer* const output_buffer = &config->output;
+  int ok = 0;
+
+  ClearPreviousPic();
+  output_buffer->colorspace = MODE_RGBA;
+  ok = (WebPDecode(curr->fragment.bytes, curr->fragment.size,
+                   config) == VP8_STATUS_OK);
+  if (!ok) {
+    fprintf(stderr, "Decoding of frame #%d failed!\n", curr->frame_num);
+  } else {
+    kParams.pic = output_buffer;
+    if (kParams.use_color_profile) {
+      ok = ApplyColorProfile(&kParams.iccp.chunk, output_buffer);
+      if (!ok) {
+        fprintf(stderr, "Applying color profile to frame #%d failed!\n",
+                curr->frame_num);
+      }
+    }
+  }
+  return ok;
+}
+
+static void decode_callback(int what) {
+  if (what == 0 && !kParams.done) {
+    int duration = 0;
+    if (kParams.dmux != NULL) {
+      WebPIterator* const curr = &kParams.curr_frame;
+      if (!WebPDemuxNextFrame(curr)) {
+        WebPDemuxReleaseIterator(curr);
+        if (WebPDemuxGetFrame(kParams.dmux, 1, curr)) {
+          --kParams.loop_count;
+          kParams.done = (kParams.loop_count == 0);
+        } else {
+          kParams.decoding_error = 1;
+          kParams.done = 1;
+          return;
+        }
+      }
+      duration = curr->duration;
+    }
+    if (!Decode()) {
+      kParams.decoding_error = 1;
+      kParams.done = 1;
+    } else {
+      glutPostRedisplay();
+      glutTimerFunc(duration, decode_callback, what);
+    }
+  }
 }
 
 //------------------------------------------------------------------------------
@@ -88,6 +212,24 @@ static void HandleKey(unsigned char key, int pos_x, int pos_y) {
     ClearParams();
     exit(0);
 #endif
+  } else if (key == 'c') {
+    if (kParams.has_color_profile && !kParams.decoding_error) {
+      kParams.use_color_profile = 1 - kParams.use_color_profile;
+
+      if (kParams.has_animation) {
+        // Restart the completed animation to pickup the color profile change.
+        if (kParams.done && kParams.loop_count == 0) {
+          kParams.loop_count =
+              (int)WebPDemuxGetI(kParams.dmux, WEBP_FF_LOOP_COUNT) + 1;
+          kParams.done = 0;
+          // Start the decode loop immediately.
+          glutTimerFunc(0, decode_callback, 0);
+        }
+      } else {
+        Decode();
+        glutPostRedisplay();
+      }
+    }
   } else if (key == 'i') {
     kParams.print_info = 1 - kParams.print_info;
     glutPostRedisplay();
@@ -112,6 +254,10 @@ static void PrintString(const char* const text) {
   }
 }
 
+static float GetColorf(uint32_t color, int shift) {
+  return (color >> shift) / 255.f;
+}
+
 static void DrawCheckerBoard(void) {
   const int square_size = 8;  // must be a power of 2
   int x, y;
@@ -133,104 +279,90 @@ static void DrawCheckerBoard(void) {
 }
 
 static void HandleDisplay(void) {
-  const WebPDecBuffer* pic = kParams.pic;
+  const WebPDecBuffer* const pic = kParams.pic;
+  const WebPIterator* const curr = &kParams.curr_frame;
+  WebPIterator* const prev = &kParams.prev_frame;
+  GLfloat xoff, yoff;
   if (pic == NULL) return;
-  glClear(GL_COLOR_BUFFER_BIT);
   glPushMatrix();
   glPixelZoom(1, -1);
-  glRasterPos2f(-1, 1);
+  xoff = (GLfloat)(2. * curr->x_offset / kParams.canvas_width);
+  yoff = (GLfloat)(2. * curr->y_offset / kParams.canvas_height);
+  glRasterPos2f(-1.f + xoff, 1.f - yoff);
   glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
   glPixelStorei(GL_UNPACK_ROW_LENGTH, pic->u.RGBA.stride / 4);
-  DrawCheckerBoard();
+
+  if (prev->dispose_method == WEBP_MUX_DISPOSE_BACKGROUND ||
+      curr->blend_method == WEBP_MUX_NO_BLEND) {
+    // TODO(later): these offsets and those above should factor in window size.
+    //              they will be incorrect if the window is resized.
+    // glScissor() takes window coordinates (0,0 at bottom left).
+    int window_x, window_y;
+    if (prev->dispose_method == WEBP_MUX_DISPOSE_BACKGROUND) {
+      // Clear the previous frame rectangle.
+      window_x = prev->x_offset;
+      window_y = kParams.canvas_height - prev->y_offset - prev->height;
+    } else {  // curr->blend_method == WEBP_MUX_NO_BLEND.
+      // We simulate no-blending behavior by first clearing the current frame
+      // rectangle (to a checker-board) and then alpha-blending against it.
+      window_x = curr->x_offset;
+      window_y = kParams.canvas_height - curr->y_offset - curr->height;
+    }
+    glEnable(GL_SCISSOR_TEST);
+    // Only update the requested area, not the whole canvas.
+    glScissor(window_x, window_y, prev->width, prev->height);
+
+    glClear(GL_COLOR_BUFFER_BIT);  // use clear color
+    DrawCheckerBoard();
+
+    glDisable(GL_SCISSOR_TEST);
+  }
+
+  *prev = *curr;
+
   glDrawPixels(pic->width, pic->height,
                GL_RGBA, GL_UNSIGNED_BYTE,
                (GLvoid*)pic->u.RGBA.rgba);
   if (kParams.print_info) {
     char tmp[32];
 
-    glColor4f(0.0, 0.0, 0.0, 1.0);
+    glColor4f(0.90f, 0.0f, 0.90f, 1.0f);
     glRasterPos2f(-0.95f, 0.90f);
     PrintString(kParams.file_name);
 
     snprintf(tmp, sizeof(tmp), "Dimension:%d x %d", pic->width, pic->height);
-    glColor4f(0.0, 0.0, 0.0, 1.0);
+    glColor4f(0.90f, 0.0f, 0.90f, 1.0f);
     glRasterPos2f(-0.95f, 0.80f);
     PrintString(tmp);
+    if (curr->x_offset != 0 || curr->y_offset != 0) {
+      snprintf(tmp, sizeof(tmp), " (offset:%d,%d)",
+               curr->x_offset, curr->y_offset);
+      glRasterPos2f(-0.95f, 0.70f);
+      PrintString(tmp);
+    }
   }
   glPopMatrix();
   glFlush();
 }
 
-static void StartDisplay(const WebPDecBuffer* const pic) {
+static void StartDisplay(void) {
+  const int width = kParams.canvas_width;
+  const int height = kParams.canvas_height;
   glutInitDisplayMode(GLUT_RGBA);
-  glutInitWindowSize(pic->width, pic->height);
+  glutInitWindowSize(width, height);
   glutCreateWindow("WebP viewer");
   glutDisplayFunc(HandleDisplay);
   glutIdleFunc(NULL);
   glutKeyboardFunc(HandleKey);
   glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
   glEnable(GL_BLEND);
-  glClearColor(0.0, 0.0, 0.0, 0.0);
-  HandleReshape(pic->width, pic->height);
-}
-
-//------------------------------------------------------------------------------
-// File decoding
-
-static int Decode(const int frame_number, int* const duration) {
-  WebPDecoderConfig* const config = kParams.config;
-  WebPData *data, image_data;
-  int x_off = 0, y_off = 0;
-  WebPDecBuffer* const output_buffer = &config->output;
-  int ok = 0;
-
-  ClearPreviousPic();
-  if (kParams.has_animation) {
-    if (WebPMuxGetFrame(kParams.mux, frame_number, &image_data,
-                        &x_off, &y_off, duration) != WEBP_MUX_OK) {
-      goto end;
-    }
-    if (x_off != 0 || y_off != 0) {
-      fprintf(stderr,
-              "Frame offsets not yet supported! Forcing offset to 0,0\n");
-      x_off = y_off = 0;
-    }
-    data = &image_data;
-  } else {
-    data = &kParams.data;
-  }
-
-  output_buffer->colorspace = MODE_RGBA;
-  ok = (WebPDecode(data->bytes_, data->size_, config) == VP8_STATUS_OK);
-
- end:
-  if (!ok) {
-    fprintf(stderr, "Decoding of frame #%d failed!\n", frame_number);
-  } else {
-    kParams.pic = output_buffer;
-  }
-  return ok;
-}
-
-static void decode_callback(int what) {
-  if (what == 0 && !kParams.done) {
-    int duration = 0;
-    if (kParams.mux != NULL) {
-      if (!Decode(kParams.frame_num, &duration)) {
-        kParams.decoding_error = 1;
-        kParams.done = 1;
-      } else {
-        ++kParams.frame_num;
-        if (kParams.frame_num > kParams.frame_max) {
-          kParams.frame_num = 1;
-          --kParams.loop_count;
-          kParams.done = (kParams.loop_count == 0);
-        }
-      }
-    }
-    glutPostRedisplay();
-    glutTimerFunc(duration, decode_callback, what);
-  }
+  glClearColor(GetColorf(kParams.bg_color, 0),
+               GetColorf(kParams.bg_color, 8),
+               GetColorf(kParams.bg_color, 16),
+               GetColorf(kParams.bg_color, 24));
+  HandleReshape(width, height);
+  glClear(GL_COLOR_BUFFER_BIT);
+  DrawCheckerBoard();
 }
 
 //------------------------------------------------------------------------------
@@ -240,52 +372,74 @@ static void Help(void) {
   printf("Usage: vwebp in_file [options]\n\n"
          "Decodes the WebP image file and visualize it using OpenGL\n"
          "Options are:\n"
-         "  -version  .... print version number and exit.\n"
-         "  -nofancy ..... don't use the fancy YUV420 upscaler.\n"
-         "  -nofilter .... disable in-loop filtering.\n"
+         "  -version  .... print version number and exit\n"
+         "  -noicc ....... don't use the icc profile if present\n"
+         "  -nofancy ..... don't use the fancy YUV420 upscaler\n"
+         "  -nofilter .... disable in-loop filtering\n"
+         "  -dither <int>  dithering strength (0..100), default=50\n"
+#if WEBP_DECODER_ABI_VERSION > 0x0204
+         "  -noalphadither disable alpha plane dithering\n"
+#endif
          "  -mt .......... use multi-threading\n"
-         "  -crop <x> <y> <w> <h> ... crop output with the given rectangle\n"
-         "  -scale <w> <h> .......... scale the output (*after* any cropping)\n"
-         "  -h     ....... this help message.\n"
+         "  -info ........ print info\n"
+         "  -h     ....... this help message\n"
+         "\n"
+         "Keyboard shortcuts:\n"
+         "  'c' ................ toggle use of color profile\n"
+         "  'i' ................ overlay file information\n"
+         "  'q' / 'Q' / ESC .... quit\n"
         );
 }
 
 int main(int argc, char *argv[]) {
-  WebPDecoderConfig config;
-  WebPMuxError mux_err;
   int c;
+  WebPDecoderConfig* const config = &kParams.config;
+  WebPIterator* const curr = &kParams.curr_frame;
+  WebPIterator* const prev = &kParams.prev_frame;
 
-  if (!WebPInitDecoderConfig(&config)) {
+  if (!WebPInitDecoderConfig(config)) {
     fprintf(stderr, "Library version mismatch!\n");
     return -1;
   }
-  kParams.config = &config;
+  config->options.dithering_strength = 50;
+#if WEBP_DECODER_ABI_VERSION > 0x0204
+  config->options.alpha_dithering_strength = 100;
+#endif
+  kParams.use_color_profile = 1;
 
   for (c = 1; c < argc; ++c) {
+    int parse_error = 0;
     if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       Help();
       return 0;
+    } else if (!strcmp(argv[c], "-noicc")) {
+      kParams.use_color_profile = 0;
     } else if (!strcmp(argv[c], "-nofancy")) {
-      config.options.no_fancy_upsampling = 1;
+      config->options.no_fancy_upsampling = 1;
     } else if (!strcmp(argv[c], "-nofilter")) {
-      config.options.bypass_filtering = 1;
+      config->options.bypass_filtering = 1;
+#if WEBP_DECODER_ABI_VERSION > 0x0204
+    } else if (!strcmp(argv[c], "-noalphadither")) {
+      config->options.alpha_dithering_strength = 0;
+#endif
+    } else if (!strcmp(argv[c], "-dither") && c + 1 < argc) {
+      config->options.dithering_strength =
+          ExUtilGetInt(argv[++c], 0, &parse_error);
+    } else if (!strcmp(argv[c], "-info")) {
+      kParams.print_info = 1;
     } else if (!strcmp(argv[c], "-version")) {
-      const int version = WebPGetDecoderVersion();
-      printf("%d.%d.%d\n",
-        (version >> 16) & 0xff, (version >> 8) & 0xff, version & 0xff);
+      const int dec_version = WebPGetDecoderVersion();
+      const int dmux_version = WebPGetDemuxVersion();
+      printf("WebP Decoder version: %d.%d.%d\nWebP Demux version: %d.%d.%d\n",
+             (dec_version >> 16) & 0xff, (dec_version >> 8) & 0xff,
+             dec_version & 0xff, (dmux_version >> 16) & 0xff,
+             (dmux_version >> 8) & 0xff, dmux_version & 0xff);
       return 0;
     } else if (!strcmp(argv[c], "-mt")) {
-      config.options.use_threads = 1;
-    } else if (!strcmp(argv[c], "-crop") && c < argc - 4) {
-      config.options.use_cropping = 1;
-      config.options.crop_left   = strtol(argv[++c], NULL, 0);
-      config.options.crop_top    = strtol(argv[++c], NULL, 0);
-      config.options.crop_width  = strtol(argv[++c], NULL, 0);
-      config.options.crop_height = strtol(argv[++c], NULL, 0);
-    } else if (!strcmp(argv[c], "-scale") && c < argc - 2) {
-      config.options.use_scaling = 1;
-      config.options.scaled_width  = strtol(argv[++c], NULL, 0);
-      config.options.scaled_height = strtol(argv[++c], NULL, 0);
+      config->options.use_threads = 1;
+    } else if (!strcmp(argv[c], "--")) {
+      if (c < argc - 1) kParams.file_name = argv[++c];
+      break;
     } else if (argv[c][0] == '-') {
       printf("Unknown option '%s'\n", argv[c]);
       Help();
@@ -293,6 +447,11 @@ int main(int argc, char *argv[]) {
     } else {
       kParams.file_name = argv[c];
     }
+
+    if (parse_error) {
+      Help();
+      return -1;
+    }
   }
 
   if (kParams.file_name == NULL) {
@@ -302,53 +461,73 @@ int main(int argc, char *argv[]) {
   }
 
   if (!ExUtilReadFile(kParams.file_name,
-                      &kParams.data.bytes_, &kParams.data.size_)) {
+                      &kParams.data.bytes, &kParams.data.size)) {
     goto Error;
   }
 
-  kParams.mux = WebPMuxCreate(&kParams.data, 0);
-  if (kParams.mux == NULL) {
+  if (!WebPGetInfo(kParams.data.bytes, kParams.data.size, NULL, NULL)) {
+    fprintf(stderr, "Input file doesn't appear to be WebP format.\n");
+    goto Error;
+  }
+
+  kParams.dmux = WebPDemux(&kParams.data);
+  if (kParams.dmux == NULL) {
     fprintf(stderr, "Could not create demuxing object!\n");
     goto Error;
   }
 
-  mux_err = WebPMuxGetFeatures(kParams.mux, &kParams.flags);
-  if (mux_err != WEBP_MUX_OK) {
+  if (WebPDemuxGetI(kParams.dmux, WEBP_FF_FORMAT_FLAGS) & FRAGMENTS_FLAG) {
+    fprintf(stderr, "Image fragments are not supported for now!\n");
     goto Error;
   }
-  if (kParams.flags & TILE_FLAG) {
-    fprintf(stderr, "Tiling is not supported for now!\n");
-    goto Error;
+  kParams.canvas_width = WebPDemuxGetI(kParams.dmux, WEBP_FF_CANVAS_WIDTH);
+  kParams.canvas_height = WebPDemuxGetI(kParams.dmux, WEBP_FF_CANVAS_HEIGHT);
+  if (kParams.print_info) {
+    printf("Canvas: %d x %d\n", kParams.canvas_width, kParams.canvas_height);
   }
 
-  kParams.has_animation = !!(kParams.flags & ANIMATION_FLAG);
+  prev->width = kParams.canvas_width;
+  prev->height = kParams.canvas_height;
+  prev->x_offset = prev->y_offset = 0;
+  prev->dispose_method = WEBP_MUX_DISPOSE_BACKGROUND;
 
-  if (kParams.has_animation) {
-    mux_err = WebPMuxGetLoopCount(kParams.mux, &kParams.loop_count);
-    if (mux_err != WEBP_MUX_OK && mux_err != WEBP_MUX_NOT_FOUND) {
-      goto Error;
-    }
-    mux_err = WebPMuxNumChunks(kParams.mux, WEBP_CHUNK_IMAGE,
-                                      &kParams.frame_max);
-    if (mux_err != WEBP_MUX_OK) {
-      goto Error;
-    }
-    printf("VP8X: Found %d images in file (loop count = %d)\n",
-           kParams.frame_max, kParams.loop_count);
+  memset(&kParams.iccp, 0, sizeof(kParams.iccp));
+  kParams.has_color_profile =
+      !!(WebPDemuxGetI(kParams.dmux, WEBP_FF_FORMAT_FLAGS) & ICCP_FLAG);
+  if (kParams.has_color_profile) {
+#ifdef WEBP_HAVE_QCMS
+    if (!WebPDemuxGetChunk(kParams.dmux, "ICCP", 1, &kParams.iccp)) goto Error;
+    printf("VP8X: Found color profile\n");
+#else
+    fprintf(stderr, "Warning: color profile present, but qcms is unavailable!\n"
+            "Build libqcms from Mozilla or Chromium and define WEBP_HAVE_QCMS "
+            "before building.\n");
+#endif
   }
 
+  if (!WebPDemuxGetFrame(kParams.dmux, 1, curr)) goto Error;
+
+  kParams.has_animation = (curr->num_frames > 1);
+  kParams.loop_count = (int)WebPDemuxGetI(kParams.dmux, WEBP_FF_LOOP_COUNT);
+  kParams.bg_color = WebPDemuxGetI(kParams.dmux, WEBP_FF_BACKGROUND_COLOR);
+  printf("VP8X: Found %d images in file (loop count = %d)\n",
+         curr->num_frames, kParams.loop_count);
+
   // Decode first frame
-  {
-    int duration;
-    if (!Decode(1, &duration)) goto Error;
-  }
+  if (!Decode()) goto Error;
+
+  // Position iterator to last frame. Next call to HandleDisplay will wrap over.
+  // We take this into account by bumping up loop_count.
+  WebPDemuxGetFrame(kParams.dmux, 0, curr);
+  if (kParams.loop_count) ++kParams.loop_count;
 
   // Start display (and timer)
   glutInit(&argc, argv);
 #ifdef FREEGLUT
   glutSetOption(GLUT_ACTION_ON_WINDOW_CLOSE, GLUT_ACTION_CONTINUE_EXECUTION);
 #endif
-  StartDisplay(kParams.pic);
+  StartDisplay();
+
   if (kParams.has_animation) glutTimerFunc(0, decode_callback, 0);
   glutMainLoop();
 
@@ -361,4 +540,14 @@ int main(int argc, char *argv[]) {
   return -1;
 }
 
+#else   // !WEBP_HAVE_GL
+
+int main(int argc, const char *argv[]) {
+  fprintf(stderr, "OpenGL support not enabled in %s.\n", argv[0]);
+  (void)argc;
+  return 0;
+}
+
+#endif
+
 //------------------------------------------------------------------------------

examples/webpdec.c

@@ -0,0 +1,67 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// WebP decode.
+
+#include "./webpdec.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "webp/decode.h"
+#include "webp/encode.h"
+#include "./example_util.h"
+#include "./metadata.h"
+
+int ReadWebP(const char* const in_file, WebPPicture* const pic,
+             int keep_alpha, Metadata* const metadata) {
+  int ok = 0;
+  size_t data_size = 0;
+  const uint8_t* data = NULL;
+  VP8StatusCode status = VP8_STATUS_OK;
+  WebPDecoderConfig config;
+  WebPDecBuffer* const output_buffer = &config.output;
+  WebPBitstreamFeatures* const bitstream = &config.input;
+
+  // TODO(jzern): add Exif/XMP/ICC extraction.
+  if (metadata != NULL) {
+    fprintf(stderr, "Warning: metadata extraction from WebP is unsupported.\n");
+  }
+
+  if (!WebPInitDecoderConfig(&config)) {
+    fprintf(stderr, "Library version mismatch!\n");
+    return 0;
+  }
+
+  if (ExUtilLoadWebP(in_file, &data, &data_size, bitstream)) {
+    const int has_alpha = keep_alpha && bitstream->has_alpha;
+    output_buffer->colorspace = has_alpha ? MODE_RGBA : MODE_RGB;
+
+    status = ExUtilDecodeWebP(data, data_size, 0, &config);
+    if (status == VP8_STATUS_OK) {
+      const uint8_t* const rgba = output_buffer->u.RGBA.rgba;
+      const int stride = output_buffer->u.RGBA.stride;
+      pic->width = output_buffer->width;
+      pic->height = output_buffer->height;
+      pic->use_argb = 1;
+      ok = has_alpha ? WebPPictureImportRGBA(pic, rgba, stride)
+                     : WebPPictureImportRGB(pic, rgba, stride);
+    }
+  }
+
+  if (status != VP8_STATUS_OK) {
+    ExUtilPrintWebPError(in_file, status);
+  }
+
+  free((void*)data);
+  WebPFreeDecBuffer(output_buffer);
+  return ok;
+}
+
+// -----------------------------------------------------------------------------

examples/webpdec.h

@@ -0,0 +1,33 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// WebP decode.
+
+#ifndef WEBP_EXAMPLES_WEBPDEC_H_
+#define WEBP_EXAMPLES_WEBPDEC_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct Metadata;
+struct WebPPicture;
+
+// Reads a WebP from 'in_file', returning the decoded output in 'pic'.
+// If 'keep_alpha' is true and the WebP has an alpha channel, the output is
+// RGBA otherwise it will be RGB.
+// Returns true on success.
+int ReadWebP(const char* const in_file, struct WebPPicture* const pic,
+             int keep_alpha, struct Metadata* const metadata);
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_EXAMPLES_WEBPDEC_H_

examples/wicdec.c

@@ -0,0 +1,349 @@
+// Copyright 2013 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Windows Imaging Component (WIC) decode.
+
+#include "./wicdec.h"
+
+#ifdef HAVE_CONFIG_H
+#include "webp/config.h"
+#endif
+
+#include <stdio.h>
+
+#ifdef HAVE_WINCODEC_H
+#ifdef __MINGW32__
+#define INITGUID  // Without this GUIDs are declared extern and fail to link
+#endif
+#define CINTERFACE
+#define COBJMACROS
+#define _WIN32_IE 0x500  // Workaround bug in shlwapi.h when compiling C++
+                         // code with COBJMACROS.
+#include <shlwapi.h>
+#include <windows.h>
+#include <wincodec.h>
+
+#include "webp/encode.h"
+#include "./metadata.h"
+
+#define IFS(fn)                                                     \
+  do {                                                              \
+    if (SUCCEEDED(hr)) {                                            \
+      hr = (fn);                                                    \
+      if (FAILED(hr)) fprintf(stderr, #fn " failed %08lx\n", hr);   \
+    }                                                               \
+  } while (0)
+
+// modified version of DEFINE_GUID from guiddef.h.
+#define WEBP_DEFINE_GUID(name, l, w1, w2, b1, b2, b3, b4, b5, b6, b7, b8) \
+  static const GUID name = \
+      { l, w1, w2, { b1, b2,  b3,  b4,  b5,  b6,  b7,  b8 } }
+
+#ifdef __cplusplus
+#define MAKE_REFGUID(x) (x)
+#else
+#define MAKE_REFGUID(x) &(x)
+#endif
+
+typedef struct WICFormatImporter {
+  const GUID* pixel_format;
+  int bytes_per_pixel;
+  int (*import)(WebPPicture* const, const uint8_t* const, int);
+} WICFormatImporter;
+
+// From Microsoft SDK 7.0a -- wincodec.h
+// Create local copies for compatibility when building against earlier
+// versions of the SDK.
+WEBP_DEFINE_GUID(GUID_WICPixelFormat24bppBGR_,
+                 0x6fddc324, 0x4e03, 0x4bfe,
+                 0xb1, 0x85, 0x3d, 0x77, 0x76, 0x8d, 0xc9, 0x0c);
+WEBP_DEFINE_GUID(GUID_WICPixelFormat24bppRGB_,
+                 0x6fddc324, 0x4e03, 0x4bfe,
+                 0xb1, 0x85, 0x3d, 0x77, 0x76, 0x8d, 0xc9, 0x0d);
+WEBP_DEFINE_GUID(GUID_WICPixelFormat32bppBGRA_,
+                 0x6fddc324, 0x4e03, 0x4bfe,
+                 0xb1, 0x85, 0x3d, 0x77, 0x76, 0x8d, 0xc9, 0x0f);
+WEBP_DEFINE_GUID(GUID_WICPixelFormat32bppRGBA_,
+                 0xf5c7ad2d, 0x6a8d, 0x43dd,
+                 0xa7, 0xa8, 0xa2, 0x99, 0x35, 0x26, 0x1a, 0xe9);
+
+static HRESULT OpenInputStream(const char* filename, IStream** stream) {
+  HRESULT hr = S_OK;
+  IFS(SHCreateStreamOnFileA(filename, STGM_READ, stream));
+  if (FAILED(hr)) {
+    fprintf(stderr, "Error opening input file %s (%08lx)\n", filename, hr);
+  }
+  return hr;
+}
+
+// -----------------------------------------------------------------------------
+// Metadata processing
+
+// Stores the first non-zero sized color profile from 'frame' to 'iccp'.
+// Returns an HRESULT to indicate success or failure. The caller is responsible
+// for freeing 'iccp->bytes' in either case.
+static HRESULT ExtractICCP(IWICImagingFactory* const factory,
+                           IWICBitmapFrameDecode* const frame,
+                           MetadataPayload* const iccp) {
+  HRESULT hr = S_OK;
+  UINT i, count;
+  IWICColorContext** color_contexts;
+
+  IFS(IWICBitmapFrameDecode_GetColorContexts(frame, 0, NULL, &count));
+  if (FAILED(hr) || count == 0) return hr;
+
+  color_contexts = (IWICColorContext**)calloc(count, sizeof(*color_contexts));
+  if (color_contexts == NULL) return E_OUTOFMEMORY;
+  for (i = 0; SUCCEEDED(hr) && i < count; ++i) {
+    IFS(IWICImagingFactory_CreateColorContext(factory, &color_contexts[i]));
+  }
+
+  if (SUCCEEDED(hr)) {
+    UINT num_color_contexts;
+    IFS(IWICBitmapFrameDecode_GetColorContexts(frame,
+                                               count, color_contexts,
+                                               &num_color_contexts));
+    for (i = 0; SUCCEEDED(hr) && i < num_color_contexts; ++i) {
+      WICColorContextType type;
+      IFS(IWICColorContext_GetType(color_contexts[i], &type));
+      if (SUCCEEDED(hr) && type == WICColorContextProfile) {
+        UINT size;
+        IFS(IWICColorContext_GetProfileBytes(color_contexts[i],
+                                             0, NULL, &size));
+        if (size > 0) {
+          iccp->bytes = (uint8_t*)malloc(size);
+          if (iccp->bytes == NULL) {
+            hr = E_OUTOFMEMORY;
+            break;
+          }
+          iccp->size = size;
+          IFS(IWICColorContext_GetProfileBytes(color_contexts[i],
+                                               (UINT)iccp->size, iccp->bytes,
+                                               &size));
+          if (SUCCEEDED(hr) && size != iccp->size) {
+            fprintf(stderr, "Warning! ICC profile size (%u) != expected (%u)\n",
+                    size, (uint32_t)iccp->size);
+            iccp->size = size;
+          }
+          break;
+        }
+      }
+    }
+  }
+  for (i = 0; i < count; ++i) {
+    if (color_contexts[i] != NULL) IUnknown_Release(color_contexts[i]);
+  }
+  free(color_contexts);
+  return hr;
+}
+
+static HRESULT ExtractMetadata(IWICImagingFactory* const factory,
+                               IWICBitmapFrameDecode* const frame,
+                               Metadata* const metadata) {
+  // TODO(jzern): add XMP/EXIF extraction.
+  const HRESULT hr = ExtractICCP(factory, frame, &metadata->iccp);
+  if (FAILED(hr)) MetadataFree(metadata);
+  return hr;
+}
+
+// -----------------------------------------------------------------------------
+
+static int HasPalette(GUID pixel_format) {
+  return (IsEqualGUID(MAKE_REFGUID(pixel_format),
+                      MAKE_REFGUID(GUID_WICPixelFormat1bppIndexed)) ||
+          IsEqualGUID(MAKE_REFGUID(pixel_format),
+                      MAKE_REFGUID(GUID_WICPixelFormat2bppIndexed)) ||
+          IsEqualGUID(MAKE_REFGUID(pixel_format),
+                      MAKE_REFGUID(GUID_WICPixelFormat4bppIndexed)) ||
+          IsEqualGUID(MAKE_REFGUID(pixel_format),
+                      MAKE_REFGUID(GUID_WICPixelFormat8bppIndexed)));
+}
+
+static int HasAlpha(IWICImagingFactory* const factory,
+                    IWICBitmapDecoder* const decoder,
+                    IWICBitmapFrameDecode* const frame,
+                    GUID pixel_format) {
+  int has_alpha;
+  if (HasPalette(pixel_format)) {
+    IWICPalette* frame_palette = NULL;
+    IWICPalette* global_palette = NULL;
+    BOOL frame_palette_has_alpha = FALSE;
+    BOOL global_palette_has_alpha = FALSE;
+
+    // A palette may exist at the frame or container level,
+    // check IWICPalette::HasAlpha() for both if present.
+    if (SUCCEEDED(IWICImagingFactory_CreatePalette(factory, &frame_palette)) &&
+        SUCCEEDED(IWICBitmapFrameDecode_CopyPalette(frame, frame_palette))) {
+      IWICPalette_HasAlpha(frame_palette, &frame_palette_has_alpha);
+    }
+    if (SUCCEEDED(IWICImagingFactory_CreatePalette(factory, &global_palette)) &&
+        SUCCEEDED(IWICBitmapDecoder_CopyPalette(decoder, global_palette))) {
+      IWICPalette_HasAlpha(global_palette, &global_palette_has_alpha);
+    }
+    has_alpha = frame_palette_has_alpha || global_palette_has_alpha;
+
+    if (frame_palette != NULL) IUnknown_Release(frame_palette);
+    if (global_palette != NULL) IUnknown_Release(global_palette);
+  } else {
+    has_alpha = IsEqualGUID(MAKE_REFGUID(pixel_format),
+                            MAKE_REFGUID(GUID_WICPixelFormat32bppRGBA_)) ||
+                IsEqualGUID(MAKE_REFGUID(pixel_format),
+                            MAKE_REFGUID(GUID_WICPixelFormat32bppBGRA_));
+  }
+  return has_alpha;
+}
+
+int ReadPictureWithWIC(const char* const filename,
+                       WebPPicture* const pic, int keep_alpha,
+                       Metadata* const metadata) {
+  // From Microsoft SDK 6.0a -- ks.h
+  // Define a local copy to avoid link errors under mingw.
+  WEBP_DEFINE_GUID(GUID_NULL_, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
+  static const WICFormatImporter kAlphaFormatImporters[] = {
+    { &GUID_WICPixelFormat32bppBGRA_, 4, WebPPictureImportBGRA },
+    { &GUID_WICPixelFormat32bppRGBA_, 4, WebPPictureImportRGBA },
+    { NULL, 0, NULL },
+  };
+  static const WICFormatImporter kNonAlphaFormatImporters[] = {
+    { &GUID_WICPixelFormat24bppBGR_, 3, WebPPictureImportBGR },
+    { &GUID_WICPixelFormat24bppRGB_, 3, WebPPictureImportRGB },
+    { NULL, 0, NULL },
+  };
+  HRESULT hr = S_OK;
+  IWICBitmapFrameDecode* frame = NULL;
+  IWICFormatConverter* converter = NULL;
+  IWICImagingFactory* factory = NULL;
+  IWICBitmapDecoder* decoder = NULL;
+  IStream* stream = NULL;
+  UINT frame_count = 0;
+  UINT width = 0, height = 0;
+  BYTE* rgb = NULL;
+  WICPixelFormatGUID src_pixel_format = GUID_WICPixelFormatUndefined;
+  const WICFormatImporter* importer = NULL;
+  GUID src_container_format = GUID_NULL_;
+  static const GUID* kAlphaContainers[] = {
+    &GUID_ContainerFormatBmp,
+    &GUID_ContainerFormatPng,
+    &GUID_ContainerFormatTiff,
+    NULL
+  };
+  int has_alpha = 0;
+  int stride;
+
+  IFS(CoInitialize(NULL));
+  IFS(CoCreateInstance(MAKE_REFGUID(CLSID_WICImagingFactory), NULL,
+                       CLSCTX_INPROC_SERVER,
+                       MAKE_REFGUID(IID_IWICImagingFactory),
+                       (LPVOID*)&factory));
+  if (hr == REGDB_E_CLASSNOTREG) {
+    fprintf(stderr,
+            "Couldn't access Windows Imaging Component (are you running "
+            "Windows XP SP3 or newer?). Most formats not available. "
+            "Use -s for the available YUV input.\n");
+  }
+  // Prepare for image decoding.
+  IFS(OpenInputStream(filename, &stream));
+  IFS(IWICImagingFactory_CreateDecoderFromStream(
+          factory, stream, NULL,
+          WICDecodeMetadataCacheOnDemand, &decoder));
+  IFS(IWICBitmapDecoder_GetFrameCount(decoder, &frame_count));
+  if (SUCCEEDED(hr) && frame_count == 0) {
+    fprintf(stderr, "No frame found in input file.\n");
+    hr = E_FAIL;
+  }
+  IFS(IWICBitmapDecoder_GetFrame(decoder, 0, &frame));
+  IFS(IWICBitmapFrameDecode_GetPixelFormat(frame, &src_pixel_format));
+  IFS(IWICBitmapDecoder_GetContainerFormat(decoder, &src_container_format));
+
+  if (keep_alpha) {
+    const GUID** guid;
+    for (guid = kAlphaContainers; *guid != NULL; ++guid) {
+      if (IsEqualGUID(MAKE_REFGUID(src_container_format),
+                      MAKE_REFGUID(**guid))) {
+        has_alpha = HasAlpha(factory, decoder, frame, src_pixel_format);
+        break;
+      }
+    }
+  }
+
+  // Prepare for pixel format conversion (if necessary).
+  IFS(IWICImagingFactory_CreateFormatConverter(factory, &converter));
+
+  for (importer = has_alpha ? kAlphaFormatImporters : kNonAlphaFormatImporters;
+       hr == S_OK && importer->import != NULL; ++importer) {
+    BOOL can_convert;
+    const HRESULT cchr = IWICFormatConverter_CanConvert(
+        converter,
+        MAKE_REFGUID(src_pixel_format),
+        MAKE_REFGUID(*importer->pixel_format),
+        &can_convert);
+    if (SUCCEEDED(cchr) && can_convert) break;
+  }
+  if (importer->import == NULL) hr = E_FAIL;
+
+  IFS(IWICFormatConverter_Initialize(converter, (IWICBitmapSource*)frame,
+                                     importer->pixel_format,
+                                     WICBitmapDitherTypeNone,
+                                     NULL, 0.0, WICBitmapPaletteTypeCustom));
+
+  // Decode.
+  IFS(IWICFormatConverter_GetSize(converter, &width, &height));
+  stride = importer->bytes_per_pixel * width * sizeof(*rgb);
+  if (SUCCEEDED(hr)) {
+    rgb = (BYTE*)malloc(stride * height);
+    if (rgb == NULL)
+      hr = E_OUTOFMEMORY;
+  }
+  IFS(IWICFormatConverter_CopyPixels(converter, NULL,
+                                     stride, stride * height, rgb));
+
+  // WebP conversion.
+  if (SUCCEEDED(hr)) {
+    int ok;
+    pic->width = width;
+    pic->height = height;
+    pic->use_argb = 1;
+    ok = importer->import(pic, rgb, stride);
+    if (!ok) hr = E_FAIL;
+  }
+  if (SUCCEEDED(hr)) {
+    if (metadata != NULL) {
+      hr = ExtractMetadata(factory, frame, metadata);
+      if (FAILED(hr)) {
+        fprintf(stderr, "Error extracting image metadata using WIC!\n");
+      }
+    }
+  }
+
+  // Cleanup.
+  if (converter != NULL) IUnknown_Release(converter);
+  if (frame != NULL) IUnknown_Release(frame);
+  if (decoder != NULL) IUnknown_Release(decoder);
+  if (factory != NULL) IUnknown_Release(factory);
+  if (stream != NULL) IUnknown_Release(stream);
+  free(rgb);
+  return SUCCEEDED(hr);
+}
+#else  // !HAVE_WINCODEC_H
+int ReadPictureWithWIC(const char* const filename,
+                       struct WebPPicture* const pic, int keep_alpha,
+                       struct Metadata* const metadata) {
+  (void)filename;
+  (void)pic;
+  (void)keep_alpha;
+  (void)metadata;
+  fprintf(stderr, "Windows Imaging Component (WIC) support not compiled. "
+                  "Visual Studio and mingw-w64 builds support WIC. Make sure "
+                  "wincodec.h detection is working correctly if using autoconf "
+                  "and HAVE_WINCODEC_H is defined before building.\n");
+  return 0;
+}
+#endif  // HAVE_WINCODEC_H
+
+// -----------------------------------------------------------------------------

examples/wicdec.h

@@ -0,0 +1,34 @@
+// Copyright 2013 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Windows Imaging Component (WIC) decode.
+
+#ifndef WEBP_EXAMPLES_WICDEC_H_
+#define WEBP_EXAMPLES_WICDEC_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct Metadata;
+struct WebPPicture;
+
+// Reads an image from 'filename', returning the decoded output in 'pic'.
+// If 'keep_alpha' is true and the image has an alpha channel, the output is
+// RGBA otherwise it will be RGB.
+// Returns true on success.
+int ReadPictureWithWIC(const char* const filename,
+                       struct WebPPicture* const pic, int keep_alpha,
+                       struct Metadata* const metadata);
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_EXAMPLES_WICDEC_H_

iosbuild.sh

@@ -0,0 +1,117 @@
+#!/bin/bash
+#
+# This script generates 'WebP.framework'. An iOS app can decode WebP images
+# by including 'WebP.framework'.
+#
+# Run ./iosbuild.sh to generate 'WebP.framework' under the current directory
+# (previous build will be erased if it exists).
+#
+# This script is inspired by the build script written by Carson McDonald.
+# (http://www.ioncannon.net/programming/1483/using-webp-to-reduce-native-ios-app-size/).
+
+set -e
+
+# Extract the latest SDK version from the final field of the form: iphoneosX.Y
+readonly SDK=$(xcodebuild -showsdks \
+  | grep iphoneos | sort | tail -n 1 | awk '{print substr($NF, 9)}'
+)
+# Extract Xcode version.
+readonly XCODE=$(xcodebuild -version | grep Xcode | cut -d " " -f2)
+if [[ -z "${XCODE}" ]]; then
+  echo "Xcode not available"
+  exit 1
+fi
+
+readonly OLDPATH=${PATH}
+
+# Add iPhoneOS-V6 to the list of platforms below if you need armv6 support.
+# Note that iPhoneOS-V6 support is not available with the iOS6 SDK.
+readonly PLATFORMS="iPhoneSimulator iPhoneOS-V7 iPhoneOS-V7s iPhoneOS-V7-arm64"
+readonly SRCDIR=$(dirname $0)
+readonly TOPDIR=$(pwd)
+readonly BUILDDIR="${TOPDIR}/iosbuild"
+readonly TARGETDIR="${TOPDIR}/WebP.framework"
+readonly DEVELOPER=$(xcode-select --print-path)
+readonly PLATFORMSROOT="${DEVELOPER}/Platforms"
+readonly LIPO=$(xcrun -sdk iphoneos${SDK} -find lipo)
+LIBLIST=''
+
+if [[ -z "${SDK}" ]]; then
+  echo "iOS SDK not available"
+  exit 1
+elif [[ ${SDK} < 6.0 ]]; then
+  echo "You need iOS SDK version 6.0 or above"
+  exit 1
+else
+  echo "iOS SDK Version ${SDK}"
+fi
+
+rm -rf ${BUILDDIR}
+rm -rf ${TARGETDIR}
+mkdir -p ${BUILDDIR}
+mkdir -p ${TARGETDIR}/Headers/
+
+if [[ ! -e ${SRCDIR}/configure ]]; then
+  if ! (cd ${SRCDIR} && sh autogen.sh); then
+    cat <<EOT
+Error creating configure script!
+This script requires the autoconf/automake and libtool to build. MacPorts can
+be used to obtain these:
+http://www.macports.org/install.php
+EOT
+    exit 1
+  fi
+fi
+
+for PLATFORM in ${PLATFORMS}; do
+  ARCH2=""
+  if [[ "${PLATFORM}" == "iPhoneOS-V7-arm64" ]]; then
+    PLATFORM="iPhoneOS"
+    ARCH="aarch64"
+    ARCH2="arm64"
+  elif [[ "${PLATFORM}" == "iPhoneOS-V7s" ]]; then
+    PLATFORM="iPhoneOS"
+    ARCH="armv7s"
+  elif [[ "${PLATFORM}" == "iPhoneOS-V7" ]]; then
+    PLATFORM="iPhoneOS"
+    ARCH="armv7"
+  elif [[ "${PLATFORM}" == "iPhoneOS-V6" ]]; then
+    PLATFORM="iPhoneOS"
+    ARCH="armv6"
+  else
+    ARCH="i386"
+  fi
+
+  ROOTDIR="${BUILDDIR}/${PLATFORM}-${SDK}-${ARCH}"
+  mkdir -p "${ROOTDIR}"
+
+  DEVROOT="${DEVELOPER}/Toolchains/XcodeDefault.xctoolchain"
+  SDKROOT="${PLATFORMSROOT}/"
+  SDKROOT+="${PLATFORM}.platform/Developer/SDKs/${PLATFORM}${SDK}.sdk/"
+  CFLAGS="-arch ${ARCH2:-${ARCH}} -pipe -isysroot ${SDKROOT} -O3 -DNDEBUG"
+  CFLAGS+=" -miphoneos-version-min=6.0"
+
+  set -x
+  export PATH="${DEVROOT}/usr/bin:${OLDPATH}"
+  ${SRCDIR}/configure --host=${ARCH}-apple-darwin --prefix=${ROOTDIR} \
+    --build=$(${SRCDIR}/config.guess) \
+    --disable-shared --enable-static \
+    --enable-libwebpdecoder --enable-swap-16bit-csp \
+    CFLAGS="${CFLAGS}"
+  set +x
+
+  # run make only in the src/ directory to create libwebpdecoder.a
+  cd src/
+  make V=0
+  make install
+
+  LIBLIST+=" ${ROOTDIR}/lib/libwebpdecoder.a"
+
+  make clean
+  cd ..
+
+  export PATH=${OLDPATH}
+done
+
+cp -a ${SRCDIR}/src/webp/*.h ${TARGETDIR}/Headers/
+${LIPO} -create ${LIBLIST} -output ${TARGETDIR}/WebP

m4/ax_pthread.m4

@@ -82,7 +82,7 @@
 #   modified version of the Autoconf Macro, you may extend this special
 #   exception to the GPL to apply to your modified version as well.
 
-#serial 18
+#serial 21
 
 AU_ALIAS([ACX_PTHREAD], [AX_PTHREAD])
 AC_DEFUN([AX_PTHREAD], [
@@ -103,8 +103,8 @@ if test x"$PTHREAD_LIBS$PTHREAD_CFLAGS" != x; then
         save_LIBS="$LIBS"
         LIBS="$PTHREAD_LIBS $LIBS"
         AC_MSG_CHECKING([for pthread_join in LIBS=$PTHREAD_LIBS with CFLAGS=$PTHREAD_CFLAGS])
-        AC_TRY_LINK_FUNC(pthread_join, ax_pthread_ok=yes)
-        AC_MSG_RESULT($ax_pthread_ok)
+        AC_TRY_LINK_FUNC([pthread_join], [ax_pthread_ok=yes])
+        AC_MSG_RESULT([$ax_pthread_ok])
         if test x"$ax_pthread_ok" = xno; then
                 PTHREAD_LIBS=""
                 PTHREAD_CFLAGS=""
@@ -164,6 +164,20 @@ case ${host_os} in
         ;;
 esac
 
+# Clang doesn't consider unrecognized options an error unless we specify
+# -Werror. We throw in some extra Clang-specific options to ensure that
+# this doesn't happen for GCC, which also accepts -Werror.
+
+AC_MSG_CHECKING([if compiler needs -Werror to reject unknown flags])
+save_CFLAGS="$CFLAGS"
+ax_pthread_extra_flags="-Werror"
+CFLAGS="$CFLAGS $ax_pthread_extra_flags -Wunknown-warning-option -Wsizeof-array-argument"
+AC_COMPILE_IFELSE([AC_LANG_PROGRAM([int foo(void);],[foo()])],
+                  [AC_MSG_RESULT([yes])],
+                  [ax_pthread_extra_flags=
+                   AC_MSG_RESULT([no])])
+CFLAGS="$save_CFLAGS"
+
 if test x"$ax_pthread_ok" = xno; then
 for flag in $ax_pthread_flags; do
 
@@ -178,7 +192,7 @@ for flag in $ax_pthread_flags; do
                 ;;
 
                 pthread-config)
-                AC_CHECK_PROG(ax_pthread_config, pthread-config, yes, no)
+                AC_CHECK_PROG([ax_pthread_config], [pthread-config], [yes], [no])
                 if test x"$ax_pthread_config" = xno; then continue; fi
                 PTHREAD_CFLAGS="`pthread-config --cflags`"
                 PTHREAD_LIBS="`pthread-config --ldflags` `pthread-config --libs`"
@@ -193,7 +207,7 @@ for flag in $ax_pthread_flags; do
         save_LIBS="$LIBS"
         save_CFLAGS="$CFLAGS"
         LIBS="$PTHREAD_LIBS $LIBS"
-        CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+        CFLAGS="$CFLAGS $PTHREAD_CFLAGS $ax_pthread_extra_flags"
 
         # Check for various functions.  We must include pthread.h,
         # since some functions may be macros.  (On the Sequent, we
@@ -219,7 +233,7 @@ for flag in $ax_pthread_flags; do
         LIBS="$save_LIBS"
         CFLAGS="$save_CFLAGS"
 
-        AC_MSG_RESULT($ax_pthread_ok)
+        AC_MSG_RESULT([$ax_pthread_ok])
         if test "x$ax_pthread_ok" = xyes; then
                 break;
         fi
@@ -245,9 +259,9 @@ if test "x$ax_pthread_ok" = xyes; then
                 [attr_name=$attr; break],
                 [])
         done
-        AC_MSG_RESULT($attr_name)
+        AC_MSG_RESULT([$attr_name])
         if test "$attr_name" != PTHREAD_CREATE_JOINABLE; then
-            AC_DEFINE_UNQUOTED(PTHREAD_CREATE_JOINABLE, $attr_name,
+            AC_DEFINE_UNQUOTED([PTHREAD_CREATE_JOINABLE], [$attr_name],
                                [Define to necessary symbol if this constant
                                 uses a non-standard name on your system.])
         fi
@@ -261,45 +275,54 @@ if test "x$ax_pthread_ok" = xyes; then
             if test "$GCC" = "yes"; then
                 flag="-D_REENTRANT"
             else
+                # TODO: What about Clang on Solaris?
                 flag="-mt -D_REENTRANT"
             fi
             ;;
         esac
-        AC_MSG_RESULT(${flag})
+        AC_MSG_RESULT([$flag])
         if test "x$flag" != xno; then
             PTHREAD_CFLAGS="$flag $PTHREAD_CFLAGS"
         fi
 
         AC_CACHE_CHECK([for PTHREAD_PRIO_INHERIT],
-            ax_cv_PTHREAD_PRIO_INHERIT, [
-                AC_LINK_IFELSE([
-                    AC_LANG_PROGRAM([[#include <pthread.h>]], [[int i = PTHREAD_PRIO_INHERIT;]])],
+            [ax_cv_PTHREAD_PRIO_INHERIT], [
+                AC_LINK_IFELSE([AC_LANG_PROGRAM([[#include <pthread.h>]],
+                                                [[int i = PTHREAD_PRIO_INHERIT;]])],
                     [ax_cv_PTHREAD_PRIO_INHERIT=yes],
                     [ax_cv_PTHREAD_PRIO_INHERIT=no])
             ])
         AS_IF([test "x$ax_cv_PTHREAD_PRIO_INHERIT" = "xyes"],
-            AC_DEFINE([HAVE_PTHREAD_PRIO_INHERIT], 1, [Have PTHREAD_PRIO_INHERIT.]))
+            [AC_DEFINE([HAVE_PTHREAD_PRIO_INHERIT], [1], [Have PTHREAD_PRIO_INHERIT.])])
 
         LIBS="$save_LIBS"
         CFLAGS="$save_CFLAGS"
 
-        # More AIX lossage: must compile with xlc_r or cc_r
-        if test x"$GCC" != xyes; then
-          AC_CHECK_PROGS(PTHREAD_CC, xlc_r cc_r, ${CC})
-        else
-          PTHREAD_CC=$CC
+        # More AIX lossage: compile with *_r variant
+        if test "x$GCC" != xyes; then
+            case $host_os in
+                aix*)
+                AS_CASE(["x/$CC"],
+                  [x*/c89|x*/c89_128|x*/c99|x*/c99_128|x*/cc|x*/cc128|x*/xlc|x*/xlc_v6|x*/xlc128|x*/xlc128_v6],
+                  [#handle absolute path differently from PATH based program lookup
+                   AS_CASE(["x$CC"],
+                     [x/*],
+                     [AS_IF([AS_EXECUTABLE_P([${CC}_r])],[PTHREAD_CC="${CC}_r"])],
+                     [AC_CHECK_PROGS([PTHREAD_CC],[${CC}_r],[$CC])])])
+                ;;
+            esac
         fi
-else
-        PTHREAD_CC="$CC"
 fi
 
-AC_SUBST(PTHREAD_LIBS)
-AC_SUBST(PTHREAD_CFLAGS)
-AC_SUBST(PTHREAD_CC)
+test -n "$PTHREAD_CC" || PTHREAD_CC="$CC"
+
+AC_SUBST([PTHREAD_LIBS])
+AC_SUBST([PTHREAD_CFLAGS])
+AC_SUBST([PTHREAD_CC])
 
 # Finally, execute ACTION-IF-FOUND/ACTION-IF-NOT-FOUND:
 if test x"$ax_pthread_ok" = xyes; then
-        ifelse([$1],,AC_DEFINE(HAVE_PTHREAD,1,[Define if you have POSIX threads libraries and header files.]),[$1])
+        ifelse([$1],,[AC_DEFINE([HAVE_PTHREAD],[1],[Define if you have POSIX threads libraries and header files.])],[$1])
         :
 else
         ax_pthread_ok=no

makefile.unix

@@ -2,7 +2,8 @@
 # system, for simple local building of the libraries and tools.
 # It will not install the libraries system-wide, but just create the 'cwebp'
 # and 'dwebp' tools in the examples/ directory, along with the static
-# libraries 'src/libwebp.a' and 'src/mux/libwebpmux.a'.
+# libraries 'src/libwebp.a', 'src/libwebpdecoder.a', 'src/mux/libwebpmux.a' and
+# 'src/demux/libwebpdemux.a'.
 #
 # To build the library and examples, use:
 #    make -f makefile.unix
@@ -10,15 +11,20 @@
 
 #### Customizable part ####
 
-# These flag assume you have libpng and libjpeg installed. If not, either
-# follow below install instructions or just comment out the next lines.
+# These flags assume you have libpng, libjpeg, libtiff and libgif installed. If
+# not, either follow the install instructions below or just comment out the next
+# four lines.
 EXTRA_FLAGS= -DWEBP_HAVE_PNG -DWEBP_HAVE_JPEG -DWEBP_HAVE_TIFF
-EXTRA_LIBS= -lpng -ltiff -ljpeg -lz
+DWEBP_LIBS= -lpng -lz
+CWEBP_LIBS= $(DWEBP_LIBS) -ljpeg -ltiff
+GIF_LIBS = -lgif
+
 ifeq ($(strip $(shell uname)), Darwin)
   # Work around a problem linking tables marked as common symbols,
   # cf., src/enc/yuv.[hc]
   # Failure observed with: gcc 4.2.1 and 4.0.1.
   EXTRA_FLAGS += -fno-common
+  EXTRA_FLAGS += -DHAVE_GLUT_GLUT_H
   EXTRA_FLAGS += -I/opt/local/include
   EXTRA_LIBS  += -L/opt/local/lib
   GL_LIBS = -framework GLUT -framework OpenGL
@@ -26,16 +32,19 @@ else
   GL_LIBS = -lglut -lGL
 endif
 
+
 # To install libraries on Mac OS X:
 # 1. Install MacPorts (http://www.macports.org/install.php)
 # 2. Run "sudo port install jpeg"
 # 3. Run "sudo port install libpng"
 # 4. Run "sudo port install tiff"
+# 5. Run "sudo port install giflib"
 
 # To install libraries on Linux:
 # 1. Run "sudo apt-get install libjpeg62-dev"
 # 2. Run "sudo apt-get install libpng12-dev"
 # 3. Run "sudo apt-get install libtiff4-dev"
+# 4. Run "sudo apt-get install libgif-dev"
 
 # Uncomment for build for 32bit platform
 # Alternatively, you can just use the command
@@ -45,6 +54,9 @@ endif
 # Extra flags to enable experimental features and code
 # EXTRA_FLAGS += -DWEBP_EXPERIMENTAL_FEATURES
 
+# Extra flags to enable byte swap for 16 bit colorspaces.
+# EXTRA_FLAGS += -DWEBP_SWAP_16BIT_CSP
+
 # Extra flags to enable multi-threading
 EXTRA_FLAGS += -DWEBP_USE_THREAD
 EXTRA_LIBS += -lpthread
@@ -54,8 +66,19 @@ EXTRA_FLAGS += -Wextra -Wold-style-definition
 EXTRA_FLAGS += -Wmissing-prototypes
 EXTRA_FLAGS += -Wmissing-declarations
 EXTRA_FLAGS += -Wdeclaration-after-statement
+EXTRA_FLAGS += -Wshadow
 # EXTRA_FLAGS += -Wvla
 
+# AVX2-specific flags:
+ifeq ($(HAVE_AVX2), 1)
+EXTRA_FLAGS += -DWEBP_HAVE_AVX2
+src/dsp/%_avx2.o: EXTRA_FLAGS += -mavx2
+endif
+
+# NEON-specific flags:
+# EXTRA_FLAGS += -march=armv7-a -mfloat-abi=hard -mfpu=neon -mtune=cortex-a8
+# -> seems to make the overall lib slower: -fno-split-wide-types
+
 #### Nothing should normally be changed below this line ####
 
 AR = ar
@@ -66,7 +89,7 @@ CFLAGS = -O3 -DNDEBUG $(EXTRA_FLAGS)
 INSTALL = install
 GROFF = /usr/bin/groff
 COL = /usr/bin/col
-LDFLAGS = $(EXTRA_LIBS) -lm
+LDFLAGS = $(EXTRA_LIBS) $(EXTRA_FLAGS) -lm
 
 DEC_OBJS = \
     src/dec/alpha.o \
@@ -74,24 +97,41 @@ DEC_OBJS = \
     src/dec/frame.o \
     src/dec/idec.o \
     src/dec/io.o \
-    src/dec/layer.o \
     src/dec/quant.o \
     src/dec/tree.o \
     src/dec/vp8.o \
     src/dec/vp8l.o \
     src/dec/webp.o \
 
-DSP_OBJS = \
+DEMUX_OBJS = \
+    src/demux/demux.o \
+
+DSP_DEC_OBJS = \
+    src/dsp/alpha_processing.o \
+    src/dsp/alpha_processing_sse2.o \
     src/dsp/cpu.o \
     src/dsp/dec.o \
+    src/dsp/dec_clip_tables.o \
+    src/dsp/dec_mips32.o \
     src/dsp/dec_neon.o \
     src/dsp/dec_sse2.o \
-    src/dsp/enc.o \
-    src/dsp/enc_sse2.o \
     src/dsp/lossless.o \
+    src/dsp/lossless_mips32.o \
+    src/dsp/lossless_neon.o \
+    src/dsp/lossless_sse2.o \
     src/dsp/upsampling.o \
+    src/dsp/upsampling_neon.o \
     src/dsp/upsampling_sse2.o \
     src/dsp/yuv.o \
+    src/dsp/yuv_mips32.o \
+    src/dsp/yuv_sse2.o \
+
+DSP_ENC_OBJS = \
+    src/dsp/enc.o \
+    src/dsp/enc_avx2.o \
+    src/dsp/enc_mips32.o \
+    src/dsp/enc_neon.o \
+    src/dsp/enc_sse2.o \
 
 ENC_OBJS = \
     src/enc/alpha.o \
@@ -103,53 +143,83 @@ ENC_OBJS = \
     src/enc/frame.o \
     src/enc/histogram.o \
     src/enc/iterator.o \
-    src/enc/layer.o \
     src/enc/picture.o \
+    src/enc/picture_csp.o \
+    src/enc/picture_psnr.o \
+    src/enc/picture_rescale.o \
+    src/enc/picture_tools.o \
     src/enc/quant.o \
     src/enc/syntax.o \
+    src/enc/token.o \
     src/enc/tree.o \
     src/enc/vp8l.o \
     src/enc/webpenc.o \
 
+EX_FORMAT_DEC_OBJS = \
+    examples/jpegdec.o \
+    examples/metadata.o \
+    examples/pngdec.o \
+    examples/tiffdec.o \
+    examples/webpdec.o \
+
 EX_UTIL_OBJS = \
     examples/example_util.o \
 
+GIF2WEBP_UTIL_OBJS = \
+    examples/gif2webp_util.o \
+
 MUX_OBJS = \
-    src/mux/demux.o \
     src/mux/muxedit.o \
     src/mux/muxinternal.o \
     src/mux/muxread.o \
 
-UTILS_OBJS = \
+UTILS_DEC_OBJS = \
     src/utils/bit_reader.o \
-    src/utils/bit_writer.o \
     src/utils/color_cache.o \
     src/utils/filters.o \
     src/utils/huffman.o \
-    src/utils/huffman_encode.o \
-    src/utils/quant_levels.o \
+    src/utils/quant_levels_dec.o \
+    src/utils/random.o \
     src/utils/rescaler.o \
     src/utils/thread.o \
     src/utils/utils.o \
 
-LIBWEBP_OBJS = $(DEC_OBJS) $(DSP_OBJS) $(ENC_OBJS) $(UTILS_OBJS)
+UTILS_ENC_OBJS = \
+    src/utils/bit_writer.o \
+    src/utils/huffman_encode.o \
+    src/utils/quant_levels.o \
+
+LIBWEBPDECODER_OBJS = $(DEC_OBJS) $(DSP_DEC_OBJS) $(UTILS_DEC_OBJS)
+LIBWEBP_OBJS = $(LIBWEBPDECODER_OBJS) $(ENC_OBJS) $(DSP_ENC_OBJS) \
+               $(UTILS_ENC_OBJS)
 LIBWEBPMUX_OBJS = $(MUX_OBJS)
+LIBWEBPDEMUX_OBJS = $(DEMUX_OBJS)
 
 HDRS_INSTALLED = \
     src/webp/decode.h \
+    src/webp/demux.h \
     src/webp/encode.h \
+    src/webp/mux.h \
+    src/webp/mux_types.h \
     src/webp/types.h \
 
 HDRS = \
+    src/dec/alphai.h \
     src/dec/decode_vp8.h \
     src/dec/vp8i.h \
     src/dec/vp8li.h \
     src/dec/webpi.h \
     src/dsp/dsp.h \
     src/dsp/lossless.h \
+    src/dsp/neon.h \
     src/dsp/yuv.h \
+    src/dsp/yuv_tables_sse2.h \
+    src/enc/backward_references.h \
     src/enc/cost.h \
+    src/enc/histogram.h \
     src/enc/vp8enci.h \
+    src/enc/vp8li.h \
+    src/mux/muxi.h \
     src/utils/bit_reader.h \
     src/utils/bit_writer.h \
     src/utils/color_cache.h \
@@ -157,57 +227,87 @@ HDRS = \
     src/utils/huffman.h \
     src/utils/huffman_encode.h \
     src/utils/quant_levels.h \
+    src/utils/quant_levels_dec.h \
+    src/utils/random.h \
     src/utils/rescaler.h \
     src/utils/thread.h \
+    src/utils/utils.h \
     src/webp/format_constants.h \
-    src/webp/mux.h \
     $(HDRS_INSTALLED) \
 
-OUT_LIBS = examples/libexample_util.a src/libwebp.a
+OUT_LIBS = examples/libexample_util.a src/libwebpdecoder.a src/libwebp.a
 OUT_EXAMPLES = examples/cwebp examples/dwebp
+EXTRA_EXAMPLES = examples/gif2webp examples/vwebp examples/webpmux
 
 OUTPUT = $(OUT_LIBS) $(OUT_EXAMPLES)
 ifeq ($(MAKECMDGOALS),clean)
-  OUTPUT += examples/vwebp examples/webpmux src/mux/libwebpmux.a
+  OUTPUT += $(EXTRA_EXAMPLES)
+  OUTPUT += src/demux/libwebpdemux.a src/mux/libwebpmux.a
+  OUTPUT += examples/libgif2webp_util.a
 endif
 
-all: ex
+ex: $(OUT_EXAMPLES)
+all: ex $(EXTRA_EXAMPLES)
+
+$(EX_FORMAT_DEC_OBJS): %.o: %.h
+
+# special dependencies:
+#   tree.c/vp8.c/bit_reader.c <-> bit_reader_inl.h, endian_inl.h
+#   bit_writer.c <-> endian_inl.h
+src/dec/tree.o: src/utils/bit_reader_inl.h src/utils/endian_inl.h
+src/dec/vp8.o: src/utils/bit_reader_inl.h src/utils/endian_inl.h
+src/utils/bit_reader.o: src/utils/bit_reader_inl.h src/utils/endian_inl.h
+src/utils/bit_writer.o: src/utils/endian_inl.h
 
 %.o: %.c $(HDRS)
 	$(CC) $(CFLAGS) $(CPPFLAGS) -c $< -o $@
 
 examples/libexample_util.a: $(EX_UTIL_OBJS)
+examples/libgif2webp_util.a: $(GIF2WEBP_UTIL_OBJS)
+src/libwebpdecoder.a: $(LIBWEBPDECODER_OBJS)
 src/libwebp.a: $(LIBWEBP_OBJS)
 src/mux/libwebpmux.a: $(LIBWEBPMUX_OBJS)
+src/demux/libwebpdemux.a: $(LIBWEBPDEMUX_OBJS)
 
 %.a:
 	$(AR) $(ARFLAGS) $@ $^
 
-ex: $(OUT_EXAMPLES)
-
-examples/cwebp: examples/cwebp.o
+examples/cwebp: examples/cwebp.o $(EX_FORMAT_DEC_OBJS)
 examples/dwebp: examples/dwebp.o
+examples/gif2webp: examples/gif2webp.o
 examples/vwebp: examples/vwebp.o
 examples/webpmux: examples/webpmux.o
 
-examples/cwebp: src/libwebp.a
-examples/dwebp: examples/libexample_util.a src/libwebp.a
-examples/vwebp: examples/libexample_util.a src/mux/libwebpmux.a src/libwebp.a
+examples/cwebp: examples/libexample_util.a src/libwebp.a
+examples/cwebp: EXTRA_LIBS += $(CWEBP_LIBS)
+examples/dwebp: examples/libexample_util.a src/libwebpdecoder.a
+examples/dwebp: EXTRA_LIBS += $(DWEBP_LIBS)
+examples/gif2webp: examples/libexample_util.a examples/libgif2webp_util.a
+examples/gif2webp: src/mux/libwebpmux.a src/libwebp.a
+examples/gif2webp: EXTRA_LIBS += $(GIF_LIBS)
+examples/gif2webp: EXTRA_FLAGS += -DWEBP_HAVE_GIF
+examples/vwebp: examples/libexample_util.a src/demux/libwebpdemux.a
+examples/vwebp: src/libwebp.a
 examples/vwebp: EXTRA_LIBS += $(GL_LIBS)
-examples/webpmux: examples/libexample_util.a src/mux/libwebpmux.a src/libwebp.a
+examples/vwebp: EXTRA_FLAGS += -DWEBP_HAVE_GL
+examples/webpmux: examples/libexample_util.a src/mux/libwebpmux.a
+examples/webpmux: src/libwebpdecoder.a
 
-$(OUT_EXAMPLES) examples/vwebp examples/webpmux:
+$(OUT_EXAMPLES) $(EXTRA_EXAMPLES):
 	$(CC) -o $@ $^ $(LDFLAGS)
 
 dist: DESTDIR := dist
+dist: OUT_EXAMPLES += $(EXTRA_EXAMPLES)
 dist: all
 	$(INSTALL) -m755 -d $(DESTDIR)/include/webp \
-	           $(DESTDIR)/doc $(DESTDIR)/lib
-	$(INSTALL) -m755 -s $(OUT_EXAMPLES) $(DESTDIR)
+	           $(DESTDIR)/bin $(DESTDIR)/doc $(DESTDIR)/lib
+	$(INSTALL) -m755 -s $(OUT_EXAMPLES) $(DESTDIR)/bin
 	$(INSTALL) -m644 $(HDRS_INSTALLED) $(DESTDIR)/include/webp
 	$(INSTALL) -m644 src/libwebp.a $(DESTDIR)/lib
+	$(INSTALL) -m644 src/demux/libwebpdemux.a $(DESTDIR)/lib
+	$(INSTALL) -m644 src/mux/libwebpmux.a $(DESTDIR)/lib
 	umask 022; \
-	for m in man/[cd]webp.1; do \
+	for m in man/[cdv]webp.1 man/gif2webp.1 man/webpmux.1; do \
 	  basenam=$$(basename $$m .1); \
 	  $(GROFF) -t -e -man -T utf8 $$m \
 	    | $(COL) -bx >$(DESTDIR)/doc/$${basenam}.txt; \
@@ -219,6 +319,7 @@ clean:
 	$(RM) $(OUTPUT) *~ \
               examples/*.o examples/*~ \
               src/dec/*.o src/dec/*~ \
+              src/demux/*.o src/demux/*~ \
               src/dsp/*.o src/dsp/*~ \
               src/enc/*.o src/enc/*~ \
               src/mux/*.o src/mux/*~ \
@@ -234,7 +335,8 @@ superclean: clean
 	$(RM) Makefile */Makefile */*/Makefile
 	$(RM) Makefile.in */Makefile.in */*/Makefile.in
 	$(RM) config.log autom4te.cache libtool config.h stamp-h1
-	$(RM) aclocal.m4 compile config.guess config.h.in config.sub config.status
+	$(RM) aclocal.m4 compile
+	$(RM) config.guess config.h.in config.sub config.status
 	$(RM) configure depcomp install-sh ltmain.sh missing src/libwebp.pc
 	$(RM) m4/*
 

man/Makefile.am

@@ -2,4 +2,10 @@ man_MANS = cwebp.1 dwebp.1
 if WANT_MUX
   man_MANS += webpmux.1
 endif
+if BUILD_GIF2WEBP
+  man_MANS += gif2webp.1
+endif
+if BUILD_VWEBP
+  man_MANS += vwebp.1
+endif
 EXTRA_DIST = $(man_MANS)

man/cwebp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH CWEBP 1 "July 19, 2012"
+.TH CWEBP 1 "Oct 13, 2014"
 .SH NAME
 cwebp \- compress an image file to a WebP file
 .SH SYNOPSIS
@@ -12,13 +12,19 @@ This manual page documents the
 command.
 .PP
 \fBcwebp\fP compresses an image using the WebP format.
-Input format can be either PNG, JPEG, TIFF or raw Y'CbCr samples.
+Input format can be either PNG, JPEG, TIFF, WebP or raw Y'CbCr samples.
 .SH OPTIONS
 The basic options are:
 .TP
-.B \-o string
+.BI \-o " string
 Specify the name of the output WebP file. If omitted, \fBcwebp\fP will
 perform compression but only report statistics.
+Using "\-" as output name will direct output to 'stdout'.
+.TP
+.BI \-\- " string
+Explicitly specify the input file. This option is useful if the input
+file starts with an '\-' for instance. This option must appear \fBlast\fP.
+Any other options afterward will be ignored.
 .TP
 .B \-h, \-help
 A short usage summary.
@@ -29,24 +35,39 @@ A summary of all the possible options.
 .B \-version
 Print the version number (as major.minor.revision) and exit.
 .TP
-.B \-q float
-Specify the compression factor for RGB channels between 0 and 100. A small
-factor produces a smaller file with lower quality. Best quality is achieved
-using a value of 100. The default is 75.
+.BI \-q " float
+Specify the compression factor for RGB channels between 0 and 100. The default
+is 75.
+.br
+In case of lossy compression (default), a small factor produces a smaller file
+with lower quality. Best quality is achieved by using a value of 100.
+.br
+In case of lossless compression (specified by the \-lossless option), a small
+factor enables faster compression speed, but produces a larger file. Maximum
+compression is achieved by using a value of 100.
+.\" TODO(jzern): restore post-v0.4.1
+.\" .TP
+.\" .BI \-z " int
+.\" Switch on \fBlossless\fP compression mode with the specified level between 0
+.\" and 9, with level 0 being the fastest, 9 being the slowest. Fast mode
+.\" produces larger file size than slower ones. A good default is \-z 6.
+.\" This option is actually a shortcut for some predefined settings for quality
+.\" and method. If options \-q  or \-m are subsequently used, they will invalidate
+.\" the effect of this \-z option.
 .TP
-.B \-alpha_q int
+.BI \-alpha_q " int
 Specify the compression factor for alpha compression between 0 and 100.
 Lossless compression of alpha is achieved using a value of 100, while the lower
 values result in a lossy compression. The default is 100.
 .TP
-.B \-f int
+.BI \-f " int
 Specify the strength of the deblocking filter, between 0 (no filtering)
 and 100 (maximum filtering). A value of 0 will turn off any filtering.
 Higher value will increase the strength of the filtering process applied
 after decoding the picture. The higher the value the smoother the picture will
 appear. Typical values are usually in the range of 20 to 50.
 .TP
-.B \-preset string
+.BI \-preset " string
 Specify a set of pre-defined parameters to suit a particular type of
 source material. Possible values are:  \fBdefault\fP, \fBphoto\fP,
 \fBpicture\fP, \fBdrawing\fP, \fBicon\fP, \fBtext\fP. Since
@@ -54,22 +75,41 @@ source material. Possible values are:  \fBdefault\fP, \fBphoto\fP,
 \fB\-q\fP one), this option should preferably appear first in the
 order of the arguments.
 .TP
-.B \-sns int
+.BI \-sns " int
 Specify the amplitude of the spatial noise shaping. Spatial noise shaping
 (or \fBsns\fP for short) refers to a general collection of built-in algorithms
 used to decide which area of the picture should use relatively less bits,
 and where else to better transfer these bits. The possible range goes from
 0 (algorithm is off) to 100 (the maximal effect). The default value is 80.
 .TP
-.B \-m int
+.BI \-m " int
 Specify the compression method to use. This parameter controls the
 trade off between encoding speed and the compressed file size and quality.
 Possible values range from 0 to 6. Default value is 4.
 When higher values are used, the encoder will spend more time inspecting
 additional encoding possibilities and decide on the quality gain.
-Lower value can result is faster processing time at the expense of
+Lower value can result in faster processing time at the expense of
 larger file size and lower compression quality.
 .TP
+.B \-jpeg_like
+Change the internal parameter mapping to better match the expected size
+of JPEG compression. This flag will generally produce an output file of
+similar size to its JPEG equivalent (for the same \fB\-q\fP setting), but
+with less visual distortion.
+.TP
+.B \-mt
+Use multi-threading for encoding, if possible. This option is only effective
+when using lossy compression on a source with a transparency channel.
+.TP
+.B \-low_memory
+Reduce memory usage of lossy encoding by saving four times the compressed
+size (typically). This will make the encoding slower and the output slightly
+different in size and distortion. This flag is only effective for methods
+3 and up, and is off by default. Note that leaving this flag off will have
+some side effects on the bitstream: it forces certain bitstream features
+like number of partitions (forced to 1). Note that a more detailed report
+of bitstream size is printed by \fBcwebp\fP when using this option.
+.TP
 .B \-af
 Turns auto-filter on. This algorithm will spend additional time optimizing
 the filtering strength to reach a well-balanced quality.
@@ -77,19 +117,25 @@ the filtering strength to reach a well-balanced quality.
 .SH ADDITIONAL OPTIONS
 More advanced options are:
 .TP
-.B \-sharpness int
+.BI \-sharpness " int
 Specify the sharpness of the filtering (if used).
 Range is 0 (sharpest) to 7 (least sharp). Default is 0.
 .TP
 .B \-strong
-Use a stronger filtering than the default one (if filtering is being
-used thanks to the \fB\-f\fP option). Strong filtering is off by default.
+Use strong filtering (if filtering is being used thanks to the
+\fB\-f\fP option). Strong filtering is on by default.
 .TP
-.B \-segments int
+.B \-nostrong
+Disable strong filtering (if filtering is being used thanks to the
+\fB\-f\fP option) and use simple filtering instead.
+.TP
+.BI \-segments " int
 Change the number of partitions to use during the segmentation of the
 sns algorithm. Segments should be in range 1 to 4. Default value is 4.
+This option has no effect for methods 3 and up, unless \fB\-low_memory\fP
+is used.
 .TP
-.B \-partition_limit int
+.BI \-partition_limit " int
 Degrade quality by limiting the number of bits used by some macroblocks.
 Range is 0 (no degradation, the default) to 100 (full degradation).
 Useful values are usually around 30-70 for moderately large images.
@@ -108,39 +154,45 @@ If using \fB-partition_limit\fP is not enough to meet the 512k constraint, one
 should use less segments in order to save more header bits per macroblock.
 See the \fB-segments\fP option.
 .TP
-.B \-size int
+.BI \-size " int
 Specify a target size (in bytes) to try and reach for the compressed output.
 Compressor will make several pass of partial encoding in order to get as
 close as possible to this target.
 .TP
-.B \-psnr float
+.BI \-psnr " float
 Specify a target PSNR (in dB) to try and reach for the compressed output.
 Compressor will make several pass of partial encoding in order to get as
 close as possible to this target.
 .TP
-.B \-pass int
+.BI \-pass " int
 Set a maximum number of passes to use during the dichotomy used by
 options \fB\-size\fP or \fB\-psnr\fP. Maximum value is 10.
 .TP
-.B \-crop x_position y_position width height
+.BI \-resize " width height
+Resize the source to a rectangle with size \fBwidth\fP x \fBheight\fP.
+If either (but not both) of the \fBwidth\fP or \fBheight\fP parameters is 0,
+the value will be calculated preserving the aspect-ratio.
+.TP
+.BI \-crop " x_position y_position width height
 Crop the source to a rectangle with top-left corner at coordinates
 (\fBx_position\fP, \fBy_position\fP) and size \fBwidth\fP x \fBheight\fP.
 This cropping area must be fully contained within the source rectangle.
 .TP
-.B \-s width height
+.BI \-s " width height
 Specify that the input file actually consists of raw Y'CbCr samples following
 the ITU-R BT.601 recommendation, in 4:2:0 linear format.
 The luma plane has size \fBwidth\fP x \fBheight\fP.
 .TP
-.B \-map int
+.BI \-map " int
 Output additional ASCII-map of encoding information. Possible map values
 range from 1 to 6. This is only meant to help debugging.
 .TP
-.B \-pre int
-Specify a pre-processing filter. This option is a placeholder
-and has currently no effect.
+.BI \-pre " int
+Specify some pre-processing steps. Using a value of '2' will trigger
+quality-dependent pseudo-random dithering during RGBA->YUVA conversion
+(lossy compression only).
 .TP
-.B \-alpha_filter string
+.BI \-alpha_filter " string
 Specify the predictive filtering method for the alpha plane. One of 'none',
 \&'fast' or 'best', in increasing complexity and slowness order. Default is
 \&'fast'. Internally, alpha filtering is performed using four possible
@@ -148,7 +200,7 @@ predictions (none, horizontal, vertical, gradient). The 'best' mode will try
 each mode in turn and pick the one which gives the smaller size. The 'fast'
 mode will just try to form an a-priori guess without testing all modes.
 .TP
-.B \-alpha_method int
+.BI \-alpha_method " int
 Specify the algorithm used for alpha compression: 0 or 1. Algorithm 0 denotes
 no compression, 1 uses WebP lossless format for compression. The default is 1.
 .TP
@@ -156,16 +208,29 @@ no compression, 1 uses WebP lossless format for compression. The default is 1.
 Modify unseen RGB values under fully transparent area, to help compressibility.
 The default is off.
 .TP
+.BI \-blend_alpha " int
+This option blends the alpha channel (if present) with the source using the
+background color specified in hexadecimal as 0xrrggbb. The alpha channel is
+afterward reset to the opaque value 255.
+.TP
 .B \-noalpha
 Using this option will discard the alpha channel.
 .TP
 .B \-lossless
 Encode the image without any loss.
 .TP
-.B \-hint string
+.BI \-hint " string
 Specify the hint about input image type. Possible values are:
 \fBphoto\fP, \fBpicture\fP or \fBgraph\fP.
 .TP
+.BI \-metadata " string
+A comma separated list of metadata to copy from the input to the output if
+present.
+Valid values: \fBall\fP, \fBnone\fP, \fBexif\fP, \fBicc\fP, \fBxmp\fP.
+The default is \fBnone\fP.
+
+Note: each input format may not support all combinations.
+.TP
 .B \-noasm
 Disable all assembly optimizations.
 .TP
@@ -176,7 +241,12 @@ Print extra information (encoding time in particular).
 Compute and report average PSNR (Peak-Signal-To-Noise ratio).
 .TP
 .B \-print_ssim
-Compute and report average SSIM (structural similarity metric)
+Compute and report average SSIM (structural similarity
+metric, see http://en.wikipedia.org/wiki/SSIM for additional details).
+.TP
+.B \-print_lsim
+Compute and report local similarity metric (sum of lowest error amongst the
+collocated pixel neighbors).
 .TP
 .B \-progress
 Report encoding progress in percent.
@@ -199,7 +269,9 @@ cwebp \-q 50 -lossless picture.png \-o picture_lossless.webp
 .br
 cwebp \-q 70 picture_with_alpha.png \-o picture_with_alpha.webp
 .br
-cwebp \-sns 70 \-f 50 \-strong \-af \-size 60000 picture.png \-o picture.webp
+cwebp \-sns 70 \-f 50 \-size 60000 picture.png \-o picture.webp
+.br
+cwebp \-o picture.webp \-\- \-\-\-picture.png
 
 .SH AUTHORS
 \fBcwebp\fP was written by the WebP team.
@@ -210,7 +282,8 @@ This manual page was written by Pascal Massimino <pascal.massimino@gmail.com>,
 for the Debian project (and may be used by others).
 
 .SH SEE ALSO
-.BR dwebp (1).
+.BR dwebp (1),
+.BR gif2webp (1)
 .br
 Please refer to http://developers.google.com/speed/webp/ for additional
 information.

man/dwebp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH DWEBP 1 "August 2, 2012"
+.TH DWEBP 1 "July 22, 2014"
 .SH NAME
 dwebp \- decompress a WebP file to an image file
 .SH SYNOPSIS
@@ -21,8 +21,21 @@ Print usage summary.
 .B \-version
 Print the version number (as major.minor.revision) and exit.
 .TP
-.B \-o string
+.BI \-o " string
 Specify the name of the output file (as PNG format by default).
+Using "-" as output name will direct output to 'stdout'.
+.TP
+.BI \-\- " string
+Explicitly specify the input file. This option is useful if the input
+file starts with an '\-' for instance. This option must appear \fBlast\fP.
+Any other options afterward will be ignored. If the input file is "\-",
+the data will be read from \fIstdin\fP instead of a file.
+.TP
+.B \-bmp
+Change the output format to uncompressed BMP.
+.TP
+.B \-tiff
+Change the output format to uncompressed TIFF.
 .TP
 .B \-pam
 Change the output format to PAM (retains alpha).
@@ -31,9 +44,15 @@ Change the output format to PAM (retains alpha).
 Change the output format to PPM (discards alpha).
 .TP
 .B \-pgm
-Change the output format to PGM. The output consist of luma/chroma
-samples instead of RGB, using the ICM4 layout. This option is mainly
-for verification and debugging purpose.
+Change the output format to PGM. The output consists of luma/chroma
+samples instead of RGB, using the IMC4 layout. This option is mainly
+for verification and debugging purposes.
+.TP
+.B \-yuv
+Change the output format to raw YUV. The output consists of
+luma/chroma-U/chroma-V samples instead of RGB, saved sequentially as
+individual planes. This option is mainly for verification and debugging
+purposes.
 .TP
 .B \-nofancy
 Don't use the fancy upscaler for YUV420. This may lead to jaggy
@@ -42,20 +61,38 @@ edges (especially the red ones), but should be faster.
 .B \-nofilter
 Don't use the in-loop filtering process even if it is required by
 the bitstream. This may produce visible blocks on the non-compliant output,
-but will make the decoding faster.
+but it will make the decoding faster.
+.TP
+.BI \-dither " strength
+Specify a dithering \fBstrength\fP between 0 and 100. Dithering is a
+post-processing effect applied to chroma components in lossy compression.
+It helps by smoothing gradients and avoiding banding artifacts.
+.\" TODO(jzern): restore post-v0.4.1
+.\" .TP
+.\" .BI \-alpha_dither
+.\" If the compressed file contains a transparency plane that was quantized
+.\" during compression, this flag will allow dithering the reconstructed plane
+.\" in order to generate smoother transparency gradients.
+.TP
+.B \-nodither
+Disable all dithering (default).
 .TP
 .B \-mt
 Use multi-threading for decoding, if possible.
 .TP
-.B \-crop x_position y_position width height
+.BI \-crop " x_position y_position width height
 Crop the decoded picture to a rectangle with top-left corner at coordinates
 (\fBx_position\fP, \fBy_position\fP) and size \fBwidth\fP x \fBheight\fP.
 This cropping area must be fully contained within the source rectangle.
 The top-left corner will be snapped to even coordinates if needed.
 This option is meant to reduce the memory needed for cropping large images.
 Note: the cropping is applied \fIbefore\fP any scaling.
+.\" TODO(jzern): restore post-v0.4.1
+.\" .TP
+.\" .B \-flip
+.\" Flip decoded image vertically (can be useful for OpenGL textures for instance).
 .TP
-.B \-scale width height
+.BI \-scale " width height
 Rescale the decoded picture to dimension \fBwidth\fP x \fBheight\fP. This
 option is mostly intended to reducing the memory needed to decode large images,
 when only a small version is needed (thumbnail, preview, etc.).  Note: scaling
@@ -78,6 +115,10 @@ http://www.webmproject.org/code/contribute/submitting-patches/
 dwebp picture.webp \-o output.png
 .br
 dwebp picture.webp \-ppm \-o output.ppm
+.br
+dwebp \-o output.ppm \-\- \-\-\-picture.webp
+.br
+cat picture.webp | dwebp \-o \- \-\- \- > output.ppm
 
 .SH AUTHORS
 \fBdwebp\fP was written by the WebP team.
@@ -88,7 +129,9 @@ This manual page was written by Pascal Massimino <pascal.massimino@gmail.com>,
 for the Debian project (and may be used by others).
 
 .SH SEE ALSO
-.BR cwebp (1).
+.BR cwebp (1),
+.BR gif2webp (1),
+.BR webpmux (1)
 .br
 Please refer to http://developers.google.com/speed/webp/ for additional
 information.

man/gif2webp.1

@@ -0,0 +1,144 @@
+.\"                                      Hey, EMACS: -*- nroff -*-
+.TH GIF2WEBP 1 "March 7, 2014"
+.SH NAME
+gif2webp \- Convert a GIF image to WebP
+.SH SYNOPSIS
+.B gif2webp
+.RI [ options ] " input_file.gif \-o output_file.webp
+.br
+.SH DESCRIPTION
+This manual page documents the
+.B gif2webp
+command.
+.PP
+\fBgif2webp\fP converts a GIF image to a WebP image.
+.SH OPTIONS
+The basic options are:
+.TP
+.BI \-o " string
+Specify the name of the output WebP file. If omitted, \fBgif2webp\fP will
+perform conversion but only report statistics.
+Using "\-" as output name will direct output to 'stdout'.
+.TP
+.B \-h, \-help
+Usage information.
+.TP
+.B \-version
+Print the version number (as major.minor.revision) and exit.
+.TP
+.B \-lossy
+Encode the image using lossy compression.
+.TP
+.B \-mixed
+Mixed compression mode: optimize compression of the image by picking either
+lossy or lossless compression for each frame heuristically.
+.TP
+.BI \-q " float
+Specify the compression factor for RGB channels between 0 and 100. The default
+is 75.
+.br
+In case of lossless compression (default), a small factor enables faster
+compression speed, but produces a larger file. Maximum compression is achieved
+by using a value of 100.
+.br
+In case of lossy compression (specified by the \-lossy option), a small factor
+produces a smaller file with lower quality. Best quality is achieved by using a
+value of 100.
+.TP
+.BI \-m " int
+Specify the compression method to use. This parameter controls the
+trade off between encoding speed and the compressed file size and quality.
+Possible values range from 0 to 6. Default value is 4.
+When higher values are used, the encoder will spend more time inspecting
+additional encoding possibilities and decide on the quality gain.
+Lower value can result is faster processing time at the expense of
+larger file size and lower compression quality.
+.TP
+.BI \-kmin " int
+.TP
+.BI \-kmax " int
+Specify the minimum and maximum distance between consecutive key frames
+(independently decodable frames) in the output animation. The tool will insert
+some key frames into the output animation as needed so that this criteria is
+satisfied.
+.br
+A 'kmin' value of 0 will turn off insertion of key frames.
+Typical values are in the range 3 to 30. Default values are kmin = 9,
+kmax = 17 for lossless compression and kmin = 3, kmax = 5 for lossy compression.
+.br
+These two options are relevant only for animated images with large number of
+frames (>50).
+.br
+When lower values are used, more frames will be converted to key frames. This
+may lead to smaller number of frames required to decode a frame on average,
+thereby improving the decoding performance. But this may lead to slightly bigger
+file sizes.
+Higher values may lead to worse decoding performance, but smaller file sizes.
+.br
+Some restrictions:
+.br
+(i) kmin < kmax,
+.br
+(ii) kmin >= kmax / 2 + 1 and
+.br
+(iii) kmax - kmin <= 30.
+.br
+If any of these restrictions are not met, they will be enforced automatically.
+.TP
+.BI \-metadata " string
+A comma separated list of metadata to copy from the input to the output if
+present.
+Valid values: \fBall\fP, \fBnone\fP, \fBicc\fP, \fBxmp\fP.
+The default is \fBxmp\fP.
+.TP
+.BI \-f " int
+For lossy encoding only (specified by the \-lossy option). Specify the strength
+of the deblocking filter, between 0 (no filtering) and 100 (maximum filtering).
+A value of 0 will turn off any filtering. Higher value will increase the
+strength of the filtering process applied after decoding the picture. The higher
+the value the smoother the picture will appear. Typical values are usually in
+the range of 20 to 50.
+.TP
+.B \-mt
+Use multi-threading for encoding, if possible. This option is only effective
+when using lossy compression.
+.TP
+.B \-v
+Print extra information.
+.TP
+.B \-quiet
+Do not print anything.
+
+.SH BUGS
+Please report all bugs to our issue tracker:
+http://code.google.com/p/webp/issues
+.br
+Patches welcome! See this page to get started:
+http://www.webmproject.org/code/contribute/submitting-patches/
+
+.SH EXAMPLES
+gif2webp picture.gif \-o picture.webp
+.br
+gif2webp \-q 70 picture.gif \-o picture.webp
+.br
+gif2webp \-lossy \-m 3 picture.gif \-o picture_lossy.webp
+.br
+gif2webp \-lossy \-f 50 picture.gif \-o picture.webp
+.br
+gif2webp \-q 70 \-o picture.webp \-\- \-\-\-picture.gif
+
+.SH AUTHORS
+\fBgif2webp\fP was written by the WebP team.
+.br
+The latest source tree is available at http://www.webmproject.org/code
+.PP
+This manual page was written by Urvang Joshi <urvang@google.com>, for the
+Debian project (and may be used by others).
+
+.SH SEE ALSO
+.BR cwebp (1),
+.BR dwebp (1),
+.BR webpmux (1)
+.br
+Please refer to http://developers.google.com/speed/webp/ for additional
+information.

man/vwebp.1

@@ -0,0 +1,91 @@
+.\"                                      Hey, EMACS: -*- nroff -*-
+.TH VWEBP 1 "July 23, 2014"
+.SH NAME
+vwebp \- decompress a WebP file and display it in a window
+.SH SYNOPSIS
+.B vwebp
+.RI [ options ] " input_file.webp
+.br
+.SH DESCRIPTION
+This manual page documents the
+.B vwebp
+command.
+.PP
+\fBvwebp\fP decompresses a WebP file and displays it in a window using OpenGL.
+.SH OPTIONS
+.TP
+.B \-h
+Print usage summary.
+.TP
+.B \-version
+Print version number and exit.
+.TP
+.B \-noicc
+Don't use the ICC profile if present.
+.TP
+.B \-nofancy
+Don't use the fancy YUV420 upscaler.
+.TP
+.B \-nofilter
+Disable in-loop filtering.
+.TP
+.BI \-dither " strength
+Specify a dithering \fBstrength\fP between 0 and 100. Dithering is a
+post-processing effect applied to chroma components in lossy compression.
+It helps by smoothing gradients and avoiding banding artifacts. Default: 50.
+.\" TODO(jzern): restore post-v0.4.1
+.\" .TP
+.\" .BI \-noalphadither
+.\" By default, quantized transparency planes are dithered during decompression,
+.\" to smooth the gradients. This flag will prevent this dithering.
+.TP
+.B \-mt
+Use multi-threading for decoding, if possible.
+.TP
+.B \-info
+Display image information on top of the decoded image.
+.TP
+.BI \-\- " string
+Explicitly specify the input file. This option is useful if the input
+file starts with an '\-' for instance. This option must appear \fBlast\fP.
+Any other options afterward will be ignored. If the input file is "\-",
+the data will be read from \fIstdin\fP instead of a file.
+.TP
+
+.SH KEYBOARD SHORTCUTS
+.TP
+.B 'c'
+Toggle use of color profile.
+.TP
+.B 'i'
+Overlay file information.
+.TP
+.B 'q' / 'Q' / ESC
+Quit.
+
+.SH BUGS
+Please report all bugs to our issue tracker:
+http://code.google.com/p/webp/issues
+.br
+Patches welcome! See this page to get started:
+http://www.webmproject.org/code/contribute/submitting-patches/
+
+.SH EXAMPLES
+vwebp picture.webp
+.br
+vwebp picture.webp -mt -dither 0
+.br
+vwebp \-\- \-\-\-picture.webp
+
+.SH AUTHORS
+\fBvwebp\fP was written by the WebP team.
+.br
+The latest source tree is available at http://www.webmproject.org/code
+.PP
+This manual page was written for the Debian project (and may be used by others).
+
+.SH SEE ALSO
+.BR dwebp (1)
+.br
+Please refer to http://developers.google.com/speed/webp/ for additional
+information.

man/webpmux.1

@@ -1,7 +1,8 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH WEBPMUX 1 "January 24, 2012"
+.TH WEBPMUX 1 "August 28, 2014"
 .SH NAME
-webpmux \- command line tool to create WebP Mux/container file.
+webpmux \- create animated WebP files from non\-animated WebP images, extract
+frames from animated WebP images, and manage XMP/EXIF metadata and ICC profile.
 .SH SYNOPSIS
 .B webpmux \-get
 .I GET_OPTIONS
@@ -21,74 +22,98 @@ webpmux \- command line tool to create WebP Mux/container file.
 .B \-o
 .I OUTPUT
 .br
-.B webpmux \-tile
-.I TILE_OPTIONS
-.B [\-tile...] \-o
-.I OUTPUT
-.br
 .B webpmux \-frame
 .I FRAME_OPTIONS
-.B [\-frame...] \-loop
+.B [ \-frame ... ] [ \-loop
 .I LOOP_COUNT
-.B \-o
+.B ]
+.br
+.RS 8
+.B [ \-bgcolor
+.I BACKGROUND_COLOR
+.B ] \-o
 .I OUTPUT
+.RE
 .br
 .B webpmux \-info
 .I INPUT
 .br
 .B webpmux [\-h|\-help]
+.br
+.B webpmux \-version
 .SH DESCRIPTION
 This manual page documents the
 .B webpmux
 command.
 .PP
-\fBwebpmux\fP can be used to create a WebP container file
-and extract/strip relevant data from the container file.
+\fBwebpmux\fP can be used to create/extract from animated WebP files, as well as
+to add/extract/strip XMP/EXIF metadata and ICC profile.
 .SH OPTIONS
 .SS GET_OPTIONS (\-get):
 .TP
 .B icc
-Get ICC Color profile.
+Get ICC profile.
+.TP
+.B exif
+Get EXIF metadata.
 .TP
 .B xmp
 Get XMP metadata.
 .TP
-.B tile n
-Get nth tile.
-.TP
-.B frame n
-Get nth frame.
+.BI frame " n
+Get nth frame from an animated image. (n = 0 has a special meaning: last frame).
 
 .SS SET_OPTIONS (\-set)
 .TP
-.B icc
-Set ICC Color profile.
+.BI icc " file.icc
+Set ICC profile.
+.P
+Where: 'file.icc' contains the ICC profile to be set.
 .TP
-.B xmp
+.BI exif " file.exif
+Set EXIF metadata.
+.P
+Where: 'file.exif' contains the EXIF metadata to be set.
+.TP
+.BI xmp " file.xmp
 Set XMP metadata.
+.P
+Where: 'file.xmp' contains the XMP metadata to be set.
 
 .SS STRIP_OPTIONS (\-strip)
 .TP
 .B icc
-Strip ICC Color profile.
+Strip ICC profile.
+.TP
+.B exif
+Strip EXIF metadata.
 .TP
 .B xmp
 Strip XMP metadata.
 
-.SS TILE_OPTIONS (\-tile)
-.TP
-.B file_i +xi+yi
-Where: 'file_i' is the i'th tile (webp format) and 'xi','yi' specify the image
-offset for this tile.
-
 .SS FRAME_OPTIONS (\-frame)
+Create an animated WebP file from multiple (non\-animated) WebP images.
 .TP
-.B file_i +xi+yi+di
-Where: 'file_i' is the i'th frame (webp format), 'xi','yi' specify the image
-offset for this frame and 'di' is the pause duration before next frame.
+.I file_i +di[+xi+yi[+mi[bi]]]
+Where: 'file_i' is the i'th frame (WebP format), 'xi','yi' specify the image
+offset for this frame, 'di' is the pause duration before next frame, 'mi' is
+the dispose method for this frame (0 for NONE or 1 for BACKGROUND) and 'bi' is
+the blending method for this frame (+b for BLEND or \-b for NO_BLEND).
+Argument 'bi' can be omitted and will default to +b (BLEND).
+Also, 'mi' can be omitted if 'bi' is omitted and will default to 0 (NONE).
+Finally, if 'mi' and 'bi' are omitted then 'xi' and 'yi' can be omitted and will
+default to +0+0.
 .TP
-.B \-loop n
+.BI \-loop " n
 Loop the frames n number of times. 0 indicates the frames should loop forever.
+Valid range is 0 to 65535 [Default: 0 (infinite)].
+.TP
+.BI \-bgcolor " A,R,G,B
+Background color of the canvas.
+.br
+where: 'A', 'R', 'G' and 'B' are integers in the range 0 to 255 specifying the
+Alpha, Red, Green and Blue component values respectively
+[Default: 255,255,255,255].
 
 .SS INPUT
 .TP
@@ -98,26 +123,76 @@ Input file in WebP format.
 .TP
 Output file in WebP format.
 
+.SS Note:
+.TP
+The nature of EXIF, XMP and ICC data is not checked and is assumed to be valid.
+
 .SH BUGS
 Please report all bugs to our issue tracker:
 http://code.google.com/p/webp/issues
 .br
 Patches welcome! See this page to get started:
-http://www.webmproject.org/code/contribute/submitting-patches/
+http://www.webmproject.org/code/contribute/submitting\-patches/
 
 .SH EXAMPLES
+.P
+Add ICC profile:
+.br
 webpmux \-set icc image_profile.icc in.webp \-o icc_container.webp
+.P
+Extract ICC profile:
 .br
 webpmux \-get icc icc_container.webp \-o image_profile.icc
+.P
+Strip ICC profile:
+.br
+webpmux \-strip icc icc_container.webp \-o without_icc.webp
+.P
+Add XMP metadata:
 .br
 webpmux \-set xmp image_metadata.xmp in.webp \-o xmp_container.webp
+.P
+Extract XMP metadata:
 .br
 webpmux \-get xmp xmp_container.webp \-o image_metadata.xmp
+.P
+Strip XMP metadata:
+.br
+webpmux \-strip xmp xmp_container.webp \-o without_xmp.webp
+.P
+Add EXIF metadata:
+.br
+webpmux \-set exif image_metadata.exif in.webp \-o exif_container.webp
+.P
+Extract EXIF metadata:
+.br
+webpmux \-get exif exif_container.webp \-o image_metadata.exif
+.P
+Strip EXIF metadata:
+.br
+webpmux \-strip exif exif_container.webp \-o without_exif.webp
+.P
+Create an animated WebP file from 3 (non\-animated) WebP images:
+.br
+webpmux \-frame 1.webp +100 \-frame 2.webp +100+50+50
+.br
+.RS 8
+\-frame 3.webp +100+50+50+1+b \-loop 10 \-bgcolor 255,255,255,255
 .br
-webpmux \-frame anim_1.webp +0+0+0 \-frame anim_2.webp +50+50+0 \-loop 10
 \-o anim_container.webp
+.RE
+.P
+Get the 2nd frame from an animated WebP file:
 .br
 webpmux \-get frame 2 anim_container.webp \-o frame_2.webp
+.P
+Using \-get/\-set/\-strip with input file name starting with '\-':
+.br
+webpmux \-set icc image_profile.icc \-o icc_container.webp \-\- \-\-\-in.webp
+.br
+webpmux \-get icc \-o image_profile.icc \-\- \-\-\-icc_container.webp
+.br
+webpmux \-strip icc \-o without_icc.webp \-\- \-\-\-icc_container.webp
 
 .SH AUTHORS
 \fBwebpmux\fP is written by the WebP team.
@@ -128,8 +203,9 @@ This manual page was written by Vikas Arora <vikaas.arora@gmail.com>,
 for the Debian project (and may be used by others).
 
 .SH SEE ALSO
+.BR cwebp (1),
 .BR dwebp (1),
-.BR cwebp (1).
+.BR gif2webp (1)
 .br
 Please refer to http://developers.google.com/speed/webp/ for additional
 information.

src/.gitignore

@@ -1 +0,0 @@
-/*.pc

src/Makefile.am

@@ -1,16 +1,27 @@
-SUBDIRS = dec enc dsp utils
+# The mux and demux libraries depend on libwebp, thus the '.' to force the
+# build order so it's available to them.
+SUBDIRS = dec enc dsp utils .
 if WANT_MUX
   SUBDIRS += mux
 endif
+if WANT_DEMUX
+  SUBDIRS += demux
+endif
 
-AM_CPPFLAGS = -I$(top_srcdir)/src
 lib_LTLIBRARIES = libwebp.la
 
+if BUILD_LIBWEBPDECODER
+  lib_LTLIBRARIES += libwebpdecoder.la
+endif
+
+common_HEADERS =
+common_HEADERS += webp/decode.h
+common_HEADERS += webp/types.h
+commondir = $(includedir)/webp
+
 libwebp_la_SOURCES =
 libwebpinclude_HEADERS =
-libwebpinclude_HEADERS += webp/decode.h
 libwebpinclude_HEADERS += webp/encode.h
-libwebpinclude_HEADERS += webp/types.h
 noinst_HEADERS =
 noinst_HEADERS += webp/format_constants.h
 
@@ -24,8 +35,20 @@ libwebp_la_LIBADD += utils/libwebputils.la
 # other than the ones listed on the command line, i.e., after linking, it will
 # not have unresolved symbols. Some platforms (Windows among them) require all
 # symbols in shared libraries to be resolved at library creation.
-libwebp_la_LDFLAGS = -no-undefined -version-info 4:1:0
+libwebp_la_LDFLAGS = -no-undefined -version-info 5:2:0
 libwebpincludedir = $(includedir)/webp
-
 pkgconfig_DATA = libwebp.pc
+
+if BUILD_LIBWEBPDECODER
+  libwebpdecoder_la_SOURCES =
+
+  libwebpdecoder_la_LIBADD =
+  libwebpdecoder_la_LIBADD += dec/libwebpdecode.la
+  libwebpdecoder_la_LIBADD += dsp/libwebpdspdecode.la
+  libwebpdecoder_la_LIBADD += utils/libwebputilsdecode.la
+
+  libwebpdecoder_la_LDFLAGS = -no-undefined -version-info 1:2:0
+  pkgconfig_DATA += libwebpdecoder.pc
+endif
+
 ${pkgconfig_DATA}: ${top_builddir}/config.status

src/dec/Makefile.am

@@ -1,14 +1,13 @@
-AM_CPPFLAGS = -I$(top_srcdir)/src
 noinst_LTLIBRARIES = libwebpdecode.la
 
 libwebpdecode_la_SOURCES =
 libwebpdecode_la_SOURCES += alpha.c
+libwebpdecode_la_SOURCES += alphai.h
 libwebpdecode_la_SOURCES += buffer.c
 libwebpdecode_la_SOURCES += decode_vp8.h
 libwebpdecode_la_SOURCES += frame.c
 libwebpdecode_la_SOURCES += idec.c
 libwebpdecode_la_SOURCES += io.c
-libwebpdecode_la_SOURCES += layer.c
 libwebpdecode_la_SOURCES += quant.c
 libwebpdecode_la_SOURCES += tree.c
 libwebpdecode_la_SOURCES += vp8.c

src/dec/alpha.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Alpha-plane decompression.
@@ -10,131 +12,154 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include <stdlib.h>
+#include "./alphai.h"
 #include "./vp8i.h"
 #include "./vp8li.h"
-#include "../utils/filters.h"
-#include "../utils/quant_levels.h"
+#include "../utils/quant_levels_dec.h"
+#include "../utils/utils.h"
 #include "../webp/format_constants.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
+//------------------------------------------------------------------------------
+// ALPHDecoder object.
 
-// TODO(skal): move to dsp/ ?
-static void CopyPlane(const uint8_t* src, int src_stride,
-                      uint8_t* dst, int dst_stride, int width, int height) {
-  while (height-- > 0) {
-    memcpy(dst, src, width);
-    src += src_stride;
-    dst += dst_stride;
+ALPHDecoder* ALPHNew(void) {
+  ALPHDecoder* const dec = (ALPHDecoder*)WebPSafeCalloc(1ULL, sizeof(*dec));
+  return dec;
+}
+
+void ALPHDelete(ALPHDecoder* const dec) {
+  if (dec != NULL) {
+    VP8LDelete(dec->vp8l_dec_);
+    dec->vp8l_dec_ = NULL;
+    WebPSafeFree(dec);
   }
 }
 
 //------------------------------------------------------------------------------
-// Decodes the compressed data 'data' of size 'data_size' into the 'output'.
-// The 'output' buffer should be pre-allocated and must be of the same
-// dimension 'height'x'stride', as that of the image.
-//
-// Returns 1 on successfully decoding the compressed alpha and
-//         0 if either:
-//           error in bit-stream header (invalid compression mode or filter), or
-//           error returned by appropriate compression method.
+// Decoding.
 
-static int DecodeAlpha(const uint8_t* data, size_t data_size,
-                       int width, int height, int stride, uint8_t* output) {
-  uint8_t* decoded_data = NULL;
-  const size_t decoded_size = height * width;
-  uint8_t* unfiltered_data = NULL;
-  WEBP_FILTER_TYPE filter;
-  int pre_processing;
-  int rsrv;
+// Initialize alpha decoding by parsing the alpha header and decoding the image
+// header for alpha data stored using lossless compression.
+// Returns false in case of error in alpha header (data too short, invalid
+// compression method or filter, error in lossless header data etc).
+static int ALPHInit(ALPHDecoder* const dec, const uint8_t* data,
+                    size_t data_size, int width, int height, uint8_t* output) {
   int ok = 0;
-  int method;
+  const uint8_t* const alpha_data = data + ALPHA_HEADER_LEN;
+  const size_t alpha_data_size = data_size - ALPHA_HEADER_LEN;
+  int rsrv;
 
-  assert(width > 0 && height > 0 && stride >= width);
+  assert(width > 0 && height > 0);
   assert(data != NULL && output != NULL);
 
+  dec->width_ = width;
+  dec->height_ = height;
+
   if (data_size <= ALPHA_HEADER_LEN) {
     return 0;
   }
 
-  method = (data[0] >> 0) & 0x03;
-  filter = (data[0] >> 2) & 0x03;
-  pre_processing = (data[0] >> 4) & 0x03;
+  dec->method_ = (data[0] >> 0) & 0x03;
+  dec->filter_ = (data[0] >> 2) & 0x03;
+  dec->pre_processing_ = (data[0] >> 4) & 0x03;
   rsrv = (data[0] >> 6) & 0x03;
-  if (method < ALPHA_NO_COMPRESSION ||
-      method > ALPHA_LOSSLESS_COMPRESSION ||
-      filter >= WEBP_FILTER_LAST ||
-      pre_processing > ALPHA_PREPROCESSED_LEVELS ||
+  if (dec->method_ < ALPHA_NO_COMPRESSION ||
+      dec->method_ > ALPHA_LOSSLESS_COMPRESSION ||
+      dec->filter_ >= WEBP_FILTER_LAST ||
+      dec->pre_processing_ > ALPHA_PREPROCESSED_LEVELS ||
       rsrv != 0) {
     return 0;
   }
 
-  if (method == ALPHA_NO_COMPRESSION) {
-    ok = (data_size >= decoded_size);
-    decoded_data = (uint8_t*)data + ALPHA_HEADER_LEN;
+  if (dec->method_ == ALPHA_NO_COMPRESSION) {
+    const size_t alpha_decoded_size = dec->width_ * dec->height_;
+    ok = (alpha_data_size >= alpha_decoded_size);
   } else {
-    decoded_data = (uint8_t*)malloc(decoded_size);
-    if (decoded_data == NULL) return 0;
-    ok = VP8LDecodeAlphaImageStream(width, height,
-                                    data + ALPHA_HEADER_LEN,
-                                    data_size - ALPHA_HEADER_LEN,
-                                    decoded_data);
-  }
-
-  if (ok) {
-    WebPFilterFunc unfilter_func = WebPUnfilters[filter];
-    if (unfilter_func != NULL) {
-      unfiltered_data = (uint8_t*)malloc(decoded_size);
-      if (unfiltered_data == NULL) {
-        ok = 0;
-        goto Error;
-      }
-      // TODO(vikas): Implement on-the-fly decoding & filter mechanism to decode
-      // and apply filter per image-row.
-      unfilter_func(decoded_data, width, height, 1, width, unfiltered_data);
-      // Construct raw_data (height x stride) from alpha data (height x width).
-      CopyPlane(unfiltered_data, width, output, stride, width, height);
-      free(unfiltered_data);
-    } else {
-      // Construct raw_data (height x stride) from alpha data (height x width).
-      CopyPlane(decoded_data, width, output, stride, width, height);
-    }
-    if (pre_processing == ALPHA_PREPROCESSED_LEVELS) {
-      ok = DequantizeLevels(decoded_data, width, height);
-    }
-  }
-
- Error:
-  if (method != ALPHA_NO_COMPRESSION) {
-    free(decoded_data);
+    assert(dec->method_ == ALPHA_LOSSLESS_COMPRESSION);
+    ok = VP8LDecodeAlphaHeader(dec, alpha_data, alpha_data_size, output);
   }
   return ok;
 }
 
+// Decodes, unfilters and dequantizes *at least* 'num_rows' rows of alpha
+// starting from row number 'row'. It assumes that rows up to (row - 1) have
+// already been decoded.
+// Returns false in case of bitstream error.
+static int ALPHDecode(VP8Decoder* const dec, int row, int num_rows) {
+  ALPHDecoder* const alph_dec = dec->alph_dec_;
+  const int width = alph_dec->width_;
+  const int height = alph_dec->height_;
+  WebPUnfilterFunc unfilter_func = WebPUnfilters[alph_dec->filter_];
+  uint8_t* const output = dec->alpha_plane_;
+  if (alph_dec->method_ == ALPHA_NO_COMPRESSION) {
+    const size_t offset = row * width;
+    const size_t num_pixels = num_rows * width;
+    assert(dec->alpha_data_size_ >= ALPHA_HEADER_LEN + offset + num_pixels);
+    memcpy(dec->alpha_plane_ + offset,
+           dec->alpha_data_ + ALPHA_HEADER_LEN + offset, num_pixels);
+  } else {  // alph_dec->method_ == ALPHA_LOSSLESS_COMPRESSION
+    assert(alph_dec->vp8l_dec_ != NULL);
+    if (!VP8LDecodeAlphaImageStream(alph_dec, row + num_rows)) {
+      return 0;
+    }
+  }
+
+  if (unfilter_func != NULL) {
+    unfilter_func(width, height, width, row, num_rows, output);
+  }
+
+  if (row + num_rows == dec->pic_hdr_.height_) {
+    dec->is_alpha_decoded_ = 1;
+  }
+  return 1;
+}
+
 //------------------------------------------------------------------------------
+// Main entry point.
 
 const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec,
                                       int row, int num_rows) {
-  const int stride = dec->pic_hdr_.width_;
+  const int width = dec->pic_hdr_.width_;
+  const int height = dec->pic_hdr_.height_;
 
-  if (row < 0 || num_rows < 0 || row + num_rows > dec->pic_hdr_.height_) {
+  if (row < 0 || num_rows <= 0 || row + num_rows > height) {
     return NULL;    // sanity check.
   }
 
   if (row == 0) {
-    // Decode everything during the first call.
-    if (!DecodeAlpha(dec->alpha_data_, (size_t)dec->alpha_data_size_,
-                     dec->pic_hdr_.width_, dec->pic_hdr_.height_, stride,
-                     dec->alpha_plane_)) {
-      return NULL;  // Error.
+    // Initialize decoding.
+    assert(dec->alpha_plane_ != NULL);
+    dec->alph_dec_ = ALPHNew();
+    if (dec->alph_dec_ == NULL) return NULL;
+    if (!ALPHInit(dec->alph_dec_, dec->alpha_data_, dec->alpha_data_size_,
+                  width, height, dec->alpha_plane_)) {
+      ALPHDelete(dec->alph_dec_);
+      dec->alph_dec_ = NULL;
+      return NULL;
+    }
+    // if we allowed use of alpha dithering, check whether it's needed at all
+    if (dec->alph_dec_->pre_processing_ != ALPHA_PREPROCESSED_LEVELS) {
+      dec->alpha_dithering_ = 0;  // disable dithering
+    } else {
+      num_rows = height;          // decode everything in one pass
     }
   }
 
-  // Return a pointer to the current decoded row.
-  return dec->alpha_plane_ + row * stride;
-}
+  if (!dec->is_alpha_decoded_) {
+    int ok = 0;
+    assert(dec->alph_dec_ != NULL);
+    ok = ALPHDecode(dec, row, num_rows);
+    if (ok && dec->alpha_dithering_ > 0) {
+      ok = WebPDequantizeLevels(dec->alpha_plane_, width, height,
+                                dec->alpha_dithering_);
+    }
+    if (!ok || dec->is_alpha_decoded_) {
+      ALPHDelete(dec->alph_dec_);
+      dec->alph_dec_ = NULL;
+    }
+    if (!ok) return NULL;  // Error.
+  }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif
+  // Return a pointer to the current decoded row.
+  return dec->alpha_plane_ + row * width;
+}

src/dec/alphai.h

@@ -0,0 +1,55 @@
+// Copyright 2013 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Alpha decoder: internal header.
+//
+// Author: Urvang (urvang@google.com)
+
+#ifndef WEBP_DEC_ALPHAI_H_
+#define WEBP_DEC_ALPHAI_H_
+
+#include "./webpi.h"
+#include "../utils/filters.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct VP8LDecoder;  // Defined in dec/vp8li.h.
+
+typedef struct ALPHDecoder ALPHDecoder;
+struct ALPHDecoder {
+  int width_;
+  int height_;
+  int method_;
+  WEBP_FILTER_TYPE filter_;
+  int pre_processing_;
+  struct VP8LDecoder* vp8l_dec_;
+  VP8Io io_;
+  int use_8b_decode;  // Although alpha channel requires only 1 byte per
+                      // pixel, sometimes VP8LDecoder may need to allocate
+                      // 4 bytes per pixel internally during decode.
+};
+
+//------------------------------------------------------------------------------
+// internal functions. Not public.
+
+// Allocates a new alpha decoder instance.
+ALPHDecoder* ALPHNew(void);
+
+// Clears and deallocates an alpha decoder instance.
+void ALPHDelete(ALPHDecoder* const dec);
+
+//------------------------------------------------------------------------------
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  /* WEBP_DEC_ALPHAI_H_ */

src/dec/buffer.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Everything about WebPDecBuffer
@@ -15,10 +17,6 @@
 #include "./webpi.h"
 #include "../utils/utils.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 //------------------------------------------------------------------------------
 // WebPDecBuffer
 
@@ -44,29 +42,34 @@ static VP8StatusCode CheckDecBuffer(const WebPDecBuffer* const buffer) {
     ok = 0;
   } else if (!WebPIsRGBMode(mode)) {   // YUV checks
     const WebPYUVABuffer* const buf = &buffer->u.YUVA;
-    const uint64_t y_size = (uint64_t)buf->y_stride * height;
-    const uint64_t u_size = (uint64_t)buf->u_stride * ((height + 1) / 2);
-    const uint64_t v_size = (uint64_t)buf->v_stride * ((height + 1) / 2);
-    const uint64_t a_size = (uint64_t)buf->a_stride * height;
+    const int y_stride = abs(buf->y_stride);
+    const int u_stride = abs(buf->u_stride);
+    const int v_stride = abs(buf->v_stride);
+    const int a_stride = abs(buf->a_stride);
+    const uint64_t y_size = (uint64_t)y_stride * height;
+    const uint64_t u_size = (uint64_t)u_stride * ((height + 1) / 2);
+    const uint64_t v_size = (uint64_t)v_stride * ((height + 1) / 2);
+    const uint64_t a_size = (uint64_t)a_stride * height;
     ok &= (y_size <= buf->y_size);
     ok &= (u_size <= buf->u_size);
     ok &= (v_size <= buf->v_size);
-    ok &= (buf->y_stride >= width);
-    ok &= (buf->u_stride >= (width + 1) / 2);
-    ok &= (buf->v_stride >= (width + 1) / 2);
+    ok &= (y_stride >= width);
+    ok &= (u_stride >= (width + 1) / 2);
+    ok &= (v_stride >= (width + 1) / 2);
     ok &= (buf->y != NULL);
     ok &= (buf->u != NULL);
     ok &= (buf->v != NULL);
     if (mode == MODE_YUVA) {
-      ok &= (buf->a_stride >= width);
+      ok &= (a_stride >= width);
       ok &= (a_size <= buf->a_size);
       ok &= (buf->a != NULL);
     }
   } else {    // RGB checks
     const WebPRGBABuffer* const buf = &buffer->u.RGBA;
-    const uint64_t size = (uint64_t)buf->stride * height;
+    const int stride = abs(buf->stride);
+    const uint64_t size = (uint64_t)stride * height;
     ok &= (size <= buf->size);
-    ok &= (buf->stride >= width * kModeBpp[mode]);
+    ok &= (stride >= width * kModeBpp[mode]);
     ok &= (buf->rgba != NULL);
   }
   return ok ? VP8_STATUS_OK : VP8_STATUS_INVALID_PARAM;
@@ -133,9 +136,35 @@ static VP8StatusCode AllocateBuffer(WebPDecBuffer* const buffer) {
   return CheckDecBuffer(buffer);
 }
 
+VP8StatusCode WebPFlipBuffer(WebPDecBuffer* const buffer) {
+  if (buffer == NULL) {
+    return VP8_STATUS_INVALID_PARAM;
+  }
+  if (WebPIsRGBMode(buffer->colorspace)) {
+    WebPRGBABuffer* const buf = &buffer->u.RGBA;
+    buf->rgba += (buffer->height - 1) * buf->stride;
+    buf->stride = -buf->stride;
+  } else {
+    WebPYUVABuffer* const buf = &buffer->u.YUVA;
+    const int H = buffer->height;
+    buf->y += (H - 1) * buf->y_stride;
+    buf->y_stride = -buf->y_stride;
+    buf->u += ((H - 1) >> 1) * buf->u_stride;
+    buf->u_stride = -buf->u_stride;
+    buf->v += ((H - 1) >> 1) * buf->v_stride;
+    buf->v_stride = -buf->v_stride;
+    if (buf->a != NULL) {
+      buf->a += (H - 1) * buf->a_stride;
+      buf->a_stride = -buf->a_stride;
+    }
+  }
+  return VP8_STATUS_OK;
+}
+
 VP8StatusCode WebPAllocateDecBuffer(int w, int h,
                                     const WebPDecoderOptions* const options,
                                     WebPDecBuffer* const out) {
+  VP8StatusCode status;
   if (out == NULL || w <= 0 || h <= 0) {
     return VP8_STATUS_INVALID_PARAM;
   }
@@ -162,8 +191,17 @@ VP8StatusCode WebPAllocateDecBuffer(int w, int h,
   out->width = w;
   out->height = h;
 
-  // Then, allocate buffer for real
-  return AllocateBuffer(out);
+  // Then, allocate buffer for real.
+  status = AllocateBuffer(out);
+  if (status != VP8_STATUS_OK) return status;
+
+#if WEBP_DECODER_ABI_VERSION > 0x0203
+  // Use the stride trick if vertical flip is needed.
+  if (options != NULL && options->flip) {
+    status = WebPFlipBuffer(out);
+  }
+#endif
+  return status;
 }
 
 //------------------------------------------------------------------------------
@@ -180,8 +218,9 @@ int WebPInitDecBufferInternal(WebPDecBuffer* buffer, int version) {
 
 void WebPFreeDecBuffer(WebPDecBuffer* buffer) {
   if (buffer != NULL) {
-    if (!buffer->is_external_memory)
-      free(buffer->private_memory);
+    if (!buffer->is_external_memory) {
+      WebPSafeFree(buffer->private_memory);
+    }
     buffer->private_memory = NULL;
   }
 }
@@ -210,6 +249,3 @@ void WebPGrabDecBuffer(WebPDecBuffer* const src, WebPDecBuffer* const dst) {
 
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/decode_vp8.h

@@ -1,8 +1,10 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 //  Low-level API for VP8 decoder
@@ -14,7 +16,7 @@
 
 #include "../webp/decode.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 extern "C" {
 #endif
 
@@ -130,7 +132,8 @@ static WEBP_INLINE int VP8InitIo(VP8Io* const io) {
   return VP8InitIoInternal(io, WEBP_DECODER_ABI_VERSION);
 }
 
-// Start decoding a new picture. Returns true if ok.
+// Decode the VP8 frame header. Returns true if ok.
+// Note: 'io->data' must be pointing to the start of the VP8 frame header.
 int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io);
 
 // Decode a picture. Will call VP8GetHeaders() if it wasn't done already.
@@ -175,7 +178,7 @@ WEBP_EXTERN(int) VP8LGetInfo(
     const uint8_t* data, size_t data_size,  // data available so far
     int* const width, int* const height, int* const has_alpha);
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 

src/dec/frame.c

@@ -1,8 +1,10 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Frame-reconstruction function. Memory allocation.
@@ -13,12 +15,11 @@
 #include "./vp8i.h"
 #include "../utils/utils.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 #define ALIGN_MASK (32 - 1)
 
+static void ReconstructRow(const VP8Decoder* const dec,
+                           const VP8ThreadContext* ctx);  // TODO(skal): remove
+
 //------------------------------------------------------------------------------
 // Filtering
 
@@ -29,25 +30,18 @@ extern "C" {
 //                 U/V, so it's 8 samples total (because of the 2x upsampling).
 static const uint8_t kFilterExtraRows[3] = { 0, 2, 8 };
 
-static WEBP_INLINE int hev_thresh_from_level(int level, int keyframe) {
-  if (keyframe) {
-    return (level >= 40) ? 2 : (level >= 15) ? 1 : 0;
-  } else {
-    return (level >= 40) ? 3 : (level >= 20) ? 2 : (level >= 15) ? 1 : 0;
-  }
-}
-
 static void DoFilter(const VP8Decoder* const dec, int mb_x, int mb_y) {
   const VP8ThreadContext* const ctx = &dec->thread_ctx_;
+  const int cache_id = ctx->id_;
   const int y_bps = dec->cache_y_stride_;
-  VP8FInfo* const f_info = ctx->f_info_ + mb_x;
-  uint8_t* const y_dst = dec->cache_y_ + ctx->id_ * 16 * y_bps + mb_x * 16;
-  const int level = f_info->f_level_;
+  const VP8FInfo* const f_info = ctx->f_info_ + mb_x;
+  uint8_t* const y_dst = dec->cache_y_ + cache_id * 16 * y_bps + mb_x * 16;
   const int ilevel = f_info->f_ilevel_;
-  const int limit = 2 * level + ilevel;
-  if (level == 0) {
+  const int limit = f_info->f_limit_;
+  if (limit == 0) {
     return;
   }
+  assert(limit >= 3);
   if (dec->filter_type_ == 1) {   // simple
     if (mb_x > 0) {
       VP8SimpleHFilter16(y_dst, y_bps, limit + 4);
@@ -63,10 +57,9 @@ static void DoFilter(const VP8Decoder* const dec, int mb_x, int mb_y) {
     }
   } else {    // complex
     const int uv_bps = dec->cache_uv_stride_;
-    uint8_t* const u_dst = dec->cache_u_ + ctx->id_ * 8 * uv_bps + mb_x * 8;
-    uint8_t* const v_dst = dec->cache_v_ + ctx->id_ * 8 * uv_bps + mb_x * 8;
-    const int hev_thresh =
-        hev_thresh_from_level(level, dec->frm_hdr_.key_frame_);
+    uint8_t* const u_dst = dec->cache_u_ + cache_id * 8 * uv_bps + mb_x * 8;
+    uint8_t* const v_dst = dec->cache_v_ + cache_id * 8 * uv_bps + mb_x * 8;
+    const int hev_thresh = f_info->hev_thresh_;
     if (mb_x > 0) {
       VP8HFilter16(y_dst, y_bps, limit + 4, ilevel, hev_thresh);
       VP8HFilter8(u_dst, v_dst, uv_bps, limit + 4, ilevel, hev_thresh);
@@ -97,53 +90,141 @@ static void FilterRow(const VP8Decoder* const dec) {
 }
 
 //------------------------------------------------------------------------------
+// Precompute the filtering strength for each segment and each i4x4/i16x16 mode.
 
-void VP8StoreBlock(VP8Decoder* const dec) {
+static void PrecomputeFilterStrengths(VP8Decoder* const dec) {
   if (dec->filter_type_ > 0) {
-    VP8FInfo* const info = dec->f_info_ + dec->mb_x_;
-    const int skip = dec->mb_info_[dec->mb_x_].skip_;
-    int level = dec->filter_levels_[dec->segment_];
-    if (dec->filter_hdr_.use_lf_delta_) {
-      // TODO(skal): only CURRENT is handled for now.
-      level += dec->filter_hdr_.ref_lf_delta_[0];
-      if (dec->is_i4x4_) {
-        level += dec->filter_hdr_.mode_lf_delta_[0];
-      }
-    }
-    level = (level < 0) ? 0 : (level > 63) ? 63 : level;
-    info->f_level_ = level;
-
-    if (dec->filter_hdr_.sharpness_ > 0) {
-      if (dec->filter_hdr_.sharpness_ > 4) {
-        level >>= 2;
+    int s;
+    const VP8FilterHeader* const hdr = &dec->filter_hdr_;
+    for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
+      int i4x4;
+      // First, compute the initial level
+      int base_level;
+      if (dec->segment_hdr_.use_segment_) {
+        base_level = dec->segment_hdr_.filter_strength_[s];
+        if (!dec->segment_hdr_.absolute_delta_) {
+          base_level += hdr->level_;
+        }
       } else {
-        level >>= 1;
+        base_level = hdr->level_;
       }
-      if (level > 9 - dec->filter_hdr_.sharpness_) {
-        level = 9 - dec->filter_hdr_.sharpness_;
+      for (i4x4 = 0; i4x4 <= 1; ++i4x4) {
+        VP8FInfo* const info = &dec->fstrengths_[s][i4x4];
+        int level = base_level;
+        if (hdr->use_lf_delta_) {
+          // TODO(skal): only CURRENT is handled for now.
+          level += hdr->ref_lf_delta_[0];
+          if (i4x4) {
+            level += hdr->mode_lf_delta_[0];
+          }
+        }
+        level = (level < 0) ? 0 : (level > 63) ? 63 : level;
+        if (level > 0) {
+          int ilevel = level;
+          if (hdr->sharpness_ > 0) {
+            if (hdr->sharpness_ > 4) {
+              ilevel >>= 2;
+            } else {
+              ilevel >>= 1;
+            }
+            if (ilevel > 9 - hdr->sharpness_) {
+              ilevel = 9 - hdr->sharpness_;
+            }
+          }
+          if (ilevel < 1) ilevel = 1;
+          info->f_ilevel_ = ilevel;
+          info->f_limit_ = 2 * level + ilevel;
+          info->hev_thresh_ = (level >= 40) ? 2 : (level >= 15) ? 1 : 0;
+        } else {
+          info->f_limit_ = 0;  // no filtering
+        }
+        info->f_inner_ = i4x4;
       }
     }
-
-    info->f_ilevel_ = (level < 1) ? 1 : level;
-    info->f_inner_ = (!skip || dec->is_i4x4_);
   }
-  {
-    // Transfer samples to row cache
-    int y;
-    const int y_offset = dec->cache_id_ * 16 * dec->cache_y_stride_;
-    const int uv_offset = dec->cache_id_ * 8 * dec->cache_uv_stride_;
-    uint8_t* const ydst = dec->cache_y_ + dec->mb_x_ * 16 + y_offset;
-    uint8_t* const udst = dec->cache_u_ + dec->mb_x_ * 8 + uv_offset;
-    uint8_t* const vdst = dec->cache_v_ + dec->mb_x_ * 8 + uv_offset;
-    for (y = 0; y < 16; ++y) {
-      memcpy(ydst + y * dec->cache_y_stride_,
-             dec->yuv_b_ + Y_OFF + y * BPS, 16);
+}
+
+//------------------------------------------------------------------------------
+// Dithering
+
+#define DITHER_AMP_TAB_SIZE 12
+static const int kQuantToDitherAmp[DITHER_AMP_TAB_SIZE] = {
+  // roughly, it's dqm->uv_mat_[1]
+  8, 7, 6, 4, 4, 2, 2, 2, 1, 1, 1, 1
+};
+
+void VP8InitDithering(const WebPDecoderOptions* const options,
+                      VP8Decoder* const dec) {
+  assert(dec != NULL);
+  if (options != NULL) {
+    const int d = options->dithering_strength;
+    const int max_amp = (1 << VP8_RANDOM_DITHER_FIX) - 1;
+    const int f = (d < 0) ? 0 : (d > 100) ? max_amp : (d * max_amp / 100);
+    if (f > 0) {
+      int s;
+      int all_amp = 0;
+      for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
+        VP8QuantMatrix* const dqm = &dec->dqm_[s];
+        if (dqm->uv_quant_ < DITHER_AMP_TAB_SIZE) {
+          // TODO(skal): should we specially dither more for uv_quant_ < 0?
+          const int idx = (dqm->uv_quant_ < 0) ? 0 : dqm->uv_quant_;
+          dqm->dither_ = (f * kQuantToDitherAmp[idx]) >> 3;
+        }
+        all_amp |= dqm->dither_;
+      }
+      if (all_amp != 0) {
+        VP8InitRandom(&dec->dithering_rg_, 1.0f);
+        dec->dither_ = 1;
+      }
     }
-    for (y = 0; y < 8; ++y) {
-      memcpy(udst + y * dec->cache_uv_stride_,
-           dec->yuv_b_ + U_OFF + y * BPS, 8);
-      memcpy(vdst + y * dec->cache_uv_stride_,
-           dec->yuv_b_ + V_OFF + y * BPS, 8);
+#if WEBP_DECODER_ABI_VERSION > 0x0204
+    // potentially allow alpha dithering
+    dec->alpha_dithering_ = options->alpha_dithering_strength;
+    if (dec->alpha_dithering_ > 100) {
+      dec->alpha_dithering_ = 100;
+    } else if (dec->alpha_dithering_ < 0) {
+      dec->alpha_dithering_ = 0;
+    }
+#endif
+  }
+}
+
+// minimal amp that will provide a non-zero dithering effect
+#define MIN_DITHER_AMP 4
+#define DITHER_DESCALE 4
+#define DITHER_DESCALE_ROUNDER (1 << (DITHER_DESCALE - 1))
+#define DITHER_AMP_BITS 8
+#define DITHER_AMP_CENTER (1 << DITHER_AMP_BITS)
+
+static void Dither8x8(VP8Random* const rg, uint8_t* dst, int bps, int amp) {
+  int i, j;
+  for (j = 0; j < 8; ++j) {
+    for (i = 0; i < 8; ++i) {
+      // TODO: could be made faster with SSE2
+      const int bits =
+          VP8RandomBits2(rg, DITHER_AMP_BITS + 1, amp) - DITHER_AMP_CENTER;
+      // Convert to range: [-2,2] for dither=50, [-4,4] for dither=100
+      const int delta = (bits + DITHER_DESCALE_ROUNDER) >> DITHER_DESCALE;
+      const int v = (int)dst[i] + delta;
+      dst[i] = (v < 0) ? 0 : (v > 255) ? 255u : (uint8_t)v;
+    }
+    dst += bps;
+  }
+}
+
+static void DitherRow(VP8Decoder* const dec) {
+  int mb_x;
+  assert(dec->dither_);
+  for (mb_x = dec->tl_mb_x_; mb_x < dec->br_mb_x_; ++mb_x) {
+    const VP8ThreadContext* const ctx = &dec->thread_ctx_;
+    const VP8MBData* const data = ctx->mb_data_ + mb_x;
+    const int cache_id = ctx->id_;
+    const int uv_bps = dec->cache_uv_stride_;
+    if (data->dither_ >= MIN_DITHER_AMP) {
+      uint8_t* const u_dst = dec->cache_u_ + cache_id * 8 * uv_bps + mb_x * 8;
+      uint8_t* const v_dst = dec->cache_v_ + cache_id * 8 * uv_bps + mb_x * 8;
+      Dither8x8(&dec->dithering_rg_, u_dst, uv_bps, data->dither_);
+      Dither8x8(&dec->dithering_rg_, v_dst, uv_bps, data->dither_);
     }
   }
 }
@@ -165,25 +246,35 @@ void VP8StoreBlock(VP8Decoder* const dec) {
 static int FinishRow(VP8Decoder* const dec, VP8Io* const io) {
   int ok = 1;
   const VP8ThreadContext* const ctx = &dec->thread_ctx_;
+  const int cache_id = ctx->id_;
   const int extra_y_rows = kFilterExtraRows[dec->filter_type_];
   const int ysize = extra_y_rows * dec->cache_y_stride_;
   const int uvsize = (extra_y_rows / 2) * dec->cache_uv_stride_;
-  const int y_offset = ctx->id_ * 16 * dec->cache_y_stride_;
-  const int uv_offset = ctx->id_ * 8 * dec->cache_uv_stride_;
+  const int y_offset = cache_id * 16 * dec->cache_y_stride_;
+  const int uv_offset = cache_id * 8 * dec->cache_uv_stride_;
   uint8_t* const ydst = dec->cache_y_ - ysize + y_offset;
   uint8_t* const udst = dec->cache_u_ - uvsize + uv_offset;
   uint8_t* const vdst = dec->cache_v_ - uvsize + uv_offset;
-  const int first_row = (ctx->mb_y_ == 0);
-  const int last_row = (ctx->mb_y_ >= dec->br_mb_y_ - 1);
-  int y_start = MACROBLOCK_VPOS(ctx->mb_y_);
-  int y_end = MACROBLOCK_VPOS(ctx->mb_y_ + 1);
+  const int mb_y = ctx->mb_y_;
+  const int is_first_row = (mb_y == 0);
+  const int is_last_row = (mb_y >= dec->br_mb_y_ - 1);
+
+  if (dec->mt_method_ == 2) {
+    ReconstructRow(dec, ctx);
+  }
 
   if (ctx->filter_row_) {
     FilterRow(dec);
   }
 
-  if (io->put) {
-    if (!first_row) {
+  if (dec->dither_) {
+    DitherRow(dec);
+  }
+
+  if (io->put != NULL) {
+    int y_start = MACROBLOCK_VPOS(mb_y);
+    int y_end = MACROBLOCK_VPOS(mb_y + 1);
+    if (!is_first_row) {
       y_start -= extra_y_rows;
       io->y = ydst;
       io->u = udst;
@@ -194,7 +285,7 @@ static int FinishRow(VP8Decoder* const dec, VP8Io* const io) {
       io->v = dec->cache_v_ + uv_offset;
     }
 
-    if (!last_row) {
+    if (!is_last_row) {
       y_end -= extra_y_rows;
     }
     if (y_end > io->crop_bottom) {
@@ -202,11 +293,8 @@ static int FinishRow(VP8Decoder* const dec, VP8Io* const io) {
     }
     io->a = NULL;
     if (dec->alpha_data_ != NULL && y_start < y_end) {
-      // TODO(skal): several things to correct here:
-      // * testing presence of alpha with dec->alpha_data_ is not a good idea
-      // * we're actually decompressing the full plane only once. It should be
-      //   more obvious from signature.
-      // * we could free alpha_data_ right after this call, but we don't own.
+      // TODO(skal): testing presence of alpha with dec->alpha_data_ is not a
+      // good idea.
       io->a = VP8DecompressAlphaRows(dec, y_start, y_end - y_start);
       if (io->a == NULL) {
         return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
@@ -238,8 +326,8 @@ static int FinishRow(VP8Decoder* const dec, VP8Io* const io) {
     }
   }
   // rotate top samples if needed
-  if (ctx->id_ + 1 == dec->num_caches_) {
-    if (!last_row) {
+  if (cache_id + 1 == dec->num_caches_) {
+    if (!is_last_row) {
       memcpy(dec->cache_y_ - ysize, ydst + 16 * dec->cache_y_stride_, ysize);
       memcpy(dec->cache_u_ - uvsize, udst + 8 * dec->cache_uv_stride_, uvsize);
       memcpy(dec->cache_v_ - uvsize, vdst + 8 * dec->cache_uv_stride_, uvsize);
@@ -256,27 +344,40 @@ static int FinishRow(VP8Decoder* const dec, VP8Io* const io) {
 int VP8ProcessRow(VP8Decoder* const dec, VP8Io* const io) {
   int ok = 1;
   VP8ThreadContext* const ctx = &dec->thread_ctx_;
-  if (!dec->use_threads_) {
+  const int filter_row =
+      (dec->filter_type_ > 0) &&
+      (dec->mb_y_ >= dec->tl_mb_y_) && (dec->mb_y_ <= dec->br_mb_y_);
+  if (dec->mt_method_ == 0) {
     // ctx->id_ and ctx->f_info_ are already set
     ctx->mb_y_ = dec->mb_y_;
-    ctx->filter_row_ = dec->filter_row_;
+    ctx->filter_row_ = filter_row;
+    ReconstructRow(dec, ctx);
     ok = FinishRow(dec, io);
   } else {
     WebPWorker* const worker = &dec->worker_;
     // Finish previous job *before* updating context
-    ok &= WebPWorkerSync(worker);
+    ok &= WebPGetWorkerInterface()->Sync(worker);
     assert(worker->status_ == OK);
     if (ok) {   // spawn a new deblocking/output job
       ctx->io_ = *io;
       ctx->id_ = dec->cache_id_;
       ctx->mb_y_ = dec->mb_y_;
-      ctx->filter_row_ = dec->filter_row_;
-      if (ctx->filter_row_) {    // just swap filter info
+      ctx->filter_row_ = filter_row;
+      if (dec->mt_method_ == 2) {  // swap macroblock data
+        VP8MBData* const tmp = ctx->mb_data_;
+        ctx->mb_data_ = dec->mb_data_;
+        dec->mb_data_ = tmp;
+      } else {
+        // perform reconstruction directly in main thread
+        ReconstructRow(dec, ctx);
+      }
+      if (filter_row) {            // swap filter info
         VP8FInfo* const tmp = ctx->f_info_;
         ctx->f_info_ = dec->f_info_;
         dec->f_info_ = tmp;
       }
-      WebPWorkerLaunch(worker);
+      // (reconstruct)+filter in parallel
+      WebPGetWorkerInterface()->Launch(worker);
       if (++dec->cache_id_ == dec->num_caches_) {
         dec->cache_id_ = 0;
       }
@@ -290,8 +391,8 @@ int VP8ProcessRow(VP8Decoder* const dec, VP8Io* const io) {
 
 VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io) {
   // Call setup() first. This may trigger additional decoding features on 'io'.
-  // Note: Afterward, we must call teardown() not matter what.
-  if (io->setup && !io->setup(io)) {
+  // Note: Afterward, we must call teardown() no matter what.
+  if (io->setup != NULL && !io->setup(io)) {
     VP8SetError(dec, VP8_STATUS_USER_ABORT, "Frame setup failed");
     return dec->status_;
   }
@@ -304,7 +405,7 @@ VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io) {
 
   // Define the area where we can skip in-loop filtering, in case of cropping.
   //
-  // 'Simple' filter reads two luma samples outside of the macroblock and
+  // 'Simple' filter reads two luma samples outside of the macroblock
   // and filters one. It doesn't filter the chroma samples. Hence, we can
   // avoid doing the in-loop filtering before crop_top/crop_left position.
   // For the 'Complex' filter, 3 samples are read and up to 3 are filtered.
@@ -339,16 +440,17 @@ VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io) {
       dec->br_mb_y_ = dec->mb_h_;
     }
   }
+  PrecomputeFilterStrengths(dec);
   return VP8_STATUS_OK;
 }
 
 int VP8ExitCritical(VP8Decoder* const dec, VP8Io* const io) {
   int ok = 1;
-  if (dec->use_threads_) {
-    ok = WebPWorkerSync(&dec->worker_);
+  if (dec->mt_method_ > 0) {
+    ok = WebPGetWorkerInterface()->Sync(&dec->worker_);
   }
 
-  if (io->teardown) {
+  if (io->teardown != NULL) {
     io->teardown(io);
   }
   return ok;
@@ -384,9 +486,9 @@ int VP8ExitCritical(VP8Decoder* const dec, VP8Io* const io) {
 // Initialize multi/single-thread worker
 static int InitThreadContext(VP8Decoder* const dec) {
   dec->cache_id_ = 0;
-  if (dec->use_threads_) {
+  if (dec->mt_method_ > 0) {
     WebPWorker* const worker = &dec->worker_;
-    if (!WebPWorkerReset(worker)) {
+    if (!WebPGetWorkerInterface()->Reset(worker)) {
       return VP8SetError(dec, VP8_STATUS_OUT_OF_MEMORY,
                          "thread initialization failed.");
     }
@@ -401,6 +503,28 @@ static int InitThreadContext(VP8Decoder* const dec) {
   return 1;
 }
 
+int VP8GetThreadMethod(const WebPDecoderOptions* const options,
+                       const WebPHeaderStructure* const headers,
+                       int width, int height) {
+  if (options == NULL || options->use_threads == 0) {
+    return 0;
+  }
+  (void)headers;
+  (void)width;
+  (void)height;
+  assert(headers == NULL || !headers->is_lossless);
+#if defined(WEBP_USE_THREAD)
+  if (width < MIN_WIDTH_FOR_THREADS) return 0;
+  // TODO(skal): tune the heuristic further
+#if 0
+  if (height < 2 * width) return 2;
+#endif
+  return 2;
+#else   // !WEBP_USE_THREAD
+  return 0;
+#endif
+}
+
 #undef MT_CACHE_LINES
 #undef ST_CACHE_LINES
 
@@ -412,14 +536,15 @@ static int AllocateMemory(VP8Decoder* const dec) {
   const int mb_w = dec->mb_w_;
   // Note: we use 'size_t' when there's no overflow risk, uint64_t otherwise.
   const size_t intra_pred_mode_size = 4 * mb_w * sizeof(uint8_t);
-  const size_t top_size = (16 + 8 + 8) * mb_w;
+  const size_t top_size = sizeof(VP8TopSamples) * mb_w;
   const size_t mb_info_size = (mb_w + 1) * sizeof(VP8MB);
   const size_t f_info_size =
       (dec->filter_type_ > 0) ?
-          mb_w * (dec->use_threads_ ? 2 : 1) * sizeof(VP8FInfo)
+          mb_w * (dec->mt_method_ > 0 ? 2 : 1) * sizeof(VP8FInfo)
         : 0;
   const size_t yuv_size = YUV_SIZE * sizeof(*dec->yuv_b_);
-  const size_t coeffs_size = 384 * sizeof(*dec->coeffs_);
+  const size_t mb_data_size =
+      (dec->mt_method_ == 2 ? 2 : 1) * mb_w * sizeof(*dec->mb_data_);
   const size_t cache_height = (16 * num_caches
                             + kFilterExtraRows[dec->filter_type_]) * 3 / 2;
   const size_t cache_size = top_size * cache_height;
@@ -428,13 +553,13 @@ static int AllocateMemory(VP8Decoder* const dec) {
       (uint64_t)dec->pic_hdr_.width_ * dec->pic_hdr_.height_ : 0ULL;
   const uint64_t needed = (uint64_t)intra_pred_mode_size
                         + top_size + mb_info_size + f_info_size
-                        + yuv_size + coeffs_size
+                        + yuv_size + mb_data_size
                         + cache_size + alpha_size + ALIGN_MASK;
   uint8_t* mem;
 
   if (needed != (size_t)needed) return 0;  // check for overflow
   if (needed > dec->mem_size_) {
-    free(dec->mem_);
+    WebPSafeFree(dec->mem_);
     dec->mem_size_ = 0;
     dec->mem_ = WebPSafeMalloc(needed, sizeof(uint8_t));
     if (dec->mem_ == NULL) {
@@ -449,12 +574,8 @@ static int AllocateMemory(VP8Decoder* const dec) {
   dec->intra_t_ = (uint8_t*)mem;
   mem += intra_pred_mode_size;
 
-  dec->y_t_ = (uint8_t*)mem;
-  mem += 16 * mb_w;
-  dec->u_t_ = (uint8_t*)mem;
-  mem += 8 * mb_w;
-  dec->v_t_ = (uint8_t*)mem;
-  mem += 8 * mb_w;
+  dec->yuv_t_ = (VP8TopSamples*)mem;
+  mem += top_size;
 
   dec->mb_info_ = ((VP8MB*)mem) + 1;
   mem += mb_info_size;
@@ -463,7 +584,7 @@ static int AllocateMemory(VP8Decoder* const dec) {
   mem += f_info_size;
   dec->thread_ctx_.id_ = 0;
   dec->thread_ctx_.f_info_ = dec->f_info_;
-  if (dec->use_threads_) {
+  if (dec->mt_method_ > 0) {
     // secondary cache line. The deblocking process need to make use of the
     // filtering strength from previous macroblock row, while the new ones
     // are being decoded in parallel. We'll just swap the pointers.
@@ -475,8 +596,12 @@ static int AllocateMemory(VP8Decoder* const dec) {
   dec->yuv_b_ = (uint8_t*)mem;
   mem += yuv_size;
 
-  dec->coeffs_ = (int16_t*)mem;
-  mem += coeffs_size;
+  dec->mb_data_ = (VP8MBData*)mem;
+  dec->thread_ctx_.mb_data_ = (VP8MBData*)mem;
+  if (dec->mt_method_ == 2) {
+    dec->thread_ctx_.mb_data_ += mb_w;
+  }
+  mem += mb_data_size;
 
   dec->cache_y_stride_ = 16 * mb_w;
   dec->cache_uv_stride_ = 8 * mb_w;
@@ -496,9 +621,11 @@ static int AllocateMemory(VP8Decoder* const dec) {
   // alpha plane
   dec->alpha_plane_ = alpha_size ? (uint8_t*)mem : NULL;
   mem += alpha_size;
+  assert(mem <= (uint8_t*)dec->mem_ + dec->mem_size_);
 
-  // note: left-info is initialized once for all.
+  // note: left/top-info is initialized once for all.
   memset(dec->mb_info_ - 1, 0, mb_info_size);
+  VP8InitScanline(dec);   // initialize left too.
 
   // initialize top
   memset(dec->intra_t_, B_DC_PRED, intra_pred_mode_size);
@@ -535,138 +662,163 @@ static const int kScan[16] = {
   0 + 12 * BPS,  4 + 12 * BPS, 8 + 12 * BPS, 12 + 12 * BPS
 };
 
-static WEBP_INLINE int CheckMode(VP8Decoder* const dec, int mode) {
+static int CheckMode(int mb_x, int mb_y, int mode) {
   if (mode == B_DC_PRED) {
-    if (dec->mb_x_ == 0) {
-      return (dec->mb_y_ == 0) ? B_DC_PRED_NOTOPLEFT : B_DC_PRED_NOLEFT;
+    if (mb_x == 0) {
+      return (mb_y == 0) ? B_DC_PRED_NOTOPLEFT : B_DC_PRED_NOLEFT;
     } else {
-      return (dec->mb_y_ == 0) ? B_DC_PRED_NOTOP : B_DC_PRED;
+      return (mb_y == 0) ? B_DC_PRED_NOTOP : B_DC_PRED;
     }
   }
   return mode;
 }
 
-static WEBP_INLINE void Copy32b(uint8_t* dst, uint8_t* src) {
-  *(uint32_t*)dst = *(uint32_t*)src;
+static void Copy32b(uint8_t* dst, uint8_t* src) {
+  memcpy(dst, src, 4);
 }
 
-void VP8ReconstructBlock(VP8Decoder* const dec) {
+static WEBP_INLINE void DoTransform(uint32_t bits, const int16_t* const src,
+                                    uint8_t* const dst) {
+  switch (bits >> 30) {
+    case 3:
+      VP8Transform(src, dst, 0);
+      break;
+    case 2:
+      VP8TransformAC3(src, dst);
+      break;
+    case 1:
+      VP8TransformDC(src, dst);
+      break;
+    default:
+      break;
+  }
+}
+
+static void DoUVTransform(uint32_t bits, const int16_t* const src,
+                          uint8_t* const dst) {
+  if (bits & 0xff) {    // any non-zero coeff at all?
+    if (bits & 0xaa) {  // any non-zero AC coefficient?
+      VP8TransformUV(src, dst);   // note we don't use the AC3 variant for U/V
+    } else {
+      VP8TransformDCUV(src, dst);
+    }
+  }
+}
+
+static void ReconstructRow(const VP8Decoder* const dec,
+                           const VP8ThreadContext* ctx) {
+  int j;
+  int mb_x;
+  const int mb_y = ctx->mb_y_;
+  const int cache_id = ctx->id_;
   uint8_t* const y_dst = dec->yuv_b_ + Y_OFF;
   uint8_t* const u_dst = dec->yuv_b_ + U_OFF;
   uint8_t* const v_dst = dec->yuv_b_ + V_OFF;
+  for (mb_x = 0; mb_x < dec->mb_w_; ++mb_x) {
+    const VP8MBData* const block = ctx->mb_data_ + mb_x;
 
-  // Rotate in the left samples from previously decoded block. We move four
-  // pixels at a time for alignment reason, and because of in-loop filter.
-  if (dec->mb_x_ > 0) {
-    int j;
-    for (j = -1; j < 16; ++j) {
-      Copy32b(&y_dst[j * BPS - 4], &y_dst[j * BPS + 12]);
-    }
-    for (j = -1; j < 8; ++j) {
-      Copy32b(&u_dst[j * BPS - 4], &u_dst[j * BPS + 4]);
-      Copy32b(&v_dst[j * BPS - 4], &v_dst[j * BPS + 4]);
-    }
-  } else {
-    int j;
-    for (j = 0; j < 16; ++j) {
-      y_dst[j * BPS - 1] = 129;
-    }
-    for (j = 0; j < 8; ++j) {
-      u_dst[j * BPS - 1] = 129;
-      v_dst[j * BPS - 1] = 129;
-    }
-    // Init top-left sample on left column too
-    if (dec->mb_y_ > 0) {
-      y_dst[-1 - BPS] = u_dst[-1 - BPS] = v_dst[-1 - BPS] = 129;
-    }
-  }
-  {
-    // bring top samples into the cache
-    uint8_t* const top_y = dec->y_t_ + dec->mb_x_ * 16;
-    uint8_t* const top_u = dec->u_t_ + dec->mb_x_ * 8;
-    uint8_t* const top_v = dec->v_t_ + dec->mb_x_ * 8;
-    const int16_t* coeffs = dec->coeffs_;
-    int n;
-
-    if (dec->mb_y_ > 0) {
-      memcpy(y_dst - BPS, top_y, 16);
-      memcpy(u_dst - BPS, top_u, 8);
-      memcpy(v_dst - BPS, top_v, 8);
-    } else if (dec->mb_x_ == 0) {
-      // we only need to do this init once at block (0,0).
-      // Afterward, it remains valid for the whole topmost row.
-      memset(y_dst - BPS - 1, 127, 16 + 4 + 1);
-      memset(u_dst - BPS - 1, 127, 8 + 1);
-      memset(v_dst - BPS - 1, 127, 8 + 1);
-    }
-
-    // predict and add residuals
-
-    if (dec->is_i4x4_) {   // 4x4
-      uint32_t* const top_right = (uint32_t*)(y_dst - BPS + 16);
-
-      if (dec->mb_y_ > 0) {
-        if (dec->mb_x_ >= dec->mb_w_ - 1) {    // on rightmost border
-          top_right[0] = top_y[15] * 0x01010101u;
-        } else {
-          memcpy(top_right, top_y + 16, sizeof(*top_right));
-        }
+    // Rotate in the left samples from previously decoded block. We move four
+    // pixels at a time for alignment reason, and because of in-loop filter.
+    if (mb_x > 0) {
+      for (j = -1; j < 16; ++j) {
+        Copy32b(&y_dst[j * BPS - 4], &y_dst[j * BPS + 12]);
       }
-      // replicate the top-right pixels below
-      top_right[BPS] = top_right[2 * BPS] = top_right[3 * BPS] = top_right[0];
-
-      // predict and add residues for all 4x4 blocks in turn.
-      for (n = 0; n < 16; n++) {
-        uint8_t* const dst = y_dst + kScan[n];
-        VP8PredLuma4[dec->imodes_[n]](dst);
-        if (dec->non_zero_ac_ & (1 << n)) {
-          VP8Transform(coeffs + n * 16, dst, 0);
-        } else if (dec->non_zero_ & (1 << n)) {  // only DC is present
-          VP8TransformDC(coeffs + n * 16, dst);
-        }
+      for (j = -1; j < 8; ++j) {
+        Copy32b(&u_dst[j * BPS - 4], &u_dst[j * BPS + 4]);
+        Copy32b(&v_dst[j * BPS - 4], &v_dst[j * BPS + 4]);
       }
-    } else {    // 16x16
-      const int pred_func = CheckMode(dec, dec->imodes_[0]);
-      VP8PredLuma16[pred_func](y_dst);
-      if (dec->non_zero_) {
-        for (n = 0; n < 16; n++) {
-          uint8_t* const dst = y_dst + kScan[n];
-          if (dec->non_zero_ac_ & (1 << n)) {
-            VP8Transform(coeffs + n * 16, dst, 0);
-          } else if (dec->non_zero_ & (1 << n)) {  // only DC is present
-            VP8TransformDC(coeffs + n * 16, dst);
-          }
-        }
+    } else {
+      for (j = 0; j < 16; ++j) {
+        y_dst[j * BPS - 1] = 129;
+      }
+      for (j = 0; j < 8; ++j) {
+        u_dst[j * BPS - 1] = 129;
+        v_dst[j * BPS - 1] = 129;
+      }
+      // Init top-left sample on left column too
+      if (mb_y > 0) {
+        y_dst[-1 - BPS] = u_dst[-1 - BPS] = v_dst[-1 - BPS] = 129;
       }
     }
     {
-      // Chroma
-      const int pred_func = CheckMode(dec, dec->uvmode_);
-      VP8PredChroma8[pred_func](u_dst);
-      VP8PredChroma8[pred_func](v_dst);
+      // bring top samples into the cache
+      VP8TopSamples* const top_yuv = dec->yuv_t_ + mb_x;
+      const int16_t* const coeffs = block->coeffs_;
+      uint32_t bits = block->non_zero_y_;
+      int n;
 
-      if (dec->non_zero_ & 0x0f0000) {   // chroma-U
-        const int16_t* const u_coeffs = dec->coeffs_ + 16 * 16;
-        if (dec->non_zero_ac_ & 0x0f0000) {
-          VP8TransformUV(u_coeffs, u_dst);
-        } else {
-          VP8TransformDCUV(u_coeffs, u_dst);
+      if (mb_y > 0) {
+        memcpy(y_dst - BPS, top_yuv[0].y, 16);
+        memcpy(u_dst - BPS, top_yuv[0].u, 8);
+        memcpy(v_dst - BPS, top_yuv[0].v, 8);
+      } else if (mb_x == 0) {
+        // we only need to do this init once at block (0,0).
+        // Afterward, it remains valid for the whole topmost row.
+        memset(y_dst - BPS - 1, 127, 16 + 4 + 1);
+        memset(u_dst - BPS - 1, 127, 8 + 1);
+        memset(v_dst - BPS - 1, 127, 8 + 1);
+      }
+
+      // predict and add residuals
+      if (block->is_i4x4_) {   // 4x4
+        uint32_t* const top_right = (uint32_t*)(y_dst - BPS + 16);
+
+        if (mb_y > 0) {
+          if (mb_x >= dec->mb_w_ - 1) {    // on rightmost border
+            memset(top_right, top_yuv[0].y[15], sizeof(*top_right));
+          } else {
+            memcpy(top_right, top_yuv[1].y, sizeof(*top_right));
+          }
+        }
+        // replicate the top-right pixels below
+        top_right[BPS] = top_right[2 * BPS] = top_right[3 * BPS] = top_right[0];
+
+        // predict and add residuals for all 4x4 blocks in turn.
+        for (n = 0; n < 16; ++n, bits <<= 2) {
+          uint8_t* const dst = y_dst + kScan[n];
+          VP8PredLuma4[block->imodes_[n]](dst);
+          DoTransform(bits, coeffs + n * 16, dst);
+        }
+      } else {    // 16x16
+        const int pred_func = CheckMode(mb_x, mb_y,
+                                        block->imodes_[0]);
+        VP8PredLuma16[pred_func](y_dst);
+        if (bits != 0) {
+          for (n = 0; n < 16; ++n, bits <<= 2) {
+            DoTransform(bits, coeffs + n * 16, y_dst + kScan[n]);
+          }
         }
       }
-      if (dec->non_zero_ & 0xf00000) {   // chroma-V
-        const int16_t* const v_coeffs = dec->coeffs_ + 20 * 16;
-        if (dec->non_zero_ac_ & 0xf00000) {
-          VP8TransformUV(v_coeffs, v_dst);
-        } else {
-          VP8TransformDCUV(v_coeffs, v_dst);
-        }
+      {
+        // Chroma
+        const uint32_t bits_uv = block->non_zero_uv_;
+        const int pred_func = CheckMode(mb_x, mb_y, block->uvmode_);
+        VP8PredChroma8[pred_func](u_dst);
+        VP8PredChroma8[pred_func](v_dst);
+        DoUVTransform(bits_uv >> 0, coeffs + 16 * 16, u_dst);
+        DoUVTransform(bits_uv >> 8, coeffs + 20 * 16, v_dst);
       }
 
       // stash away top samples for next block
-      if (dec->mb_y_ < dec->mb_h_ - 1) {
-        memcpy(top_y, y_dst + 15 * BPS, 16);
-        memcpy(top_u, u_dst +  7 * BPS,  8);
-        memcpy(top_v, v_dst +  7 * BPS,  8);
+      if (mb_y < dec->mb_h_ - 1) {
+        memcpy(top_yuv[0].y, y_dst + 15 * BPS, 16);
+        memcpy(top_yuv[0].u, u_dst +  7 * BPS,  8);
+        memcpy(top_yuv[0].v, v_dst +  7 * BPS,  8);
+      }
+    }
+    // Transfer reconstructed samples from yuv_b_ cache to final destination.
+    {
+      const int y_offset = cache_id * 16 * dec->cache_y_stride_;
+      const int uv_offset = cache_id * 8 * dec->cache_uv_stride_;
+      uint8_t* const y_out = dec->cache_y_ + mb_x * 16 + y_offset;
+      uint8_t* const u_out = dec->cache_u_ + mb_x * 8 + uv_offset;
+      uint8_t* const v_out = dec->cache_v_ + mb_x * 8 + uv_offset;
+      for (j = 0; j < 16; ++j) {
+        memcpy(y_out + j * dec->cache_y_stride_, y_dst + j * BPS, 16);
+      }
+      for (j = 0; j < 8; ++j) {
+        memcpy(u_out + j * dec->cache_uv_stride_, u_dst + j * BPS, 8);
+        memcpy(v_out + j * dec->cache_uv_stride_, v_dst + j * BPS, 8);
       }
     }
   }
@@ -674,6 +826,3 @@ void VP8ReconstructBlock(VP8Decoder* const dec) {
 
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/idec.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Incremental decoding
@@ -13,14 +15,11 @@
 #include <string.h>
 #include <stdlib.h>
 
+#include "./alphai.h"
 #include "./webpi.h"
 #include "./vp8i.h"
 #include "../utils/utils.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 // In append mode, buffer allocations increase as multiples of this value.
 // Needs to be a power of 2.
 #define CHUNK_SIZE 4096
@@ -29,11 +28,13 @@ extern "C" {
 //------------------------------------------------------------------------------
 // Data structures for memory and states
 
-// Decoding states. State normally flows like HEADER->PARTS0->DATA->DONE.
+// Decoding states. State normally flows as:
+// WEBP_HEADER->VP8_HEADER->VP8_PARTS0->VP8_DATA->DONE for a lossy image, and
+// WEBP_HEADER->VP8L_HEADER->VP8L_DATA->DONE for a lossless image.
 // If there is any error the decoder goes into state ERROR.
 typedef enum {
-  STATE_PRE_VP8,  // All data before that of the first VP8 chunk.
-  STATE_VP8_FRAME_HEADER,  // For VP8 Frame header (within VP8 chunk).
+  STATE_WEBP_HEADER,  // All the data before that of the VP8/VP8L chunk.
+  STATE_VP8_HEADER,   // The VP8 Frame header (within the VP8 chunk).
   STATE_VP8_PARTS0,
   STATE_VP8_DATA,
   STATE_VP8L_HEADER,
@@ -71,32 +72,41 @@ struct WebPIDecoder {
   MemBuffer mem_;          // input memory buffer.
   WebPDecBuffer output_;   // output buffer (when no external one is supplied)
   size_t chunk_size_;      // Compressed VP8/VP8L size extracted from Header.
+
+  int last_mb_y_;          // last row reached for intra-mode decoding
 };
 
 // MB context to restore in case VP8DecodeMB() fails
 typedef struct {
   VP8MB left_;
   VP8MB info_;
-  uint8_t intra_t_[4];
-  uint8_t intra_l_[4];
-  VP8BitReader br_;
   VP8BitReader token_br_;
 } MBContext;
 
 //------------------------------------------------------------------------------
 // MemBuffer: incoming data handling
 
-static void RemapBitReader(VP8BitReader* const br, ptrdiff_t offset) {
-  if (br->buf_ != NULL) {
-    br->buf_ += offset;
-    br->buf_end_ += offset;
-  }
-}
-
 static WEBP_INLINE size_t MemDataSize(const MemBuffer* mem) {
   return (mem->end_ - mem->start_);
 }
 
+// Check if we need to preserve the compressed alpha data, as it may not have
+// been decoded yet.
+static int NeedCompressedAlpha(const WebPIDecoder* const idec) {
+  if (idec->state_ == STATE_WEBP_HEADER) {
+    // We haven't parsed the headers yet, so we don't know whether the image is
+    // lossy or lossless. This also means that we haven't parsed the ALPH chunk.
+    return 0;
+  }
+  if (idec->is_lossless_) {
+    return 0;  // ALPH chunk is not present for lossless images.
+  } else {
+    const VP8Decoder* const dec = (VP8Decoder*)idec->dec_;
+    assert(dec != NULL);  // Must be true as idec->state_ != STATE_WEBP_HEADER.
+    return (dec->alpha_data_ != NULL) && !dec->is_alpha_decoded_;
+  }
+}
+
 static void DoRemap(WebPIDecoder* const idec, ptrdiff_t offset) {
   MemBuffer* const mem = &idec->mem_;
   const uint8_t* const new_base = mem->buf_ + mem->start_;
@@ -112,16 +122,32 @@ static void DoRemap(WebPIDecoder* const idec, ptrdiff_t offset) {
       if (offset != 0) {
         int p;
         for (p = 0; p <= last_part; ++p) {
-          RemapBitReader(dec->parts_ + p, offset);
+          VP8RemapBitReader(dec->parts_ + p, offset);
         }
         // Remap partition #0 data pointer to new offset, but only in MAP
         // mode (in APPEND mode, partition #0 is copied into a fixed memory).
         if (mem->mode_ == MEM_MODE_MAP) {
-          RemapBitReader(&dec->br_, offset);
+          VP8RemapBitReader(&dec->br_, offset);
         }
       }
       assert(last_part >= 0);
       dec->parts_[last_part].buf_end_ = mem->buf_ + mem->end_;
+      if (NeedCompressedAlpha(idec)) {
+        ALPHDecoder* const alph_dec = dec->alph_dec_;
+        dec->alpha_data_ += offset;
+        if (alph_dec != NULL) {
+          if (alph_dec->method_ == ALPHA_LOSSLESS_COMPRESSION) {
+            VP8LDecoder* const alph_vp8l_dec = alph_dec->vp8l_dec_;
+            assert(alph_vp8l_dec != NULL);
+            assert(dec->alpha_data_size_ >= ALPHA_HEADER_LEN);
+            VP8LBitReaderSetBuffer(&alph_vp8l_dec->br_,
+                                   dec->alpha_data_ + ALPHA_HEADER_LEN,
+                                   dec->alpha_data_size_ - ALPHA_HEADER_LEN);
+          } else {  // alph_dec->method_ == ALPHA_NO_COMPRESSION
+            // Nothing special to do in this case.
+          }
+        }
+      }
     } else {    // Resize lossless bitreader
       VP8LDecoder* const dec = (VP8LDecoder*)idec->dec_;
       VP8LBitReaderSetBuffer(&dec->br_, new_base, MemDataSize(mem));
@@ -133,8 +159,12 @@ static void DoRemap(WebPIDecoder* const idec, ptrdiff_t offset) {
 // size if required and also updates VP8BitReader's if new memory is allocated.
 static int AppendToMemBuffer(WebPIDecoder* const idec,
                              const uint8_t* const data, size_t data_size) {
+  VP8Decoder* const dec = (VP8Decoder*)idec->dec_;
   MemBuffer* const mem = &idec->mem_;
-  const uint8_t* const old_base = mem->buf_ + mem->start_;
+  const int need_compressed_alpha = NeedCompressedAlpha(idec);
+  const uint8_t* const old_start = mem->buf_ + mem->start_;
+  const uint8_t* const old_base =
+      need_compressed_alpha ? dec->alpha_data_ : old_start;
   assert(mem->mode_ == MEM_MODE_APPEND);
   if (data_size > MAX_CHUNK_PAYLOAD) {
     // security safeguard: trying to allocate more than what the format
@@ -143,17 +173,18 @@ static int AppendToMemBuffer(WebPIDecoder* const idec,
   }
 
   if (mem->end_ + data_size > mem->buf_size_) {  // Need some free memory
-    const size_t current_size = MemDataSize(mem);
+    const size_t new_mem_start = old_start - old_base;
+    const size_t current_size = MemDataSize(mem) + new_mem_start;
     const uint64_t new_size = (uint64_t)current_size + data_size;
     const uint64_t extra_size = (new_size + CHUNK_SIZE - 1) & ~(CHUNK_SIZE - 1);
     uint8_t* const new_buf =
         (uint8_t*)WebPSafeMalloc(extra_size, sizeof(*new_buf));
     if (new_buf == NULL) return 0;
     memcpy(new_buf, old_base, current_size);
-    free(mem->buf_);
+    WebPSafeFree(mem->buf_);
     mem->buf_ = new_buf;
     mem->buf_size_ = (size_t)extra_size;
-    mem->start_ = 0;
+    mem->start_ = new_mem_start;
     mem->end_ = current_size;
   }
 
@@ -161,14 +192,15 @@ static int AppendToMemBuffer(WebPIDecoder* const idec,
   mem->end_ += data_size;
   assert(mem->end_ <= mem->buf_size_);
 
-  DoRemap(idec, mem->buf_ + mem->start_ - old_base);
+  DoRemap(idec, mem->buf_ + mem->start_ - old_start);
   return 1;
 }
 
 static int RemapMemBuffer(WebPIDecoder* const idec,
                           const uint8_t* const data, size_t data_size) {
   MemBuffer* const mem = &idec->mem_;
-  const uint8_t* const old_base = mem->buf_ + mem->start_;
+  const uint8_t* const old_buf = mem->buf_;
+  const uint8_t* const old_start = old_buf + mem->start_;
   assert(mem->mode_ == MEM_MODE_MAP);
 
   if (data_size < mem->buf_size_) return 0;  // can't remap to a shorter buffer!
@@ -176,7 +208,7 @@ static int RemapMemBuffer(WebPIDecoder* const idec,
   mem->buf_ = (uint8_t*)data;
   mem->end_ = mem->buf_size_ = data_size;
 
-  DoRemap(idec, mem->buf_ + mem->start_ - old_base);
+  DoRemap(idec, mem->buf_ + mem->start_ - old_start);
   return 1;
 }
 
@@ -191,8 +223,8 @@ static void InitMemBuffer(MemBuffer* const mem) {
 static void ClearMemBuffer(MemBuffer* const mem) {
   assert(mem);
   if (mem->mode_ == MEM_MODE_APPEND) {
-    free(mem->buf_);
-    free((void*)mem->part0_buf_);
+    WebPSafeFree(mem->buf_);
+    WebPSafeFree((void*)mem->part0_buf_);
   }
 }
 
@@ -206,35 +238,36 @@ static int CheckMemBufferMode(MemBuffer* const mem, MemBufferMode expected) {
   return 1;
 }
 
+// To be called last.
+static VP8StatusCode FinishDecoding(WebPIDecoder* const idec) {
+#if WEBP_DECODER_ABI_VERSION > 0x0203
+  const WebPDecoderOptions* const options = idec->params_.options;
+  WebPDecBuffer* const output = idec->params_.output;
+
+  idec->state_ = STATE_DONE;
+  if (options != NULL && options->flip) {
+    return WebPFlipBuffer(output);
+  }
+#endif
+  idec->state_ = STATE_DONE;
+  return VP8_STATUS_OK;
+}
+
 //------------------------------------------------------------------------------
 // Macroblock-decoding contexts
 
 static void SaveContext(const VP8Decoder* dec, const VP8BitReader* token_br,
                         MBContext* const context) {
-  const VP8BitReader* const br = &dec->br_;
-  const VP8MB* const left = dec->mb_info_ - 1;
-  const VP8MB* const info = dec->mb_info_ + dec->mb_x_;
-
-  context->left_ = *left;
-  context->info_ = *info;
-  context->br_ = *br;
+  context->left_ = dec->mb_info_[-1];
+  context->info_ = dec->mb_info_[dec->mb_x_];
   context->token_br_ = *token_br;
-  memcpy(context->intra_t_, dec->intra_t_ + 4 * dec->mb_x_, 4);
-  memcpy(context->intra_l_, dec->intra_l_, 4);
 }
 
 static void RestoreContext(const MBContext* context, VP8Decoder* const dec,
                            VP8BitReader* const token_br) {
-  VP8BitReader* const br = &dec->br_;
-  VP8MB* const left = dec->mb_info_ - 1;
-  VP8MB* const info = dec->mb_info_ + dec->mb_x_;
-
-  *left = context->left_;
-  *info = context->info_;
-  *br = context->br_;
+  dec->mb_info_[-1] = context->left_;
+  dec->mb_info_[dec->mb_x_] = context->info_;
   *token_br = context->token_br_;
-  memcpy(dec->intra_t_ + 4 * dec->mb_x_, context->intra_t_, 4);
-  memcpy(dec->intra_l_, context->intra_l_, 4);
 }
 
 //------------------------------------------------------------------------------
@@ -242,7 +275,7 @@ static void RestoreContext(const MBContext* context, VP8Decoder* const dec,
 static VP8StatusCode IDecError(WebPIDecoder* const idec, VP8StatusCode error) {
   if (idec->state_ == STATE_VP8_DATA) {
     VP8Io* const io = &idec->io_;
-    if (io->teardown) {
+    if (io->teardown != NULL) {
       io->teardown(io);
     }
   }
@@ -270,6 +303,7 @@ static VP8StatusCode DecodeWebPHeaders(WebPIDecoder* const idec) {
 
   headers.data = data;
   headers.data_size = curr_size;
+  headers.have_all_data = 0;
   status = WebPParseHeaders(&headers);
   if (status == VP8_STATUS_NOT_ENOUGH_DATA) {
     return VP8_STATUS_SUSPENDED;  // We haven't found a VP8 chunk yet.
@@ -285,15 +319,9 @@ static VP8StatusCode DecodeWebPHeaders(WebPIDecoder* const idec) {
       return VP8_STATUS_OUT_OF_MEMORY;
     }
     idec->dec_ = dec;
-#ifdef WEBP_USE_THREAD
-    dec->use_threads_ = (idec->params_.options != NULL) &&
-                        (idec->params_.options->use_threads > 0);
-#else
-    dec->use_threads_ = 0;
-#endif
     dec->alpha_data_ = headers.alpha_data;
     dec->alpha_data_size_ = headers.alpha_data_size;
-    ChangeState(idec, STATE_VP8_FRAME_HEADER, headers.offset);
+    ChangeState(idec, STATE_VP8_HEADER, headers.offset);
   } else {
     VP8LDecoder* const dec = VP8LNew();
     if (dec == NULL) {
@@ -308,13 +336,14 @@ static VP8StatusCode DecodeWebPHeaders(WebPIDecoder* const idec) {
 static VP8StatusCode DecodeVP8FrameHeader(WebPIDecoder* const idec) {
   const uint8_t* data = idec->mem_.buf_ + idec->mem_.start_;
   const size_t curr_size = MemDataSize(&idec->mem_);
+  int width, height;
   uint32_t bits;
 
   if (curr_size < VP8_FRAME_HEADER_SIZE) {
     // Not enough data bytes to extract VP8 Frame Header.
     return VP8_STATUS_SUSPENDED;
   }
-  if (!VP8GetInfo(data, curr_size, idec->chunk_size_, NULL, NULL)) {
+  if (!VP8GetInfo(data, curr_size, idec->chunk_size_, &width, &height)) {
     return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
   }
 
@@ -339,7 +368,7 @@ static int CopyParts0Data(WebPIDecoder* const idec) {
   assert(psize <= mem->part0_size_);  // Format limit: no need for runtime check
   if (mem->mode_ == MEM_MODE_APPEND) {
     // We copy and grab ownership of the partition #0 data.
-    uint8_t* const part0_buf = (uint8_t*)malloc(psize);
+    uint8_t* const part0_buf = (uint8_t*)WebPSafeMalloc(1ULL, psize);
     if (part0_buf == NULL) {
       return 0;
     }
@@ -381,7 +410,10 @@ static VP8StatusCode DecodePartition0(WebPIDecoder* const idec) {
   if (dec->status_ != VP8_STATUS_OK) {
     return IDecError(idec, dec->status_);
   }
-
+  // This change must be done before calling VP8InitFrame()
+  dec->mt_method_ = VP8GetThreadMethod(params->options, NULL,
+                                       io->width, io->height);
+  VP8InitDithering(params->options, dec);
   if (!CopyParts0Data(idec)) {
     return IDecError(idec, VP8_STATUS_OUT_OF_MEMORY);
   }
@@ -407,50 +439,52 @@ static VP8StatusCode DecodeRemaining(WebPIDecoder* const idec) {
   VP8Io* const io = &idec->io_;
 
   assert(dec->ready_);
-
   for (; dec->mb_y_ < dec->mb_h_; ++dec->mb_y_) {
-    VP8BitReader* token_br = &dec->parts_[dec->mb_y_ & (dec->num_parts_ - 1)];
-    if (dec->mb_x_ == 0) {
-      VP8InitScanline(dec);
+    if (idec->last_mb_y_ != dec->mb_y_) {
+      if (!VP8ParseIntraModeRow(&dec->br_, dec)) {
+        // note: normally, error shouldn't occur since we already have the whole
+        // partition0 available here in DecodeRemaining(). Reaching EOF while
+        // reading intra modes really means a BITSTREAM_ERROR.
+        return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
+      }
+      idec->last_mb_y_ = dec->mb_y_;
     }
-    for (; dec->mb_x_ < dec->mb_w_;  dec->mb_x_++) {
+    for (; dec->mb_x_ < dec->mb_w_; ++dec->mb_x_) {
+      VP8BitReader* const token_br =
+          &dec->parts_[dec->mb_y_ & (dec->num_parts_ - 1)];
       MBContext context;
       SaveContext(dec, token_br, &context);
-
       if (!VP8DecodeMB(dec, token_br)) {
-        RestoreContext(&context, dec, token_br);
         // We shouldn't fail when MAX_MB data was available
         if (dec->num_parts_ == 1 && MemDataSize(&idec->mem_) > MAX_MB_SIZE) {
           return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
         }
+        RestoreContext(&context, dec, token_br);
         return VP8_STATUS_SUSPENDED;
       }
-      VP8ReconstructBlock(dec);
-      // Store data and save block's filtering params
-      VP8StoreBlock(dec);
-
       // Release buffer only if there is only one partition
       if (dec->num_parts_ == 1) {
         idec->mem_.start_ = token_br->buf_ - idec->mem_.buf_;
         assert(idec->mem_.start_ <= idec->mem_.end_);
       }
     }
+    VP8InitScanline(dec);   // Prepare for next scanline
+
+    // Reconstruct, filter and emit the row.
     if (!VP8ProcessRow(dec, io)) {
       return IDecError(idec, VP8_STATUS_USER_ABORT);
     }
-    dec->mb_x_ = 0;
   }
   // Synchronize the thread and check for errors.
   if (!VP8ExitCritical(dec, io)) {
     return IDecError(idec, VP8_STATUS_USER_ABORT);
   }
   dec->ready_ = 0;
-  idec->state_ = STATE_DONE;
-
-  return VP8_STATUS_OK;
+  return FinishDecoding(idec);
 }
 
-static int ErrorStatusLossless(WebPIDecoder* const idec, VP8StatusCode status) {
+static VP8StatusCode ErrorStatusLossless(WebPIDecoder* const idec,
+                                         VP8StatusCode status) {
   if (status == VP8_STATUS_SUSPENDED || status == VP8_STATUS_NOT_ENOUGH_DATA) {
     return VP8_STATUS_SUSPENDED;
   }
@@ -495,26 +529,30 @@ static VP8StatusCode DecodeVP8LData(WebPIDecoder* const idec) {
   }
 
   if (!VP8LDecodeImage(dec)) {
+    // The decoding is called after all the data-bytes are aggregated. Change
+    // the error to VP8_BITSTREAM_ERROR in case lossless decoder fails to decode
+    // all the pixels (VP8_STATUS_SUSPENDED).
+    if (dec->status_ == VP8_STATUS_SUSPENDED) {
+      dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
+    }
     return ErrorStatusLossless(idec, dec->status_);
   }
 
-  idec->state_ = STATE_DONE;
-
-  return VP8_STATUS_OK;
+  return FinishDecoding(idec);
 }
 
   // Main decoding loop
 static VP8StatusCode IDecode(WebPIDecoder* idec) {
   VP8StatusCode status = VP8_STATUS_SUSPENDED;
 
-  if (idec->state_ == STATE_PRE_VP8) {
+  if (idec->state_ == STATE_WEBP_HEADER) {
     status = DecodeWebPHeaders(idec);
   } else {
     if (idec->dec_ == NULL) {
       return VP8_STATUS_SUSPENDED;    // can't continue if we have no decoder.
     }
   }
-  if (idec->state_ == STATE_VP8_FRAME_HEADER) {
+  if (idec->state_ == STATE_VP8_HEADER) {
     status = DecodeVP8FrameHeader(idec);
   }
   if (idec->state_ == STATE_VP8_PARTS0) {
@@ -536,20 +574,23 @@ static VP8StatusCode IDecode(WebPIDecoder* idec) {
 // Public functions
 
 WebPIDecoder* WebPINewDecoder(WebPDecBuffer* output_buffer) {
-  WebPIDecoder* idec = (WebPIDecoder*)calloc(1, sizeof(*idec));
+  WebPIDecoder* idec = (WebPIDecoder*)WebPSafeCalloc(1ULL, sizeof(*idec));
   if (idec == NULL) {
     return NULL;
   }
 
-  idec->state_ = STATE_PRE_VP8;
+  idec->state_ = STATE_WEBP_HEADER;
   idec->chunk_size_ = 0;
 
+  idec->last_mb_y_ = -1;
+
   InitMemBuffer(&idec->mem_);
   WebPInitDecBuffer(&idec->output_);
   VP8InitIo(&idec->io_);
 
   WebPResetDecParams(&idec->params_);
-  idec->params_.output = output_buffer ? output_buffer : &idec->output_;
+  idec->params_.output = (output_buffer != NULL) ? output_buffer
+                                                 : &idec->output_;
   WebPInitCustomIo(&idec->params_, &idec->io_);  // Plug the I/O functions.
 
   return idec;
@@ -581,14 +622,18 @@ void WebPIDelete(WebPIDecoder* idec) {
   if (idec == NULL) return;
   if (idec->dec_ != NULL) {
     if (!idec->is_lossless_) {
-      VP8Delete(idec->dec_);
+      if (idec->state_ == STATE_VP8_DATA) {
+        // Synchronize the thread, clean-up and check for errors.
+        VP8ExitCritical((VP8Decoder*)idec->dec_, &idec->io_);
+      }
+      VP8Delete((VP8Decoder*)idec->dec_);
     } else {
-      VP8LDelete(idec->dec_);
+      VP8LDelete((VP8LDecoder*)idec->dec_);
     }
   }
   ClearMemBuffer(&idec->mem_);
   WebPFreeDecBuffer(&idec->output_);
-  free(idec);
+  WebPSafeFree(idec);
 }
 
 //------------------------------------------------------------------------------
@@ -596,12 +641,22 @@ void WebPIDelete(WebPIDecoder* idec) {
 
 WebPIDecoder* WebPINewRGB(WEBP_CSP_MODE mode, uint8_t* output_buffer,
                           size_t output_buffer_size, int output_stride) {
+  const int is_external_memory = (output_buffer != NULL);
   WebPIDecoder* idec;
+
   if (mode >= MODE_YUV) return NULL;
+  if (!is_external_memory) {    // Overwrite parameters to sane values.
+    output_buffer_size = 0;
+    output_stride = 0;
+  } else {  // A buffer was passed. Validate the other params.
+    if (output_stride == 0 || output_buffer_size == 0) {
+      return NULL;   // invalid parameter.
+    }
+  }
   idec = WebPINewDecoder(NULL);
   if (idec == NULL) return NULL;
   idec->output_.colorspace = mode;
-  idec->output_.is_external_memory = 1;
+  idec->output_.is_external_memory = is_external_memory;
   idec->output_.u.RGBA.rgba = output_buffer;
   idec->output_.u.RGBA.stride = output_stride;
   idec->output_.u.RGBA.size = output_buffer_size;
@@ -612,10 +667,30 @@ WebPIDecoder* WebPINewYUVA(uint8_t* luma, size_t luma_size, int luma_stride,
                            uint8_t* u, size_t u_size, int u_stride,
                            uint8_t* v, size_t v_size, int v_stride,
                            uint8_t* a, size_t a_size, int a_stride) {
-  WebPIDecoder* const idec = WebPINewDecoder(NULL);
+  const int is_external_memory = (luma != NULL);
+  WebPIDecoder* idec;
+  WEBP_CSP_MODE colorspace;
+
+  if (!is_external_memory) {    // Overwrite parameters to sane values.
+    luma_size = u_size = v_size = a_size = 0;
+    luma_stride = u_stride = v_stride = a_stride = 0;
+    u = v = a = NULL;
+    colorspace = MODE_YUVA;
+  } else {  // A luma buffer was passed. Validate the other parameters.
+    if (u == NULL || v == NULL) return NULL;
+    if (luma_size == 0 || u_size == 0 || v_size == 0) return NULL;
+    if (luma_stride == 0 || u_stride == 0 || v_stride == 0) return NULL;
+    if (a != NULL) {
+      if (a_size == 0 || a_stride == 0) return NULL;
+    }
+    colorspace = (a == NULL) ? MODE_YUV : MODE_YUVA;
+  }
+
+  idec = WebPINewDecoder(NULL);
   if (idec == NULL) return NULL;
-  idec->output_.colorspace = (a == NULL) ? MODE_YUV : MODE_YUVA;
-  idec->output_.is_external_memory = 1;
+
+  idec->output_.colorspace = colorspace;
+  idec->output_.is_external_memory = is_external_memory;
   idec->output_.u.YUVA.y = luma;
   idec->output_.u.YUVA.y_stride = luma_stride;
   idec->output_.u.YUVA.y_size = luma_size;
@@ -768,7 +843,7 @@ int WebPISetIOHooks(WebPIDecoder* const idec,
                     VP8IoSetupHook setup,
                     VP8IoTeardownHook teardown,
                     void* user_data) {
-  if (idec == NULL || idec->state_ > STATE_PRE_VP8) {
+  if (idec == NULL || idec->state_ > STATE_WEBP_HEADER) {
     return 0;
   }
 
@@ -780,6 +855,3 @@ int WebPISetIOHooks(WebPIDecoder* const idec,
   return 1;
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/io.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // functions for sample output.
@@ -15,10 +17,7 @@
 #include "./webpi.h"
 #include "../dsp/dsp.h"
 #include "../dsp/yuv.h"
-
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
+#include "../utils/utils.h"
 
 //------------------------------------------------------------------------------
 // Main YUV<->RGB conversion functions
@@ -46,27 +45,13 @@ static int EmitYUV(const VP8Io* const io, WebPDecParams* const p) {
 
 // Point-sampling U/V sampler.
 static int EmitSampledRGB(const VP8Io* const io, WebPDecParams* const p) {
-  WebPDecBuffer* output = p->output;
-  const WebPRGBABuffer* const buf = &output->u.RGBA;
-  uint8_t* dst = buf->rgba + io->mb_y * buf->stride;
-  const uint8_t* y_src = io->y;
-  const uint8_t* u_src = io->u;
-  const uint8_t* v_src = io->v;
-  const WebPSampleLinePairFunc sample = WebPSamplers[output->colorspace];
-  const int mb_w = io->mb_w;
-  const int last = io->mb_h - 1;
-  int j;
-  for (j = 0; j < last; j += 2) {
-    sample(y_src, y_src + io->y_stride, u_src, v_src,
-           dst, dst + buf->stride, mb_w);
-    y_src += 2 * io->y_stride;
-    u_src += io->uv_stride;
-    v_src += io->uv_stride;
-    dst += 2 * buf->stride;
-  }
-  if (j == last) {  // Just do the last line twice
-    sample(y_src, y_src, u_src, v_src, dst, dst, mb_w);
-  }
+  WebPDecBuffer* const output = p->output;
+  WebPRGBABuffer* const buf = &output->u.RGBA;
+  uint8_t* const dst = buf->rgba + io->mb_y * buf->stride;
+  WebPSamplerProcessPlane(io->y, io->y_stride,
+                          io->u, io->v, io->uv_stride,
+                          dst, buf->stride, io->mb_w, io->mb_h,
+                          WebPSamplers[output->colorspace]);
   return io->mb_h;
 }
 
@@ -117,7 +102,7 @@ static int EmitFancyRGB(const VP8Io* const io, WebPDecParams* const p) {
 
   if (y == 0) {
     // First line is special cased. We mirror the u/v samples at boundary.
-    upsample(NULL, cur_y, cur_u, cur_v, cur_u, cur_v, NULL, dst, mb_w);
+    upsample(cur_y, NULL, cur_u, cur_v, cur_u, cur_v, dst, NULL, mb_w);
   } else {
     // We can finish the left-over line from previous call.
     upsample(p->tmp_y, cur_y, top_u, top_v, cur_u, cur_v,
@@ -252,7 +237,11 @@ static int EmitAlphaRGBA4444(const VP8Io* const io, WebPDecParams* const p) {
     int num_rows;
     const int start_y = GetAlphaSourceRow(io, &alpha, &num_rows);
     uint8_t* const base_rgba = buf->rgba + start_y * buf->stride;
+#ifdef WEBP_SWAP_16BIT_CSP
+    uint8_t* alpha_dst = base_rgba;
+#else
     uint8_t* alpha_dst = base_rgba + 1;
+#endif
     uint32_t alpha_mask = 0x0f;
     int i, j;
 
@@ -291,7 +280,17 @@ static int Rescale(const uint8_t* src, int src_stride,
 static int EmitRescaledYUV(const VP8Io* const io, WebPDecParams* const p) {
   const int mb_h = io->mb_h;
   const int uv_mb_h = (mb_h + 1) >> 1;
-  const int num_lines_out = Rescale(io->y, io->y_stride, mb_h, &p->scaler_y);
+  WebPRescaler* const scaler = &p->scaler_y;
+  int num_lines_out = 0;
+  if (WebPIsAlphaMode(p->output->colorspace) && io->a != NULL) {
+    // Before rescaling, we premultiply the luma directly into the io->y
+    // internal buffer. This is OK since these samples are not used for
+    // intra-prediction (the top samples are saved in cache_y_/u_/v_).
+    // But we need to cast the const away, though.
+    WebPMultRows((uint8_t*)io->y, io->y_stride,
+                 io->a, io->width, io->mb_w, mb_h, 0);
+  }
+  num_lines_out = Rescale(io->y, io->y_stride, mb_h, scaler);
   Rescale(io->u, io->uv_stride, uv_mb_h, &p->scaler_u);
   Rescale(io->v, io->uv_stride, uv_mb_h, &p->scaler_v);
   return num_lines_out;
@@ -299,7 +298,14 @@ static int EmitRescaledYUV(const VP8Io* const io, WebPDecParams* const p) {
 
 static int EmitRescaledAlphaYUV(const VP8Io* const io, WebPDecParams* const p) {
   if (io->a != NULL) {
-    Rescale(io->a, io->width, io->mb_h, &p->scaler_a);
+    const WebPYUVABuffer* const buf = &p->output->u.YUVA;
+    uint8_t* dst_y = buf->y + p->last_y * buf->y_stride;
+    const uint8_t* src_a = buf->a + p->last_y * buf->a_stride;
+    const int num_lines_out = Rescale(io->a, io->width, io->mb_h, &p->scaler_a);
+    if (num_lines_out > 0) {   // unmultiply the Y
+      WebPMultRows(dst_y, buf->y_stride, src_a, buf->a_stride,
+                   p->scaler_a.dst_width, num_lines_out, 1);
+    }
   }
   return 0;
 }
@@ -318,11 +324,11 @@ static int InitYUVRescaler(const VP8Io* const io, WebPDecParams* const p) {
   size_t tmp_size;
   int32_t* work;
 
-  tmp_size = work_size + 2 * uv_work_size;
+  tmp_size = (work_size + 2 * uv_work_size) * sizeof(*work);
   if (has_alpha) {
-    tmp_size += work_size;
+    tmp_size += work_size * sizeof(*work);
   }
-  p->memory = calloc(1, tmp_size * sizeof(*work));
+  p->memory = WebPSafeCalloc(1ULL, tmp_size);
   if (p->memory == NULL) {
     return 0;   // memory error
   }
@@ -349,6 +355,7 @@ static int InitYUVRescaler(const VP8Io* const io, WebPDecParams* const p) {
                      io->mb_w, out_width, io->mb_h, out_height,
                      work + work_size + 2 * uv_work_size);
     p->emit_alpha = EmitRescaledAlphaYUV;
+    WebPInitAlphaProcessing();
   }
   return 1;
 }
@@ -368,9 +375,9 @@ static int ExportRGB(WebPDecParams* const p, int y_pos) {
          WebPRescalerHasPendingOutput(&p->scaler_u)) {
     assert(p->last_y + y_pos + num_lines_out < p->output->height);
     assert(p->scaler_u.y_accum == p->scaler_v.y_accum);
-    WebPRescalerExportRow(&p->scaler_y);
-    WebPRescalerExportRow(&p->scaler_u);
-    WebPRescalerExportRow(&p->scaler_v);
+    WebPRescalerExportRow(&p->scaler_y, 0);
+    WebPRescalerExportRow(&p->scaler_u, 0);
+    WebPRescalerExportRow(&p->scaler_v, 0);
     convert(p->scaler_y.dst, p->scaler_u.dst, p->scaler_v.dst,
             dst, p->scaler_y.dst_width);
     dst += buf->stride;
@@ -418,7 +425,7 @@ static int ExportAlpha(WebPDecParams* const p, int y_pos) {
   while (WebPRescalerHasPendingOutput(&p->scaler_a)) {
     int i;
     assert(p->last_y + y_pos + num_lines_out < p->output->height);
-    WebPRescalerExportRow(&p->scaler_a);
+    WebPRescalerExportRow(&p->scaler_a, 0);
     for (i = 0; i < width; ++i) {
       const uint32_t alpha_value = p->scaler_a.dst[i];
       dst[4 * i] = alpha_value;
@@ -437,7 +444,11 @@ static int ExportAlpha(WebPDecParams* const p, int y_pos) {
 static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos) {
   const WebPRGBABuffer* const buf = &p->output->u.RGBA;
   uint8_t* const base_rgba = buf->rgba + (p->last_y + y_pos) * buf->stride;
+#ifdef WEBP_SWAP_16BIT_CSP
+  uint8_t* alpha_dst = base_rgba;
+#else
   uint8_t* alpha_dst = base_rgba + 1;
+#endif
   int num_lines_out = 0;
   const WEBP_CSP_MODE colorspace = p->output->colorspace;
   const int width = p->scaler_a.dst_width;
@@ -447,7 +458,7 @@ static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos) {
   while (WebPRescalerHasPendingOutput(&p->scaler_a)) {
     int i;
     assert(p->last_y + y_pos + num_lines_out < p->output->height);
-    WebPRescalerExportRow(&p->scaler_a);
+    WebPRescalerExportRow(&p->scaler_a, 0);
     for (i = 0; i < width; ++i) {
       // Fill in the alpha value (converted to 4 bits).
       const uint32_t alpha_value = p->scaler_a.dst[i] >> 4;
@@ -486,7 +497,7 @@ static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) {
   const size_t work_size = 2 * out_width;   // scratch memory for one rescaler
   int32_t* work;  // rescalers work area
   uint8_t* tmp;   // tmp storage for scaled YUV444 samples before RGB conversion
-  size_t tmp_size1, tmp_size2;
+  size_t tmp_size1, tmp_size2, total_size;
 
   tmp_size1 = 3 * work_size;
   tmp_size2 = 3 * out_width;
@@ -494,7 +505,8 @@ static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) {
     tmp_size1 += work_size;
     tmp_size2 += out_width;
   }
-  p->memory = calloc(1, tmp_size1 * sizeof(*work) + tmp_size2 * sizeof(*tmp));
+  total_size = tmp_size1 * sizeof(*work) + tmp_size2 * sizeof(*tmp);
+  p->memory = WebPSafeCalloc(1ULL, total_size);
   if (p->memory == NULL) {
     return 0;   // memory error
   }
@@ -526,6 +538,7 @@ static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) {
     } else {
       p->emit_alpha_row = ExportAlpha;
     }
+    WebPInitAlphaProcessing();
   }
   return 1;
 }
@@ -546,7 +559,9 @@ static int CustomSetup(VP8Io* io) {
   if (!WebPIoInitFromOptions(p->options, io, is_alpha ? MODE_YUV : MODE_YUVA)) {
     return 0;
   }
-
+  if (is_alpha && WebPIsPremultipliedMode(colorspace)) {
+    WebPInitUpsamplers();
+  }
   if (io->use_scaling) {
     const int ok = is_rgb ? InitRGBRescaler(io, p) : InitYUVRescaler(io, p);
     if (!ok) {
@@ -555,10 +570,10 @@ static int CustomSetup(VP8Io* io) {
   } else {
     if (is_rgb) {
       p->emit = EmitSampledRGB;   // default
-#ifdef FANCY_UPSAMPLING
       if (io->fancy_upsampling) {
+#ifdef FANCY_UPSAMPLING
         const int uv_width = (io->mb_w + 1) >> 1;
-        p->memory = malloc(io->mb_w + 2 * uv_width);
+        p->memory = WebPSafeMalloc(1ULL, (size_t)(io->mb_w + 2 * uv_width));
         if (p->memory == NULL) {
           return 0;   // memory error.
         }
@@ -567,18 +582,22 @@ static int CustomSetup(VP8Io* io) {
         p->tmp_v = p->tmp_u + uv_width;
         p->emit = EmitFancyRGB;
         WebPInitUpsamplers();
-      }
 #endif
+      } else {
+        WebPInitSamplers();
+      }
     } else {
       p->emit = EmitYUV;
     }
     if (is_alpha) {  // need transparency output
-      if (WebPIsPremultipliedMode(colorspace)) WebPInitPremultiply();
       p->emit_alpha =
           (colorspace == MODE_RGBA_4444 || colorspace == MODE_rgbA_4444) ?
               EmitAlphaRGBA4444
           : is_rgb ? EmitAlphaRGB
           : EmitAlphaYUV;
+      if (is_rgb) {
+        WebPInitAlphaProcessing();
+      }
     }
   }
 
@@ -601,7 +620,7 @@ static int CustomPut(const VP8Io* io) {
     return 0;
   }
   num_lines_out = p->emit(io, p);
-  if (p->emit_alpha) {
+  if (p->emit_alpha != NULL) {
     p->emit_alpha(io, p);
   }
   p->last_y += num_lines_out;
@@ -612,7 +631,7 @@ static int CustomPut(const VP8Io* io) {
 
 static void CustomTeardown(const VP8Io* io) {
   WebPDecParams* const p = (WebPDecParams*)io->opaque;
-  free(p->memory);
+  WebPSafeFree(p->memory);
   p->memory = NULL;
 }
 
@@ -627,7 +646,3 @@ void WebPInitCustomIo(WebPDecParams* const params, VP8Io* const io) {
 }
 
 //------------------------------------------------------------------------------
-
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/layer.c

@@ -1,35 +0,0 @@
-// Copyright 2011 Google Inc. All Rights Reserved.
-//
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
-// -----------------------------------------------------------------------------
-//
-// Enhancement layer (for YUV444/422)
-//
-// Author: Skal (pascal.massimino@gmail.com)
-
-#include <assert.h>
-#include <stdlib.h>
-
-#include "./vp8i.h"
-
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
-//------------------------------------------------------------------------------
-
-int VP8DecodeLayer(VP8Decoder* const dec) {
-  assert(dec);
-  assert(dec->layer_data_size_ > 0);
-  (void)dec;
-
-  // TODO: handle enhancement layer here.
-
-  return 1;
-}
-
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/quant.c

@@ -1,8 +1,10 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Quantizer initialization
@@ -11,10 +13,6 @@
 
 #include "./vp8i.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 static WEBP_INLINE int clip(int v, int M) {
   return v < 0 ? 0 : v > M ? M : v;
 }
@@ -102,12 +100,11 @@ void VP8ParseQuant(VP8Decoder* const dec) {
 
       m->uv_mat_[0] = kDcTable[clip(q + dquv_dc, 117)];
       m->uv_mat_[1] = kAcTable[clip(q + dquv_ac, 127)];
+
+      m->uv_quant_ = q + dquv_ac;   // for dithering strength evaluation
     }
   }
 }
 
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/tree.c

@@ -1,22 +1,21 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Coding trees and probas
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "vp8i.h"
+#include "./vp8i.h"
+#include "../utils/bit_reader_inl.h"
 
 #define USE_GENERIC_TREE
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 #ifdef USE_GENERIC_TREE
 static const int8_t kYModesIntra4[18] = {
   -B_DC_PRED, 1,
@@ -31,61 +30,12 @@ static const int8_t kYModesIntra4[18] = {
 };
 #endif
 
-#ifndef ONLY_KEYFRAME_CODE
-
-// inter prediction modes
-enum {
-  LEFT4 = 0, ABOVE4 = 1, ZERO4 = 2, NEW4 = 3,
-  NEARESTMV, NEARMV, ZEROMV, NEWMV, SPLITMV };
-
-static const int8_t kYModesInter[8] = {
-  -DC_PRED, 1,
-    2, 3,
-      -V_PRED, -H_PRED,
-      -TM_PRED, -B_PRED
-};
-
-static const int8_t kMBSplit[6] = {
-  -3, 1,
-    -2, 2,
-      -0, -1
-};
-
-static const int8_t kMVRef[8] = {
-  -ZEROMV, 1,
-    -NEARESTMV, 2,
-      -NEARMV, 3,
-        -NEWMV, -SPLITMV
-};
-
-static const int8_t kMVRef4[6] = {
-  -LEFT4, 1,
-    -ABOVE4, 2,
-      -ZERO4, -NEW4
-};
-#endif
-
 //------------------------------------------------------------------------------
 // Default probabilities
 
-// Inter
-#ifndef ONLY_KEYFRAME_CODE
-static const uint8_t kYModeProbaInter0[4] = { 112, 86, 140, 37 };
-static const uint8_t kUVModeProbaInter0[3] = { 162, 101, 204 };
-static const uint8_t kMVProba0[2][NUM_MV_PROBAS] = {
-  { 162, 128, 225, 146, 172, 147, 214,  39,
-    156, 128, 129, 132,  75, 145, 178, 206,
-    239, 254, 254 },
-  { 164, 128, 204, 170, 119, 235, 140, 230,
-    228, 128, 130, 130,  74, 148, 180, 203,
-    236, 254, 254 }
-};
-#endif
-
 // Paragraph 13.5
 static const uint8_t
   CoeffsProba0[NUM_TYPES][NUM_BANDS][NUM_CTX][NUM_PROBAS] = {
-  // genereated using vp8_default_coef_probs() in entropy.c:129
   { { { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
       { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
       { 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 }
@@ -326,28 +276,38 @@ static const uint8_t kBModesProba[NUM_BMODES][NUM_BMODES][NUM_BMODES - 1] = {
 
 void VP8ResetProba(VP8Proba* const proba) {
   memset(proba->segments_, 255u, sizeof(proba->segments_));
-  memcpy(proba->coeffs_, CoeffsProba0, sizeof(CoeffsProba0));
-#ifndef ONLY_KEYFRAME_CODE
-  memcpy(proba->mv_, kMVProba0, sizeof(kMVProba0));
-  memcpy(proba->ymode_, kYModeProbaInter0, sizeof(kYModeProbaInter0));
-  memcpy(proba->uvmode_, kUVModeProbaInter0, sizeof(kUVModeProbaInter0));
-#endif
+  // proba->bands_[][] is initialized later
 }
 
-void VP8ParseIntraMode(VP8BitReader* const br,  VP8Decoder* const dec) {
-  uint8_t* const top = dec->intra_t_ + 4 * dec->mb_x_;
+static void ParseIntraMode(VP8BitReader* const br,
+                           VP8Decoder* const dec, int mb_x) {
+  uint8_t* const top = dec->intra_t_ + 4 * mb_x;
   uint8_t* const left = dec->intra_l_;
-  // Hardcoded 16x16 intra-mode decision tree.
-  dec->is_i4x4_ = !VP8GetBit(br, 145);   // decide for B_PRED first
-  if (!dec->is_i4x4_) {
+  VP8MBData* const block = dec->mb_data_ + mb_x;
+
+  // Note: we don't save segment map (yet), as we don't expect
+  // to decode more than 1 keyframe.
+  if (dec->segment_hdr_.update_map_) {
+    // Hardcoded tree parsing
+    block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0])
+                    ? VP8GetBit(br, dec->proba_.segments_[1])
+                    : 2 + VP8GetBit(br, dec->proba_.segments_[2]);
+  } else {
+    block->segment_ = 0;  // default for intra
+  }
+  if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_);
+
+  block->is_i4x4_ = !VP8GetBit(br, 145);   // decide for B_PRED first
+  if (!block->is_i4x4_) {
+    // Hardcoded 16x16 intra-mode decision tree.
     const int ymode =
         VP8GetBit(br, 156) ? (VP8GetBit(br, 128) ? TM_PRED : H_PRED)
                            : (VP8GetBit(br, 163) ? V_PRED : DC_PRED);
-    dec->imodes_[0] = ymode;
-    memset(top, ymode, 4 * sizeof(top[0]));
-    memset(left, ymode, 4 * sizeof(left[0]));
+    block->imodes_[0] = ymode;
+    memset(top, ymode, 4 * sizeof(*top));
+    memset(left, ymode, 4 * sizeof(*left));
   } else {
-    uint8_t* modes = dec->imodes_;
+    uint8_t* modes = block->imodes_;
     int y;
     for (y = 0; y < 4; ++y) {
       int ymode = left[y];
@@ -356,10 +316,10 @@ void VP8ParseIntraMode(VP8BitReader* const br,  VP8Decoder* const dec) {
         const uint8_t* const prob = kBModesProba[top[x]][ymode];
 #ifdef USE_GENERIC_TREE
         // Generic tree-parsing
-        int i = 0;
-        do {
+        int i = kYModesIntra4[VP8GetBit(br, prob[0])];
+        while (i > 0) {
           i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i])];
-        } while (i > 0);
+        }
         ymode = -i;
 #else
         // Hardcoded tree parsing
@@ -374,15 +334,24 @@ void VP8ParseIntraMode(VP8BitReader* const br,  VP8Decoder* const dec) {
                             (!VP8GetBit(br, prob[8]) ? B_HD_PRED : B_HU_PRED)));
 #endif    // USE_GENERIC_TREE
         top[x] = ymode;
-        *modes++ = ymode;
       }
+      memcpy(modes, top, 4 * sizeof(*top));
+      modes += 4;
       left[y] = ymode;
     }
   }
   // Hardcoded UVMode decision tree
-  dec->uvmode_ = !VP8GetBit(br, 142) ? DC_PRED
-               : !VP8GetBit(br, 114) ? V_PRED
-               : VP8GetBit(br, 183) ? TM_PRED : H_PRED;
+  block->uvmode_ = !VP8GetBit(br, 142) ? DC_PRED
+                 : !VP8GetBit(br, 114) ? V_PRED
+                 : VP8GetBit(br, 183) ? TM_PRED : H_PRED;
+}
+
+int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec) {
+  int mb_x;
+  for (mb_x = 0; mb_x < dec->mb_w_; ++mb_x) {
+    ParseIntraMode(br, dec, mb_x);
+  }
+  return !dec->br_.eof_;
 }
 
 //------------------------------------------------------------------------------
@@ -524,17 +493,6 @@ static const uint8_t
   }
 };
 
-#ifndef ONLY_KEYFRAME_CODE
-static const uint8_t MVUpdateProba[2][NUM_MV_PROBAS] = {
-  { 237, 246, 253, 253, 254, 254, 254, 254,
-    254, 254, 254, 254, 254, 254, 250, 250,
-    252, 254, 254 },
-  { 231, 243, 245, 253, 254, 254, 254, 254,
-    254, 254, 254, 254, 254, 254, 251, 251,
-    254, 254, 254 }
-};
-#endif
-
 // Paragraph 9.9
 void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
   VP8Proba* const proba = &dec->proba_;
@@ -543,9 +501,9 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
     for (b = 0; b < NUM_BANDS; ++b) {
       for (c = 0; c < NUM_CTX; ++c) {
         for (p = 0; p < NUM_PROBAS; ++p) {
-          if (VP8GetBit(br, CoeffsUpdateProba[t][b][c][p])) {
-            proba->coeffs_[t][b][c][p] = VP8GetValue(br, 8);
-          }
+          const int v = VP8GetBit(br, CoeffsUpdateProba[t][b][c][p]) ?
+                        VP8GetValue(br, 8) : CoeffsProba0[t][b][c][p];
+          proba->bands_[t][b].probas_[c][p] = v;
         }
       }
     }
@@ -554,36 +512,5 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
   if (dec->use_skip_proba_) {
     dec->skip_p_ = VP8GetValue(br, 8);
   }
-#ifndef ONLY_KEYFRAME_CODE
-  if (!dec->frm_hdr_.key_frame_) {
-    int i;
-    dec->intra_p_ = VP8GetValue(br, 8);
-    dec->last_p_ = VP8GetValue(br, 8);
-    dec->golden_p_ = VP8GetValue(br, 8);
-    if (VP8Get(br)) {   // update y-mode
-      for (i = 0; i < 4; ++i) {
-        proba->ymode_[i] = VP8GetValue(br, 8);
-      }
-    }
-    if (VP8Get(br)) {   // update uv-mode
-      for (i = 0; i < 3; ++i) {
-        proba->uvmode_[i] = VP8GetValue(br, 8);
-      }
-    }
-    // update MV
-    for (i = 0; i < 2; ++i) {
-      int k;
-      for (k = 0; k < NUM_MV_PROBAS; ++k) {
-        if (VP8GetBit(br, MVUpdateProba[i][k])) {
-          const int v = VP8GetValue(br, 7);
-          proba->mv_[i][k] = v ? v << 1 : 1;
-        }
-      }
-    }
-  }
-#endif
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/vp8.c

@@ -1,8 +1,10 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // main entry for the decoder
@@ -11,14 +13,12 @@
 
 #include <stdlib.h>
 
+#include "./alphai.h"
 #include "./vp8i.h"
 #include "./vp8li.h"
 #include "./webpi.h"
-#include "../utils/bit_reader.h"
-
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
+#include "../utils/bit_reader_inl.h"
+#include "../utils/utils.h"
 
 //------------------------------------------------------------------------------
 
@@ -45,10 +45,10 @@ int VP8InitIoInternal(VP8Io* const io, int version) {
 }
 
 VP8Decoder* VP8New(void) {
-  VP8Decoder* const dec = (VP8Decoder*)calloc(1, sizeof(*dec));
+  VP8Decoder* const dec = (VP8Decoder*)WebPSafeCalloc(1ULL, sizeof(*dec));
   if (dec != NULL) {
     SetOk(dec);
-    WebPWorkerInit(&dec->worker_);
+    WebPGetWorkerInterface()->Init(&dec->worker_);
     dec->ready_ = 0;
     dec->num_parts_ = 1;
   }
@@ -69,7 +69,7 @@ const char* VP8StatusMessage(VP8Decoder* const dec) {
 void VP8Delete(VP8Decoder* const dec) {
   if (dec != NULL) {
     VP8Clear(dec);
-    free(dec);
+    WebPSafeFree(dec);
   }
 }
 
@@ -121,6 +121,9 @@ int VP8GetInfo(const uint8_t* data, size_t data_size, size_t chunk_size,
     if (((bits >> 5)) >= chunk_size) {  // partition_length
       return 0;         // inconsistent size information.
     }
+    if (w == 0 || h == 0) {
+      return 0;         // We don't support both width and height to be zero.
+    }
 
     if (width) {
       *width = w;
@@ -236,20 +239,6 @@ static int ParseFilterHeader(VP8BitReader* br, VP8Decoder* const dec) {
     }
   }
   dec->filter_type_ = (hdr->level_ == 0) ? 0 : hdr->simple_ ? 1 : 2;
-  if (dec->filter_type_ > 0) {    // precompute filter levels per segment
-    if (dec->segment_hdr_.use_segment_) {
-      int s;
-      for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
-        int strength = dec->segment_hdr_.filter_strength_[s];
-        if (!dec->segment_hdr_.absolute_delta_) {
-          strength += hdr->level_;
-        }
-        dec->filter_levels_[s] = strength;
-      }
-    } else {
-      dec->filter_levels_[0] = hdr->level_;
-    }
-  }
   return !br->eof_;
 }
 
@@ -261,7 +250,6 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
   VP8PictureHeader* pic_hdr;
   VP8BitReader* br;
   VP8StatusCode status;
-  WebPHeaderStructure headers;
 
   if (dec == NULL) {
     return 0;
@@ -271,33 +259,8 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
     return VP8SetError(dec, VP8_STATUS_INVALID_PARAM,
                        "null VP8Io passed to VP8GetHeaders()");
   }
-
-  // Process Pre-VP8 chunks.
-  headers.data = io->data;
-  headers.data_size = io->data_size;
-  status = WebPParseHeaders(&headers);
-  if (status != VP8_STATUS_OK) {
-    return VP8SetError(dec, status, "Incorrect/incomplete header.");
-  }
-  if (headers.is_lossless) {
-    return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
-                       "Unexpected lossless format encountered.");
-  }
-
-  if (dec->alpha_data_ == NULL) {
-    assert(dec->alpha_data_size_ == 0);
-    // We have NOT set alpha data yet. Set it now.
-    // (This is to ensure that dec->alpha_data_ is NOT reset to NULL if
-    // WebPParseHeaders() is called more than once, as in incremental decoding
-    // case.)
-    dec->alpha_data_ = headers.alpha_data;
-    dec->alpha_data_size_ = headers.alpha_data_size;
-  }
-
-  // Process the VP8 frame header.
-  buf = headers.data + headers.offset;
-  buf_size = headers.data_size - headers.offset;
-  assert(headers.data_size >= headers.offset);  // WebPParseHeaders' guarantee
+  buf = io->data;
+  buf_size = io->data_size;
   if (buf_size < 4) {
     return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA,
                        "Truncated header.");
@@ -355,7 +318,6 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
 
     VP8ResetProba(&dec->proba_);
     ResetSegmentHeader(&dec->segment_hdr_);
-    dec->segment_ = 0;    // default for intra
   }
 
   // Check if we have all the partition #0 available, and initialize dec->br_
@@ -393,63 +355,14 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
 
   // Frame buffer marking
   if (!frm_hdr->key_frame_) {
-    // Paragraph 9.7
-#ifndef ONLY_KEYFRAME_CODE
-    dec->buffer_flags_ = VP8Get(br) << 0;   // update golden
-    dec->buffer_flags_ |= VP8Get(br) << 1;  // update alt ref
-    if (!(dec->buffer_flags_ & 1)) {
-      dec->buffer_flags_ |= VP8GetValue(br, 2) << 2;
-    }
-    if (!(dec->buffer_flags_ & 2)) {
-      dec->buffer_flags_ |= VP8GetValue(br, 2) << 4;
-    }
-    dec->buffer_flags_ |= VP8Get(br) << 6;    // sign bias golden
-    dec->buffer_flags_ |= VP8Get(br) << 7;    // sign bias alt ref
-#else
     return VP8SetError(dec, VP8_STATUS_UNSUPPORTED_FEATURE,
                        "Not a key frame.");
-#endif
-  } else {
-    dec->buffer_flags_ = 0x003 | 0x100;
   }
 
-  // Paragraph 9.8
-#ifndef ONLY_KEYFRAME_CODE
-  dec->update_proba_ = VP8Get(br);
-  if (!dec->update_proba_) {    // save for later restore
-    dec->proba_saved_ = dec->proba_;
-  }
-  dec->buffer_flags_ &= 1 << 8;
-  dec->buffer_flags_ |=
-      (frm_hdr->key_frame_ || VP8Get(br)) << 8;    // refresh last frame
-#else
-  VP8Get(br);   // just ignore the value of update_proba_
-#endif
+  VP8Get(br);   // ignore the value of update_proba_
 
   VP8ParseProba(br, dec);
 
-#ifdef WEBP_EXPERIMENTAL_FEATURES
-  // Extensions
-  if (dec->pic_hdr_.colorspace_) {
-    const size_t kTrailerSize = 8;
-    const uint8_t kTrailerMarker = 0x01;
-    const uint8_t* ext_buf = buf - kTrailerSize;
-    size_t size;
-
-    if (frm_hdr->partition_length_ < kTrailerSize ||
-        ext_buf[kTrailerSize - 1] != kTrailerMarker) {
-      return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
-                         "RIFF: Inconsistent extra information.");
-    }
-
-    // Layer
-    size = (ext_buf[0] << 0) | (ext_buf[1] << 8) | (ext_buf[2] << 16);
-    dec->layer_data_size_ = size;
-    dec->layer_data_ = NULL;  // will be set later
-    dec->layer_colorspace_ = ext_buf[3];
-  }
-#endif
-
   // sanitized state
   dec->ready_ = 1;
   return 1;
@@ -458,7 +371,7 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
 //------------------------------------------------------------------------------
 // Residual decoding (Paragraph 13.2 / 13.3)
 
-static const uint8_t kBands[16 + 1] = {
+static const int kBands[16 + 1] = {
   0, 1, 2, 3, 6, 4, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7,
   0  // extra entry as sentinel
 };
@@ -473,254 +386,227 @@ static const uint8_t kZigzag[16] = {
   0, 1, 4, 8,  5, 2, 3, 6,  9, 12, 13, 10,  7, 11, 14, 15
 };
 
-typedef const uint8_t (*ProbaArray)[NUM_CTX][NUM_PROBAS];  // for const-casting
-
-// Returns the position of the last non-zero coeff plus one
-// (and 0 if there's no coeff at all)
-static int GetCoeffs(VP8BitReader* const br, ProbaArray prob,
-                     int ctx, const quant_t dq, int n, int16_t* out) {
-  // n is either 0 or 1 here. kBands[n] is not necessary for extracting '*p'.
-  const uint8_t* p = prob[n][ctx];
-  if (!VP8GetBit(br, p[0])) {   // first EOB is more a 'CBP' bit.
-    return 0;
-  }
-  while (1) {
-    ++n;
-    if (!VP8GetBit(br, p[1])) {
-      p = prob[kBands[n]][0];
-    } else {  // non zero coeff
-      int v, j;
-      if (!VP8GetBit(br, p[2])) {
-        p = prob[kBands[n]][1];
-        v = 1;
+// See section 13-2: http://tools.ietf.org/html/rfc6386#section-13.2
+static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) {
+  int v;
+  if (!VP8GetBit(br, p[3])) {
+    if (!VP8GetBit(br, p[4])) {
+      v = 2;
+    } else {
+      v = 3 + VP8GetBit(br, p[5]);
+    }
+  } else {
+    if (!VP8GetBit(br, p[6])) {
+      if (!VP8GetBit(br, p[7])) {
+        v = 5 + VP8GetBit(br, 159);
       } else {
-        if (!VP8GetBit(br, p[3])) {
-          if (!VP8GetBit(br, p[4])) {
-            v = 2;
-          } else {
-            v = 3 + VP8GetBit(br, p[5]);
-          }
-        } else {
-          if (!VP8GetBit(br, p[6])) {
-            if (!VP8GetBit(br, p[7])) {
-              v = 5 + VP8GetBit(br, 159);
-            } else {
-              v = 7 + 2 * VP8GetBit(br, 165);
-              v += VP8GetBit(br, 145);
-            }
-          } else {
-            const uint8_t* tab;
-            const int bit1 = VP8GetBit(br, p[8]);
-            const int bit0 = VP8GetBit(br, p[9 + bit1]);
-            const int cat = 2 * bit1 + bit0;
-            v = 0;
-            for (tab = kCat3456[cat]; *tab; ++tab) {
-              v += v + VP8GetBit(br, *tab);
-            }
-            v += 3 + (8 << cat);
-          }
-        }
-        p = prob[kBands[n]][2];
+        v = 7 + 2 * VP8GetBit(br, 165);
+        v += VP8GetBit(br, 145);
       }
-      j = kZigzag[n - 1];
-      out[j] = VP8GetSigned(br, v) * dq[j > 0];
-      if (n == 16 || !VP8GetBit(br, p[0])) {   // EOB
-        return n;
+    } else {
+      const uint8_t* tab;
+      const int bit1 = VP8GetBit(br, p[8]);
+      const int bit0 = VP8GetBit(br, p[9 + bit1]);
+      const int cat = 2 * bit1 + bit0;
+      v = 0;
+      for (tab = kCat3456[cat]; *tab; ++tab) {
+        v += v + VP8GetBit(br, *tab);
       }
-    }
-    if (n == 16) {
-      return 16;
+      v += 3 + (8 << cat);
     }
   }
+  return v;
 }
 
-// Alias-safe way of converting 4bytes to 32bits.
-typedef union {
-  uint8_t  i8[4];
-  uint32_t i32;
-} PackedNz;
+// Returns the position of the last non-zero coeff plus one
+static int GetCoeffs(VP8BitReader* const br, const VP8BandProbas* const prob,
+                     int ctx, const quant_t dq, int n, int16_t* out) {
+  // n is either 0 or 1 here. kBands[n] is not necessary for extracting '*p'.
+  const uint8_t* p = prob[n].probas_[ctx];
+  for (; n < 16; ++n) {
+    if (!VP8GetBit(br, p[0])) {
+      return n;  // previous coeff was last non-zero coeff
+    }
+    while (!VP8GetBit(br, p[1])) {       // sequence of zero coeffs
+      p = prob[kBands[++n]].probas_[0];
+      if (n == 16) return 16;
+    }
+    {        // non zero coeff
+      const VP8ProbaArray* const p_ctx = &prob[kBands[n + 1]].probas_[0];
+      int v;
+      if (!VP8GetBit(br, p[2])) {
+        v = 1;
+        p = p_ctx[1];
+      } else {
+        v = GetLargeValue(br, p);
+        p = p_ctx[2];
+      }
+      out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
+    }
+  }
+  return 16;
+}
 
-// Table to unpack four bits into four bytes
-static const PackedNz kUnpackTab[16] = {
-  {{0, 0, 0, 0}},  {{1, 0, 0, 0}},  {{0, 1, 0, 0}},  {{1, 1, 0, 0}},
-  {{0, 0, 1, 0}},  {{1, 0, 1, 0}},  {{0, 1, 1, 0}},  {{1, 1, 1, 0}},
-  {{0, 0, 0, 1}},  {{1, 0, 0, 1}},  {{0, 1, 0, 1}},  {{1, 1, 0, 1}},
-  {{0, 0, 1, 1}},  {{1, 0, 1, 1}},  {{0, 1, 1, 1}},  {{1, 1, 1, 1}} };
+static WEBP_INLINE uint32_t NzCodeBits(uint32_t nz_coeffs, int nz, int dc_nz) {
+  nz_coeffs <<= 2;
+  nz_coeffs |= (nz > 3) ? 3 : (nz > 1) ? 2 : dc_nz;
+  return nz_coeffs;
+}
 
-// Macro to pack four LSB of four bytes into four bits.
-#if defined(__PPC__) || defined(_M_PPC) || defined(_ARCH_PPC) || \
-    defined(__BIG_ENDIAN__)
-#define PACK_CST 0x08040201U
-#else
-#define PACK_CST 0x01020408U
-#endif
-#define PACK(X, S) ((((X).i32 * PACK_CST) & 0xff000000) >> (S))
-
-static void ParseResiduals(VP8Decoder* const dec,
-                           VP8MB* const mb, VP8BitReader* const token_br) {
-  int out_t_nz, out_l_nz, first;
-  ProbaArray ac_prob;
-  const VP8QuantMatrix* q = &dec->dqm_[dec->segment_];
-  int16_t* dst = dec->coeffs_;
+static int ParseResiduals(VP8Decoder* const dec,
+                          VP8MB* const mb, VP8BitReader* const token_br) {
+  VP8BandProbas (* const bands)[NUM_BANDS] = dec->proba_.bands_;
+  const VP8BandProbas* ac_proba;
+  VP8MBData* const block = dec->mb_data_ + dec->mb_x_;
+  const VP8QuantMatrix* const q = &dec->dqm_[block->segment_];
+  int16_t* dst = block->coeffs_;
   VP8MB* const left_mb = dec->mb_info_ - 1;
-  PackedNz nz_ac, nz_dc;
-  PackedNz tnz, lnz;
-  uint32_t non_zero_ac = 0;
-  uint32_t non_zero_dc = 0;
+  uint8_t tnz, lnz;
+  uint32_t non_zero_y = 0;
+  uint32_t non_zero_uv = 0;
   int x, y, ch;
+  uint32_t out_t_nz, out_l_nz;
+  int first;
 
-  nz_dc.i32 = nz_ac.i32 = 0;
   memset(dst, 0, 384 * sizeof(*dst));
-  if (!dec->is_i4x4_) {    // parse DC
+  if (!block->is_i4x4_) {    // parse DC
     int16_t dc[16] = { 0 };
-    const int ctx = mb->dc_nz_ + left_mb->dc_nz_;
-    mb->dc_nz_ = left_mb->dc_nz_ =
-        (GetCoeffs(token_br, (ProbaArray)dec->proba_.coeffs_[1],
-                   ctx, q->y2_mat_, 0, dc) > 0);
+    const int ctx = mb->nz_dc_ + left_mb->nz_dc_;
+    const int nz = GetCoeffs(token_br, bands[1], ctx, q->y2_mat_, 0, dc);
+    mb->nz_dc_ = left_mb->nz_dc_ = (nz > 0);
+    if (nz > 1) {   // more than just the DC -> perform the full transform
+      VP8TransformWHT(dc, dst);
+    } else {        // only DC is non-zero -> inlined simplified transform
+      int i;
+      const int dc0 = (dc[0] + 3) >> 3;
+      for (i = 0; i < 16 * 16; i += 16) dst[i] = dc0;
+    }
     first = 1;
-    ac_prob = (ProbaArray)dec->proba_.coeffs_[0];
-    VP8TransformWHT(dc, dst);
+    ac_proba = bands[0];
   } else {
     first = 0;
-    ac_prob = (ProbaArray)dec->proba_.coeffs_[3];
+    ac_proba = bands[3];
   }
 
-  tnz = kUnpackTab[mb->nz_ & 0xf];
-  lnz = kUnpackTab[left_mb->nz_ & 0xf];
+  tnz = mb->nz_ & 0x0f;
+  lnz = left_mb->nz_ & 0x0f;
   for (y = 0; y < 4; ++y) {
-    int l = lnz.i8[y];
+    int l = lnz & 1;
+    uint32_t nz_coeffs = 0;
     for (x = 0; x < 4; ++x) {
-      const int ctx = l + tnz.i8[x];
-      const int nz = GetCoeffs(token_br, ac_prob, ctx,
-                               q->y1_mat_, first, dst);
-      tnz.i8[x] = l = (nz > 0);
-      nz_dc.i8[x] = (dst[0] != 0);
-      nz_ac.i8[x] = (nz > 1);
+      const int ctx = l + (tnz & 1);
+      const int nz = GetCoeffs(token_br, ac_proba, ctx, q->y1_mat_, first, dst);
+      l = (nz > first);
+      tnz = (tnz >> 1) | (l << 7);
+      nz_coeffs = NzCodeBits(nz_coeffs, nz, dst[0] != 0);
       dst += 16;
     }
-    lnz.i8[y] = l;
-    non_zero_dc |= PACK(nz_dc, 24 - y * 4);
-    non_zero_ac |= PACK(nz_ac, 24 - y * 4);
+    tnz >>= 4;
+    lnz = (lnz >> 1) | (l << 7);
+    non_zero_y = (non_zero_y << 8) | nz_coeffs;
   }
-  out_t_nz = PACK(tnz, 24);
-  out_l_nz = PACK(lnz, 24);
+  out_t_nz = tnz;
+  out_l_nz = lnz >> 4;
 
-  tnz = kUnpackTab[mb->nz_ >> 4];
-  lnz = kUnpackTab[left_mb->nz_ >> 4];
   for (ch = 0; ch < 4; ch += 2) {
+    uint32_t nz_coeffs = 0;
+    tnz = mb->nz_ >> (4 + ch);
+    lnz = left_mb->nz_ >> (4 + ch);
     for (y = 0; y < 2; ++y) {
-      int l = lnz.i8[ch + y];
+      int l = lnz & 1;
       for (x = 0; x < 2; ++x) {
-        const int ctx = l + tnz.i8[ch + x];
-        const int nz =
-            GetCoeffs(token_br, (ProbaArray)dec->proba_.coeffs_[2],
-                      ctx, q->uv_mat_, 0, dst);
-        tnz.i8[ch + x] = l = (nz > 0);
-        nz_dc.i8[y * 2 + x] = (dst[0] != 0);
-        nz_ac.i8[y * 2 + x] = (nz > 1);
+        const int ctx = l + (tnz & 1);
+        const int nz = GetCoeffs(token_br, bands[2], ctx, q->uv_mat_, 0, dst);
+        l = (nz > 0);
+        tnz = (tnz >> 1) | (l << 3);
+        nz_coeffs = NzCodeBits(nz_coeffs, nz, dst[0] != 0);
         dst += 16;
       }
-      lnz.i8[ch + y] = l;
+      tnz >>= 2;
+      lnz = (lnz >> 1) | (l << 5);
     }
-    non_zero_dc |= PACK(nz_dc, 8 - ch * 2);
-    non_zero_ac |= PACK(nz_ac, 8 - ch * 2);
+    // Note: we don't really need the per-4x4 details for U/V blocks.
+    non_zero_uv |= nz_coeffs << (4 * ch);
+    out_t_nz |= (tnz << 4) << ch;
+    out_l_nz |= (lnz & 0xf0) << ch;
   }
-  out_t_nz |= PACK(tnz, 20);
-  out_l_nz |= PACK(lnz, 20);
   mb->nz_ = out_t_nz;
   left_mb->nz_ = out_l_nz;
 
-  dec->non_zero_ac_ = non_zero_ac;
-  dec->non_zero_ = non_zero_ac | non_zero_dc;
-  mb->skip_ = !dec->non_zero_;
+  block->non_zero_y_ = non_zero_y;
+  block->non_zero_uv_ = non_zero_uv;
+
+  // We look at the mode-code of each block and check if some blocks have less
+  // than three non-zero coeffs (code < 2). This is to avoid dithering flat and
+  // empty blocks.
+  block->dither_ = (non_zero_uv & 0xaaaa) ? 0 : q->dither_;
+
+  return !(non_zero_y | non_zero_uv);  // will be used for further optimization
 }
-#undef PACK
 
 //------------------------------------------------------------------------------
 // Main loop
 
 int VP8DecodeMB(VP8Decoder* const dec, VP8BitReader* const token_br) {
-  VP8BitReader* const br = &dec->br_;
   VP8MB* const left = dec->mb_info_ - 1;
-  VP8MB* const info = dec->mb_info_ + dec->mb_x_;
+  VP8MB* const mb = dec->mb_info_ + dec->mb_x_;
+  VP8MBData* const block = dec->mb_data_ + dec->mb_x_;
+  int skip = dec->use_skip_proba_ ? block->skip_ : 0;
 
-  // Note: we don't save segment map (yet), as we don't expect
-  // to decode more than 1 keyframe.
-  if (dec->segment_hdr_.update_map_) {
-    // Hardcoded tree parsing
-    dec->segment_ = !VP8GetBit(br, dec->proba_.segments_[0]) ?
-        VP8GetBit(br, dec->proba_.segments_[1]) :
-        2 + VP8GetBit(br, dec->proba_.segments_[2]);
-  }
-  info->skip_ = dec->use_skip_proba_ ? VP8GetBit(br, dec->skip_p_) : 0;
-
-  VP8ParseIntraMode(br, dec);
-  if (br->eof_) {
-    return 0;
-  }
-
-  if (!info->skip_) {
-    ParseResiduals(dec, info, token_br);
+  if (!skip) {
+    skip = ParseResiduals(dec, mb, token_br);
   } else {
-    left->nz_ = info->nz_ = 0;
-    if (!dec->is_i4x4_) {
-      left->dc_nz_ = info->dc_nz_ = 0;
+    left->nz_ = mb->nz_ = 0;
+    if (!block->is_i4x4_) {
+      left->nz_dc_ = mb->nz_dc_ = 0;
     }
-    dec->non_zero_ = 0;
-    dec->non_zero_ac_ = 0;
+    block->non_zero_y_ = 0;
+    block->non_zero_uv_ = 0;
   }
 
-  return (!token_br->eof_);
+  if (dec->filter_type_ > 0) {  // store filter info
+    VP8FInfo* const finfo = dec->f_info_ + dec->mb_x_;
+    *finfo = dec->fstrengths_[block->segment_][block->is_i4x4_];
+    finfo->f_inner_ |= !skip;
+  }
+
+  return !token_br->eof_;
 }
 
 void VP8InitScanline(VP8Decoder* const dec) {
   VP8MB* const left = dec->mb_info_ - 1;
   left->nz_ = 0;
-  left->dc_nz_ = 0;
+  left->nz_dc_ = 0;
   memset(dec->intra_l_, B_DC_PRED, sizeof(dec->intra_l_));
-  dec->filter_row_ =
-    (dec->filter_type_ > 0) &&
-    (dec->mb_y_ >= dec->tl_mb_y_) && (dec->mb_y_ <= dec->br_mb_y_);
+  dec->mb_x_ = 0;
 }
 
 static int ParseFrame(VP8Decoder* const dec, VP8Io* io) {
   for (dec->mb_y_ = 0; dec->mb_y_ < dec->br_mb_y_; ++dec->mb_y_) {
+    // Parse bitstream for this row.
     VP8BitReader* const token_br =
         &dec->parts_[dec->mb_y_ & (dec->num_parts_ - 1)];
-    VP8InitScanline(dec);
-    for (dec->mb_x_ = 0; dec->mb_x_ < dec->mb_w_;  dec->mb_x_++) {
+    if (!VP8ParseIntraModeRow(&dec->br_, dec)) {
+      return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA,
+                         "Premature end-of-partition0 encountered.");
+    }
+    for (; dec->mb_x_ < dec->mb_w_; ++dec->mb_x_) {
       if (!VP8DecodeMB(dec, token_br)) {
         return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA,
                            "Premature end-of-file encountered.");
       }
-      VP8ReconstructBlock(dec);
-
-      // Store data and save block's filtering params
-      VP8StoreBlock(dec);
     }
+    VP8InitScanline(dec);   // Prepare for next scanline
+
+    // Reconstruct, filter and emit the row.
     if (!VP8ProcessRow(dec, io)) {
       return VP8SetError(dec, VP8_STATUS_USER_ABORT, "Output aborted.");
     }
   }
-  if (dec->use_threads_ && !WebPWorkerSync(&dec->worker_)) {
-    return 0;
+  if (dec->mt_method_ > 0) {
+    if (!WebPGetWorkerInterface()->Sync(&dec->worker_)) return 0;
   }
 
-  // Finish
-#ifndef ONLY_KEYFRAME_CODE
-  if (!dec->update_proba_) {
-    dec->proba_ = dec->proba_saved_;
-  }
-#endif
-
-#ifdef WEBP_EXPERIMENTAL_FEATURES
-  if (dec->layer_data_size_ > 0) {
-    if (!VP8DecodeLayer(dec)) {
-      return 0;
-    }
-  }
-#endif
-
   return 1;
 }
 
@@ -768,12 +654,10 @@ void VP8Clear(VP8Decoder* const dec) {
   if (dec == NULL) {
     return;
   }
-  if (dec->use_threads_) {
-    WebPWorkerEnd(&dec->worker_);
-  }
-  if (dec->mem_) {
-    free(dec->mem_);
-  }
+  WebPGetWorkerInterface()->End(&dec->worker_);
+  ALPHDelete(dec->alph_dec_);
+  dec->alph_dec_ = NULL;
+  WebPSafeFree(dec->mem_);
   dec->mem_ = NULL;
   dec->mem_size_ = 0;
   memset(&dec->br_, 0, sizeof(dec->br_));
@@ -782,6 +666,3 @@ void VP8Clear(VP8Decoder* const dec) {
 
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/vp8i.h

@@ -1,8 +1,10 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // VP8 decoder: internal header.
@@ -15,10 +17,11 @@
 #include <string.h>     // for memcpy()
 #include "./vp8li.h"
 #include "../utils/bit_reader.h"
+#include "../utils/random.h"
 #include "../utils/thread.h"
 #include "../dsp/dsp.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 extern "C" {
 #endif
 
@@ -27,10 +30,8 @@ extern "C" {
 
 // version numbers
 #define DEC_MAJ_VERSION 0
-#define DEC_MIN_VERSION 2
-#define DEC_REV_VERSION 1
-
-#define ONLY_KEYFRAME_CODE      // to remove any code related to P-Frames
+#define DEC_MIN_VERSION 4
+#define DEC_REV_VERSION 2
 
 // intra prediction modes
 enum { B_DC_PRED = 0,   // 4x4 modes
@@ -98,6 +99,9 @@ enum { MB_FEATURE_TREE_PROBS = 3,
 #define U_OFF    (Y_OFF + BPS * 16 + BPS)
 #define V_OFF    (U_OFF + 16)
 
+// minimal width under which lossy multi-threading is always disabled
+#define MIN_WIDTH_FOR_THREADS 512
+
 //------------------------------------------------------------------------------
 // Headers
 
@@ -126,15 +130,19 @@ typedef struct {
   int8_t filter_strength_[NUM_MB_SEGMENTS];  // filter strength for segments
 } VP8SegmentHeader;
 
+
+// probas associated to one of the contexts
+typedef uint8_t VP8ProbaArray[NUM_PROBAS];
+
+typedef struct {   // all the probas associated to one band
+  VP8ProbaArray probas_[NUM_CTX];
+} VP8BandProbas;
+
 // Struct collecting all frame-persistent probabilities.
 typedef struct {
   uint8_t segments_[MB_FEATURE_TREE_PROBS];
   // Type: 0:Intra16-AC  1:Intra16-DC   2:Chroma   3:Intra4
-  uint8_t coeffs_[NUM_TYPES][NUM_BANDS][NUM_CTX][NUM_PROBAS];
-#ifndef ONLY_KEYFRAME_CODE
-  uint8_t ymode_[4], uvmode_[3];
-  uint8_t mv_[2][NUM_MV_PROBAS];
-#endif
+  VP8BandProbas bands_[NUM_TYPES][NUM_BANDS];
 } VP8Proba;
 
 // Filter parameters
@@ -151,32 +159,61 @@ typedef struct {
 // Informations about the macroblocks.
 
 typedef struct {  // filter specs
-  unsigned int f_level_:6;      // filter strength: 0..63
-  unsigned int f_ilevel_:6;     // inner limit: 1..63
-  unsigned int f_inner_:1;      // do inner filtering?
+  uint8_t f_limit_;      // filter limit in [3..189], or 0 if no filtering
+  uint8_t f_ilevel_;     // inner limit in [1..63]
+  uint8_t f_inner_;      // do inner filtering?
+  uint8_t hev_thresh_;   // high edge variance threshold in [0..2]
 } VP8FInfo;
 
-typedef struct {  // used for syntax-parsing
-  unsigned int nz_;          // non-zero AC/DC coeffs
-  unsigned int dc_nz_:1;     // non-zero DC coeffs
-  unsigned int skip_:1;      // block type
+typedef struct {  // Top/Left Contexts used for syntax-parsing
+  uint8_t nz_;        // non-zero AC/DC coeffs (4bit for luma + 4bit for chroma)
+  uint8_t nz_dc_;     // non-zero DC coeff (1bit)
 } VP8MB;
 
 // Dequantization matrices
 typedef int quant_t[2];      // [DC / AC].  Can be 'uint16_t[2]' too (~slower).
 typedef struct {
   quant_t y1_mat_, y2_mat_, uv_mat_;
+
+  int uv_quant_;   // U/V quantizer value
+  int dither_;     // dithering amplitude (0 = off, max=255)
 } VP8QuantMatrix;
 
+// Data needed to reconstruct a macroblock
+typedef struct {
+  int16_t coeffs_[384];   // 384 coeffs = (16+4+4) * 4*4
+  uint8_t is_i4x4_;       // true if intra4x4
+  uint8_t imodes_[16];    // one 16x16 mode (#0) or sixteen 4x4 modes
+  uint8_t uvmode_;        // chroma prediction mode
+  // bit-wise info about the content of each sub-4x4 blocks (in decoding order).
+  // Each of the 4x4 blocks for y/u/v is associated with a 2b code according to:
+  //   code=0 -> no coefficient
+  //   code=1 -> only DC
+  //   code=2 -> first three coefficients are non-zero
+  //   code=3 -> more than three coefficients are non-zero
+  // This allows to call specialized transform functions.
+  uint32_t non_zero_y_;
+  uint32_t non_zero_uv_;
+  uint8_t dither_;      // local dithering strength (deduced from non_zero_*)
+  uint8_t skip_;
+  uint8_t segment_;
+} VP8MBData;
+
 // Persistent information needed by the parallel processing
 typedef struct {
-  int id_;            // cache row to process (in [0..2])
-  int mb_y_;          // macroblock position of the row
-  int filter_row_;    // true if row-filtering is needed
-  VP8FInfo* f_info_;  // filter strengths
-  VP8Io io_;          // copy of the VP8Io to pass to put()
+  int id_;              // cache row to process (in [0..2])
+  int mb_y_;            // macroblock position of the row
+  int filter_row_;      // true if row-filtering is needed
+  VP8FInfo* f_info_;    // filter strengths (swapped with dec->f_info_)
+  VP8MBData* mb_data_;  // reconstruction data (swapped with dec->mb_data_)
+  VP8Io io_;            // copy of the VP8Io to pass to put()
 } VP8ThreadContext;
 
+// Saved top samples, per macroblock. Fits into a cache-line.
+typedef struct {
+  uint8_t y[16], u[8], v[8];
+} VP8TopSamples;
+
 //------------------------------------------------------------------------------
 // VP8Decoder: the main opaque structure handed over to user
 
@@ -196,7 +233,8 @@ struct VP8Decoder {
 
   // Worker
   WebPWorker worker_;
-  int use_threads_;    // use multi-thread
+  int mt_method_;      // multi-thread method: 0=off, 1=[parse+recon][filter]
+                       // 2=[parse][recon+filter]
   int cache_id_;       // current cache row
   int num_caches_;     // number of cached rows of 16 pixels (1, 2 or 3)
   VP8ThreadContext thread_ctx_;  // Thread context
@@ -213,12 +251,9 @@ struct VP8Decoder {
   // per-partition boolean decoders.
   VP8BitReader parts_[MAX_NUM_PARTITIONS];
 
-  // buffer refresh flags
-  //   bit 0: refresh Gold, bit 1: refresh Alt
-  //   bit 2-3: copy to Gold, bit 4-5: copy to Alt
-  //   bit 6: Gold sign bias, bit 7: Alt sign bias
-  //   bit 8: refresh last frame
-  uint32_t buffer_flags_;
+  // Dithering strength, deduced from decoding options
+  int dither_;                // whether to use dithering or not
+  VP8Random dithering_rg_;    // random generator for dithering
 
   // dequantization (one set of DC/AC dequant factor per segment)
   VP8QuantMatrix dqm_[NUM_MB_SEGMENTS];
@@ -227,24 +262,18 @@ struct VP8Decoder {
   VP8Proba proba_;
   int use_skip_proba_;
   uint8_t skip_p_;
-#ifndef ONLY_KEYFRAME_CODE
-  uint8_t intra_p_, last_p_, golden_p_;
-  VP8Proba proba_saved_;
-  int update_proba_;
-#endif
 
   // Boundary data cache and persistent buffers.
-  uint8_t* intra_t_;     // top intra modes values: 4 * mb_w_
-  uint8_t  intra_l_[4];  // left intra modes values
-  uint8_t* y_t_;         // top luma samples: 16 * mb_w_
-  uint8_t* u_t_, *v_t_;  // top u/v samples: 8 * mb_w_ each
+  uint8_t* intra_t_;      // top intra modes values: 4 * mb_w_
+  uint8_t  intra_l_[4];   // left intra modes values
 
-  VP8MB* mb_info_;       // contextual macroblock info (mb_w_ + 1)
-  VP8FInfo* f_info_;     // filter strength info
-  uint8_t* yuv_b_;       // main block for Y/U/V (size = YUV_SIZE)
-  int16_t* coeffs_;      // 384 coeffs = (16+8+8) * 4*4
+  VP8TopSamples* yuv_t_;  // top y/u/v samples
 
-  uint8_t* cache_y_;     // macroblock row for storing unfiltered samples
+  VP8MB* mb_info_;        // contextual macroblock info (mb_w_ + 1)
+  VP8FInfo* f_info_;      // filter strength info
+  uint8_t* yuv_b_;        // main block for Y/U/V (size = YUV_SIZE)
+
+  uint8_t* cache_y_;      // macroblock row for storing unfiltered samples
   uint8_t* cache_u_;
   uint8_t* cache_v_;
   int cache_y_stride_;
@@ -256,31 +285,19 @@ struct VP8Decoder {
 
   // Per macroblock non-persistent infos.
   int mb_x_, mb_y_;       // current position, in macroblock units
-  uint8_t is_i4x4_;       // true if intra4x4
-  uint8_t imodes_[16];    // one 16x16 mode (#0) or sixteen 4x4 modes
-  uint8_t uvmode_;        // chroma prediction mode
-  uint8_t segment_;       // block's segment
-
-  // bit-wise info about the content of each sub-4x4 blocks: there are 16 bits
-  // for luma (bits #0->#15), then 4 bits for chroma-u (#16->#19) and 4 bits for
-  // chroma-v (#20->#23), each corresponding to one 4x4 block in decoding order.
-  // If the bit is set, the 4x4 block contains some non-zero coefficients.
-  uint32_t non_zero_;
-  uint32_t non_zero_ac_;
+  VP8MBData* mb_data_;    // parsed reconstruction data
 
   // Filtering side-info
-  int filter_type_;                         // 0=off, 1=simple, 2=complex
-  int filter_row_;                          // per-row flag
-  uint8_t filter_levels_[NUM_MB_SEGMENTS];  // precalculated per-segment
+  int filter_type_;                          // 0=off, 1=simple, 2=complex
+  VP8FInfo fstrengths_[NUM_MB_SEGMENTS][2];  // precalculated per-segment/type
 
-  // extensions
-  const uint8_t* alpha_data_;   // compressed alpha data (if present)
+  // Alpha
+  struct ALPHDecoder* alph_dec_;  // alpha-plane decoder object
+  const uint8_t* alpha_data_;     // compressed alpha data (if present)
   size_t alpha_data_size_;
-  uint8_t* alpha_plane_;        // output. Persistent, contains the whole data.
-
-  int layer_colorspace_;
-  const uint8_t* layer_data_;   // compressed layer data (if present)
-  size_t layer_data_size_;
+  int is_alpha_decoded_;  // true if alpha_data_ is decoded in alpha_plane_
+  uint8_t* alpha_plane_;  // output. Persistent, contains the whole data.
+  int alpha_dithering_;   // derived from decoding options (0=off, 100=full).
 };
 
 //------------------------------------------------------------------------------
@@ -293,15 +310,14 @@ int VP8SetError(VP8Decoder* const dec,
 // in tree.c
 void VP8ResetProba(VP8Proba* const proba);
 void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec);
-void VP8ParseIntraMode(VP8BitReader* const br,  VP8Decoder* const dec);
+// parses one row of intra mode data in partition 0, returns !eof
+int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec);
 
 // in quant.c
 void VP8ParseQuant(VP8Decoder* const dec);
 
 // in frame.c
 int VP8InitFrame(VP8Decoder* const dec, VP8Io* io);
-// Predict a block and add residual
-void VP8ReconstructBlock(VP8Decoder* const dec);
 // Call io->setup() and finish setting up scan parameters.
 // After this call returns, one must always call VP8ExitCritical() with the
 // same parameters. Both functions should be used in pair. Returns VP8_STATUS_OK
@@ -310,10 +326,16 @@ VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io);
 // Must always be called in pair with VP8EnterCritical().
 // Returns false in case of error.
 int VP8ExitCritical(VP8Decoder* const dec, VP8Io* const io);
-// Process the last decoded row (filtering + output)
+// Return the multi-threading method to use (0=off), depending
+// on options and bitstream size. Only for lossy decoding.
+int VP8GetThreadMethod(const WebPDecoderOptions* const options,
+                       const WebPHeaderStructure* const headers,
+                       int width, int height);
+// Initialize dithering post-process if needed.
+void VP8InitDithering(const WebPDecoderOptions* const options,
+                      VP8Decoder* const dec);
+// Process the last decoded row (filtering + output).
 int VP8ProcessRow(VP8Decoder* const dec, VP8Io* const io);
-// Store a block, along with filtering params
-void VP8StoreBlock(VP8Decoder* const dec);
 // To be called at the start of a new scanline, to initialize predictors.
 void VP8InitScanline(VP8Decoder* const dec);
 // Decode one macroblock. Returns false if there is not enough data.
@@ -323,12 +345,9 @@ int VP8DecodeMB(VP8Decoder* const dec, VP8BitReader* const token_br);
 const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec,
                                       int row, int num_rows);
 
-// in layer.c
-int VP8DecodeLayer(VP8Decoder* const dec);
-
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 

src/dec/vp8li.h

@@ -1,8 +1,10 @@
 // Copyright 2012 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Lossless decoder: internal header.
@@ -18,9 +20,8 @@
 #include "../utils/bit_reader.h"
 #include "../utils/color_cache.h"
 #include "../utils/huffman.h"
-#include "../webp/format_constants.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 extern "C" {
 #endif
 
@@ -39,10 +40,6 @@ struct VP8LTransform {
   uint32_t              *data_;   // transform data.
 };
 
-typedef struct {
-  HuffmanTree htrees_[HUFFMAN_CODES_PER_META_CODE];
-} HTreeGroup;
-
 typedef struct {
   int             color_cache_size_;
   VP8LColorCache  color_cache_;
@@ -55,7 +52,8 @@ typedef struct {
   HTreeGroup     *htree_groups_;
 } VP8LMetadata;
 
-typedef struct {
+typedef struct VP8LDecoder VP8LDecoder;
+struct VP8LDecoder {
   VP8StatusCode    status_;
   VP8LDecodeState  action_;
   VP8LDecodeState  state_;
@@ -63,7 +61,8 @@ typedef struct {
 
   const WebPDecBuffer *output_;    // shortcut to io->opaque->output
 
-  uint32_t        *argb_;          // Internal data: always in BGRA color mode.
+  uint32_t        *pixels_;        // Internal data: either uint8_t* for alpha
+                                   // or uint32_t* for BGRA.
   uint32_t        *argb_cache_;    // Scratch buffer for temporary BGRA storage.
 
   VP8LBitReader    br_;
@@ -71,6 +70,9 @@ typedef struct {
   int              width_;
   int              height_;
   int              last_row_;      // last input row decoded so far.
+  int              last_pixel_;    // last pixel decoded so far. However, it may
+                                   // not be transformed, scaled and
+                                   // color-converted yet.
   int              last_out_row_;  // last row output so far.
 
   VP8LMetadata     hdr_;
@@ -82,18 +84,27 @@ typedef struct {
 
   uint8_t         *rescaler_memory;  // Working memory for rescaling work.
   WebPRescaler    *rescaler;         // Common rescaler for all channels.
-} VP8LDecoder;
+};
 
 //------------------------------------------------------------------------------
 // internal functions. Not public.
 
+struct ALPHDecoder;  // Defined in dec/alphai.h.
+
 // in vp8l.c
 
-// Decodes a raw image stream (without header) and store the alpha data
-// into *output, which must be of size width x height. Returns false in case
-// of error.
-int VP8LDecodeAlphaImageStream(int width, int height, const uint8_t* const data,
-                               size_t data_size, uint8_t* const output);
+// Decodes image header for alpha data stored using lossless compression.
+// Returns false in case of error.
+int VP8LDecodeAlphaHeader(struct ALPHDecoder* const alph_dec,
+                          const uint8_t* const data, size_t data_size,
+                          uint8_t* const output);
+
+// Decodes *at least* 'last_row' rows of alpha. If some of the initial rows are
+// already decoded in previous call(s), it will resume decoding from where it
+// was paused.
+// Returns false in case of bitstream error.
+int VP8LDecodeAlphaImageStream(struct ALPHDecoder* const alph_dec,
+                               int last_row);
 
 // Allocates and initialize a new lossless decoder instance.
 VP8LDecoder* VP8LNew(void);
@@ -114,7 +125,7 @@ void VP8LDelete(VP8LDecoder* const dec);
 
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 

src/dec/webp.c

@@ -1,8 +1,10 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Main decoding functions for WEBP images.
@@ -14,11 +16,7 @@
 #include "./vp8i.h"
 #include "./vp8li.h"
 #include "./webpi.h"
-#include "../webp/format_constants.h"
-
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
+#include "../webp/mux_types.h"  // ALPHA_FLAG
 
 //------------------------------------------------------------------------------
 // RIFF layout is:
@@ -40,8 +38,8 @@ extern "C" {
 //   20..23  VP8X flags bit-map corresponding to the chunk-types present.
 //   24..26  Width of the Canvas Image.
 //   27..29  Height of the Canvas Image.
-// There can be extra chunks after the "VP8X" chunk (ICCP, TILE, FRM, VP8,
-// META  ...)
+// There can be extra chunks after the "VP8X" chunk (ICCP, FRGM, ANMF, VP8,
+// VP8L, XMP, EXIF  ...)
 // All sizes are in little-endian order.
 // Note: chunk data size must be padded to multiple of 2 when written.
 
@@ -54,13 +52,14 @@ static WEBP_INLINE uint32_t get_le32(const uint8_t* const data) {
 }
 
 // Validates the RIFF container (if detected) and skips over it.
-// If a RIFF container is detected,
-// Returns VP8_STATUS_BITSTREAM_ERROR for invalid header, and
-//         VP8_STATUS_OK otherwise.
+// If a RIFF container is detected, returns:
+//     VP8_STATUS_BITSTREAM_ERROR for invalid header,
+//     VP8_STATUS_NOT_ENOUGH_DATA for truncated data if have_all_data is true,
+// and VP8_STATUS_OK otherwise.
 // In case there are not enough bytes (partial RIFF container), return 0 for
 // *riff_size. Else return the RIFF size extracted from the header.
 static VP8StatusCode ParseRIFF(const uint8_t** const data,
-                               size_t* const data_size,
+                               size_t* const data_size, int have_all_data,
                                size_t* const riff_size) {
   assert(data != NULL);
   assert(data_size != NULL);
@@ -79,6 +78,9 @@ static VP8StatusCode ParseRIFF(const uint8_t** const data,
       if (size > MAX_CHUNK_PAYLOAD) {
         return VP8_STATUS_BITSTREAM_ERROR;
       }
+      if (have_all_data && (size > *data_size - CHUNK_HEADER_SIZE)) {
+        return VP8_STATUS_NOT_ENOUGH_DATA;  // Truncated bitstream.
+      }
       // We have a RIFF container. Skip it.
       *riff_size = size;
       *data += RIFF_HEADER_SIZE;
@@ -192,6 +194,15 @@ static VP8StatusCode ParseOptionalChunks(const uint8_t** const data,
       return VP8_STATUS_BITSTREAM_ERROR;          // Not a valid chunk size.
     }
 
+    // Start of a (possibly incomplete) VP8/VP8L chunk implies that we have
+    // parsed all the optional chunks.
+    // Note: This check must occur before the check 'buf_size < disk_chunk_size'
+    // below to allow incomplete VP8/VP8L chunks.
+    if (!memcmp(buf, "VP8 ", TAG_SIZE) ||
+        !memcmp(buf, "VP8L", TAG_SIZE)) {
+      return VP8_STATUS_OK;
+    }
+
     if (buf_size < disk_chunk_size) {             // Insufficient data.
       return VP8_STATUS_NOT_ENOUGH_DATA;
     }
@@ -199,9 +210,6 @@ static VP8StatusCode ParseOptionalChunks(const uint8_t** const data,
     if (!memcmp(buf, "ALPH", TAG_SIZE)) {         // A valid ALPH header.
       *alpha_data = buf + CHUNK_HEADER_SIZE;
       *alpha_size = chunk_size;
-    } else if (!memcmp(buf, "VP8 ", TAG_SIZE) ||
-               !memcmp(buf, "VP8L", TAG_SIZE)) {  // A valid VP8/VP8L header.
-      return VP8_STATUS_OK;  // Found.
     }
 
     // We have a full and valid chunk; skip it.
@@ -219,9 +227,8 @@ static VP8StatusCode ParseOptionalChunks(const uint8_t** const data,
 // extracted from the VP8/VP8L chunk header.
 // The flag '*is_lossless' is set to 1 in case of VP8L chunk / raw VP8L data.
 static VP8StatusCode ParseVP8Header(const uint8_t** const data_ptr,
-                                    size_t* const data_size,
-                                    size_t riff_size,
-                                    size_t* const chunk_size,
+                                    size_t* const data_size, int have_all_data,
+                                    size_t riff_size, size_t* const chunk_size,
                                     int* const is_lossless) {
   const uint8_t* const data = *data_ptr;
   const int is_vp8 = !memcmp(data, "VP8 ", TAG_SIZE);
@@ -244,6 +251,9 @@ static VP8StatusCode ParseVP8Header(const uint8_t** const data_ptr,
     if ((riff_size >= minimal_size) && (size > riff_size - minimal_size)) {
       return VP8_STATUS_BITSTREAM_ERROR;  // Inconsistent size information.
     }
+    if (have_all_data && (size > *data_size - CHUNK_HEADER_SIZE)) {
+      return VP8_STATUS_NOT_ENOUGH_DATA;  // Truncated bitstream.
+    }
     // Skip over CHUNK_HEADER_SIZE bytes from VP8/VP8L Header.
     *chunk_size = size;
     *data_ptr += CHUNK_HEADER_SIZE;
@@ -276,9 +286,19 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
                                           int* const width,
                                           int* const height,
                                           int* const has_alpha,
+                                          int* const has_animation,
+                                          int* const format,
                                           WebPHeaderStructure* const headers) {
+  int canvas_width = 0;
+  int canvas_height = 0;
+  int image_width = 0;
+  int image_height = 0;
   int found_riff = 0;
   int found_vp8x = 0;
+  int animation_present = 0;
+  int fragments_present = 0;
+  const int have_all_data = (headers != NULL) ? headers->have_all_data : 0;
+
   VP8StatusCode status;
   WebPHeaderStructure hdrs;
 
@@ -290,7 +310,7 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
   hdrs.data_size = data_size;
 
   // Skip over RIFF header.
-  status = ParseRIFF(&data, &data_size, &hdrs.riff_size);
+  status = ParseRIFF(&data, &data_size, have_all_data, &hdrs.riff_size);
   if (status != VP8_STATUS_OK) {
     return status;   // Wrong RIFF header / insufficient data.
   }
@@ -299,22 +319,35 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
   // Skip over VP8X.
   {
     uint32_t flags = 0;
-    status = ParseVP8X(&data, &data_size, &found_vp8x, width, height, &flags);
+    status = ParseVP8X(&data, &data_size, &found_vp8x,
+                       &canvas_width, &canvas_height, &flags);
     if (status != VP8_STATUS_OK) {
       return status;  // Wrong VP8X / insufficient data.
     }
+    animation_present = !!(flags & ANIMATION_FLAG);
+    fragments_present = !!(flags & FRAGMENTS_FLAG);
     if (!found_riff && found_vp8x) {
       // Note: This restriction may be removed in the future, if it becomes
       // necessary to send VP8X chunk to the decoder.
       return VP8_STATUS_BITSTREAM_ERROR;
     }
-    if (has_alpha != NULL) *has_alpha = !!(flags & ALPHA_FLAG_BIT);
-    if (found_vp8x && headers == NULL) {
-      return VP8_STATUS_OK;  // Return features from VP8X header.
+    if (has_alpha != NULL) *has_alpha = !!(flags & ALPHA_FLAG);
+    if (has_animation != NULL) *has_animation = animation_present;
+    if (format != NULL) *format = 0;   // default = undefined
+
+    image_width = canvas_width;
+    image_height = canvas_height;
+    if (found_vp8x && (animation_present || fragments_present) &&
+        headers == NULL) {
+      status = VP8_STATUS_OK;
+      goto ReturnWidthHeight;  // Just return features from VP8X header.
     }
   }
 
-  if (data_size < TAG_SIZE) return VP8_STATUS_NOT_ENOUGH_DATA;
+  if (data_size < TAG_SIZE) {
+    status = VP8_STATUS_NOT_ENOUGH_DATA;
+    goto ReturnWidthHeight;
+  }
 
   // Skip over optional chunks if data started with "RIFF + VP8X" or "ALPH".
   if ((found_riff && found_vp8x) ||
@@ -322,43 +355,49 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
     status = ParseOptionalChunks(&data, &data_size, hdrs.riff_size,
                                  &hdrs.alpha_data, &hdrs.alpha_data_size);
     if (status != VP8_STATUS_OK) {
-      return status;  // Found an invalid chunk size / insufficient data.
+      goto ReturnWidthHeight;  // Invalid chunk size / insufficient data.
     }
   }
 
   // Skip over VP8/VP8L header.
-  status = ParseVP8Header(&data, &data_size, hdrs.riff_size,
+  status = ParseVP8Header(&data, &data_size, have_all_data, hdrs.riff_size,
                           &hdrs.compressed_size, &hdrs.is_lossless);
   if (status != VP8_STATUS_OK) {
-    return status;  // Wrong VP8/VP8L chunk-header / insufficient data.
+    goto ReturnWidthHeight;  // Wrong VP8/VP8L chunk-header / insufficient data.
   }
   if (hdrs.compressed_size > MAX_CHUNK_PAYLOAD) {
     return VP8_STATUS_BITSTREAM_ERROR;
   }
 
+  if (format != NULL && !(animation_present || fragments_present)) {
+    *format = hdrs.is_lossless ? 2 : 1;
+  }
+
   if (!hdrs.is_lossless) {
     if (data_size < VP8_FRAME_HEADER_SIZE) {
-      return VP8_STATUS_NOT_ENOUGH_DATA;
+      status = VP8_STATUS_NOT_ENOUGH_DATA;
+      goto ReturnWidthHeight;
     }
     // Validates raw VP8 data.
-    if (!VP8GetInfo(data, data_size,
-                    (uint32_t)hdrs.compressed_size, width, height)) {
+    if (!VP8GetInfo(data, data_size, (uint32_t)hdrs.compressed_size,
+                    &image_width, &image_height)) {
       return VP8_STATUS_BITSTREAM_ERROR;
     }
   } else {
     if (data_size < VP8L_FRAME_HEADER_SIZE) {
-      return VP8_STATUS_NOT_ENOUGH_DATA;
+      status = VP8_STATUS_NOT_ENOUGH_DATA;
+      goto ReturnWidthHeight;
     }
     // Validates raw VP8L data.
-    if (!VP8LGetInfo(data, data_size, width, height, has_alpha)) {
+    if (!VP8LGetInfo(data, data_size, &image_width, &image_height, has_alpha)) {
       return VP8_STATUS_BITSTREAM_ERROR;
     }
   }
-
-  if (has_alpha != NULL) {
-    // If the data did not contain a VP8X/VP8L chunk the only definitive way
-    // to set this is by looking for alpha data (from an ALPH chunk).
-    *has_alpha |= (hdrs.alpha_data != NULL);
+  // Validates image size coherency.
+  if (found_vp8x) {
+    if (canvas_width != image_width || canvas_height != image_height) {
+      return VP8_STATUS_BITSTREAM_ERROR;
+    }
   }
   if (headers != NULL) {
     *headers = hdrs;
@@ -366,21 +405,44 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
     assert((uint64_t)(data - headers->data) < MAX_CHUNK_PAYLOAD);
     assert(headers->offset == headers->data_size - data_size);
   }
-  return VP8_STATUS_OK;  // Return features from VP8 header.
+ ReturnWidthHeight:
+  if (status == VP8_STATUS_OK ||
+      (status == VP8_STATUS_NOT_ENOUGH_DATA && found_vp8x && headers == NULL)) {
+    if (has_alpha != NULL) {
+      // If the data did not contain a VP8X/VP8L chunk the only definitive way
+      // to set this is by looking for alpha data (from an ALPH chunk).
+      *has_alpha |= (hdrs.alpha_data != NULL);
+    }
+    if (width != NULL) *width = image_width;
+    if (height != NULL) *height = image_height;
+    return VP8_STATUS_OK;
+  } else {
+    return status;
+  }
 }
 
 VP8StatusCode WebPParseHeaders(WebPHeaderStructure* const headers) {
+  VP8StatusCode status;
+  int has_animation = 0;
   assert(headers != NULL);
   // fill out headers, ignore width/height/has_alpha.
-  return ParseHeadersInternal(headers->data, headers->data_size,
-                              NULL, NULL, NULL, headers);
+  status = ParseHeadersInternal(headers->data, headers->data_size,
+                                NULL, NULL, NULL, &has_animation,
+                                NULL, headers);
+  if (status == VP8_STATUS_OK || status == VP8_STATUS_NOT_ENOUGH_DATA) {
+    // TODO(jzern): full support of animation frames will require API additions.
+    if (has_animation) {
+      status = VP8_STATUS_UNSUPPORTED_FEATURE;
+    }
+  }
+  return status;
 }
 
 //------------------------------------------------------------------------------
 // WebPDecParams
 
 void WebPResetDecParams(WebPDecParams* const params) {
-  if (params) {
+  if (params != NULL) {
     memset(params, 0, sizeof(*params));
   }
 }
@@ -397,6 +459,7 @@ static VP8StatusCode DecodeInto(const uint8_t* const data, size_t data_size,
 
   headers.data = data;
   headers.data_size = data_size;
+  headers.have_all_data = 1;
   status = WebPParseHeaders(&headers);   // Process Pre-VP8 chunks.
   if (status != VP8_STATUS_OK) {
     return status;
@@ -413,11 +476,6 @@ static VP8StatusCode DecodeInto(const uint8_t* const data, size_t data_size,
     if (dec == NULL) {
       return VP8_STATUS_OUT_OF_MEMORY;
     }
-#ifdef WEBP_USE_THREAD
-    dec->use_threads_ = params->options && (params->options->use_threads > 0);
-#else
-    dec->use_threads_ = 0;
-#endif
     dec->alpha_data_ = headers.alpha_data;
     dec->alpha_data_size_ = headers.alpha_data_size;
 
@@ -429,6 +487,10 @@ static VP8StatusCode DecodeInto(const uint8_t* const data, size_t data_size,
       status = WebPAllocateDecBuffer(io.width, io.height, params->options,
                                      params->output);
       if (status == VP8_STATUS_OK) {  // Decode
+        // This change must be done before calling VP8Decode()
+        dec->mt_method_ = VP8GetThreadMethod(params->options, &headers,
+                                             io.width, io.height);
+        VP8InitDithering(params->options, dec);
         if (!VP8Decode(dec, &io)) {
           status = dec->status_;
         }
@@ -458,6 +520,12 @@ static VP8StatusCode DecodeInto(const uint8_t* const data, size_t data_size,
   if (status != VP8_STATUS_OK) {
     WebPFreeDecBuffer(params->output);
   }
+
+#if WEBP_DECODER_ABI_VERSION > 0x0203
+  if (params->options != NULL && params->options->flip) {
+    status = WebPFlipBuffer(params->output);
+  }
+#endif
   return status;
 }
 
@@ -615,7 +683,6 @@ uint8_t* WebPDecodeYUV(const uint8_t* data, size_t data_size,
 static void DefaultFeatures(WebPBitstreamFeatures* const features) {
   assert(features != NULL);
   memset(features, 0, sizeof(*features));
-  features->bitstream_version = 0;
 }
 
 static VP8StatusCode GetFeatures(const uint8_t* const data, size_t data_size,
@@ -625,10 +692,11 @@ static VP8StatusCode GetFeatures(const uint8_t* const data, size_t data_size,
   }
   DefaultFeatures(features);
 
-  // Only parse enough of the data to retrieve width/height/has_alpha.
+  // Only parse enough of the data to retrieve the features.
   return ParseHeadersInternal(data, data_size,
                               &features->width, &features->height,
-                              &features->has_alpha, NULL);
+                              &features->has_alpha, &features->has_animation,
+                              &features->format, NULL);
 }
 
 //------------------------------------------------------------------------------
@@ -672,19 +740,13 @@ int WebPInitDecoderConfigInternal(WebPDecoderConfig* config,
 VP8StatusCode WebPGetFeaturesInternal(const uint8_t* data, size_t data_size,
                                       WebPBitstreamFeatures* features,
                                       int version) {
-  VP8StatusCode status;
   if (WEBP_ABI_IS_INCOMPATIBLE(version, WEBP_DECODER_ABI_VERSION)) {
     return VP8_STATUS_INVALID_PARAM;   // version mismatch
   }
   if (features == NULL) {
     return VP8_STATUS_INVALID_PARAM;
   }
-
-  status = GetFeatures(data, data_size, features);
-  if (status == VP8_STATUS_NOT_ENOUGH_DATA) {
-    return VP8_STATUS_BITSTREAM_ERROR;  // Not-enough-data treated as error.
-  }
-  return status;
+  return GetFeatures(data, data_size, features);
 }
 
 VP8StatusCode WebPDecode(const uint8_t* data, size_t data_size,
@@ -728,9 +790,9 @@ int WebPIoInitFromOptions(const WebPDecoderOptions* const options,
     h = options->crop_height;
     x = options->crop_left;
     y = options->crop_top;
-    if (!WebPIsRGBMode(src_colorspace)) {   // only snap for YUV420 or YUV422
+    if (!WebPIsRGBMode(src_colorspace)) {   // only snap for YUV420
       x &= ~1;
-      y &= ~1;    // TODO(later): only for YUV420, not YUV422.
+      y &= ~1;
     }
     if (x < 0 || y < 0 || w <= 0 || h <= 0 || x + w > W || y + h > H) {
       return 0;  // out of frame boundary error
@@ -772,6 +834,3 @@ int WebPIoInitFromOptions(const WebPDecoderOptions* const options,
 
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dec/webpi.h

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Internal header: WebP decoding parameters and custom IO on buffer
@@ -12,7 +14,7 @@
 #ifndef WEBP_DEC_WEBPI_H_
 #define WEBP_DEC_WEBPI_H_
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 extern "C" {
 #endif
 
@@ -52,6 +54,7 @@ void WebPResetDecParams(WebPDecParams* const params);
 typedef struct {
   const uint8_t* data;         // input buffer
   size_t data_size;            // input buffer size
+  int have_all_data;           // true if all data is known to be available
   size_t offset;               // offset to main data chunk (VP8 or VP8L)
   const uint8_t* alpha_data;   // points to alpha chunk (if present)
   size_t alpha_data_size;      // alpha chunk size
@@ -61,10 +64,10 @@ typedef struct {
 } WebPHeaderStructure;
 
 // Skips over all valid chunks prior to the first VP8/VP8L frame header.
-// Returns VP8_STATUS_OK on success,
-//         VP8_STATUS_BITSTREAM_ERROR if an invalid header/chunk is found, and
-//         VP8_STATUS_NOT_ENOUGH_DATA if case of insufficient data.
-// In 'headers', compressed_size, offset, alpha_data, alpha_size and lossless
+// Returns: VP8_STATUS_OK, VP8_STATUS_BITSTREAM_ERROR (invalid header/chunk),
+// VP8_STATUS_NOT_ENOUGH_DATA (partial input) or VP8_STATUS_UNSUPPORTED_FEATURE
+// in the case of non-decodable features (animation for instance).
+// In 'headers', compressed_size, offset, alpha_data, alpha_size, and lossless
 // fields are updated appropriately upon success.
 VP8StatusCode WebPParseHeaders(WebPHeaderStructure* const headers);
 
@@ -91,10 +94,15 @@ int WebPIoInitFromOptions(const WebPDecoderOptions* const options,
 // dimension / etc.). If *options is not NULL, also verify that the options'
 // parameters are valid and apply them to the width/height dimensions of the
 // output buffer. This takes cropping / scaling / rotation into account.
+// Also incorporates the options->flip flag to flip the buffer parameters if
+// needed.
 VP8StatusCode WebPAllocateDecBuffer(int width, int height,
                                     const WebPDecoderOptions* const options,
                                     WebPDecBuffer* const buffer);
 
+// Flip buffer vertically by negating the various strides.
+VP8StatusCode WebPFlipBuffer(WebPDecBuffer* const buffer);
+
 // Copy 'src' into 'dst' buffer, making sure 'dst' is not marked as owner of the
 // memory (still held by 'src').
 void WebPCopyDecBuffer(const WebPDecBuffer* const src,
@@ -103,11 +111,9 @@ void WebPCopyDecBuffer(const WebPDecBuffer* const src,
 // Copy and transfer ownership from src to dst (beware of parameter order!)
 void WebPGrabDecBuffer(WebPDecBuffer* const src, WebPDecBuffer* const dst);
 
-
-
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 

src/demux/Makefile.am

@@ -0,0 +1,14 @@
+lib_LTLIBRARIES = libwebpdemux.la
+
+libwebpdemux_la_SOURCES =
+libwebpdemux_la_SOURCES += demux.c
+
+libwebpdemuxinclude_HEADERS =
+libwebpdemuxinclude_HEADERS += ../webp/demux.h
+libwebpdemuxinclude_HEADERS += ../webp/mux_types.h
+libwebpdemuxinclude_HEADERS += ../webp/types.h
+
+libwebpdemux_la_LIBADD = ../libwebp.la
+libwebpdemux_la_LDFLAGS = -no-undefined -version-info 1:2:0
+libwebpdemuxincludedir = $(includedir)/webp
+pkgconfig_DATA = libwebpdemux.pc

src/demux/demux.c

@@ -1,26 +1,31 @@
 // Copyright 2012 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 //  WebP container demux.
 //
 
-#include "../webp/mux.h"
+#ifdef HAVE_CONFIG_H
+#include "../webp/config.h"
+#endif
 
+#include <assert.h>
 #include <stdlib.h>
 #include <string.h>
 
-#include "../webp/decode.h"  // WebPGetInfo
+#include "../utils/utils.h"
+#include "../webp/decode.h"     // WebPGetFeatures
+#include "../webp/demux.h"
 #include "../webp/format_constants.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
-#define MKFOURCC(a, b, c, d) ((uint32_t)(a) | (b) << 8 | (c) << 16 | (d) << 24)
+#define DMUX_MAJ_VERSION 0
+#define DMUX_MIN_VERSION 2
+#define DMUX_REV_VERSION 2
 
 typedef struct {
   size_t start_;        // start location of the data
@@ -38,9 +43,12 @@ typedef struct {
 typedef struct Frame {
   int x_offset_, y_offset_;
   int width_, height_;
+  int has_alpha_;
   int duration_;
-  int is_tile_;    // this is an image fragment from a 'TILE'.
-  int frame_num_;  // the referent frame number for use in assembling tiles.
+  WebPMuxAnimDispose dispose_method_;
+  WebPMuxAnimBlend blend_method_;
+  int is_fragment_;  // this is a frame fragment (and not a full frame).
+  int frame_num_;  // the referent frame number for use in assembling fragments.
   int complete_;   // img_components_ contains a full image.
   ChunkData img_components_[2];  // 0=VP8{,L} 1=ALPH
   struct Frame* next_;
@@ -58,9 +66,12 @@ struct WebPDemuxer {
   uint32_t feature_flags_;
   int canvas_width_, canvas_height_;
   int loop_count_;
+  uint32_t bgcolor_;
   int num_frames_;
   Frame* frames_;
+  Frame** frames_tail_;
   Chunk* chunks_;  // non-image chunks
+  Chunk** chunks_tail_;
 };
 
 typedef enum {
@@ -87,6 +98,12 @@ static const ChunkParser kMasterChunks[] = {
   { { '0', '0', '0', '0' }, NULL,             NULL },
 };
 
+//------------------------------------------------------------------------------
+
+int WebPGetDemuxVersion(void) {
+  return (DMUX_MAJ_VERSION << 16) | (DMUX_MIN_VERSION << 8) | DMUX_REV_VERSION;
+}
+
 // -----------------------------------------------------------------------------
 // MemBuffer
 
@@ -127,43 +144,30 @@ static WEBP_INLINE const uint8_t* GetBuffer(MemBuffer* const mem) {
   return mem->buf_ + mem->start_;
 }
 
-static WEBP_INLINE uint8_t GetByte(MemBuffer* const mem) {
+// Read from 'mem' and skip the read bytes.
+static WEBP_INLINE uint8_t ReadByte(MemBuffer* const mem) {
   const uint8_t byte = mem->buf_[mem->start_];
   Skip(mem, 1);
   return byte;
 }
 
-// Read 16, 24 or 32 bits stored in little-endian order.
-static WEBP_INLINE int ReadLE16s(const uint8_t* const data) {
-  return (int)(data[0] << 0) | (data[1] << 8);
-}
-
-static WEBP_INLINE int ReadLE24s(const uint8_t* const data) {
-  return ReadLE16s(data) | (data[2] << 16);
-}
-
-static WEBP_INLINE uint32_t ReadLE32(const uint8_t* const data) {
-  return (uint32_t)ReadLE24s(data) | (data[3] << 24);
-}
-
-// In addition to reading, skip the read bytes.
-static WEBP_INLINE int GetLE16s(MemBuffer* const mem) {
+static WEBP_INLINE int ReadLE16s(MemBuffer* const mem) {
   const uint8_t* const data = mem->buf_ + mem->start_;
-  const int val = ReadLE16s(data);
+  const int val = GetLE16(data);
   Skip(mem, 2);
   return val;
 }
 
-static WEBP_INLINE int GetLE24s(MemBuffer* const mem) {
+static WEBP_INLINE int ReadLE24s(MemBuffer* const mem) {
   const uint8_t* const data = mem->buf_ + mem->start_;
-  const int val = ReadLE24s(data);
+  const int val = GetLE24(data);
   Skip(mem, 3);
   return val;
 }
 
-static WEBP_INLINE uint32_t GetLE32(MemBuffer* const mem) {
+static WEBP_INLINE uint32_t ReadLE32(MemBuffer* const mem) {
   const uint8_t* const data = mem->buf_ + mem->start_;
-  const uint32_t val = ReadLE32(data);
+  const uint32_t val = GetLE32(data);
   Skip(mem, 4);
   return val;
 }
@@ -172,41 +176,37 @@ static WEBP_INLINE uint32_t GetLE32(MemBuffer* const mem) {
 // Secondary chunk parsing
 
 static void AddChunk(WebPDemuxer* const dmux, Chunk* const chunk) {
-  Chunk** c = &dmux->chunks_;
-  while (*c != NULL) c = &(*c)->next_;
-  *c = chunk;
+  *dmux->chunks_tail_ = chunk;
   chunk->next_ = NULL;
+  dmux->chunks_tail_ = &chunk->next_;
 }
 
 // Add a frame to the end of the list, ensuring the last frame is complete.
 // Returns true on success, false otherwise.
 static int AddFrame(WebPDemuxer* const dmux, Frame* const frame) {
-  const Frame* last_frame = NULL;
-  Frame** f = &dmux->frames_;
-  while (*f != NULL) {
-    last_frame = *f;
-    f = &(*f)->next_;
-  }
+  const Frame* const last_frame = *dmux->frames_tail_;
   if (last_frame != NULL && !last_frame->complete_) return 0;
-  *f = frame;
+
+  *dmux->frames_tail_ = frame;
   frame->next_ = NULL;
+  dmux->frames_tail_ = &frame->next_;
   return 1;
 }
 
 // Store image bearing chunks to 'frame'.
-static ParseStatus StoreFrame(int frame_num, MemBuffer* const mem,
-                              Frame* const frame) {
+static ParseStatus StoreFrame(int frame_num, uint32_t min_size,
+                              MemBuffer* const mem, Frame* const frame) {
   int alpha_chunks = 0;
   int image_chunks = 0;
-  int done = (MemDataSize(mem) < CHUNK_HEADER_SIZE);
+  int done = (MemDataSize(mem) < min_size);
   ParseStatus status = PARSE_OK;
 
   if (done) return PARSE_NEED_MORE_DATA;
 
   do {
     const size_t chunk_start_offset = mem->start_;
-    const uint32_t fourcc = GetLE32(mem);
-    const uint32_t payload_size = GetLE32(mem);
+    const uint32_t fourcc = ReadLE32(mem);
+    const uint32_t payload_size = ReadLE32(mem);
     const uint32_t payload_size_padded = payload_size + (payload_size & 1);
     const size_t payload_available = (payload_size_padded > MemDataSize(mem))
                                    ? MemDataSize(mem) : payload_size_padded;
@@ -222,29 +222,37 @@ static ParseStatus StoreFrame(int frame_num, MemBuffer* const mem,
           ++alpha_chunks;
           frame->img_components_[1].offset_ = chunk_start_offset;
           frame->img_components_[1].size_ = chunk_size;
+          frame->has_alpha_ = 1;
           frame->frame_num_ = frame_num;
           Skip(mem, payload_available);
         } else {
           goto Done;
         }
         break;
-      case MKFOURCC('V', 'P', '8', ' '):
       case MKFOURCC('V', 'P', '8', 'L'):
+        if (alpha_chunks > 0) return PARSE_ERROR;  // VP8L has its own alpha
+        // fall through
+      case MKFOURCC('V', 'P', '8', ' '):
         if (image_chunks == 0) {
-          int width = 0, height = 0;
+          // Extract the bitstream features, tolerating failures when the data
+          // is incomplete.
+          WebPBitstreamFeatures features;
+          const VP8StatusCode vp8_status =
+              WebPGetFeatures(mem->buf_ + chunk_start_offset, chunk_size,
+                              &features);
+          if (status == PARSE_NEED_MORE_DATA &&
+              vp8_status == VP8_STATUS_NOT_ENOUGH_DATA) {
+            return PARSE_NEED_MORE_DATA;
+          } else if (vp8_status != VP8_STATUS_OK) {
+            // We have enough data, and yet WebPGetFeatures() failed.
+            return PARSE_ERROR;
+          }
           ++image_chunks;
           frame->img_components_[0].offset_ = chunk_start_offset;
           frame->img_components_[0].size_ = chunk_size;
-          // Extract the width and height from the bitstream, tolerating
-          // failures when the data is incomplete.
-          if (!WebPGetInfo(mem->buf_ + frame->img_components_[0].offset_,
-                           frame->img_components_[0].size_, &width, &height) &&
-              status != PARSE_NEED_MORE_DATA) {
-            return PARSE_ERROR;
-          }
-
-          frame->width_ = width;
-          frame->height_ = height;
+          frame->width_ = features.width;
+          frame->height_ = features.height;
+          frame->has_alpha_ |= features.has_alpha;
           frame->frame_num_ = frame_num;
           frame->complete_ = (status == PARSE_OK);
           Skip(mem, payload_available);
@@ -275,43 +283,48 @@ static ParseStatus StoreFrame(int frame_num, MemBuffer* const mem,
 // Returns PARSE_OK on success with *frame pointing to the new Frame.
 // Returns PARSE_NEED_MORE_DATA with insufficient data, PARSE_ERROR otherwise.
 static ParseStatus NewFrame(const MemBuffer* const mem,
-                            uint32_t min_size, uint32_t expected_size,
-                            uint32_t actual_size, Frame** frame) {
+                            uint32_t min_size, uint32_t actual_size,
+                            Frame** frame) {
   if (SizeIsInvalid(mem, min_size)) return PARSE_ERROR;
-  if (actual_size < expected_size) return PARSE_ERROR;
+  if (actual_size < min_size) return PARSE_ERROR;
   if (MemDataSize(mem) < min_size)  return PARSE_NEED_MORE_DATA;
 
-  *frame = (Frame*)calloc(1, sizeof(**frame));
+  *frame = (Frame*)WebPSafeCalloc(1ULL, sizeof(**frame));
   return (*frame == NULL) ? PARSE_ERROR : PARSE_OK;
 }
 
-// Parse a 'FRM ' chunk and any image bearing chunks that immediately follow.
+// Parse a 'ANMF' chunk and any image bearing chunks that immediately follow.
 // 'frame_chunk_size' is the previously validated, padded chunk size.
-static ParseStatus ParseFrame(
+static ParseStatus ParseAnimationFrame(
     WebPDemuxer* const dmux, uint32_t frame_chunk_size) {
-  const int has_frames = !!(dmux->feature_flags_ & ANIMATION_FLAG);
-  const uint32_t min_size = frame_chunk_size + CHUNK_HEADER_SIZE;
+  const int is_animation = !!(dmux->feature_flags_ & ANIMATION_FLAG);
+  const uint32_t anmf_payload_size = frame_chunk_size - ANMF_CHUNK_SIZE;
   int added_frame = 0;
+  int bits;
   MemBuffer* const mem = &dmux->mem_;
   Frame* frame;
   ParseStatus status =
-      NewFrame(mem, min_size, FRAME_CHUNK_SIZE, frame_chunk_size, &frame);
+      NewFrame(mem, ANMF_CHUNK_SIZE, frame_chunk_size, &frame);
   if (status != PARSE_OK) return status;
 
-  frame->x_offset_ = 2 * GetLE24s(mem);
-  frame->y_offset_ = 2 * GetLE24s(mem);
-  frame->width_    = 1 + GetLE24s(mem);
-  frame->height_   = 1 + GetLE24s(mem);
-  frame->duration_ = 1 + GetLE24s(mem);
-  Skip(mem, frame_chunk_size - FRAME_CHUNK_SIZE);  // skip any trailing data.
+  frame->x_offset_       = 2 * ReadLE24s(mem);
+  frame->y_offset_       = 2 * ReadLE24s(mem);
+  frame->width_          = 1 + ReadLE24s(mem);
+  frame->height_         = 1 + ReadLE24s(mem);
+  frame->duration_       = ReadLE24s(mem);
+  bits = ReadByte(mem);
+  frame->dispose_method_ =
+      (bits & 1) ? WEBP_MUX_DISPOSE_BACKGROUND : WEBP_MUX_DISPOSE_NONE;
+  frame->blend_method_ = (bits & 2) ? WEBP_MUX_NO_BLEND : WEBP_MUX_BLEND;
   if (frame->width_ * (uint64_t)frame->height_ >= MAX_IMAGE_AREA) {
+    WebPSafeFree(frame);
     return PARSE_ERROR;
   }
 
-  // Store a (potentially partial) frame only if the animation flag is set
-  // and there is some data in 'frame'.
-  status = StoreFrame(dmux->num_frames_ + 1, mem, frame);
-  if (status != PARSE_ERROR && has_frames && frame->frame_num_ > 0) {
+  // Store a frame only if the animation flag is set there is some data for
+  // this frame is available.
+  status = StoreFrame(dmux->num_frames_ + 1, anmf_payload_size, mem, frame);
+  if (status != PARSE_ERROR && is_animation && frame->frame_num_ > 0) {
     added_frame = AddFrame(dmux, frame);
     if (added_frame) {
       ++dmux->num_frames_;
@@ -320,48 +333,53 @@ static ParseStatus ParseFrame(
     }
   }
 
-  if (!added_frame) free(frame);
+  if (!added_frame) WebPSafeFree(frame);
   return status;
 }
 
-// Parse a 'TILE' chunk and any image bearing chunks that immediately follow.
-// 'tile_chunk_size' is the previously validated, padded chunk size.
-static ParseStatus ParseTile(WebPDemuxer* const dmux,
-                             uint32_t tile_chunk_size) {
-  const int has_tiles = !!(dmux->feature_flags_ & TILE_FLAG);
-  const uint32_t min_size = tile_chunk_size + CHUNK_HEADER_SIZE;
-  int added_tile = 0;
+#ifdef WEBP_EXPERIMENTAL_FEATURES
+// Parse a 'FRGM' chunk and any image bearing chunks that immediately follow.
+// 'fragment_chunk_size' is the previously validated, padded chunk size.
+static ParseStatus ParseFragment(WebPDemuxer* const dmux,
+                                 uint32_t fragment_chunk_size) {
+  const int frame_num = 1;  // All fragments belong to the 1st (and only) frame.
+  const int is_fragmented = !!(dmux->feature_flags_ & FRAGMENTS_FLAG);
+  const uint32_t frgm_payload_size = fragment_chunk_size - FRGM_CHUNK_SIZE;
+  int added_fragment = 0;
   MemBuffer* const mem = &dmux->mem_;
   Frame* frame;
   ParseStatus status =
-      NewFrame(mem, min_size, TILE_CHUNK_SIZE, tile_chunk_size, &frame);
+      NewFrame(mem, FRGM_CHUNK_SIZE, fragment_chunk_size, &frame);
   if (status != PARSE_OK) return status;
 
-  frame->is_tile_  = 1;
-  frame->x_offset_ = 2 * GetLE24s(mem);
-  frame->y_offset_ = 2 * GetLE24s(mem);
-  Skip(mem, tile_chunk_size - TILE_CHUNK_SIZE);  // skip any trailing data.
+  frame->is_fragment_ = 1;
+  frame->x_offset_ = 2 * ReadLE24s(mem);
+  frame->y_offset_ = 2 * ReadLE24s(mem);
 
-  // Store a (potentially partial) tile only if the tile flag is set
-  // and the tile contains some data.
-  status = StoreFrame(dmux->num_frames_, mem, frame);
-  if (status != PARSE_ERROR && has_tiles && frame->frame_num_ > 0) {
-    // Note num_frames_ is incremented only when all tiles have been consumed.
-    added_tile = AddFrame(dmux, frame);
-    if (!added_tile) status = PARSE_ERROR;
+  // Store a fragment only if the 'fragments' flag is set and there is some
+  // data available.
+  status = StoreFrame(frame_num, frgm_payload_size, mem, frame);
+  if (status != PARSE_ERROR && is_fragmented && frame->frame_num_ > 0) {
+    added_fragment = AddFrame(dmux, frame);
+    if (!added_fragment) {
+      status = PARSE_ERROR;
+    } else {
+      dmux->num_frames_ = 1;
+    }
   }
 
-  if (!added_tile) free(frame);
+  if (!added_fragment) WebPSafeFree(frame);
   return status;
 }
+#endif  // WEBP_EXPERIMENTAL_FEATURES
 
-// General chunk storage starting with the header at 'start_offset' allowing
+// General chunk storage, starting with the header at 'start_offset', allowing
 // the user to request the payload via a fourcc string. 'size' includes the
 // header and the unpadded payload size.
 // Returns true on success, false otherwise.
 static int StoreChunk(WebPDemuxer* const dmux,
                       size_t start_offset, uint32_t size) {
-  Chunk* const chunk = (Chunk*)calloc(1, sizeof(*chunk));
+  Chunk* const chunk = (Chunk*)WebPSafeCalloc(1ULL, sizeof(*chunk));
   if (chunk == NULL) return 0;
 
   chunk->data_.offset_ = start_offset;
@@ -373,20 +391,20 @@ static int StoreChunk(WebPDemuxer* const dmux,
 // -----------------------------------------------------------------------------
 // Primary chunk parsing
 
-static int ReadHeader(MemBuffer* const mem) {
+static ParseStatus ReadHeader(MemBuffer* const mem) {
   const size_t min_size = RIFF_HEADER_SIZE + CHUNK_HEADER_SIZE;
   uint32_t riff_size;
 
   // Basic file level validation.
-  if (MemDataSize(mem) < min_size) return 0;
+  if (MemDataSize(mem) < min_size) return PARSE_NEED_MORE_DATA;
   if (memcmp(GetBuffer(mem), "RIFF", CHUNK_SIZE_BYTES) ||
       memcmp(GetBuffer(mem) + CHUNK_HEADER_SIZE, "WEBP", CHUNK_SIZE_BYTES)) {
-    return 0;
+    return PARSE_ERROR;
   }
 
-  riff_size = ReadLE32(GetBuffer(mem) + TAG_SIZE);
-  if (riff_size < CHUNK_HEADER_SIZE) return 0;
-  if (riff_size > MAX_CHUNK_PAYLOAD) return 0;
+  riff_size = GetLE32(GetBuffer(mem) + TAG_SIZE);
+  if (riff_size < CHUNK_HEADER_SIZE) return PARSE_ERROR;
+  if (riff_size > MAX_CHUNK_PAYLOAD) return PARSE_ERROR;
 
   // There's no point in reading past the end of the RIFF chunk
   mem->riff_end_ = riff_size + CHUNK_HEADER_SIZE;
@@ -395,7 +413,7 @@ static int ReadHeader(MemBuffer* const mem) {
   }
 
   Skip(mem, RIFF_HEADER_SIZE);
-  return 1;
+  return PARSE_OK;
 }
 
 static ParseStatus ParseSingleImage(WebPDemuxer* const dmux) {
@@ -403,73 +421,58 @@ static ParseStatus ParseSingleImage(WebPDemuxer* const dmux) {
   MemBuffer* const mem = &dmux->mem_;
   Frame* frame;
   ParseStatus status;
+  int image_added = 0;
 
   if (dmux->frames_ != NULL) return PARSE_ERROR;
   if (SizeIsInvalid(mem, min_size)) return PARSE_ERROR;
   if (MemDataSize(mem) < min_size) return PARSE_NEED_MORE_DATA;
 
-  frame = (Frame*)calloc(1, sizeof(*frame));
+  frame = (Frame*)WebPSafeCalloc(1ULL, sizeof(*frame));
   if (frame == NULL) return PARSE_ERROR;
 
-  status = StoreFrame(1, &dmux->mem_, frame);
+  // For the single image case we allow parsing of a partial frame, but we need
+  // at least CHUNK_HEADER_SIZE for parsing.
+  status = StoreFrame(1, CHUNK_HEADER_SIZE, &dmux->mem_, frame);
   if (status != PARSE_ERROR) {
     const int has_alpha = !!(dmux->feature_flags_ & ALPHA_FLAG);
     // Clear any alpha when the alpha flag is missing.
     if (!has_alpha && frame->img_components_[1].size_ > 0) {
       frame->img_components_[1].offset_ = 0;
       frame->img_components_[1].size_ = 0;
+      frame->has_alpha_ = 0;
     }
 
     // Use the frame width/height as the canvas values for non-vp8x files.
+    // Also, set ALPHA_FLAG if this is a lossless image with alpha.
     if (!dmux->is_ext_format_ && frame->width_ > 0 && frame->height_ > 0) {
       dmux->state_ = WEBP_DEMUX_PARSED_HEADER;
       dmux->canvas_width_ = frame->width_;
       dmux->canvas_height_ = frame->height_;
+      dmux->feature_flags_ |= frame->has_alpha_ ? ALPHA_FLAG : 0;
+    }
+    if (!AddFrame(dmux, frame)) {
+      status = PARSE_ERROR;  // last frame was left incomplete
+    } else {
+      image_added = 1;
+      dmux->num_frames_ = 1;
     }
-    AddFrame(dmux, frame);
-    dmux->num_frames_ = 1;
-  } else {
-    free(frame);
   }
 
+  if (!image_added) WebPSafeFree(frame);
   return status;
 }
 
-static ParseStatus ParseVP8X(WebPDemuxer* const dmux) {
+static ParseStatus ParseVP8XChunks(WebPDemuxer* const dmux) {
+  const int is_animation = !!(dmux->feature_flags_ & ANIMATION_FLAG);
   MemBuffer* const mem = &dmux->mem_;
-  int loop_chunks = 0;
-  uint32_t vp8x_size;
+  int anim_chunks = 0;
   ParseStatus status = PARSE_OK;
 
-  if (MemDataSize(mem) < CHUNK_HEADER_SIZE) return PARSE_NEED_MORE_DATA;
-
-  dmux->is_ext_format_ = 1;
-  Skip(mem, TAG_SIZE);  // VP8X
-  vp8x_size = GetLE32(mem);
-  if (vp8x_size > MAX_CHUNK_PAYLOAD) return PARSE_ERROR;
-  if (vp8x_size < VP8X_CHUNK_SIZE) return PARSE_ERROR;
-  vp8x_size += vp8x_size & 1;
-  if (SizeIsInvalid(mem, vp8x_size)) return PARSE_ERROR;
-  if (MemDataSize(mem) < vp8x_size) return PARSE_NEED_MORE_DATA;
-
-  dmux->feature_flags_ = GetByte(mem);
-  Skip(mem, 3);  // Reserved.
-  dmux->canvas_width_  = 1 + GetLE24s(mem);
-  dmux->canvas_height_ = 1 + GetLE24s(mem);
-  if (dmux->canvas_width_ * (uint64_t)dmux->canvas_height_ >= MAX_IMAGE_AREA) {
-    return PARSE_ERROR;  // image final dimension is too large
-  }
-  Skip(mem, vp8x_size - VP8X_CHUNK_SIZE);  // skip any trailing data.
-  dmux->state_ = WEBP_DEMUX_PARSED_HEADER;
-
-  if (SizeIsInvalid(mem, CHUNK_HEADER_SIZE)) return PARSE_ERROR;
-  if (MemDataSize(mem) < CHUNK_HEADER_SIZE) return PARSE_NEED_MORE_DATA;
-
   do {
     int store_chunk = 1;
     const size_t chunk_start_offset = mem->start_;
-    const uint32_t fourcc = GetLE32(mem);
-    const uint32_t chunk_size = GetLE32(mem);
+    const uint32_t fourcc = ReadLE32(mem);
+    const uint32_t chunk_size = ReadLE32(mem);
     const uint32_t chunk_size_padded = chunk_size + (chunk_size & 1);
 
     if (chunk_size > MAX_CHUNK_PAYLOAD) return PARSE_ERROR;
@@ -482,40 +485,50 @@ static ParseStatus ParseVP8X(WebPDemuxer* const dmux) {
       case MKFOURCC('A', 'L', 'P', 'H'):
       case MKFOURCC('V', 'P', '8', ' '):
       case MKFOURCC('V', 'P', '8', 'L'): {
+        // check that this isn't an animation (all frames should be in an ANMF).
+        if (anim_chunks > 0 || is_animation) return PARSE_ERROR;
+
         Rewind(mem, CHUNK_HEADER_SIZE);
         status = ParseSingleImage(dmux);
         break;
       }
-      case MKFOURCC('L', 'O', 'O', 'P'): {
-        if (chunk_size_padded < LOOP_CHUNK_SIZE) return PARSE_ERROR;
+      case MKFOURCC('A', 'N', 'I', 'M'): {
+        if (chunk_size_padded < ANIM_CHUNK_SIZE) return PARSE_ERROR;
 
         if (MemDataSize(mem) < chunk_size_padded) {
           status = PARSE_NEED_MORE_DATA;
-        } else if (loop_chunks == 0) {
-          ++loop_chunks;
-          dmux->loop_count_ = GetLE16s(mem);
-          Skip(mem, chunk_size_padded - LOOP_CHUNK_SIZE);
+        } else if (anim_chunks == 0) {
+          ++anim_chunks;
+          dmux->bgcolor_ = ReadLE32(mem);
+          dmux->loop_count_ = ReadLE16s(mem);
+          Skip(mem, chunk_size_padded - ANIM_CHUNK_SIZE);
         } else {
           store_chunk = 0;
           goto Skip;
         }
         break;
       }
-      case MKFOURCC('F', 'R', 'M', ' '): {
-        status = ParseFrame(dmux, chunk_size_padded);
+      case MKFOURCC('A', 'N', 'M', 'F'): {
+        if (anim_chunks == 0) return PARSE_ERROR;  // 'ANIM' precedes frames.
+        status = ParseAnimationFrame(dmux, chunk_size_padded);
         break;
       }
-      case MKFOURCC('T', 'I', 'L', 'E'): {
-        if (dmux->num_frames_ == 0) dmux->num_frames_ = 1;
-        status = ParseTile(dmux, chunk_size_padded);
+#ifdef WEBP_EXPERIMENTAL_FEATURES
+      case MKFOURCC('F', 'R', 'G', 'M'): {
+        status = ParseFragment(dmux, chunk_size_padded);
         break;
       }
+#endif
       case MKFOURCC('I', 'C', 'C', 'P'): {
         store_chunk = !!(dmux->feature_flags_ & ICCP_FLAG);
         goto Skip;
       }
-      case MKFOURCC('M', 'E', 'T', 'A'): {
-        store_chunk = !!(dmux->feature_flags_ & META_FLAG);
+      case MKFOURCC('E', 'X', 'I', 'F'): {
+        store_chunk = !!(dmux->feature_flags_ & EXIF_FLAG);
+        goto Skip;
+      }
+      case MKFOURCC('X', 'M', 'P', ' '): {
+        store_chunk = !!(dmux->feature_flags_ & XMP_FLAG);
         goto Skip;
       }
  Skip:
@@ -546,6 +559,37 @@ static ParseStatus ParseVP8X(WebPDemuxer* const dmux) {
   return status;
 }
 
+static ParseStatus ParseVP8X(WebPDemuxer* const dmux) {
+  MemBuffer* const mem = &dmux->mem_;
+  uint32_t vp8x_size;
+
+  if (MemDataSize(mem) < CHUNK_HEADER_SIZE) return PARSE_NEED_MORE_DATA;
+
+  dmux->is_ext_format_ = 1;
+  Skip(mem, TAG_SIZE);  // VP8X
+  vp8x_size = ReadLE32(mem);
+  if (vp8x_size > MAX_CHUNK_PAYLOAD) return PARSE_ERROR;
+  if (vp8x_size < VP8X_CHUNK_SIZE) return PARSE_ERROR;
+  vp8x_size += vp8x_size & 1;
+  if (SizeIsInvalid(mem, vp8x_size)) return PARSE_ERROR;
+  if (MemDataSize(mem) < vp8x_size) return PARSE_NEED_MORE_DATA;
+
+  dmux->feature_flags_ = ReadByte(mem);
+  Skip(mem, 3);  // Reserved.
+  dmux->canvas_width_  = 1 + ReadLE24s(mem);
+  dmux->canvas_height_ = 1 + ReadLE24s(mem);
+  if (dmux->canvas_width_ * (uint64_t)dmux->canvas_height_ >= MAX_IMAGE_AREA) {
+    return PARSE_ERROR;  // image final dimension is too large
+  }
+  Skip(mem, vp8x_size - VP8X_CHUNK_SIZE);  // skip any trailing data.
+  dmux->state_ = WEBP_DEMUX_PARSED_HEADER;
+
+  if (SizeIsInvalid(mem, CHUNK_HEADER_SIZE)) return PARSE_ERROR;
+  if (MemDataSize(mem) < CHUNK_HEADER_SIZE) return PARSE_NEED_MORE_DATA;
+
+  return ParseVP8XChunks(dmux);
+}
+
 // -----------------------------------------------------------------------------
 // Format validation
 
@@ -560,30 +604,55 @@ static int IsValidSimpleFormat(const WebPDemuxer* const dmux) {
   return 1;
 }
 
+// If 'exact' is true, check that the image resolution matches the canvas.
+// If 'exact' is false, check that the x/y offsets do not exceed the canvas.
+// TODO(jzern): this is insufficient in the fragmented image case if the
+// expectation is that the fragments completely cover the canvas.
+static int CheckFrameBounds(const Frame* const frame, int exact,
+                            int canvas_width, int canvas_height) {
+  if (exact) {
+    if (frame->x_offset_ != 0 || frame->y_offset_ != 0) {
+      return 0;
+    }
+    if (frame->width_ != canvas_width || frame->height_ != canvas_height) {
+      return 0;
+    }
+  } else {
+    if (frame->x_offset_ < 0 || frame->y_offset_ < 0) return 0;
+    if (frame->width_ + frame->x_offset_ > canvas_width) return 0;
+    if (frame->height_ + frame->y_offset_ > canvas_height) return 0;
+  }
+  return 1;
+}
+
 static int IsValidExtendedFormat(const WebPDemuxer* const dmux) {
-  const int has_tiles = !!(dmux->feature_flags_ & TILE_FLAG);
-  const int has_frames = !!(dmux->feature_flags_ & ANIMATION_FLAG);
-  const Frame* f;
+  const int is_animation = !!(dmux->feature_flags_ & ANIMATION_FLAG);
+  const int is_fragmented = !!(dmux->feature_flags_ & FRAGMENTS_FLAG);
+  const Frame* f = dmux->frames_;
 
   if (dmux->state_ == WEBP_DEMUX_PARSING_HEADER) return 1;
 
   if (dmux->canvas_width_ <= 0 || dmux->canvas_height_ <= 0) return 0;
   if (dmux->loop_count_ < 0) return 0;
   if (dmux->state_ == WEBP_DEMUX_DONE && dmux->frames_ == NULL) return 0;
+#ifndef WEBP_EXPERIMENTAL_FEATURES
+  if (is_fragmented) return 0;
+#endif
 
-  for (f = dmux->frames_; f != NULL; f = f->next_) {
+  while (f != NULL) {
     const int cur_frame_set = f->frame_num_;
-    int frame_count = 0, tile_count = 0;
+    int frame_count = 0, fragment_count = 0;
 
-    // Check frame properties and if the image is composed of tiles that each
-    // fragment came from a 'TILE'.
+    // Check frame properties and if the image is composed of fragments that
+    // each fragment came from a fragment.
     for (; f != NULL && f->frame_num_ == cur_frame_set; f = f->next_) {
       const ChunkData* const image = f->img_components_;
       const ChunkData* const alpha = f->img_components_ + 1;
 
-      if (!has_tiles && f->is_tile_) return 0;
-      if (!has_frames && f->frame_num_ > 1) return 0;
-      if (f->x_offset_ < 0 || f->y_offset_ < 0) return 0;
+      if (is_fragmented && !f->is_fragment_) return 0;
+      if (!is_fragmented && f->is_fragment_) return 0;
+      if (!is_animation && f->frame_num_ > 1) return 0;
+
       if (f->complete_) {
         if (alpha->size_ == 0 && image->size_ == 0) return 0;
         // Ensure alpha precedes image bitstream.
@@ -593,6 +662,9 @@ static int IsValidExtendedFormat(const WebPDemuxer* const dmux) {
 
         if (f->width_ <= 0 || f->height_ <= 0) return 0;
       } else {
+        // There shouldn't be a partial frame in a complete file.
+        if (dmux->state_ == WEBP_DEMUX_DONE) return 0;
+
         // Ensure alpha precedes image bitstream.
         if (alpha->size_ > 0 && image->size_ > 0 &&
             alpha->offset_ > image->offset_) {
@@ -602,12 +674,17 @@ static int IsValidExtendedFormat(const WebPDemuxer* const dmux) {
         if (f->next_ != NULL) return 0;
       }
 
-      tile_count += f->is_tile_;
+      if (f->width_ > 0 && f->height_ > 0 &&
+          !CheckFrameBounds(f, !(is_animation || is_fragmented),
+                            dmux->canvas_width_, dmux->canvas_height_)) {
+        return 0;
+      }
+
+      fragment_count += f->is_fragment_;
       ++frame_count;
     }
-    if (!has_tiles && frame_count > 1) return 0;
-    if (tile_count > 0 && frame_count != tile_count) return 0;
-    if (f == NULL) break;
+    if (!is_fragmented && frame_count > 1) return 0;
+    if (fragment_count > 0 && frame_count != fragment_count) return 0;
   }
   return 1;
 }
@@ -618,8 +695,11 @@ static int IsValidExtendedFormat(const WebPDemuxer* const dmux) {
 static void InitDemux(WebPDemuxer* const dmux, const MemBuffer* const mem) {
   dmux->state_ = WEBP_DEMUX_PARSING_HEADER;
   dmux->loop_count_ = 1;
+  dmux->bgcolor_ = 0xFFFFFFFF;  // White background by default.
   dmux->canvas_width_ = -1;
   dmux->canvas_height_ = -1;
+  dmux->frames_tail_ = &dmux->frames_;
+  dmux->chunks_tail_ = &dmux->chunks_;
   dmux->mem_ = *mem;
 }
 
@@ -631,28 +711,40 @@ WebPDemuxer* WebPDemuxInternal(const WebPData* data, int allow_partial,
   MemBuffer mem;
   WebPDemuxer* dmux;
 
-  if (WEBP_ABI_IS_INCOMPATIBLE(version, WEBP_DEMUX_ABI_VERSION)) return NULL;
-  if (data == NULL || data->bytes_ == NULL || data->size_ == 0) return NULL;
+  if (state != NULL) *state = WEBP_DEMUX_PARSE_ERROR;
 
-  if (!InitMemBuffer(&mem, data->bytes_, data->size_)) return NULL;
-  if (!ReadHeader(&mem)) return NULL;
+  if (WEBP_ABI_IS_INCOMPATIBLE(version, WEBP_DEMUX_ABI_VERSION)) return NULL;
+  if (data == NULL || data->bytes == NULL || data->size == 0) return NULL;
+
+  if (!InitMemBuffer(&mem, data->bytes, data->size)) return NULL;
+  status = ReadHeader(&mem);
+  if (status != PARSE_OK) {
+    if (state != NULL) {
+      *state = (status == PARSE_NEED_MORE_DATA) ? WEBP_DEMUX_PARSING_HEADER
+                                                : WEBP_DEMUX_PARSE_ERROR;
+    }
+    return NULL;
+  }
 
   partial = (mem.buf_size_ < mem.riff_end_);
   if (!allow_partial && partial) return NULL;
 
-  dmux = (WebPDemuxer*)calloc(1, sizeof(*dmux));
+  dmux = (WebPDemuxer*)WebPSafeCalloc(1ULL, sizeof(*dmux));
   if (dmux == NULL) return NULL;
   InitDemux(dmux, &mem);
 
+  status = PARSE_ERROR;
   for (parser = kMasterChunks; parser->parse != NULL; ++parser) {
     if (!memcmp(parser->id, GetBuffer(&dmux->mem_), TAG_SIZE)) {
       status = parser->parse(dmux);
       if (status == PARSE_OK) dmux->state_ = WEBP_DEMUX_DONE;
+      if (status == PARSE_NEED_MORE_DATA && !partial) status = PARSE_ERROR;
       if (status != PARSE_ERROR && !parser->valid(dmux)) status = PARSE_ERROR;
+      if (status == PARSE_ERROR) dmux->state_ = WEBP_DEMUX_PARSE_ERROR;
       break;
     }
   }
-  if (state) *state = dmux->state_;
+  if (state != NULL) *state = dmux->state_;
 
   if (status == PARSE_ERROR) {
     WebPDemuxDelete(dmux);
@@ -669,14 +761,14 @@ void WebPDemuxDelete(WebPDemuxer* dmux) {
   for (f = dmux->frames_; f != NULL;) {
     Frame* const cur_frame = f;
     f = f->next_;
-    free(cur_frame);
+    WebPSafeFree(cur_frame);
   }
   for (c = dmux->chunks_; c != NULL;) {
     Chunk* const cur_chunk = c;
     c = c->next_;
-    free(cur_chunk);
+    WebPSafeFree(cur_chunk);
   }
-  free(dmux);
+  WebPSafeFree(dmux);
 }
 
 // -----------------------------------------------------------------------------
@@ -685,10 +777,12 @@ uint32_t WebPDemuxGetI(const WebPDemuxer* dmux, WebPFormatFeature feature) {
   if (dmux == NULL) return 0;
 
   switch (feature) {
-    case WEBP_FF_FORMAT_FLAGS:  return dmux->feature_flags_;
-    case WEBP_FF_CANVAS_WIDTH:  return (uint32_t)dmux->canvas_width_;
-    case WEBP_FF_CANVAS_HEIGHT: return (uint32_t)dmux->canvas_height_;
-    case WEBP_FF_LOOP_COUNT:    return (uint32_t)dmux->loop_count_;
+    case WEBP_FF_FORMAT_FLAGS:     return dmux->feature_flags_;
+    case WEBP_FF_CANVAS_WIDTH:     return (uint32_t)dmux->canvas_width_;
+    case WEBP_FF_CANVAS_HEIGHT:    return (uint32_t)dmux->canvas_height_;
+    case WEBP_FF_LOOP_COUNT:       return (uint32_t)dmux->loop_count_;
+    case WEBP_FF_BACKGROUND_COLOR: return dmux->bgcolor_;
+    case WEBP_FF_FRAME_COUNT:      return (uint32_t)dmux->num_frames_;
   }
   return 0;
 }
@@ -696,7 +790,8 @@ uint32_t WebPDemuxGetI(const WebPDemuxer* dmux, WebPFormatFeature feature) {
 // -----------------------------------------------------------------------------
 // Frame iteration
 
-// Find the first 'frame_num' frame. There may be multiple in a tiled frame.
+// Find the first 'frame_num' frame. There may be multiple such frames in a
+// fragmented frame.
 static const Frame* GetFrame(const WebPDemuxer* const dmux, int frame_num) {
   const Frame* f;
   for (f = dmux->frames_; f != NULL; f = f->next_) {
@@ -705,19 +800,19 @@ static const Frame* GetFrame(const WebPDemuxer* const dmux, int frame_num) {
   return f;
 }
 
-// Returns tile 'tile_num' and the total count.
-static const Frame* GetTile(
-    const Frame* const frame_set, int tile_num, int* const count) {
+// Returns fragment 'fragment_num' and the total count.
+static const Frame* GetFragment(
+    const Frame* const frame_set, int fragment_num, int* const count) {
   const int this_frame = frame_set->frame_num_;
   const Frame* f = frame_set;
-  const Frame* tile = NULL;
+  const Frame* fragment = NULL;
   int total;
 
   for (total = 0; f != NULL && f->frame_num_ == this_frame; f = f->next_) {
-    if (++total == tile_num) tile = f;
+    if (++total == fragment_num) fragment = f;
   }
   *count = total;
-  return tile;
+  return fragment;
 }
 
 static const uint8_t* GetFramePayload(const uint8_t* const mem_buf,
@@ -747,27 +842,33 @@ static const uint8_t* GetFramePayload(const uint8_t* const mem_buf,
 // Create a whole 'frame' from VP8 (+ alpha) or lossless.
 static int SynthesizeFrame(const WebPDemuxer* const dmux,
                            const Frame* const first_frame,
-                           int tile_num, WebPIterator* const iter) {
+                           int fragment_num, WebPIterator* const iter) {
   const uint8_t* const mem_buf = dmux->mem_.buf_;
-  int num_tiles;
+  int num_fragments;
   size_t payload_size = 0;
-  const Frame* const tile = GetTile(first_frame, tile_num, &num_tiles);
-  const uint8_t* const payload = GetFramePayload(mem_buf, tile, &payload_size);
+  const Frame* const fragment =
+      GetFragment(first_frame, fragment_num, &num_fragments);
+  const uint8_t* const payload =
+      GetFramePayload(mem_buf, fragment, &payload_size);
   if (payload == NULL) return 0;
+  assert(first_frame != NULL);
 
-  iter->frame_num_   = first_frame->frame_num_;
-  iter->num_frames_  = dmux->num_frames_;
-  iter->tile_num_    = tile_num;
-  iter->num_tiles_   = num_tiles;
-  iter->x_offset_    = tile->x_offset_;
-  iter->y_offset_    = tile->y_offset_;
-  iter->width_       = tile->width_;
-  iter->height_      = tile->height_;
-  iter->duration_    = tile->duration_;
-  iter->complete_    = tile->complete_;
-  iter->tile_.bytes_ = payload;
-  iter->tile_.size_  = payload_size;
-  // TODO(jzern): adjust offsets for 'TILE's embedded in 'FRM 's
+  iter->frame_num      = first_frame->frame_num_;
+  iter->num_frames     = dmux->num_frames_;
+  iter->fragment_num   = fragment_num;
+  iter->num_fragments  = num_fragments;
+  iter->x_offset       = fragment->x_offset_;
+  iter->y_offset       = fragment->y_offset_;
+  iter->width          = fragment->width_;
+  iter->height         = fragment->height_;
+  iter->has_alpha      = fragment->has_alpha_;
+  iter->duration       = fragment->duration_;
+  iter->dispose_method = fragment->dispose_method_;
+  iter->blend_method   = fragment->blend_method_;
+  iter->complete       = fragment->complete_;
+  iter->fragment.bytes = payload;
+  iter->fragment.size  = payload_size;
+  // TODO(jzern): adjust offsets for 'FRGM's embedded in 'ANMF's
   return 1;
 }
 
@@ -779,6 +880,8 @@ static int SetFrame(int frame_num, WebPIterator* const iter) {
   if (frame_num == 0) frame_num = dmux->num_frames_;
 
   frame = GetFrame(dmux, frame_num);
+  if (frame == NULL) return 0;
+
   return SynthesizeFrame(dmux, frame, 1, iter);
 }
 
@@ -792,22 +895,22 @@ int WebPDemuxGetFrame(const WebPDemuxer* dmux, int frame, WebPIterator* iter) {
 
 int WebPDemuxNextFrame(WebPIterator* iter) {
   if (iter == NULL) return 0;
-  return SetFrame(iter->frame_num_ + 1, iter);
+  return SetFrame(iter->frame_num + 1, iter);
 }
 
 int WebPDemuxPrevFrame(WebPIterator* iter) {
   if (iter == NULL) return 0;
-  if (iter->frame_num_ <= 1) return 0;
-  return SetFrame(iter->frame_num_ - 1, iter);
+  if (iter->frame_num <= 1) return 0;
+  return SetFrame(iter->frame_num - 1, iter);
 }
 
-int WebPDemuxSelectTile(WebPIterator* iter, int tile) {
-  if (iter != NULL && iter->private_ != NULL && tile > 0) {
+int WebPDemuxSelectFragment(WebPIterator* iter, int fragment_num) {
+  if (iter != NULL && iter->private_ != NULL && fragment_num > 0) {
     const WebPDemuxer* const dmux = (WebPDemuxer*)iter->private_;
-    const Frame* const frame = GetFrame(dmux, iter->frame_num_);
+    const Frame* const frame = GetFrame(dmux, iter->frame_num);
     if (frame == NULL) return 0;
 
-    return SynthesizeFrame(dmux, frame, tile, iter);
+    return SynthesizeFrame(dmux, frame, fragment_num, iter);
   }
   return 0;
 }
@@ -856,10 +959,10 @@ static int SetChunk(const char fourcc[4], int chunk_num,
   if (chunk_num <= count) {
     const uint8_t* const mem_buf = dmux->mem_.buf_;
     const Chunk* const chunk = GetChunk(dmux, fourcc, chunk_num);
-    iter->chunk_.bytes_ = mem_buf + chunk->data_.offset_ + CHUNK_HEADER_SIZE;
-    iter->chunk_.size_  = chunk->data_.size_ - CHUNK_HEADER_SIZE;
-    iter->num_chunks_   = count;
-    iter->chunk_num_    = chunk_num;
+    iter->chunk.bytes = mem_buf + chunk->data_.offset_ + CHUNK_HEADER_SIZE;
+    iter->chunk.size  = chunk->data_.size_ - CHUNK_HEADER_SIZE;
+    iter->num_chunks  = count;
+    iter->chunk_num   = chunk_num;
     return 1;
   }
   return 0;
@@ -878,17 +981,17 @@ int WebPDemuxGetChunk(const WebPDemuxer* dmux,
 int WebPDemuxNextChunk(WebPChunkIterator* iter) {
   if (iter != NULL) {
     const char* const fourcc =
-        (const char*)iter->chunk_.bytes_ - CHUNK_HEADER_SIZE;
-    return SetChunk(fourcc, iter->chunk_num_ + 1, iter);
+        (const char*)iter->chunk.bytes - CHUNK_HEADER_SIZE;
+    return SetChunk(fourcc, iter->chunk_num + 1, iter);
   }
   return 0;
 }
 
 int WebPDemuxPrevChunk(WebPChunkIterator* iter) {
-  if (iter != NULL && iter->chunk_num_ > 1) {
+  if (iter != NULL && iter->chunk_num > 1) {
     const char* const fourcc =
-        (const char*)iter->chunk_.bytes_ - CHUNK_HEADER_SIZE;
-    return SetChunk(fourcc, iter->chunk_num_ - 1, iter);
+        (const char*)iter->chunk.bytes - CHUNK_HEADER_SIZE;
+    return SetChunk(fourcc, iter->chunk_num - 1, iter);
   }
   return 0;
 }
@@ -897,6 +1000,3 @@ void WebPDemuxReleaseChunkIterator(WebPChunkIterator* iter) {
   (void)iter;
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}  // extern "C"
-#endif

src/demux/libwebpdemux.pc.in

@@ -0,0 +1,11 @@
+prefix=@prefix@
+exec_prefix=@exec_prefix@
+libdir=@libdir@
+includedir=@includedir@
+
+Name: libwebpdemux
+Description: Library for parsing the WebP graphics format container
+Version: @PACKAGE_VERSION@
+Requires: libwebp >= 0.2.0
+Cflags: -I${includedir}
+Libs: -L${libdir} -lwebpdemux

src/dsp/Makefile.am

@@ -1,26 +1,72 @@
-AM_CPPFLAGS = -I$(top_srcdir)/src
-noinst_LTLIBRARIES = libwebpdsp.la
+noinst_LTLIBRARIES = libwebpdsp.la libwebpdsp_avx2.la
+noinst_LTLIBRARIES += libwebpdsp_sse2.la libwebpdspdecode_sse2.la
 
-libwebpdsp_la_SOURCES =
-libwebpdsp_la_SOURCES += cpu.c
-libwebpdsp_la_SOURCES += dec.c
-libwebpdsp_la_SOURCES += dec_neon.c
-libwebpdsp_la_SOURCES += dec_sse2.c
-libwebpdsp_la_SOURCES += dsp.h
-libwebpdsp_la_SOURCES += enc.c
-libwebpdsp_la_SOURCES += enc_sse2.c
-libwebpdsp_la_SOURCES += lossless.c
-libwebpdsp_la_SOURCES += lossless.h
-libwebpdsp_la_SOURCES += upsampling.c
-libwebpdsp_la_SOURCES += upsampling_sse2.c
-libwebpdsp_la_SOURCES += yuv.c
-libwebpdsp_la_SOURCES += yuv.h
+if BUILD_LIBWEBPDECODER
+  noinst_LTLIBRARIES += libwebpdspdecode.la
+endif
+
+common_HEADERS = ../webp/types.h
+commondir = $(includedir)/webp
+
+COMMON_SOURCES =
+COMMON_SOURCES += alpha_processing.c
+COMMON_SOURCES += cpu.c
+COMMON_SOURCES += dec.c
+COMMON_SOURCES += dec_clip_tables.c
+COMMON_SOURCES += dec_mips32.c
+COMMON_SOURCES += dec_neon.c
+COMMON_SOURCES += dsp.h
+COMMON_SOURCES += lossless.c
+COMMON_SOURCES += lossless.h
+COMMON_SOURCES += lossless_mips32.c
+COMMON_SOURCES += lossless_neon.c
+COMMON_SOURCES += neon.h
+COMMON_SOURCES += upsampling.c
+COMMON_SOURCES += upsampling_neon.c
+COMMON_SOURCES += yuv.c
+COMMON_SOURCES += yuv.h
+COMMON_SOURCES += yuv_mips32.c
+
+ENC_SOURCES =
+ENC_SOURCES += enc.c
+ENC_SOURCES += enc_mips32.c
+ENC_SOURCES += enc_neon.c
+
+libwebpdsp_avx2_la_SOURCES =
+libwebpdsp_avx2_la_SOURCES += enc_avx2.c
+libwebpdsp_avx2_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
+libwebpdsp_avx2_la_CFLAGS = $(AM_CFLAGS) $(AVX2_FLAGS)
+
+libwebpdspdecode_sse2_la_SOURCES =
+libwebpdspdecode_sse2_la_SOURCES += alpha_processing_sse2.c
+libwebpdspdecode_sse2_la_SOURCES += dec_sse2.c
+libwebpdspdecode_sse2_la_SOURCES += lossless_sse2.c
+libwebpdspdecode_sse2_la_SOURCES += upsampling_sse2.c
+libwebpdspdecode_sse2_la_SOURCES += yuv_sse2.c
+libwebpdspdecode_sse2_la_SOURCES += yuv_tables_sse2.h
+libwebpdspdecode_sse2_la_CPPFLAGS = $(libwebpdsp_sse2_la_CPPFLAGS)
+libwebpdspdecode_sse2_la_CFLAGS = $(libwebpdsp_sse2_la_CFLAGS)
+
+libwebpdsp_sse2_la_SOURCES =
+libwebpdsp_sse2_la_SOURCES += enc_sse2.c
+libwebpdsp_sse2_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
+libwebpdsp_sse2_la_CFLAGS = $(AM_CFLAGS) $(SSE2_FLAGS)
+libwebpdsp_sse2_la_LIBADD = libwebpdspdecode_sse2.la
+
+libwebpdsp_la_SOURCES = $(COMMON_SOURCES) $(ENC_SOURCES)
 
-libwebpdspinclude_HEADERS = ../webp/types.h
 noinst_HEADERS =
 noinst_HEADERS += ../dec/decode_vp8.h
 noinst_HEADERS += ../webp/decode.h
 
+libwebpdsp_la_CPPFLAGS = $(USE_EXPERIMENTAL_CODE) $(USE_SWAP_16BIT_CSP)
 libwebpdsp_la_LDFLAGS = -lm
-libwebpdsp_la_CPPFLAGS = $(USE_EXPERIMENTAL_CODE)
-libwebpdspincludedir = $(includedir)/webp
+libwebpdsp_la_LIBADD = libwebpdsp_avx2.la libwebpdsp_sse2.la
+
+if BUILD_LIBWEBPDECODER
+  libwebpdspdecode_la_SOURCES = $(COMMON_SOURCES)
+
+  libwebpdspdecode_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
+  libwebpdspdecode_la_LDFLAGS = $(libwebpdsp_la_LDFLAGS)
+  libwebpdspdecode_la_LIBADD = libwebpdspdecode_sse2.la
+endif

src/dsp/alpha_processing.c

@@ -0,0 +1,329 @@
+// Copyright 2013 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Utilities for processing transparent channel.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include <assert.h>
+#include "./dsp.h"
+
+// Tables can be faster on some platform but incur some extra binary size (~2k).
+// #define USE_TABLES_FOR_ALPHA_MULT
+
+// -----------------------------------------------------------------------------
+
+#define MFIX 24    // 24bit fixed-point arithmetic
+#define HALF ((1u << MFIX) >> 1)
+#define KINV_255 ((1u << MFIX) / 255u)
+
+static uint32_t Mult(uint8_t x, uint32_t mult) {
+  const uint32_t v = (x * mult + HALF) >> MFIX;
+  assert(v <= 255);  // <- 24bit precision is enough to ensure that.
+  return v;
+}
+
+#ifdef USE_TABLES_FOR_ALPHA_MULT
+
+static const uint32_t kMultTables[2][256] = {
+  {    // (255u << MFIX) / alpha
+    0x00000000, 0xff000000, 0x7f800000, 0x55000000, 0x3fc00000, 0x33000000,
+    0x2a800000, 0x246db6db, 0x1fe00000, 0x1c555555, 0x19800000, 0x172e8ba2,
+    0x15400000, 0x139d89d8, 0x1236db6d, 0x11000000, 0x0ff00000, 0x0f000000,
+    0x0e2aaaaa, 0x0d6bca1a, 0x0cc00000, 0x0c249249, 0x0b9745d1, 0x0b1642c8,
+    0x0aa00000, 0x0a333333, 0x09cec4ec, 0x0971c71c, 0x091b6db6, 0x08cb08d3,
+    0x08800000, 0x0839ce73, 0x07f80000, 0x07ba2e8b, 0x07800000, 0x07492492,
+    0x07155555, 0x06e45306, 0x06b5e50d, 0x0689d89d, 0x06600000, 0x063831f3,
+    0x06124924, 0x05ee23b8, 0x05cba2e8, 0x05aaaaaa, 0x058b2164, 0x056cefa8,
+    0x05500000, 0x05343eb1, 0x05199999, 0x05000000, 0x04e76276, 0x04cfb2b7,
+    0x04b8e38e, 0x04a2e8ba, 0x048db6db, 0x0479435e, 0x04658469, 0x045270d0,
+    0x04400000, 0x042e29f7, 0x041ce739, 0x040c30c3, 0x03fc0000, 0x03ec4ec4,
+    0x03dd1745, 0x03ce540f, 0x03c00000, 0x03b21642, 0x03a49249, 0x03976fc6,
+    0x038aaaaa, 0x037e3f1f, 0x03722983, 0x03666666, 0x035af286, 0x034fcace,
+    0x0344ec4e, 0x033a5440, 0x03300000, 0x0325ed09, 0x031c18f9, 0x0312818a,
+    0x03092492, 0x03000000, 0x02f711dc, 0x02ee5846, 0x02e5d174, 0x02dd7baf,
+    0x02d55555, 0x02cd5cd5, 0x02c590b2, 0x02bdef7b, 0x02b677d4, 0x02af286b,
+    0x02a80000, 0x02a0fd5c, 0x029a1f58, 0x029364d9, 0x028ccccc, 0x0286562d,
+    0x02800000, 0x0279c952, 0x0273b13b, 0x026db6db, 0x0267d95b, 0x026217ec,
+    0x025c71c7, 0x0256e62a, 0x0251745d, 0x024c1bac, 0x0246db6d, 0x0241b2f9,
+    0x023ca1af, 0x0237a6f4, 0x0232c234, 0x022df2df, 0x02293868, 0x02249249,
+    0x02200000, 0x021b810e, 0x021714fb, 0x0212bb51, 0x020e739c, 0x020a3d70,
+    0x02061861, 0x02020408, 0x01fe0000, 0x01fa0be8, 0x01f62762, 0x01f25213,
+    0x01ee8ba2, 0x01ead3ba, 0x01e72a07, 0x01e38e38, 0x01e00000, 0x01dc7f10,
+    0x01d90b21, 0x01d5a3e9, 0x01d24924, 0x01cefa8d, 0x01cbb7e3, 0x01c880e5,
+    0x01c55555, 0x01c234f7, 0x01bf1f8f, 0x01bc14e5, 0x01b914c1, 0x01b61eed,
+    0x01b33333, 0x01b05160, 0x01ad7943, 0x01aaaaaa, 0x01a7e567, 0x01a5294a,
+    0x01a27627, 0x019fcbd2, 0x019d2a20, 0x019a90e7, 0x01980000, 0x01957741,
+    0x0192f684, 0x01907da4, 0x018e0c7c, 0x018ba2e8, 0x018940c5, 0x0186e5f0,
+    0x01849249, 0x018245ae, 0x01800000, 0x017dc11f, 0x017b88ee, 0x0179574e,
+    0x01772c23, 0x01750750, 0x0172e8ba, 0x0170d045, 0x016ebdd7, 0x016cb157,
+    0x016aaaaa, 0x0168a9b9, 0x0166ae6a, 0x0164b8a7, 0x0162c859, 0x0160dd67,
+    0x015ef7bd, 0x015d1745, 0x015b3bea, 0x01596596, 0x01579435, 0x0155c7b4,
+    0x01540000, 0x01523d03, 0x01507eae, 0x014ec4ec, 0x014d0fac, 0x014b5edc,
+    0x0149b26c, 0x01480a4a, 0x01466666, 0x0144c6af, 0x01432b16, 0x0141938b,
+    0x01400000, 0x013e7063, 0x013ce4a9, 0x013b5cc0, 0x0139d89d, 0x01385830,
+    0x0136db6d, 0x01356246, 0x0133ecad, 0x01327a97, 0x01310bf6, 0x012fa0be,
+    0x012e38e3, 0x012cd459, 0x012b7315, 0x012a150a, 0x0128ba2e, 0x01276276,
+    0x01260dd6, 0x0124bc44, 0x01236db6, 0x01222222, 0x0120d97c, 0x011f93bc,
+    0x011e50d7, 0x011d10c4, 0x011bd37a, 0x011a98ef, 0x0119611a, 0x01182bf2,
+    0x0116f96f, 0x0115c988, 0x01149c34, 0x0113716a, 0x01124924, 0x01112358,
+    0x01100000, 0x010edf12, 0x010dc087, 0x010ca458, 0x010b8a7d, 0x010a72f0,
+    0x01095da8, 0x01084a9f, 0x010739ce, 0x01062b2e, 0x01051eb8, 0x01041465,
+    0x01030c30, 0x01020612, 0x01010204, 0x01000000 },
+  {   // alpha * KINV_255
+    0x00000000, 0x00010101, 0x00020202, 0x00030303, 0x00040404, 0x00050505,
+    0x00060606, 0x00070707, 0x00080808, 0x00090909, 0x000a0a0a, 0x000b0b0b,
+    0x000c0c0c, 0x000d0d0d, 0x000e0e0e, 0x000f0f0f, 0x00101010, 0x00111111,
+    0x00121212, 0x00131313, 0x00141414, 0x00151515, 0x00161616, 0x00171717,
+    0x00181818, 0x00191919, 0x001a1a1a, 0x001b1b1b, 0x001c1c1c, 0x001d1d1d,
+    0x001e1e1e, 0x001f1f1f, 0x00202020, 0x00212121, 0x00222222, 0x00232323,
+    0x00242424, 0x00252525, 0x00262626, 0x00272727, 0x00282828, 0x00292929,
+    0x002a2a2a, 0x002b2b2b, 0x002c2c2c, 0x002d2d2d, 0x002e2e2e, 0x002f2f2f,
+    0x00303030, 0x00313131, 0x00323232, 0x00333333, 0x00343434, 0x00353535,
+    0x00363636, 0x00373737, 0x00383838, 0x00393939, 0x003a3a3a, 0x003b3b3b,
+    0x003c3c3c, 0x003d3d3d, 0x003e3e3e, 0x003f3f3f, 0x00404040, 0x00414141,
+    0x00424242, 0x00434343, 0x00444444, 0x00454545, 0x00464646, 0x00474747,
+    0x00484848, 0x00494949, 0x004a4a4a, 0x004b4b4b, 0x004c4c4c, 0x004d4d4d,
+    0x004e4e4e, 0x004f4f4f, 0x00505050, 0x00515151, 0x00525252, 0x00535353,
+    0x00545454, 0x00555555, 0x00565656, 0x00575757, 0x00585858, 0x00595959,
+    0x005a5a5a, 0x005b5b5b, 0x005c5c5c, 0x005d5d5d, 0x005e5e5e, 0x005f5f5f,
+    0x00606060, 0x00616161, 0x00626262, 0x00636363, 0x00646464, 0x00656565,
+    0x00666666, 0x00676767, 0x00686868, 0x00696969, 0x006a6a6a, 0x006b6b6b,
+    0x006c6c6c, 0x006d6d6d, 0x006e6e6e, 0x006f6f6f, 0x00707070, 0x00717171,
+    0x00727272, 0x00737373, 0x00747474, 0x00757575, 0x00767676, 0x00777777,
+    0x00787878, 0x00797979, 0x007a7a7a, 0x007b7b7b, 0x007c7c7c, 0x007d7d7d,
+    0x007e7e7e, 0x007f7f7f, 0x00808080, 0x00818181, 0x00828282, 0x00838383,
+    0x00848484, 0x00858585, 0x00868686, 0x00878787, 0x00888888, 0x00898989,
+    0x008a8a8a, 0x008b8b8b, 0x008c8c8c, 0x008d8d8d, 0x008e8e8e, 0x008f8f8f,
+    0x00909090, 0x00919191, 0x00929292, 0x00939393, 0x00949494, 0x00959595,
+    0x00969696, 0x00979797, 0x00989898, 0x00999999, 0x009a9a9a, 0x009b9b9b,
+    0x009c9c9c, 0x009d9d9d, 0x009e9e9e, 0x009f9f9f, 0x00a0a0a0, 0x00a1a1a1,
+    0x00a2a2a2, 0x00a3a3a3, 0x00a4a4a4, 0x00a5a5a5, 0x00a6a6a6, 0x00a7a7a7,
+    0x00a8a8a8, 0x00a9a9a9, 0x00aaaaaa, 0x00ababab, 0x00acacac, 0x00adadad,
+    0x00aeaeae, 0x00afafaf, 0x00b0b0b0, 0x00b1b1b1, 0x00b2b2b2, 0x00b3b3b3,
+    0x00b4b4b4, 0x00b5b5b5, 0x00b6b6b6, 0x00b7b7b7, 0x00b8b8b8, 0x00b9b9b9,
+    0x00bababa, 0x00bbbbbb, 0x00bcbcbc, 0x00bdbdbd, 0x00bebebe, 0x00bfbfbf,
+    0x00c0c0c0, 0x00c1c1c1, 0x00c2c2c2, 0x00c3c3c3, 0x00c4c4c4, 0x00c5c5c5,
+    0x00c6c6c6, 0x00c7c7c7, 0x00c8c8c8, 0x00c9c9c9, 0x00cacaca, 0x00cbcbcb,
+    0x00cccccc, 0x00cdcdcd, 0x00cecece, 0x00cfcfcf, 0x00d0d0d0, 0x00d1d1d1,
+    0x00d2d2d2, 0x00d3d3d3, 0x00d4d4d4, 0x00d5d5d5, 0x00d6d6d6, 0x00d7d7d7,
+    0x00d8d8d8, 0x00d9d9d9, 0x00dadada, 0x00dbdbdb, 0x00dcdcdc, 0x00dddddd,
+    0x00dedede, 0x00dfdfdf, 0x00e0e0e0, 0x00e1e1e1, 0x00e2e2e2, 0x00e3e3e3,
+    0x00e4e4e4, 0x00e5e5e5, 0x00e6e6e6, 0x00e7e7e7, 0x00e8e8e8, 0x00e9e9e9,
+    0x00eaeaea, 0x00ebebeb, 0x00ececec, 0x00ededed, 0x00eeeeee, 0x00efefef,
+    0x00f0f0f0, 0x00f1f1f1, 0x00f2f2f2, 0x00f3f3f3, 0x00f4f4f4, 0x00f5f5f5,
+    0x00f6f6f6, 0x00f7f7f7, 0x00f8f8f8, 0x00f9f9f9, 0x00fafafa, 0x00fbfbfb,
+    0x00fcfcfc, 0x00fdfdfd, 0x00fefefe, 0x00ffffff }
+};
+
+static WEBP_INLINE uint32_t GetScale(uint32_t a, int inverse) {
+  return kMultTables[!inverse][a];
+}
+
+#else
+
+static WEBP_INLINE uint32_t GetScale(uint32_t a, int inverse) {
+  return inverse ? (255u << MFIX) / a : a * KINV_255;
+}
+
+#endif    // USE_TABLES_FOR_ALPHA_MULT
+
+static void MultARGBRow(uint32_t* const ptr, int width, int inverse) {
+  int x;
+  for (x = 0; x < width; ++x) {
+    const uint32_t argb = ptr[x];
+    if (argb < 0xff000000u) {      // alpha < 255
+      if (argb <= 0x00ffffffu) {   // alpha == 0
+        ptr[x] = 0;
+      } else {
+        const uint32_t alpha = (argb >> 24) & 0xff;
+        const uint32_t scale = GetScale(alpha, inverse);
+        uint32_t out = argb & 0xff000000u;
+        out |= Mult(argb >>  0, scale) <<  0;
+        out |= Mult(argb >>  8, scale) <<  8;
+        out |= Mult(argb >> 16, scale) << 16;
+        ptr[x] = out;
+      }
+    }
+  }
+}
+
+static void MultRow(uint8_t* const ptr, const uint8_t* const alpha,
+                    int width, int inverse) {
+  int x;
+  for (x = 0; x < width; ++x) {
+    const uint32_t a = alpha[x];
+    if (a != 255) {
+      if (a == 0) {
+        ptr[x] = 0;
+      } else {
+        const uint32_t scale = GetScale(a, inverse);
+        ptr[x] = Mult(ptr[x], scale);
+      }
+    }
+  }
+}
+
+#undef KINV_255
+#undef HALF
+#undef MFIX
+
+void (*WebPMultARGBRow)(uint32_t* const ptr, int width, int inverse);
+void (*WebPMultRow)(uint8_t* const ptr, const uint8_t* const alpha,
+                    int width, int inverse);
+
+//------------------------------------------------------------------------------
+// Generic per-plane calls
+
+void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows,
+                      int inverse) {
+  int n;
+  for (n = 0; n < num_rows; ++n) {
+    WebPMultARGBRow((uint32_t*)ptr, width, inverse);
+    ptr += stride;
+  }
+}
+
+void WebPMultRows(uint8_t* ptr, int stride,
+                  const uint8_t* alpha, int alpha_stride,
+                  int width, int num_rows, int inverse) {
+  int n;
+  for (n = 0; n < num_rows; ++n) {
+    WebPMultRow(ptr, alpha, width, inverse);
+    ptr += stride;
+    alpha += alpha_stride;
+  }
+}
+
+//------------------------------------------------------------------------------
+// Premultiplied modes
+
+// non dithered-modes
+
+// (x * a * 32897) >> 23 is bit-wise equivalent to (int)(x * a / 255.)
+// for all 8bit x or a. For bit-wise equivalence to (int)(x * a / 255. + .5),
+// one can use instead: (x * a * 65793 + (1 << 23)) >> 24
+#if 1     // (int)(x * a / 255.)
+#define MULTIPLIER(a)   ((a) * 32897U)
+#define PREMULTIPLY(x, m) (((x) * (m)) >> 23)
+#else     // (int)(x * a / 255. + .5)
+#define MULTIPLIER(a) ((a) * 65793U)
+#define PREMULTIPLY(x, m) (((x) * (m) + (1U << 23)) >> 24)
+#endif
+
+static void ApplyAlphaMultiply(uint8_t* rgba, int alpha_first,
+                               int w, int h, int stride) {
+  while (h-- > 0) {
+    uint8_t* const rgb = rgba + (alpha_first ? 1 : 0);
+    const uint8_t* const alpha = rgba + (alpha_first ? 0 : 3);
+    int i;
+    for (i = 0; i < w; ++i) {
+      const uint32_t a = alpha[4 * i];
+      if (a != 0xff) {
+        const uint32_t mult = MULTIPLIER(a);
+        rgb[4 * i + 0] = PREMULTIPLY(rgb[4 * i + 0], mult);
+        rgb[4 * i + 1] = PREMULTIPLY(rgb[4 * i + 1], mult);
+        rgb[4 * i + 2] = PREMULTIPLY(rgb[4 * i + 2], mult);
+      }
+    }
+    rgba += stride;
+  }
+}
+#undef MULTIPLIER
+#undef PREMULTIPLY
+
+// rgbA4444
+
+#define MULTIPLIER(a)  ((a) * 0x1111)    // 0x1111 ~= (1 << 16) / 15
+
+static WEBP_INLINE uint8_t dither_hi(uint8_t x) {
+  return (x & 0xf0) | (x >> 4);
+}
+
+static WEBP_INLINE uint8_t dither_lo(uint8_t x) {
+  return (x & 0x0f) | (x << 4);
+}
+
+static WEBP_INLINE uint8_t multiply(uint8_t x, uint32_t m) {
+  return (x * m) >> 16;
+}
+
+static WEBP_INLINE void ApplyAlphaMultiply4444(uint8_t* rgba4444,
+                                               int w, int h, int stride,
+                                               int rg_byte_pos /* 0 or 1 */) {
+  while (h-- > 0) {
+    int i;
+    for (i = 0; i < w; ++i) {
+      const uint32_t rg = rgba4444[2 * i + rg_byte_pos];
+      const uint32_t ba = rgba4444[2 * i + (rg_byte_pos ^ 1)];
+      const uint8_t a = ba & 0x0f;
+      const uint32_t mult = MULTIPLIER(a);
+      const uint8_t r = multiply(dither_hi(rg), mult);
+      const uint8_t g = multiply(dither_lo(rg), mult);
+      const uint8_t b = multiply(dither_hi(ba), mult);
+      rgba4444[2 * i + rg_byte_pos] = (r & 0xf0) | ((g >> 4) & 0x0f);
+      rgba4444[2 * i + (rg_byte_pos ^ 1)] = (b & 0xf0) | a;
+    }
+    rgba4444 += stride;
+  }
+}
+#undef MULTIPLIER
+
+static void ApplyAlphaMultiply_16b(uint8_t* rgba4444,
+                                   int w, int h, int stride) {
+#ifdef WEBP_SWAP_16BIT_CSP
+  ApplyAlphaMultiply4444(rgba4444, w, h, stride, 1);
+#else
+  ApplyAlphaMultiply4444(rgba4444, w, h, stride, 0);
+#endif
+}
+
+static int ExtractAlpha(const uint8_t* argb, int argb_stride,
+                        int width, int height,
+                        uint8_t* alpha, int alpha_stride) {
+  uint8_t alpha_mask = 0xff;
+  int i, j;
+
+  for (j = 0; j < height; ++j) {
+    for (i = 0; i < width; ++i) {
+      const uint8_t alpha_value = argb[4 * i];
+      alpha[i] = alpha_value;
+      alpha_mask &= alpha_value;
+    }
+    argb += argb_stride;
+    alpha += alpha_stride;
+  }
+  return (alpha_mask == 0xff);
+}
+
+void (*WebPApplyAlphaMultiply)(uint8_t*, int, int, int, int);
+void (*WebPApplyAlphaMultiply4444)(uint8_t*, int, int, int);
+int (*WebPExtractAlpha)(const uint8_t*, int, int, int, uint8_t*, int);
+
+//------------------------------------------------------------------------------
+// Init function
+
+extern void WebPInitAlphaProcessingSSE2(void);
+
+void WebPInitAlphaProcessing(void) {
+  WebPMultARGBRow = MultARGBRow;
+  WebPMultRow = MultRow;
+  WebPApplyAlphaMultiply = ApplyAlphaMultiply;
+  WebPApplyAlphaMultiply4444 = ApplyAlphaMultiply_16b;
+  WebPExtractAlpha = ExtractAlpha;
+
+  // If defined, use CPUInfo() to overwrite some pointers with faster versions.
+  if (VP8GetCPUInfo != NULL) {
+#if defined(WEBP_USE_SSE2)
+    if (VP8GetCPUInfo(kSSE2)) {
+      WebPInitAlphaProcessingSSE2();
+    }
+#endif
+  }
+}

src/dsp/alpha_processing_sse2.c

@@ -0,0 +1,77 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Utilities for processing transparent channel.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_SSE2)
+#include <emmintrin.h>
+
+//------------------------------------------------------------------------------
+
+static int ExtractAlpha(const uint8_t* argb, int argb_stride,
+                        int width, int height,
+                        uint8_t* alpha, int alpha_stride) {
+  // alpha_and stores an 'and' operation of all the alpha[] values. The final
+  // value is not 0xff if any of the alpha[] is not equal to 0xff.
+  uint32_t alpha_and = 0xff;
+  int i, j;
+  const __m128i a_mask = _mm_set1_epi32(0xffu);  // to preserve alpha
+  const __m128i all_0xff = _mm_set_epi32(0, 0, ~0u, ~0u);
+  __m128i all_alphas = all_0xff;
+
+  // We must be able to access 3 extra bytes after the last written byte
+  // 'src[4 * width - 4]', because we don't know if alpha is the first or the
+  // last byte of the quadruplet.
+  const int limit = (width - 1) & ~7;
+
+  for (j = 0; j < height; ++j) {
+    const __m128i* src = (const __m128i*)argb;
+    for (i = 0; i < limit; i += 8) {
+      // load 32 argb bytes
+      const __m128i a0 = _mm_loadu_si128(src + 0);
+      const __m128i a1 = _mm_loadu_si128(src + 1);
+      const __m128i b0 = _mm_and_si128(a0, a_mask);
+      const __m128i b1 = _mm_and_si128(a1, a_mask);
+      const __m128i c0 = _mm_packs_epi32(b0, b1);
+      const __m128i d0 = _mm_packus_epi16(c0, c0);
+      // store
+      _mm_storel_epi64((__m128i*)&alpha[i], d0);
+      // accumulate eight alpha 'and' in parallel
+      all_alphas = _mm_and_si128(all_alphas, d0);
+      src += 2;
+    }
+    for (; i < width; ++i) {
+      const uint32_t alpha_value = argb[4 * i];
+      alpha[i] = alpha_value;
+      alpha_and &= alpha_value;
+    }
+    argb += argb_stride;
+    alpha += alpha_stride;
+  }
+  // Combine the eight alpha 'and' into a 8-bit mask.
+  alpha_and &= _mm_movemask_epi8(_mm_cmpeq_epi8(all_alphas, all_0xff));
+  return (alpha_and == 0xff);
+}
+
+#endif   // WEBP_USE_SSE2
+
+//------------------------------------------------------------------------------
+// Init function
+
+extern void WebPInitAlphaProcessingSSE2(void);
+
+void WebPInitAlphaProcessingSSE2(void) {
+#if defined(WEBP_USE_SSE2)
+  WebPExtractAlpha = ExtractAlpha;
+#endif
+}

src/dsp/cpu.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // CPU detection
@@ -15,10 +17,6 @@
 #include <cpu-features.h>
 #endif
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 //------------------------------------------------------------------------------
 // SSE2 detection.
 //
@@ -40,10 +38,41 @@ static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
     : "=a"(cpu_info[0]), "=b"(cpu_info[1]), "=c"(cpu_info[2]), "=d"(cpu_info[3])
     : "a"(info_type));
 }
+#elif defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 150030729  // >= VS2008 SP1
+#define GetCPUInfo(info, type) __cpuidex(info, type, 0)  // set ecx=0
 #elif defined(WEBP_MSC_SSE2)
 #define GetCPUInfo __cpuid
 #endif
 
+// NaCl has no support for xgetbv or the raw opcode.
+#if !defined(__native_client__) && (defined(__i386__) || defined(__x86_64__))
+static WEBP_INLINE uint64_t xgetbv(void) {
+  const uint32_t ecx = 0;
+  uint32_t eax, edx;
+  // Use the raw opcode for xgetbv for compatibility with older toolchains.
+  __asm__ volatile (
+    ".byte 0x0f, 0x01, 0xd0\n"
+    : "=a"(eax), "=d"(edx) : "c" (ecx));
+  return ((uint64_t)edx << 32) | eax;
+}
+#elif defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 160040219  // >= VS2010 SP1
+#define xgetbv() _xgetbv(0)
+#elif defined(_MSC_VER) && defined(_M_IX86)
+static WEBP_INLINE uint64_t xgetbv(void) {
+  uint32_t eax_, edx_;
+  __asm {
+    xor ecx, ecx  // ecx = 0
+    // Use the raw opcode for xgetbv for compatibility with older toolchains.
+    __asm _emit 0x0f __asm _emit 0x01 __asm _emit 0xd0
+    mov eax_, eax
+    mov edx_, edx
+  }
+  return ((uint64_t)edx_ << 32) | eax_;
+}
+#else
+#define xgetbv() 0U  // no AVX for older x64 or unrecognized toolchains.
+#endif
+
 #if defined(__i386__) || defined(__x86_64__) || defined(WEBP_MSC_SSE2)
 static int x86CPUInfo(CPUFeature feature) {
   int cpu_info[4];
@@ -54,10 +83,23 @@ static int x86CPUInfo(CPUFeature feature) {
   if (feature == kSSE3) {
     return 0 != (cpu_info[2] & 0x00000001);
   }
+  if (feature == kAVX) {
+    // bits 27 (OSXSAVE) & 28 (256-bit AVX)
+    if ((cpu_info[2] & 0x18000000) == 0x18000000) {
+      // XMM state and YMM state enabled by the OS.
+      return (xgetbv() & 0x6) == 0x6;
+    }
+  }
+  if (feature == kAVX2) {
+    if (x86CPUInfo(kAVX)) {
+      GetCPUInfo(cpu_info, 7);
+      return ((cpu_info[1] & 0x00000020) == 0x00000020);
+    }
+  }
   return 0;
 }
 VP8CPUInfo VP8GetCPUInfo = x86CPUInfo;
-#elif defined(WEBP_ANDROID_NEON)
+#elif defined(WEBP_ANDROID_NEON)  // NB: needs to be before generic NEON test.
 static int AndroidCPUInfo(CPUFeature feature) {
   const AndroidCpuFamily cpu_family = android_getCpuFamily();
   const uint64_t cpu_features = android_getCpuFeatures();
@@ -68,7 +110,7 @@ static int AndroidCPUInfo(CPUFeature feature) {
   return 0;
 }
 VP8CPUInfo VP8GetCPUInfo = AndroidCPUInfo;
-#elif defined(__ARM_NEON__)
+#elif defined(WEBP_USE_NEON)
 // define a dummy function to enable turning off NEON at runtime by setting
 // VP8DecGetCPUInfo = NULL
 static int armCPUInfo(CPUFeature feature) {
@@ -76,10 +118,13 @@ static int armCPUInfo(CPUFeature feature) {
   return 1;
 }
 VP8CPUInfo VP8GetCPUInfo = armCPUInfo;
+#elif defined(WEBP_USE_MIPS32)
+static int mipsCPUInfo(CPUFeature feature) {
+  (void)feature;
+  return 1;
+}
+VP8CPUInfo VP8GetCPUInfo = mipsCPUInfo;
 #else
 VP8CPUInfo VP8GetCPUInfo = NULL;
 #endif
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dsp/dec.c

@@ -1,8 +1,10 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Speed-critical decoding functions.
@@ -12,42 +14,7 @@
 #include "./dsp.h"
 #include "../dec/vp8i.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 //------------------------------------------------------------------------------
-// run-time tables (~4k)
-
-static uint8_t abs0[255 + 255 + 1];     // abs(i)
-static uint8_t abs1[255 + 255 + 1];     // abs(i)>>1
-static int8_t sclip1[1020 + 1020 + 1];  // clips [-1020, 1020] to [-128, 127]
-static int8_t sclip2[112 + 112 + 1];    // clips [-112, 112] to [-16, 15]
-static uint8_t clip1[255 + 510 + 1];    // clips [-255,510] to [0,255]
-
-// We declare this variable 'volatile' to prevent instruction reordering
-// and make sure it's set to true _last_ (so as to be thread-safe)
-static volatile int tables_ok = 0;
-
-static void DspInitTables(void) {
-  if (!tables_ok) {
-    int i;
-    for (i = -255; i <= 255; ++i) {
-      abs0[255 + i] = (i < 0) ? -i : i;
-      abs1[255 + i] = abs0[255 + i] >> 1;
-    }
-    for (i = -1020; i <= 1020; ++i) {
-      sclip1[1020 + i] = (i < -128) ? -128 : (i > 127) ? 127 : i;
-    }
-    for (i = -112; i <= 112; ++i) {
-      sclip2[112 + i] = (i < -16) ? -16 : (i > 15) ? 15 : i;
-    }
-    for (i = -255; i <= 255 + 255; ++i) {
-      clip1[255 + i] = (i < 0) ? 0 : (i > 255) ? 255 : i;
-    }
-    tables_ok = 1;
-  }
-}
 
 static WEBP_INLINE uint8_t clip_8b(int v) {
   return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
@@ -59,6 +26,14 @@ static WEBP_INLINE uint8_t clip_8b(int v) {
 #define STORE(x, y, v) \
   dst[x + y * BPS] = clip_8b(dst[x + y * BPS] + ((v) >> 3))
 
+#define STORE2(y, dc, d, c) do {    \
+  const int DC = (dc);              \
+  STORE(0, y, DC + (d));            \
+  STORE(1, y, DC + (c));            \
+  STORE(2, y, DC - (c));            \
+  STORE(3, y, DC - (d));            \
+} while (0)
+
 static const int kC1 = 20091 + (1 << 16);
 static const int kC2 = 35468;
 #define MUL(a, b) (((a) * (b)) >> 16)
@@ -101,7 +76,21 @@ static void TransformOne(const int16_t* in, uint8_t* dst) {
     dst += BPS;
   }
 }
+
+// Simplified transform when only in[0], in[1] and in[4] are non-zero
+static void TransformAC3(const int16_t* in, uint8_t* dst) {
+  const int a = in[0] + 4;
+  const int c4 = MUL(in[4], kC2);
+  const int d4 = MUL(in[4], kC1);
+  const int c1 = MUL(in[1], kC2);
+  const int d1 = MUL(in[1], kC1);
+  STORE2(0, a + d4, d1, c1);
+  STORE2(1, a + c4, d1, c1);
+  STORE2(2, a - c4, d1, c1);
+  STORE2(3, a - d4, d1, c1);
+}
 #undef MUL
+#undef STORE2
 
 static void TransformTwo(const int16_t* in, uint8_t* dst, int do_two) {
   TransformOne(in, dst);
@@ -126,10 +115,10 @@ static void TransformDC(const int16_t *in, uint8_t* dst) {
 }
 
 static void TransformDCUV(const int16_t* in, uint8_t* dst) {
-  if (in[0 * 16]) TransformDC(in + 0 * 16, dst);
-  if (in[1 * 16]) TransformDC(in + 1 * 16, dst + 4);
-  if (in[2 * 16]) TransformDC(in + 2 * 16, dst + 4 * BPS);
-  if (in[3 * 16]) TransformDC(in + 3 * 16, dst + 4 * BPS + 4);
+  if (in[0 * 16]) VP8TransformDC(in + 0 * 16, dst);
+  if (in[1 * 16]) VP8TransformDC(in + 1 * 16, dst + 4);
+  if (in[2 * 16]) VP8TransformDC(in + 2 * 16, dst + 4 * BPS);
+  if (in[3 * 16]) VP8TransformDC(in + 3 * 16, dst + 4 * BPS + 4);
 }
 
 #undef STORE
@@ -164,7 +153,7 @@ static void TransformWHT(const int16_t* in, int16_t* out) {
   }
 }
 
-void (*VP8TransformWHT)(const int16_t* in, int16_t* out) = TransformWHT;
+void (*VP8TransformWHT)(const int16_t* in, int16_t* out);
 
 //------------------------------------------------------------------------------
 // Intra predictions
@@ -173,7 +162,7 @@ void (*VP8TransformWHT)(const int16_t* in, int16_t* out) = TransformWHT;
 
 static WEBP_INLINE void TrueMotion(uint8_t *dst, int size) {
   const uint8_t* top = dst - BPS;
-  const uint8_t* const clip0 = clip1 + 255 - top[-1];
+  const uint8_t* const clip0 = VP8kclip1 - top[-1];
   int y;
   for (y = 0; y < size; ++y) {
     const uint8_t* const clip = clip0 + dst[-1];
@@ -426,10 +415,10 @@ static void HE8uv(uint8_t *dst) {    // horizontal
 }
 
 // helper for chroma-DC predictions
-static WEBP_INLINE void Put8x8uv(uint64_t v, uint8_t* dst) {
+static WEBP_INLINE void Put8x8uv(uint8_t value, uint8_t* dst) {
   int j;
   for (j = 0; j < 8; ++j) {
-    *(uint64_t*)(dst + j * BPS) = v;
+    memset(dst + j * BPS, value, 8);
   }
 }
 
@@ -439,7 +428,7 @@ static void DC8uv(uint8_t *dst) {     // DC
   for (i = 0; i < 8; ++i) {
     dc0 += dst[i - BPS] + dst[-1 + i * BPS];
   }
-  Put8x8uv((uint64_t)((dc0 >> 4) * 0x0101010101010101ULL), dst);
+  Put8x8uv(dc0 >> 4, dst);
 }
 
 static void DC8uvNoLeft(uint8_t *dst) {   // DC with no left samples
@@ -448,7 +437,7 @@ static void DC8uvNoLeft(uint8_t *dst) {   // DC with no left samples
   for (i = 0; i < 8; ++i) {
     dc0 += dst[i - BPS];
   }
-  Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst);
+  Put8x8uv(dc0 >> 3, dst);
 }
 
 static void DC8uvNoTop(uint8_t *dst) {  // DC with no top samples
@@ -457,11 +446,11 @@ static void DC8uvNoTop(uint8_t *dst) {  // DC with no top samples
   for (i = 0; i < 8; ++i) {
     dc0 += dst[-1 + i * BPS];
   }
-  Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst);
+  Put8x8uv(dc0 >> 3, dst);
 }
 
 static void DC8uvNoTopLeft(uint8_t *dst) {    // DC with nothing
-  Put8x8uv(0x8080808080808080ULL, dst);
+  Put8x8uv(0x80, dst);
 }
 
 //------------------------------------------------------------------------------
@@ -487,61 +476,62 @@ const VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES] = {
 // 4 pixels in, 2 pixels out
 static WEBP_INLINE void do_filter2(uint8_t* p, int step) {
   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
-  const int a = 3 * (q0 - p0) + sclip1[1020 + p1 - q1];
-  const int a1 = sclip2[112 + ((a + 4) >> 3)];
-  const int a2 = sclip2[112 + ((a + 3) >> 3)];
-  p[-step] = clip1[255 + p0 + a2];
-  p[    0] = clip1[255 + q0 - a1];
+  const int a = 3 * (q0 - p0) + VP8ksclip1[p1 - q1];  // in [-893,892]
+  const int a1 = VP8ksclip2[(a + 4) >> 3];            // in [-16,15]
+  const int a2 = VP8ksclip2[(a + 3) >> 3];
+  p[-step] = VP8kclip1[p0 + a2];
+  p[    0] = VP8kclip1[q0 - a1];
 }
 
 // 4 pixels in, 4 pixels out
 static WEBP_INLINE void do_filter4(uint8_t* p, int step) {
   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
   const int a = 3 * (q0 - p0);
-  const int a1 = sclip2[112 + ((a + 4) >> 3)];
-  const int a2 = sclip2[112 + ((a + 3) >> 3)];
+  const int a1 = VP8ksclip2[(a + 4) >> 3];
+  const int a2 = VP8ksclip2[(a + 3) >> 3];
   const int a3 = (a1 + 1) >> 1;
-  p[-2*step] = clip1[255 + p1 + a3];
-  p[-  step] = clip1[255 + p0 + a2];
-  p[      0] = clip1[255 + q0 - a1];
-  p[   step] = clip1[255 + q1 - a3];
+  p[-2*step] = VP8kclip1[p1 + a3];
+  p[-  step] = VP8kclip1[p0 + a2];
+  p[      0] = VP8kclip1[q0 - a1];
+  p[   step] = VP8kclip1[q1 - a3];
 }
 
 // 6 pixels in, 6 pixels out
 static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
   const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
   const int q0 = p[0], q1 = p[step], q2 = p[2*step];
-  const int a = sclip1[1020 + 3 * (q0 - p0) + sclip1[1020 + p1 - q1]];
+  const int a = VP8ksclip1[3 * (q0 - p0) + VP8ksclip1[p1 - q1]];
+  // a is in [-128,127], a1 in [-27,27], a2 in [-18,18] and a3 in [-9,9]
   const int a1 = (27 * a + 63) >> 7;  // eq. to ((3 * a + 7) * 9) >> 7
   const int a2 = (18 * a + 63) >> 7;  // eq. to ((2 * a + 7) * 9) >> 7
   const int a3 = (9  * a + 63) >> 7;  // eq. to ((1 * a + 7) * 9) >> 7
-  p[-3*step] = clip1[255 + p2 + a3];
-  p[-2*step] = clip1[255 + p1 + a2];
-  p[-  step] = clip1[255 + p0 + a1];
-  p[      0] = clip1[255 + q0 - a1];
-  p[   step] = clip1[255 + q1 - a2];
-  p[ 2*step] = clip1[255 + q2 - a3];
+  p[-3*step] = VP8kclip1[p2 + a3];
+  p[-2*step] = VP8kclip1[p1 + a2];
+  p[-  step] = VP8kclip1[p0 + a1];
+  p[      0] = VP8kclip1[q0 - a1];
+  p[   step] = VP8kclip1[q1 - a2];
+  p[ 2*step] = VP8kclip1[q2 - a3];
 }
 
 static WEBP_INLINE int hev(const uint8_t* p, int step, int thresh) {
   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
-  return (abs0[255 + p1 - p0] > thresh) || (abs0[255 + q1 - q0] > thresh);
+  return (VP8kabs0[p1 - p0] > thresh) || (VP8kabs0[q1 - q0] > thresh);
 }
 
-static WEBP_INLINE int needs_filter(const uint8_t* p, int step, int thresh) {
-  const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
-  return (2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) <= thresh;
+static WEBP_INLINE int needs_filter(const uint8_t* p, int step, int t) {
+  const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
+  return ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) <= t);
 }
 
 static WEBP_INLINE int needs_filter2(const uint8_t* p,
                                      int step, int t, int it) {
-  const int p3 = p[-4*step], p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
-  const int q0 = p[0], q1 = p[step], q2 = p[2*step], q3 = p[3*step];
-  if ((2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) > t)
-    return 0;
-  return abs0[255 + p3 - p2] <= it && abs0[255 + p2 - p1] <= it &&
-         abs0[255 + p1 - p0] <= it && abs0[255 + q3 - q2] <= it &&
-         abs0[255 + q2 - q1] <= it && abs0[255 + q1 - q0] <= it;
+  const int p3 = p[-4 * step], p2 = p[-3 * step], p1 = p[-2 * step];
+  const int p0 = p[-step], q0 = p[0];
+  const int q1 = p[step], q2 = p[2 * step], q3 = p[3 * step];
+  if ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) > t) return 0;
+  return VP8kabs0[p3 - p2] <= it && VP8kabs0[p2 - p1] <= it &&
+         VP8kabs0[p1 - p0] <= it && VP8kabs0[q3 - q2] <= it &&
+         VP8kabs0[q2 - q1] <= it && VP8kabs0[q1 - q0] <= it;
 }
 
 //------------------------------------------------------------------------------
@@ -549,8 +539,9 @@ static WEBP_INLINE int needs_filter2(const uint8_t* p,
 
 static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
   int i;
+  const int thresh2 = 2 * thresh + 1;
   for (i = 0; i < 16; ++i) {
-    if (needs_filter(p + i, stride, thresh)) {
+    if (needs_filter(p + i, stride, thresh2)) {
       do_filter2(p + i, stride);
     }
   }
@@ -558,8 +549,9 @@ static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
 
 static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
   int i;
+  const int thresh2 = 2 * thresh + 1;
   for (i = 0; i < 16; ++i) {
-    if (needs_filter(p + i * stride, 1, thresh)) {
+    if (needs_filter(p + i * stride, 1, thresh2)) {
       do_filter2(p + i * stride, 1);
     }
   }
@@ -587,8 +579,9 @@ static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
 static WEBP_INLINE void FilterLoop26(uint8_t* p,
                                      int hstride, int vstride, int size,
                                      int thresh, int ithresh, int hev_thresh) {
+  const int thresh2 = 2 * thresh + 1;
   while (size-- > 0) {
-    if (needs_filter2(p, hstride, thresh, ithresh)) {
+    if (needs_filter2(p, hstride, thresh2, ithresh)) {
       if (hev(p, hstride, hev_thresh)) {
         do_filter2(p, hstride);
       } else {
@@ -602,8 +595,9 @@ static WEBP_INLINE void FilterLoop26(uint8_t* p,
 static WEBP_INLINE void FilterLoop24(uint8_t* p,
                                      int hstride, int vstride, int size,
                                      int thresh, int ithresh, int hev_thresh) {
+  const int thresh2 = 2 * thresh + 1;
   while (size-- > 0) {
-    if (needs_filter2(p, hstride, thresh, ithresh)) {
+    if (needs_filter2(p, hstride, thresh2, ithresh)) {
       if (hev(p, hstride, hev_thresh)) {
         do_filter2(p, hstride);
       } else {
@@ -672,6 +666,7 @@ static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
 //------------------------------------------------------------------------------
 
 VP8DecIdct2 VP8Transform;
+VP8DecIdct VP8TransformAC3;
 VP8DecIdct VP8TransformUV;
 VP8DecIdct VP8TransformDC;
 VP8DecIdct VP8TransformDCUV;
@@ -691,14 +686,17 @@ VP8SimpleFilterFunc VP8SimpleHFilter16i;
 
 extern void VP8DspInitSSE2(void);
 extern void VP8DspInitNEON(void);
+extern void VP8DspInitMIPS32(void);
 
 void VP8DspInit(void) {
-  DspInitTables();
+  VP8InitClipTables();
 
+  VP8TransformWHT = TransformWHT;
   VP8Transform = TransformTwo;
   VP8TransformUV = TransformUV;
   VP8TransformDC = TransformDC;
   VP8TransformDCUV = TransformDCUV;
+  VP8TransformAC3 = TransformAC3;
 
   VP8VFilter16 = VFilter16;
   VP8HFilter16 = HFilter16;
@@ -714,7 +712,7 @@ void VP8DspInit(void) {
   VP8SimpleHFilter16i = SimpleHFilter16i;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
-  if (VP8GetCPUInfo) {
+  if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8DspInitSSE2();
@@ -723,10 +721,11 @@ void VP8DspInit(void) {
     if (VP8GetCPUInfo(kNEON)) {
       VP8DspInitNEON();
     }
+#elif defined(WEBP_USE_MIPS32)
+    if (VP8GetCPUInfo(kMIPS32)) {
+      VP8DspInitMIPS32();
+    }
 #endif
   }
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dsp/dec_clip_tables.c

@@ -0,0 +1,366 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Clipping tables for filtering
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./dsp.h"
+
+#define USE_STATIC_TABLES     // undefine to have run-time table initialization
+
+#ifdef USE_STATIC_TABLES
+
+static const uint8_t abs0[255 + 255 + 1] = {
+  0xff, 0xfe, 0xfd, 0xfc, 0xfb, 0xfa, 0xf9, 0xf8, 0xf7, 0xf6, 0xf5, 0xf4,
+  0xf3, 0xf2, 0xf1, 0xf0, 0xef, 0xee, 0xed, 0xec, 0xeb, 0xea, 0xe9, 0xe8,
+  0xe7, 0xe6, 0xe5, 0xe4, 0xe3, 0xe2, 0xe1, 0xe0, 0xdf, 0xde, 0xdd, 0xdc,
+  0xdb, 0xda, 0xd9, 0xd8, 0xd7, 0xd6, 0xd5, 0xd4, 0xd3, 0xd2, 0xd1, 0xd0,
+  0xcf, 0xce, 0xcd, 0xcc, 0xcb, 0xca, 0xc9, 0xc8, 0xc7, 0xc6, 0xc5, 0xc4,
+  0xc3, 0xc2, 0xc1, 0xc0, 0xbf, 0xbe, 0xbd, 0xbc, 0xbb, 0xba, 0xb9, 0xb8,
+  0xb7, 0xb6, 0xb5, 0xb4, 0xb3, 0xb2, 0xb1, 0xb0, 0xaf, 0xae, 0xad, 0xac,
+  0xab, 0xaa, 0xa9, 0xa8, 0xa7, 0xa6, 0xa5, 0xa4, 0xa3, 0xa2, 0xa1, 0xa0,
+  0x9f, 0x9e, 0x9d, 0x9c, 0x9b, 0x9a, 0x99, 0x98, 0x97, 0x96, 0x95, 0x94,
+  0x93, 0x92, 0x91, 0x90, 0x8f, 0x8e, 0x8d, 0x8c, 0x8b, 0x8a, 0x89, 0x88,
+  0x87, 0x86, 0x85, 0x84, 0x83, 0x82, 0x81, 0x80, 0x7f, 0x7e, 0x7d, 0x7c,
+  0x7b, 0x7a, 0x79, 0x78, 0x77, 0x76, 0x75, 0x74, 0x73, 0x72, 0x71, 0x70,
+  0x6f, 0x6e, 0x6d, 0x6c, 0x6b, 0x6a, 0x69, 0x68, 0x67, 0x66, 0x65, 0x64,
+  0x63, 0x62, 0x61, 0x60, 0x5f, 0x5e, 0x5d, 0x5c, 0x5b, 0x5a, 0x59, 0x58,
+  0x57, 0x56, 0x55, 0x54, 0x53, 0x52, 0x51, 0x50, 0x4f, 0x4e, 0x4d, 0x4c,
+  0x4b, 0x4a, 0x49, 0x48, 0x47, 0x46, 0x45, 0x44, 0x43, 0x42, 0x41, 0x40,
+  0x3f, 0x3e, 0x3d, 0x3c, 0x3b, 0x3a, 0x39, 0x38, 0x37, 0x36, 0x35, 0x34,
+  0x33, 0x32, 0x31, 0x30, 0x2f, 0x2e, 0x2d, 0x2c, 0x2b, 0x2a, 0x29, 0x28,
+  0x27, 0x26, 0x25, 0x24, 0x23, 0x22, 0x21, 0x20, 0x1f, 0x1e, 0x1d, 0x1c,
+  0x1b, 0x1a, 0x19, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10,
+  0x0f, 0x0e, 0x0d, 0x0c, 0x0b, 0x0a, 0x09, 0x08, 0x07, 0x06, 0x05, 0x04,
+  0x03, 0x02, 0x01, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+  0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, 0x11, 0x12, 0x13, 0x14,
+  0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20,
+  0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c,
+  0x2d, 0x2e, 0x2f, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38,
+  0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f, 0x40, 0x41, 0x42, 0x43, 0x44,
+  0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0x50,
+  0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5a, 0x5b, 0x5c,
+  0x5d, 0x5e, 0x5f, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,
+  0x69, 0x6a, 0x6b, 0x6c, 0x6d, 0x6e, 0x6f, 0x70, 0x71, 0x72, 0x73, 0x74,
+  0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f, 0x80,
+  0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 0x8a, 0x8b, 0x8c,
+  0x8d, 0x8e, 0x8f, 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98,
+  0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f, 0xa0, 0xa1, 0xa2, 0xa3, 0xa4,
+  0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, 0xaf, 0xb0,
+  0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xbb, 0xbc,
+  0xbd, 0xbe, 0xbf, 0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7, 0xc8,
+  0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, 0xcf, 0xd0, 0xd1, 0xd2, 0xd3, 0xd4,
+  0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf, 0xe0,
+  0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec,
+  0xed, 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8,
+  0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff
+};
+
+static const int8_t sclip1[1020 + 1020 + 1] = {
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+  0x80, 0x80, 0x80, 0x80, 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87,
+  0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f, 0x90, 0x91, 0x92, 0x93,
+  0x94, 0x95, 0x96, 0x97, 0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f,
+  0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xab,
+  0xac, 0xad, 0xae, 0xaf, 0xb0, 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7,
+  0xb8, 0xb9, 0xba, 0xbb, 0xbc, 0xbd, 0xbe, 0xbf, 0xc0, 0xc1, 0xc2, 0xc3,
+  0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, 0xcf,
+  0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb,
+  0xdc, 0xdd, 0xde, 0xdf, 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7,
+  0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3,
+  0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff,
+  0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0a, 0x0b,
+  0x0c, 0x0d, 0x0e, 0x0f, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
+  0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20, 0x21, 0x22, 0x23,
+  0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f,
+  0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3a, 0x3b,
+  0x3c, 0x3d, 0x3e, 0x3f, 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47,
+  0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0x50, 0x51, 0x52, 0x53,
+  0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5a, 0x5b, 0x5c, 0x5d, 0x5e, 0x5f,
+  0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6a, 0x6b,
+  0x6c, 0x6d, 0x6e, 0x6f, 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77,
+  0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f,
+  0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f
+};
+
+static const int8_t sclip2[112 + 112 + 1] = {
+  0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0,
+  0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0,
+  0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0,
+  0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0,
+  0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0,
+  0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0,
+  0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0,
+  0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0,
+  0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb,
+  0xfc, 0xfd, 0xfe, 0xff, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+  0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f,
+  0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f,
+  0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f,
+  0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f,
+  0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f,
+  0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f,
+  0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f,
+  0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f,
+  0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f
+};
+
+static const uint8_t clip1[255 + 511 + 1] = {
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+  0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, 0x11, 0x12, 0x13, 0x14,
+  0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20,
+  0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c,
+  0x2d, 0x2e, 0x2f, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38,
+  0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f, 0x40, 0x41, 0x42, 0x43, 0x44,
+  0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0x50,
+  0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5a, 0x5b, 0x5c,
+  0x5d, 0x5e, 0x5f, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,
+  0x69, 0x6a, 0x6b, 0x6c, 0x6d, 0x6e, 0x6f, 0x70, 0x71, 0x72, 0x73, 0x74,
+  0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f, 0x80,
+  0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 0x8a, 0x8b, 0x8c,
+  0x8d, 0x8e, 0x8f, 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98,
+  0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f, 0xa0, 0xa1, 0xa2, 0xa3, 0xa4,
+  0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, 0xaf, 0xb0,
+  0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xbb, 0xbc,
+  0xbd, 0xbe, 0xbf, 0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7, 0xc8,
+  0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, 0xcf, 0xd0, 0xd1, 0xd2, 0xd3, 0xd4,
+  0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf, 0xe0,
+  0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec,
+  0xed, 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8,
+  0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+  0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
+};
+
+#else
+
+// uninitialized tables
+static uint8_t abs0[255 + 255 + 1];
+static int8_t sclip1[1020 + 1020 + 1];
+static int8_t sclip2[112 + 112 + 1];
+static uint8_t clip1[255 + 511 + 1];
+
+// We declare this variable 'volatile' to prevent instruction reordering
+// and make sure it's set to true _last_ (so as to be thread-safe)
+static volatile int tables_ok = 0;
+
+#endif
+
+const int8_t* const VP8ksclip1 = &sclip1[1020];
+const int8_t* const VP8ksclip2 = &sclip2[112];
+const uint8_t* const VP8kclip1 = &clip1[255];
+const uint8_t* const VP8kabs0 = &abs0[255];
+
+void VP8InitClipTables(void) {
+#if !defined(USE_STATIC_TABLES)
+  int i;
+  if (!tables_ok) {
+    for (i = -255; i <= 255; ++i) {
+      abs0[255 + i] = (i < 0) ? -i : i;
+    }
+    for (i = -1020; i <= 1020; ++i) {
+      sclip1[1020 + i] = (i < -128) ? -128 : (i > 127) ? 127 : i;
+    }
+    for (i = -112; i <= 112; ++i) {
+      sclip2[112 + i] = (i < -16) ? -16 : (i > 15) ? 15 : i;
+    }
+    for (i = -255; i <= 255 + 255; ++i) {
+      clip1[255 + i] = (i < 0) ? 0 : (i > 255) ? 255 : i;
+    }
+    tables_ok = 1;
+  }
+#endif    // USE_STATIC_TABLES
+}

src/dsp/dec_mips32.c

@@ -0,0 +1,578 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// MIPS version of dsp functions
+//
+// Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
+//             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_MIPS32)
+
+static const int kC1 = 20091 + (1 << 16);
+static const int kC2 = 35468;
+
+static WEBP_INLINE int abs_mips32(int x) {
+  const int sign = x >> 31;
+  return (x ^ sign) - sign;
+}
+
+// 4 pixels in, 2 pixels out
+static WEBP_INLINE void do_filter2(uint8_t* p, int step) {
+  const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
+  const int a = 3 * (q0 - p0) + VP8ksclip1[p1 - q1];
+  const int a1 = VP8ksclip2[(a + 4) >> 3];
+  const int a2 = VP8ksclip2[(a + 3) >> 3];
+  p[-step] = VP8kclip1[p0 + a2];
+  p[    0] = VP8kclip1[q0 - a1];
+}
+
+// 4 pixels in, 4 pixels out
+static WEBP_INLINE void do_filter4(uint8_t* p, int step) {
+  const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
+  const int a = 3 * (q0 - p0);
+  const int a1 = VP8ksclip2[(a + 4) >> 3];
+  const int a2 = VP8ksclip2[(a + 3) >> 3];
+  const int a3 = (a1 + 1) >> 1;
+  p[-2 * step] = VP8kclip1[p1 + a3];
+  p[-    step] = VP8kclip1[p0 + a2];
+  p[        0] = VP8kclip1[q0 - a1];
+  p[     step] = VP8kclip1[q1 - a3];
+}
+
+// 6 pixels in, 6 pixels out
+static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
+  const int p2 = p[-3 * step], p1 = p[-2 * step], p0 = p[-step];
+  const int q0 = p[0], q1 = p[step], q2 = p[2 * step];
+  const int a = VP8ksclip1[3 * (q0 - p0) + VP8ksclip1[p1 - q1]];
+  const int a1 = (27 * a + 63) >> 7;  // eq. to ((3 * a + 7) * 9) >> 7
+  const int a2 = (18 * a + 63) >> 7;  // eq. to ((2 * a + 7) * 9) >> 7
+  const int a3 = (9  * a + 63) >> 7;  // eq. to ((1 * a + 7) * 9) >> 7
+  p[-3 * step] = VP8kclip1[p2 + a3];
+  p[-2 * step] = VP8kclip1[p1 + a2];
+  p[-    step] = VP8kclip1[p0 + a1];
+  p[        0] = VP8kclip1[q0 - a1];
+  p[     step] = VP8kclip1[q1 - a2];
+  p[ 2 * step] = VP8kclip1[q2 - a3];
+}
+
+static WEBP_INLINE int hev(const uint8_t* p, int step, int thresh) {
+  const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
+  return (abs_mips32(p1 - p0) > thresh) || (abs_mips32(q1 - q0) > thresh);
+}
+
+static WEBP_INLINE int needs_filter(const uint8_t* p, int step, int thresh) {
+  const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
+  return ((2 * abs_mips32(p0 - q0) + (abs_mips32(p1 - q1) >> 1)) <= thresh);
+}
+
+static WEBP_INLINE int needs_filter2(const uint8_t* p,
+                                     int step, int t, int it) {
+  const int p3 = p[-4 * step], p2 = p[-3 * step];
+  const int p1 = p[-2 * step], p0 = p[-step];
+  const int q0 = p[0], q1 = p[step], q2 = p[2 * step], q3 = p[3 * step];
+  if ((2 * abs_mips32(p0 - q0) + (abs_mips32(p1 - q1) >> 1)) > t) {
+    return 0;
+  }
+  return abs_mips32(p3 - p2) <= it && abs_mips32(p2 - p1) <= it &&
+         abs_mips32(p1 - p0) <= it && abs_mips32(q3 - q2) <= it &&
+         abs_mips32(q2 - q1) <= it && abs_mips32(q1 - q0) <= it;
+}
+
+static WEBP_INLINE void FilterLoop26(uint8_t* p,
+                                     int hstride, int vstride, int size,
+                                     int thresh, int ithresh, int hev_thresh) {
+  while (size-- > 0) {
+    if (needs_filter2(p, hstride, thresh, ithresh)) {
+      if (hev(p, hstride, hev_thresh)) {
+        do_filter2(p, hstride);
+      } else {
+        do_filter6(p, hstride);
+      }
+    }
+    p += vstride;
+  }
+}
+
+static WEBP_INLINE void FilterLoop24(uint8_t* p,
+                                     int hstride, int vstride, int size,
+                                     int thresh, int ithresh, int hev_thresh) {
+  while (size-- > 0) {
+    if (needs_filter2(p, hstride, thresh, ithresh)) {
+      if (hev(p, hstride, hev_thresh)) {
+        do_filter2(p, hstride);
+      } else {
+        do_filter4(p, hstride);
+      }
+    }
+    p += vstride;
+  }
+}
+
+// on macroblock edges
+static void VFilter16(uint8_t* p, int stride,
+                      int thresh, int ithresh, int hev_thresh) {
+  FilterLoop26(p, stride, 1, 16, thresh, ithresh, hev_thresh);
+}
+
+static void HFilter16(uint8_t* p, int stride,
+                      int thresh, int ithresh, int hev_thresh) {
+  FilterLoop26(p, 1, stride, 16, thresh, ithresh, hev_thresh);
+}
+
+// 8-pixels wide variant, for chroma filtering
+static void VFilter8(uint8_t* u, uint8_t* v, int stride,
+                     int thresh, int ithresh, int hev_thresh) {
+  FilterLoop26(u, stride, 1, 8, thresh, ithresh, hev_thresh);
+  FilterLoop26(v, stride, 1, 8, thresh, ithresh, hev_thresh);
+}
+
+static void HFilter8(uint8_t* u, uint8_t* v, int stride,
+                     int thresh, int ithresh, int hev_thresh) {
+  FilterLoop26(u, 1, stride, 8, thresh, ithresh, hev_thresh);
+  FilterLoop26(v, 1, stride, 8, thresh, ithresh, hev_thresh);
+}
+
+static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
+                      int thresh, int ithresh, int hev_thresh) {
+  FilterLoop24(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
+  FilterLoop24(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
+}
+
+static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
+                      int thresh, int ithresh, int hev_thresh) {
+  FilterLoop24(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
+  FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
+}
+
+// on three inner edges
+static void VFilter16i(uint8_t* p, int stride,
+                       int thresh, int ithresh, int hev_thresh) {
+  int k;
+  for (k = 3; k > 0; --k) {
+    p += 4 * stride;
+    FilterLoop24(p, stride, 1, 16, thresh, ithresh, hev_thresh);
+  }
+}
+
+static void HFilter16i(uint8_t* p, int stride,
+                       int thresh, int ithresh, int hev_thresh) {
+  int k;
+  for (k = 3; k > 0; --k) {
+    p += 4;
+    FilterLoop24(p, 1, stride, 16, thresh, ithresh, hev_thresh);
+  }
+}
+
+//------------------------------------------------------------------------------
+// Simple In-loop filtering (Paragraph 15.2)
+
+static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
+  int i;
+  for (i = 0; i < 16; ++i) {
+    if (needs_filter(p + i, stride, thresh)) {
+      do_filter2(p + i, stride);
+    }
+  }
+}
+
+static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
+  int i;
+  for (i = 0; i < 16; ++i) {
+    if (needs_filter(p + i * stride, 1, thresh)) {
+      do_filter2(p + i * stride, 1);
+    }
+  }
+}
+
+static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
+  int k;
+  for (k = 3; k > 0; --k) {
+    p += 4 * stride;
+    SimpleVFilter16(p, stride, thresh);
+  }
+}
+
+static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
+  int k;
+  for (k = 3; k > 0; --k) {
+    p += 4;
+    SimpleHFilter16(p, stride, thresh);
+  }
+}
+
+static void TransformOne(const int16_t* in, uint8_t* dst) {
+  int temp0, temp1, temp2, temp3, temp4;
+  int temp5, temp6, temp7, temp8, temp9;
+  int temp10, temp11, temp12, temp13, temp14;
+  int temp15, temp16, temp17, temp18;
+  int16_t* p_in = (int16_t*)in;
+
+  // loops unrolled and merged to avoid usage of tmp buffer
+  // and to reduce number of stalls. MUL macro is written
+  // in assembler and inlined
+  __asm__ volatile(
+    "lh       %[temp0],  0(%[in])                      \n\t"
+    "lh       %[temp8],  16(%[in])                     \n\t"
+    "lh       %[temp4],  8(%[in])                      \n\t"
+    "lh       %[temp12], 24(%[in])                     \n\t"
+    "addu     %[temp16], %[temp0],  %[temp8]           \n\t"
+    "subu     %[temp0],  %[temp0],  %[temp8]           \n\t"
+    "mul      %[temp8],  %[temp4],  %[kC2]             \n\t"
+    "mul      %[temp17], %[temp12], %[kC1]             \n\t"
+    "mul      %[temp4],  %[temp4],  %[kC1]             \n\t"
+    "mul      %[temp12], %[temp12], %[kC2]             \n\t"
+    "lh       %[temp1],  2(%[in])                      \n\t"
+    "lh       %[temp5],  10(%[in])                     \n\t"
+    "lh       %[temp9],  18(%[in])                     \n\t"
+    "lh       %[temp13], 26(%[in])                     \n\t"
+    "sra      %[temp8],  %[temp8],  16                 \n\t"
+    "sra      %[temp17], %[temp17], 16                 \n\t"
+    "sra      %[temp4],  %[temp4],  16                 \n\t"
+    "sra      %[temp12], %[temp12], 16                 \n\t"
+    "lh       %[temp2],  4(%[in])                      \n\t"
+    "lh       %[temp6],  12(%[in])                     \n\t"
+    "lh       %[temp10], 20(%[in])                     \n\t"
+    "lh       %[temp14], 28(%[in])                     \n\t"
+    "subu     %[temp17], %[temp8],  %[temp17]          \n\t"
+    "addu     %[temp4],  %[temp4],  %[temp12]          \n\t"
+    "addu     %[temp8],  %[temp16], %[temp4]           \n\t"
+    "subu     %[temp4],  %[temp16], %[temp4]           \n\t"
+    "addu     %[temp16], %[temp1],  %[temp9]           \n\t"
+    "subu     %[temp1],  %[temp1],  %[temp9]           \n\t"
+    "lh       %[temp3],  6(%[in])                      \n\t"
+    "lh       %[temp7],  14(%[in])                     \n\t"
+    "lh       %[temp11], 22(%[in])                     \n\t"
+    "lh       %[temp15], 30(%[in])                     \n\t"
+    "addu     %[temp12], %[temp0],  %[temp17]          \n\t"
+    "subu     %[temp0],  %[temp0],  %[temp17]          \n\t"
+    "mul      %[temp9],  %[temp5],  %[kC2]             \n\t"
+    "mul      %[temp17], %[temp13], %[kC1]             \n\t"
+    "mul      %[temp5],  %[temp5],  %[kC1]             \n\t"
+    "mul      %[temp13], %[temp13], %[kC2]             \n\t"
+    "sra      %[temp9],  %[temp9],  16                 \n\t"
+    "sra      %[temp17], %[temp17], 16                 \n\t"
+    "subu     %[temp17], %[temp9],  %[temp17]          \n\t"
+    "sra      %[temp5],  %[temp5],  16                 \n\t"
+    "sra      %[temp13], %[temp13], 16                 \n\t"
+    "addu     %[temp5],  %[temp5],  %[temp13]          \n\t"
+    "addu     %[temp13], %[temp1],  %[temp17]          \n\t"
+    "subu     %[temp1],  %[temp1],  %[temp17]          \n\t"
+    "mul      %[temp17], %[temp14], %[kC1]             \n\t"
+    "mul      %[temp14], %[temp14], %[kC2]             \n\t"
+    "addu     %[temp9],  %[temp16], %[temp5]           \n\t"
+    "subu     %[temp5],  %[temp16], %[temp5]           \n\t"
+    "addu     %[temp16], %[temp2],  %[temp10]          \n\t"
+    "subu     %[temp2],  %[temp2],  %[temp10]          \n\t"
+    "mul      %[temp10], %[temp6],  %[kC2]             \n\t"
+    "mul      %[temp6],  %[temp6],  %[kC1]             \n\t"
+    "sra      %[temp17], %[temp17], 16                 \n\t"
+    "sra      %[temp14], %[temp14], 16                 \n\t"
+    "sra      %[temp10], %[temp10], 16                 \n\t"
+    "sra      %[temp6],  %[temp6],  16                 \n\t"
+    "subu     %[temp17], %[temp10], %[temp17]          \n\t"
+    "addu     %[temp6],  %[temp6],  %[temp14]          \n\t"
+    "addu     %[temp10], %[temp16], %[temp6]           \n\t"
+    "subu     %[temp6],  %[temp16], %[temp6]           \n\t"
+    "addu     %[temp14], %[temp2],  %[temp17]          \n\t"
+    "subu     %[temp2],  %[temp2],  %[temp17]          \n\t"
+    "mul      %[temp17], %[temp15], %[kC1]             \n\t"
+    "mul      %[temp15], %[temp15], %[kC2]             \n\t"
+    "addu     %[temp16], %[temp3],  %[temp11]          \n\t"
+    "subu     %[temp3],  %[temp3],  %[temp11]          \n\t"
+    "mul      %[temp11], %[temp7],  %[kC2]             \n\t"
+    "mul      %[temp7],  %[temp7],  %[kC1]             \n\t"
+    "addiu    %[temp8],  %[temp8],  4                  \n\t"
+    "addiu    %[temp12], %[temp12], 4                  \n\t"
+    "addiu    %[temp0],  %[temp0],  4                  \n\t"
+    "addiu    %[temp4],  %[temp4],  4                  \n\t"
+    "sra      %[temp17], %[temp17], 16                 \n\t"
+    "sra      %[temp15], %[temp15], 16                 \n\t"
+    "sra      %[temp11], %[temp11], 16                 \n\t"
+    "sra      %[temp7],  %[temp7],  16                 \n\t"
+    "subu     %[temp17], %[temp11], %[temp17]          \n\t"
+    "addu     %[temp7],  %[temp7],  %[temp15]          \n\t"
+    "addu     %[temp15], %[temp3],  %[temp17]          \n\t"
+    "subu     %[temp3],  %[temp3],  %[temp17]          \n\t"
+    "addu     %[temp11], %[temp16], %[temp7]           \n\t"
+    "subu     %[temp7],  %[temp16], %[temp7]           \n\t"
+    "addu     %[temp16], %[temp8],  %[temp10]          \n\t"
+    "subu     %[temp8],  %[temp8],  %[temp10]          \n\t"
+    "mul      %[temp10], %[temp9],  %[kC2]             \n\t"
+    "mul      %[temp17], %[temp11], %[kC1]             \n\t"
+    "mul      %[temp9],  %[temp9],  %[kC1]             \n\t"
+    "mul      %[temp11], %[temp11], %[kC2]             \n\t"
+    "sra      %[temp10], %[temp10], 16                 \n\t"
+    "sra      %[temp17], %[temp17], 16                 \n\t"
+    "sra      %[temp9],  %[temp9],  16                 \n\t"
+    "sra      %[temp11], %[temp11], 16                 \n\t"
+    "subu     %[temp17], %[temp10], %[temp17]          \n\t"
+    "addu     %[temp11], %[temp9],  %[temp11]          \n\t"
+    "addu     %[temp10], %[temp12], %[temp14]          \n\t"
+    "subu     %[temp12], %[temp12], %[temp14]          \n\t"
+    "mul      %[temp14], %[temp13], %[kC2]             \n\t"
+    "mul      %[temp9],  %[temp15], %[kC1]             \n\t"
+    "mul      %[temp13], %[temp13], %[kC1]             \n\t"
+    "mul      %[temp15], %[temp15], %[kC2]             \n\t"
+    "sra      %[temp14], %[temp14], 16                 \n\t"
+    "sra      %[temp9],  %[temp9],  16                 \n\t"
+    "sra      %[temp13], %[temp13], 16                 \n\t"
+    "sra      %[temp15], %[temp15], 16                 \n\t"
+    "subu     %[temp9],  %[temp14], %[temp9]           \n\t"
+    "addu     %[temp15], %[temp13], %[temp15]          \n\t"
+    "addu     %[temp14], %[temp0],  %[temp2]           \n\t"
+    "subu     %[temp0],  %[temp0],  %[temp2]           \n\t"
+    "mul      %[temp2],  %[temp1],  %[kC2]             \n\t"
+    "mul      %[temp13], %[temp3],  %[kC1]             \n\t"
+    "mul      %[temp1],  %[temp1],  %[kC1]             \n\t"
+    "mul      %[temp3],  %[temp3],  %[kC2]             \n\t"
+    "sra      %[temp2],  %[temp2],  16                 \n\t"
+    "sra      %[temp13], %[temp13], 16                 \n\t"
+    "sra      %[temp1],  %[temp1],  16                 \n\t"
+    "sra      %[temp3],  %[temp3],  16                 \n\t"
+    "subu     %[temp13], %[temp2],  %[temp13]          \n\t"
+    "addu     %[temp3],  %[temp1],  %[temp3]           \n\t"
+    "addu     %[temp2],  %[temp4],  %[temp6]           \n\t"
+    "subu     %[temp4],  %[temp4],  %[temp6]           \n\t"
+    "mul      %[temp6],  %[temp5],  %[kC2]             \n\t"
+    "mul      %[temp1],  %[temp7],  %[kC1]             \n\t"
+    "mul      %[temp5],  %[temp5],  %[kC1]             \n\t"
+    "mul      %[temp7],  %[temp7],  %[kC2]             \n\t"
+    "sra      %[temp6],  %[temp6],  16                 \n\t"
+    "sra      %[temp1],  %[temp1],  16                 \n\t"
+    "sra      %[temp5],  %[temp5],  16                 \n\t"
+    "sra      %[temp7],  %[temp7],  16                 \n\t"
+    "subu     %[temp1],  %[temp6],  %[temp1]           \n\t"
+    "addu     %[temp7],  %[temp5],  %[temp7]           \n\t"
+    "addu     %[temp5],  %[temp16], %[temp11]          \n\t"
+    "subu     %[temp16], %[temp16], %[temp11]          \n\t"
+    "addu     %[temp11], %[temp8],  %[temp17]          \n\t"
+    "subu     %[temp8],  %[temp8],  %[temp17]          \n\t"
+    "sra      %[temp5],  %[temp5],  3                  \n\t"
+    "sra      %[temp16], %[temp16], 3                  \n\t"
+    "sra      %[temp11], %[temp11], 3                  \n\t"
+    "sra      %[temp8],  %[temp8],  3                  \n\t"
+    "addu     %[temp17], %[temp10], %[temp15]          \n\t"
+    "subu     %[temp10], %[temp10], %[temp15]          \n\t"
+    "addu     %[temp15], %[temp12], %[temp9]           \n\t"
+    "subu     %[temp12], %[temp12], %[temp9]           \n\t"
+    "sra      %[temp17], %[temp17], 3                  \n\t"
+    "sra      %[temp10], %[temp10], 3                  \n\t"
+    "sra      %[temp15], %[temp15], 3                  \n\t"
+    "sra      %[temp12], %[temp12], 3                  \n\t"
+    "addu     %[temp9],  %[temp14], %[temp3]           \n\t"
+    "subu     %[temp14], %[temp14], %[temp3]           \n\t"
+    "addu     %[temp3],  %[temp0],  %[temp13]          \n\t"
+    "subu     %[temp0],  %[temp0],  %[temp13]          \n\t"
+    "sra      %[temp9],  %[temp9],  3                  \n\t"
+    "sra      %[temp14], %[temp14], 3                  \n\t"
+    "sra      %[temp3],  %[temp3],  3                  \n\t"
+    "sra      %[temp0],  %[temp0],  3                  \n\t"
+    "addu     %[temp13], %[temp2],  %[temp7]           \n\t"
+    "subu     %[temp2],  %[temp2],  %[temp7]           \n\t"
+    "addu     %[temp7],  %[temp4],  %[temp1]           \n\t"
+    "subu     %[temp4],  %[temp4],  %[temp1]           \n\t"
+    "sra      %[temp13], %[temp13], 3                  \n\t"
+    "sra      %[temp2],  %[temp2],  3                  \n\t"
+    "sra      %[temp7],  %[temp7],  3                  \n\t"
+    "sra      %[temp4],  %[temp4],  3                  \n\t"
+    "addiu    %[temp6],  $zero,     255                \n\t"
+    "lbu      %[temp1],  0(%[dst])                     \n\t"
+    "addu     %[temp1],  %[temp1],  %[temp5]           \n\t"
+    "sra      %[temp5],  %[temp1],  8                  \n\t"
+    "sra      %[temp18], %[temp1],  31                 \n\t"
+    "beqz     %[temp5],  1f                            \n\t"
+    "xor      %[temp1],  %[temp1],  %[temp1]           \n\t"
+    "movz     %[temp1],  %[temp6],  %[temp18]          \n\t"
+  "1:                                                  \n\t"
+    "lbu      %[temp18], 1(%[dst])                     \n\t"
+    "sb       %[temp1],  0(%[dst])                     \n\t"
+    "addu     %[temp18], %[temp18], %[temp11]          \n\t"
+    "sra      %[temp11], %[temp18], 8                  \n\t"
+    "sra      %[temp1],  %[temp18], 31                 \n\t"
+    "beqz     %[temp11], 2f                            \n\t"
+    "xor      %[temp18], %[temp18], %[temp18]          \n\t"
+    "movz     %[temp18], %[temp6],  %[temp1]           \n\t"
+  "2:                                                  \n\t"
+    "lbu      %[temp1],  2(%[dst])                     \n\t"
+    "sb       %[temp18], 1(%[dst])                     \n\t"
+    "addu     %[temp1],  %[temp1],  %[temp8]           \n\t"
+    "sra      %[temp8],  %[temp1],  8                  \n\t"
+    "sra      %[temp18], %[temp1],  31                 \n\t"
+    "beqz     %[temp8],  3f                            \n\t"
+    "xor      %[temp1],  %[temp1],  %[temp1]           \n\t"
+    "movz     %[temp1],  %[temp6],  %[temp18]          \n\t"
+  "3:                                                  \n\t"
+    "lbu      %[temp18], 3(%[dst])                     \n\t"
+    "sb       %[temp1],  2(%[dst])                     \n\t"
+    "addu     %[temp18], %[temp18], %[temp16]          \n\t"
+    "sra      %[temp16], %[temp18], 8                  \n\t"
+    "sra      %[temp1],  %[temp18], 31                 \n\t"
+    "beqz     %[temp16], 4f                            \n\t"
+    "xor      %[temp18], %[temp18], %[temp18]          \n\t"
+    "movz     %[temp18], %[temp6],  %[temp1]           \n\t"
+  "4:                                                  \n\t"
+    "sb       %[temp18], 3(%[dst])                     \n\t"
+    "lbu      %[temp5],  32(%[dst])                    \n\t"
+    "lbu      %[temp8],  33(%[dst])                    \n\t"
+    "lbu      %[temp11], 34(%[dst])                    \n\t"
+    "lbu      %[temp16], 35(%[dst])                    \n\t"
+    "addu     %[temp5],  %[temp5],  %[temp17]          \n\t"
+    "addu     %[temp8],  %[temp8],  %[temp15]          \n\t"
+    "addu     %[temp11], %[temp11], %[temp12]          \n\t"
+    "addu     %[temp16], %[temp16], %[temp10]          \n\t"
+    "sra      %[temp18], %[temp5],  8                  \n\t"
+    "sra      %[temp1],  %[temp5],  31                 \n\t"
+    "beqz     %[temp18], 5f                            \n\t"
+    "xor      %[temp5],  %[temp5],  %[temp5]           \n\t"
+    "movz     %[temp5],  %[temp6],  %[temp1]           \n\t"
+  "5:                                                  \n\t"
+    "sra      %[temp18], %[temp8],  8                  \n\t"
+    "sra      %[temp1],  %[temp8],  31                 \n\t"
+    "beqz     %[temp18], 6f                            \n\t"
+    "xor      %[temp8],  %[temp8],  %[temp8]           \n\t"
+    "movz     %[temp8],  %[temp6],  %[temp1]           \n\t"
+  "6:                                                  \n\t"
+    "sra      %[temp18], %[temp11], 8                  \n\t"
+    "sra      %[temp1],  %[temp11], 31                 \n\t"
+    "sra      %[temp17], %[temp16], 8                  \n\t"
+    "sra      %[temp15], %[temp16], 31                 \n\t"
+    "beqz     %[temp18], 7f                            \n\t"
+    "xor      %[temp11], %[temp11], %[temp11]          \n\t"
+    "movz     %[temp11], %[temp6],  %[temp1]           \n\t"
+  "7:                                                  \n\t"
+    "beqz     %[temp17], 8f                            \n\t"
+    "xor      %[temp16], %[temp16], %[temp16]          \n\t"
+    "movz     %[temp16], %[temp6],  %[temp15]          \n\t"
+  "8:                                                  \n\t"
+    "sb       %[temp5],  32(%[dst])                    \n\t"
+    "sb       %[temp8],  33(%[dst])                    \n\t"
+    "sb       %[temp11], 34(%[dst])                    \n\t"
+    "sb       %[temp16], 35(%[dst])                    \n\t"
+    "lbu      %[temp5],  64(%[dst])                    \n\t"
+    "lbu      %[temp8],  65(%[dst])                    \n\t"
+    "lbu      %[temp11], 66(%[dst])                    \n\t"
+    "lbu      %[temp16], 67(%[dst])                    \n\t"
+    "addu     %[temp5],  %[temp5],  %[temp9]           \n\t"
+    "addu     %[temp8],  %[temp8],  %[temp3]           \n\t"
+    "addu     %[temp11], %[temp11], %[temp0]           \n\t"
+    "addu     %[temp16], %[temp16], %[temp14]          \n\t"
+    "sra      %[temp18], %[temp5],  8                  \n\t"
+    "sra      %[temp1],  %[temp5],  31                 \n\t"
+    "sra      %[temp17], %[temp8],  8                  \n\t"
+    "sra      %[temp15], %[temp8],  31                 \n\t"
+    "sra      %[temp12], %[temp11], 8                  \n\t"
+    "sra      %[temp10], %[temp11], 31                 \n\t"
+    "sra      %[temp9],  %[temp16], 8                  \n\t"
+    "sra      %[temp3],  %[temp16], 31                 \n\t"
+    "beqz     %[temp18], 9f                            \n\t"
+    "xor      %[temp5],  %[temp5],  %[temp5]           \n\t"
+    "movz     %[temp5],  %[temp6],  %[temp1]           \n\t"
+  "9:                                                  \n\t"
+    "beqz     %[temp17], 10f                           \n\t"
+    "xor      %[temp8],  %[temp8],  %[temp8]           \n\t"
+    "movz     %[temp8],  %[temp6],  %[temp15]          \n\t"
+  "10:                                                 \n\t"
+    "beqz     %[temp12], 11f                           \n\t"
+    "xor      %[temp11], %[temp11], %[temp11]          \n\t"
+    "movz     %[temp11], %[temp6],  %[temp10]          \n\t"
+  "11:                                                 \n\t"
+    "beqz     %[temp9],  12f                           \n\t"
+    "xor      %[temp16], %[temp16], %[temp16]          \n\t"
+    "movz     %[temp16], %[temp6],  %[temp3]           \n\t"
+  "12:                                                 \n\t"
+    "sb       %[temp5],  64(%[dst])                    \n\t"
+    "sb       %[temp8],  65(%[dst])                    \n\t"
+    "sb       %[temp11], 66(%[dst])                    \n\t"
+    "sb       %[temp16], 67(%[dst])                    \n\t"
+    "lbu      %[temp5],  96(%[dst])                    \n\t"
+    "lbu      %[temp8],  97(%[dst])                    \n\t"
+    "lbu      %[temp11], 98(%[dst])                    \n\t"
+    "lbu      %[temp16], 99(%[dst])                    \n\t"
+    "addu     %[temp5],  %[temp5],  %[temp13]          \n\t"
+    "addu     %[temp8],  %[temp8],  %[temp7]           \n\t"
+    "addu     %[temp11], %[temp11], %[temp4]           \n\t"
+    "addu     %[temp16], %[temp16], %[temp2]           \n\t"
+    "sra      %[temp18], %[temp5],  8                  \n\t"
+    "sra      %[temp1],  %[temp5],  31                 \n\t"
+    "sra      %[temp17], %[temp8],  8                  \n\t"
+    "sra      %[temp15], %[temp8],  31                 \n\t"
+    "sra      %[temp12], %[temp11], 8                  \n\t"
+    "sra      %[temp10], %[temp11], 31                 \n\t"
+    "sra      %[temp9],  %[temp16], 8                  \n\t"
+    "sra      %[temp3],  %[temp16], 31                 \n\t"
+    "beqz     %[temp18], 13f                           \n\t"
+    "xor      %[temp5],  %[temp5],  %[temp5]           \n\t"
+    "movz     %[temp5],  %[temp6],  %[temp1]           \n\t"
+  "13:                                                 \n\t"
+    "beqz     %[temp17], 14f                           \n\t"
+    "xor      %[temp8],  %[temp8],  %[temp8]           \n\t"
+    "movz     %[temp8],  %[temp6],  %[temp15]          \n\t"
+  "14:                                                 \n\t"
+    "beqz     %[temp12], 15f                           \n\t"
+    "xor      %[temp11], %[temp11], %[temp11]          \n\t"
+    "movz     %[temp11], %[temp6],  %[temp10]          \n\t"
+  "15:                                                 \n\t"
+    "beqz     %[temp9],  16f                           \n\t"
+    "xor      %[temp16], %[temp16], %[temp16]          \n\t"
+    "movz     %[temp16], %[temp6],  %[temp3]           \n\t"
+  "16:                                                 \n\t"
+    "sb       %[temp5],  96(%[dst])                    \n\t"
+    "sb       %[temp8],  97(%[dst])                    \n\t"
+    "sb       %[temp11], 98(%[dst])                    \n\t"
+    "sb       %[temp16], 99(%[dst])                    \n\t"
+
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
+      [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11),
+      [temp12]"=&r"(temp12), [temp13]"=&r"(temp13), [temp14]"=&r"(temp14),
+      [temp15]"=&r"(temp15), [temp16]"=&r"(temp16), [temp17]"=&r"(temp17),
+      [temp18]"=&r"(temp18)
+    : [in]"r"(p_in), [kC1]"r"(kC1), [kC2]"r"(kC2), [dst]"r"(dst)
+    : "memory", "hi", "lo"
+  );
+}
+
+static void TransformTwo(const int16_t* in, uint8_t* dst, int do_two) {
+  TransformOne(in, dst);
+  if (do_two) {
+    TransformOne(in + 16, dst + 4);
+  }
+}
+
+#endif  // WEBP_USE_MIPS32
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void VP8DspInitMIPS32(void);
+
+void VP8DspInitMIPS32(void) {
+#if defined(WEBP_USE_MIPS32)
+  VP8InitClipTables();
+
+  VP8Transform = TransformTwo;
+
+  VP8VFilter16 = VFilter16;
+  VP8HFilter16 = HFilter16;
+  VP8VFilter8 = VFilter8;
+  VP8HFilter8 = HFilter8;
+  VP8VFilter16i = VFilter16i;
+  VP8HFilter16i = HFilter16i;
+  VP8VFilter8i = VFilter8i;
+  VP8HFilter8i = HFilter8i;
+
+  VP8SimpleVFilter16 = SimpleVFilter16;
+  VP8SimpleHFilter16 = SimpleHFilter16;
+  VP8SimpleVFilter16i = SimpleVFilter16i;
+  VP8SimpleHFilter16i = SimpleHFilter16i;
+#endif  // WEBP_USE_MIPS32
+}

src/dsp/dec_sse2.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // SSE2 version of some decoding functions (idct, loop filtering).
@@ -12,19 +14,19 @@
 
 #include "./dsp.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 #if defined(WEBP_USE_SSE2)
 
+// The 3-coeff sparse transform in SSE2 is not really faster than the plain-C
+// one it seems => disable it by default. Uncomment the following to enable:
+// #define USE_TRANSFORM_AC3
+
 #include <emmintrin.h>
 #include "../dec/vp8i.h"
 
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
-static void TransformSSE2(const int16_t* in, uint8_t* dst, int do_two) {
+static void Transform(const int16_t* in, uint8_t* dst, int do_two) {
   // This implementation makes use of 16-bit fixed point versions of two
   // multiply constants:
   //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
@@ -194,21 +196,21 @@ static void TransformSSE2(const int16_t* in, uint8_t* dst, int do_two) {
 
   // Add inverse transform to 'dst' and store.
   {
-    const __m128i zero = _mm_set1_epi16(0);
+    const __m128i zero = _mm_setzero_si128();
     // Load the reference(s).
     __m128i dst0, dst1, dst2, dst3;
     if (do_two) {
       // Load eight bytes/pixels per line.
-      dst0 = _mm_loadl_epi64((__m128i*)&dst[0 * BPS]);
-      dst1 = _mm_loadl_epi64((__m128i*)&dst[1 * BPS]);
-      dst2 = _mm_loadl_epi64((__m128i*)&dst[2 * BPS]);
-      dst3 = _mm_loadl_epi64((__m128i*)&dst[3 * BPS]);
+      dst0 = _mm_loadl_epi64((__m128i*)(dst + 0 * BPS));
+      dst1 = _mm_loadl_epi64((__m128i*)(dst + 1 * BPS));
+      dst2 = _mm_loadl_epi64((__m128i*)(dst + 2 * BPS));
+      dst3 = _mm_loadl_epi64((__m128i*)(dst + 3 * BPS));
     } else {
       // Load four bytes/pixels per line.
-      dst0 = _mm_cvtsi32_si128(*(int*)&dst[0 * BPS]);
-      dst1 = _mm_cvtsi32_si128(*(int*)&dst[1 * BPS]);
-      dst2 = _mm_cvtsi32_si128(*(int*)&dst[2 * BPS]);
-      dst3 = _mm_cvtsi32_si128(*(int*)&dst[3 * BPS]);
+      dst0 = _mm_cvtsi32_si128(*(int*)(dst + 0 * BPS));
+      dst1 = _mm_cvtsi32_si128(*(int*)(dst + 1 * BPS));
+      dst2 = _mm_cvtsi32_si128(*(int*)(dst + 2 * BPS));
+      dst3 = _mm_cvtsi32_si128(*(int*)(dst + 3 * BPS));
     }
     // Convert to 16b.
     dst0 = _mm_unpacklo_epi8(dst0, zero);
@@ -228,20 +230,66 @@ static void TransformSSE2(const int16_t* in, uint8_t* dst, int do_two) {
     // Store the results.
     if (do_two) {
       // Store eight bytes/pixels per line.
-      _mm_storel_epi64((__m128i*)&dst[0 * BPS], dst0);
-      _mm_storel_epi64((__m128i*)&dst[1 * BPS], dst1);
-      _mm_storel_epi64((__m128i*)&dst[2 * BPS], dst2);
-      _mm_storel_epi64((__m128i*)&dst[3 * BPS], dst3);
+      _mm_storel_epi64((__m128i*)(dst + 0 * BPS), dst0);
+      _mm_storel_epi64((__m128i*)(dst + 1 * BPS), dst1);
+      _mm_storel_epi64((__m128i*)(dst + 2 * BPS), dst2);
+      _mm_storel_epi64((__m128i*)(dst + 3 * BPS), dst3);
     } else {
       // Store four bytes/pixels per line.
-      *((int32_t *)&dst[0 * BPS]) = _mm_cvtsi128_si32(dst0);
-      *((int32_t *)&dst[1 * BPS]) = _mm_cvtsi128_si32(dst1);
-      *((int32_t *)&dst[2 * BPS]) = _mm_cvtsi128_si32(dst2);
-      *((int32_t *)&dst[3 * BPS]) = _mm_cvtsi128_si32(dst3);
+      *(int*)(dst + 0 * BPS) = _mm_cvtsi128_si32(dst0);
+      *(int*)(dst + 1 * BPS) = _mm_cvtsi128_si32(dst1);
+      *(int*)(dst + 2 * BPS) = _mm_cvtsi128_si32(dst2);
+      *(int*)(dst + 3 * BPS) = _mm_cvtsi128_si32(dst3);
     }
   }
 }
 
+#if defined(USE_TRANSFORM_AC3)
+#define MUL(a, b) (((a) * (b)) >> 16)
+static void TransformAC3(const int16_t* in, uint8_t* dst) {
+  static const int kC1 = 20091 + (1 << 16);
+  static const int kC2 = 35468;
+  const __m128i A = _mm_set1_epi16(in[0] + 4);
+  const __m128i c4 = _mm_set1_epi16(MUL(in[4], kC2));
+  const __m128i d4 = _mm_set1_epi16(MUL(in[4], kC1));
+  const int c1 = MUL(in[1], kC2);
+  const int d1 = MUL(in[1], kC1);
+  const __m128i CD = _mm_set_epi16(0, 0, 0, 0, -d1, -c1, c1, d1);
+  const __m128i B = _mm_adds_epi16(A, CD);
+  const __m128i m0 = _mm_adds_epi16(B, d4);
+  const __m128i m1 = _mm_adds_epi16(B, c4);
+  const __m128i m2 = _mm_subs_epi16(B, c4);
+  const __m128i m3 = _mm_subs_epi16(B, d4);
+  const __m128i zero = _mm_setzero_si128();
+  // Load the source pixels.
+  __m128i dst0 = _mm_cvtsi32_si128(*(int*)(dst + 0 * BPS));
+  __m128i dst1 = _mm_cvtsi32_si128(*(int*)(dst + 1 * BPS));
+  __m128i dst2 = _mm_cvtsi32_si128(*(int*)(dst + 2 * BPS));
+  __m128i dst3 = _mm_cvtsi32_si128(*(int*)(dst + 3 * BPS));
+  // Convert to 16b.
+  dst0 = _mm_unpacklo_epi8(dst0, zero);
+  dst1 = _mm_unpacklo_epi8(dst1, zero);
+  dst2 = _mm_unpacklo_epi8(dst2, zero);
+  dst3 = _mm_unpacklo_epi8(dst3, zero);
+  // Add the inverse transform.
+  dst0 = _mm_adds_epi16(dst0, _mm_srai_epi16(m0, 3));
+  dst1 = _mm_adds_epi16(dst1, _mm_srai_epi16(m1, 3));
+  dst2 = _mm_adds_epi16(dst2, _mm_srai_epi16(m2, 3));
+  dst3 = _mm_adds_epi16(dst3, _mm_srai_epi16(m3, 3));
+  // Unsigned saturate to 8b.
+  dst0 = _mm_packus_epi16(dst0, dst0);
+  dst1 = _mm_packus_epi16(dst1, dst1);
+  dst2 = _mm_packus_epi16(dst2, dst2);
+  dst3 = _mm_packus_epi16(dst3, dst3);
+  // Store the results.
+  *(int*)(dst + 0 * BPS) = _mm_cvtsi128_si32(dst0);
+  *(int*)(dst + 1 * BPS) = _mm_cvtsi128_si32(dst1);
+  *(int*)(dst + 2 * BPS) = _mm_cvtsi128_si32(dst2);
+  *(int*)(dst + 3 * BPS) = _mm_cvtsi128_si32(dst3);
+}
+#undef MUL
+#endif   // USE_TRANSFORM_AC3
+
 //------------------------------------------------------------------------------
 // Loop Filter (Paragraph 15)
 
@@ -250,20 +298,15 @@ static void TransformSSE2(const int16_t* in, uint8_t* dst, int do_two) {
     _mm_subs_epu8((q), (p)),                                                   \
     _mm_subs_epu8((p), (q)))
 
-// Shift each byte of "a" by N bits while preserving by the sign bit.
-//
-// It first shifts the lower bytes of the words and then the upper bytes and
-// then merges the results together.
-#define SIGNED_SHIFT_N(a, N) {                                                 \
-  __m128i t = a;                                                               \
-  t = _mm_slli_epi16(t, 8);                                                    \
-  t = _mm_srai_epi16(t, N);                                                    \
-  t = _mm_srli_epi16(t, 8);                                                    \
-                                                                               \
-  a = _mm_srai_epi16(a, N + 8);                                                \
-  a = _mm_slli_epi16(a, 8);                                                    \
-                                                                               \
-  a = _mm_or_si128(t, a);                                                      \
+// Shift each byte of "x" by 3 bits while preserving by the sign bit.
+static WEBP_INLINE void SignedShift8b(__m128i* const x) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i signs = _mm_cmpgt_epi8(zero, *x);
+  const __m128i lo_0 = _mm_unpacklo_epi8(*x, signs);  // s8 -> s16 sign extend
+  const __m128i hi_0 = _mm_unpackhi_epi8(*x, signs);
+  const __m128i lo_1 = _mm_srai_epi16(lo_0, 3);
+  const __m128i hi_1 = _mm_srai_epi16(hi_0, 3);
+  *x = _mm_packs_epi16(lo_1, hi_1);
 }
 
 #define FLIP_SIGN_BIT2(a, b) {                                                 \
@@ -276,103 +319,123 @@ static void TransformSSE2(const int16_t* in, uint8_t* dst, int do_two) {
   FLIP_SIGN_BIT2(c, d);                                                        \
 }
 
-#define GET_NOTHEV(p1, p0, q0, q1, hev_thresh, not_hev) {                      \
-  const __m128i zero = _mm_setzero_si128();                                    \
-  const __m128i t1 = MM_ABS(p1, p0);                                           \
-  const __m128i t2 = MM_ABS(q1, q0);                                           \
-                                                                               \
-  const __m128i h = _mm_set1_epi8(hev_thresh);                                 \
-  const __m128i t3 = _mm_subs_epu8(t1, h);  /* abs(p1 - p0) - hev_tresh */     \
-  const __m128i t4 = _mm_subs_epu8(t2, h);  /* abs(q1 - q0) - hev_tresh */     \
-                                                                               \
-  not_hev = _mm_or_si128(t3, t4);                                              \
-  not_hev = _mm_cmpeq_epi8(not_hev, zero); /* not_hev <= t1 && not_hev <= t2 */\
+// input/output is uint8_t
+static WEBP_INLINE void GetNotHEV(const __m128i* const p1,
+                                  const __m128i* const p0,
+                                  const __m128i* const q0,
+                                  const __m128i* const q1,
+                                  int hev_thresh, __m128i* const not_hev) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i t_1 = MM_ABS(*p1, *p0);
+  const __m128i t_2 = MM_ABS(*q1, *q0);
+
+  const __m128i h = _mm_set1_epi8(hev_thresh);
+  const __m128i t_3 = _mm_subs_epu8(t_1, h);  // abs(p1 - p0) - hev_tresh
+  const __m128i t_4 = _mm_subs_epu8(t_2, h);  // abs(q1 - q0) - hev_tresh
+
+  *not_hev = _mm_or_si128(t_3, t_4);
+  *not_hev = _mm_cmpeq_epi8(*not_hev, zero);  // not_hev <= t1 && not_hev <= t2
 }
 
-#define GET_BASE_DELTA(p1, p0, q0, q1, o) {                                    \
-  const __m128i qp0 = _mm_subs_epi8(q0, p0);  /* q0 - p0 */                    \
-  o = _mm_subs_epi8(p1, q1);            /* p1 - q1 */                          \
-  o = _mm_adds_epi8(o, qp0);            /* p1 - q1 + 1 * (q0 - p0) */          \
-  o = _mm_adds_epi8(o, qp0);            /* p1 - q1 + 2 * (q0 - p0) */          \
-  o = _mm_adds_epi8(o, qp0);            /* p1 - q1 + 3 * (q0 - p0) */          \
+// input pixels are int8_t
+static WEBP_INLINE void GetBaseDelta(const __m128i* const p1,
+                                     const __m128i* const p0,
+                                     const __m128i* const q0,
+                                     const __m128i* const q1,
+                                     __m128i* const delta) {
+  // beware of addition order, for saturation!
+  const __m128i p1_q1 = _mm_subs_epi8(*p1, *q1);   // p1 - q1
+  const __m128i q0_p0 = _mm_subs_epi8(*q0, *p0);   // q0 - p0
+  const __m128i s1 = _mm_adds_epi8(p1_q1, q0_p0);  // p1 - q1 + 1 * (q0 - p0)
+  const __m128i s2 = _mm_adds_epi8(q0_p0, s1);     // p1 - q1 + 2 * (q0 - p0)
+  const __m128i s3 = _mm_adds_epi8(q0_p0, s2);     // p1 - q1 + 3 * (q0 - p0)
+  *delta = s3;
 }
 
-#define DO_SIMPLE_FILTER(p0, q0, fl) {                                         \
-  const __m128i three = _mm_set1_epi8(3);                                      \
-  const __m128i four = _mm_set1_epi8(4);                                       \
-  __m128i v3 = _mm_adds_epi8(fl, three);                                       \
-  __m128i v4 = _mm_adds_epi8(fl, four);                                        \
-                                                                               \
-  /* Do +4 side */                                                             \
-  SIGNED_SHIFT_N(v4, 3);                /* v4 >> 3  */                         \
-  q0 = _mm_subs_epi8(q0, v4);           /* q0 -= v4 */                         \
-                                                                               \
-  /* Now do +3 side */                                                         \
-  SIGNED_SHIFT_N(v3, 3);                /* v3 >> 3  */                         \
-  p0 = _mm_adds_epi8(p0, v3);           /* p0 += v3 */                         \
+// input and output are int8_t
+static WEBP_INLINE void DoSimpleFilter(__m128i* const p0, __m128i* const q0,
+                                       const __m128i* const fl) {
+  const __m128i k3 = _mm_set1_epi8(3);
+  const __m128i k4 = _mm_set1_epi8(4);
+  __m128i v3 = _mm_adds_epi8(*fl, k3);
+  __m128i v4 = _mm_adds_epi8(*fl, k4);
+
+  SignedShift8b(&v4);                  // v4 >> 3
+  SignedShift8b(&v3);                  // v3 >> 3
+  *q0 = _mm_subs_epi8(*q0, v4);        // q0 -= v4
+  *p0 = _mm_adds_epi8(*p0, v3);        // p0 += v3
 }
 
 // Updates values of 2 pixels at MB edge during complex filtering.
 // Update operations:
-// q = q - a and p = p + a; where a = [(a_hi >> 7), (a_lo >> 7)]
-#define UPDATE_2PIXELS(pi, qi, a_lo, a_hi) {                                   \
-  const __m128i a_lo7 = _mm_srai_epi16(a_lo, 7);                               \
-  const __m128i a_hi7 = _mm_srai_epi16(a_hi, 7);                               \
-  const __m128i a = _mm_packs_epi16(a_lo7, a_hi7);                             \
-  pi = _mm_adds_epi8(pi, a);                                                   \
-  qi = _mm_subs_epi8(qi, a);                                                   \
+// q = q - delta and p = p + delta; where delta = [(a_hi >> 7), (a_lo >> 7)]
+// Pixels 'pi' and 'qi' are int8_t on input, uint8_t on output (sign flip).
+static WEBP_INLINE void Update2Pixels(__m128i* const pi, __m128i* const qi,
+                                      const __m128i* const a0_lo,
+                                      const __m128i* const a0_hi) {
+  const __m128i a1_lo = _mm_srai_epi16(*a0_lo, 7);
+  const __m128i a1_hi = _mm_srai_epi16(*a0_hi, 7);
+  const __m128i delta = _mm_packs_epi16(a1_lo, a1_hi);
+  const __m128i sign_bit = _mm_set1_epi8(0x80);
+  *pi = _mm_adds_epi8(*pi, delta);
+  *qi = _mm_subs_epi8(*qi, delta);
+  FLIP_SIGN_BIT2(*pi, *qi);
 }
 
-static void NeedsFilter(const __m128i* p1, const __m128i* p0, const __m128i* q0,
-                        const __m128i* q1, int thresh, __m128i *mask) {
-  __m128i t1 = MM_ABS(*p1, *q1);        // abs(p1 - q1)
-  *mask = _mm_set1_epi8(0xFE);
-  t1 = _mm_and_si128(t1, *mask);        // set lsb of each byte to zero
-  t1 = _mm_srli_epi16(t1, 1);           // abs(p1 - q1) / 2
+// input pixels are uint8_t
+static WEBP_INLINE void NeedsFilter(const __m128i* const p1,
+                                    const __m128i* const p0,
+                                    const __m128i* const q0,
+                                    const __m128i* const q1,
+                                    int thresh, __m128i* const mask) {
+  const __m128i m_thresh = _mm_set1_epi8(thresh);
+  const __m128i t1 = MM_ABS(*p1, *q1);        // abs(p1 - q1)
+  const __m128i kFE = _mm_set1_epi8(0xFE);
+  const __m128i t2 = _mm_and_si128(t1, kFE);  // set lsb of each byte to zero
+  const __m128i t3 = _mm_srli_epi16(t2, 1);   // abs(p1 - q1) / 2
 
-  *mask = MM_ABS(*p0, *q0);             // abs(p0 - q0)
-  *mask = _mm_adds_epu8(*mask, *mask);  // abs(p0 - q0) * 2
-  *mask = _mm_adds_epu8(*mask, t1);     // abs(p0 - q0) * 2 + abs(p1 - q1) / 2
+  const __m128i t4 = MM_ABS(*p0, *q0);        // abs(p0 - q0)
+  const __m128i t5 = _mm_adds_epu8(t4, t4);   // abs(p0 - q0) * 2
+  const __m128i t6 = _mm_adds_epu8(t5, t3);   // abs(p0-q0)*2 + abs(p1-q1)/2
 
-  t1 = _mm_set1_epi8(thresh);
-  *mask = _mm_subs_epu8(*mask, t1);     // mask <= thresh
-  *mask = _mm_cmpeq_epi8(*mask, _mm_setzero_si128());
+  const __m128i t7 = _mm_subs_epu8(t6, m_thresh);  // mask <= m_thresh
+  *mask = _mm_cmpeq_epi8(t7, _mm_setzero_si128());
 }
 
 //------------------------------------------------------------------------------
 // Edge filtering functions
 
 // Applies filter on 2 pixels (p0 and q0)
-static WEBP_INLINE void DoFilter2(const __m128i* p1, __m128i* p0, __m128i* q0,
-                                  const __m128i* q1, int thresh) {
+static WEBP_INLINE void DoFilter2(__m128i* const p1, __m128i* const p0,
+                                  __m128i* const q0, __m128i* const q1,
+                                  int thresh) {
   __m128i a, mask;
   const __m128i sign_bit = _mm_set1_epi8(0x80);
+  // convert p1/q1 to int8_t (for GetBaseDelta)
   const __m128i p1s = _mm_xor_si128(*p1, sign_bit);
   const __m128i q1s = _mm_xor_si128(*q1, sign_bit);
 
   NeedsFilter(p1, p0, q0, q1, thresh, &mask);
 
-  // convert to signed values
   FLIP_SIGN_BIT2(*p0, *q0);
-
-  GET_BASE_DELTA(p1s, *p0, *q0, q1s, a);
+  GetBaseDelta(&p1s, p0, q0, &q1s, &a);
   a = _mm_and_si128(a, mask);     // mask filter values we don't care about
-  DO_SIMPLE_FILTER(*p0, *q0, a);
-
-  // unoffset
+  DoSimpleFilter(p0, q0, &a);
   FLIP_SIGN_BIT2(*p0, *q0);
 }
 
 // Applies filter on 4 pixels (p1, p0, q0 and q1)
-static WEBP_INLINE void DoFilter4(__m128i* p1, __m128i *p0,
-                                  __m128i* q0, __m128i* q1,
-                                  const __m128i* mask, int hev_thresh) {
+static WEBP_INLINE void DoFilter4(__m128i* const p1, __m128i* const p0,
+                                  __m128i* const q0, __m128i* const q1,
+                                  const __m128i* const mask, int hev_thresh) {
+  const __m128i sign_bit = _mm_set1_epi8(0x80);
+  const __m128i k64 = _mm_set1_epi8(0x40);
+  const __m128i zero = _mm_setzero_si128();
   __m128i not_hev;
   __m128i t1, t2, t3;
-  const __m128i sign_bit = _mm_set1_epi8(0x80);
 
   // compute hev mask
-  GET_NOTHEV(*p1, *p0, *q0, *q1, hev_thresh, not_hev);
+  GetNotHEV(p1, p0, q0, q1, hev_thresh, &not_hev);
 
   // convert to signed values
   FLIP_SIGN_BIT4(*p1, *p0, *q0, *q1);
@@ -385,92 +448,83 @@ static WEBP_INLINE void DoFilter4(__m128i* p1, __m128i *p0,
   t1 = _mm_adds_epi8(t1, t2);          // hev(p1 - q1) + 3 * (q0 - p0)
   t1 = _mm_and_si128(t1, *mask);       // mask filter values we don't care about
 
-  // Do +4 side
-  t2 = _mm_set1_epi8(4);
-  t2 = _mm_adds_epi8(t1, t2);        // 3 * (q0 - p0) + (p1 - q1) + 4
-  SIGNED_SHIFT_N(t2, 3);             // (3 * (q0 - p0) + hev(p1 - q1) + 4) >> 3
-  t3 = t2;                           // save t2
-  *q0 = _mm_subs_epi8(*q0, t2);      // q0 -= t2
-
-  // Now do +3 side
   t2 = _mm_set1_epi8(3);
-  t2 = _mm_adds_epi8(t1, t2);        // +3 instead of +4
-  SIGNED_SHIFT_N(t2, 3);             // (3 * (q0 - p0) + hev(p1 - q1) + 3) >> 3
+  t3 = _mm_set1_epi8(4);
+  t2 = _mm_adds_epi8(t1, t2);        // 3 * (q0 - p0) + (p1 - q1) + 3
+  t3 = _mm_adds_epi8(t1, t3);        // 3 * (q0 - p0) + (p1 - q1) + 4
+  SignedShift8b(&t2);                // (3 * (q0 - p0) + hev(p1 - q1) + 3) >> 3
+  SignedShift8b(&t3);                // (3 * (q0 - p0) + hev(p1 - q1) + 4) >> 3
   *p0 = _mm_adds_epi8(*p0, t2);      // p0 += t2
+  *q0 = _mm_subs_epi8(*q0, t3);      // q0 -= t3
+  FLIP_SIGN_BIT2(*p0, *q0);
 
-  t2 = _mm_set1_epi8(1);
-  t3 = _mm_adds_epi8(t3, t2);
-  SIGNED_SHIFT_N(t3, 1);             // (3 * (q0 - p0) + hev(p1 - q1) + 4) >> 4
+  // this is equivalent to signed (a + 1) >> 1 calculation
+  t2 = _mm_add_epi8(t3, sign_bit);
+  t3 = _mm_avg_epu8(t2, zero);
+  t3 = _mm_sub_epi8(t3, k64);
 
   t3 = _mm_and_si128(not_hev, t3);   // if !hev
   *q1 = _mm_subs_epi8(*q1, t3);      // q1 -= t3
   *p1 = _mm_adds_epi8(*p1, t3);      // p1 += t3
-
-  // unoffset
-  FLIP_SIGN_BIT4(*p1, *p0, *q0, *q1);
+  FLIP_SIGN_BIT2(*p1, *q1);
 }
 
 // Applies filter on 6 pixels (p2, p1, p0, q0, q1 and q2)
-static WEBP_INLINE void DoFilter6(__m128i *p2, __m128i* p1, __m128i *p0,
-                                  __m128i* q0, __m128i* q1, __m128i *q2,
-                                  const __m128i* mask, int hev_thresh) {
-  __m128i a, not_hev;
+static WEBP_INLINE void DoFilter6(__m128i* const p2, __m128i* const p1,
+                                  __m128i* const p0, __m128i* const q0,
+                                  __m128i* const q1, __m128i* const q2,
+                                  const __m128i* const mask, int hev_thresh) {
+  const __m128i zero = _mm_setzero_si128();
   const __m128i sign_bit = _mm_set1_epi8(0x80);
+  __m128i a, not_hev;
 
   // compute hev mask
-  GET_NOTHEV(*p1, *p0, *q0, *q1, hev_thresh, not_hev);
+  GetNotHEV(p1, p0, q0, q1, hev_thresh, &not_hev);
 
-  // convert to signed values
   FLIP_SIGN_BIT4(*p1, *p0, *q0, *q1);
   FLIP_SIGN_BIT2(*p2, *q2);
-
-  GET_BASE_DELTA(*p1, *p0, *q0, *q1, a);
+  GetBaseDelta(p1, p0, q0, q1, &a);
 
   { // do simple filter on pixels with hev
     const __m128i m = _mm_andnot_si128(not_hev, *mask);
     const __m128i f = _mm_and_si128(a, m);
-    DO_SIMPLE_FILTER(*p0, *q0, f);
+    DoSimpleFilter(p0, q0, &f);
   }
+
   { // do strong filter on pixels with not hev
-    const __m128i zero = _mm_setzero_si128();
-    const __m128i nine = _mm_set1_epi16(0x0900);
-    const __m128i sixty_three = _mm_set1_epi16(63);
+    const __m128i k9 = _mm_set1_epi16(0x0900);
+    const __m128i k63 = _mm_set1_epi16(63);
 
     const __m128i m = _mm_and_si128(not_hev, *mask);
     const __m128i f = _mm_and_si128(a, m);
+
     const __m128i f_lo = _mm_unpacklo_epi8(zero, f);
     const __m128i f_hi = _mm_unpackhi_epi8(zero, f);
 
-    const __m128i f9_lo = _mm_mulhi_epi16(f_lo, nine);   // Filter (lo) * 9
-    const __m128i f9_hi = _mm_mulhi_epi16(f_hi, nine);   // Filter (hi) * 9
-    const __m128i f18_lo = _mm_add_epi16(f9_lo, f9_lo);  // Filter (lo) * 18
-    const __m128i f18_hi = _mm_add_epi16(f9_hi, f9_hi);  // Filter (hi) * 18
+    const __m128i f9_lo = _mm_mulhi_epi16(f_lo, k9);    // Filter (lo) * 9
+    const __m128i f9_hi = _mm_mulhi_epi16(f_hi, k9);    // Filter (hi) * 9
 
-    const __m128i a2_lo = _mm_add_epi16(f9_lo, sixty_three);  // Filter * 9 + 63
-    const __m128i a2_hi = _mm_add_epi16(f9_hi, sixty_three);  // Filter * 9 + 63
+    const __m128i a2_lo = _mm_add_epi16(f9_lo, k63);    // Filter * 9 + 63
+    const __m128i a2_hi = _mm_add_epi16(f9_hi, k63);    // Filter * 9 + 63
 
-    const __m128i a1_lo = _mm_add_epi16(f18_lo, sixty_three);  // F... * 18 + 63
-    const __m128i a1_hi = _mm_add_epi16(f18_hi, sixty_three);  // F... * 18 + 63
+    const __m128i a1_lo = _mm_add_epi16(a2_lo, f9_lo);  // Filter * 18 + 63
+    const __m128i a1_hi = _mm_add_epi16(a2_hi, f9_hi);  // Filter * 18 + 63
 
-    const __m128i a0_lo = _mm_add_epi16(f18_lo, a2_lo);  // Filter * 27 + 63
-    const __m128i a0_hi = _mm_add_epi16(f18_hi, a2_hi);  // Filter * 27 + 63
+    const __m128i a0_lo = _mm_add_epi16(a1_lo, f9_lo);  // Filter * 27 + 63
+    const __m128i a0_hi = _mm_add_epi16(a1_hi, f9_hi);  // Filter * 27 + 63
 
-    UPDATE_2PIXELS(*p2, *q2, a2_lo, a2_hi);
-    UPDATE_2PIXELS(*p1, *q1, a1_lo, a1_hi);
-    UPDATE_2PIXELS(*p0, *q0, a0_lo, a0_hi);
+    Update2Pixels(p2, q2, &a2_lo, &a2_hi);
+    Update2Pixels(p1, q1, &a1_lo, &a1_hi);
+    Update2Pixels(p0, q0, &a0_lo, &a0_hi);
   }
-
-  // unoffset
-  FLIP_SIGN_BIT4(*p1, *p0, *q0, *q1);
-  FLIP_SIGN_BIT2(*p2, *q2);
 }
 
 // reads 8 rows across a vertical edge.
 //
 // TODO(somnath): Investigate _mm_shuffle* also see if it can be broken into
 // two Load4x4() to avoid code duplication.
-static WEBP_INLINE void Load8x4(const uint8_t* b, int stride,
-                                __m128i* p, __m128i* q) {
+static WEBP_INLINE void Load8x4(const uint8_t* const b, int stride,
+                                __m128i* const p, __m128i* const q) {
   __m128i t1, t2;
 
   // Load 0th, 1st, 4th and 5th rows
@@ -509,10 +563,11 @@ static WEBP_INLINE void Load8x4(const uint8_t* b, int stride,
   *q = _mm_unpackhi_epi32(t1, t2);
 }
 
-static WEBP_INLINE void Load16x4(const uint8_t* r0, const uint8_t* r8,
+static WEBP_INLINE void Load16x4(const uint8_t* const r0,
+                                 const uint8_t* const r8,
                                  int stride,
-                                 __m128i* p1, __m128i* p0,
-                                 __m128i* q0, __m128i* q1) {
+                                 __m128i* const p1, __m128i* const p0,
+                                 __m128i* const q0, __m128i* const q1) {
   __m128i t1, t2;
   // Assume the pixels around the edge (|) are numbered as follows
   //                00 01 | 02 03
@@ -544,7 +599,7 @@ static WEBP_INLINE void Load16x4(const uint8_t* r0, const uint8_t* r8,
   *q1 = _mm_unpackhi_epi64(t2, *q1);
 }
 
-static WEBP_INLINE void Store4x4(__m128i* x, uint8_t* dst, int stride) {
+static WEBP_INLINE void Store4x4(__m128i* const x, uint8_t* dst, int stride) {
   int i;
   for (i = 0; i < 4; ++i, dst += stride) {
     *((int32_t*)dst) = _mm_cvtsi128_si32(*x);
@@ -553,48 +608,51 @@ static WEBP_INLINE void Store4x4(__m128i* x, uint8_t* dst, int stride) {
 }
 
 // Transpose back and store
-static WEBP_INLINE void Store16x4(uint8_t* r0, uint8_t* r8, int stride,
-                                  __m128i* p1, __m128i* p0,
-                                  __m128i* q0, __m128i* q1) {
-  __m128i t1;
+static WEBP_INLINE void Store16x4(const __m128i* const p1,
+                                  const __m128i* const p0,
+                                  const __m128i* const q0,
+                                  const __m128i* const q1,
+                                  uint8_t* r0, uint8_t* r8,
+                                  int stride) {
+  __m128i t1, p1_s, p0_s, q0_s, q1_s;
 
   // p0 = 71 70 61 60 51 50 41 40 31 30 21 20 11 10 01 00
   // p1 = f1 f0 e1 e0 d1 d0 c1 c0 b1 b0 a1 a0 91 90 81 80
   t1 = *p0;
-  *p0 = _mm_unpacklo_epi8(*p1, t1);
-  *p1 = _mm_unpackhi_epi8(*p1, t1);
+  p0_s = _mm_unpacklo_epi8(*p1, t1);
+  p1_s = _mm_unpackhi_epi8(*p1, t1);
 
   // q0 = 73 72 63 62 53 52 43 42 33 32 23 22 13 12 03 02
   // q1 = f3 f2 e3 e2 d3 d2 c3 c2 b3 b2 a3 a2 93 92 83 82
   t1 = *q0;
-  *q0 = _mm_unpacklo_epi8(t1, *q1);
-  *q1 = _mm_unpackhi_epi8(t1, *q1);
+  q0_s = _mm_unpacklo_epi8(t1, *q1);
+  q1_s = _mm_unpackhi_epi8(t1, *q1);
 
   // p0 = 33 32 31 30 23 22 21 20 13 12 11 10 03 02 01 00
   // q0 = 73 72 71 70 63 62 61 60 53 52 51 50 43 42 41 40
-  t1 = *p0;
-  *p0 = _mm_unpacklo_epi16(t1, *q0);
-  *q0 = _mm_unpackhi_epi16(t1, *q0);
+  t1 = p0_s;
+  p0_s = _mm_unpacklo_epi16(t1, q0_s);
+  q0_s = _mm_unpackhi_epi16(t1, q0_s);
 
   // p1 = b3 b2 b1 b0 a3 a2 a1 a0 93 92 91 90 83 82 81 80
   // q1 = f3 f2 f1 f0 e3 e2 e1 e0 d3 d2 d1 d0 c3 c2 c1 c0
-  t1 = *p1;
-  *p1 = _mm_unpacklo_epi16(t1, *q1);
-  *q1 = _mm_unpackhi_epi16(t1, *q1);
+  t1 = p1_s;
+  p1_s = _mm_unpacklo_epi16(t1, q1_s);
+  q1_s = _mm_unpackhi_epi16(t1, q1_s);
 
-  Store4x4(p0, r0, stride);
+  Store4x4(&p0_s, r0, stride);
   r0 += 4 * stride;
-  Store4x4(q0, r0, stride);
+  Store4x4(&q0_s, r0, stride);
 
-  Store4x4(p1, r8, stride);
+  Store4x4(&p1_s, r8, stride);
   r8 += 4 * stride;
-  Store4x4(q1, r8, stride);
+  Store4x4(&q1_s, r8, stride);
 }
 
 //------------------------------------------------------------------------------
 // Simple In-loop filtering (Paragraph 15.2)
 
-static void SimpleVFilter16SSE2(uint8_t* p, int stride, int thresh) {
+static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
   // Load
   __m128i p1 = _mm_loadu_si128((__m128i*)&p[-2 * stride]);
   __m128i p0 = _mm_loadu_si128((__m128i*)&p[-stride]);
@@ -605,49 +663,49 @@ static void SimpleVFilter16SSE2(uint8_t* p, int stride, int thresh) {
 
   // Store
   _mm_storeu_si128((__m128i*)&p[-stride], p0);
-  _mm_storeu_si128((__m128i*)p, q0);
+  _mm_storeu_si128((__m128i*)&p[0], q0);
 }
 
-static void SimpleHFilter16SSE2(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
   __m128i p1, p0, q0, q1;
 
   p -= 2;  // beginning of p1
 
-  Load16x4(p, p + 8 * stride,  stride, &p1, &p0, &q0, &q1);
+  Load16x4(p, p + 8 * stride, stride, &p1, &p0, &q0, &q1);
   DoFilter2(&p1, &p0, &q0, &q1, thresh);
-  Store16x4(p, p + 8 * stride, stride, &p1, &p0, &q0, &q1);
+  Store16x4(&p1, &p0, &q0, &q1, p, p + 8 * stride, stride);
 }
 
-static void SimpleVFilter16iSSE2(uint8_t* p, int stride, int thresh) {
+static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
   int k;
   for (k = 3; k > 0; --k) {
     p += 4 * stride;
-    SimpleVFilter16SSE2(p, stride, thresh);
+    SimpleVFilter16(p, stride, thresh);
   }
 }
 
-static void SimpleHFilter16iSSE2(uint8_t* p, int stride, int thresh) {
+static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
   int k;
   for (k = 3; k > 0; --k) {
     p += 4;
-    SimpleHFilter16SSE2(p, stride, thresh);
+    SimpleHFilter16(p, stride, thresh);
   }
 }
 
 //------------------------------------------------------------------------------
 // Complex In-loop filtering (Paragraph 15.3)
 
-#define MAX_DIFF1(p3, p2, p1, p0, m) {                                         \
-  m = MM_ABS(p3, p2);                                                          \
-  m = _mm_max_epu8(m, MM_ABS(p2, p1));                                         \
-  m = _mm_max_epu8(m, MM_ABS(p1, p0));                                         \
-}
-
-#define MAX_DIFF2(p3, p2, p1, p0, m) {                                         \
+#define MAX_DIFF1(p3, p2, p1, p0, m) do {                                      \
+  m = MM_ABS(p1, p0);                                                          \
   m = _mm_max_epu8(m, MM_ABS(p3, p2));                                         \
   m = _mm_max_epu8(m, MM_ABS(p2, p1));                                         \
+} while (0)
+
+#define MAX_DIFF2(p3, p2, p1, p0, m) do {                                      \
   m = _mm_max_epu8(m, MM_ABS(p1, p0));                                         \
-}
+  m = _mm_max_epu8(m, MM_ABS(p3, p2));                                         \
+  m = _mm_max_epu8(m, MM_ABS(p2, p1));                                         \
+} while (0)
 
 #define LOAD_H_EDGES4(p, stride, e1, e2, e3, e4) {                             \
   e1 = _mm_loadu_si128((__m128i*)&(p)[0 * stride]);                            \
@@ -656,10 +714,11 @@ static void SimpleHFilter16iSSE2(uint8_t* p, int stride, int thresh) {
   e4 = _mm_loadu_si128((__m128i*)&(p)[3 * stride]);                            \
 }
 
-#define LOADUV_H_EDGE(p, u, v, stride) {                                       \
-  p = _mm_loadl_epi64((__m128i*)&(u)[(stride)]);                               \
-  p = _mm_unpacklo_epi64(p, _mm_loadl_epi64((__m128i*)&(v)[(stride)]));        \
-}
+#define LOADUV_H_EDGE(p, u, v, stride) do {                                    \
+  const __m128i U = _mm_loadl_epi64((__m128i*)&(u)[(stride)]);                 \
+  const __m128i V = _mm_loadl_epi64((__m128i*)&(v)[(stride)]);                 \
+  p = _mm_unpacklo_epi64(U, V);                                                \
+} while (0)
 
 #define LOADUV_H_EDGES4(u, v, stride, e1, e2, e3, e4) {                        \
   LOADUV_H_EDGE(e1, u, v, 0 * stride);                                         \
@@ -674,18 +733,23 @@ static void SimpleHFilter16iSSE2(uint8_t* p, int stride, int thresh) {
   _mm_storel_epi64((__m128i*)&v[(stride)], p);                                 \
 }
 
-#define COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask) {               \
-  __m128i fl_yes;                                                              \
-  const __m128i it = _mm_set1_epi8(ithresh);                                   \
-  mask = _mm_subs_epu8(mask, it);                                              \
-  mask = _mm_cmpeq_epi8(mask, _mm_setzero_si128());                            \
-  NeedsFilter(&p1, &p0, &q0, &q1, thresh, &fl_yes);                            \
-  mask = _mm_and_si128(mask, fl_yes);                                          \
+static WEBP_INLINE void ComplexMask(const __m128i* const p1,
+                                    const __m128i* const p0,
+                                    const __m128i* const q0,
+                                    const __m128i* const q1,
+                                    int thresh, int ithresh,
+                                    __m128i* const mask) {
+  const __m128i it = _mm_set1_epi8(ithresh);
+  const __m128i diff = _mm_subs_epu8(*mask, it);
+  const __m128i thresh_mask = _mm_cmpeq_epi8(diff, _mm_setzero_si128());
+  __m128i filter_mask;
+  NeedsFilter(p1, p0, q0, q1, thresh, &filter_mask);
+  *mask = _mm_and_si128(thresh_mask, filter_mask);
 }
 
 // on macroblock edges
-static void VFilter16SSE2(uint8_t* p, int stride,
-                          int thresh, int ithresh, int hev_thresh) {
+static void VFilter16(uint8_t* p, int stride,
+                      int thresh, int ithresh, int hev_thresh) {
   __m128i t1;
   __m128i mask;
   __m128i p2, p1, p0, q0, q1, q2;
@@ -698,20 +762,20 @@ static void VFilter16SSE2(uint8_t* p, int stride,
   LOAD_H_EDGES4(p, stride, q0, q1, q2, t1);
   MAX_DIFF2(t1, q2, q1, q0, mask);
 
-  COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask);
+  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
   DoFilter6(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
 
   // Store
   _mm_storeu_si128((__m128i*)&p[-3 * stride], p2);
   _mm_storeu_si128((__m128i*)&p[-2 * stride], p1);
   _mm_storeu_si128((__m128i*)&p[-1 * stride], p0);
-  _mm_storeu_si128((__m128i*)&p[0 * stride], q0);
-  _mm_storeu_si128((__m128i*)&p[1 * stride], q1);
-  _mm_storeu_si128((__m128i*)&p[2 * stride], q2);
+  _mm_storeu_si128((__m128i*)&p[+0 * stride], q0);
+  _mm_storeu_si128((__m128i*)&p[+1 * stride], q1);
+  _mm_storeu_si128((__m128i*)&p[+2 * stride], q2);
 }
 
-static void HFilter16SSE2(uint8_t* p, int stride,
-                          int thresh, int ithresh, int hev_thresh) {
+static void HFilter16(uint8_t* p, int stride,
+                      int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i p3, p2, p1, p0, q0, q1, q2, q3;
 
@@ -722,71 +786,78 @@ static void HFilter16SSE2(uint8_t* p, int stride,
   Load16x4(p, p + 8 * stride, stride, &q0, &q1, &q2, &q3);  // q0, q1, q2, q3
   MAX_DIFF2(q3, q2, q1, q0, mask);
 
-  COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask);
+  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
   DoFilter6(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
 
-  Store16x4(b, b + 8 * stride, stride, &p3, &p2, &p1, &p0);
-  Store16x4(p, p + 8 * stride, stride, &q0, &q1, &q2, &q3);
+  Store16x4(&p3, &p2, &p1, &p0, b, b + 8 * stride, stride);
+  Store16x4(&q0, &q1, &q2, &q3, p, p + 8 * stride, stride);
 }
 
 // on three inner edges
-static void VFilter16iSSE2(uint8_t* p, int stride,
-                           int thresh, int ithresh, int hev_thresh) {
+static void VFilter16i(uint8_t* p, int stride,
+                       int thresh, int ithresh, int hev_thresh) {
   int k;
-  __m128i mask;
-  __m128i t1, t2, p1, p0, q0, q1;
+  __m128i p3, p2, p1, p0;   // loop invariants
+
+  LOAD_H_EDGES4(p, stride, p3, p2, p1, p0);  // prologue
 
   for (k = 3; k > 0; --k) {
-    // Load p3, p2, p1, p0
-    LOAD_H_EDGES4(p, stride, t2, t1, p1, p0);
-    MAX_DIFF1(t2, t1, p1, p0, mask);
-
+    __m128i mask, tmp1, tmp2;
+    uint8_t* const b = p + 2 * stride;   // beginning of p1
     p += 4 * stride;
 
-    // Load q0, q1, q2, q3
-    LOAD_H_EDGES4(p, stride, q0, q1, t1, t2);
-    MAX_DIFF2(t2, t1, q1, q0, mask);
+    MAX_DIFF1(p3, p2, p1, p0, mask);   // compute partial mask
+    LOAD_H_EDGES4(p, stride, p3, p2, tmp1, tmp2);
+    MAX_DIFF2(p3, p2, tmp1, tmp2, mask);
 
-    COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask);
-    DoFilter4(&p1, &p0, &q0, &q1, &mask, hev_thresh);
+    // p3 and p2 are not just temporary variables here: they will be
+    // re-used for next span. And q2/q3 will become p1/p0 accordingly.
+    ComplexMask(&p1, &p0, &p3, &p2, thresh, ithresh, &mask);
+    DoFilter4(&p1, &p0, &p3, &p2, &mask, hev_thresh);
 
     // Store
-    _mm_storeu_si128((__m128i*)&p[-2 * stride], p1);
-    _mm_storeu_si128((__m128i*)&p[-1 * stride], p0);
-    _mm_storeu_si128((__m128i*)&p[0 * stride], q0);
-    _mm_storeu_si128((__m128i*)&p[1 * stride], q1);
+    _mm_storeu_si128((__m128i*)&b[0 * stride], p1);
+    _mm_storeu_si128((__m128i*)&b[1 * stride], p0);
+    _mm_storeu_si128((__m128i*)&b[2 * stride], p3);
+    _mm_storeu_si128((__m128i*)&b[3 * stride], p2);
+
+    // rotate samples
+    p1 = tmp1;
+    p0 = tmp2;
   }
 }
 
-static void HFilter16iSSE2(uint8_t* p, int stride,
-                           int thresh, int ithresh, int hev_thresh) {
+static void HFilter16i(uint8_t* p, int stride,
+                       int thresh, int ithresh, int hev_thresh) {
   int k;
-  uint8_t* b;
-  __m128i mask;
-  __m128i t1, t2, p1, p0, q0, q1;
+  __m128i p3, p2, p1, p0;   // loop invariants
+
+  Load16x4(p, p + 8 * stride, stride, &p3, &p2, &p1, &p0);  // prologue
 
   for (k = 3; k > 0; --k) {
-    b = p;
-    Load16x4(b, b + 8 * stride, stride, &t2, &t1, &p1, &p0);  // p3, p2, p1, p0
-    MAX_DIFF1(t2, t1, p1, p0, mask);
+    __m128i mask, tmp1, tmp2;
+    uint8_t* const b = p + 2;   // beginning of p1
 
-    b += 4;  // beginning of q0
-    Load16x4(b, b + 8 * stride, stride, &q0, &q1, &t1, &t2);  // q0, q1, q2, q3
-    MAX_DIFF2(t2, t1, q1, q0, mask);
+    p += 4;  // beginning of q0 (and next span)
 
-    COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask);
-    DoFilter4(&p1, &p0, &q0, &q1, &mask, hev_thresh);
+    MAX_DIFF1(p3, p2, p1, p0, mask);   // compute partial mask
+    Load16x4(p, p + 8 * stride, stride, &p3, &p2, &tmp1, &tmp2);
+    MAX_DIFF2(p3, p2, tmp1, tmp2, mask);
 
-    b -= 2;  // beginning of p1
-    Store16x4(b, b + 8 * stride, stride, &p1, &p0, &q0, &q1);
+    ComplexMask(&p1, &p0, &p3, &p2, thresh, ithresh, &mask);
+    DoFilter4(&p1, &p0, &p3, &p2, &mask, hev_thresh);
 
-    p += 4;
+    Store16x4(&p1, &p0, &p3, &p2, b, b + 8 * stride, stride);
+
+    // rotate samples
+    p1 = tmp1;
+    p0 = tmp2;
   }
 }
 
 // 8-pixels wide variant, for chroma filtering
-static void VFilter8SSE2(uint8_t* u, uint8_t* v, int stride,
-                         int thresh, int ithresh, int hev_thresh) {
+static void VFilter8(uint8_t* u, uint8_t* v, int stride,
+                     int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i t1, p2, p1, p0, q0, q1, q2;
 
@@ -798,7 +869,7 @@ static void VFilter8SSE2(uint8_t* u, uint8_t* v, int stride,
   LOADUV_H_EDGES4(u, v, stride, q0, q1, q2, t1);
   MAX_DIFF2(t1, q2, q1, q0, mask);
 
-  COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask);
+  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
   DoFilter6(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
 
   // Store
@@ -810,8 +881,8 @@ static void VFilter8SSE2(uint8_t* u, uint8_t* v, int stride,
   STOREUV(q2, u, v, 2 * stride);
 }
 
-static void HFilter8SSE2(uint8_t* u, uint8_t* v, int stride,
-                         int thresh, int ithresh, int hev_thresh) {
+static void HFilter8(uint8_t* u, uint8_t* v, int stride,
+                     int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i p3, p2, p1, p0, q0, q1, q2, q3;
 
@@ -823,15 +894,15 @@ static void HFilter8SSE2(uint8_t* u, uint8_t* v, int stride,
   Load16x4(u, v, stride, &q0, &q1, &q2, &q3);    // q0, q1, q2, q3
   MAX_DIFF2(q3, q2, q1, q0, mask);
 
-  COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask);
+  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
   DoFilter6(&p2, &p1, &p0, &q0, &q1, &q2, &mask, hev_thresh);
 
-  Store16x4(tu, tv, stride, &p3, &p2, &p1, &p0);
-  Store16x4(u, v, stride, &q0, &q1, &q2, &q3);
+  Store16x4(&p3, &p2, &p1, &p0, tu, tv, stride);
+  Store16x4(&q0, &q1, &q2, &q3, u, v, stride);
 }
 
-static void VFilter8iSSE2(uint8_t* u, uint8_t* v, int stride,
-                          int thresh, int ithresh, int hev_thresh) {
+static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
+                      int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i t1, t2, p1, p0, q0, q1;
 
@@ -846,7 +917,7 @@ static void VFilter8iSSE2(uint8_t* u, uint8_t* v, int stride,
   LOADUV_H_EDGES4(u, v, stride, q0, q1, t1, t2);
   MAX_DIFF2(t2, t1, q1, q0, mask);
 
-  COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask);
+  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
   DoFilter4(&p1, &p0, &q0, &q1, &mask, hev_thresh);
 
   // Store
@@ -856,8 +927,8 @@ static void VFilter8iSSE2(uint8_t* u, uint8_t* v, int stride,
   STOREUV(q1, u, v, 1 * stride);
 }
 
-static void HFilter8iSSE2(uint8_t* u, uint8_t* v, int stride,
-                          int thresh, int ithresh, int hev_thresh) {
+static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
+                      int thresh, int ithresh, int hev_thresh) {
   __m128i mask;
   __m128i t1, t2, p1, p0, q0, q1;
   Load16x4(u, v, stride, &t2, &t1, &p1, &p0);   // p3, p2, p1, p0
@@ -868,12 +939,12 @@ static void HFilter8iSSE2(uint8_t* u, uint8_t* v, int stride,
   Load16x4(u, v, stride, &q0, &q1, &t1, &t2);  // q0, q1, q2, q3
   MAX_DIFF2(t2, t1, q1, q0, mask);
 
-  COMPLEX_FL_MASK(p1, p0, q0, q1, thresh, ithresh, mask);
+  ComplexMask(&p1, &p0, &q0, &q1, thresh, ithresh, &mask);
   DoFilter4(&p1, &p0, &q0, &q1, &mask, hev_thresh);
 
   u -= 2;  // beginning of p1
   v -= 2;
-  Store16x4(u, v, stride, &p1, &p0, &q0, &q1);
+  Store16x4(&p1, &p0, &q0, &q1, u, v, stride);
 }
 
 #endif   // WEBP_USE_SSE2
@@ -885,24 +956,23 @@ extern void VP8DspInitSSE2(void);
 
 void VP8DspInitSSE2(void) {
 #if defined(WEBP_USE_SSE2)
-  VP8Transform = TransformSSE2;
+  VP8Transform = Transform;
+#if defined(USE_TRANSFORM_AC3)
+  VP8TransformAC3 = TransformAC3;
+#endif
 
-  VP8VFilter16 = VFilter16SSE2;
-  VP8HFilter16 = HFilter16SSE2;
-  VP8VFilter8 = VFilter8SSE2;
-  VP8HFilter8 = HFilter8SSE2;
-  VP8VFilter16i = VFilter16iSSE2;
-  VP8HFilter16i = HFilter16iSSE2;
-  VP8VFilter8i = VFilter8iSSE2;
-  VP8HFilter8i = HFilter8iSSE2;
+  VP8VFilter16 = VFilter16;
+  VP8HFilter16 = HFilter16;
+  VP8VFilter8 = VFilter8;
+  VP8HFilter8 = HFilter8;
+  VP8VFilter16i = VFilter16i;
+  VP8HFilter16i = HFilter16i;
+  VP8VFilter8i = VFilter8i;
+  VP8HFilter8i = HFilter8i;
 
-  VP8SimpleVFilter16 = SimpleVFilter16SSE2;
-  VP8SimpleHFilter16 = SimpleHFilter16SSE2;
-  VP8SimpleVFilter16i = SimpleVFilter16iSSE2;
-  VP8SimpleHFilter16i = SimpleHFilter16iSSE2;
+  VP8SimpleVFilter16 = SimpleVFilter16;
+  VP8SimpleHFilter16 = SimpleHFilter16;
+  VP8SimpleVFilter16i = SimpleVFilter16i;
+  VP8SimpleHFilter16i = SimpleHFilter16i;
 #endif   // WEBP_USE_SSE2
 }
-
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dsp/dsp.h

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 //   Speed-critical functions.
@@ -12,35 +14,76 @@
 #ifndef WEBP_DSP_DSP_H_
 #define WEBP_DSP_DSP_H_
 
+#ifdef HAVE_CONFIG_H
+#include "../webp/config.h"
+#endif
+
 #include "../webp/types.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 extern "C" {
 #endif
 
 //------------------------------------------------------------------------------
 // CPU detection
 
-#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
+#if defined(__GNUC__)
+# define LOCAL_GCC_VERSION ((__GNUC__ << 8) | __GNUC_MINOR__)
+# define LOCAL_GCC_PREREQ(maj, min) \
+    (LOCAL_GCC_VERSION >= (((maj) << 8) | (min)))
+#else
+# define LOCAL_GCC_VERSION 0
+# define LOCAL_GCC_PREREQ(maj, min) 0
+#endif
+
+#ifdef __clang__
+# define LOCAL_CLANG_VERSION ((__clang_major__ << 8) | __clang_minor__)
+# define LOCAL_CLANG_PREREQ(maj, min) \
+    (LOCAL_CLANG_VERSION >= (((maj) << 8) | (min)))
+#else
+# define LOCAL_CLANG_VERSION 0
+# define LOCAL_CLANG_PREREQ(maj, min) 0
+#endif  // __clang__
+
+#if defined(_MSC_VER) && _MSC_VER > 1310 && \
+    (defined(_M_X64) || defined(_M_IX86))
 #define WEBP_MSC_SSE2  // Visual C++ SSE2 targets
 #endif
 
-#if defined(__SSE2__) || defined(WEBP_MSC_SSE2)
+// WEBP_HAVE_* are used to indicate the presence of the instruction set in dsp
+// files without intrinsics, allowing the corresponding Init() to be called.
+// Files containing intrinsics will need to be built targeting the instruction
+// set so should succeed on one of the earlier tests.
+#if defined(__SSE2__) || defined(WEBP_MSC_SSE2) || defined(WEBP_HAVE_SSE2)
 #define WEBP_USE_SSE2
 #endif
 
+#if defined(__AVX2__) || defined(WEBP_HAVE_AVX2)
+#define WEBP_USE_AVX2
+#endif
+
 #if defined(__ANDROID__) && defined(__ARM_ARCH_7A__)
 #define WEBP_ANDROID_NEON  // Android targets that might support NEON
 #endif
 
-#if defined(__ARM_NEON__) || defined(WEBP_ANDROID_NEON)
+#if defined(__ARM_NEON__) || defined(WEBP_ANDROID_NEON) || defined(__aarch64__)
 #define WEBP_USE_NEON
 #endif
 
+#if defined(__mips__) && !defined(__mips64) && (__mips_isa_rev < 6)
+#define WEBP_USE_MIPS32
+#if (__mips_isa_rev >= 2)
+#define WEBP_USE_MIPS32_R2
+#endif
+#endif
+
 typedef enum {
   kSSE2,
   kSSE3,
-  kNEON
+  kAVX,
+  kAVX2,
+  kNEON,
+  kMIPS32
 } CPUFeature;
 // returns true if the CPU supports the feature.
 typedef int (*VP8CPUInfo)(CPUFeature feature);
@@ -49,8 +92,6 @@ extern VP8CPUInfo VP8GetCPUInfo;
 //------------------------------------------------------------------------------
 // Encoding
 
-int VP8GetAlpha(const int histo[]);
-
 // Transforms
 // VP8Idct: Does one of two inverse transforms. If do_two is set, the transforms
 //          will be done for (ref, in, dst) and (ref + 4, in + 16, dst + 4).
@@ -60,7 +101,6 @@ typedef void (*VP8Fdct)(const uint8_t* src, const uint8_t* ref, int16_t* out);
 typedef void (*VP8WHT)(const int16_t* in, int16_t* out);
 extern VP8Idct VP8ITransform;
 extern VP8Fdct VP8FTransform;
-extern VP8WHT VP8ITransformWHT;
 extern VP8WHT VP8FTransformWHT;
 // Predictions
 // *dst is the destination block. *top and *left can be NULL.
@@ -82,13 +122,19 @@ extern VP8BlockCopy VP8Copy4x4;
 // Quantization
 struct VP8Matrix;   // forward declaration
 typedef int (*VP8QuantizeBlock)(int16_t in[16], int16_t out[16],
-                                int n, const struct VP8Matrix* const mtx);
+                                const struct VP8Matrix* const mtx);
 extern VP8QuantizeBlock VP8EncQuantizeBlock;
 
-// Compute susceptibility based on DCT-coeff histograms:
-// the higher, the "easier" the macroblock is to compress.
-typedef int (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred,
-                         int start_block, int end_block);
+// specific to 2nd transform:
+typedef int (*VP8QuantizeBlockWHT)(int16_t in[16], int16_t out[16],
+                                   const struct VP8Matrix* const mtx);
+extern VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT;
+
+// Collect histogram for susceptibility calculation and accumulate in histo[].
+struct VP8Histogram;
+typedef void (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred,
+                          int start_block, int end_block,
+                          struct VP8Histogram* const histo);
 extern const int VP8DspScan[16 + 4 + 4];
 extern VP8CHisto VP8CollectHistogram;
 
@@ -101,10 +147,11 @@ typedef void (*VP8DecIdct)(const int16_t* coeffs, uint8_t* dst);
 // when doing two transforms, coeffs is actually int16_t[2][16].
 typedef void (*VP8DecIdct2)(const int16_t* coeffs, uint8_t* dst, int do_two);
 extern VP8DecIdct2 VP8Transform;
+extern VP8DecIdct VP8TransformAC3;
 extern VP8DecIdct VP8TransformUV;
 extern VP8DecIdct VP8TransformDC;
 extern VP8DecIdct VP8TransformDCUV;
-extern void (*VP8TransformWHT)(const int16_t* in, int16_t* out);
+extern VP8WHT VP8TransformWHT;
 
 // *dst is the destination block, with stride BPS. Boundary samples are
 // assumed accessible when needed.
@@ -113,6 +160,13 @@ extern const VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */];
 extern const VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */];
 extern const VP8PredFunc VP8PredLuma4[/* NUM_BMODES */];
 
+// clipping tables (for filtering)
+extern const int8_t* const VP8ksclip1;  // clips [-1020, 1020] to [-128, 127]
+extern const int8_t* const VP8ksclip2;  // clips [-112, 112] to [-16, 15]
+extern const uint8_t* const VP8kclip1;  // clips [-255,511] to [0,255]
+extern const uint8_t* const VP8kabs0;   // abs(x) for x in [-255,255]
+void VP8InitClipTables(void);           // must be called first
+
 // simple filter (only for luma)
 typedef void (*VP8SimpleFilterFunc)(uint8_t* p, int stride, int thresh);
 extern VP8SimpleFilterFunc VP8SimpleVFilter16;
@@ -145,6 +199,8 @@ void VP8DspInit(void);
 
 #define FANCY_UPSAMPLING   // undefined to remove fancy upsampling support
 
+// Convert a pair of y/u/v lines together to the output rgb/a colorspace.
+// bottom_y can be NULL if only one line of output is needed (at top/bottom).
 typedef void (*WebPUpsampleLinePairFunc)(
     const uint8_t* top_y, const uint8_t* bottom_y,
     const uint8_t* top_u, const uint8_t* top_v,
@@ -156,18 +212,20 @@ typedef void (*WebPUpsampleLinePairFunc)(
 // Fancy upsampling functions to convert YUV to RGB(A) modes
 extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
 
-// Initializes SSE2 version of the fancy upsamplers.
-void WebPInitUpsamplersSSE2(void);
-
 #endif    // FANCY_UPSAMPLING
 
-// Point-sampling methods.
-typedef void (*WebPSampleLinePairFunc)(
-    const uint8_t* top_y, const uint8_t* bottom_y,
-    const uint8_t* u, const uint8_t* v,
-    uint8_t* top_dst, uint8_t* bottom_dst, int len);
+// Per-row point-sampling methods.
+typedef void (*WebPSamplerRowFunc)(const uint8_t* y,
+                                   const uint8_t* u, const uint8_t* v,
+                                   uint8_t* dst, int len);
+// Generic function to apply 'WebPSamplerRowFunc' to the whole plane:
+void WebPSamplerProcessPlane(const uint8_t* y, int y_stride,
+                             const uint8_t* u, const uint8_t* v, int uv_stride,
+                             uint8_t* dst, int dst_stride,
+                             int width, int height, WebPSamplerRowFunc func);
 
-extern const WebPSampleLinePairFunc WebPSamplers[/* MODE_LAST */];
+// Sampling functions to convert rows of YUV to RGB(A)
+extern WebPSamplerRowFunc WebPSamplers[/* MODE_LAST */];
 
 // General function for converting two lines of ARGB or RGBA.
 // 'alpha_is_last' should be true if 0xff000000 is stored in memory as
@@ -181,11 +239,14 @@ typedef void (*WebPYUV444Converter)(const uint8_t* y,
 
 extern const WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */];
 
-// Main function to be called
+// Must be called before using the WebPUpsamplers[] (and for premultiplied
+// colorspaces like rgbA, rgbA4444, etc)
 void WebPInitUpsamplers(void);
+// Must be called before using WebPSamplers[]
+void WebPInitSamplers(void);
 
 //------------------------------------------------------------------------------
-// Pre-multiply planes with alpha values
+// Utilities for processing transparent channel.
 
 // Apply alpha pre-multiply on an rgba, bgra or argb plane of size w * h.
 // alpha_first should be 0 for argb, 1 for rgba or bgra (where alpha is last).
@@ -196,14 +257,36 @@ extern void (*WebPApplyAlphaMultiply)(
 extern void (*WebPApplyAlphaMultiply4444)(
     uint8_t* rgba4444, int w, int h, int stride);
 
+// Extract the alpha values from 32b values in argb[] and pack them into alpha[]
+// (this is the opposite of WebPDispatchAlpha).
+// Returns true if there's only trivial 0xff alpha values.
+extern int (*WebPExtractAlpha)(const uint8_t* argb, int argb_stride,
+                               int width, int height,
+                               uint8_t* alpha, int alpha_stride);
+
+// Pre-Multiply operation transforms x into x * A / 255  (where x=Y,R,G or B).
+// Un-Multiply operation transforms x into x * 255 / A.
+
+// Pre-Multiply or Un-Multiply (if 'inverse' is true) argb values in a row.
+extern void (*WebPMultARGBRow)(uint32_t* const ptr, int width, int inverse);
+
+// Same a WebPMultARGBRow(), but for several rows.
+void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows,
+                      int inverse);
+
+// Same for a row of single values, with side alpha values.
+extern void (*WebPMultRow)(uint8_t* const ptr, const uint8_t* const alpha,
+                           int width, int inverse);
+
+// Same a WebPMultRow(), but for several 'num_rows' rows.
+void WebPMultRows(uint8_t* ptr, int stride,
+                  const uint8_t* alpha, int alpha_stride,
+                  int width, int num_rows, int inverse);
+
 // To be called first before using the above.
-void WebPInitPremultiply(void);
+void WebPInitAlphaProcessing(void);
 
-void WebPInitPremultiplySSE2(void);   // should not be called directly.
-
-//------------------------------------------------------------------------------
-
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 

src/dsp/enc.c

@@ -1,47 +1,34 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Speed-critical encoding functions.
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
+#include <assert.h>
 #include <stdlib.h>  // for abs()
+
 #include "./dsp.h"
 #include "../enc/vp8enci.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
+static WEBP_INLINE uint8_t clip_8b(int v) {
+  return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
+}
+
+static WEBP_INLINE int clip_max(int v, int max) {
+  return (v > max) ? max : v;
+}
 
 //------------------------------------------------------------------------------
 // Compute susceptibility based on DCT-coeff histograms:
 // the higher, the "easier" the macroblock is to compress.
 
-static int ClipAlpha(int alpha) {
-  return alpha < 0 ? 0 : alpha > 255 ? 255 : alpha;
-}
-
-int VP8GetAlpha(const int histo[MAX_COEFF_THRESH + 1]) {
-  int num = 0, den = 0, val = 0;
-  int k;
-  int alpha;
-  // note: changing this loop to avoid the numerous "k + 1" slows things down.
-  for (k = 0; k < MAX_COEFF_THRESH; ++k) {
-    if (histo[k + 1]) {
-      val += histo[k + 1];
-      num += val * (k + 1);
-      den += (k + 1) * (k + 1);
-    }
-  }
-  // we scale the value to a usable [0..255] range
-  alpha = den ? 10 * num / den - 5 : 0;
-  return ClipAlpha(alpha);
-}
-
 const int VP8DspScan[16 + 4 + 4] = {
   // Luma
   0 +  0 * BPS,  4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS,
@@ -53,27 +40,23 @@ const int VP8DspScan[16 + 4 + 4] = {
   8 + 0 * BPS,  12 + 0 * BPS, 8 + 4 * BPS, 12 + 4 * BPS     // V
 };
 
-static int CollectHistogram(const uint8_t* ref, const uint8_t* pred,
-                            int start_block, int end_block) {
-  int histo[MAX_COEFF_THRESH + 1] = { 0 };
-  int16_t out[16];
-  int j, k;
+static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
+                             int start_block, int end_block,
+                             VP8Histogram* const histo) {
+  int j;
   for (j = start_block; j < end_block; ++j) {
+    int k;
+    int16_t out[16];
+
     VP8FTransform(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
 
-    // Convert coefficients to bin (within out[]).
+    // Convert coefficients to bin.
     for (k = 0; k < 16; ++k) {
-      const int v = abs(out[k]) >> 2;
-      out[k] = (v > MAX_COEFF_THRESH) ? MAX_COEFF_THRESH : v;
-    }
-
-    // Use bin to update histogram.
-    for (k = 0; k < 16; ++k) {
-      histo[out[k]]++;
+      const int v = abs(out[k]) >> 3;  // TODO(skal): add rounding?
+      const int clipped_value = clip_max(v, MAX_COEFF_THRESH);
+      histo->distribution[clipped_value]++;
     }
   }
-
-  return VP8GetAlpha(histo);
 }
 
 //------------------------------------------------------------------------------
@@ -89,15 +72,12 @@ static void InitTables(void) {
   if (!tables_ok) {
     int i;
     for (i = -255; i <= 255 + 255; ++i) {
-      clip1[255 + i] = (i < 0) ? 0 : (i > 255) ? 255 : i;
+      clip1[255 + i] = clip_8b(i);
     }
     tables_ok = 1;
   }
 }
 
-static WEBP_INLINE uint8_t clip_8b(int v) {
-  return (!(v & ~0xff)) ? v : v < 0 ? 0 : 255;
-}
 
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
@@ -154,84 +134,58 @@ static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
   int i;
   int tmp[16];
   for (i = 0; i < 4; ++i, src += BPS, ref += BPS) {
-    const int d0 = src[0] - ref[0];
+    const int d0 = src[0] - ref[0];   // 9bit dynamic range ([-255,255])
     const int d1 = src[1] - ref[1];
     const int d2 = src[2] - ref[2];
     const int d3 = src[3] - ref[3];
-    const int a0 = (d0 + d3) << 3;
-    const int a1 = (d1 + d2) << 3;
-    const int a2 = (d1 - d2) << 3;
-    const int a3 = (d0 - d3) << 3;
-    tmp[0 + i * 4] = (a0 + a1);
-    tmp[1 + i * 4] = (a2 * 2217 + a3 * 5352 + 14500) >> 12;
-    tmp[2 + i * 4] = (a0 - a1);
-    tmp[3 + i * 4] = (a3 * 2217 - a2 * 5352 +  7500) >> 12;
+    const int a0 = (d0 + d3);         // 10b                      [-510,510]
+    const int a1 = (d1 + d2);
+    const int a2 = (d1 - d2);
+    const int a3 = (d0 - d3);
+    tmp[0 + i * 4] = (a0 + a1) * 8;   // 14b                      [-8160,8160]
+    tmp[1 + i * 4] = (a2 * 2217 + a3 * 5352 + 1812) >> 9;      // [-7536,7542]
+    tmp[2 + i * 4] = (a0 - a1) * 8;
+    tmp[3 + i * 4] = (a3 * 2217 - a2 * 5352 +  937) >> 9;
   }
   for (i = 0; i < 4; ++i) {
-    const int a0 = (tmp[0 + i] + tmp[12 + i]);
+    const int a0 = (tmp[0 + i] + tmp[12 + i]);  // 15b
     const int a1 = (tmp[4 + i] + tmp[ 8 + i]);
     const int a2 = (tmp[4 + i] - tmp[ 8 + i]);
     const int a3 = (tmp[0 + i] - tmp[12 + i]);
-    out[0 + i] = (a0 + a1 + 7) >> 4;
+    out[0 + i] = (a0 + a1 + 7) >> 4;            // 12b
     out[4 + i] = ((a2 * 2217 + a3 * 5352 + 12000) >> 16) + (a3 != 0);
     out[8 + i] = (a0 - a1 + 7) >> 4;
     out[12+ i] = ((a3 * 2217 - a2 * 5352 + 51000) >> 16);
   }
 }
 
-static void ITransformWHT(const int16_t* in, int16_t* out) {
-  int tmp[16];
-  int i;
-  for (i = 0; i < 4; ++i) {
-    const int a0 = in[0 + i] + in[12 + i];
-    const int a1 = in[4 + i] + in[ 8 + i];
-    const int a2 = in[4 + i] - in[ 8 + i];
-    const int a3 = in[0 + i] - in[12 + i];
-    tmp[0  + i] = a0 + a1;
-    tmp[8  + i] = a0 - a1;
-    tmp[4  + i] = a3 + a2;
-    tmp[12 + i] = a3 - a2;
-  }
-  for (i = 0; i < 4; ++i) {
-    const int dc = tmp[0 + i * 4] + 3;    // w/ rounder
-    const int a0 = dc             + tmp[3 + i * 4];
-    const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4];
-    const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4];
-    const int a3 = dc             - tmp[3 + i * 4];
-    out[ 0] = (a0 + a1) >> 3;
-    out[16] = (a3 + a2) >> 3;
-    out[32] = (a0 - a1) >> 3;
-    out[48] = (a3 - a2) >> 3;
-    out += 64;
-  }
-}
-
 static void FTransformWHT(const int16_t* in, int16_t* out) {
-  int tmp[16];
+  // input is 12b signed
+  int32_t tmp[16];
   int i;
   for (i = 0; i < 4; ++i, in += 64) {
-    const int a0 = (in[0 * 16] + in[2 * 16]) << 2;
-    const int a1 = (in[1 * 16] + in[3 * 16]) << 2;
-    const int a2 = (in[1 * 16] - in[3 * 16]) << 2;
-    const int a3 = (in[0 * 16] - in[2 * 16]) << 2;
-    tmp[0 + i * 4] = (a0 + a1) + (a0 != 0);
+    const int a0 = (in[0 * 16] + in[2 * 16]);  // 13b
+    const int a1 = (in[1 * 16] + in[3 * 16]);
+    const int a2 = (in[1 * 16] - in[3 * 16]);
+    const int a3 = (in[0 * 16] - in[2 * 16]);
+    tmp[0 + i * 4] = a0 + a1;   // 14b
     tmp[1 + i * 4] = a3 + a2;
     tmp[2 + i * 4] = a3 - a2;
     tmp[3 + i * 4] = a0 - a1;
   }
   for (i = 0; i < 4; ++i) {
-    const int a0 = (tmp[0 + i] + tmp[8 + i]);
+    const int a0 = (tmp[0 + i] + tmp[8 + i]);  // 15b
     const int a1 = (tmp[4 + i] + tmp[12+ i]);
     const int a2 = (tmp[4 + i] - tmp[12+ i]);
     const int a3 = (tmp[0 + i] - tmp[8 + i]);
-    const int b0 = a0 + a1;
+    const int b0 = a0 + a1;    // 16b
     const int b1 = a3 + a2;
     const int b2 = a3 - a2;
     const int b3 = a0 - a1;
-    out[ 0 + i] = (b0 + (b0 > 0) + 3) >> 3;
-    out[ 4 + i] = (b1 + (b1 > 0) + 3) >> 3;
-    out[ 8 + i] = (b2 + (b2 > 0) + 3) >> 3;
-    out[12 + i] = (b3 + (b3 > 0) + 3) >> 3;
+    out[ 0 + i] = b0 >> 1;     // 15b
+    out[ 4 + i] = b1 >> 1;
+    out[ 8 + i] = b2 >> 1;
+    out[12 + i] = b3 >> 1;
   }
 }
 
@@ -589,30 +543,30 @@ static int TTransform(const uint8_t* in, const uint16_t* w) {
   int i;
   // horizontal pass
   for (i = 0; i < 4; ++i, in += BPS) {
-    const int a0 = (in[0] + in[2]) << 2;
-    const int a1 = (in[1] + in[3]) << 2;
-    const int a2 = (in[1] - in[3]) << 2;
-    const int a3 = (in[0] - in[2]) << 2;
-    tmp[0 + i * 4] = a0 + a1 + (a0 != 0);
+    const int a0 = in[0] + in[2];
+    const int a1 = in[1] + in[3];
+    const int a2 = in[1] - in[3];
+    const int a3 = in[0] - in[2];
+    tmp[0 + i * 4] = a0 + a1;
     tmp[1 + i * 4] = a3 + a2;
     tmp[2 + i * 4] = a3 - a2;
     tmp[3 + i * 4] = a0 - a1;
   }
   // vertical pass
   for (i = 0; i < 4; ++i, ++w) {
-    const int a0 = (tmp[0 + i] + tmp[8 + i]);
-    const int a1 = (tmp[4 + i] + tmp[12+ i]);
-    const int a2 = (tmp[4 + i] - tmp[12+ i]);
-    const int a3 = (tmp[0 + i] - tmp[8 + i]);
+    const int a0 = tmp[0 + i] + tmp[8 + i];
+    const int a1 = tmp[4 + i] + tmp[12+ i];
+    const int a2 = tmp[4 + i] - tmp[12+ i];
+    const int a3 = tmp[0 + i] - tmp[8 + i];
     const int b0 = a0 + a1;
     const int b1 = a3 + a2;
     const int b2 = a3 - a2;
     const int b3 = a0 - a1;
-    // abs((b + (b<0) + 3) >> 3) = (abs(b) + 3) >> 3
-    sum += w[ 0] * ((abs(b0) + 3) >> 3);
-    sum += w[ 4] * ((abs(b1) + 3) >> 3);
-    sum += w[ 8] * ((abs(b2) + 3) >> 3);
-    sum += w[12] * ((abs(b3) + 3) >> 3);
+
+    sum += w[ 0] * abs(b0);
+    sum += w[ 4] * abs(b1);
+    sum += w[ 8] * abs(b2);
+    sum += w[12] * abs(b3);
   }
   return sum;
 }
@@ -621,7 +575,7 @@ static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
                     const uint16_t* const w) {
   const int sum1 = TTransform(a, w);
   const int sum2 = TTransform(b, w);
-  return (abs(sum2 - sum1) + 8) >> 4;
+  return abs(sum2 - sum1) >> 5;
 }
 
 static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
@@ -646,21 +600,49 @@ static const uint8_t kZigzag[16] = {
 
 // Simple quantization
 static int QuantizeBlock(int16_t in[16], int16_t out[16],
-                         int n, const VP8Matrix* const mtx) {
+                         const VP8Matrix* const mtx) {
   int last = -1;
-  for (; n < 16; ++n) {
+  int n;
+  for (n = 0; n < 16; ++n) {
     const int j = kZigzag[n];
     const int sign = (in[j] < 0);
-    int coeff = (sign ? -in[j] : in[j]) + mtx->sharpen_[j];
-    if (coeff > 2047) coeff = 2047;
+    const uint32_t coeff = (sign ? -in[j] : in[j]) + mtx->sharpen_[j];
     if (coeff > mtx->zthresh_[j]) {
-      const int Q = mtx->q_[j];
-      const int iQ = mtx->iq_[j];
-      const int B = mtx->bias_[j];
-      out[n] = QUANTDIV(coeff, iQ, B);
-      if (sign) out[n] = -out[n];
-      in[j] = out[n] * Q;
-      if (out[n]) last = n;
+      const uint32_t Q = mtx->q_[j];
+      const uint32_t iQ = mtx->iq_[j];
+      const uint32_t B = mtx->bias_[j];
+      int level = QUANTDIV(coeff, iQ, B);
+      if (level > MAX_LEVEL) level = MAX_LEVEL;
+      if (sign) level = -level;
+      in[j] = level * Q;
+      out[n] = level;
+      if (level) last = n;
+    } else {
+      out[n] = 0;
+      in[j] = 0;
+    }
+  }
+  return (last >= 0);
+}
+
+static int QuantizeBlockWHT(int16_t in[16], int16_t out[16],
+                            const VP8Matrix* const mtx) {
+  int n, last = -1;
+  for (n = 0; n < 16; ++n) {
+    const int j = kZigzag[n];
+    const int sign = (in[j] < 0);
+    const uint32_t coeff = sign ? -in[j] : in[j];
+    assert(mtx->sharpen_[j] == 0);
+    if (coeff > mtx->zthresh_[j]) {
+      const uint32_t Q = mtx->q_[j];
+      const uint32_t iQ = mtx->iq_[j];
+      const uint32_t B = mtx->bias_[j];
+      int level = QUANTDIV(coeff, iQ, B);
+      if (level > MAX_LEVEL) level = MAX_LEVEL;
+      if (sign) level = -level;
+      in[j] = level * Q;
+      out[n] = level;
+      if (level) last = n;
     } else {
       out[n] = 0;
       in[j] = 0;
@@ -691,7 +673,6 @@ static void Copy4x4(const uint8_t* src, uint8_t* dst) { Copy(src, dst, 4); }
 VP8CHisto VP8CollectHistogram;
 VP8Idct VP8ITransform;
 VP8Fdct VP8FTransform;
-VP8WHT VP8ITransformWHT;
 VP8WHT VP8FTransformWHT;
 VP8Intra4Preds VP8EncPredLuma4;
 VP8IntraPreds VP8EncPredLuma16;
@@ -703,18 +684,22 @@ VP8Metric VP8SSE4x4;
 VP8WMetric VP8TDisto4x4;
 VP8WMetric VP8TDisto16x16;
 VP8QuantizeBlock VP8EncQuantizeBlock;
+VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT;
 VP8BlockCopy VP8Copy4x4;
 
 extern void VP8EncDspInitSSE2(void);
+extern void VP8EncDspInitAVX2(void);
+extern void VP8EncDspInitNEON(void);
+extern void VP8EncDspInitMIPS32(void);
 
 void VP8EncDspInit(void) {
+  VP8DspInit();  // common inverse transforms
   InitTables();
 
   // default C implementations
   VP8CollectHistogram = CollectHistogram;
   VP8ITransform = ITransform;
   VP8FTransform = FTransform;
-  VP8ITransformWHT = ITransformWHT;
   VP8FTransformWHT = FTransformWHT;
   VP8EncPredLuma4 = Intra4Preds;
   VP8EncPredLuma16 = Intra16Preds;
@@ -726,18 +711,31 @@ void VP8EncDspInit(void) {
   VP8TDisto4x4 = Disto4x4;
   VP8TDisto16x16 = Disto16x16;
   VP8EncQuantizeBlock = QuantizeBlock;
+  VP8EncQuantizeBlockWHT = QuantizeBlockWHT;
   VP8Copy4x4 = Copy4x4;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
-  if (VP8GetCPUInfo) {
+  if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8EncDspInitSSE2();
     }
+#endif
+#if defined(WEBP_USE_AVX2)
+    if (VP8GetCPUInfo(kAVX2)) {
+      VP8EncDspInitAVX2();
+    }
+#endif
+#if defined(WEBP_USE_NEON)
+    if (VP8GetCPUInfo(kNEON)) {
+      VP8EncDspInitNEON();
+    }
+#endif
+#if defined(WEBP_USE_MIPS32)
+    if (VP8GetCPUInfo(kMIPS32)) {
+      VP8EncDspInitMIPS32();
+    }
 #endif
   }
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dsp/enc_avx2.c

@@ -0,0 +1,24 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// AVX2 version of speed-critical encoding functions.
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_AVX2)
+
+#endif  // WEBP_USE_AVX2
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void VP8EncDspInitAVX2(void);
+
+void VP8EncDspInitAVX2(void) {
+}

src/dsp/enc_mips32.c

@@ -0,0 +1,776 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// MIPS version of speed-critical encoding functions.
+//
+// Author(s): Djordje Pesut    (djordje.pesut@imgtec.com)
+//            Jovan Zelincevic (jovan.zelincevic@imgtec.com)
+//            Slobodan Prijic  (slobodan.prijic@imgtec.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_MIPS32)
+
+#include "../enc/vp8enci.h"
+#include "../enc/cost.h"
+
+#if defined(__GNUC__) && defined(__ANDROID__) && LOCAL_GCC_VERSION == 0x409
+#define WORK_AROUND_GCC
+#endif
+
+static const int kC1 = 20091 + (1 << 16);
+static const int kC2 = 35468;
+
+// macro for one vertical pass in ITransformOne
+// MUL macro inlined
+// temp0..temp15 holds tmp[0]..tmp[15]
+// A..D - offsets in bytes to load from in buffer
+// TEMP0..TEMP3 - registers for corresponding tmp elements
+// TEMP4..TEMP5 - temporary registers
+#define VERTICAL_PASS(A, B, C, D, TEMP4, TEMP0, TEMP1, TEMP2, TEMP3)        \
+  "lh      %[temp16],      "#A"(%[temp20])                 \n\t"            \
+  "lh      %[temp18],      "#B"(%[temp20])                 \n\t"            \
+  "lh      %[temp17],      "#C"(%[temp20])                 \n\t"            \
+  "lh      %[temp19],      "#D"(%[temp20])                 \n\t"            \
+  "addu    %["#TEMP4"],    %[temp16],      %[temp18]       \n\t"            \
+  "subu    %[temp16],      %[temp16],      %[temp18]       \n\t"            \
+  "mul     %["#TEMP0"],    %[temp17],      %[kC2]          \n\t"            \
+  "mul     %[temp18],      %[temp19],      %[kC1]          \n\t"            \
+  "mul     %[temp17],      %[temp17],      %[kC1]          \n\t"            \
+  "mul     %[temp19],      %[temp19],      %[kC2]          \n\t"            \
+  "sra     %["#TEMP0"],    %["#TEMP0"],    16              \n\n"            \
+  "sra     %[temp18],      %[temp18],      16              \n\n"            \
+  "sra     %[temp17],      %[temp17],      16              \n\n"            \
+  "sra     %[temp19],      %[temp19],      16              \n\n"            \
+  "subu    %["#TEMP2"],    %["#TEMP0"],    %[temp18]       \n\t"            \
+  "addu    %["#TEMP3"],    %[temp17],      %[temp19]       \n\t"            \
+  "addu    %["#TEMP0"],    %["#TEMP4"],    %["#TEMP3"]     \n\t"            \
+  "addu    %["#TEMP1"],    %[temp16],      %["#TEMP2"]     \n\t"            \
+  "subu    %["#TEMP2"],    %[temp16],      %["#TEMP2"]     \n\t"            \
+  "subu    %["#TEMP3"],    %["#TEMP4"],    %["#TEMP3"]     \n\t"
+
+// macro for one horizontal pass in ITransformOne
+// MUL and STORE macros inlined
+// a = clip_8b(a) is replaced with: a = max(a, 0); a = min(a, 255)
+// temp0..temp15 holds tmp[0]..tmp[15]
+// A..D - offsets in bytes to load from ref and store to dst buffer
+// TEMP0, TEMP4, TEMP8 and TEMP12 - registers for corresponding tmp elements
+#define HORIZONTAL_PASS(A, B, C, D, TEMP0, TEMP4, TEMP8, TEMP12)            \
+  "addiu   %["#TEMP0"],    %["#TEMP0"],    4               \n\t"            \
+  "addu    %[temp16],      %["#TEMP0"],    %["#TEMP8"]     \n\t"            \
+  "subu    %[temp17],      %["#TEMP0"],    %["#TEMP8"]     \n\t"            \
+  "mul     %["#TEMP0"],    %["#TEMP4"],    %[kC2]          \n\t"            \
+  "mul     %["#TEMP8"],    %["#TEMP12"],   %[kC1]          \n\t"            \
+  "mul     %["#TEMP4"],    %["#TEMP4"],    %[kC1]          \n\t"            \
+  "mul     %["#TEMP12"],   %["#TEMP12"],   %[kC2]          \n\t"            \
+  "sra     %["#TEMP0"],    %["#TEMP0"],    16              \n\t"            \
+  "sra     %["#TEMP8"],    %["#TEMP8"],    16              \n\t"            \
+  "sra     %["#TEMP4"],    %["#TEMP4"],    16              \n\t"            \
+  "sra     %["#TEMP12"],   %["#TEMP12"],   16              \n\t"            \
+  "subu    %[temp18],      %["#TEMP0"],    %["#TEMP8"]     \n\t"            \
+  "addu    %[temp19],      %["#TEMP4"],    %["#TEMP12"]    \n\t"            \
+  "addu    %["#TEMP0"],    %[temp16],      %[temp19]       \n\t"            \
+  "addu    %["#TEMP4"],    %[temp17],      %[temp18]       \n\t"            \
+  "subu    %["#TEMP8"],    %[temp17],      %[temp18]       \n\t"            \
+  "subu    %["#TEMP12"],   %[temp16],      %[temp19]       \n\t"            \
+  "lw      %[temp20],      0(%[args])                      \n\t"            \
+  "sra     %["#TEMP0"],    %["#TEMP0"],    3               \n\t"            \
+  "sra     %["#TEMP4"],    %["#TEMP4"],    3               \n\t"            \
+  "sra     %["#TEMP8"],    %["#TEMP8"],    3               \n\t"            \
+  "sra     %["#TEMP12"],   %["#TEMP12"],   3               \n\t"            \
+  "lbu     %[temp16],      "#A"(%[temp20])                 \n\t"            \
+  "lbu     %[temp17],      "#B"(%[temp20])                 \n\t"            \
+  "lbu     %[temp18],      "#C"(%[temp20])                 \n\t"            \
+  "lbu     %[temp19],      "#D"(%[temp20])                 \n\t"            \
+  "addu    %["#TEMP0"],    %[temp16],      %["#TEMP0"]     \n\t"            \
+  "addu    %["#TEMP4"],    %[temp17],      %["#TEMP4"]     \n\t"            \
+  "addu    %["#TEMP8"],    %[temp18],      %["#TEMP8"]     \n\t"            \
+  "addu    %["#TEMP12"],   %[temp19],      %["#TEMP12"]    \n\t"            \
+  "slt     %[temp16],      %["#TEMP0"],    $zero           \n\t"            \
+  "slt     %[temp17],      %["#TEMP4"],    $zero           \n\t"            \
+  "slt     %[temp18],      %["#TEMP8"],    $zero           \n\t"            \
+  "slt     %[temp19],      %["#TEMP12"],   $zero           \n\t"            \
+  "movn    %["#TEMP0"],    $zero,          %[temp16]       \n\t"            \
+  "movn    %["#TEMP4"],    $zero,          %[temp17]       \n\t"            \
+  "movn    %["#TEMP8"],    $zero,          %[temp18]       \n\t"            \
+  "movn    %["#TEMP12"],   $zero,          %[temp19]       \n\t"            \
+  "addiu   %[temp20],      $zero,          255             \n\t"            \
+  "slt     %[temp16],      %["#TEMP0"],    %[temp20]       \n\t"            \
+  "slt     %[temp17],      %["#TEMP4"],    %[temp20]       \n\t"            \
+  "slt     %[temp18],      %["#TEMP8"],    %[temp20]       \n\t"            \
+  "slt     %[temp19],      %["#TEMP12"],   %[temp20]       \n\t"            \
+  "movz    %["#TEMP0"],    %[temp20],      %[temp16]       \n\t"            \
+  "movz    %["#TEMP4"],    %[temp20],      %[temp17]       \n\t"            \
+  "lw      %[temp16],      8(%[args])                      \n\t"            \
+  "movz    %["#TEMP8"],    %[temp20],      %[temp18]       \n\t"            \
+  "movz    %["#TEMP12"],   %[temp20],      %[temp19]       \n\t"            \
+  "sb      %["#TEMP0"],    "#A"(%[temp16])                 \n\t"            \
+  "sb      %["#TEMP4"],    "#B"(%[temp16])                 \n\t"            \
+  "sb      %["#TEMP8"],    "#C"(%[temp16])                 \n\t"            \
+  "sb      %["#TEMP12"],   "#D"(%[temp16])                 \n\t"
+
+// Does one or two inverse transforms.
+static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
+                                      uint8_t* dst) {
+  int temp0, temp1, temp2, temp3, temp4, temp5, temp6;
+  int temp7, temp8, temp9, temp10, temp11, temp12, temp13;
+  int temp14, temp15, temp16, temp17, temp18, temp19, temp20;
+  const int* args[3] = {(const int*)ref, (const int*)in, (const int*)dst};
+
+  __asm__ volatile(
+    "lw      %[temp20],      4(%[args])                      \n\t"
+    VERTICAL_PASS(0, 16,  8, 24, temp4,  temp0,  temp1,  temp2,  temp3)
+    VERTICAL_PASS(2, 18, 10, 26, temp8,  temp4,  temp5,  temp6,  temp7)
+    VERTICAL_PASS(4, 20, 12, 28, temp12, temp8,  temp9,  temp10, temp11)
+    VERTICAL_PASS(6, 22, 14, 30, temp20, temp12, temp13, temp14, temp15)
+
+    HORIZONTAL_PASS( 0,  1,  2,  3, temp0, temp4, temp8,  temp12)
+    HORIZONTAL_PASS(16, 17, 18, 19, temp1, temp5, temp9,  temp13)
+    HORIZONTAL_PASS(32, 33, 34, 35, temp2, temp6, temp10, temp14)
+    HORIZONTAL_PASS(48, 49, 50, 51, temp3, temp7, temp11, temp15)
+
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
+      [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11),
+      [temp12]"=&r"(temp12), [temp13]"=&r"(temp13), [temp14]"=&r"(temp14),
+      [temp15]"=&r"(temp15), [temp16]"=&r"(temp16), [temp17]"=&r"(temp17),
+      [temp18]"=&r"(temp18), [temp19]"=&r"(temp19), [temp20]"=&r"(temp20)
+    : [args]"r"(args), [kC1]"r"(kC1), [kC2]"r"(kC2)
+    : "memory", "hi", "lo"
+  );
+}
+
+static void ITransform(const uint8_t* ref, const int16_t* in,
+                       uint8_t* dst, int do_two) {
+  ITransformOne(ref, in, dst);
+  if (do_two) {
+    ITransformOne(ref + 4, in + 16, dst + 4);
+  }
+}
+
+#undef VERTICAL_PASS
+#undef HORIZONTAL_PASS
+
+// macro for one pass through for loop in QuantizeBlock
+// QUANTDIV macro inlined
+// J - offset in bytes (kZigzag[n] * 2)
+// K - offset in bytes (kZigzag[n] * 4)
+// N - offset in bytes (n * 2)
+#define QUANTIZE_ONE(J, K, N)                                               \
+  "lh           %[temp0],       "#J"(%[ppin])                       \n\t"   \
+  "lhu          %[temp1],       "#J"(%[ppsharpen])                  \n\t"   \
+  "lw           %[temp2],       "#K"(%[ppzthresh])                  \n\t"   \
+  "sra          %[sign],        %[temp0],           15              \n\t"   \
+  "xor          %[coeff],       %[temp0],           %[sign]         \n\t"   \
+  "subu         %[coeff],       %[coeff],           %[sign]         \n\t"   \
+  "addu         %[coeff],       %[coeff],           %[temp1]        \n\t"   \
+  "slt          %[temp4],       %[temp2],           %[coeff]        \n\t"   \
+  "addiu        %[temp5],       $zero,              0               \n\t"   \
+  "addiu        %[level],       $zero,              0               \n\t"   \
+  "beqz         %[temp4],       2f                                  \n\t"   \
+  "lhu          %[temp1],       "#J"(%[ppiq])                       \n\t"   \
+  "lw           %[temp2],       "#K"(%[ppbias])                     \n\t"   \
+  "lhu          %[temp3],       "#J"(%[ppq])                        \n\t"   \
+  "mul          %[level],       %[coeff],           %[temp1]        \n\t"   \
+  "addu         %[level],       %[level],           %[temp2]        \n\t"   \
+  "sra          %[level],       %[level],           17              \n\t"   \
+  "slt          %[temp4],       %[max_level],       %[level]        \n\t"   \
+  "movn         %[level],       %[max_level],       %[temp4]        \n\t"   \
+  "xor          %[level],       %[level],           %[sign]         \n\t"   \
+  "subu         %[level],       %[level],           %[sign]         \n\t"   \
+  "mul          %[temp5],       %[level],           %[temp3]        \n\t"   \
+"2:                                                                 \n\t"   \
+  "sh           %[temp5],       "#J"(%[ppin])                       \n\t"   \
+  "sh           %[level],       "#N"(%[pout])                       \n\t"
+
+static int QuantizeBlock(int16_t in[16], int16_t out[16],
+                         const VP8Matrix* const mtx) {
+  int temp0, temp1, temp2, temp3, temp4, temp5;
+  int sign, coeff, level, i;
+  int max_level = MAX_LEVEL;
+
+  int16_t* ppin             = &in[0];
+  int16_t* pout             = &out[0];
+  const uint16_t* ppsharpen = &mtx->sharpen_[0];
+  const uint32_t* ppzthresh = &mtx->zthresh_[0];
+  const uint16_t* ppq       = &mtx->q_[0];
+  const uint16_t* ppiq      = &mtx->iq_[0];
+  const uint32_t* ppbias    = &mtx->bias_[0];
+
+  __asm__ volatile(
+    QUANTIZE_ONE( 0,  0,  0)
+    QUANTIZE_ONE( 2,  4,  2)
+    QUANTIZE_ONE( 8, 16,  4)
+    QUANTIZE_ONE(16, 32,  6)
+    QUANTIZE_ONE(10, 20,  8)
+    QUANTIZE_ONE( 4,  8, 10)
+    QUANTIZE_ONE( 6, 12, 12)
+    QUANTIZE_ONE(12, 24, 14)
+    QUANTIZE_ONE(18, 36, 16)
+    QUANTIZE_ONE(24, 48, 18)
+    QUANTIZE_ONE(26, 52, 20)
+    QUANTIZE_ONE(20, 40, 22)
+    QUANTIZE_ONE(14, 28, 24)
+    QUANTIZE_ONE(22, 44, 26)
+    QUANTIZE_ONE(28, 56, 28)
+    QUANTIZE_ONE(30, 60, 30)
+
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
+      [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
+      [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [sign]"=&r"(sign), [coeff]"=&r"(coeff),
+      [level]"=&r"(level)
+    : [pout]"r"(pout), [ppin]"r"(ppin),
+      [ppiq]"r"(ppiq), [max_level]"r"(max_level),
+      [ppbias]"r"(ppbias), [ppzthresh]"r"(ppzthresh),
+      [ppsharpen]"r"(ppsharpen), [ppq]"r"(ppq)
+    : "memory", "hi", "lo"
+  );
+
+  // moved out from macro to increase possibility for earlier breaking
+  for (i = 15; i >= 0; i--) {
+    if (out[i]) return 1;
+  }
+  return 0;
+}
+
+#undef QUANTIZE_ONE
+
+// macro for one horizontal pass in Disto4x4 (TTransform)
+// two calls of function TTransform are merged into single one
+// A..D - offsets in bytes to load from a and b buffers
+// E..H - offsets in bytes to store first results to tmp buffer
+// E1..H1 - offsets in bytes to store second results to tmp buffer
+#define HORIZONTAL_PASS(A, B, C, D, E, F, G, H, E1, F1, G1, H1)   \
+  "lbu    %[temp0],  "#A"(%[a])              \n\t"                \
+  "lbu    %[temp1],  "#B"(%[a])              \n\t"                \
+  "lbu    %[temp2],  "#C"(%[a])              \n\t"                \
+  "lbu    %[temp3],  "#D"(%[a])              \n\t"                \
+  "lbu    %[temp4],  "#A"(%[b])              \n\t"                \
+  "lbu    %[temp5],  "#B"(%[b])              \n\t"                \
+  "lbu    %[temp6],  "#C"(%[b])              \n\t"                \
+  "lbu    %[temp7],  "#D"(%[b])              \n\t"                \
+  "addu   %[temp8],  %[temp0],    %[temp2]   \n\t"                \
+  "subu   %[temp0],  %[temp0],    %[temp2]   \n\t"                \
+  "addu   %[temp2],  %[temp1],    %[temp3]   \n\t"                \
+  "subu   %[temp1],  %[temp1],    %[temp3]   \n\t"                \
+  "addu   %[temp3],  %[temp4],    %[temp6]   \n\t"                \
+  "subu   %[temp4],  %[temp4],    %[temp6]   \n\t"                \
+  "addu   %[temp6],  %[temp5],    %[temp7]   \n\t"                \
+  "subu   %[temp5],  %[temp5],    %[temp7]   \n\t"                \
+  "addu   %[temp7],  %[temp8],    %[temp2]   \n\t"                \
+  "subu   %[temp2],  %[temp8],    %[temp2]   \n\t"                \
+  "addu   %[temp8],  %[temp0],    %[temp1]   \n\t"                \
+  "subu   %[temp0],  %[temp0],    %[temp1]   \n\t"                \
+  "addu   %[temp1],  %[temp3],    %[temp6]   \n\t"                \
+  "subu   %[temp3],  %[temp3],    %[temp6]   \n\t"                \
+  "addu   %[temp6],  %[temp4],    %[temp5]   \n\t"                \
+  "subu   %[temp4],  %[temp4],    %[temp5]   \n\t"                \
+  "sw     %[temp7],  "#E"(%[tmp])            \n\t"                \
+  "sw     %[temp2],  "#H"(%[tmp])            \n\t"                \
+  "sw     %[temp8],  "#F"(%[tmp])            \n\t"                \
+  "sw     %[temp0],  "#G"(%[tmp])            \n\t"                \
+  "sw     %[temp1],  "#E1"(%[tmp])           \n\t"                \
+  "sw     %[temp3],  "#H1"(%[tmp])           \n\t"                \
+  "sw     %[temp6],  "#F1"(%[tmp])           \n\t"                \
+  "sw     %[temp4],  "#G1"(%[tmp])           \n\t"
+
+// macro for one vertical pass in Disto4x4 (TTransform)
+// two calls of function TTransform are merged into single one
+// since only one accu is available in mips32r1 instruction set
+//   first is done second call of function TTransform and after
+//   that first one.
+//   const int sum1 = TTransform(a, w);
+//   const int sum2 = TTransform(b, w);
+//   return abs(sum2 - sum1) >> 5;
+//   (sum2 - sum1) is calculated with madds (sub2) and msubs (sub1)
+// A..D - offsets in bytes to load first results from tmp buffer
+// A1..D1 - offsets in bytes to load second results from tmp buffer
+// E..H - offsets in bytes to load from w buffer
+#define VERTICAL_PASS(A, B, C, D, A1, B1, C1, D1, E, F, G, H)     \
+  "lw     %[temp0],  "#A1"(%[tmp])           \n\t"                \
+  "lw     %[temp1],  "#C1"(%[tmp])           \n\t"                \
+  "lw     %[temp2],  "#B1"(%[tmp])           \n\t"                \
+  "lw     %[temp3],  "#D1"(%[tmp])           \n\t"                \
+  "addu   %[temp8],  %[temp0],    %[temp1]   \n\t"                \
+  "subu   %[temp0],  %[temp0],    %[temp1]   \n\t"                \
+  "addu   %[temp1],  %[temp2],    %[temp3]   \n\t"                \
+  "subu   %[temp2],  %[temp2],    %[temp3]   \n\t"                \
+  "addu   %[temp3],  %[temp8],    %[temp1]   \n\t"                \
+  "subu   %[temp8],  %[temp8],    %[temp1]   \n\t"                \
+  "addu   %[temp1],  %[temp0],    %[temp2]   \n\t"                \
+  "subu   %[temp0],  %[temp0],    %[temp2]   \n\t"                \
+  "sra    %[temp4],  %[temp3],    31         \n\t"                \
+  "sra    %[temp5],  %[temp1],    31         \n\t"                \
+  "sra    %[temp6],  %[temp0],    31         \n\t"                \
+  "sra    %[temp7],  %[temp8],    31         \n\t"                \
+  "xor    %[temp3],  %[temp3],    %[temp4]   \n\t"                \
+  "xor    %[temp1],  %[temp1],    %[temp5]   \n\t"                \
+  "xor    %[temp0],  %[temp0],    %[temp6]   \n\t"                \
+  "xor    %[temp8],  %[temp8],    %[temp7]   \n\t"                \
+  "subu   %[temp3],  %[temp3],    %[temp4]   \n\t"                \
+  "subu   %[temp1],  %[temp1],    %[temp5]   \n\t"                \
+  "subu   %[temp0],  %[temp0],    %[temp6]   \n\t"                \
+  "subu   %[temp8],  %[temp8],    %[temp7]   \n\t"                \
+  "lhu    %[temp4],  "#E"(%[w])              \n\t"                \
+  "lhu    %[temp5],  "#F"(%[w])              \n\t"                \
+  "lhu    %[temp6],  "#G"(%[w])              \n\t"                \
+  "lhu    %[temp7],  "#H"(%[w])              \n\t"                \
+  "madd   %[temp4],  %[temp3]                \n\t"                \
+  "madd   %[temp5],  %[temp1]                \n\t"                \
+  "madd   %[temp6],  %[temp0]                \n\t"                \
+  "madd   %[temp7],  %[temp8]                \n\t"                \
+  "lw     %[temp0],  "#A"(%[tmp])            \n\t"                \
+  "lw     %[temp1],  "#C"(%[tmp])            \n\t"                \
+  "lw     %[temp2],  "#B"(%[tmp])            \n\t"                \
+  "lw     %[temp3],  "#D"(%[tmp])            \n\t"                \
+  "addu   %[temp8],  %[temp0],    %[temp1]   \n\t"                \
+  "subu   %[temp0],  %[temp0],    %[temp1]   \n\t"                \
+  "addu   %[temp1],  %[temp2],    %[temp3]   \n\t"                \
+  "subu   %[temp2],  %[temp2],    %[temp3]   \n\t"                \
+  "addu   %[temp3],  %[temp8],    %[temp1]   \n\t"                \
+  "subu   %[temp1],  %[temp8],    %[temp1]   \n\t"                \
+  "addu   %[temp8],  %[temp0],    %[temp2]   \n\t"                \
+  "subu   %[temp0],  %[temp0],    %[temp2]   \n\t"                \
+  "sra    %[temp2],  %[temp3],    31         \n\t"                \
+  "xor    %[temp3],  %[temp3],    %[temp2]   \n\t"                \
+  "subu   %[temp3],  %[temp3],    %[temp2]   \n\t"                \
+  "msub   %[temp4],  %[temp3]                \n\t"                \
+  "sra    %[temp2],  %[temp8],    31         \n\t"                \
+  "sra    %[temp3],  %[temp0],    31         \n\t"                \
+  "sra    %[temp4],  %[temp1],    31         \n\t"                \
+  "xor    %[temp8],  %[temp8],    %[temp2]   \n\t"                \
+  "xor    %[temp0],  %[temp0],    %[temp3]   \n\t"                \
+  "xor    %[temp1],  %[temp1],    %[temp4]   \n\t"                \
+  "subu   %[temp8],  %[temp8],    %[temp2]   \n\t"                \
+  "subu   %[temp0],  %[temp0],    %[temp3]   \n\t"                \
+  "subu   %[temp1],  %[temp1],    %[temp4]   \n\t"                \
+  "msub   %[temp5],  %[temp8]                \n\t"                \
+  "msub   %[temp6],  %[temp0]                \n\t"                \
+  "msub   %[temp7],  %[temp1]                \n\t"
+
+static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
+                    const uint16_t* const w) {
+  int tmp[32];
+  int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
+
+  __asm__ volatile(
+    HORIZONTAL_PASS( 0,  1,  2,  3,    0,  4,  8, 12,    64,  68,  72,  76)
+    HORIZONTAL_PASS(16, 17, 18, 19,   16, 20, 24, 28,    80,  84,  88,  92)
+    HORIZONTAL_PASS(32, 33, 34, 35,   32, 36, 40, 44,    96, 100, 104, 108)
+    HORIZONTAL_PASS(48, 49, 50, 51,   48, 52, 56, 60,   112, 116, 120, 124)
+    "mthi   $zero                             \n\t"
+    "mtlo   $zero                             \n\t"
+    VERTICAL_PASS( 0, 16, 32, 48,     64, 80,  96, 112,   0,  8, 16, 24)
+    VERTICAL_PASS( 4, 20, 36, 52,     68, 84, 100, 116,   2, 10, 18, 26)
+    VERTICAL_PASS( 8, 24, 40, 56,     72, 88, 104, 120,   4, 12, 20, 28)
+    VERTICAL_PASS(12, 28, 44, 60,     76, 92, 108, 124,   6, 14, 22, 30)
+    "mflo   %[temp0]                          \n\t"
+    "sra    %[temp1],  %[temp0],  31          \n\t"
+    "xor    %[temp0],  %[temp0],  %[temp1]    \n\t"
+    "subu   %[temp0],  %[temp0],  %[temp1]    \n\t"
+    "sra    %[temp0],  %[temp0],  5           \n\t"
+
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8)
+    : [a]"r"(a), [b]"r"(b), [w]"r"(w), [tmp]"r"(tmp)
+    : "memory", "hi", "lo"
+  );
+
+  return temp0;
+}
+
+#undef VERTICAL_PASS
+#undef HORIZONTAL_PASS
+
+static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
+                      const uint16_t* const w) {
+  int D = 0;
+  int x, y;
+  for (y = 0; y < 16 * BPS; y += 4 * BPS) {
+    for (x = 0; x < 16; x += 4) {
+      D += Disto4x4(a + x + y, b + x + y, w);
+    }
+  }
+  return D;
+}
+
+// macro for one horizontal pass in FTransform
+// temp0..temp15 holds tmp[0]..tmp[15]
+// A..D - offsets in bytes to load from src and ref buffers
+// TEMP0..TEMP3 - registers for corresponding tmp elements
+#define HORIZONTAL_PASS(A, B, C, D, TEMP0, TEMP1, TEMP2, TEMP3) \
+  "lw     %["#TEMP1"],  0(%[args])                     \n\t"    \
+  "lw     %["#TEMP2"],  4(%[args])                     \n\t"    \
+  "lbu    %[temp16],    "#A"(%["#TEMP1"])              \n\t"    \
+  "lbu    %[temp17],    "#A"(%["#TEMP2"])              \n\t"    \
+  "lbu    %[temp18],    "#B"(%["#TEMP1"])              \n\t"    \
+  "lbu    %[temp19],    "#B"(%["#TEMP2"])              \n\t"    \
+  "subu   %[temp20],    %[temp16],    %[temp17]        \n\t"    \
+  "lbu    %[temp16],    "#C"(%["#TEMP1"])              \n\t"    \
+  "lbu    %[temp17],    "#C"(%["#TEMP2"])              \n\t"    \
+  "subu   %["#TEMP0"],  %[temp18],    %[temp19]        \n\t"    \
+  "lbu    %[temp18],    "#D"(%["#TEMP1"])              \n\t"    \
+  "lbu    %[temp19],    "#D"(%["#TEMP2"])              \n\t"    \
+  "subu   %["#TEMP1"],  %[temp16],    %[temp17]        \n\t"    \
+  "subu   %["#TEMP2"],  %[temp18],    %[temp19]        \n\t"    \
+  "addu   %["#TEMP3"],  %[temp20],    %["#TEMP2"]      \n\t"    \
+  "subu   %["#TEMP2"],  %[temp20],    %["#TEMP2"]      \n\t"    \
+  "addu   %[temp20],    %["#TEMP0"],  %["#TEMP1"]      \n\t"    \
+  "subu   %["#TEMP0"],  %["#TEMP0"],  %["#TEMP1"]      \n\t"    \
+  "mul    %[temp16],    %["#TEMP2"],  %[c5352]         \n\t"    \
+  "mul    %[temp17],    %["#TEMP2"],  %[c2217]         \n\t"    \
+  "mul    %[temp18],    %["#TEMP0"],  %[c5352]         \n\t"    \
+  "mul    %[temp19],    %["#TEMP0"],  %[c2217]         \n\t"    \
+  "addu   %["#TEMP1"],  %["#TEMP3"],  %[temp20]        \n\t"    \
+  "subu   %[temp20],    %["#TEMP3"],  %[temp20]        \n\t"    \
+  "sll    %["#TEMP0"],  %["#TEMP1"],  3                \n\t"    \
+  "sll    %["#TEMP2"],  %[temp20],    3                \n\t"    \
+  "addiu  %[temp16],    %[temp16],    1812             \n\t"    \
+  "addiu  %[temp17],    %[temp17],    937              \n\t"    \
+  "addu   %[temp16],    %[temp16],    %[temp19]        \n\t"    \
+  "subu   %[temp17],    %[temp17],    %[temp18]        \n\t"    \
+  "sra    %["#TEMP1"],  %[temp16],    9                \n\t"    \
+  "sra    %["#TEMP3"],  %[temp17],    9                \n\t"
+
+// macro for one vertical pass in FTransform
+// temp0..temp15 holds tmp[0]..tmp[15]
+// A..D - offsets in bytes to store to out buffer
+// TEMP0, TEMP4, TEMP8 and TEMP12 - registers for corresponding tmp elements
+#define VERTICAL_PASS(A, B, C, D, TEMP0, TEMP4, TEMP8, TEMP12)  \
+  "addu   %[temp16],    %["#TEMP0"],  %["#TEMP12"]     \n\t"    \
+  "subu   %[temp19],    %["#TEMP0"],  %["#TEMP12"]     \n\t"    \
+  "addu   %[temp17],    %["#TEMP4"],  %["#TEMP8"]      \n\t"    \
+  "subu   %[temp18],    %["#TEMP4"],  %["#TEMP8"]      \n\t"    \
+  "mul    %["#TEMP8"],  %[temp19],    %[c2217]         \n\t"    \
+  "mul    %["#TEMP12"], %[temp18],    %[c2217]         \n\t"    \
+  "mul    %["#TEMP4"],  %[temp19],    %[c5352]         \n\t"    \
+  "mul    %[temp18],    %[temp18],    %[c5352]         \n\t"    \
+  "addiu  %[temp16],    %[temp16],    7                \n\t"    \
+  "addu   %["#TEMP0"],  %[temp16],    %[temp17]        \n\t"    \
+  "sra    %["#TEMP0"],  %["#TEMP0"],  4                \n\t"    \
+  "addu   %["#TEMP12"], %["#TEMP12"], %["#TEMP4"]      \n\t"    \
+  "subu   %["#TEMP4"],  %[temp16],    %[temp17]        \n\t"    \
+  "sra    %["#TEMP4"],  %["#TEMP4"],  4                \n\t"    \
+  "addiu  %["#TEMP8"],  %["#TEMP8"],  30000            \n\t"    \
+  "addiu  %["#TEMP12"], %["#TEMP12"], 12000            \n\t"    \
+  "addiu  %["#TEMP8"],  %["#TEMP8"],  21000            \n\t"    \
+  "subu   %["#TEMP8"],  %["#TEMP8"],  %[temp18]        \n\t"    \
+  "sra    %["#TEMP12"], %["#TEMP12"], 16               \n\t"    \
+  "sra    %["#TEMP8"],  %["#TEMP8"],  16               \n\t"    \
+  "addiu  %[temp16],    %["#TEMP12"], 1                \n\t"    \
+  "movn   %["#TEMP12"], %[temp16],    %[temp19]        \n\t"    \
+  "sh     %["#TEMP0"],  "#A"(%[temp20])                \n\t"    \
+  "sh     %["#TEMP4"],  "#C"(%[temp20])                \n\t"    \
+  "sh     %["#TEMP8"],  "#D"(%[temp20])                \n\t"    \
+  "sh     %["#TEMP12"], "#B"(%[temp20])                \n\t"
+
+static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+  int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
+  int temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
+  int temp17, temp18, temp19, temp20;
+  const int c2217 = 2217;
+  const int c5352 = 5352;
+  const int* const args[3] =
+      { (const int*)src, (const int*)ref, (const int*)out };
+
+  __asm__ volatile(
+    HORIZONTAL_PASS( 0,  1,  2,  3, temp0,  temp1,  temp2,  temp3)
+    HORIZONTAL_PASS(16, 17, 18, 19, temp4,  temp5,  temp6,  temp7)
+    HORIZONTAL_PASS(32, 33, 34, 35, temp8,  temp9,  temp10, temp11)
+    HORIZONTAL_PASS(48, 49, 50, 51, temp12, temp13, temp14, temp15)
+    "lw   %[temp20],    8(%[args])                     \n\t"
+    VERTICAL_PASS(0,  8, 16, 24, temp0, temp4, temp8,  temp12)
+    VERTICAL_PASS(2, 10, 18, 26, temp1, temp5, temp9,  temp13)
+    VERTICAL_PASS(4, 12, 20, 28, temp2, temp6, temp10, temp14)
+    VERTICAL_PASS(6, 14, 22, 30, temp3, temp7, temp11, temp15)
+
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
+      [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11),
+      [temp12]"=&r"(temp12), [temp13]"=&r"(temp13), [temp14]"=&r"(temp14),
+      [temp15]"=&r"(temp15), [temp16]"=&r"(temp16), [temp17]"=&r"(temp17),
+      [temp18]"=&r"(temp18), [temp19]"=&r"(temp19), [temp20]"=&r"(temp20)
+    : [args]"r"(args), [c2217]"r"(c2217), [c5352]"r"(c5352)
+    : "memory", "hi", "lo"
+  );
+}
+
+#undef VERTICAL_PASS
+#undef HORIZONTAL_PASS
+
+// Forward declaration.
+extern int VP8GetResidualCostMIPS32(int ctx0, const VP8Residual* const res);
+
+int VP8GetResidualCostMIPS32(int ctx0, const VP8Residual* const res) {
+  int n = res->first;
+  // should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1
+  int p0 = res->prob[n][ctx0][0];
+  const uint16_t* t = res->cost[n][ctx0];
+  int cost;
+  const int const_2 = 2;
+  const int const_255 = 255;
+  const int const_max_level = MAX_VARIABLE_LEVEL;
+  int res_cost;
+  int res_prob;
+  int res_coeffs;
+  int res_last;
+  int v_reg;
+  int b_reg;
+  int ctx_reg;
+  int cost_add, temp_1, temp_2, temp_3;
+
+  if (res->last < 0) {
+    return VP8BitCost(0, p0);
+  }
+
+  cost = (ctx0 == 0) ? VP8BitCost(1, p0) : 0;
+
+  res_cost = (int)res->cost;
+  res_prob = (int)res->prob;
+  res_coeffs = (int)res->coeffs;
+  res_last = (int)res->last;
+
+  __asm__ volatile(
+    ".set   push                                                           \n\t"
+    ".set   noreorder                                                      \n\t"
+
+    "sll    %[temp_1],     %[n],              1                            \n\t"
+    "addu   %[res_coeffs], %[res_coeffs],     %[temp_1]                    \n\t"
+    "slt    %[temp_2],     %[n],              %[res_last]                  \n\t"
+    "bnez   %[temp_2],     1f                                              \n\t"
+    " li    %[cost_add],   0                                               \n\t"
+    "b      2f                                                             \n\t"
+    " nop                                                                  \n\t"
+  "1:                                                                      \n\t"
+    "lh     %[v_reg],      0(%[res_coeffs])                                \n\t"
+    "addu   %[b_reg],      %[n],              %[VP8EncBands]               \n\t"
+    "move   %[temp_1],     %[const_max_level]                              \n\t"
+    "addu   %[cost],       %[cost],           %[cost_add]                  \n\t"
+    "negu   %[temp_2],     %[v_reg]                                        \n\t"
+    "slti   %[temp_3],     %[v_reg],          0                            \n\t"
+    "movn   %[v_reg],      %[temp_2],         %[temp_3]                    \n\t"
+    "lbu    %[b_reg],      1(%[b_reg])                                     \n\t"
+    "li     %[cost_add],   0                                               \n\t"
+
+    "sltiu  %[temp_3],     %[v_reg],          2                            \n\t"
+    "move   %[ctx_reg],    %[v_reg]                                        \n\t"
+    "movz   %[ctx_reg],    %[const_2],        %[temp_3]                    \n\t"
+    //  cost += VP8LevelCost(t, v);
+    "slt    %[temp_3],     %[v_reg],          %[const_max_level]           \n\t"
+    "movn   %[temp_1],     %[v_reg],          %[temp_3]                    \n\t"
+    "sll    %[temp_2],     %[v_reg],          1                            \n\t"
+    "addu   %[temp_2],     %[temp_2],         %[VP8LevelFixedCosts]        \n\t"
+    "lhu    %[temp_2],     0(%[temp_2])                                    \n\t"
+    "sll    %[temp_1],     %[temp_1],         1                            \n\t"
+    "addu   %[temp_1],     %[temp_1],         %[t]                         \n\t"
+    "lhu    %[temp_3],     0(%[temp_1])                                    \n\t"
+    "addu   %[cost],       %[cost],           %[temp_2]                    \n\t"
+
+    //  t = res->cost[b][ctx];
+    "sll    %[temp_1],     %[ctx_reg],        7                            \n\t"
+    "sll    %[temp_2],     %[ctx_reg],        3                            \n\t"
+    "addu   %[cost],       %[cost],           %[temp_3]                    \n\t"
+    "addu   %[temp_1],     %[temp_1],         %[temp_2]                    \n\t"
+    "sll    %[temp_2],     %[b_reg],          3                            \n\t"
+    "sll    %[temp_3],     %[b_reg],          5                            \n\t"
+    "sub    %[temp_2],     %[temp_3],         %[temp_2]                    \n\t"
+    "sll    %[temp_3],     %[temp_2],         4                            \n\t"
+    "addu   %[temp_1],     %[temp_1],         %[temp_3]                    \n\t"
+    "addu   %[temp_2],     %[temp_2],         %[res_cost]                  \n\t"
+    "addiu  %[n],          %[n],              1                            \n\t"
+    "addu   %[t],          %[temp_1],         %[temp_2]                    \n\t"
+    "slt    %[temp_1],     %[n],              %[res_last]                  \n\t"
+    "bnez   %[temp_1],     1b                                              \n\t"
+    " addiu %[res_coeffs], %[res_coeffs],     2                            \n\t"
+   "2:                                                                     \n\t"
+
+    ".set   pop                                                            \n\t"
+    : [cost]"+r"(cost), [t]"+r"(t), [n]"+r"(n), [v_reg]"=&r"(v_reg),
+      [ctx_reg]"=&r"(ctx_reg), [b_reg]"=&r"(b_reg), [cost_add]"=&r"(cost_add),
+      [temp_1]"=&r"(temp_1), [temp_2]"=&r"(temp_2), [temp_3]"=&r"(temp_3)
+    : [const_2]"r"(const_2), [const_255]"r"(const_255), [res_last]"r"(res_last),
+      [VP8EntropyCost]"r"(VP8EntropyCost), [VP8EncBands]"r"(VP8EncBands),
+      [const_max_level]"r"(const_max_level), [res_prob]"r"(res_prob),
+      [VP8LevelFixedCosts]"r"(VP8LevelFixedCosts), [res_coeffs]"r"(res_coeffs),
+      [res_cost]"r"(res_cost)
+    : "memory"
+  );
+
+  // Last coefficient is always non-zero
+  {
+    const int v = abs(res->coeffs[n]);
+    assert(v != 0);
+    cost += VP8LevelCost(t, v);
+    if (n < 15) {
+      const int b = VP8EncBands[n + 1];
+      const int ctx = (v == 1) ? 1 : 2;
+      const int last_p0 = res->prob[b][ctx][0];
+      cost += VP8BitCost(0, last_p0);
+    }
+  }
+  return cost;
+}
+
+#define GET_SSE_INNER(A, B, C, D)                               \
+  "lbu     %[temp0],    "#A"(%[a])                   \n\t"      \
+  "lbu     %[temp1],    "#A"(%[b])                   \n\t"      \
+  "lbu     %[temp2],    "#B"(%[a])                   \n\t"      \
+  "lbu     %[temp3],    "#B"(%[b])                   \n\t"      \
+  "lbu     %[temp4],    "#C"(%[a])                   \n\t"      \
+  "lbu     %[temp5],    "#C"(%[b])                   \n\t"      \
+  "lbu     %[temp6],    "#D"(%[a])                   \n\t"      \
+  "lbu     %[temp7],    "#D"(%[b])                   \n\t"      \
+  "subu    %[temp0],    %[temp0],     %[temp1]       \n\t"      \
+  "subu    %[temp2],    %[temp2],     %[temp3]       \n\t"      \
+  "subu    %[temp4],    %[temp4],     %[temp5]       \n\t"      \
+  "subu    %[temp6],    %[temp6],     %[temp7]       \n\t"      \
+  "madd    %[temp0],    %[temp0]                     \n\t"      \
+  "madd    %[temp2],    %[temp2]                     \n\t"      \
+  "madd    %[temp4],    %[temp4]                     \n\t"      \
+  "madd    %[temp6],    %[temp6]                     \n\t"
+
+#define GET_SSE(A, B, C, D)               \
+  GET_SSE_INNER(A, A + 1, A + 2, A + 3)   \
+  GET_SSE_INNER(B, B + 1, B + 2, B + 3)   \
+  GET_SSE_INNER(C, C + 1, C + 2, C + 3)   \
+  GET_SSE_INNER(D, D + 1, D + 2, D + 3)
+
+#if !defined(WORK_AROUND_GCC)
+static int SSE16x16(const uint8_t* a, const uint8_t* b) {
+  int count;
+  int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
+
+  __asm__ volatile(
+     "mult   $zero,    $zero                            \n\t"
+
+     GET_SSE(  0,   4,   8,  12)
+     GET_SSE( 16,  20,  24,  28)
+     GET_SSE( 32,  36,  40,  44)
+     GET_SSE( 48,  52,  56,  60)
+     GET_SSE( 64,  68,  72,  76)
+     GET_SSE( 80,  84,  88,  92)
+     GET_SSE( 96, 100, 104, 108)
+     GET_SSE(112, 116, 120, 124)
+     GET_SSE(128, 132, 136, 140)
+     GET_SSE(144, 148, 152, 156)
+     GET_SSE(160, 164, 168, 172)
+     GET_SSE(176, 180, 184, 188)
+     GET_SSE(192, 196, 200, 204)
+     GET_SSE(208, 212, 216, 220)
+     GET_SSE(224, 228, 232, 236)
+     GET_SSE(240, 244, 248, 252)
+
+    "mflo    %[count]                                   \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [count]"=&r"(count)
+    : [a]"r"(a), [b]"r"(b)
+    : "memory", "hi" , "lo"
+  );
+  return count;
+}
+
+static int SSE16x8(const uint8_t* a, const uint8_t* b) {
+  int count;
+  int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
+
+  __asm__ volatile(
+     "mult   $zero,    $zero                            \n\t"
+
+     GET_SSE(  0,   4,   8,  12)
+     GET_SSE( 16,  20,  24,  28)
+     GET_SSE( 32,  36,  40,  44)
+     GET_SSE( 48,  52,  56,  60)
+     GET_SSE( 64,  68,  72,  76)
+     GET_SSE( 80,  84,  88,  92)
+     GET_SSE( 96, 100, 104, 108)
+     GET_SSE(112, 116, 120, 124)
+
+    "mflo    %[count]                                   \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [count]"=&r"(count)
+    : [a]"r"(a), [b]"r"(b)
+    : "memory", "hi" , "lo"
+  );
+  return count;
+}
+
+static int SSE8x8(const uint8_t* a, const uint8_t* b) {
+  int count;
+  int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
+
+  __asm__ volatile(
+     "mult   $zero,    $zero                            \n\t"
+
+     GET_SSE( 0,   4,  16,  20)
+     GET_SSE(32,  36,  48,  52)
+     GET_SSE(64,  68,  80,  84)
+     GET_SSE(96, 100, 112, 116)
+
+    "mflo    %[count]                                   \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [count]"=&r"(count)
+    : [a]"r"(a), [b]"r"(b)
+    : "memory", "hi" , "lo"
+  );
+  return count;
+}
+
+static int SSE4x4(const uint8_t* a, const uint8_t* b) {
+  int count;
+  int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
+
+  __asm__ volatile(
+     "mult   $zero,    $zero                            \n\t"
+
+     GET_SSE(0, 16, 32, 48)
+
+    "mflo    %[count]                                   \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
+      [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [count]"=&r"(count)
+    : [a]"r"(a), [b]"r"(b)
+    : "memory", "hi" , "lo"
+  );
+  return count;
+}
+
+#endif  // WORK_AROUND_GCC
+
+#undef GET_SSE_MIPS32
+#undef GET_SSE_MIPS32_INNER
+
+#endif  // WEBP_USE_MIPS32
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void VP8EncDspInitMIPS32(void);
+
+void VP8EncDspInitMIPS32(void) {
+#if defined(WEBP_USE_MIPS32)
+  VP8ITransform = ITransform;
+  VP8EncQuantizeBlock = QuantizeBlock;
+  VP8TDisto4x4 = Disto4x4;
+  VP8TDisto16x16 = Disto16x16;
+  VP8FTransform = FTransform;
+#if !defined(WORK_AROUND_GCC)
+  VP8SSE16x16 = SSE16x16;
+  VP8SSE8x8 = SSE8x8;
+  VP8SSE16x8 = SSE16x8;
+  VP8SSE4x4 = SSE4x4;
+#endif
+#endif  // WEBP_USE_MIPS32
+}

src/dsp/enc_sse2.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // SSE2 version of speed-critical encoding functions.
@@ -11,27 +13,56 @@
 
 #include "./dsp.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 #if defined(WEBP_USE_SSE2)
 #include <stdlib.h>  // for abs()
 #include <emmintrin.h>
 
+#include "../enc/cost.h"
 #include "../enc/vp8enci.h"
+#include "../utils/utils.h"
+
+//------------------------------------------------------------------------------
+// Quite useful macro for debugging. Left here for convenience.
+
+#if 0
+#include <stdio.h>
+static void PrintReg(const __m128i r, const char* const name, int size) {
+  int n;
+  union {
+    __m128i r;
+    uint8_t i8[16];
+    uint16_t i16[8];
+    uint32_t i32[4];
+    uint64_t i64[2];
+  } tmp;
+  tmp.r = r;
+  printf("%s\t: ", name);
+  if (size == 8) {
+    for (n = 0; n < 16; ++n) printf("%.2x ", tmp.i8[n]);
+  } else if (size == 16) {
+    for (n = 0; n < 8; ++n) printf("%.4x ", tmp.i16[n]);
+  } else if (size == 32) {
+    for (n = 0; n < 4; ++n) printf("%.8x ", tmp.i32[n]);
+  } else {
+    for (n = 0; n < 2; ++n) printf("%.16lx ", tmp.i64[n]);
+  }
+  printf("\n");
+}
+#endif
 
 //------------------------------------------------------------------------------
 // Compute susceptibility based on DCT-coeff histograms:
 // the higher, the "easier" the macroblock is to compress.
 
-static int CollectHistogramSSE2(const uint8_t* ref, const uint8_t* pred,
-                                int start_block, int end_block) {
-  int histo[MAX_COEFF_THRESH + 1] = { 0 };
-  int16_t out[16];
-  int j, k;
+static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
+                             int start_block, int end_block,
+                             VP8Histogram* const histo) {
   const __m128i max_coeff_thresh = _mm_set1_epi16(MAX_COEFF_THRESH);
+  int j;
   for (j = start_block; j < end_block; ++j) {
+    int16_t out[16];
+    int k;
+
     VP8FTransform(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
 
     // Convert coefficients to bin (within out[]).
@@ -47,9 +78,9 @@ static int CollectHistogramSSE2(const uint8_t* ref, const uint8_t* pred,
       const __m128i xor1 = _mm_xor_si128(out1, sign1);
       const __m128i abs0 = _mm_sub_epi16(xor0, sign0);
       const __m128i abs1 = _mm_sub_epi16(xor1, sign1);
-      // v = abs(out) >> 2
-      const __m128i v0 = _mm_srai_epi16(abs0, 2);
-      const __m128i v1 = _mm_srai_epi16(abs1, 2);
+      // v = abs(out) >> 3
+      const __m128i v0 = _mm_srai_epi16(abs0, 3);
+      const __m128i v1 = _mm_srai_epi16(abs1, 3);
       // bin = min(v, MAX_COEFF_THRESH)
       const __m128i bin0 = _mm_min_epi16(v0, max_coeff_thresh);
       const __m128i bin1 = _mm_min_epi16(v1, max_coeff_thresh);
@@ -58,21 +89,19 @@ static int CollectHistogramSSE2(const uint8_t* ref, const uint8_t* pred,
       _mm_storeu_si128((__m128i*)&out[8], bin1);
     }
 
-    // Use bin to update histogram.
+    // Convert coefficients to bin.
     for (k = 0; k < 16; ++k) {
-      histo[out[k]]++;
+      histo->distribution[out[k]]++;
     }
   }
-
-  return VP8GetAlpha(histo);
 }
 
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
 // Does one or two inverse transforms.
-static void ITransformSSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
-                           int do_two) {
+static void ITransform(const uint8_t* ref, const int16_t* in, uint8_t* dst,
+                       int do_two) {
   // This implementation makes use of 16-bit fixed point versions of two
   // multiply constants:
   //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
@@ -243,7 +272,7 @@ static void ITransformSSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
 
   // Add inverse transform to 'ref' and store.
   {
-    const __m128i zero = _mm_set1_epi16(0);
+    const __m128i zero = _mm_setzero_si128();
     // Load the reference(s).
     __m128i ref0, ref1, ref2, ref3;
     if (do_two) {
@@ -291,21 +320,26 @@ static void ITransformSSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
   }
 }
 
-static void FTransformSSE2(const uint8_t* src, const uint8_t* ref,
-                           int16_t* out) {
+static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i seven = _mm_set1_epi16(7);
-  const __m128i k7500 = _mm_set1_epi32(7500);
-  const __m128i k14500 = _mm_set1_epi32(14500);
+  const __m128i k937 = _mm_set1_epi32(937);
+  const __m128i k1812 = _mm_set1_epi32(1812);
   const __m128i k51000 = _mm_set1_epi32(51000);
   const __m128i k12000_plus_one = _mm_set1_epi32(12000 + (1 << 16));
   const __m128i k5352_2217 = _mm_set_epi16(5352,  2217, 5352,  2217,
                                            5352,  2217, 5352,  2217);
   const __m128i k2217_5352 = _mm_set_epi16(2217, -5352, 2217, -5352,
                                            2217, -5352, 2217, -5352);
-
+  const __m128i k88p = _mm_set_epi16(8, 8, 8, 8, 8, 8, 8, 8);
+  const __m128i k88m = _mm_set_epi16(-8, 8, -8, 8, -8, 8, -8, 8);
+  const __m128i k5352_2217p = _mm_set_epi16(2217, 5352, 2217, 5352,
+                                            2217, 5352, 2217, 5352);
+  const __m128i k5352_2217m = _mm_set_epi16(-5352, 2217, -5352, 2217,
+                                            -5352, 2217, -5352, 2217);
   __m128i v01, v32;
 
+
   // Difference between src and ref and initial transpose.
   {
     // Load src and convert to 16b.
@@ -326,73 +360,52 @@ static void FTransformSSE2(const uint8_t* src, const uint8_t* ref,
     const __m128i ref_1 = _mm_unpacklo_epi8(ref1, zero);
     const __m128i ref_2 = _mm_unpacklo_epi8(ref2, zero);
     const __m128i ref_3 = _mm_unpacklo_epi8(ref3, zero);
-    // Compute difference.
+    // Compute difference. -> 00 01 02 03 00 00 00 00
     const __m128i diff0 = _mm_sub_epi16(src_0, ref_0);
     const __m128i diff1 = _mm_sub_epi16(src_1, ref_1);
     const __m128i diff2 = _mm_sub_epi16(src_2, ref_2);
     const __m128i diff3 = _mm_sub_epi16(src_3, ref_3);
 
-    // Transpose.
+
+    // Unpack and shuffle
     // 00 01 02 03   0 0 0 0
     // 10 11 12 13   0 0 0 0
     // 20 21 22 23   0 0 0 0
     // 30 31 32 33   0 0 0 0
-    const __m128i transpose0_0 = _mm_unpacklo_epi16(diff0, diff1);
-    const __m128i transpose0_1 = _mm_unpacklo_epi16(diff2, diff3);
-    // 00 10 01 11   02 12 03 13
-    // 20 30 21 31   22 32 23 33
-    const __m128i v23 = _mm_unpackhi_epi32(transpose0_0, transpose0_1);
-    v01 = _mm_unpacklo_epi32(transpose0_0, transpose0_1);
-    v32 = _mm_shuffle_epi32(v23, _MM_SHUFFLE(1, 0, 3, 2));
-    // a02 a12 a22 a32   a03 a13 a23 a33
-    // a00 a10 a20 a30   a01 a11 a21 a31
-    // a03 a13 a23 a33   a02 a12 a22 a32
-  }
+    const __m128i shuf01 = _mm_unpacklo_epi32(diff0, diff1);
+    const __m128i shuf23 = _mm_unpacklo_epi32(diff2, diff3);
+    // 00 01 10 11 02 03 12 13
+    // 20 21 30 31 22 23 32 33
+    const __m128i shuf01_p =
+        _mm_shufflehi_epi16(shuf01, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128i shuf23_p =
+        _mm_shufflehi_epi16(shuf23, _MM_SHUFFLE(2, 3, 0, 1));
+    // 00 01 10 11 03 02 13 12
+    // 20 21 30 31 23 22 33 32
+    const __m128i s01 = _mm_unpacklo_epi64(shuf01_p, shuf23_p);
+    const __m128i s32 = _mm_unpackhi_epi64(shuf01_p, shuf23_p);
+    // 00 01 10 11 20 21 30 31
+    // 03 02 13 12 23 22 33 32
+    const __m128i a01 = _mm_add_epi16(s01, s32);
+    const __m128i a32 = _mm_sub_epi16(s01, s32);
+    // [d0 + d3 | d1 + d2 | ...] = [a0 a1 | a0' a1' | ... ]
+    // [d0 - d3 | d1 - d2 | ...] = [a3 a2 | a3' a2' | ... ]
 
-  // First pass and subsequent transpose.
-  {
-    // Same operations are done on the (0,3) and (1,2) pairs.
-    // b0 = (a0 + a3) << 3
-    // b1 = (a1 + a2) << 3
-    // b3 = (a0 - a3) << 3
-    // b2 = (a1 - a2) << 3
-    const __m128i a01 = _mm_add_epi16(v01, v32);
-    const __m128i a32 = _mm_sub_epi16(v01, v32);
-    const __m128i b01 = _mm_slli_epi16(a01, 3);
-    const __m128i b32 = _mm_slli_epi16(a32, 3);
-    const __m128i b11 = _mm_unpackhi_epi64(b01, b01);
-    const __m128i b22 = _mm_unpackhi_epi64(b32, b32);
-
-    // e0 = b0 + b1
-    // e2 = b0 - b1
-    const __m128i e0 = _mm_add_epi16(b01, b11);
-    const __m128i e2 = _mm_sub_epi16(b01, b11);
-    const __m128i e02 = _mm_unpacklo_epi64(e0, e2);
-
-    // e1 = (b3 * 5352 + b2 * 2217 + 14500) >> 12
-    // e3 = (b3 * 2217 - b2 * 5352 +  7500) >> 12
-    const __m128i b23 = _mm_unpacklo_epi16(b22, b32);
-    const __m128i c1 = _mm_madd_epi16(b23, k5352_2217);
-    const __m128i c3 = _mm_madd_epi16(b23, k2217_5352);
-    const __m128i d1 = _mm_add_epi32(c1, k14500);
-    const __m128i d3 = _mm_add_epi32(c3, k7500);
-    const __m128i e1 = _mm_srai_epi32(d1, 12);
-    const __m128i e3 = _mm_srai_epi32(d3, 12);
-    const __m128i e13 = _mm_packs_epi32(e1, e3);
-
-    // Transpose.
-    // 00 01 02 03  20 21 22 23
-    // 10 11 12 13  30 31 32 33
-    const __m128i transpose0_0 = _mm_unpacklo_epi16(e02, e13);
-    const __m128i transpose0_1 = _mm_unpackhi_epi16(e02, e13);
-    // 00 10 01 11   02 12 03 13
-    // 20 30 21 31   22 32 23 33
-    const __m128i v23 = _mm_unpackhi_epi32(transpose0_0, transpose0_1);
-    v01 = _mm_unpacklo_epi32(transpose0_0, transpose0_1);
-    v32 = _mm_shuffle_epi32(v23, _MM_SHUFFLE(1, 0, 3, 2));
-    // 02 12 22 32   03 13 23 33
-    // 00 10 20 30   01 11 21 31
-    // 03 13 23 33   02 12 22 32
+    const __m128i tmp0 = _mm_madd_epi16(a01, k88p);  // [ (a0 + a1) << 3, ... ]
+    const __m128i tmp2 = _mm_madd_epi16(a01, k88m);  // [ (a0 - a1) << 3, ... ]
+    const __m128i tmp1_1 = _mm_madd_epi16(a32, k5352_2217p);
+    const __m128i tmp3_1 = _mm_madd_epi16(a32, k5352_2217m);
+    const __m128i tmp1_2 = _mm_add_epi32(tmp1_1, k1812);
+    const __m128i tmp3_2 = _mm_add_epi32(tmp3_1, k937);
+    const __m128i tmp1   = _mm_srai_epi32(tmp1_2, 9);
+    const __m128i tmp3   = _mm_srai_epi32(tmp3_2, 9);
+    const __m128i s03 = _mm_packs_epi32(tmp0, tmp2);
+    const __m128i s12 = _mm_packs_epi32(tmp1, tmp3);
+    const __m128i s_lo = _mm_unpacklo_epi16(s03, s12);   // 0 1 0 1 0 1...
+    const __m128i s_hi = _mm_unpackhi_epi16(s03, s12);   // 2 3 2 3 2 3
+    const __m128i v23 = _mm_unpackhi_epi32(s_lo, s_hi);
+    v01 = _mm_unpacklo_epi32(s_lo, s_hi);
+    v32 = _mm_shuffle_epi32(v23, _MM_SHUFFLE(1, 0, 3, 2));  // 3 2 3 2 3 2..
   }
 
   // Second pass
@@ -406,13 +419,12 @@ static void FTransformSSE2(const uint8_t* src, const uint8_t* ref,
     const __m128i a32 = _mm_sub_epi16(v01, v32);
     const __m128i a11 = _mm_unpackhi_epi64(a01, a01);
     const __m128i a22 = _mm_unpackhi_epi64(a32, a32);
+    const __m128i a01_plus_7 = _mm_add_epi16(a01, seven);
 
     // d0 = (a0 + a1 + 7) >> 4;
     // d2 = (a0 - a1 + 7) >> 4;
-    const __m128i b0 = _mm_add_epi16(a01, a11);
-    const __m128i b2 = _mm_sub_epi16(a01, a11);
-    const __m128i c0 = _mm_add_epi16(b0, seven);
-    const __m128i c2 = _mm_add_epi16(b2, seven);
+    const __m128i c0 = _mm_add_epi16(a01_plus_7, a11);
+    const __m128i c2 = _mm_sub_epi16(a01_plus_7, a11);
     const __m128i d0 = _mm_srai_epi16(c0, 4);
     const __m128i d2 = _mm_srai_epi16(c2, 4);
 
@@ -430,22 +442,147 @@ static void FTransformSSE2(const uint8_t* src, const uint8_t* ref,
     // f1 = f1 + (a3 != 0);
     // The compare will return (0xffff, 0) for (==0, !=0). To turn that into the
     // desired (0, 1), we add one earlier through k12000_plus_one.
+    // -> f1 = f1 + 1 - (a3 == 0)
     const __m128i g1 = _mm_add_epi16(f1, _mm_cmpeq_epi16(a32, zero));
 
-    _mm_storel_epi64((__m128i*)&out[ 0], d0);
-    _mm_storel_epi64((__m128i*)&out[ 4], g1);
-    _mm_storel_epi64((__m128i*)&out[ 8], d2);
-    _mm_storel_epi64((__m128i*)&out[12], f3);
+    const __m128i d0_g1 = _mm_unpacklo_epi64(d0, g1);
+    const __m128i d2_f3 = _mm_unpacklo_epi64(d2, f3);
+    _mm_storeu_si128((__m128i*)&out[0], d0_g1);
+    _mm_storeu_si128((__m128i*)&out[8], d2_f3);
+  }
+}
+
+static void FTransformWHT(const int16_t* in, int16_t* out) {
+  int32_t tmp[16];
+  int i;
+  for (i = 0; i < 4; ++i, in += 64) {
+    const int a0 = (in[0 * 16] + in[2 * 16]);
+    const int a1 = (in[1 * 16] + in[3 * 16]);
+    const int a2 = (in[1 * 16] - in[3 * 16]);
+    const int a3 = (in[0 * 16] - in[2 * 16]);
+    tmp[0 + i * 4] = a0 + a1;
+    tmp[1 + i * 4] = a3 + a2;
+    tmp[2 + i * 4] = a3 - a2;
+    tmp[3 + i * 4] = a0 - a1;
+  }
+  {
+    const __m128i src0 = _mm_loadu_si128((__m128i*)&tmp[0]);
+    const __m128i src1 = _mm_loadu_si128((__m128i*)&tmp[4]);
+    const __m128i src2 = _mm_loadu_si128((__m128i*)&tmp[8]);
+    const __m128i src3 = _mm_loadu_si128((__m128i*)&tmp[12]);
+    const __m128i a0 = _mm_add_epi32(src0, src2);
+    const __m128i a1 = _mm_add_epi32(src1, src3);
+    const __m128i a2 = _mm_sub_epi32(src1, src3);
+    const __m128i a3 = _mm_sub_epi32(src0, src2);
+    const __m128i b0 = _mm_srai_epi32(_mm_add_epi32(a0, a1), 1);
+    const __m128i b1 = _mm_srai_epi32(_mm_add_epi32(a3, a2), 1);
+    const __m128i b2 = _mm_srai_epi32(_mm_sub_epi32(a3, a2), 1);
+    const __m128i b3 = _mm_srai_epi32(_mm_sub_epi32(a0, a1), 1);
+    const __m128i out0 = _mm_packs_epi32(b0, b1);
+    const __m128i out1 = _mm_packs_epi32(b2, b3);
+    _mm_storeu_si128((__m128i*)&out[0], out0);
+    _mm_storeu_si128((__m128i*)&out[8], out1);
   }
 }
 
 //------------------------------------------------------------------------------
 // Metric
 
-static int SSE4x4SSE2(const uint8_t* a, const uint8_t* b) {
-  const __m128i zero = _mm_set1_epi16(0);
+static int SSE_Nx4(const uint8_t* a, const uint8_t* b,
+                   int num_quads, int do_16) {
+  const __m128i zero = _mm_setzero_si128();
+  __m128i sum1 = zero;
+  __m128i sum2 = zero;
 
-  // Load values.
+  while (num_quads-- > 0) {
+    // Note: for the !do_16 case, we read 16 pixels instead of 8 but that's ok,
+    // thanks to buffer over-allocation to that effect.
+    const __m128i a0 = _mm_loadu_si128((__m128i*)&a[BPS * 0]);
+    const __m128i a1 = _mm_loadu_si128((__m128i*)&a[BPS * 1]);
+    const __m128i a2 = _mm_loadu_si128((__m128i*)&a[BPS * 2]);
+    const __m128i a3 = _mm_loadu_si128((__m128i*)&a[BPS * 3]);
+    const __m128i b0 = _mm_loadu_si128((__m128i*)&b[BPS * 0]);
+    const __m128i b1 = _mm_loadu_si128((__m128i*)&b[BPS * 1]);
+    const __m128i b2 = _mm_loadu_si128((__m128i*)&b[BPS * 2]);
+    const __m128i b3 = _mm_loadu_si128((__m128i*)&b[BPS * 3]);
+
+    // compute clip0(a-b) and clip0(b-a)
+    const __m128i a0p = _mm_subs_epu8(a0, b0);
+    const __m128i a0m = _mm_subs_epu8(b0, a0);
+    const __m128i a1p = _mm_subs_epu8(a1, b1);
+    const __m128i a1m = _mm_subs_epu8(b1, a1);
+    const __m128i a2p = _mm_subs_epu8(a2, b2);
+    const __m128i a2m = _mm_subs_epu8(b2, a2);
+    const __m128i a3p = _mm_subs_epu8(a3, b3);
+    const __m128i a3m = _mm_subs_epu8(b3, a3);
+
+    // compute |a-b| with 8b arithmetic as clip0(a-b) | clip0(b-a)
+    const __m128i diff0 = _mm_or_si128(a0p, a0m);
+    const __m128i diff1 = _mm_or_si128(a1p, a1m);
+    const __m128i diff2 = _mm_or_si128(a2p, a2m);
+    const __m128i diff3 = _mm_or_si128(a3p, a3m);
+
+    // unpack (only four operations, instead of eight)
+    const __m128i low0 = _mm_unpacklo_epi8(diff0, zero);
+    const __m128i low1 = _mm_unpacklo_epi8(diff1, zero);
+    const __m128i low2 = _mm_unpacklo_epi8(diff2, zero);
+    const __m128i low3 = _mm_unpacklo_epi8(diff3, zero);
+
+    // multiply with self
+    const __m128i low_madd0 = _mm_madd_epi16(low0, low0);
+    const __m128i low_madd1 = _mm_madd_epi16(low1, low1);
+    const __m128i low_madd2 = _mm_madd_epi16(low2, low2);
+    const __m128i low_madd3 = _mm_madd_epi16(low3, low3);
+
+    // collect in a cascading way
+    const __m128i low_sum0 = _mm_add_epi32(low_madd0, low_madd1);
+    const __m128i low_sum1 = _mm_add_epi32(low_madd2, low_madd3);
+    sum1 = _mm_add_epi32(sum1, low_sum0);
+    sum2 = _mm_add_epi32(sum2, low_sum1);
+
+    if (do_16) {  // if necessary, process the higher 8 bytes similarly
+      const __m128i hi0 = _mm_unpackhi_epi8(diff0, zero);
+      const __m128i hi1 = _mm_unpackhi_epi8(diff1, zero);
+      const __m128i hi2 = _mm_unpackhi_epi8(diff2, zero);
+      const __m128i hi3 = _mm_unpackhi_epi8(diff3, zero);
+
+      const __m128i hi_madd0 = _mm_madd_epi16(hi0, hi0);
+      const __m128i hi_madd1 = _mm_madd_epi16(hi1, hi1);
+      const __m128i hi_madd2 = _mm_madd_epi16(hi2, hi2);
+      const __m128i hi_madd3 = _mm_madd_epi16(hi3, hi3);
+      const __m128i hi_sum0 = _mm_add_epi32(hi_madd0, hi_madd1);
+      const __m128i hi_sum1 = _mm_add_epi32(hi_madd2, hi_madd3);
+      sum1 = _mm_add_epi32(sum1, hi_sum0);
+      sum2 = _mm_add_epi32(sum2, hi_sum1);
+    }
+    a += 4 * BPS;
+    b += 4 * BPS;
+  }
+  {
+    int32_t tmp[4];
+    const __m128i sum = _mm_add_epi32(sum1, sum2);
+    _mm_storeu_si128((__m128i*)tmp, sum);
+    return (tmp[3] + tmp[2] + tmp[1] + tmp[0]);
+  }
+}
+
+static int SSE16x16(const uint8_t* a, const uint8_t* b) {
+  return SSE_Nx4(a, b, 4, 1);
+}
+
+static int SSE16x8(const uint8_t* a, const uint8_t* b) {
+  return SSE_Nx4(a, b, 2, 1);
+}
+
+static int SSE8x8(const uint8_t* a, const uint8_t* b) {
+  return SSE_Nx4(a, b, 2, 0);
+}
+
+static int SSE4x4(const uint8_t* a, const uint8_t* b) {
+  const __m128i zero = _mm_setzero_si128();
+
+  // Load values. Note that we read 8 pixels instead of 4,
+  // but the a/b buffers are over-allocated to that effect.
   const __m128i a0 = _mm_loadl_epi64((__m128i*)&a[BPS * 0]);
   const __m128i a1 = _mm_loadl_epi64((__m128i*)&a[BPS * 1]);
   const __m128i a2 = _mm_loadl_epi64((__m128i*)&a[BPS * 2]);
@@ -483,6 +620,7 @@ static int SSE4x4SSE2(const uint8_t* a, const uint8_t* b) {
   const __m128i sum0 = _mm_add_epi32(madd0, madd1);
   const __m128i sum1 = _mm_add_epi32(madd2, madd3);
   const __m128i sum2 = _mm_add_epi32(sum0, sum1);
+
   int32_t tmp[4];
   _mm_storeu_si128((__m128i*)tmp, sum2);
   return (tmp[3] + tmp[2] + tmp[1] + tmp[0]);
@@ -497,15 +635,13 @@ static int SSE4x4SSE2(const uint8_t* a, const uint8_t* b) {
 // Hadamard transform
 // Returns the difference between the weighted sum of the absolute value of
 // transformed coefficients.
-static int TTransformSSE2(const uint8_t* inA, const uint8_t* inB,
-                          const uint16_t* const w) {
+static int TTransform(const uint8_t* inA, const uint8_t* inB,
+                      const uint16_t* const w) {
   int32_t sum[4];
   __m128i tmp_0, tmp_1, tmp_2, tmp_3;
   const __m128i zero = _mm_setzero_si128();
-  const __m128i one = _mm_set1_epi16(1);
-  const __m128i three = _mm_set1_epi16(3);
 
-  // Load, combine and tranpose inputs.
+  // Load, combine and transpose inputs.
   {
     const __m128i inA_0 = _mm_loadl_epi64((__m128i*)&inA[BPS * 0]);
     const __m128i inA_1 = _mm_loadl_epi64((__m128i*)&inA[BPS * 1]);
@@ -550,17 +686,14 @@ static int TTransformSSE2(const uint8_t* inA, const uint8_t* inB,
   // Horizontal pass and subsequent transpose.
   {
     // Calculate a and b (two 4x4 at once).
-    const __m128i a0 = _mm_slli_epi16(_mm_add_epi16(tmp_0, tmp_2), 2);
-    const __m128i a1 = _mm_slli_epi16(_mm_add_epi16(tmp_1, tmp_3), 2);
-    const __m128i a2 = _mm_slli_epi16(_mm_sub_epi16(tmp_1, tmp_3), 2);
-    const __m128i a3 = _mm_slli_epi16(_mm_sub_epi16(tmp_0, tmp_2), 2);
-    // b0_extra = (a0 != 0);
-    const __m128i b0_extra = _mm_andnot_si128(_mm_cmpeq_epi16 (a0, zero), one);
-    const __m128i b0_base = _mm_add_epi16(a0, a1);
+    const __m128i a0 = _mm_add_epi16(tmp_0, tmp_2);
+    const __m128i a1 = _mm_add_epi16(tmp_1, tmp_3);
+    const __m128i a2 = _mm_sub_epi16(tmp_1, tmp_3);
+    const __m128i a3 = _mm_sub_epi16(tmp_0, tmp_2);
+    const __m128i b0 = _mm_add_epi16(a0, a1);
     const __m128i b1 = _mm_add_epi16(a3, a2);
     const __m128i b2 = _mm_sub_epi16(a3, a2);
     const __m128i b3 = _mm_sub_epi16(a0, a1);
-    const __m128i b0 = _mm_add_epi16(b0_base, b0_extra);
     // a00 a01 a02 a03   b00 b01 b02 b03
     // a10 a11 a12 a13   b10 b11 b12 b13
     // a20 a21 a22 a23   b20 b21 b22 b23
@@ -635,19 +768,6 @@ static int TTransformSSE2(const uint8_t* inA, const uint8_t* inB,
       B_b2 = _mm_sub_epi16(B_b2, sign_B_b2);
     }
 
-    // b = abs(b) + 3
-    A_b0 = _mm_add_epi16(A_b0, three);
-    A_b2 = _mm_add_epi16(A_b2, three);
-    B_b0 = _mm_add_epi16(B_b0, three);
-    B_b2 = _mm_add_epi16(B_b2, three);
-
-    // abs((b + (b<0) + 3) >> 3) = (abs(b) + 3) >> 3
-    // b = (abs(b) + 3) >> 3
-    A_b0 = _mm_srai_epi16(A_b0, 3);
-    A_b2 = _mm_srai_epi16(A_b2, 3);
-    B_b0 = _mm_srai_epi16(B_b0, 3);
-    B_b2 = _mm_srai_epi16(B_b2, 3);
-
     // weighted sums
     A_b0 = _mm_madd_epi16(A_b0, w_0);
     A_b2 = _mm_madd_epi16(A_b2, w_8);
@@ -663,35 +783,33 @@ static int TTransformSSE2(const uint8_t* inA, const uint8_t* inB,
   return sum[0] + sum[1] + sum[2] + sum[3];
 }
 
-static int Disto4x4SSE2(const uint8_t* const a, const uint8_t* const b,
-                        const uint16_t* const w) {
-  const int diff_sum = TTransformSSE2(a, b, w);
-  return (abs(diff_sum) + 8) >> 4;
+static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
+                    const uint16_t* const w) {
+  const int diff_sum = TTransform(a, b, w);
+  return abs(diff_sum) >> 5;
 }
 
-static int Disto16x16SSE2(const uint8_t* const a, const uint8_t* const b,
-                          const uint16_t* const w) {
+static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
+                      const uint16_t* const w) {
   int D = 0;
   int x, y;
   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
     for (x = 0; x < 16; x += 4) {
-      D += Disto4x4SSE2(a + x + y, b + x + y, w);
+      D += Disto4x4(a + x + y, b + x + y, w);
     }
   }
   return D;
 }
 
-
 //------------------------------------------------------------------------------
 // Quantization
 //
 
-// Simple quantization
-static int QuantizeBlockSSE2(int16_t in[16], int16_t out[16],
-                             int n, const VP8Matrix* const mtx) {
-  const __m128i max_coeff_2047 = _mm_set1_epi16(2047);
-  const __m128i zero = _mm_set1_epi16(0);
-  __m128i sign0, sign8;
+static WEBP_INLINE int DoQuantizeBlock(int16_t in[16], int16_t out[16],
+                                       const uint16_t* const sharpen,
+                                       const VP8Matrix* const mtx) {
+  const __m128i max_coeff_2047 = _mm_set1_epi16(MAX_LEVEL);
+  const __m128i zero = _mm_setzero_si128();
   __m128i coeff0, coeff8;
   __m128i out0, out8;
   __m128i packed_out;
@@ -701,20 +819,14 @@ static int QuantizeBlockSSE2(int16_t in[16], int16_t out[16],
   //                  we can use _mm_load_si128 instead of _mm_loadu_si128.
   __m128i in0 = _mm_loadu_si128((__m128i*)&in[0]);
   __m128i in8 = _mm_loadu_si128((__m128i*)&in[8]);
-  const __m128i sharpen0 = _mm_loadu_si128((__m128i*)&mtx->sharpen_[0]);
-  const __m128i sharpen8 = _mm_loadu_si128((__m128i*)&mtx->sharpen_[8]);
   const __m128i iq0 = _mm_loadu_si128((__m128i*)&mtx->iq_[0]);
   const __m128i iq8 = _mm_loadu_si128((__m128i*)&mtx->iq_[8]);
-  const __m128i bias0 = _mm_loadu_si128((__m128i*)&mtx->bias_[0]);
-  const __m128i bias8 = _mm_loadu_si128((__m128i*)&mtx->bias_[8]);
   const __m128i q0 = _mm_loadu_si128((__m128i*)&mtx->q_[0]);
   const __m128i q8 = _mm_loadu_si128((__m128i*)&mtx->q_[8]);
-  const __m128i zthresh0 = _mm_loadu_si128((__m128i*)&mtx->zthresh_[0]);
-  const __m128i zthresh8 = _mm_loadu_si128((__m128i*)&mtx->zthresh_[8]);
 
-  // sign(in) = in >> 15  (0x0000 if positive, 0xffff if negative)
-  sign0 = _mm_srai_epi16(in0, 15);
-  sign8 = _mm_srai_epi16(in8, 15);
+  // extract sign(in)  (0x0000 if positive, 0xffff if negative)
+  const __m128i sign0 = _mm_cmpgt_epi16(zero, in0);
+  const __m128i sign8 = _mm_cmpgt_epi16(zero, in8);
 
   // coeff = abs(in) = (in ^ sign) - sign
   coeff0 = _mm_xor_si128(in0, sign0);
@@ -723,43 +835,47 @@ static int QuantizeBlockSSE2(int16_t in[16], int16_t out[16],
   coeff8 = _mm_sub_epi16(coeff8, sign8);
 
   // coeff = abs(in) + sharpen
-  coeff0 = _mm_add_epi16(coeff0, sharpen0);
-  coeff8 = _mm_add_epi16(coeff8, sharpen8);
+  if (sharpen != NULL) {
+    const __m128i sharpen0 = _mm_loadu_si128((__m128i*)&sharpen[0]);
+    const __m128i sharpen8 = _mm_loadu_si128((__m128i*)&sharpen[8]);
+    coeff0 = _mm_add_epi16(coeff0, sharpen0);
+    coeff8 = _mm_add_epi16(coeff8, sharpen8);
+  }
 
-  // if (coeff > 2047) coeff = 2047
-  coeff0 = _mm_min_epi16(coeff0, max_coeff_2047);
-  coeff8 = _mm_min_epi16(coeff8, max_coeff_2047);
-
-  // out = (coeff * iQ + B) >> QFIX;
+  // out = (coeff * iQ + B) >> QFIX
   {
     // doing calculations with 32b precision (QFIX=17)
     // out = (coeff * iQ)
-    __m128i coeff_iQ0H = _mm_mulhi_epu16(coeff0, iq0);
-    __m128i coeff_iQ0L = _mm_mullo_epi16(coeff0, iq0);
-    __m128i coeff_iQ8H = _mm_mulhi_epu16(coeff8, iq8);
-    __m128i coeff_iQ8L = _mm_mullo_epi16(coeff8, iq8);
+    const __m128i coeff_iQ0H = _mm_mulhi_epu16(coeff0, iq0);
+    const __m128i coeff_iQ0L = _mm_mullo_epi16(coeff0, iq0);
+    const __m128i coeff_iQ8H = _mm_mulhi_epu16(coeff8, iq8);
+    const __m128i coeff_iQ8L = _mm_mullo_epi16(coeff8, iq8);
     __m128i out_00 = _mm_unpacklo_epi16(coeff_iQ0L, coeff_iQ0H);
     __m128i out_04 = _mm_unpackhi_epi16(coeff_iQ0L, coeff_iQ0H);
     __m128i out_08 = _mm_unpacklo_epi16(coeff_iQ8L, coeff_iQ8H);
     __m128i out_12 = _mm_unpackhi_epi16(coeff_iQ8L, coeff_iQ8H);
-    // expand bias from 16b to 32b
-    __m128i bias_00 = _mm_unpacklo_epi16(bias0, zero);
-    __m128i bias_04 = _mm_unpackhi_epi16(bias0, zero);
-    __m128i bias_08 = _mm_unpacklo_epi16(bias8, zero);
-    __m128i bias_12 = _mm_unpackhi_epi16(bias8, zero);
     // out = (coeff * iQ + B)
+    const __m128i bias_00 = _mm_loadu_si128((__m128i*)&mtx->bias_[0]);
+    const __m128i bias_04 = _mm_loadu_si128((__m128i*)&mtx->bias_[4]);
+    const __m128i bias_08 = _mm_loadu_si128((__m128i*)&mtx->bias_[8]);
+    const __m128i bias_12 = _mm_loadu_si128((__m128i*)&mtx->bias_[12]);
     out_00 = _mm_add_epi32(out_00, bias_00);
     out_04 = _mm_add_epi32(out_04, bias_04);
     out_08 = _mm_add_epi32(out_08, bias_08);
     out_12 = _mm_add_epi32(out_12, bias_12);
-    // out = (coeff * iQ + B) >> QFIX;
+    // out = QUANTDIV(coeff, iQ, B, QFIX)
     out_00 = _mm_srai_epi32(out_00, QFIX);
     out_04 = _mm_srai_epi32(out_04, QFIX);
     out_08 = _mm_srai_epi32(out_08, QFIX);
     out_12 = _mm_srai_epi32(out_12, QFIX);
+
     // pack result as 16b
     out0 = _mm_packs_epi32(out_00, out_04);
     out8 = _mm_packs_epi32(out_08, out_12);
+
+    // if (coeff > 2047) coeff = 2047
+    out0 = _mm_min_epi16(out0, max_coeff_2047);
+    out8 = _mm_min_epi16(out8, max_coeff_2047);
   }
 
   // get sign back (if (sign[j]) out_n = -out_n)
@@ -772,17 +888,8 @@ static int QuantizeBlockSSE2(int16_t in[16], int16_t out[16],
   in0 = _mm_mullo_epi16(out0, q0);
   in8 = _mm_mullo_epi16(out8, q8);
 
-  // if (coeff <= mtx->zthresh_) {in=0; out=0;}
-  {
-    __m128i cmp0 = _mm_cmpgt_epi16(coeff0, zthresh0);
-    __m128i cmp8 = _mm_cmpgt_epi16(coeff8, zthresh8);
-    in0 = _mm_and_si128(in0, cmp0);
-    in8 = _mm_and_si128(in8, cmp8);
-    _mm_storeu_si128((__m128i*)&in[0], in0);
-    _mm_storeu_si128((__m128i*)&in[8], in8);
-    out0 = _mm_and_si128(out0, cmp0);
-    out8 = _mm_and_si128(out8, cmp8);
-  }
+  _mm_storeu_si128((__m128i*)&in[0], in0);
+  _mm_storeu_si128((__m128i*)&in[8], in8);
 
   // zigzag the output before storing it.
   //
@@ -809,14 +916,44 @@ static int QuantizeBlockSSE2(int16_t in[16], int16_t out[16],
   }
 
   // detect if all 'out' values are zeroes or not
-  {
-    int32_t tmp[4];
-    _mm_storeu_si128((__m128i*)tmp, packed_out);
-    if (n) {
-      tmp[0] &= ~0xff;
-    }
-    return (tmp[3] || tmp[2] || tmp[1] || tmp[0]);
-  }
+  return (_mm_movemask_epi8(_mm_cmpeq_epi8(packed_out, zero)) != 0xffff);
+}
+
+static int QuantizeBlock(int16_t in[16], int16_t out[16],
+                         const VP8Matrix* const mtx) {
+  return DoQuantizeBlock(in, out, &mtx->sharpen_[0], mtx);
+}
+
+static int QuantizeBlockWHT(int16_t in[16], int16_t out[16],
+                            const VP8Matrix* const mtx) {
+  return DoQuantizeBlock(in, out, NULL, mtx);
+}
+
+// Forward declaration.
+void VP8SetResidualCoeffsSSE2(const int16_t* const coeffs,
+                              VP8Residual* const res);
+
+void VP8SetResidualCoeffsSSE2(const int16_t* const coeffs,
+                              VP8Residual* const res) {
+  const __m128i c0 = _mm_loadu_si128((const __m128i*)coeffs);
+  const __m128i c1 = _mm_loadu_si128((const __m128i*)(coeffs + 8));
+  // Use SSE to compare 8 values with a single instruction.
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i m0 = _mm_cmpeq_epi16(c0, zero);
+  const __m128i m1 = _mm_cmpeq_epi16(c1, zero);
+  // Get the comparison results as a bitmask, consisting of two times 16 bits:
+  // two identical bits for each result. Concatenate both bitmasks to get a
+  // single 32 bit value. Negate the mask to get the position of entries that
+  // are not equal to zero. We don't need to mask out least significant bits
+  // according to res->first, since coeffs[0] is 0 if res->first > 0
+  const uint32_t mask =
+      ~(((uint32_t)_mm_movemask_epi8(m1) << 16) | _mm_movemask_epi8(m0));
+  // The position of the most significant non-zero bit indicates the position of
+  // the last non-zero value. Divide the result by two because __movemask_epi8
+  // operates on 8 bit values instead of 16 bit values.
+  assert(res->first == 0 || coeffs[0] == 0);
+  res->last = mask ? (BitsLog2Floor(mask) >> 1) : -1;
+  res->coeffs = coeffs;
 }
 
 #endif   // WEBP_USE_SSE2
@@ -828,16 +965,18 @@ extern void VP8EncDspInitSSE2(void);
 
 void VP8EncDspInitSSE2(void) {
 #if defined(WEBP_USE_SSE2)
-  VP8CollectHistogram = CollectHistogramSSE2;
-  VP8EncQuantizeBlock = QuantizeBlockSSE2;
-  VP8ITransform = ITransformSSE2;
-  VP8FTransform = FTransformSSE2;
-  VP8SSE4x4 = SSE4x4SSE2;
-  VP8TDisto4x4 = Disto4x4SSE2;
-  VP8TDisto16x16 = Disto16x16SSE2;
+  VP8CollectHistogram = CollectHistogram;
+  VP8EncQuantizeBlock = QuantizeBlock;
+  VP8EncQuantizeBlockWHT = QuantizeBlockWHT;
+  VP8ITransform = ITransform;
+  VP8FTransform = FTransform;
+  VP8FTransformWHT = FTransformWHT;
+  VP8SSE16x16 = SSE16x16;
+  VP8SSE16x8 = SSE16x8;
+  VP8SSE8x8 = SSE8x8;
+  VP8SSE4x4 = SSE4x4;
+  VP8TDisto4x4 = Disto4x4;
+  VP8TDisto16x16 = Disto16x16;
 #endif   // WEBP_USE_SSE2
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif

src/dsp/lossless.h

@@ -1,8 +1,10 @@
 // Copyright 2012 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Image transforms and color space conversion methods for lossless decoder.
@@ -16,10 +18,62 @@
 #include "../webp/types.h"
 #include "../webp/decode.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#include "../enc/histogram.h"
+#include "../utils/utils.h"
+
+#ifdef __cplusplus
 extern "C" {
 #endif
 
+//------------------------------------------------------------------------------
+// Signatures and generic function-pointers
+
+typedef uint32_t (*VP8LPredictorFunc)(uint32_t left, const uint32_t* const top);
+extern VP8LPredictorFunc VP8LPredictors[16];
+
+typedef void (*VP8LProcessBlueAndRedFunc)(uint32_t* argb_data, int num_pixels);
+extern VP8LProcessBlueAndRedFunc VP8LSubtractGreenFromBlueAndRed;
+extern VP8LProcessBlueAndRedFunc VP8LAddGreenToBlueAndRed;
+
+typedef struct {
+  // Note: the members are uint8_t, so that any negative values are
+  // automatically converted to "mod 256" values.
+  uint8_t green_to_red_;
+  uint8_t green_to_blue_;
+  uint8_t red_to_blue_;
+} VP8LMultipliers;
+typedef void (*VP8LTransformColorFunc)(const VP8LMultipliers* const m,
+                                       uint32_t* argb_data, int num_pixels);
+extern VP8LTransformColorFunc VP8LTransformColor;
+extern VP8LTransformColorFunc VP8LTransformColorInverse;
+
+typedef void (*VP8LConvertFunc)(const uint32_t* src, int num_pixels,
+                                uint8_t* dst);
+extern VP8LConvertFunc VP8LConvertBGRAToRGB;
+extern VP8LConvertFunc VP8LConvertBGRAToRGBA;
+extern VP8LConvertFunc VP8LConvertBGRAToRGBA4444;
+extern VP8LConvertFunc VP8LConvertBGRAToRGB565;
+extern VP8LConvertFunc VP8LConvertBGRAToBGR;
+
+// Expose some C-only fallback functions
+void VP8LTransformColor_C(const VP8LMultipliers* const m,
+                          uint32_t* data, int num_pixels);
+void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
+                                 uint32_t* data, int num_pixels);
+
+void VP8LConvertBGRAToRGB_C(const uint32_t* src, int num_pixels, uint8_t* dst);
+void VP8LConvertBGRAToRGBA_C(const uint32_t* src, int num_pixels, uint8_t* dst);
+void VP8LConvertBGRAToRGBA4444_C(const uint32_t* src,
+                                 int num_pixels, uint8_t* dst);
+void VP8LConvertBGRAToRGB565_C(const uint32_t* src,
+                               int num_pixels, uint8_t* dst);
+void VP8LConvertBGRAToBGR_C(const uint32_t* src, int num_pixels, uint8_t* dst);
+void VP8LSubtractGreenFromBlueAndRed_C(uint32_t* argb_data, int num_pixels);
+void VP8LAddGreenToBlueAndRed_C(uint32_t* data, int num_pixels);
+
+// Must be called before calling any of the above methods.
+void VP8LDspInit(void);
+
 //------------------------------------------------------------------------------
 // Image transforms.
 
@@ -33,14 +87,18 @@ void VP8LInverseTransform(const struct VP8LTransform* const transform,
                           int row_start, int row_end,
                           const uint32_t* const in, uint32_t* const out);
 
-// Subtracts green from blue and red channels.
-void VP8LSubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixs);
+// Similar to the static method ColorIndexInverseTransform() that is part of
+// lossless.c, but used only for alpha decoding. It takes uint8_t (rather than
+// uint32_t) arguments for 'src' and 'dst'.
+void VP8LColorIndexInverseTransformAlpha(
+    const struct VP8LTransform* const transform, int y_start, int y_end,
+    const uint8_t* src, uint8_t* dst);
 
 void VP8LResidualImage(int width, int height, int bits,
                        uint32_t* const argb, uint32_t* const argb_scratch,
                        uint32_t* const image);
 
-void VP8LColorSpaceTransform(int width, int height, int bits, int step,
+void VP8LColorSpaceTransform(int width, int height, int bits, int quality,
                              uint32_t* const argb, uint32_t* image);
 
 //------------------------------------------------------------------------------
@@ -59,10 +117,116 @@ static WEBP_INLINE uint32_t VP8LSubSampleSize(uint32_t size,
   return (size + (1 << sampling_bits) - 1) >> sampling_bits;
 }
 
-// Faster logarithm for integers, with the property of log2(0) == 0.
-float VP8LFastLog2(int v);
+// -----------------------------------------------------------------------------
+// Faster logarithm for integers. Small values use a look-up table.
+#define LOG_LOOKUP_IDX_MAX 256
+extern const float kLog2Table[LOG_LOOKUP_IDX_MAX];
+extern const float kSLog2Table[LOG_LOOKUP_IDX_MAX];
+typedef float (*VP8LFastLog2SlowFunc)(uint32_t v);
+
+extern VP8LFastLog2SlowFunc VP8LFastLog2Slow;
+extern VP8LFastLog2SlowFunc VP8LFastSLog2Slow;
+
+static WEBP_INLINE float VP8LFastLog2(uint32_t v) {
+  return (v < LOG_LOOKUP_IDX_MAX) ? kLog2Table[v] : VP8LFastLog2Slow(v);
+}
 // Fast calculation of v * log2(v) for integer input.
-static WEBP_INLINE float VP8LFastSLog2(int v) { return VP8LFastLog2(v) * v; }
+static WEBP_INLINE float VP8LFastSLog2(uint32_t v) {
+  return (v < LOG_LOOKUP_IDX_MAX) ? kSLog2Table[v] : VP8LFastSLog2Slow(v);
+}
+
+// -----------------------------------------------------------------------------
+// Huffman-cost related functions.
+
+typedef double (*VP8LCostFunc)(const uint32_t* population, int length);
+typedef double (*VP8LCostCombinedFunc)(const uint32_t* X, const uint32_t* Y,
+                                       int length);
+
+extern VP8LCostFunc VP8LExtraCost;
+extern VP8LCostCombinedFunc VP8LExtraCostCombined;
+
+typedef struct {        // small struct to hold counters
+  int counts[2];        // index: 0=zero steak, 1=non-zero streak
+  int streaks[2][2];    // [zero/non-zero][streak<3 / streak>=3]
+} VP8LStreaks;
+
+typedef VP8LStreaks (*VP8LCostCountFunc)(const uint32_t* population,
+                                         int length);
+typedef VP8LStreaks (*VP8LCostCombinedCountFunc)(const uint32_t* X,
+                                                 const uint32_t* Y, int length);
+
+extern VP8LCostCountFunc VP8LHuffmanCostCount;
+extern VP8LCostCombinedCountFunc VP8LHuffmanCostCombinedCount;
+
+typedef void (*VP8LHistogramAddFunc)(const VP8LHistogram* const a,
+                                     const VP8LHistogram* const b,
+                                     VP8LHistogram* const out);
+extern VP8LHistogramAddFunc VP8LHistogramAdd;
+
+// -----------------------------------------------------------------------------
+// PrefixEncode()
+
+static WEBP_INLINE int VP8LBitsLog2Ceiling(uint32_t n) {
+  const int log_floor = BitsLog2Floor(n);
+  if (n == (n & ~(n - 1)))  // zero or a power of two.
+    return log_floor;
+  else
+    return log_floor + 1;
+}
+
+// Splitting of distance and length codes into prefixes and
+// extra bits. The prefixes are encoded with an entropy code
+// while the extra bits are stored just as normal bits.
+static WEBP_INLINE void VP8LPrefixEncodeBitsNoLUT(int distance, int* const code,
+                                                  int* const extra_bits) {
+  const int highest_bit = BitsLog2Floor(--distance);
+  const int second_highest_bit = (distance >> (highest_bit - 1)) & 1;
+  *extra_bits = highest_bit - 1;
+  *code = 2 * highest_bit + second_highest_bit;
+}
+
+static WEBP_INLINE void VP8LPrefixEncodeNoLUT(int distance, int* const code,
+                                              int* const extra_bits,
+                                              int* const extra_bits_value) {
+  const int highest_bit = BitsLog2Floor(--distance);
+  const int second_highest_bit = (distance >> (highest_bit - 1)) & 1;
+  *extra_bits = highest_bit - 1;
+  *extra_bits_value = distance & ((1 << *extra_bits) - 1);
+  *code = 2 * highest_bit + second_highest_bit;
+}
+
+#define PREFIX_LOOKUP_IDX_MAX   512
+typedef struct {
+  int8_t code_;
+  int8_t extra_bits_;
+} VP8LPrefixCode;
+
+// These tables are derived using VP8LPrefixEncodeNoLUT.
+extern const VP8LPrefixCode kPrefixEncodeCode[PREFIX_LOOKUP_IDX_MAX];
+extern const uint8_t kPrefixEncodeExtraBitsValue[PREFIX_LOOKUP_IDX_MAX];
+static WEBP_INLINE void VP8LPrefixEncodeBits(int distance, int* const code,
+                                             int* const extra_bits) {
+  if (distance < PREFIX_LOOKUP_IDX_MAX) {
+    const VP8LPrefixCode prefix_code = kPrefixEncodeCode[distance];
+    *code = prefix_code.code_;
+    *extra_bits = prefix_code.extra_bits_;
+  } else {
+    VP8LPrefixEncodeBitsNoLUT(distance, code, extra_bits);
+  }
+}
+
+static WEBP_INLINE void VP8LPrefixEncode(int distance, int* const code,
+                                         int* const extra_bits,
+                                         int* const extra_bits_value) {
+  if (distance < PREFIX_LOOKUP_IDX_MAX) {
+    const VP8LPrefixCode prefix_code = kPrefixEncodeCode[distance];
+    *code = prefix_code.code_;
+    *extra_bits = prefix_code.extra_bits_;
+    *extra_bits_value = kPrefixEncodeExtraBitsValue[distance];
+  } else {
+    VP8LPrefixEncodeNoLUT(distance, code, extra_bits, extra_bits_value);
+  }
+}
 
 // In-place difference of each component with mod 256.
 static WEBP_INLINE uint32_t VP8LSubPixels(uint32_t a, uint32_t b) {
@@ -73,9 +237,12 @@ static WEBP_INLINE uint32_t VP8LSubPixels(uint32_t a, uint32_t b) {
   return (alpha_and_green & 0xff00ff00u) | (red_and_blue & 0x00ff00ffu);
 }
 
+void VP8LBundleColorMap(const uint8_t* const row, int width,
+                        int xbits, uint32_t* const dst);
+
 //------------------------------------------------------------------------------
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 

src/dsp/lossless_mips32.c

@@ -0,0 +1,416 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// MIPS version of lossless functions
+//
+// Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
+//             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
+
+#include "./dsp.h"
+#include "./lossless.h"
+
+#if defined(WEBP_USE_MIPS32)
+
+#include <assert.h>
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#define APPROX_LOG_WITH_CORRECTION_MAX  65536
+#define APPROX_LOG_MAX                   4096
+#define LOG_2_RECIPROCAL 1.44269504088896338700465094007086
+
+static float FastSLog2Slow(uint32_t v) {
+  assert(v >= LOG_LOOKUP_IDX_MAX);
+  if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
+    uint32_t log_cnt, y, correction;
+    const int c24 = 24;
+    const float v_f = (float)v;
+    uint32_t temp;
+
+    // Xf = 256 = 2^8
+    // log_cnt is index of leading one in upper 24 bits
+    __asm__ volatile(
+      "clz      %[log_cnt], %[v]                      \n\t"
+      "addiu    %[y],       $zero,        1           \n\t"
+      "subu     %[log_cnt], %[c24],       %[log_cnt]  \n\t"
+      "sllv     %[y],       %[y],         %[log_cnt]  \n\t"
+      "srlv     %[temp],    %[v],         %[log_cnt]  \n\t"
+      : [log_cnt]"=&r"(log_cnt), [y]"=&r"(y),
+        [temp]"=r"(temp)
+      : [c24]"r"(c24), [v]"r"(v)
+    );
+
+    // vf = (2^log_cnt) * Xf; where y = 2^log_cnt and Xf < 256
+    // Xf = floor(Xf) * (1 + (v % y) / v)
+    // log2(Xf) = log2(floor(Xf)) + log2(1 + (v % y) / v)
+    // The correction factor: log(1 + d) ~ d; for very small d values, so
+    // log2(1 + (v % y) / v) ~ LOG_2_RECIPROCAL * (v % y)/v
+    // LOG_2_RECIPROCAL ~ 23/16
+
+    // (v % y) = (v % 2^log_cnt) = v & (2^log_cnt - 1)
+    correction = (23 * (v & (y - 1))) >> 4;
+    return v_f * (kLog2Table[temp] + log_cnt) + correction;
+  } else {
+    return (float)(LOG_2_RECIPROCAL * v * log((double)v));
+  }
+}
+
+static float FastLog2Slow(uint32_t v) {
+  assert(v >= LOG_LOOKUP_IDX_MAX);
+  if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
+    uint32_t log_cnt, y;
+    const int c24 = 24;
+    double log_2;
+    uint32_t temp;
+
+    __asm__ volatile(
+      "clz      %[log_cnt], %[v]                      \n\t"
+      "addiu    %[y],       $zero,        1           \n\t"
+      "subu     %[log_cnt], %[c24],       %[log_cnt]  \n\t"
+      "sllv     %[y],       %[y],         %[log_cnt]  \n\t"
+      "srlv     %[temp],    %[v],         %[log_cnt]  \n\t"
+      : [log_cnt]"=&r"(log_cnt), [y]"=&r"(y),
+        [temp]"=r"(temp)
+      : [c24]"r"(c24), [v]"r"(v)
+    );
+
+    log_2 = kLog2Table[temp] + log_cnt;
+    if (v >= APPROX_LOG_MAX) {
+      // Since the division is still expensive, add this correction factor only
+      // for large values of 'v'.
+
+      const uint32_t correction = (23 * (v & (y - 1))) >> 4;
+      log_2 += (double)correction / v;
+    }
+    return (float)log_2;
+  } else {
+    return (float)(LOG_2_RECIPROCAL * log((double)v));
+  }
+}
+
+// C version of this function:
+//   int i = 0;
+//   int64_t cost = 0;
+//   const uint32_t* pop = &population[4];
+//   const uint32_t* LoopEnd = &population[length];
+//   while (pop != LoopEnd) {
+//     ++i;
+//     cost += i * *pop;
+//     cost += i * *(pop + 1);
+//     pop += 2;
+//   }
+//   return (double)cost;
+static double ExtraCost(const uint32_t* const population, int length) {
+  int i, temp0, temp1;
+  const uint32_t* pop = &population[4];
+  const uint32_t* const LoopEnd = &population[length];
+
+  __asm__ volatile(
+    "mult   $zero,    $zero                  \n\t"
+    "xor    %[i],     %[i],       %[i]       \n\t"
+    "beq    %[pop],   %[LoopEnd], 2f         \n\t"
+  "1:                                        \n\t"
+    "lw     %[temp0], 0(%[pop])              \n\t"
+    "lw     %[temp1], 4(%[pop])              \n\t"
+    "addiu  %[i],     %[i],       1          \n\t"
+    "addiu  %[pop],   %[pop],     8          \n\t"
+    "madd   %[i],     %[temp0]               \n\t"
+    "madd   %[i],     %[temp1]               \n\t"
+    "bne    %[pop],   %[LoopEnd], 1b         \n\t"
+  "2:                                        \n\t"
+    "mfhi   %[temp0]                         \n\t"
+    "mflo   %[temp1]                         \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
+      [i]"=&r"(i), [pop]"+r"(pop)
+    : [LoopEnd]"r"(LoopEnd)
+    : "memory", "hi", "lo"
+  );
+
+  return (double)((int64_t)temp0 << 32 | temp1);
+}
+
+// C version of this function:
+//   int i = 0;
+//   int64_t cost = 0;
+//   const uint32_t* pX = &X[4];
+//   const uint32_t* pY = &Y[4];
+//   const uint32_t* LoopEnd = &X[length];
+//   while (pX != LoopEnd) {
+//     const uint32_t xy0 = *pX + *pY;
+//     const uint32_t xy1 = *(pX + 1) + *(pY + 1);
+//     ++i;
+//     cost += i * xy0;
+//     cost += i * xy1;
+//     pX += 2;
+//     pY += 2;
+//   }
+//   return (double)cost;
+static double ExtraCostCombined(const uint32_t* const X,
+                                const uint32_t* const Y, int length) {
+  int i, temp0, temp1, temp2, temp3;
+  const uint32_t* pX = &X[4];
+  const uint32_t* pY = &Y[4];
+  const uint32_t* const LoopEnd = &X[length];
+
+  __asm__ volatile(
+    "mult   $zero,    $zero                  \n\t"
+    "xor    %[i],     %[i],       %[i]       \n\t"
+    "beq    %[pX],    %[LoopEnd], 2f         \n\t"
+  "1:                                        \n\t"
+    "lw     %[temp0], 0(%[pX])               \n\t"
+    "lw     %[temp1], 0(%[pY])               \n\t"
+    "lw     %[temp2], 4(%[pX])               \n\t"
+    "lw     %[temp3], 4(%[pY])               \n\t"
+    "addiu  %[i],     %[i],       1          \n\t"
+    "addu   %[temp0], %[temp0],   %[temp1]   \n\t"
+    "addu   %[temp2], %[temp2],   %[temp3]   \n\t"
+    "addiu  %[pX],    %[pX],      8          \n\t"
+    "addiu  %[pY],    %[pY],      8          \n\t"
+    "madd   %[i],     %[temp0]               \n\t"
+    "madd   %[i],     %[temp2]               \n\t"
+    "bne    %[pX],    %[LoopEnd], 1b         \n\t"
+  "2:                                        \n\t"
+    "mfhi   %[temp0]                         \n\t"
+    "mflo   %[temp1]                         \n\t"
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
+      [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
+      [i]"=&r"(i), [pX]"+r"(pX), [pY]"+r"(pY)
+    : [LoopEnd]"r"(LoopEnd)
+    : "memory", "hi", "lo"
+  );
+
+  return (double)((int64_t)temp0 << 32 | temp1);
+}
+
+#define HUFFMAN_COST_PASS                                 \
+  __asm__ volatile(                                       \
+    "sll   %[temp1],  %[temp0],    3           \n\t"      \
+    "addiu %[temp3],  %[streak],   -3          \n\t"      \
+    "addu  %[temp2],  %[pstreaks], %[temp1]    \n\t"      \
+    "blez  %[temp3],  1f                       \n\t"      \
+    "srl   %[temp1],  %[temp1],    1           \n\t"      \
+    "addu  %[temp3],  %[pcnts],    %[temp1]    \n\t"      \
+    "lw    %[temp0],  4(%[temp2])              \n\t"      \
+    "lw    %[temp1],  0(%[temp3])              \n\t"      \
+    "addu  %[temp0],  %[temp0],    %[streak]   \n\t"      \
+    "addiu %[temp1],  %[temp1],    1           \n\t"      \
+    "sw    %[temp0],  4(%[temp2])              \n\t"      \
+    "sw    %[temp1],  0(%[temp3])              \n\t"      \
+    "b     2f                                  \n\t"      \
+  "1:                                          \n\t"      \
+    "lw    %[temp0],  0(%[temp2])              \n\t"      \
+    "addu  %[temp0],  %[temp0],    %[streak]   \n\t"      \
+    "sw    %[temp0],  0(%[temp2])              \n\t"      \
+  "2:                                          \n\t"      \
+    : [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),           \
+      [temp3]"=&r"(temp3), [temp0]"+r"(temp0)             \
+    : [pstreaks]"r"(pstreaks), [pcnts]"r"(pcnts),         \
+      [streak]"r"(streak)                                 \
+    : "memory"                                            \
+  );
+
+// Returns the various RLE counts
+static VP8LStreaks HuffmanCostCount(const uint32_t* population, int length) {
+  int i;
+  int streak = 0;
+  VP8LStreaks stats;
+  int* const pstreaks = &stats.streaks[0][0];
+  int* const pcnts = &stats.counts[0];
+  int temp0, temp1, temp2, temp3;
+  memset(&stats, 0, sizeof(stats));
+  for (i = 0; i < length - 1; ++i) {
+    ++streak;
+    if (population[i] == population[i + 1]) {
+      continue;
+    }
+    temp0 = (population[i] != 0);
+    HUFFMAN_COST_PASS
+    streak = 0;
+  }
+  ++streak;
+  temp0 = (population[i] != 0);
+  HUFFMAN_COST_PASS
+
+  return stats;
+}
+
+static VP8LStreaks HuffmanCostCombinedCount(const uint32_t* X,
+                                            const uint32_t* Y, int length) {
+  int i;
+  int streak = 0;
+  VP8LStreaks stats;
+  int* const pstreaks = &stats.streaks[0][0];
+  int* const pcnts = &stats.counts[0];
+  int temp0, temp1, temp2, temp3;
+  memset(&stats, 0, sizeof(stats));
+  for (i = 0; i < length - 1; ++i) {
+    const uint32_t xy = X[i] + Y[i];
+    const uint32_t xy_next = X[i + 1] + Y[i + 1];
+    ++streak;
+    if (xy == xy_next) {
+      continue;
+    }
+    temp0 = (xy != 0);
+    HUFFMAN_COST_PASS
+    streak = 0;
+  }
+  {
+    const uint32_t xy = X[i] + Y[i];
+    ++streak;
+    temp0 = (xy != 0);
+    HUFFMAN_COST_PASS
+  }
+
+  return stats;
+}
+
+#define ASM_START                                       \
+  __asm__ volatile(                                     \
+    ".set   push                            \n\t"       \
+    ".set   at                              \n\t"       \
+    ".set   macro                           \n\t"       \
+  "1:                                       \n\t"
+
+// P2 = P0 + P1
+// A..D - offsets
+// E - temp variable to tell macro
+//     if pointer should be incremented
+// literal_ and successive histograms could be unaligned
+// so we must use ulw and usw
+#define ADD_TO_OUT(A, B, C, D, E, P0, P1, P2)           \
+    "ulw    %[temp0], "#A"(%["#P0"])        \n\t"       \
+    "ulw    %[temp1], "#B"(%["#P0"])        \n\t"       \
+    "ulw    %[temp2], "#C"(%["#P0"])        \n\t"       \
+    "ulw    %[temp3], "#D"(%["#P0"])        \n\t"       \
+    "ulw    %[temp4], "#A"(%["#P1"])        \n\t"       \
+    "ulw    %[temp5], "#B"(%["#P1"])        \n\t"       \
+    "ulw    %[temp6], "#C"(%["#P1"])        \n\t"       \
+    "ulw    %[temp7], "#D"(%["#P1"])        \n\t"       \
+    "addu   %[temp4], %[temp4],   %[temp0]  \n\t"       \
+    "addu   %[temp5], %[temp5],   %[temp1]  \n\t"       \
+    "addu   %[temp6], %[temp6],   %[temp2]  \n\t"       \
+    "addu   %[temp7], %[temp7],   %[temp3]  \n\t"       \
+    "addiu  %["#P0"],  %["#P0"],  16        \n\t"       \
+  ".if "#E" == 1                            \n\t"       \
+    "addiu  %["#P1"],  %["#P1"],  16        \n\t"       \
+  ".endif                                   \n\t"       \
+    "usw    %[temp4], "#A"(%["#P2"])        \n\t"       \
+    "usw    %[temp5], "#B"(%["#P2"])        \n\t"       \
+    "usw    %[temp6], "#C"(%["#P2"])        \n\t"       \
+    "usw    %[temp7], "#D"(%["#P2"])        \n\t"       \
+    "addiu  %["#P2"], %["#P2"],   16        \n\t"       \
+    "bne    %["#P0"], %[LoopEnd], 1b        \n\t"       \
+    ".set   pop                             \n\t"       \
+
+#define ASM_END_COMMON_0                                \
+    : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),         \
+      [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),         \
+      [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),         \
+      [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),         \
+      [pa]"+r"(pa), [pout]"+r"(pout)
+
+#define ASM_END_COMMON_1                                \
+    : [LoopEnd]"r"(LoopEnd)                             \
+    : "memory", "at"                                    \
+  );
+
+#define ASM_END_0                                       \
+    ASM_END_COMMON_0                                    \
+      , [pb]"+r"(pb)                                    \
+    ASM_END_COMMON_1
+
+#define ASM_END_1                                       \
+    ASM_END_COMMON_0                                    \
+    ASM_END_COMMON_1
+
+#define ADD_VECTOR(A, B, OUT, SIZE, EXTRA_SIZE)  do {   \
+  const uint32_t* pa = (const uint32_t*)(A);            \
+  const uint32_t* pb = (const uint32_t*)(B);            \
+  uint32_t* pout = (uint32_t*)(OUT);                    \
+  const uint32_t* const LoopEnd = pa + (SIZE);          \
+  assert((SIZE) % 4 == 0);                              \
+  ASM_START                                             \
+  ADD_TO_OUT(0, 4, 8, 12, 1, pa, pb, pout)              \
+  ASM_END_0                                             \
+  if ((EXTRA_SIZE) > 0) {                               \
+    const int last = (EXTRA_SIZE);                      \
+    int i;                                              \
+    for (i = 0; i < last; ++i) pout[i] = pa[i] + pb[i]; \
+  }                                                     \
+} while (0)
+
+#define ADD_VECTOR_EQ(A, OUT, SIZE, EXTRA_SIZE)  do {   \
+  const uint32_t* pa = (const uint32_t*)(A);            \
+  uint32_t* pout = (uint32_t*)(OUT);                    \
+  const uint32_t* const LoopEnd = pa + (SIZE);          \
+  assert((SIZE) % 4 == 0);                              \
+  ASM_START                                             \
+  ADD_TO_OUT(0, 4, 8, 12, 0, pa, pout, pout)            \
+  ASM_END_1                                             \
+  if ((EXTRA_SIZE) > 0) {                               \
+    const int last = (EXTRA_SIZE);                      \
+    int i;                                              \
+    for (i = 0; i < last; ++i) pout[i] += pa[i];        \
+  }                                                     \
+} while (0)
+
+static void HistogramAdd(const VP8LHistogram* const a,
+                         const VP8LHistogram* const b,
+                         VP8LHistogram* const out) {
+  uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
+  const int extra_cache_size = VP8LHistogramNumCodes(a->palette_code_bits_)
+                             - (NUM_LITERAL_CODES + NUM_LENGTH_CODES);
+  assert(a->palette_code_bits_ == b->palette_code_bits_);
+
+  if (b != out) {
+    ADD_VECTOR(a->literal_, b->literal_, out->literal_,
+               NUM_LITERAL_CODES + NUM_LENGTH_CODES, extra_cache_size);
+    ADD_VECTOR(a->distance_, b->distance_, out->distance_,
+               NUM_DISTANCE_CODES, 0);
+    ADD_VECTOR(a->red_, b->red_, out->red_, NUM_LITERAL_CODES, 0);
+    ADD_VECTOR(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES, 0);
+    ADD_VECTOR(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES, 0);
+  } else {
+    ADD_VECTOR_EQ(a->literal_, out->literal_,
+                  NUM_LITERAL_CODES + NUM_LENGTH_CODES, extra_cache_size);
+    ADD_VECTOR_EQ(a->distance_, out->distance_, NUM_DISTANCE_CODES, 0);
+    ADD_VECTOR_EQ(a->red_, out->red_, NUM_LITERAL_CODES, 0);
+    ADD_VECTOR_EQ(a->blue_, out->blue_, NUM_LITERAL_CODES, 0);
+    ADD_VECTOR_EQ(a->alpha_, out->alpha_, NUM_LITERAL_CODES, 0);
+  }
+}
+
+#undef ADD_VECTOR_EQ
+#undef ADD_VECTOR
+#undef ASM_END_1
+#undef ASM_END_0
+#undef ASM_END_COMMON_1
+#undef ASM_END_COMMON_0
+#undef ADD_TO_OUT
+#undef ASM_START
+
+#endif  // WEBP_USE_MIPS32
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void VP8LDspInitMIPS32(void);
+
+void VP8LDspInitMIPS32(void) {
+#if defined(WEBP_USE_MIPS32)
+  VP8LFastSLog2Slow = FastSLog2Slow;
+  VP8LFastLog2Slow = FastLog2Slow;
+  VP8LExtraCost = ExtraCost;
+  VP8LExtraCostCombined = ExtraCostCombined;
+  VP8LHuffmanCostCount = HuffmanCostCount;
+  VP8LHuffmanCostCombinedCount = HuffmanCostCombinedCount;
+  VP8LHistogramAdd = HistogramAdd;
+#endif  // WEBP_USE_MIPS32
+}

src/dsp/lossless_neon.c

@@ -0,0 +1,332 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// NEON variant of methods for lossless decoder
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_NEON)
+
+#include <arm_neon.h>
+
+#include "./lossless.h"
+#include "./neon.h"
+
+//------------------------------------------------------------------------------
+// Colorspace conversion functions
+
+#if !defined(WORK_AROUND_GCC)
+// gcc 4.6.0 had some trouble (NDK-r9) with this code. We only use it for
+// gcc-4.8.x at least.
+static void ConvertBGRAToRGBA(const uint32_t* src,
+                              int num_pixels, uint8_t* dst) {
+  const uint32_t* const end = src + (num_pixels & ~15);
+  for (; src < end; src += 16) {
+    uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
+    // swap B and R. (VSWP d0,d2 has no intrinsics equivalent!)
+    const uint8x16_t tmp = pixel.val[0];
+    pixel.val[0] = pixel.val[2];
+    pixel.val[2] = tmp;
+    vst4q_u8(dst, pixel);
+    dst += 64;
+  }
+  VP8LConvertBGRAToRGBA_C(src, num_pixels & 15, dst);  // left-overs
+}
+
+static void ConvertBGRAToBGR(const uint32_t* src,
+                             int num_pixels, uint8_t* dst) {
+  const uint32_t* const end = src + (num_pixels & ~15);
+  for (; src < end; src += 16) {
+    const uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
+    const uint8x16x3_t tmp = { { pixel.val[0], pixel.val[1], pixel.val[2] } };
+    vst3q_u8(dst, tmp);
+    dst += 48;
+  }
+  VP8LConvertBGRAToBGR_C(src, num_pixels & 15, dst);  // left-overs
+}
+
+static void ConvertBGRAToRGB(const uint32_t* src,
+                             int num_pixels, uint8_t* dst) {
+  const uint32_t* const end = src + (num_pixels & ~15);
+  for (; src < end; src += 16) {
+    const uint8x16x4_t pixel = vld4q_u8((uint8_t*)src);
+    const uint8x16x3_t tmp = { { pixel.val[2], pixel.val[1], pixel.val[0] } };
+    vst3q_u8(dst, tmp);
+    dst += 48;
+  }
+  VP8LConvertBGRAToRGB_C(src, num_pixels & 15, dst);  // left-overs
+}
+
+#else  // WORK_AROUND_GCC
+
+// gcc-4.6.0 fallback
+
+static const uint8_t kRGBAShuffle[8] = { 2, 1, 0, 3, 6, 5, 4, 7 };
+
+static void ConvertBGRAToRGBA(const uint32_t* src,
+                              int num_pixels, uint8_t* dst) {
+  const uint32_t* const end = src + (num_pixels & ~1);
+  const uint8x8_t shuffle = vld1_u8(kRGBAShuffle);
+  for (; src < end; src += 2) {
+    const uint8x8_t pixels = vld1_u8((uint8_t*)src);
+    vst1_u8(dst, vtbl1_u8(pixels, shuffle));
+    dst += 8;
+  }
+  VP8LConvertBGRAToRGBA_C(src, num_pixels & 1, dst);  // left-overs
+}
+
+static const uint8_t kBGRShuffle[3][8] = {
+  {  0,  1,  2,  4,  5,  6,  8,  9 },
+  { 10, 12, 13, 14, 16, 17, 18, 20 },
+  { 21, 22, 24, 25, 26, 28, 29, 30 }
+};
+
+static void ConvertBGRAToBGR(const uint32_t* src,
+                             int num_pixels, uint8_t* dst) {
+  const uint32_t* const end = src + (num_pixels & ~7);
+  const uint8x8_t shuffle0 = vld1_u8(kBGRShuffle[0]);
+  const uint8x8_t shuffle1 = vld1_u8(kBGRShuffle[1]);
+  const uint8x8_t shuffle2 = vld1_u8(kBGRShuffle[2]);
+  for (; src < end; src += 8) {
+    uint8x8x4_t pixels;
+    INIT_VECTOR4(pixels,
+                 vld1_u8((const uint8_t*)(src + 0)),
+                 vld1_u8((const uint8_t*)(src + 2)),
+                 vld1_u8((const uint8_t*)(src + 4)),
+                 vld1_u8((const uint8_t*)(src + 6)));
+    vst1_u8(dst +  0, vtbl4_u8(pixels, shuffle0));
+    vst1_u8(dst +  8, vtbl4_u8(pixels, shuffle1));
+    vst1_u8(dst + 16, vtbl4_u8(pixels, shuffle2));
+    dst += 8 * 3;
+  }
+  VP8LConvertBGRAToBGR_C(src, num_pixels & 7, dst);  // left-overs
+}
+
+static const uint8_t kRGBShuffle[3][8] = {
+  {  2,  1,  0,  6,  5,  4, 10,  9 },
+  {  8, 14, 13, 12, 18, 17, 16, 22 },
+  { 21, 20, 26, 25, 24, 30, 29, 28 }
+};
+
+static void ConvertBGRAToRGB(const uint32_t* src,
+                             int num_pixels, uint8_t* dst) {
+  const uint32_t* const end = src + (num_pixels & ~7);
+  const uint8x8_t shuffle0 = vld1_u8(kRGBShuffle[0]);
+  const uint8x8_t shuffle1 = vld1_u8(kRGBShuffle[1]);
+  const uint8x8_t shuffle2 = vld1_u8(kRGBShuffle[2]);
+  for (; src < end; src += 8) {
+    uint8x8x4_t pixels;
+    INIT_VECTOR4(pixels,
+                 vld1_u8((const uint8_t*)(src + 0)),
+                 vld1_u8((const uint8_t*)(src + 2)),
+                 vld1_u8((const uint8_t*)(src + 4)),
+                 vld1_u8((const uint8_t*)(src + 6)));
+    vst1_u8(dst +  0, vtbl4_u8(pixels, shuffle0));
+    vst1_u8(dst +  8, vtbl4_u8(pixels, shuffle1));
+    vst1_u8(dst + 16, vtbl4_u8(pixels, shuffle2));
+    dst += 8 * 3;
+  }
+  VP8LConvertBGRAToRGB_C(src, num_pixels & 7, dst);  // left-overs
+}
+
+#endif   // !WORK_AROUND_GCC
+
+//------------------------------------------------------------------------------
+
+#ifdef USE_INTRINSICS
+
+static WEBP_INLINE uint32_t Average2(const uint32_t* const a,
+                                     const uint32_t* const b) {
+  const uint8x8_t a0 = vreinterpret_u8_u64(vcreate_u64(*a));
+  const uint8x8_t b0 = vreinterpret_u8_u64(vcreate_u64(*b));
+  const uint8x8_t avg = vhadd_u8(a0, b0);
+  return vget_lane_u32(vreinterpret_u32_u8(avg), 0);
+}
+
+static WEBP_INLINE uint32_t Average3(const uint32_t* const a,
+                                     const uint32_t* const b,
+                                     const uint32_t* const c) {
+  const uint8x8_t a0 = vreinterpret_u8_u64(vcreate_u64(*a));
+  const uint8x8_t b0 = vreinterpret_u8_u64(vcreate_u64(*b));
+  const uint8x8_t c0 = vreinterpret_u8_u64(vcreate_u64(*c));
+  const uint8x8_t avg1 = vhadd_u8(a0, c0);
+  const uint8x8_t avg2 = vhadd_u8(avg1, b0);
+  return vget_lane_u32(vreinterpret_u32_u8(avg2), 0);
+}
+
+static WEBP_INLINE uint32_t Average4(const uint32_t* const a,
+                                     const uint32_t* const b,
+                                     const uint32_t* const c,
+                                     const uint32_t* const d) {
+  const uint8x8_t a0 = vreinterpret_u8_u64(vcreate_u64(*a));
+  const uint8x8_t b0 = vreinterpret_u8_u64(vcreate_u64(*b));
+  const uint8x8_t c0 = vreinterpret_u8_u64(vcreate_u64(*c));
+  const uint8x8_t d0 = vreinterpret_u8_u64(vcreate_u64(*d));
+  const uint8x8_t avg1 = vhadd_u8(a0, b0);
+  const uint8x8_t avg2 = vhadd_u8(c0, d0);
+  const uint8x8_t avg3 = vhadd_u8(avg1, avg2);
+  return vget_lane_u32(vreinterpret_u32_u8(avg3), 0);
+}
+
+static uint32_t Predictor5(uint32_t left, const uint32_t* const top) {
+  return Average3(&left, top + 0, top + 1);
+}
+
+static uint32_t Predictor6(uint32_t left, const uint32_t* const top) {
+  return Average2(&left, top - 1);
+}
+
+static uint32_t Predictor7(uint32_t left, const uint32_t* const top) {
+  return Average2(&left, top + 0);
+}
+
+static uint32_t Predictor8(uint32_t left, const uint32_t* const top) {
+  (void)left;
+  return Average2(top - 1, top + 0);
+}
+
+static uint32_t Predictor9(uint32_t left, const uint32_t* const top) {
+  (void)left;
+  return Average2(top + 0, top + 1);
+}
+
+static uint32_t Predictor10(uint32_t left, const uint32_t* const top) {
+  return Average4(&left, top - 1, top + 0, top + 1);
+}
+
+//------------------------------------------------------------------------------
+
+static WEBP_INLINE uint32_t Select(const uint32_t* const c0,
+                                   const uint32_t* const c1,
+                                   const uint32_t* const c2) {
+  const uint8x8_t p0 = vreinterpret_u8_u64(vcreate_u64(*c0));
+  const uint8x8_t p1 = vreinterpret_u8_u64(vcreate_u64(*c1));
+  const uint8x8_t p2 = vreinterpret_u8_u64(vcreate_u64(*c2));
+  const uint8x8_t bc = vabd_u8(p1, p2);   // |b-c|
+  const uint8x8_t ac = vabd_u8(p0, p2);   // |a-c|
+  const int16x4_t sum_bc = vreinterpret_s16_u16(vpaddl_u8(bc));
+  const int16x4_t sum_ac = vreinterpret_s16_u16(vpaddl_u8(ac));
+  const int32x2_t diff = vpaddl_s16(vsub_s16(sum_bc, sum_ac));
+  const int32_t pa_minus_pb = vget_lane_s32(diff, 0);
+  return (pa_minus_pb <= 0) ? *c0 : *c1;
+}
+
+static uint32_t Predictor11(uint32_t left, const uint32_t* const top) {
+  return Select(top + 0, &left, top - 1);
+}
+
+static WEBP_INLINE uint32_t ClampedAddSubtractFull(const uint32_t* const c0,
+                                                   const uint32_t* const c1,
+                                                   const uint32_t* const c2) {
+  const uint8x8_t p0 = vreinterpret_u8_u64(vcreate_u64(*c0));
+  const uint8x8_t p1 = vreinterpret_u8_u64(vcreate_u64(*c1));
+  const uint8x8_t p2 = vreinterpret_u8_u64(vcreate_u64(*c2));
+  const uint16x8_t sum0 = vaddl_u8(p0, p1);                // add and widen
+  const uint16x8_t sum1 = vqsubq_u16(sum0, vmovl_u8(p2));  // widen and subtract
+  const uint8x8_t out = vqmovn_u16(sum1);                  // narrow and clamp
+  return vget_lane_u32(vreinterpret_u32_u8(out), 0);
+}
+
+static uint32_t Predictor12(uint32_t left, const uint32_t* const top) {
+  return ClampedAddSubtractFull(&left, top + 0, top - 1);
+}
+
+static WEBP_INLINE uint32_t ClampedAddSubtractHalf(const uint32_t* const c0,
+                                                   const uint32_t* const c1,
+                                                   const uint32_t* const c2) {
+  const uint8x8_t p0 = vreinterpret_u8_u64(vcreate_u64(*c0));
+  const uint8x8_t p1 = vreinterpret_u8_u64(vcreate_u64(*c1));
+  const uint8x8_t p2 = vreinterpret_u8_u64(vcreate_u64(*c2));
+  const uint8x8_t avg = vhadd_u8(p0, p1);                  // Average(c0,c1)
+  const uint8x8_t ab = vshr_n_u8(vqsub_u8(avg, p2), 1);    // (a-b)>>1 saturated
+  const uint8x8_t ba = vshr_n_u8(vqsub_u8(p2, avg), 1);    // (b-a)>>1 saturated
+  const uint8x8_t out = vqsub_u8(vqadd_u8(avg, ab), ba);
+  return vget_lane_u32(vreinterpret_u32_u8(out), 0);
+}
+
+static uint32_t Predictor13(uint32_t left, const uint32_t* const top) {
+  return ClampedAddSubtractHalf(&left, top + 0, top - 1);
+}
+
+//------------------------------------------------------------------------------
+// Subtract-Green Transform
+
+// vtbl? are unavailable in iOS/arm64 builds.
+#if !defined(__aarch64__)
+
+// 255 = byte will be zero'd
+static const uint8_t kGreenShuffle[8] = { 1, 255, 1, 255, 5, 255, 5, 255  };
+
+static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
+  const uint32_t* const end = argb_data + (num_pixels & ~3);
+  const uint8x8_t shuffle = vld1_u8(kGreenShuffle);
+  for (; argb_data < end; argb_data += 4) {
+    const uint8x16_t argb = vld1q_u8((uint8_t*)argb_data);
+    const uint8x16_t greens =
+        vcombine_u8(vtbl1_u8(vget_low_u8(argb), shuffle),
+                    vtbl1_u8(vget_high_u8(argb), shuffle));
+    vst1q_u8((uint8_t*)argb_data, vsubq_u8(argb, greens));
+  }
+  // fallthrough and finish off with plain-C
+  VP8LSubtractGreenFromBlueAndRed_C(argb_data, num_pixels & 3);
+}
+
+static void AddGreenToBlueAndRed(uint32_t* argb_data, int num_pixels) {
+  const uint32_t* const end = argb_data + (num_pixels & ~3);
+  const uint8x8_t shuffle = vld1_u8(kGreenShuffle);
+  for (; argb_data < end; argb_data += 4) {
+    const uint8x16_t argb = vld1q_u8((uint8_t*)argb_data);
+    const uint8x16_t greens =
+        vcombine_u8(vtbl1_u8(vget_low_u8(argb), shuffle),
+                    vtbl1_u8(vget_high_u8(argb), shuffle));
+    vst1q_u8((uint8_t*)argb_data, vaddq_u8(argb, greens));
+  }
+  // fallthrough and finish off with plain-C
+  VP8LAddGreenToBlueAndRed_C(argb_data, num_pixels & 3);
+}
+
+#endif   // !__aarch64__
+
+#endif   // USE_INTRINSICS
+
+#endif   // WEBP_USE_NEON
+
+//------------------------------------------------------------------------------
+
+extern void VP8LDspInitNEON(void);
+
+void VP8LDspInitNEON(void) {
+#if defined(WEBP_USE_NEON)
+  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
+  VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
+  VP8LConvertBGRAToRGB = ConvertBGRAToRGB;
+
+#ifdef USE_INTRINSICS
+  VP8LPredictors[5] = Predictor5;
+  VP8LPredictors[6] = Predictor6;
+  VP8LPredictors[7] = Predictor7;
+  VP8LPredictors[8] = Predictor8;
+  VP8LPredictors[9] = Predictor9;
+  VP8LPredictors[10] = Predictor10;
+  VP8LPredictors[11] = Predictor11;
+  VP8LPredictors[12] = Predictor12;
+  VP8LPredictors[13] = Predictor13;
+
+#if !defined(__aarch64__)
+  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed;
+  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
+#endif
+#endif
+
+#endif   // WEBP_USE_NEON
+}
+
+//------------------------------------------------------------------------------

src/dsp/lossless_sse2.c

@@ -0,0 +1,535 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// SSE2 variant of methods for lossless decoder
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./dsp.h"
+
+#include <assert.h>
+
+#if defined(WEBP_USE_SSE2)
+#include <emmintrin.h>
+#include "./lossless.h"
+
+//------------------------------------------------------------------------------
+// Predictor Transform
+
+static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
+                                                   uint32_t c2) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
+  const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
+  const __m128i C2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero);
+  const __m128i V1 = _mm_add_epi16(C0, C1);
+  const __m128i V2 = _mm_sub_epi16(V1, C2);
+  const __m128i b = _mm_packus_epi16(V2, V2);
+  const uint32_t output = _mm_cvtsi128_si32(b);
+  return output;
+}
+
+static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
+                                                   uint32_t c2) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
+  const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
+  const __m128i B0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero);
+  const __m128i avg = _mm_add_epi16(C1, C0);
+  const __m128i A0 = _mm_srli_epi16(avg, 1);
+  const __m128i A1 = _mm_sub_epi16(A0, B0);
+  const __m128i BgtA = _mm_cmpgt_epi16(B0, A0);
+  const __m128i A2 = _mm_sub_epi16(A1, BgtA);
+  const __m128i A3 = _mm_srai_epi16(A2, 1);
+  const __m128i A4 = _mm_add_epi16(A0, A3);
+  const __m128i A5 = _mm_packus_epi16(A4, A4);
+  const uint32_t output = _mm_cvtsi128_si32(A5);
+  return output;
+}
+
+static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
+  int pa_minus_pb;
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i A0 = _mm_cvtsi32_si128(a);
+  const __m128i B0 = _mm_cvtsi32_si128(b);
+  const __m128i C0 = _mm_cvtsi32_si128(c);
+  const __m128i AC0 = _mm_subs_epu8(A0, C0);
+  const __m128i CA0 = _mm_subs_epu8(C0, A0);
+  const __m128i BC0 = _mm_subs_epu8(B0, C0);
+  const __m128i CB0 = _mm_subs_epu8(C0, B0);
+  const __m128i AC = _mm_or_si128(AC0, CA0);
+  const __m128i BC = _mm_or_si128(BC0, CB0);
+  const __m128i pa = _mm_unpacklo_epi8(AC, zero);  // |a - c|
+  const __m128i pb = _mm_unpacklo_epi8(BC, zero);  // |b - c|
+  const __m128i diff = _mm_sub_epi16(pb, pa);
+  {
+    int16_t out[8];
+    _mm_storeu_si128((__m128i*)out, diff);
+    pa_minus_pb = out[0] + out[1] + out[2] + out[3];
+  }
+  return (pa_minus_pb <= 0) ? a : b;
+}
+
+static WEBP_INLINE __m128i Average2_128i(uint32_t a0, uint32_t a1) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a0), zero);
+  const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
+  const __m128i sum = _mm_add_epi16(A1, A0);
+  const __m128i avg = _mm_srli_epi16(sum, 1);
+  return avg;
+}
+
+static WEBP_INLINE uint32_t Average2(uint32_t a0, uint32_t a1) {
+  const __m128i avg = Average2_128i(a0, a1);
+  const __m128i A2 = _mm_packus_epi16(avg, avg);
+  const uint32_t output = _mm_cvtsi128_si32(A2);
+  return output;
+}
+
+static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i avg1 = Average2_128i(a0, a2);
+  const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
+  const __m128i sum = _mm_add_epi16(avg1, A1);
+  const __m128i avg2 = _mm_srli_epi16(sum, 1);
+  const __m128i A2 = _mm_packus_epi16(avg2, avg2);
+  const uint32_t output = _mm_cvtsi128_si32(A2);
+  return output;
+}
+
+static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
+                                     uint32_t a2, uint32_t a3) {
+  const __m128i avg1 = Average2_128i(a0, a1);
+  const __m128i avg2 = Average2_128i(a2, a3);
+  const __m128i sum = _mm_add_epi16(avg2, avg1);
+  const __m128i avg3 = _mm_srli_epi16(sum, 1);
+  const __m128i A0 = _mm_packus_epi16(avg3, avg3);
+  const uint32_t output = _mm_cvtsi128_si32(A0);
+  return output;
+}
+
+static uint32_t Predictor5(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = Average3(left, top[0], top[1]);
+  return pred;
+}
+static uint32_t Predictor6(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = Average2(left, top[-1]);
+  return pred;
+}
+static uint32_t Predictor7(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = Average2(left, top[0]);
+  return pred;
+}
+static uint32_t Predictor8(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = Average2(top[-1], top[0]);
+  (void)left;
+  return pred;
+}
+static uint32_t Predictor9(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = Average2(top[0], top[1]);
+  (void)left;
+  return pred;
+}
+static uint32_t Predictor10(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = Average4(left, top[-1], top[0], top[1]);
+  return pred;
+}
+static uint32_t Predictor11(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = Select(top[0], left, top[-1]);
+  return pred;
+}
+static uint32_t Predictor12(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]);
+  return pred;
+}
+static uint32_t Predictor13(uint32_t left, const uint32_t* const top) {
+  const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]);
+  return pred;
+}
+
+//------------------------------------------------------------------------------
+// Subtract-Green Transform
+
+static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
+  const __m128i mask = _mm_set1_epi32(0x0000ff00);
+  int i;
+  for (i = 0; i + 4 <= num_pixels; i += 4) {
+    const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
+    const __m128i in_00g0 = _mm_and_si128(in, mask);     // 00g0|00g0|...
+    const __m128i in_0g00 = _mm_slli_epi32(in_00g0, 8);  // 0g00|0g00|...
+    const __m128i in_000g = _mm_srli_epi32(in_00g0, 8);  // 000g|000g|...
+    const __m128i in_0g0g = _mm_or_si128(in_0g00, in_000g);
+    const __m128i out = _mm_sub_epi8(in, in_0g0g);
+    _mm_storeu_si128((__m128i*)&argb_data[i], out);
+  }
+  // fallthrough and finish off with plain-C
+  VP8LSubtractGreenFromBlueAndRed_C(argb_data + i, num_pixels - i);
+}
+
+static void AddGreenToBlueAndRed(uint32_t* argb_data, int num_pixels) {
+  const __m128i mask = _mm_set1_epi32(0x0000ff00);
+  int i;
+  for (i = 0; i + 4 <= num_pixels; i += 4) {
+    const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
+    const __m128i in_00g0 = _mm_and_si128(in, mask);     // 00g0|00g0|...
+    const __m128i in_0g00 = _mm_slli_epi32(in_00g0, 8);  // 0g00|0g00|...
+    const __m128i in_000g = _mm_srli_epi32(in_00g0, 8);  // 000g|000g|...
+    const __m128i in_0g0g = _mm_or_si128(in_0g00, in_000g);
+    const __m128i out = _mm_add_epi8(in, in_0g0g);
+    _mm_storeu_si128((__m128i*)&argb_data[i], out);
+  }
+  // fallthrough and finish off with plain-C
+  VP8LAddGreenToBlueAndRed_C(argb_data + i, num_pixels - i);
+}
+
+//------------------------------------------------------------------------------
+// Color Transform
+
+static WEBP_INLINE __m128i ColorTransformDelta(__m128i color_pred,
+                                               __m128i color) {
+  // We simulate signed 8-bit multiplication as:
+  // * Left shift the two (8-bit) numbers by 8 bits,
+  // * Perform a 16-bit signed multiplication and retain the higher 16-bits.
+  const __m128i color_pred_shifted = _mm_slli_epi32(color_pred, 8);
+  const __m128i color_shifted = _mm_slli_epi32(color, 8);
+  // Note: This performs multiplication on 8 packed 16-bit numbers, 4 of which
+  // happen to be zeroes.
+  const __m128i signed_mult =
+      _mm_mulhi_epi16(color_pred_shifted, color_shifted);
+  return _mm_srli_epi32(signed_mult, 5);
+}
+
+static WEBP_INLINE void TransformColor(const VP8LMultipliers* const m,
+                                       uint32_t* argb_data,
+                                       int num_pixels) {
+  const __m128i g_to_r = _mm_set1_epi32(m->green_to_red_);       // multipliers
+  const __m128i g_to_b = _mm_set1_epi32(m->green_to_blue_);
+  const __m128i r_to_b = _mm_set1_epi32(m->red_to_blue_);
+
+  int i;
+
+  for (i = 0; i + 4 <= num_pixels; i += 4) {
+    const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
+    const __m128i alpha_green_mask = _mm_set1_epi32(0xff00ff00);  // masks
+    const __m128i red_mask = _mm_set1_epi32(0x00ff0000);
+    const __m128i green_mask = _mm_set1_epi32(0x0000ff00);
+    const __m128i lower_8bit_mask  = _mm_set1_epi32(0x000000ff);
+    const __m128i ag = _mm_and_si128(in, alpha_green_mask);      // alpha, green
+    const __m128i r = _mm_srli_epi32(_mm_and_si128(in, red_mask), 16);
+    const __m128i g = _mm_srli_epi32(_mm_and_si128(in, green_mask), 8);
+    const __m128i b = in;
+
+    const __m128i r_delta = ColorTransformDelta(g_to_r, g);      // red
+    const __m128i r_new =
+        _mm_and_si128(_mm_sub_epi32(r, r_delta), lower_8bit_mask);
+    const __m128i r_new_shifted = _mm_slli_epi32(r_new, 16);
+
+    const __m128i b_delta_1 = ColorTransformDelta(g_to_b, g);    // blue
+    const __m128i b_delta_2 = ColorTransformDelta(r_to_b, r);
+    const __m128i b_delta = _mm_add_epi32(b_delta_1, b_delta_2);
+    const __m128i b_new =
+        _mm_and_si128(_mm_sub_epi32(b, b_delta), lower_8bit_mask);
+
+    const __m128i out = _mm_or_si128(_mm_or_si128(ag, r_new_shifted), b_new);
+    _mm_storeu_si128((__m128i*)&argb_data[i], out);
+  }
+
+  // Fall-back to C-version for left-overs.
+  VP8LTransformColor_C(m, argb_data + i, num_pixels - i);
+}
+
+static WEBP_INLINE void TransformColorInverse(const VP8LMultipliers* const m,
+                                              uint32_t* argb_data,
+                                              int num_pixels) {
+  const __m128i g_to_r = _mm_set1_epi32(m->green_to_red_);       // multipliers
+  const __m128i g_to_b = _mm_set1_epi32(m->green_to_blue_);
+  const __m128i r_to_b = _mm_set1_epi32(m->red_to_blue_);
+
+  int i;
+
+  for (i = 0; i + 4 <= num_pixels; i += 4) {
+    const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
+    const __m128i alpha_green_mask = _mm_set1_epi32(0xff00ff00);  // masks
+    const __m128i red_mask = _mm_set1_epi32(0x00ff0000);
+    const __m128i green_mask = _mm_set1_epi32(0x0000ff00);
+    const __m128i lower_8bit_mask  = _mm_set1_epi32(0x000000ff);
+    const __m128i ag = _mm_and_si128(in, alpha_green_mask);      // alpha, green
+    const __m128i r = _mm_srli_epi32(_mm_and_si128(in, red_mask), 16);
+    const __m128i g = _mm_srli_epi32(_mm_and_si128(in, green_mask), 8);
+    const __m128i b = in;
+
+    const __m128i r_delta = ColorTransformDelta(g_to_r, g);      // red
+    const __m128i r_new =
+        _mm_and_si128(_mm_add_epi32(r, r_delta), lower_8bit_mask);
+    const __m128i r_new_shifted = _mm_slli_epi32(r_new, 16);
+
+    const __m128i b_delta_1 = ColorTransformDelta(g_to_b, g);    // blue
+    const __m128i b_delta_2 = ColorTransformDelta(r_to_b, r_new);
+    const __m128i b_delta = _mm_add_epi32(b_delta_1, b_delta_2);
+    const __m128i b_new =
+        _mm_and_si128(_mm_add_epi32(b, b_delta), lower_8bit_mask);
+
+    const __m128i out = _mm_or_si128(_mm_or_si128(ag, r_new_shifted), b_new);
+    _mm_storeu_si128((__m128i*)&argb_data[i], out);
+  }
+
+  // Fall-back to C-version for left-overs.
+  VP8LTransformColorInverse_C(m, argb_data + i, num_pixels - i);
+}
+
+//------------------------------------------------------------------------------
+// Color-space conversion functions
+
+static void ConvertBGRAToRGBA(const uint32_t* src,
+                              int num_pixels, uint8_t* dst) {
+  const __m128i* in = (const __m128i*)src;
+  __m128i* out = (__m128i*)dst;
+  while (num_pixels >= 8) {
+    const __m128i bgra0 = _mm_loadu_si128(in++);     // bgra0|bgra1|bgra2|bgra3
+    const __m128i bgra4 = _mm_loadu_si128(in++);     // bgra4|bgra5|bgra6|bgra7
+    const __m128i v0l = _mm_unpacklo_epi8(bgra0, bgra4);  // b0b4g0g4r0r4a0a4...
+    const __m128i v0h = _mm_unpackhi_epi8(bgra0, bgra4);  // b2b6g2g6r2r6a2a6...
+    const __m128i v1l = _mm_unpacklo_epi8(v0l, v0h);   // b0b2b4b6g0g2g4g6...
+    const __m128i v1h = _mm_unpackhi_epi8(v0l, v0h);   // b1b3b5b7g1g3g5g7...
+    const __m128i v2l = _mm_unpacklo_epi8(v1l, v1h);   // b0...b7 | g0...g7
+    const __m128i v2h = _mm_unpackhi_epi8(v1l, v1h);   // r0...r7 | a0...a7
+    const __m128i ga0 = _mm_unpackhi_epi64(v2l, v2h);  // g0...g7 | a0...a7
+    const __m128i rb0 = _mm_unpacklo_epi64(v2h, v2l);  // r0...r7 | b0...b7
+    const __m128i rg0 = _mm_unpacklo_epi8(rb0, ga0);   // r0g0r1g1 ... r6g6r7g7
+    const __m128i ba0 = _mm_unpackhi_epi8(rb0, ga0);   // b0a0b1a1 ... b6a6b7a7
+    const __m128i rgba0 = _mm_unpacklo_epi16(rg0, ba0);  // rgba0|rgba1...
+    const __m128i rgba4 = _mm_unpackhi_epi16(rg0, ba0);  // rgba4|rgba5...
+    _mm_storeu_si128(out++, rgba0);
+    _mm_storeu_si128(out++, rgba4);
+    num_pixels -= 8;
+  }
+  // left-overs
+  VP8LConvertBGRAToRGBA_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
+}
+
+static void ConvertBGRAToRGBA4444(const uint32_t* src,
+                                  int num_pixels, uint8_t* dst) {
+  const __m128i mask_0x0f = _mm_set1_epi8(0x0f);
+  const __m128i mask_0xf0 = _mm_set1_epi8(0xf0);
+  const __m128i* in = (const __m128i*)src;
+  __m128i* out = (__m128i*)dst;
+  while (num_pixels >= 8) {
+    const __m128i bgra0 = _mm_loadu_si128(in++);     // bgra0|bgra1|bgra2|bgra3
+    const __m128i bgra4 = _mm_loadu_si128(in++);     // bgra4|bgra5|bgra6|bgra7
+    const __m128i v0l = _mm_unpacklo_epi8(bgra0, bgra4);  // b0b4g0g4r0r4a0a4...
+    const __m128i v0h = _mm_unpackhi_epi8(bgra0, bgra4);  // b2b6g2g6r2r6a2a6...
+    const __m128i v1l = _mm_unpacklo_epi8(v0l, v0h);    // b0b2b4b6g0g2g4g6...
+    const __m128i v1h = _mm_unpackhi_epi8(v0l, v0h);    // b1b3b5b7g1g3g5g7...
+    const __m128i v2l = _mm_unpacklo_epi8(v1l, v1h);    // b0...b7 | g0...g7
+    const __m128i v2h = _mm_unpackhi_epi8(v1l, v1h);    // r0...r7 | a0...a7
+    const __m128i ga0 = _mm_unpackhi_epi64(v2l, v2h);   // g0...g7 | a0...a7
+    const __m128i rb0 = _mm_unpacklo_epi64(v2h, v2l);   // r0...r7 | b0...b7
+    const __m128i ga1 = _mm_srli_epi16(ga0, 4);         // g0-|g1-|...|a6-|a7-
+    const __m128i rb1 = _mm_and_si128(rb0, mask_0xf0);  // -r0|-r1|...|-b6|-a7
+    const __m128i ga2 = _mm_and_si128(ga1, mask_0x0f);  // g0-|g1-|...|a6-|a7-
+    const __m128i rgba0 = _mm_or_si128(ga2, rb1);       // rg0..rg7 | ba0..ba7
+    const __m128i rgba1 = _mm_srli_si128(rgba0, 8);     // ba0..ba7 | 0
+#ifdef WEBP_SWAP_16BIT_CSP
+    const __m128i rgba = _mm_unpacklo_epi8(rgba1, rgba0);  // barg0...barg7
+#else
+    const __m128i rgba = _mm_unpacklo_epi8(rgba0, rgba1);  // rgba0...rgba7
+#endif
+    _mm_storeu_si128(out++, rgba);
+    num_pixels -= 8;
+  }
+  // left-overs
+  VP8LConvertBGRAToRGBA4444_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
+}
+
+static void ConvertBGRAToRGB565(const uint32_t* src,
+                                int num_pixels, uint8_t* dst) {
+  const __m128i mask_0xe0 = _mm_set1_epi8(0xe0);
+  const __m128i mask_0xf8 = _mm_set1_epi8(0xf8);
+  const __m128i mask_0x07 = _mm_set1_epi8(0x07);
+  const __m128i* in = (const __m128i*)src;
+  __m128i* out = (__m128i*)dst;
+  while (num_pixels >= 8) {
+    const __m128i bgra0 = _mm_loadu_si128(in++);     // bgra0|bgra1|bgra2|bgra3
+    const __m128i bgra4 = _mm_loadu_si128(in++);     // bgra4|bgra5|bgra6|bgra7
+    const __m128i v0l = _mm_unpacklo_epi8(bgra0, bgra4);  // b0b4g0g4r0r4a0a4...
+    const __m128i v0h = _mm_unpackhi_epi8(bgra0, bgra4);  // b2b6g2g6r2r6a2a6...
+    const __m128i v1l = _mm_unpacklo_epi8(v0l, v0h);      // b0b2b4b6g0g2g4g6...
+    const __m128i v1h = _mm_unpackhi_epi8(v0l, v0h);      // b1b3b5b7g1g3g5g7...
+    const __m128i v2l = _mm_unpacklo_epi8(v1l, v1h);      // b0...b7 | g0...g7
+    const __m128i v2h = _mm_unpackhi_epi8(v1l, v1h);      // r0...r7 | a0...a7
+    const __m128i ga0 = _mm_unpackhi_epi64(v2l, v2h);     // g0...g7 | a0...a7
+    const __m128i rb0 = _mm_unpacklo_epi64(v2h, v2l);     // r0...r7 | b0...b7
+    const __m128i rb1 = _mm_and_si128(rb0, mask_0xf8);    // -r0..-r7|-b0..-b7
+    const __m128i g_lo1 = _mm_srli_epi16(ga0, 5);
+    const __m128i g_lo2 = _mm_and_si128(g_lo1, mask_0x07);  // g0-...g7-|xx (3b)
+    const __m128i g_hi1 = _mm_slli_epi16(ga0, 3);
+    const __m128i g_hi2 = _mm_and_si128(g_hi1, mask_0xe0);  // -g0...-g7|xx (3b)
+    const __m128i b0 = _mm_srli_si128(rb1, 8);              // -b0...-b7|0
+    const __m128i rg1 = _mm_or_si128(rb1, g_lo2);           // gr0...gr7|xx
+    const __m128i b1 = _mm_srli_epi16(b0, 3);
+    const __m128i gb1 = _mm_or_si128(b1, g_hi2);            // bg0...bg7|xx
+#ifdef WEBP_SWAP_16BIT_CSP
+    const __m128i rgba = _mm_unpacklo_epi8(gb1, rg1);     // rggb0...rggb7
+#else
+    const __m128i rgba = _mm_unpacklo_epi8(rg1, gb1);     // bgrb0...bgrb7
+#endif
+    _mm_storeu_si128(out++, rgba);
+    num_pixels -= 8;
+  }
+  // left-overs
+  VP8LConvertBGRAToRGB565_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
+}
+
+static void ConvertBGRAToBGR(const uint32_t* src,
+                             int num_pixels, uint8_t* dst) {
+  const __m128i mask_l = _mm_set_epi32(0, 0x00ffffff, 0, 0x00ffffff);
+  const __m128i mask_h = _mm_set_epi32(0x00ffffff, 0, 0x00ffffff, 0);
+  const __m128i* in = (const __m128i*)src;
+  const uint8_t* const end = dst + num_pixels * 3;
+  // the last storel_epi64 below writes 8 bytes starting at offset 18
+  while (dst + 26 <= end) {
+    const __m128i bgra0 = _mm_loadu_si128(in++);     // bgra0|bgra1|bgra2|bgra3
+    const __m128i bgra4 = _mm_loadu_si128(in++);     // bgra4|bgra5|bgra6|bgra7
+    const __m128i a0l = _mm_and_si128(bgra0, mask_l);   // bgr0|0|bgr0|0
+    const __m128i a4l = _mm_and_si128(bgra4, mask_l);   // bgr0|0|bgr0|0
+    const __m128i a0h = _mm_and_si128(bgra0, mask_h);   // 0|bgr0|0|bgr0
+    const __m128i a4h = _mm_and_si128(bgra4, mask_h);   // 0|bgr0|0|bgr0
+    const __m128i b0h = _mm_srli_epi64(a0h, 8);         // 000b|gr00|000b|gr00
+    const __m128i b4h = _mm_srli_epi64(a4h, 8);         // 000b|gr00|000b|gr00
+    const __m128i c0 = _mm_or_si128(a0l, b0h);          // rgbrgb00|rgbrgb00
+    const __m128i c4 = _mm_or_si128(a4l, b4h);          // rgbrgb00|rgbrgb00
+    const __m128i c2 = _mm_srli_si128(c0, 8);
+    const __m128i c6 = _mm_srli_si128(c4, 8);
+    _mm_storel_epi64((__m128i*)(dst +   0), c0);
+    _mm_storel_epi64((__m128i*)(dst +   6), c2);
+    _mm_storel_epi64((__m128i*)(dst +  12), c4);
+    _mm_storel_epi64((__m128i*)(dst +  18), c6);
+    dst += 24;
+    num_pixels -= 8;
+  }
+  // left-overs
+  VP8LConvertBGRAToBGR_C((const uint32_t*)in, num_pixels, dst);
+}
+
+//------------------------------------------------------------------------------
+
+#define LINE_SIZE 16    // 8 or 16
+static void AddVector(const uint32_t* a, const uint32_t* b, uint32_t* out,
+                      int size) {
+  int i;
+  assert(size % LINE_SIZE == 0);
+  for (i = 0; i < size; i += LINE_SIZE) {
+    const __m128i a0 = _mm_loadu_si128((__m128i*)&a[i +  0]);
+    const __m128i a1 = _mm_loadu_si128((__m128i*)&a[i +  4]);
+#if (LINE_SIZE == 16)
+    const __m128i a2 = _mm_loadu_si128((__m128i*)&a[i +  8]);
+    const __m128i a3 = _mm_loadu_si128((__m128i*)&a[i + 12]);
+#endif
+    const __m128i b0 = _mm_loadu_si128((__m128i*)&b[i +  0]);
+    const __m128i b1 = _mm_loadu_si128((__m128i*)&b[i +  4]);
+#if (LINE_SIZE == 16)
+    const __m128i b2 = _mm_loadu_si128((__m128i*)&b[i +  8]);
+    const __m128i b3 = _mm_loadu_si128((__m128i*)&b[i + 12]);
+#endif
+    _mm_storeu_si128((__m128i*)&out[i +  0], _mm_add_epi32(a0, b0));
+    _mm_storeu_si128((__m128i*)&out[i +  4], _mm_add_epi32(a1, b1));
+#if (LINE_SIZE == 16)
+    _mm_storeu_si128((__m128i*)&out[i +  8], _mm_add_epi32(a2, b2));
+    _mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
+#endif
+  }
+}
+
+static void AddVectorEq(const uint32_t* a, uint32_t* out, int size) {
+  int i;
+  assert(size % LINE_SIZE == 0);
+  for (i = 0; i < size; i += LINE_SIZE) {
+    const __m128i a0 = _mm_loadu_si128((__m128i*)&a[i +  0]);
+    const __m128i a1 = _mm_loadu_si128((__m128i*)&a[i +  4]);
+#if (LINE_SIZE == 16)
+    const __m128i a2 = _mm_loadu_si128((__m128i*)&a[i +  8]);
+    const __m128i a3 = _mm_loadu_si128((__m128i*)&a[i + 12]);
+#endif
+    const __m128i b0 = _mm_loadu_si128((__m128i*)&out[i +  0]);
+    const __m128i b1 = _mm_loadu_si128((__m128i*)&out[i +  4]);
+#if (LINE_SIZE == 16)
+    const __m128i b2 = _mm_loadu_si128((__m128i*)&out[i +  8]);
+    const __m128i b3 = _mm_loadu_si128((__m128i*)&out[i + 12]);
+#endif
+    _mm_storeu_si128((__m128i*)&out[i +  0], _mm_add_epi32(a0, b0));
+    _mm_storeu_si128((__m128i*)&out[i +  4], _mm_add_epi32(a1, b1));
+#if (LINE_SIZE == 16)
+    _mm_storeu_si128((__m128i*)&out[i +  8], _mm_add_epi32(a2, b2));
+    _mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
+#endif
+  }
+}
+#undef LINE_SIZE
+
+// Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
+// that's ok since the histogram values are less than 1<<28 (max picture size).
+static void HistogramAdd(const VP8LHistogram* const a,
+                         const VP8LHistogram* const b,
+                         VP8LHistogram* const out) {
+  int i;
+  const int literal_size = VP8LHistogramNumCodes(a->palette_code_bits_);
+  assert(a->palette_code_bits_ == b->palette_code_bits_);
+  if (b != out) {
+    AddVector(a->literal_, b->literal_, out->literal_, NUM_LITERAL_CODES);
+    AddVector(a->red_, b->red_, out->red_, NUM_LITERAL_CODES);
+    AddVector(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES);
+    AddVector(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES);
+  } else {
+    AddVectorEq(a->literal_, out->literal_, NUM_LITERAL_CODES);
+    AddVectorEq(a->red_, out->red_, NUM_LITERAL_CODES);
+    AddVectorEq(a->blue_, out->blue_, NUM_LITERAL_CODES);
+    AddVectorEq(a->alpha_, out->alpha_, NUM_LITERAL_CODES);
+  }
+  for (i = NUM_LITERAL_CODES; i < literal_size; ++i) {
+    out->literal_[i] = a->literal_[i] + b->literal_[i];
+  }
+  for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
+    out->distance_[i] = a->distance_[i] + b->distance_[i];
+  }
+}
+
+#endif   // WEBP_USE_SSE2
+
+//------------------------------------------------------------------------------
+
+extern void VP8LDspInitSSE2(void);
+
+void VP8LDspInitSSE2(void) {
+#if defined(WEBP_USE_SSE2)
+  VP8LPredictors[5] = Predictor5;
+  VP8LPredictors[6] = Predictor6;
+  VP8LPredictors[7] = Predictor7;
+  VP8LPredictors[8] = Predictor8;
+  VP8LPredictors[9] = Predictor9;
+  VP8LPredictors[10] = Predictor10;
+  VP8LPredictors[11] = Predictor11;
+  VP8LPredictors[12] = Predictor12;
+  VP8LPredictors[13] = Predictor13;
+
+  VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed;
+  VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
+
+  VP8LTransformColor = TransformColor;
+  VP8LTransformColorInverse = TransformColorInverse;
+
+  VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
+  VP8LConvertBGRAToRGBA4444 = ConvertBGRAToRGBA4444;
+  VP8LConvertBGRAToRGB565 = ConvertBGRAToRGB565;
+  VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
+
+  VP8LHistogramAdd = HistogramAdd;
+#endif   // WEBP_USE_SSE2
+}
+
+//------------------------------------------------------------------------------

src/dsp/neon.h

@@ -0,0 +1,82 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//  NEON common code.
+
+#ifndef WEBP_DSP_NEON_H_
+#define WEBP_DSP_NEON_H_
+
+#include <arm_neon.h>
+
+#include "./dsp.h"
+
+// Right now, some intrinsics functions seem slower, so we disable them
+// everywhere except aarch64 where the inline assembly is incompatible.
+#if defined(__aarch64__)
+#define USE_INTRINSICS   // use intrinsics when possible
+#endif
+
+#define INIT_VECTOR2(v, a, b) do {  \
+  v.val[0] = a;                     \
+  v.val[1] = b;                     \
+} while (0)
+
+#define INIT_VECTOR3(v, a, b, c) do {  \
+  v.val[0] = a;                        \
+  v.val[1] = b;                        \
+  v.val[2] = c;                        \
+} while (0)
+
+#define INIT_VECTOR4(v, a, b, c, d) do {  \
+  v.val[0] = a;                           \
+  v.val[1] = b;                           \
+  v.val[2] = c;                           \
+  v.val[3] = d;                           \
+} while (0)
+
+// if using intrinsics, this flag avoids some functions that make gcc-4.6.3
+// crash ("internal compiler error: in immed_double_const, at emit-rtl.").
+// (probably similar to gcc.gnu.org/bugzilla/show_bug.cgi?id=48183)
+#if !(LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
+#define WORK_AROUND_GCC
+#endif
+
+static WEBP_INLINE int32x4x4_t Transpose4x4(const int32x4x4_t rows) {
+  uint64x2x2_t row01, row23;
+
+  row01.val[0] = vreinterpretq_u64_s32(rows.val[0]);
+  row01.val[1] = vreinterpretq_u64_s32(rows.val[1]);
+  row23.val[0] = vreinterpretq_u64_s32(rows.val[2]);
+  row23.val[1] = vreinterpretq_u64_s32(rows.val[3]);
+  // Transpose 64-bit values (there's no vswp equivalent)
+  {
+    const uint64x1_t row0h = vget_high_u64(row01.val[0]);
+    const uint64x1_t row2l = vget_low_u64(row23.val[0]);
+    const uint64x1_t row1h = vget_high_u64(row01.val[1]);
+    const uint64x1_t row3l = vget_low_u64(row23.val[1]);
+    row01.val[0] = vcombine_u64(vget_low_u64(row01.val[0]), row2l);
+    row23.val[0] = vcombine_u64(row0h, vget_high_u64(row23.val[0]));
+    row01.val[1] = vcombine_u64(vget_low_u64(row01.val[1]), row3l);
+    row23.val[1] = vcombine_u64(row1h, vget_high_u64(row23.val[1]));
+  }
+  {
+    const int32x4x2_t out01 = vtrnq_s32(vreinterpretq_s32_u64(row01.val[0]),
+                                        vreinterpretq_s32_u64(row01.val[1]));
+    const int32x4x2_t out23 = vtrnq_s32(vreinterpretq_s32_u64(row23.val[0]),
+                                        vreinterpretq_s32_u64(row23.val[1]));
+    int32x4x4_t out;
+    out.val[0] = out01.val[0];
+    out.val[1] = out01.val[1];
+    out.val[2] = out23.val[0];
+    out.val[3] = out23.val[1];
+    return out;
+  }
+}
+
+#endif  // WEBP_DSP_NEON_H_

src/dsp/upsampling.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // YUV to RGB upsampling functions.
@@ -12,9 +14,7 @@
 #include "./dsp.h"
 #include "./yuv.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
+#include <assert.h>
 
 //------------------------------------------------------------------------------
 // Fancy upsampler
@@ -32,7 +32,7 @@ WebPUpsampleLinePairFunc WebPUpsamplers[MODE_LAST];
 //  ([3*a +   b + 9*c + 3*d      a + 3*b + 3*c + 9*d]   [8 8]) / 16
 
 // We process u and v together stashed into 32bit (16bit each).
-#define LOAD_UV(u,v) ((u) | ((v) << 16))
+#define LOAD_UV(u, v) ((u) | ((v) << 16))
 
 #define UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP)                                  \
 static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
@@ -43,11 +43,12 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
   const int last_pixel_pair = (len - 1) >> 1;                                  \
   uint32_t tl_uv = LOAD_UV(top_u[0], top_v[0]);   /* top-left sample */        \
   uint32_t l_uv  = LOAD_UV(cur_u[0], cur_v[0]);   /* left-sample */            \
-  if (top_y) {                                                                 \
+  assert(top_y != NULL);                                                       \
+  {                                                                            \
     const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2;                \
     FUNC(top_y[0], uv0 & 0xff, (uv0 >> 16), top_dst);                          \
   }                                                                            \
-  if (bottom_y) {                                                              \
+  if (bottom_y != NULL) {                                                      \
     const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2;                \
     FUNC(bottom_y[0], uv0 & 0xff, (uv0 >> 16), bottom_dst);                    \
   }                                                                            \
@@ -58,7 +59,7 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
     const uint32_t avg = tl_uv + t_uv + l_uv + uv + 0x00080008u;               \
     const uint32_t diag_12 = (avg + 2 * (t_uv + l_uv)) >> 3;                   \
     const uint32_t diag_03 = (avg + 2 * (tl_uv + uv)) >> 3;                    \
-    if (top_y) {                                                               \
+    {                                                                          \
       const uint32_t uv0 = (diag_12 + tl_uv) >> 1;                             \
       const uint32_t uv1 = (diag_03 + t_uv) >> 1;                              \
       FUNC(top_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16),                          \
@@ -66,7 +67,7 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
       FUNC(top_y[2 * x - 0], uv1 & 0xff, (uv1 >> 16),                          \
            top_dst + (2 * x - 0) * XSTEP);                                     \
     }                                                                          \
-    if (bottom_y) {                                                            \
+    if (bottom_y != NULL) {                                                    \
       const uint32_t uv0 = (diag_03 + l_uv) >> 1;                              \
       const uint32_t uv1 = (diag_12 + uv) >> 1;                                \
       FUNC(bottom_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16),                       \
@@ -78,12 +79,12 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
     l_uv = uv;                                                                 \
   }                                                                            \
   if (!(len & 1)) {                                                            \
-    if (top_y) {                                                               \
+    {                                                                          \
       const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2;              \
       FUNC(top_y[len - 1], uv0 & 0xff, (uv0 >> 16),                            \
            top_dst + (len - 1) * XSTEP);                                       \
     }                                                                          \
-    if (bottom_y) {                                                            \
+    if (bottom_y != NULL) {                                                    \
       const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2;              \
       FUNC(bottom_y[len - 1], uv0 & 0xff, (uv0 >> 16),                         \
            bottom_dst + (len - 1) * XSTEP);                                    \
@@ -105,57 +106,6 @@ UPSAMPLE_FUNC(UpsampleRgb565LinePair,  VP8YuvToRgb565,  2)
 
 #endif  // FANCY_UPSAMPLING
 
-//------------------------------------------------------------------------------
-// simple point-sampling
-
-#define SAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP)                                    \
-static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
-                      const uint8_t* u, const uint8_t* v,                      \
-                      uint8_t* top_dst, uint8_t* bottom_dst, int len) {        \
-  int i;                                                                       \
-  for (i = 0; i < len - 1; i += 2) {                                           \
-    FUNC(top_y[0], u[0], v[0], top_dst);                                       \
-    FUNC(top_y[1], u[0], v[0], top_dst + XSTEP);                               \
-    FUNC(bottom_y[0], u[0], v[0], bottom_dst);                                 \
-    FUNC(bottom_y[1], u[0], v[0], bottom_dst + XSTEP);                         \
-    top_y += 2;                                                                \
-    bottom_y += 2;                                                             \
-    u++;                                                                       \
-    v++;                                                                       \
-    top_dst += 2 * XSTEP;                                                      \
-    bottom_dst += 2 * XSTEP;                                                   \
-  }                                                                            \
-  if (i == len - 1) {    /* last one */                                        \
-    FUNC(top_y[0], u[0], v[0], top_dst);                                       \
-    FUNC(bottom_y[0], u[0], v[0], bottom_dst);                                 \
-  }                                                                            \
-}
-
-// All variants implemented.
-SAMPLE_FUNC(SampleRgbLinePair,      VP8YuvToRgb,  3)
-SAMPLE_FUNC(SampleBgrLinePair,      VP8YuvToBgr,  3)
-SAMPLE_FUNC(SampleRgbaLinePair,     VP8YuvToRgba, 4)
-SAMPLE_FUNC(SampleBgraLinePair,     VP8YuvToBgra, 4)
-SAMPLE_FUNC(SampleArgbLinePair,     VP8YuvToArgb, 4)
-SAMPLE_FUNC(SampleRgba4444LinePair, VP8YuvToRgba4444, 2)
-SAMPLE_FUNC(SampleRgb565LinePair,   VP8YuvToRgb565, 2)
-
-#undef SAMPLE_FUNC
-
-const WebPSampleLinePairFunc WebPSamplers[MODE_LAST] = {
-  SampleRgbLinePair,       // MODE_RGB
-  SampleRgbaLinePair,      // MODE_RGBA
-  SampleBgrLinePair,       // MODE_BGR
-  SampleBgraLinePair,      // MODE_BGRA
-  SampleArgbLinePair,      // MODE_ARGB
-  SampleRgba4444LinePair,  // MODE_RGBA_4444
-  SampleRgb565LinePair,    // MODE_RGB_565
-  SampleRgbaLinePair,      // MODE_rgbA
-  SampleBgraLinePair,      // MODE_bgrA
-  SampleArgbLinePair,      // MODE_Argb
-  SampleRgba4444LinePair   // MODE_rgbA_4444
-};
-
 //------------------------------------------------------------------------------
 
 #if !defined(FANCY_UPSAMPLING)
@@ -166,7 +116,8 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y,              \
                       uint8_t* top_dst, uint8_t* bot_dst, int len) {           \
   const int half_len = len >> 1;                                               \
   int x;                                                                       \
-  if (top_dst != NULL) {                                                       \
+  assert(top_dst != NULL);                                                     \
+  {                                                                            \
     for (x = 0; x < half_len; ++x) {                                           \
       FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x + 0);         \
       FUNC(top_y[2 * x + 1], top_u[x], top_v[x], top_dst + 8 * x + 4);         \
@@ -233,83 +184,10 @@ const WebPYUV444Converter WebPYUV444Converters[MODE_LAST] = {
 };
 
 //------------------------------------------------------------------------------
-// Premultiplied modes
+// Main calls
 
-// non dithered-modes
-
-// (x * a * 32897) >> 23 is bit-wise equivalent to (int)(x * a / 255.)
-// for all 8bit x or a. For bit-wise equivalence to (int)(x * a / 255. + .5),
-// one can use instead: (x * a * 65793 + (1 << 23)) >> 24
-#if 1     // (int)(x * a / 255.)
-#define MULTIPLIER(a)   ((a) * 32897UL)
-#define PREMULTIPLY(x, m) (((x) * (m)) >> 23)
-#else     // (int)(x * a / 255. + .5)
-#define MULTIPLIER(a) ((a) * 65793UL)
-#define PREMULTIPLY(x, m) (((x) * (m) + (1UL << 23)) >> 24)
-#endif
-
-static void ApplyAlphaMultiply(uint8_t* rgba, int alpha_first,
-                               int w, int h, int stride) {
-  while (h-- > 0) {
-    uint8_t* const rgb = rgba + (alpha_first ? 1 : 0);
-    const uint8_t* const alpha = rgba + (alpha_first ? 0 : 3);
-    int i;
-    for (i = 0; i < w; ++i) {
-      const uint32_t a = alpha[4 * i];
-      if (a != 0xff) {
-        const uint32_t mult = MULTIPLIER(a);
-        rgb[4 * i + 0] = PREMULTIPLY(rgb[4 * i + 0], mult);
-        rgb[4 * i + 1] = PREMULTIPLY(rgb[4 * i + 1], mult);
-        rgb[4 * i + 2] = PREMULTIPLY(rgb[4 * i + 2], mult);
-      }
-    }
-    rgba += stride;
-  }
-}
-#undef MULTIPLIER
-#undef PREMULTIPLY
-
-// rgbA4444
-
-#define MULTIPLIER(a)  ((a) * 0x1111)    // 0x1111 ~= (1 << 16) / 15
-
-static WEBP_INLINE uint8_t dither_hi(uint8_t x) {
-  return (x & 0xf0) | (x >> 4);
-}
-
-static WEBP_INLINE uint8_t dither_lo(uint8_t x) {
-  return (x & 0x0f) | (x << 4);
-}
-
-static WEBP_INLINE uint8_t multiply(uint8_t x, uint32_t m) {
-  return (x * m) >> 16;
-}
-
-static void ApplyAlphaMultiply4444(uint8_t* rgba4444,
-                                   int w, int h, int stride) {
-  while (h-- > 0) {
-    int i;
-    for (i = 0; i < w; ++i) {
-      const uint8_t a = (rgba4444[2 * i + 1] & 0x0f);
-      const uint32_t mult = MULTIPLIER(a);
-      const uint8_t r = multiply(dither_hi(rgba4444[2 * i + 0]), mult);
-      const uint8_t g = multiply(dither_lo(rgba4444[2 * i + 0]), mult);
-      const uint8_t b = multiply(dither_hi(rgba4444[2 * i + 1]), mult);
-      rgba4444[2 * i + 0] = (r & 0xf0) | ((g >> 4) & 0x0f);
-      rgba4444[2 * i + 1] = (b & 0xf0) | a;
-    }
-    rgba4444 += stride;
-  }
-}
-#undef MULTIPLIER
-
-void (*WebPApplyAlphaMultiply)(uint8_t*, int, int, int, int)
-    = ApplyAlphaMultiply;
-void (*WebPApplyAlphaMultiply4444)(uint8_t*, int, int, int)
-    = ApplyAlphaMultiply4444;
-
-//------------------------------------------------------------------------------
-// Main call
+extern void WebPInitUpsamplersSSE2(void);
+extern void WebPInitUpsamplersNEON(void);
 
 void WebPInitUpsamplers(void) {
 #ifdef FANCY_UPSAMPLING
@@ -320,6 +198,10 @@ void WebPInitUpsamplers(void) {
   WebPUpsamplers[MODE_ARGB]      = UpsampleArgbLinePair;
   WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
   WebPUpsamplers[MODE_RGB_565]   = UpsampleRgb565LinePair;
+  WebPUpsamplers[MODE_rgbA]      = UpsampleRgbaLinePair;
+  WebPUpsamplers[MODE_bgrA]      = UpsampleBgraLinePair;
+  WebPUpsamplers[MODE_Argb]      = UpsampleArgbLinePair;
+  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
@@ -328,30 +210,13 @@ void WebPInitUpsamplers(void) {
       WebPInitUpsamplersSSE2();
     }
 #endif
-  }
-#endif  // FANCY_UPSAMPLING
-}
-
-void WebPInitPremultiply(void) {
-  WebPApplyAlphaMultiply = ApplyAlphaMultiply;
-  WebPApplyAlphaMultiply4444 = ApplyAlphaMultiply4444;
-
-#ifdef FANCY_UPSAMPLING
-  WebPUpsamplers[MODE_rgbA]      = UpsampleRgbaLinePair;
-  WebPUpsamplers[MODE_bgrA]      = UpsampleBgraLinePair;
-  WebPUpsamplers[MODE_Argb]      = UpsampleArgbLinePair;
-  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
-
-  if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
-    if (VP8GetCPUInfo(kSSE2)) {
-      WebPInitPremultiplySSE2();
+#if defined(WEBP_USE_NEON)
+    if (VP8GetCPUInfo(kNEON)) {
+      WebPInitUpsamplersNEON();
     }
 #endif
   }
 #endif  // FANCY_UPSAMPLING
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif
+//------------------------------------------------------------------------------

src/dsp/upsampling_neon.c

@@ -0,0 +1,267 @@
+// Copyright 2011 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// NEON version of YUV to RGB upsampling functions.
+//
+// Author: mans@mansr.com (Mans Rullgard)
+// Based on SSE code by: somnath@google.com (Somnath Banerjee)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_NEON)
+
+#include <assert.h>
+#include <arm_neon.h>
+#include <string.h>
+#include "./neon.h"
+#include "./yuv.h"
+
+#ifdef FANCY_UPSAMPLING
+
+//-----------------------------------------------------------------------------
+// U/V upsampling
+
+// Loads 9 pixels each from rows r1 and r2 and generates 16 pixels.
+#define UPSAMPLE_16PIXELS(r1, r2, out) {                                \
+  uint8x8_t a = vld1_u8(r1);                                            \
+  uint8x8_t b = vld1_u8(r1 + 1);                                        \
+  uint8x8_t c = vld1_u8(r2);                                            \
+  uint8x8_t d = vld1_u8(r2 + 1);                                        \
+                                                                        \
+  uint16x8_t al = vshll_n_u8(a, 1);                                     \
+  uint16x8_t bl = vshll_n_u8(b, 1);                                     \
+  uint16x8_t cl = vshll_n_u8(c, 1);                                     \
+  uint16x8_t dl = vshll_n_u8(d, 1);                                     \
+                                                                        \
+  uint8x8_t diag1, diag2;                                               \
+  uint16x8_t sl;                                                        \
+                                                                        \
+  /* a + b + c + d */                                                   \
+  sl = vaddl_u8(a,  b);                                                 \
+  sl = vaddw_u8(sl, c);                                                 \
+  sl = vaddw_u8(sl, d);                                                 \
+                                                                        \
+  al = vaddq_u16(sl, al); /* 3a +  b +  c +  d */                       \
+  bl = vaddq_u16(sl, bl); /*  a + 3b +  c +  d */                       \
+                                                                        \
+  al = vaddq_u16(al, dl); /* 3a +  b +  c + 3d */                       \
+  bl = vaddq_u16(bl, cl); /*  a + 3b + 3c +  d */                       \
+                                                                        \
+  diag2 = vshrn_n_u16(al, 3);                                           \
+  diag1 = vshrn_n_u16(bl, 3);                                           \
+                                                                        \
+  a = vrhadd_u8(a, diag1);                                              \
+  b = vrhadd_u8(b, diag2);                                              \
+  c = vrhadd_u8(c, diag2);                                              \
+  d = vrhadd_u8(d, diag1);                                              \
+                                                                        \
+  {                                                                     \
+    uint8x8x2_t a_b, c_d;                                               \
+    INIT_VECTOR2(a_b, a, b);                                            \
+    INIT_VECTOR2(c_d, c, d);                                            \
+    vst2_u8(out,      a_b);                                             \
+    vst2_u8(out + 32, c_d);                                             \
+  }                                                                     \
+}
+
+// Turn the macro into a function for reducing code-size when non-critical
+static void Upsample16Pixels(const uint8_t *r1, const uint8_t *r2,
+                             uint8_t *out) {
+  UPSAMPLE_16PIXELS(r1, r2, out);
+}
+
+#define UPSAMPLE_LAST_BLOCK(tb, bb, num_pixels, out) {                  \
+  uint8_t r1[9], r2[9];                                                 \
+  memcpy(r1, (tb), (num_pixels));                                       \
+  memcpy(r2, (bb), (num_pixels));                                       \
+  /* replicate last byte */                                             \
+  memset(r1 + (num_pixels), r1[(num_pixels) - 1], 9 - (num_pixels));    \
+  memset(r2 + (num_pixels), r2[(num_pixels) - 1], 9 - (num_pixels));    \
+  Upsample16Pixels(r1, r2, out);                                        \
+}
+
+//-----------------------------------------------------------------------------
+// YUV->RGB conversion
+
+static const int16_t kCoeffs[4] = { kYScale, kVToR, kUToG, kVToG };
+
+#define v255 vdup_n_u8(255)
+
+#define STORE_Rgb(out, r, g, b) do {                                    \
+  uint8x8x3_t r_g_b;                                                    \
+  INIT_VECTOR3(r_g_b, r, g, b);                                         \
+  vst3_u8(out, r_g_b);                                                  \
+} while (0)
+
+#define STORE_Bgr(out, r, g, b) do {                                    \
+  uint8x8x3_t b_g_r;                                                    \
+  INIT_VECTOR3(b_g_r, b, g, r);                                         \
+  vst3_u8(out, b_g_r);                                                  \
+} while (0)
+
+#define STORE_Rgba(out, r, g, b) do {                                   \
+  uint8x8x4_t r_g_b_v255;                                               \
+  INIT_VECTOR4(r_g_b_v255, r, g, b, v255);                              \
+  vst4_u8(out, r_g_b_v255);                                             \
+} while (0)
+
+#define STORE_Bgra(out, r, g, b) do {                                   \
+  uint8x8x4_t b_g_r_v255;                                               \
+  INIT_VECTOR4(b_g_r_v255, b, g, r, v255);                              \
+  vst4_u8(out, b_g_r_v255);                                             \
+} while (0)
+
+#define CONVERT8(FMT, XSTEP, N, src_y, src_uv, out, cur_x) {            \
+  int i;                                                                \
+  for (i = 0; i < N; i += 8) {                                          \
+    const int off = ((cur_x) + i) * XSTEP;                              \
+    uint8x8_t y  = vld1_u8((src_y) + (cur_x)  + i);                     \
+    uint8x8_t u  = vld1_u8((src_uv) + i);                               \
+    uint8x8_t v  = vld1_u8((src_uv) + i + 16);                          \
+    const int16x8_t yy = vreinterpretq_s16_u16(vsubl_u8(y, u16));       \
+    const int16x8_t uu = vreinterpretq_s16_u16(vsubl_u8(u, u128));      \
+    const int16x8_t vv = vreinterpretq_s16_u16(vsubl_u8(v, u128));      \
+    int32x4_t yl = vmull_lane_s16(vget_low_s16(yy),  cf16, 0);          \
+    int32x4_t yh = vmull_lane_s16(vget_high_s16(yy), cf16, 0);          \
+    const int32x4_t rl = vmlal_lane_s16(yl, vget_low_s16(vv),  cf16, 1);\
+    const int32x4_t rh = vmlal_lane_s16(yh, vget_high_s16(vv), cf16, 1);\
+    int32x4_t gl = vmlsl_lane_s16(yl, vget_low_s16(uu),  cf16, 2);      \
+    int32x4_t gh = vmlsl_lane_s16(yh, vget_high_s16(uu), cf16, 2);      \
+    const int32x4_t bl = vmovl_s16(vget_low_s16(uu));                   \
+    const int32x4_t bh = vmovl_s16(vget_high_s16(uu));                  \
+    gl = vmlsl_lane_s16(gl, vget_low_s16(vv),  cf16, 3);                \
+    gh = vmlsl_lane_s16(gh, vget_high_s16(vv), cf16, 3);                \
+    yl = vmlaq_lane_s32(yl, bl, cf32, 0);                               \
+    yh = vmlaq_lane_s32(yh, bh, cf32, 0);                               \
+    /* vrshrn_n_s32() already incorporates the rounding constant */     \
+    y = vqmovun_s16(vcombine_s16(vrshrn_n_s32(rl, YUV_FIX2),            \
+                                 vrshrn_n_s32(rh, YUV_FIX2)));          \
+    u = vqmovun_s16(vcombine_s16(vrshrn_n_s32(gl, YUV_FIX2),            \
+                                 vrshrn_n_s32(gh, YUV_FIX2)));          \
+    v = vqmovun_s16(vcombine_s16(vrshrn_n_s32(yl, YUV_FIX2),            \
+                                 vrshrn_n_s32(yh, YUV_FIX2)));          \
+    STORE_ ## FMT(out + off, y, u, v);                                  \
+  }                                                                     \
+}
+
+#define CONVERT1(FUNC, XSTEP, N, src_y, src_uv, rgb, cur_x) {           \
+  int i;                                                                \
+  for (i = 0; i < N; i++) {                                             \
+    const int off = ((cur_x) + i) * XSTEP;                              \
+    const int y = src_y[(cur_x) + i];                                   \
+    const int u = (src_uv)[i];                                          \
+    const int v = (src_uv)[i + 16];                                     \
+    FUNC(y, u, v, rgb + off);                                           \
+  }                                                                     \
+}
+
+#define CONVERT2RGB_8(FMT, XSTEP, top_y, bottom_y, uv,                  \
+                      top_dst, bottom_dst, cur_x, len) {                \
+  CONVERT8(FMT, XSTEP, len, top_y, uv, top_dst, cur_x)                  \
+  if (bottom_y != NULL) {                                               \
+    CONVERT8(FMT, XSTEP, len, bottom_y, (uv) + 32, bottom_dst, cur_x)   \
+  }                                                                     \
+}
+
+#define CONVERT2RGB_1(FUNC, XSTEP, top_y, bottom_y, uv,                 \
+                      top_dst, bottom_dst, cur_x, len) {                \
+  CONVERT1(FUNC, XSTEP, len, top_y, uv, top_dst, cur_x);                \
+  if (bottom_y != NULL) {                                               \
+    CONVERT1(FUNC, XSTEP, len, bottom_y, (uv) + 32, bottom_dst, cur_x); \
+  }                                                                     \
+}
+
+#define NEON_UPSAMPLE_FUNC(FUNC_NAME, FMT, XSTEP)                       \
+static void FUNC_NAME(const uint8_t *top_y, const uint8_t *bottom_y,    \
+                      const uint8_t *top_u, const uint8_t *top_v,       \
+                      const uint8_t *cur_u, const uint8_t *cur_v,       \
+                      uint8_t *top_dst, uint8_t *bottom_dst, int len) { \
+  int block;                                                            \
+  /* 16 byte aligned array to cache reconstructed u and v */            \
+  uint8_t uv_buf[2 * 32 + 15];                                          \
+  uint8_t *const r_uv = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~15);     \
+  const int uv_len = (len + 1) >> 1;                                    \
+  /* 9 pixels must be read-able for each block */                       \
+  const int num_blocks = (uv_len - 1) >> 3;                             \
+  const int leftover = uv_len - num_blocks * 8;                         \
+  const int last_pos = 1 + 16 * num_blocks;                             \
+                                                                        \
+  const int u_diag = ((top_u[0] + cur_u[0]) >> 1) + 1;                  \
+  const int v_diag = ((top_v[0] + cur_v[0]) >> 1) + 1;                  \
+                                                                        \
+  const int16x4_t cf16 = vld1_s16(kCoeffs);                             \
+  const int32x2_t cf32 = vdup_n_s32(kUToB);                             \
+  const uint8x8_t u16  = vdup_n_u8(16);                                 \
+  const uint8x8_t u128 = vdup_n_u8(128);                                \
+                                                                        \
+  /* Treat the first pixel in regular way */                            \
+  assert(top_y != NULL);                                                \
+  {                                                                     \
+    const int u0 = (top_u[0] + u_diag) >> 1;                            \
+    const int v0 = (top_v[0] + v_diag) >> 1;                            \
+    VP8YuvTo ## FMT(top_y[0], u0, v0, top_dst);                         \
+  }                                                                     \
+  if (bottom_y != NULL) {                                               \
+    const int u0 = (cur_u[0] + u_diag) >> 1;                            \
+    const int v0 = (cur_v[0] + v_diag) >> 1;                            \
+    VP8YuvTo ## FMT(bottom_y[0], u0, v0, bottom_dst);                   \
+  }                                                                     \
+                                                                        \
+  for (block = 0; block < num_blocks; ++block) {                        \
+    UPSAMPLE_16PIXELS(top_u, cur_u, r_uv);                              \
+    UPSAMPLE_16PIXELS(top_v, cur_v, r_uv + 16);                         \
+    CONVERT2RGB_8(FMT, XSTEP, top_y, bottom_y, r_uv,                    \
+                  top_dst, bottom_dst, 16 * block + 1, 16);             \
+    top_u += 8;                                                         \
+    cur_u += 8;                                                         \
+    top_v += 8;                                                         \
+    cur_v += 8;                                                         \
+  }                                                                     \
+                                                                        \
+  UPSAMPLE_LAST_BLOCK(top_u, cur_u, leftover, r_uv);                    \
+  UPSAMPLE_LAST_BLOCK(top_v, cur_v, leftover, r_uv + 16);               \
+  CONVERT2RGB_1(VP8YuvTo ## FMT, XSTEP, top_y, bottom_y, r_uv,          \
+                top_dst, bottom_dst, last_pos, len - last_pos);         \
+}
+
+// NEON variants of the fancy upsampler.
+NEON_UPSAMPLE_FUNC(UpsampleRgbLinePair,  Rgb,  3)
+NEON_UPSAMPLE_FUNC(UpsampleBgrLinePair,  Bgr,  3)
+NEON_UPSAMPLE_FUNC(UpsampleRgbaLinePair, Rgba, 4)
+NEON_UPSAMPLE_FUNC(UpsampleBgraLinePair, Bgra, 4)
+
+#endif  // FANCY_UPSAMPLING
+
+#endif   // WEBP_USE_NEON
+
+//------------------------------------------------------------------------------
+
+extern void WebPInitUpsamplersNEON(void);
+
+#ifdef FANCY_UPSAMPLING
+
+extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
+
+void WebPInitUpsamplersNEON(void) {
+#if defined(WEBP_USE_NEON)
+  WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePair;
+  WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
+  WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePair;
+  WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
+  WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair;
+  WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair;
+#endif   // WEBP_USE_NEON
+}
+
+#else
+
+// this empty function is to avoid an empty .o
+void WebPInitUpsamplersNEON(void) {}
+
+#endif  // FANCY_UPSAMPLING

src/dsp/upsampling_sse2.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // SSE2 version of YUV to RGB upsampling functions.
@@ -11,10 +13,6 @@
 
 #include "./dsp.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 #if defined(WEBP_USE_SSE2)
 
 #include <assert.h>
@@ -49,14 +47,14 @@ extern "C" {
   (out) = _mm_sub_epi8(tmp0, tmp4);    /* (k + in + 1) / 2 - lsb_correction */ \
 } while (0)
 
-// pack and store two alterning pixel rows
+// pack and store two alternating pixel rows
 #define PACK_AND_STORE(a, b, da, db, out) do {                                 \
-  const __m128i ta = _mm_avg_epu8(a, da);  /* (9a + 3b + 3c +  d + 8) / 16 */  \
-  const __m128i tb = _mm_avg_epu8(b, db);  /* (3a + 9b +  c + 3d + 8) / 16 */  \
-  const __m128i t1 = _mm_unpacklo_epi8(ta, tb);                                \
-  const __m128i t2 = _mm_unpackhi_epi8(ta, tb);                                \
-  _mm_store_si128(((__m128i*)(out)) + 0, t1);                                  \
-  _mm_store_si128(((__m128i*)(out)) + 1, t2);                                  \
+  const __m128i t_a = _mm_avg_epu8(a, da);  /* (9a + 3b + 3c +  d + 8) / 16 */ \
+  const __m128i t_b = _mm_avg_epu8(b, db);  /* (3a + 9b +  c + 3d + 8) / 16 */ \
+  const __m128i t_1 = _mm_unpacklo_epi8(t_a, t_b);                             \
+  const __m128i t_2 = _mm_unpackhi_epi8(t_a, t_b);                             \
+  _mm_store_si128(((__m128i*)(out)) + 0, t_1);                                 \
+  _mm_store_si128(((__m128i*)(out)) + 1, t_2);                                 \
 } while (0)
 
 // Loads 17 pixels each from rows r1 and r2 and generates 32 pixels.
@@ -85,8 +83,8 @@ extern "C" {
   GET_M(ad, s, diag2);                  /* diag2 = (3a + b + c + 3d) / 8 */    \
                                                                                \
   /* pack the alternate pixels */                                              \
-  PACK_AND_STORE(a, b, diag1, diag2, &(out)[0 * 32]);                          \
-  PACK_AND_STORE(c, d, diag2, diag1, &(out)[2 * 32]);                          \
+  PACK_AND_STORE(a, b, diag1, diag2, out +      0);  /* store top */           \
+  PACK_AND_STORE(c, d, diag2, diag1, out + 2 * 32);  /* store bottom */        \
 }
 
 // Turn the macro into a function for reducing code-size when non-critical
@@ -106,82 +104,82 @@ static void Upsample32Pixels(const uint8_t r1[], const uint8_t r2[],
   Upsample32Pixels(r1, r2, out);                                               \
 }
 
-#define CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y, uv,                          \
+#define CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y,                              \
                     top_dst, bottom_dst, cur_x, num_pixels) {                  \
   int n;                                                                       \
-  if (top_y) {                                                                 \
-    for (n = 0; n < (num_pixels); ++n) {                                       \
-      FUNC(top_y[(cur_x) + n], (uv)[n], (uv)[32 + n],                          \
-           top_dst + ((cur_x) + n) * XSTEP);                                   \
-    }                                                                          \
+  for (n = 0; n < (num_pixels); ++n) {                                         \
+    FUNC(top_y[(cur_x) + n], r_u[n], r_v[n],                                   \
+         top_dst + ((cur_x) + n) * XSTEP);                                     \
   }                                                                            \
-  if (bottom_y) {                                                              \
+  if (bottom_y != NULL) {                                                      \
     for (n = 0; n < (num_pixels); ++n) {                                       \
-      FUNC(bottom_y[(cur_x) + n], (uv)[64 + n], (uv)[64 + 32 + n],             \
+      FUNC(bottom_y[(cur_x) + n], r_u[64 + n], r_v[64 + n],                    \
            bottom_dst + ((cur_x) + n) * XSTEP);                                \
     }                                                                          \
   }                                                                            \
 }
 
+#define CONVERT2RGB_32(FUNC, XSTEP, top_y, bottom_y,                           \
+                       top_dst, bottom_dst, cur_x) do {                        \
+  FUNC##32(top_y + (cur_x), r_u, r_v, top_dst + (cur_x) * XSTEP);              \
+  if (bottom_y != NULL) {                                                      \
+    FUNC##32(bottom_y + (cur_x), r_u + 64, r_v + 64,                           \
+             bottom_dst + (cur_x) * XSTEP);                                    \
+  }                                                                            \
+} while (0)
+
 #define SSE2_UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP)                             \
 static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
                       const uint8_t* top_u, const uint8_t* top_v,              \
                       const uint8_t* cur_u, const uint8_t* cur_v,              \
                       uint8_t* top_dst, uint8_t* bottom_dst, int len) {        \
-  int b;                                                                       \
-  /* 16 byte aligned array to cache reconstructed u and v */                   \
+  int uv_pos, pos;                                                             \
+  /* 16byte-aligned array to cache reconstructed u and v */                    \
   uint8_t uv_buf[4 * 32 + 15];                                                 \
-  uint8_t* const r_uv = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~15);            \
-  const int uv_len = (len + 1) >> 1;                                           \
-  /* 17 pixels must be read-able for each block */                             \
-  const int num_blocks = (uv_len - 1) >> 4;                                    \
-  const int leftover = uv_len - num_blocks * 16;                               \
-  const int last_pos = 1 + 32 * num_blocks;                                    \
+  uint8_t* const r_u = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~15);             \
+  uint8_t* const r_v = r_u + 32;                                               \
                                                                                \
-  const int u_diag = ((top_u[0] + cur_u[0]) >> 1) + 1;                         \
-  const int v_diag = ((top_v[0] + cur_v[0]) >> 1) + 1;                         \
-                                                                               \
-  assert(len > 0);                                                             \
-  /* Treat the first pixel in regular way */                                   \
-  if (top_y) {                                                                 \
-    const int u0 = (top_u[0] + u_diag) >> 1;                                   \
-    const int v0 = (top_v[0] + v_diag) >> 1;                                   \
-    FUNC(top_y[0], u0, v0, top_dst);                                           \
+  assert(top_y != NULL);                                                       \
+  {   /* Treat the first pixel in regular way */                               \
+    const int u_diag = ((top_u[0] + cur_u[0]) >> 1) + 1;                       \
+    const int v_diag = ((top_v[0] + cur_v[0]) >> 1) + 1;                       \
+    const int u0_t = (top_u[0] + u_diag) >> 1;                                 \
+    const int v0_t = (top_v[0] + v_diag) >> 1;                                 \
+    FUNC(top_y[0], u0_t, v0_t, top_dst);                                       \
+    if (bottom_y != NULL) {                                                    \
+      const int u0_b = (cur_u[0] + u_diag) >> 1;                               \
+      const int v0_b = (cur_v[0] + v_diag) >> 1;                               \
+      FUNC(bottom_y[0], u0_b, v0_b, bottom_dst);                               \
+    }                                                                          \
   }                                                                            \
-  if (bottom_y) {                                                              \
-    const int u0 = (cur_u[0] + u_diag) >> 1;                                   \
-    const int v0 = (cur_v[0] + v_diag) >> 1;                                   \
-    FUNC(bottom_y[0], u0, v0, bottom_dst);                                     \
+  /* For UPSAMPLE_32PIXELS, 17 u/v values must be read-able for each block */  \
+  for (pos = 1, uv_pos = 0; pos + 32 + 1 <= len; pos += 32, uv_pos += 16) {    \
+    UPSAMPLE_32PIXELS(top_u + uv_pos, cur_u + uv_pos, r_u);                    \
+    UPSAMPLE_32PIXELS(top_v + uv_pos, cur_v + uv_pos, r_v);                    \
+    CONVERT2RGB_32(FUNC, XSTEP, top_y, bottom_y, top_dst, bottom_dst, pos);    \
   }                                                                            \
-                                                                               \
-  for (b = 0; b < num_blocks; ++b) {                                           \
-    UPSAMPLE_32PIXELS(top_u, cur_u, r_uv + 0 * 32);                            \
-    UPSAMPLE_32PIXELS(top_v, cur_v, r_uv + 1 * 32);                            \
-    CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y, r_uv, top_dst, bottom_dst,       \
-                32 * b + 1, 32)                                                \
-    top_u += 16;                                                               \
-    cur_u += 16;                                                               \
-    top_v += 16;                                                               \
-    cur_v += 16;                                                               \
+  if (len > 1) {                                                               \
+    const int left_over = ((len + 1) >> 1) - (pos >> 1);                       \
+    assert(left_over > 0);                                                     \
+    UPSAMPLE_LAST_BLOCK(top_u + uv_pos, cur_u + uv_pos, left_over, r_u);       \
+    UPSAMPLE_LAST_BLOCK(top_v + uv_pos, cur_v + uv_pos, left_over, r_v);       \
+    CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y, top_dst, bottom_dst,             \
+                pos, len - pos);                                               \
   }                                                                            \
-                                                                               \
-  UPSAMPLE_LAST_BLOCK(top_u, cur_u, leftover, r_uv + 0 * 32);                  \
-  UPSAMPLE_LAST_BLOCK(top_v, cur_v, leftover, r_uv + 1 * 32);                  \
-  CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y, r_uv, top_dst, bottom_dst,         \
-              last_pos, len - last_pos);                                       \
 }
 
 // SSE2 variants of the fancy upsampler.
-SSE2_UPSAMPLE_FUNC(UpsampleRgbLinePairSSE2,  VP8YuvToRgb,  3)
-SSE2_UPSAMPLE_FUNC(UpsampleBgrLinePairSSE2,  VP8YuvToBgr,  3)
-SSE2_UPSAMPLE_FUNC(UpsampleRgbaLinePairSSE2, VP8YuvToRgba, 4)
-SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePairSSE2, VP8YuvToBgra, 4)
+SSE2_UPSAMPLE_FUNC(UpsampleRgbLinePair,  VP8YuvToRgb,  3)
+SSE2_UPSAMPLE_FUNC(UpsampleBgrLinePair,  VP8YuvToBgr,  3)
+SSE2_UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4)
+SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4)
 
 #undef GET_M
 #undef PACK_AND_STORE
 #undef UPSAMPLE_32PIXELS
 #undef UPSAMPLE_LAST_BLOCK
 #undef CONVERT2RGB
+#undef CONVERT2RGB_32
 #undef SSE2_UPSAMPLE_FUNC
 
 #endif  // FANCY_UPSAMPLING
@@ -190,26 +188,27 @@ SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePairSSE2, VP8YuvToBgra, 4)
 
 //------------------------------------------------------------------------------
 
+extern void WebPInitUpsamplersSSE2(void);
+
+#ifdef FANCY_UPSAMPLING
+
 extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
 
 void WebPInitUpsamplersSSE2(void) {
 #if defined(WEBP_USE_SSE2)
-  WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePairSSE2;
-  WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePairSSE2;
-  WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePairSSE2;
-  WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePairSSE2;
+  VP8YUVInitSSE2();
+  WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePair;
+  WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
+  WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePair;
+  WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
+  WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair;
+  WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair;
 #endif   // WEBP_USE_SSE2
 }
 
-void WebPInitPremultiplySSE2(void) {
-#if defined(WEBP_USE_SSE2)
-  WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePairSSE2;
-  WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePairSSE2;
-#endif   // WEBP_USE_SSE2
-}
-
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif
+#else
 
+// this empty function is to avoid an empty .o
+void WebPInitUpsamplersSSE2(void) {}
 
+#endif  // FANCY_UPSAMPLING

src/dsp/yuv.c

@@ -1,26 +1,19 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
-// YUV->RGB conversion function
+// YUV->RGB conversion functions
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include "./yuv.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
-enum { YUV_HALF = 1 << (YUV_FIX - 1) };
-
-int16_t VP8kVToR[256], VP8kUToB[256];
-int32_t VP8kVToG[256], VP8kUToG[256];
-uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
-uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
+#if defined(WEBP_YUV_USE_TABLE)
 
 static int done = 0;
 
@@ -28,11 +21,17 @@ static WEBP_INLINE uint8_t clip(int v, int max_value) {
   return v < 0 ? 0 : v > max_value ? max_value : v;
 }
 
+int16_t VP8kVToR[256], VP8kUToB[256];
+int32_t VP8kVToG[256], VP8kUToG[256];
+uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
+uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
+
 void VP8YUVInit(void) {
   int i;
   if (done) {
     return;
   }
+#ifndef USE_YUVj
   for (i = 0; i < 256; ++i) {
     VP8kVToR[i] = (89858 * (i - 128) + YUV_HALF) >> YUV_FIX;
     VP8kUToG[i] = -22014 * (i - 128) + YUV_HALF;
@@ -44,9 +43,112 @@ void VP8YUVInit(void) {
     VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255);
     VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15);
   }
+#else
+  for (i = 0; i < 256; ++i) {
+    VP8kVToR[i] = (91881 * (i - 128) + YUV_HALF) >> YUV_FIX;
+    VP8kUToG[i] = -22554 * (i - 128) + YUV_HALF;
+    VP8kVToG[i] = -46802 * (i - 128);
+    VP8kUToB[i] = (116130 * (i - 128) + YUV_HALF) >> YUV_FIX;
+  }
+  for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) {
+    const int k = i;
+    VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255);
+    VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15);
+  }
+#endif
+
   done = 1;
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif
+#else
+
+void VP8YUVInit(void) {}
+
+#endif  // WEBP_YUV_USE_TABLE
+
+//-----------------------------------------------------------------------------
+// Plain-C version
+
+#define ROW_FUNC(FUNC_NAME, FUNC, XSTEP)                                       \
+static void FUNC_NAME(const uint8_t* y,                                        \
+                      const uint8_t* u, const uint8_t* v,                      \
+                      uint8_t* dst, int len) {                                 \
+  const uint8_t* const end = dst + (len & ~1) * XSTEP;                         \
+  while (dst != end) {                                                         \
+    FUNC(y[0], u[0], v[0], dst);                                               \
+    FUNC(y[1], u[0], v[0], dst + XSTEP);                                       \
+    y += 2;                                                                    \
+    ++u;                                                                       \
+    ++v;                                                                       \
+    dst += 2 * XSTEP;                                                          \
+  }                                                                            \
+  if (len & 1) {                                                               \
+    FUNC(y[0], u[0], v[0], dst);                                               \
+  }                                                                            \
+}                                                                              \
+
+// All variants implemented.
+ROW_FUNC(YuvToRgbRow,      VP8YuvToRgb,  3)
+ROW_FUNC(YuvToBgrRow,      VP8YuvToBgr,  3)
+ROW_FUNC(YuvToRgbaRow,     VP8YuvToRgba, 4)
+ROW_FUNC(YuvToBgraRow,     VP8YuvToBgra, 4)
+ROW_FUNC(YuvToArgbRow,     VP8YuvToArgb, 4)
+ROW_FUNC(YuvToRgba4444Row, VP8YuvToRgba4444, 2)
+ROW_FUNC(YuvToRgb565Row,   VP8YuvToRgb565, 2)
+
+#undef ROW_FUNC
+
+// Main call for processing a plane with a WebPSamplerRowFunc function:
+void WebPSamplerProcessPlane(const uint8_t* y, int y_stride,
+                             const uint8_t* u, const uint8_t* v, int uv_stride,
+                             uint8_t* dst, int dst_stride,
+                             int width, int height, WebPSamplerRowFunc func) {
+  int j;
+  for (j = 0; j < height; ++j) {
+    func(y, u, v, dst, width);
+    y += y_stride;
+    if (j & 1) {
+      u += uv_stride;
+      v += uv_stride;
+    }
+    dst += dst_stride;
+  }
+}
+
+//-----------------------------------------------------------------------------
+// Main call
+
+WebPSamplerRowFunc WebPSamplers[MODE_LAST];
+
+extern void WebPInitSamplersSSE2(void);
+extern void WebPInitSamplersMIPS32(void);
+
+void WebPInitSamplers(void) {
+  WebPSamplers[MODE_RGB]       = YuvToRgbRow;
+  WebPSamplers[MODE_RGBA]      = YuvToRgbaRow;
+  WebPSamplers[MODE_BGR]       = YuvToBgrRow;
+  WebPSamplers[MODE_BGRA]      = YuvToBgraRow;
+  WebPSamplers[MODE_ARGB]      = YuvToArgbRow;
+  WebPSamplers[MODE_RGBA_4444] = YuvToRgba4444Row;
+  WebPSamplers[MODE_RGB_565]   = YuvToRgb565Row;
+  WebPSamplers[MODE_rgbA]      = YuvToRgbaRow;
+  WebPSamplers[MODE_bgrA]      = YuvToBgraRow;
+  WebPSamplers[MODE_Argb]      = YuvToArgbRow;
+  WebPSamplers[MODE_rgbA_4444] = YuvToRgba4444Row;
+
+  // If defined, use CPUInfo() to overwrite some pointers with faster versions.
+  if (VP8GetCPUInfo != NULL) {
+#if defined(WEBP_USE_SSE2)
+    if (VP8GetCPUInfo(kSSE2)) {
+      WebPInitSamplersSSE2();
+    }
+#endif  // WEBP_USE_SSE2
+#if defined(WEBP_USE_MIPS32)
+    if (VP8GetCPUInfo(kMIPS32)) {
+      WebPInitSamplersMIPS32();
+    }
+#endif  // WEBP_USE_MIPS32
+  }
+}
+
+//-----------------------------------------------------------------------------

src/dsp/yuv.h

@@ -1,36 +1,165 @@
 // Copyright 2010 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // inline YUV<->RGB conversion function
 //
+// The exact naming is Y'CbCr, following the ITU-R BT.601 standard.
+// More information at: http://en.wikipedia.org/wiki/YCbCr
+// Y = 0.2569 * R + 0.5044 * G + 0.0979 * B + 16
+// U = -0.1483 * R - 0.2911 * G + 0.4394 * B + 128
+// V = 0.4394 * R - 0.3679 * G - 0.0715 * B + 128
+// We use 16bit fixed point operations for RGB->YUV conversion (YUV_FIX).
+//
+// For the Y'CbCr to RGB conversion, the BT.601 specification reads:
+//   R = 1.164 * (Y-16) + 1.596 * (V-128)
+//   G = 1.164 * (Y-16) - 0.813 * (V-128) - 0.391 * (U-128)
+//   B = 1.164 * (Y-16)                   + 2.018 * (U-128)
+// where Y is in the [16,235] range, and U/V in the [16,240] range.
+// In the table-lookup version (WEBP_YUV_USE_TABLE), the common factor
+// "1.164 * (Y-16)" can be handled as an offset in the VP8kClip[] table.
+// So in this case the formulae should read:
+//   R = 1.164 * [Y + 1.371 * (V-128)                  ] - 18.624
+//   G = 1.164 * [Y - 0.698 * (V-128) - 0.336 * (U-128)] - 18.624
+//   B = 1.164 * [Y                   + 1.733 * (U-128)] - 18.624
+// once factorized.
+// For YUV->RGB conversion, only 14bit fixed precision is used (YUV_FIX2).
+// That's the maximum possible for a convenient ARM implementation.
+//
 // Author: Skal (pascal.massimino@gmail.com)
 
 #ifndef WEBP_DSP_YUV_H_
 #define WEBP_DSP_YUV_H_
 
+#include "./dsp.h"
 #include "../dec/decode_vp8.h"
 
+// Define the following to use the LUT-based code:
+// #define WEBP_YUV_USE_TABLE
+
+#if defined(WEBP_EXPERIMENTAL_FEATURES)
+// Do NOT activate this feature for real compression. This is only experimental!
+// This flag is for comparison purpose against JPEG's "YUVj" natural colorspace.
+// This colorspace is close to Rec.601's Y'CbCr model with the notable
+// difference of allowing larger range for luma/chroma.
+// See http://en.wikipedia.org/wiki/YCbCr#JPEG_conversion paragraph, and its
+// difference with http://en.wikipedia.org/wiki/YCbCr#ITU-R_BT.601_conversion
+// #define USE_YUVj
+#endif
+
 //------------------------------------------------------------------------------
 // YUV -> RGB conversion
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
 extern "C" {
 #endif
 
-enum { YUV_FIX = 16,                // fixed-point precision
-       YUV_RANGE_MIN = -227,        // min value of r/g/b output
-       YUV_RANGE_MAX = 256 + 226    // max value of r/g/b output
+enum {
+  YUV_FIX = 16,                    // fixed-point precision for RGB->YUV
+  YUV_HALF = 1 << (YUV_FIX - 1),
+  YUV_MASK = (256 << YUV_FIX) - 1,
+  YUV_RANGE_MIN = -227,            // min value of r/g/b output
+  YUV_RANGE_MAX = 256 + 226,       // max value of r/g/b output
+
+  YUV_FIX2 = 14,                   // fixed-point precision for YUV->RGB
+  YUV_HALF2 = 1 << (YUV_FIX2 - 1),
+  YUV_MASK2 = (256 << YUV_FIX2) - 1
 };
+
+// These constants are 14b fixed-point version of ITU-R BT.601 constants.
+#define kYScale 19077    // 1.164 = 255 / 219
+#define kVToR   26149    // 1.596 = 255 / 112 * 0.701
+#define kUToG   6419     // 0.391 = 255 / 112 * 0.886 * 0.114 / 0.587
+#define kVToG   13320    // 0.813 = 255 / 112 * 0.701 * 0.299 / 0.587
+#define kUToB   33050    // 2.018 = 255 / 112 * 0.886
+#define kRCst (-kYScale * 16 - kVToR * 128 + YUV_HALF2)
+#define kGCst (-kYScale * 16 + kUToG * 128 + kVToG * 128 + YUV_HALF2)
+#define kBCst (-kYScale * 16 - kUToB * 128 + YUV_HALF2)
+
+//------------------------------------------------------------------------------
+
+#if !defined(WEBP_YUV_USE_TABLE)
+
+// slower on x86 by ~7-8%, but bit-exact with the SSE2 version
+
+static WEBP_INLINE int VP8Clip8(int v) {
+  return ((v & ~YUV_MASK2) == 0) ? (v >> YUV_FIX2) : (v < 0) ? 0 : 255;
+}
+
+static WEBP_INLINE int VP8YUVToR(int y, int v) {
+  return VP8Clip8(kYScale * y + kVToR * v + kRCst);
+}
+
+static WEBP_INLINE int VP8YUVToG(int y, int u, int v) {
+  return VP8Clip8(kYScale * y - kUToG * u - kVToG * v + kGCst);
+}
+
+static WEBP_INLINE int VP8YUVToB(int y, int u) {
+  return VP8Clip8(kYScale * y + kUToB * u + kBCst);
+}
+
+static WEBP_INLINE void VP8YuvToRgb(int y, int u, int v,
+                                    uint8_t* const rgb) {
+  rgb[0] = VP8YUVToR(y, v);
+  rgb[1] = VP8YUVToG(y, u, v);
+  rgb[2] = VP8YUVToB(y, u);
+}
+
+static WEBP_INLINE void VP8YuvToBgr(int y, int u, int v,
+                                    uint8_t* const bgr) {
+  bgr[0] = VP8YUVToB(y, u);
+  bgr[1] = VP8YUVToG(y, u, v);
+  bgr[2] = VP8YUVToR(y, v);
+}
+
+static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v,
+                                       uint8_t* const rgb) {
+  const int r = VP8YUVToR(y, v);      // 5 usable bits
+  const int g = VP8YUVToG(y, u, v);   // 6 usable bits
+  const int b = VP8YUVToB(y, u);      // 5 usable bits
+  const int rg = (r & 0xf8) | (g >> 5);
+  const int gb = ((g << 3) & 0xe0) | (b >> 3);
+#ifdef WEBP_SWAP_16BIT_CSP
+  rgb[0] = gb;
+  rgb[1] = rg;
+#else
+  rgb[0] = rg;
+  rgb[1] = gb;
+#endif
+}
+
+static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v,
+                                         uint8_t* const argb) {
+  const int r = VP8YUVToR(y, v);        // 4 usable bits
+  const int g = VP8YUVToG(y, u, v);     // 4 usable bits
+  const int b = VP8YUVToB(y, u);        // 4 usable bits
+  const int rg = (r & 0xf0) | (g >> 4);
+  const int ba = (b & 0xf0) | 0x0f;     // overwrite the lower 4 bits
+#ifdef WEBP_SWAP_16BIT_CSP
+  argb[0] = ba;
+  argb[1] = rg;
+#else
+  argb[0] = rg;
+  argb[1] = ba;
+#endif
+}
+
+#else
+
+// Table-based version, not totally equivalent to the SSE2 version.
+// Rounding diff is only +/-1 though.
+
 extern int16_t VP8kVToR[256], VP8kUToB[256];
 extern int32_t VP8kVToG[256], VP8kUToG[256];
 extern uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
 extern uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
 
-static WEBP_INLINE void VP8YuvToRgb(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToRgb(int y, int u, int v,
                                     uint8_t* const rgb) {
   const int r_off = VP8kVToR[v];
   const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
@@ -40,35 +169,7 @@ static WEBP_INLINE void VP8YuvToRgb(uint8_t y, uint8_t u, uint8_t v,
   rgb[2] = VP8kClip[y + b_off - YUV_RANGE_MIN];
 }
 
-static WEBP_INLINE void VP8YuvToRgb565(uint8_t y, uint8_t u, uint8_t v,
-                                       uint8_t* const rgb) {
-  const int r_off = VP8kVToR[v];
-  const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
-  const int b_off = VP8kUToB[u];
-  rgb[0] = ((VP8kClip[y + r_off - YUV_RANGE_MIN] & 0xf8) |
-            (VP8kClip[y + g_off - YUV_RANGE_MIN] >> 5));
-  rgb[1] = (((VP8kClip[y + g_off - YUV_RANGE_MIN] << 3) & 0xe0) |
-            (VP8kClip[y + b_off - YUV_RANGE_MIN] >> 3));
-}
-
-static WEBP_INLINE void VP8YuvToArgb(uint8_t y, uint8_t u, uint8_t v,
-                                     uint8_t* const argb) {
-  argb[0] = 0xff;
-  VP8YuvToRgb(y, u, v, argb + 1);
-}
-
-static WEBP_INLINE void VP8YuvToRgba4444(uint8_t y, uint8_t u, uint8_t v,
-                                         uint8_t* const argb) {
-  const int r_off = VP8kVToR[v];
-  const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
-  const int b_off = VP8kUToB[u];
-  // Don't update alpha (last 4 bits of argb[1])
-  argb[0] = ((VP8kClip4Bits[y + r_off - YUV_RANGE_MIN] << 4) |
-             VP8kClip4Bits[y + g_off - YUV_RANGE_MIN]);
-  argb[1] = 0x0f | (VP8kClip4Bits[y + b_off - YUV_RANGE_MIN] << 4);
-}
-
-static WEBP_INLINE void VP8YuvToBgr(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToBgr(int y, int u, int v,
                                     uint8_t* const bgr) {
   const int r_off = VP8kVToR[v];
   const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
@@ -78,6 +179,52 @@ static WEBP_INLINE void VP8YuvToBgr(uint8_t y, uint8_t u, uint8_t v,
   bgr[2] = VP8kClip[y + r_off - YUV_RANGE_MIN];
 }
 
+static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v,
+                                       uint8_t* const rgb) {
+  const int r_off = VP8kVToR[v];
+  const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
+  const int b_off = VP8kUToB[u];
+  const int rg = ((VP8kClip[y + r_off - YUV_RANGE_MIN] & 0xf8) |
+                  (VP8kClip[y + g_off - YUV_RANGE_MIN] >> 5));
+  const int gb = (((VP8kClip[y + g_off - YUV_RANGE_MIN] << 3) & 0xe0) |
+                   (VP8kClip[y + b_off - YUV_RANGE_MIN] >> 3));
+#ifdef WEBP_SWAP_16BIT_CSP
+  rgb[0] = gb;
+  rgb[1] = rg;
+#else
+  rgb[0] = rg;
+  rgb[1] = gb;
+#endif
+}
+
+static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v,
+                                         uint8_t* const argb) {
+  const int r_off = VP8kVToR[v];
+  const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
+  const int b_off = VP8kUToB[u];
+  const int rg = ((VP8kClip4Bits[y + r_off - YUV_RANGE_MIN] << 4) |
+                   VP8kClip4Bits[y + g_off - YUV_RANGE_MIN]);
+  const int ba = (VP8kClip4Bits[y + b_off - YUV_RANGE_MIN] << 4) | 0x0f;
+#ifdef WEBP_SWAP_16BIT_CSP
+  argb[0] = ba;
+  argb[1] = rg;
+#else
+  argb[0] = rg;
+  argb[1] = ba;
+#endif
+}
+
+#endif  // WEBP_YUV_USE_TABLE
+
+//-----------------------------------------------------------------------------
+// Alpha handling variants
+
+static WEBP_INLINE void VP8YuvToArgb(uint8_t y, uint8_t u, uint8_t v,
+                                     uint8_t* const argb) {
+  argb[0] = 0xff;
+  VP8YuvToRgb(y, u, v, argb + 1);
+}
+
 static WEBP_INLINE void VP8YuvToBgra(uint8_t y, uint8_t u, uint8_t v,
                                      uint8_t* const bgra) {
   VP8YuvToBgr(y, u, v, bgra);
@@ -93,35 +240,81 @@ static WEBP_INLINE void VP8YuvToRgba(uint8_t y, uint8_t u, uint8_t v,
 // Must be called before everything, to initialize the tables.
 void VP8YUVInit(void);
 
+//-----------------------------------------------------------------------------
+// SSE2 extra functions (mostly for upsampling_sse2.c)
+
+#if defined(WEBP_USE_SSE2)
+
+// When the following is defined, tables are initialized statically, adding ~12k
+// to the binary size. Otherwise, they are initialized at run-time (small cost).
+#define WEBP_YUV_USE_SSE2_TABLES
+
+#if defined(FANCY_UPSAMPLING)
+// Process 32 pixels and store the result (24b or 32b per pixel) in *dst.
+void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                    uint8_t* dst);
+void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                   uint8_t* dst);
+void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                    uint8_t* dst);
+void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                   uint8_t* dst);
+#endif  // FANCY_UPSAMPLING
+
+// Must be called to initialize tables before using the functions.
+void VP8YUVInitSSE2(void);
+
+#endif    // WEBP_USE_SSE2
+
 //------------------------------------------------------------------------------
 // RGB -> YUV conversion
-// The exact naming is Y'CbCr, following the ITU-R BT.601 standard.
-// More information at: http://en.wikipedia.org/wiki/YCbCr
-// Y = 0.2569 * R + 0.5044 * G + 0.0979 * B + 16
-// U = -0.1483 * R - 0.2911 * G + 0.4394 * B + 128
-// V = 0.4394 * R - 0.3679 * G - 0.0715 * B + 128
-// We use 16bit fixed point operations.
 
-static WEBP_INLINE int VP8ClipUV(int v) {
-   v = (v + (257 << (YUV_FIX + 2 - 1))) >> (YUV_FIX + 2);
-   return ((v & ~0xff) == 0) ? v : (v < 0) ? 0 : 255;
+// Stub functions that can be called with various rounding values:
+static WEBP_INLINE int VP8ClipUV(int uv, int rounding) {
+  uv = (uv + rounding + (128 << (YUV_FIX + 2))) >> (YUV_FIX + 2);
+  return ((uv & ~0xff) == 0) ? uv : (uv < 0) ? 0 : 255;
 }
 
-static WEBP_INLINE int VP8RGBToY(int r, int g, int b) {
-  const int kRound = (1 << (YUV_FIX - 1)) + (16 << YUV_FIX);
+#ifndef USE_YUVj
+
+static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
   const int luma = 16839 * r + 33059 * g + 6420 * b;
-  return (luma + kRound) >> YUV_FIX;  // no need to clip
+  return (luma + rounding + (16 << YUV_FIX)) >> YUV_FIX;  // no need to clip
 }
 
-static WEBP_INLINE int VP8RGBToU(int r, int g, int b) {
-  return VP8ClipUV(-9719 * r - 19081 * g + 28800 * b);
+static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) {
+  const int u = -9719 * r - 19081 * g + 28800 * b;
+  return VP8ClipUV(u, rounding);
 }
 
-static WEBP_INLINE int VP8RGBToV(int r, int g, int b) {
-  return VP8ClipUV(+28800 * r - 24116 * g - 4684 * b);
+static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
+  const int v = +28800 * r - 24116 * g - 4684 * b;
+  return VP8ClipUV(v, rounding);
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
+#else
+
+// This JPEG-YUV colorspace, only for comparison!
+// These are also 16bit precision coefficients from Rec.601, but with full
+// [0..255] output range.
+static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
+  const int luma = 19595 * r + 38470 * g + 7471 * b;
+  return (luma + rounding) >> YUV_FIX;  // no need to clip
+}
+
+static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) {
+  const int u = -11058 * r - 21710 * g + 32768 * b;
+  return VP8ClipUV(u, rounding);
+}
+
+static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
+  const int v = 32768 * r - 27439 * g - 5329 * b;
+  return VP8ClipUV(v, rounding);
+}
+
+#endif    // USE_YUVj
+
+#ifdef __cplusplus
 }    // extern "C"
 #endif
 

src/dsp/yuv_mips32.c

@@ -0,0 +1,100 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// MIPS version of YUV to RGB upsampling functions.
+//
+// Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
+//             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_MIPS32)
+
+#include "./yuv.h"
+
+//------------------------------------------------------------------------------
+// simple point-sampling
+
+#define ROW_FUNC(FUNC_NAME, XSTEP, R, G, B, A)                                 \
+static void FUNC_NAME(const uint8_t* y,                                        \
+                      const uint8_t* u, const uint8_t* v,                      \
+                      uint8_t* dst, int len) {                                 \
+  int i, r, g, b;                                                              \
+  int temp0, temp1, temp2, temp3, temp4;                                       \
+  for (i = 0; i < (len >> 1); i++) {                                           \
+    temp1 = kVToR * v[0];                                                      \
+    temp3 = kVToG * v[0];                                                      \
+    temp2 = kUToG * u[0];                                                      \
+    temp4 = kUToB * u[0];                                                      \
+    temp0 = kYScale * y[0];                                                    \
+    temp1 += kRCst;                                                            \
+    temp3 -= kGCst;                                                            \
+    temp2 += temp3;                                                            \
+    temp4 += kBCst;                                                            \
+    r = VP8Clip8(temp0 + temp1);                                               \
+    g = VP8Clip8(temp0 - temp2);                                               \
+    b = VP8Clip8(temp0 + temp4);                                               \
+    temp0 = kYScale * y[1];                                                    \
+    dst[R] = r;                                                                \
+    dst[G] = g;                                                                \
+    dst[B] = b;                                                                \
+    if (A) dst[A] = 0xff;                                                      \
+    r = VP8Clip8(temp0 + temp1);                                               \
+    g = VP8Clip8(temp0 - temp2);                                               \
+    b = VP8Clip8(temp0 + temp4);                                               \
+    dst[R + XSTEP] = r;                                                        \
+    dst[G + XSTEP] = g;                                                        \
+    dst[B + XSTEP] = b;                                                        \
+    if (A) dst[A + XSTEP] = 0xff;                                              \
+    y += 2;                                                                    \
+    ++u;                                                                       \
+    ++v;                                                                       \
+    dst += 2 * XSTEP;                                                          \
+  }                                                                            \
+  if (len & 1) {                                                               \
+    temp1 = kVToR * v[0];                                                      \
+    temp3 = kVToG * v[0];                                                      \
+    temp2 = kUToG * u[0];                                                      \
+    temp4 = kUToB * u[0];                                                      \
+    temp0 = kYScale * y[0];                                                    \
+    temp1 += kRCst;                                                            \
+    temp3 -= kGCst;                                                            \
+    temp2 += temp3;                                                            \
+    temp4 += kBCst;                                                            \
+    r = VP8Clip8(temp0 + temp1);                                               \
+    g = VP8Clip8(temp0 - temp2);                                               \
+    b = VP8Clip8(temp0 + temp4);                                               \
+    dst[R] = r;                                                                \
+    dst[G] = g;                                                                \
+    dst[B] = b;                                                                \
+    if (A) dst[A] = 0xff;                                                      \
+  }                                                                            \
+}
+
+ROW_FUNC(YuvToRgbRow,      3, 0, 1, 2, 0)
+ROW_FUNC(YuvToRgbaRow,     4, 0, 1, 2, 3)
+ROW_FUNC(YuvToBgrRow,      3, 2, 1, 0, 0)
+ROW_FUNC(YuvToBgraRow,     4, 2, 1, 0, 3)
+
+#undef ROW_FUNC
+
+#endif   // WEBP_USE_MIPS32
+
+//------------------------------------------------------------------------------
+
+extern void WebPInitSamplersMIPS32(void);
+
+void WebPInitSamplersMIPS32(void) {
+#if defined(WEBP_USE_MIPS32)
+  WebPSamplers[MODE_RGB]  = YuvToRgbRow;
+  WebPSamplers[MODE_RGBA] = YuvToRgbaRow;
+  WebPSamplers[MODE_BGR]  = YuvToBgrRow;
+  WebPSamplers[MODE_BGRA] = YuvToBgraRow;
+#endif  // WEBP_USE_MIPS32
+}

src/dsp/yuv_sse2.c

@@ -0,0 +1,322 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// YUV->RGB conversion functions
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./yuv.h"
+
+#if defined(WEBP_USE_SSE2)
+
+#include <emmintrin.h>
+#include <string.h>   // for memcpy
+
+typedef union {   // handy struct for converting SSE2 registers
+  int32_t i32[4];
+  uint8_t u8[16];
+  __m128i m;
+} VP8kCstSSE2;
+
+#if defined(WEBP_YUV_USE_SSE2_TABLES)
+
+#include "./yuv_tables_sse2.h"
+
+void VP8YUVInitSSE2(void) {}
+
+#else
+
+static int done_sse2 = 0;
+static VP8kCstSSE2 VP8kUtoRGBA[256], VP8kVtoRGBA[256], VP8kYtoRGBA[256];
+
+void VP8YUVInitSSE2(void) {
+  if (!done_sse2) {
+    int i;
+    for (i = 0; i < 256; ++i) {
+      VP8kYtoRGBA[i].i32[0] =
+        VP8kYtoRGBA[i].i32[1] =
+        VP8kYtoRGBA[i].i32[2] = (i - 16) * kYScale + YUV_HALF2;
+      VP8kYtoRGBA[i].i32[3] = 0xff << YUV_FIX2;
+
+      VP8kUtoRGBA[i].i32[0] = 0;
+      VP8kUtoRGBA[i].i32[1] = -kUToG * (i - 128);
+      VP8kUtoRGBA[i].i32[2] =  kUToB * (i - 128);
+      VP8kUtoRGBA[i].i32[3] = 0;
+
+      VP8kVtoRGBA[i].i32[0] =  kVToR * (i - 128);
+      VP8kVtoRGBA[i].i32[1] = -kVToG * (i - 128);
+      VP8kVtoRGBA[i].i32[2] = 0;
+      VP8kVtoRGBA[i].i32[3] = 0;
+    }
+    done_sse2 = 1;
+
+#if 0   // code used to generate 'yuv_tables_sse2.h'
+    printf("static const VP8kCstSSE2 VP8kYtoRGBA[256] = {\n");
+    for (i = 0; i < 256; ++i) {
+      printf("  {{0x%.8x, 0x%.8x, 0x%.8x, 0x%.8x}},\n",
+             VP8kYtoRGBA[i].i32[0], VP8kYtoRGBA[i].i32[1],
+             VP8kYtoRGBA[i].i32[2], VP8kYtoRGBA[i].i32[3]);
+    }
+    printf("};\n\n");
+    printf("static const VP8kCstSSE2 VP8kUtoRGBA[256] = {\n");
+    for (i = 0; i < 256; ++i) {
+      printf("  {{0, 0x%.8x, 0x%.8x, 0}},\n",
+             VP8kUtoRGBA[i].i32[1], VP8kUtoRGBA[i].i32[2]);
+    }
+    printf("};\n\n");
+    printf("static VP8kCstSSE2 VP8kVtoRGBA[256] = {\n");
+    for (i = 0; i < 256; ++i) {
+      printf("  {{0x%.8x, 0x%.8x, 0, 0}},\n",
+             VP8kVtoRGBA[i].i32[0], VP8kVtoRGBA[i].i32[1]);
+    }
+    printf("};\n\n");
+#endif
+  }
+}
+
+#endif  // WEBP_YUV_USE_SSE2_TABLES
+
+//-----------------------------------------------------------------------------
+
+static WEBP_INLINE __m128i LoadUVPart(int u, int v) {
+  const __m128i u_part = _mm_loadu_si128(&VP8kUtoRGBA[u].m);
+  const __m128i v_part = _mm_loadu_si128(&VP8kVtoRGBA[v].m);
+  const __m128i uv_part = _mm_add_epi32(u_part, v_part);
+  return uv_part;
+}
+
+static WEBP_INLINE __m128i GetRGBA32bWithUV(int y, const __m128i uv_part) {
+  const __m128i y_part = _mm_loadu_si128(&VP8kYtoRGBA[y].m);
+  const __m128i rgba1 = _mm_add_epi32(y_part, uv_part);
+  const __m128i rgba2 = _mm_srai_epi32(rgba1, YUV_FIX2);
+  return rgba2;
+}
+
+static WEBP_INLINE __m128i GetRGBA32b(int y, int u, int v) {
+  const __m128i uv_part = LoadUVPart(u, v);
+  return GetRGBA32bWithUV(y, uv_part);
+}
+
+static WEBP_INLINE void YuvToRgbSSE2(uint8_t y, uint8_t u, uint8_t v,
+                                     uint8_t* const rgb) {
+  const __m128i tmp0 = GetRGBA32b(y, u, v);
+  const __m128i tmp1 = _mm_packs_epi32(tmp0, tmp0);
+  const __m128i tmp2 = _mm_packus_epi16(tmp1, tmp1);
+  // Note: we store 8 bytes at a time, not 3 bytes! -> memory stomp
+  _mm_storel_epi64((__m128i*)rgb, tmp2);
+}
+
+static WEBP_INLINE void YuvToBgrSSE2(uint8_t y, uint8_t u, uint8_t v,
+                                     uint8_t* const bgr) {
+  const __m128i tmp0 = GetRGBA32b(y, u, v);
+  const __m128i tmp1 = _mm_shuffle_epi32(tmp0, _MM_SHUFFLE(3, 0, 1, 2));
+  const __m128i tmp2 = _mm_packs_epi32(tmp1, tmp1);
+  const __m128i tmp3 = _mm_packus_epi16(tmp2, tmp2);
+  // Note: we store 8 bytes at a time, not 3 bytes! -> memory stomp
+  _mm_storel_epi64((__m128i*)bgr, tmp3);
+}
+
+//-----------------------------------------------------------------------------
+// Convert spans of 32 pixels to various RGB formats for the fancy upsampler.
+
+#ifdef FANCY_UPSAMPLING
+
+void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                    uint8_t* dst) {
+  int n;
+  for (n = 0; n < 32; n += 4) {
+    const __m128i tmp0_1 = GetRGBA32b(y[n + 0], u[n + 0], v[n + 0]);
+    const __m128i tmp0_2 = GetRGBA32b(y[n + 1], u[n + 1], v[n + 1]);
+    const __m128i tmp0_3 = GetRGBA32b(y[n + 2], u[n + 2], v[n + 2]);
+    const __m128i tmp0_4 = GetRGBA32b(y[n + 3], u[n + 3], v[n + 3]);
+    const __m128i tmp1_1 = _mm_packs_epi32(tmp0_1, tmp0_2);
+    const __m128i tmp1_2 = _mm_packs_epi32(tmp0_3, tmp0_4);
+    const __m128i tmp2 = _mm_packus_epi16(tmp1_1, tmp1_2);
+    _mm_storeu_si128((__m128i*)dst, tmp2);
+    dst += 4 * 4;
+  }
+}
+
+void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                    uint8_t* dst) {
+  int n;
+  for (n = 0; n < 32; n += 2) {
+    const __m128i tmp0_1 = GetRGBA32b(y[n + 0], u[n + 0], v[n + 0]);
+    const __m128i tmp0_2 = GetRGBA32b(y[n + 1], u[n + 1], v[n + 1]);
+    const __m128i tmp1_1 = _mm_shuffle_epi32(tmp0_1, _MM_SHUFFLE(3, 0, 1, 2));
+    const __m128i tmp1_2 = _mm_shuffle_epi32(tmp0_2, _MM_SHUFFLE(3, 0, 1, 2));
+    const __m128i tmp2_1 = _mm_packs_epi32(tmp1_1, tmp1_2);
+    const __m128i tmp3 = _mm_packus_epi16(tmp2_1, tmp2_1);
+    _mm_storel_epi64((__m128i*)dst, tmp3);
+    dst += 4 * 2;
+  }
+}
+
+void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                   uint8_t* dst) {
+  int n;
+  uint8_t tmp0[2 * 3 + 5 + 15];
+  uint8_t* const tmp = (uint8_t*)((uintptr_t)(tmp0 + 15) & ~15);  // align
+  for (n = 0; n < 30; ++n) {   // we directly stomp the *dst memory
+    YuvToRgbSSE2(y[n], u[n], v[n], dst + n * 3);
+  }
+  // Last two pixels are special: we write in a tmp buffer before sending
+  // to dst.
+  YuvToRgbSSE2(y[n + 0], u[n + 0], v[n + 0], tmp + 0);
+  YuvToRgbSSE2(y[n + 1], u[n + 1], v[n + 1], tmp + 3);
+  memcpy(dst + n * 3, tmp, 2 * 3);
+}
+
+void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+                   uint8_t* dst) {
+  int n;
+  uint8_t tmp0[2 * 3 + 5 + 15];
+  uint8_t* const tmp = (uint8_t*)((uintptr_t)(tmp0 + 15) & ~15);  // align
+  for (n = 0; n < 30; ++n) {
+    YuvToBgrSSE2(y[n], u[n], v[n], dst + n * 3);
+  }
+  YuvToBgrSSE2(y[n + 0], u[n + 0], v[n + 0], tmp + 0);
+  YuvToBgrSSE2(y[n + 1], u[n + 1], v[n + 1], tmp + 3);
+  memcpy(dst + n * 3, tmp, 2 * 3);
+}
+
+#endif  // FANCY_UPSAMPLING
+
+//-----------------------------------------------------------------------------
+// Arbitrary-length row conversion functions
+
+static void YuvToRgbaRowSSE2(const uint8_t* y,
+                             const uint8_t* u, const uint8_t* v,
+                             uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 4 <= len; n += 4) {
+    const __m128i uv_0 = LoadUVPart(u[0], v[0]);
+    const __m128i uv_1 = LoadUVPart(u[1], v[1]);
+    const __m128i tmp0_1 = GetRGBA32bWithUV(y[0], uv_0);
+    const __m128i tmp0_2 = GetRGBA32bWithUV(y[1], uv_0);
+    const __m128i tmp0_3 = GetRGBA32bWithUV(y[2], uv_1);
+    const __m128i tmp0_4 = GetRGBA32bWithUV(y[3], uv_1);
+    const __m128i tmp1_1 = _mm_packs_epi32(tmp0_1, tmp0_2);
+    const __m128i tmp1_2 = _mm_packs_epi32(tmp0_3, tmp0_4);
+    const __m128i tmp2 = _mm_packus_epi16(tmp1_1, tmp1_2);
+    _mm_storeu_si128((__m128i*)dst, tmp2);
+    dst += 4 * 4;
+    y += 4;
+    u += 2;
+    v += 2;
+  }
+  // Finish off
+  while (n < len) {
+    VP8YuvToRgba(y[0], u[0], v[0], dst);
+    dst += 4;
+    ++y;
+    u += (n & 1);
+    v += (n & 1);
+    ++n;
+  }
+}
+
+static void YuvToBgraRowSSE2(const uint8_t* y,
+                             const uint8_t* u, const uint8_t* v,
+                             uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 2 <= len; n += 2) {
+    const __m128i uv_0 = LoadUVPart(u[0], v[0]);
+    const __m128i tmp0_1 = GetRGBA32bWithUV(y[0], uv_0);
+    const __m128i tmp0_2 = GetRGBA32bWithUV(y[1], uv_0);
+    const __m128i tmp1_1 = _mm_shuffle_epi32(tmp0_1, _MM_SHUFFLE(3, 0, 1, 2));
+    const __m128i tmp1_2 = _mm_shuffle_epi32(tmp0_2, _MM_SHUFFLE(3, 0, 1, 2));
+    const __m128i tmp2_1 = _mm_packs_epi32(tmp1_1, tmp1_2);
+    const __m128i tmp3 = _mm_packus_epi16(tmp2_1, tmp2_1);
+    _mm_storel_epi64((__m128i*)dst, tmp3);
+    dst += 4 * 2;
+    y += 2;
+    ++u;
+    ++v;
+  }
+  // Finish off
+  if (len & 1) {
+    VP8YuvToBgra(y[0], u[0], v[0], dst);
+  }
+}
+
+static void YuvToArgbRowSSE2(const uint8_t* y,
+                             const uint8_t* u, const uint8_t* v,
+                             uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 2 <= len; n += 2) {
+    const __m128i uv_0 = LoadUVPart(u[0], v[0]);
+    const __m128i tmp0_1 = GetRGBA32bWithUV(y[0], uv_0);
+    const __m128i tmp0_2 = GetRGBA32bWithUV(y[1], uv_0);
+    const __m128i tmp1_1 = _mm_shuffle_epi32(tmp0_1, _MM_SHUFFLE(2, 1, 0, 3));
+    const __m128i tmp1_2 = _mm_shuffle_epi32(tmp0_2, _MM_SHUFFLE(2, 1, 0, 3));
+    const __m128i tmp2_1 = _mm_packs_epi32(tmp1_1, tmp1_2);
+    const __m128i tmp3 = _mm_packus_epi16(tmp2_1, tmp2_1);
+    _mm_storel_epi64((__m128i*)dst, tmp3);
+    dst += 4 * 2;
+    y += 2;
+    ++u;
+    ++v;
+  }
+  // Finish off
+  if (len & 1) {
+    VP8YuvToArgb(y[0], u[0], v[0], dst);
+  }
+}
+
+static void YuvToRgbRowSSE2(const uint8_t* y,
+                            const uint8_t* u, const uint8_t* v,
+                            uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 2 < len; ++n) {   // we directly stomp the *dst memory
+    YuvToRgbSSE2(y[0], u[0], v[0], dst);  // stomps 8 bytes
+    dst += 3;
+    ++y;
+    u += (n & 1);
+    v += (n & 1);
+  }
+  VP8YuvToRgb(y[0], u[0], v[0], dst);
+  if (len > 1) {
+    VP8YuvToRgb(y[1], u[n & 1], v[n & 1], dst + 3);
+  }
+}
+
+static void YuvToBgrRowSSE2(const uint8_t* y,
+                            const uint8_t* u, const uint8_t* v,
+                            uint8_t* dst, int len) {
+  int n;
+  for (n = 0; n + 2 < len; ++n) {   // we directly stomp the *dst memory
+    YuvToBgrSSE2(y[0], u[0], v[0], dst);  // stomps 8 bytes
+    dst += 3;
+    ++y;
+    u += (n & 1);
+    v += (n & 1);
+  }
+  VP8YuvToBgr(y[0], u[0], v[0], dst + 0);
+  if (len > 1) {
+    VP8YuvToBgr(y[1], u[n & 1], v[n & 1], dst + 3);
+  }
+}
+
+#endif  // WEBP_USE_SSE2
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void WebPInitSamplersSSE2(void);
+
+void WebPInitSamplersSSE2(void) {
+#if defined(WEBP_USE_SSE2)
+  WebPSamplers[MODE_RGB]  = YuvToRgbRowSSE2;
+  WebPSamplers[MODE_RGBA] = YuvToRgbaRowSSE2;
+  WebPSamplers[MODE_BGR]  = YuvToBgrRowSSE2;
+  WebPSamplers[MODE_BGRA] = YuvToBgraRowSSE2;
+  WebPSamplers[MODE_ARGB] = YuvToArgbRowSSE2;
+#endif  // WEBP_USE_SSE2
+}

src/dsp/yuv_tables_sse2.h

@@ -0,0 +1,536 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// SSE2 tables for YUV->RGB conversion (12kB overall)
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+// This file is not compiled, but #include'd directly from yuv.c
+// Only used if WEBP_YUV_USE_SSE2_TABLES is defined.
+
+static const VP8kCstSSE2 VP8kYtoRGBA[256] = {
+  {{0xfffb77b0, 0xfffb77b0, 0xfffb77b0, 0x003fc000}},
+  {{0xfffbc235, 0xfffbc235, 0xfffbc235, 0x003fc000}},
+  {{0xfffc0cba, 0xfffc0cba, 0xfffc0cba, 0x003fc000}},
+  {{0xfffc573f, 0xfffc573f, 0xfffc573f, 0x003fc000}},
+  {{0xfffca1c4, 0xfffca1c4, 0xfffca1c4, 0x003fc000}},
+  {{0xfffcec49, 0xfffcec49, 0xfffcec49, 0x003fc000}},
+  {{0xfffd36ce, 0xfffd36ce, 0xfffd36ce, 0x003fc000}},
+  {{0xfffd8153, 0xfffd8153, 0xfffd8153, 0x003fc000}},
+  {{0xfffdcbd8, 0xfffdcbd8, 0xfffdcbd8, 0x003fc000}},
+  {{0xfffe165d, 0xfffe165d, 0xfffe165d, 0x003fc000}},
+  {{0xfffe60e2, 0xfffe60e2, 0xfffe60e2, 0x003fc000}},
+  {{0xfffeab67, 0xfffeab67, 0xfffeab67, 0x003fc000}},
+  {{0xfffef5ec, 0xfffef5ec, 0xfffef5ec, 0x003fc000}},
+  {{0xffff4071, 0xffff4071, 0xffff4071, 0x003fc000}},
+  {{0xffff8af6, 0xffff8af6, 0xffff8af6, 0x003fc000}},
+  {{0xffffd57b, 0xffffd57b, 0xffffd57b, 0x003fc000}},
+  {{0x00002000, 0x00002000, 0x00002000, 0x003fc000}},
+  {{0x00006a85, 0x00006a85, 0x00006a85, 0x003fc000}},
+  {{0x0000b50a, 0x0000b50a, 0x0000b50a, 0x003fc000}},
+  {{0x0000ff8f, 0x0000ff8f, 0x0000ff8f, 0x003fc000}},
+  {{0x00014a14, 0x00014a14, 0x00014a14, 0x003fc000}},
+  {{0x00019499, 0x00019499, 0x00019499, 0x003fc000}},
+  {{0x0001df1e, 0x0001df1e, 0x0001df1e, 0x003fc000}},
+  {{0x000229a3, 0x000229a3, 0x000229a3, 0x003fc000}},
+  {{0x00027428, 0x00027428, 0x00027428, 0x003fc000}},
+  {{0x0002bead, 0x0002bead, 0x0002bead, 0x003fc000}},
+  {{0x00030932, 0x00030932, 0x00030932, 0x003fc000}},
+  {{0x000353b7, 0x000353b7, 0x000353b7, 0x003fc000}},
+  {{0x00039e3c, 0x00039e3c, 0x00039e3c, 0x003fc000}},
+  {{0x0003e8c1, 0x0003e8c1, 0x0003e8c1, 0x003fc000}},
+  {{0x00043346, 0x00043346, 0x00043346, 0x003fc000}},
+  {{0x00047dcb, 0x00047dcb, 0x00047dcb, 0x003fc000}},
+  {{0x0004c850, 0x0004c850, 0x0004c850, 0x003fc000}},
+  {{0x000512d5, 0x000512d5, 0x000512d5, 0x003fc000}},
+  {{0x00055d5a, 0x00055d5a, 0x00055d5a, 0x003fc000}},
+  {{0x0005a7df, 0x0005a7df, 0x0005a7df, 0x003fc000}},
+  {{0x0005f264, 0x0005f264, 0x0005f264, 0x003fc000}},
+  {{0x00063ce9, 0x00063ce9, 0x00063ce9, 0x003fc000}},
+  {{0x0006876e, 0x0006876e, 0x0006876e, 0x003fc000}},
+  {{0x0006d1f3, 0x0006d1f3, 0x0006d1f3, 0x003fc000}},
+  {{0x00071c78, 0x00071c78, 0x00071c78, 0x003fc000}},
+  {{0x000766fd, 0x000766fd, 0x000766fd, 0x003fc000}},
+  {{0x0007b182, 0x0007b182, 0x0007b182, 0x003fc000}},
+  {{0x0007fc07, 0x0007fc07, 0x0007fc07, 0x003fc000}},
+  {{0x0008468c, 0x0008468c, 0x0008468c, 0x003fc000}},
+  {{0x00089111, 0x00089111, 0x00089111, 0x003fc000}},
+  {{0x0008db96, 0x0008db96, 0x0008db96, 0x003fc000}},
+  {{0x0009261b, 0x0009261b, 0x0009261b, 0x003fc000}},
+  {{0x000970a0, 0x000970a0, 0x000970a0, 0x003fc000}},
+  {{0x0009bb25, 0x0009bb25, 0x0009bb25, 0x003fc000}},
+  {{0x000a05aa, 0x000a05aa, 0x000a05aa, 0x003fc000}},
+  {{0x000a502f, 0x000a502f, 0x000a502f, 0x003fc000}},
+  {{0x000a9ab4, 0x000a9ab4, 0x000a9ab4, 0x003fc000}},
+  {{0x000ae539, 0x000ae539, 0x000ae539, 0x003fc000}},
+  {{0x000b2fbe, 0x000b2fbe, 0x000b2fbe, 0x003fc000}},
+  {{0x000b7a43, 0x000b7a43, 0x000b7a43, 0x003fc000}},
+  {{0x000bc4c8, 0x000bc4c8, 0x000bc4c8, 0x003fc000}},
+  {{0x000c0f4d, 0x000c0f4d, 0x000c0f4d, 0x003fc000}},
+  {{0x000c59d2, 0x000c59d2, 0x000c59d2, 0x003fc000}},
+  {{0x000ca457, 0x000ca457, 0x000ca457, 0x003fc000}},
+  {{0x000ceedc, 0x000ceedc, 0x000ceedc, 0x003fc000}},
+  {{0x000d3961, 0x000d3961, 0x000d3961, 0x003fc000}},
+  {{0x000d83e6, 0x000d83e6, 0x000d83e6, 0x003fc000}},
+  {{0x000dce6b, 0x000dce6b, 0x000dce6b, 0x003fc000}},
+  {{0x000e18f0, 0x000e18f0, 0x000e18f0, 0x003fc000}},
+  {{0x000e6375, 0x000e6375, 0x000e6375, 0x003fc000}},
+  {{0x000eadfa, 0x000eadfa, 0x000eadfa, 0x003fc000}},
+  {{0x000ef87f, 0x000ef87f, 0x000ef87f, 0x003fc000}},
+  {{0x000f4304, 0x000f4304, 0x000f4304, 0x003fc000}},
+  {{0x000f8d89, 0x000f8d89, 0x000f8d89, 0x003fc000}},
+  {{0x000fd80e, 0x000fd80e, 0x000fd80e, 0x003fc000}},
+  {{0x00102293, 0x00102293, 0x00102293, 0x003fc000}},
+  {{0x00106d18, 0x00106d18, 0x00106d18, 0x003fc000}},
+  {{0x0010b79d, 0x0010b79d, 0x0010b79d, 0x003fc000}},
+  {{0x00110222, 0x00110222, 0x00110222, 0x003fc000}},
+  {{0x00114ca7, 0x00114ca7, 0x00114ca7, 0x003fc000}},
+  {{0x0011972c, 0x0011972c, 0x0011972c, 0x003fc000}},
+  {{0x0011e1b1, 0x0011e1b1, 0x0011e1b1, 0x003fc000}},
+  {{0x00122c36, 0x00122c36, 0x00122c36, 0x003fc000}},
+  {{0x001276bb, 0x001276bb, 0x001276bb, 0x003fc000}},
+  {{0x0012c140, 0x0012c140, 0x0012c140, 0x003fc000}},
+  {{0x00130bc5, 0x00130bc5, 0x00130bc5, 0x003fc000}},
+  {{0x0013564a, 0x0013564a, 0x0013564a, 0x003fc000}},
+  {{0x0013a0cf, 0x0013a0cf, 0x0013a0cf, 0x003fc000}},
+  {{0x0013eb54, 0x0013eb54, 0x0013eb54, 0x003fc000}},
+  {{0x001435d9, 0x001435d9, 0x001435d9, 0x003fc000}},
+  {{0x0014805e, 0x0014805e, 0x0014805e, 0x003fc000}},
+  {{0x0014cae3, 0x0014cae3, 0x0014cae3, 0x003fc000}},
+  {{0x00151568, 0x00151568, 0x00151568, 0x003fc000}},
+  {{0x00155fed, 0x00155fed, 0x00155fed, 0x003fc000}},
+  {{0x0015aa72, 0x0015aa72, 0x0015aa72, 0x003fc000}},
+  {{0x0015f4f7, 0x0015f4f7, 0x0015f4f7, 0x003fc000}},
+  {{0x00163f7c, 0x00163f7c, 0x00163f7c, 0x003fc000}},
+  {{0x00168a01, 0x00168a01, 0x00168a01, 0x003fc000}},
+  {{0x0016d486, 0x0016d486, 0x0016d486, 0x003fc000}},
+  {{0x00171f0b, 0x00171f0b, 0x00171f0b, 0x003fc000}},
+  {{0x00176990, 0x00176990, 0x00176990, 0x003fc000}},
+  {{0x0017b415, 0x0017b415, 0x0017b415, 0x003fc000}},
+  {{0x0017fe9a, 0x0017fe9a, 0x0017fe9a, 0x003fc000}},
+  {{0x0018491f, 0x0018491f, 0x0018491f, 0x003fc000}},
+  {{0x001893a4, 0x001893a4, 0x001893a4, 0x003fc000}},
+  {{0x0018de29, 0x0018de29, 0x0018de29, 0x003fc000}},
+  {{0x001928ae, 0x001928ae, 0x001928ae, 0x003fc000}},
+  {{0x00197333, 0x00197333, 0x00197333, 0x003fc000}},
+  {{0x0019bdb8, 0x0019bdb8, 0x0019bdb8, 0x003fc000}},
+  {{0x001a083d, 0x001a083d, 0x001a083d, 0x003fc000}},
+  {{0x001a52c2, 0x001a52c2, 0x001a52c2, 0x003fc000}},
+  {{0x001a9d47, 0x001a9d47, 0x001a9d47, 0x003fc000}},
+  {{0x001ae7cc, 0x001ae7cc, 0x001ae7cc, 0x003fc000}},
+  {{0x001b3251, 0x001b3251, 0x001b3251, 0x003fc000}},
+  {{0x001b7cd6, 0x001b7cd6, 0x001b7cd6, 0x003fc000}},
+  {{0x001bc75b, 0x001bc75b, 0x001bc75b, 0x003fc000}},
+  {{0x001c11e0, 0x001c11e0, 0x001c11e0, 0x003fc000}},
+  {{0x001c5c65, 0x001c5c65, 0x001c5c65, 0x003fc000}},
+  {{0x001ca6ea, 0x001ca6ea, 0x001ca6ea, 0x003fc000}},
+  {{0x001cf16f, 0x001cf16f, 0x001cf16f, 0x003fc000}},
+  {{0x001d3bf4, 0x001d3bf4, 0x001d3bf4, 0x003fc000}},
+  {{0x001d8679, 0x001d8679, 0x001d8679, 0x003fc000}},
+  {{0x001dd0fe, 0x001dd0fe, 0x001dd0fe, 0x003fc000}},
+  {{0x001e1b83, 0x001e1b83, 0x001e1b83, 0x003fc000}},
+  {{0x001e6608, 0x001e6608, 0x001e6608, 0x003fc000}},
+  {{0x001eb08d, 0x001eb08d, 0x001eb08d, 0x003fc000}},
+  {{0x001efb12, 0x001efb12, 0x001efb12, 0x003fc000}},
+  {{0x001f4597, 0x001f4597, 0x001f4597, 0x003fc000}},
+  {{0x001f901c, 0x001f901c, 0x001f901c, 0x003fc000}},
+  {{0x001fdaa1, 0x001fdaa1, 0x001fdaa1, 0x003fc000}},
+  {{0x00202526, 0x00202526, 0x00202526, 0x003fc000}},
+  {{0x00206fab, 0x00206fab, 0x00206fab, 0x003fc000}},
+  {{0x0020ba30, 0x0020ba30, 0x0020ba30, 0x003fc000}},
+  {{0x002104b5, 0x002104b5, 0x002104b5, 0x003fc000}},
+  {{0x00214f3a, 0x00214f3a, 0x00214f3a, 0x003fc000}},
+  {{0x002199bf, 0x002199bf, 0x002199bf, 0x003fc000}},
+  {{0x0021e444, 0x0021e444, 0x0021e444, 0x003fc000}},
+  {{0x00222ec9, 0x00222ec9, 0x00222ec9, 0x003fc000}},
+  {{0x0022794e, 0x0022794e, 0x0022794e, 0x003fc000}},
+  {{0x0022c3d3, 0x0022c3d3, 0x0022c3d3, 0x003fc000}},
+  {{0x00230e58, 0x00230e58, 0x00230e58, 0x003fc000}},
+  {{0x002358dd, 0x002358dd, 0x002358dd, 0x003fc000}},
+  {{0x0023a362, 0x0023a362, 0x0023a362, 0x003fc000}},
+  {{0x0023ede7, 0x0023ede7, 0x0023ede7, 0x003fc000}},
+  {{0x0024386c, 0x0024386c, 0x0024386c, 0x003fc000}},
+  {{0x002482f1, 0x002482f1, 0x002482f1, 0x003fc000}},
+  {{0x0024cd76, 0x0024cd76, 0x0024cd76, 0x003fc000}},
+  {{0x002517fb, 0x002517fb, 0x002517fb, 0x003fc000}},
+  {{0x00256280, 0x00256280, 0x00256280, 0x003fc000}},
+  {{0x0025ad05, 0x0025ad05, 0x0025ad05, 0x003fc000}},
+  {{0x0025f78a, 0x0025f78a, 0x0025f78a, 0x003fc000}},
+  {{0x0026420f, 0x0026420f, 0x0026420f, 0x003fc000}},
+  {{0x00268c94, 0x00268c94, 0x00268c94, 0x003fc000}},
+  {{0x0026d719, 0x0026d719, 0x0026d719, 0x003fc000}},
+  {{0x0027219e, 0x0027219e, 0x0027219e, 0x003fc000}},
+  {{0x00276c23, 0x00276c23, 0x00276c23, 0x003fc000}},
+  {{0x0027b6a8, 0x0027b6a8, 0x0027b6a8, 0x003fc000}},
+  {{0x0028012d, 0x0028012d, 0x0028012d, 0x003fc000}},
+  {{0x00284bb2, 0x00284bb2, 0x00284bb2, 0x003fc000}},
+  {{0x00289637, 0x00289637, 0x00289637, 0x003fc000}},
+  {{0x0028e0bc, 0x0028e0bc, 0x0028e0bc, 0x003fc000}},
+  {{0x00292b41, 0x00292b41, 0x00292b41, 0x003fc000}},
+  {{0x002975c6, 0x002975c6, 0x002975c6, 0x003fc000}},
+  {{0x0029c04b, 0x0029c04b, 0x0029c04b, 0x003fc000}},
+  {{0x002a0ad0, 0x002a0ad0, 0x002a0ad0, 0x003fc000}},
+  {{0x002a5555, 0x002a5555, 0x002a5555, 0x003fc000}},
+  {{0x002a9fda, 0x002a9fda, 0x002a9fda, 0x003fc000}},
+  {{0x002aea5f, 0x002aea5f, 0x002aea5f, 0x003fc000}},
+  {{0x002b34e4, 0x002b34e4, 0x002b34e4, 0x003fc000}},
+  {{0x002b7f69, 0x002b7f69, 0x002b7f69, 0x003fc000}},
+  {{0x002bc9ee, 0x002bc9ee, 0x002bc9ee, 0x003fc000}},
+  {{0x002c1473, 0x002c1473, 0x002c1473, 0x003fc000}},
+  {{0x002c5ef8, 0x002c5ef8, 0x002c5ef8, 0x003fc000}},
+  {{0x002ca97d, 0x002ca97d, 0x002ca97d, 0x003fc000}},
+  {{0x002cf402, 0x002cf402, 0x002cf402, 0x003fc000}},
+  {{0x002d3e87, 0x002d3e87, 0x002d3e87, 0x003fc000}},
+  {{0x002d890c, 0x002d890c, 0x002d890c, 0x003fc000}},
+  {{0x002dd391, 0x002dd391, 0x002dd391, 0x003fc000}},
+  {{0x002e1e16, 0x002e1e16, 0x002e1e16, 0x003fc000}},
+  {{0x002e689b, 0x002e689b, 0x002e689b, 0x003fc000}},
+  {{0x002eb320, 0x002eb320, 0x002eb320, 0x003fc000}},
+  {{0x002efda5, 0x002efda5, 0x002efda5, 0x003fc000}},
+  {{0x002f482a, 0x002f482a, 0x002f482a, 0x003fc000}},
+  {{0x002f92af, 0x002f92af, 0x002f92af, 0x003fc000}},
+  {{0x002fdd34, 0x002fdd34, 0x002fdd34, 0x003fc000}},
+  {{0x003027b9, 0x003027b9, 0x003027b9, 0x003fc000}},
+  {{0x0030723e, 0x0030723e, 0x0030723e, 0x003fc000}},
+  {{0x0030bcc3, 0x0030bcc3, 0x0030bcc3, 0x003fc000}},
+  {{0x00310748, 0x00310748, 0x00310748, 0x003fc000}},
+  {{0x003151cd, 0x003151cd, 0x003151cd, 0x003fc000}},
+  {{0x00319c52, 0x00319c52, 0x00319c52, 0x003fc000}},
+  {{0x0031e6d7, 0x0031e6d7, 0x0031e6d7, 0x003fc000}},
+  {{0x0032315c, 0x0032315c, 0x0032315c, 0x003fc000}},
+  {{0x00327be1, 0x00327be1, 0x00327be1, 0x003fc000}},
+  {{0x0032c666, 0x0032c666, 0x0032c666, 0x003fc000}},
+  {{0x003310eb, 0x003310eb, 0x003310eb, 0x003fc000}},
+  {{0x00335b70, 0x00335b70, 0x00335b70, 0x003fc000}},
+  {{0x0033a5f5, 0x0033a5f5, 0x0033a5f5, 0x003fc000}},
+  {{0x0033f07a, 0x0033f07a, 0x0033f07a, 0x003fc000}},
+  {{0x00343aff, 0x00343aff, 0x00343aff, 0x003fc000}},
+  {{0x00348584, 0x00348584, 0x00348584, 0x003fc000}},
+  {{0x0034d009, 0x0034d009, 0x0034d009, 0x003fc000}},
+  {{0x00351a8e, 0x00351a8e, 0x00351a8e, 0x003fc000}},
+  {{0x00356513, 0x00356513, 0x00356513, 0x003fc000}},
+  {{0x0035af98, 0x0035af98, 0x0035af98, 0x003fc000}},
+  {{0x0035fa1d, 0x0035fa1d, 0x0035fa1d, 0x003fc000}},
+  {{0x003644a2, 0x003644a2, 0x003644a2, 0x003fc000}},
+  {{0x00368f27, 0x00368f27, 0x00368f27, 0x003fc000}},
+  {{0x0036d9ac, 0x0036d9ac, 0x0036d9ac, 0x003fc000}},
+  {{0x00372431, 0x00372431, 0x00372431, 0x003fc000}},
+  {{0x00376eb6, 0x00376eb6, 0x00376eb6, 0x003fc000}},
+  {{0x0037b93b, 0x0037b93b, 0x0037b93b, 0x003fc000}},
+  {{0x003803c0, 0x003803c0, 0x003803c0, 0x003fc000}},
+  {{0x00384e45, 0x00384e45, 0x00384e45, 0x003fc000}},
+  {{0x003898ca, 0x003898ca, 0x003898ca, 0x003fc000}},
+  {{0x0038e34f, 0x0038e34f, 0x0038e34f, 0x003fc000}},
+  {{0x00392dd4, 0x00392dd4, 0x00392dd4, 0x003fc000}},
+  {{0x00397859, 0x00397859, 0x00397859, 0x003fc000}},
+  {{0x0039c2de, 0x0039c2de, 0x0039c2de, 0x003fc000}},
+  {{0x003a0d63, 0x003a0d63, 0x003a0d63, 0x003fc000}},
+  {{0x003a57e8, 0x003a57e8, 0x003a57e8, 0x003fc000}},
+  {{0x003aa26d, 0x003aa26d, 0x003aa26d, 0x003fc000}},
+  {{0x003aecf2, 0x003aecf2, 0x003aecf2, 0x003fc000}},
+  {{0x003b3777, 0x003b3777, 0x003b3777, 0x003fc000}},
+  {{0x003b81fc, 0x003b81fc, 0x003b81fc, 0x003fc000}},
+  {{0x003bcc81, 0x003bcc81, 0x003bcc81, 0x003fc000}},
+  {{0x003c1706, 0x003c1706, 0x003c1706, 0x003fc000}},
+  {{0x003c618b, 0x003c618b, 0x003c618b, 0x003fc000}},
+  {{0x003cac10, 0x003cac10, 0x003cac10, 0x003fc000}},
+  {{0x003cf695, 0x003cf695, 0x003cf695, 0x003fc000}},
+  {{0x003d411a, 0x003d411a, 0x003d411a, 0x003fc000}},
+  {{0x003d8b9f, 0x003d8b9f, 0x003d8b9f, 0x003fc000}},
+  {{0x003dd624, 0x003dd624, 0x003dd624, 0x003fc000}},
+  {{0x003e20a9, 0x003e20a9, 0x003e20a9, 0x003fc000}},
+  {{0x003e6b2e, 0x003e6b2e, 0x003e6b2e, 0x003fc000}},
+  {{0x003eb5b3, 0x003eb5b3, 0x003eb5b3, 0x003fc000}},
+  {{0x003f0038, 0x003f0038, 0x003f0038, 0x003fc000}},
+  {{0x003f4abd, 0x003f4abd, 0x003f4abd, 0x003fc000}},
+  {{0x003f9542, 0x003f9542, 0x003f9542, 0x003fc000}},
+  {{0x003fdfc7, 0x003fdfc7, 0x003fdfc7, 0x003fc000}},
+  {{0x00402a4c, 0x00402a4c, 0x00402a4c, 0x003fc000}},
+  {{0x004074d1, 0x004074d1, 0x004074d1, 0x003fc000}},
+  {{0x0040bf56, 0x0040bf56, 0x0040bf56, 0x003fc000}},
+  {{0x004109db, 0x004109db, 0x004109db, 0x003fc000}},
+  {{0x00415460, 0x00415460, 0x00415460, 0x003fc000}},
+  {{0x00419ee5, 0x00419ee5, 0x00419ee5, 0x003fc000}},
+  {{0x0041e96a, 0x0041e96a, 0x0041e96a, 0x003fc000}},
+  {{0x004233ef, 0x004233ef, 0x004233ef, 0x003fc000}},
+  {{0x00427e74, 0x00427e74, 0x00427e74, 0x003fc000}},
+  {{0x0042c8f9, 0x0042c8f9, 0x0042c8f9, 0x003fc000}},
+  {{0x0043137e, 0x0043137e, 0x0043137e, 0x003fc000}},
+  {{0x00435e03, 0x00435e03, 0x00435e03, 0x003fc000}},
+  {{0x0043a888, 0x0043a888, 0x0043a888, 0x003fc000}},
+  {{0x0043f30d, 0x0043f30d, 0x0043f30d, 0x003fc000}},
+  {{0x00443d92, 0x00443d92, 0x00443d92, 0x003fc000}},
+  {{0x00448817, 0x00448817, 0x00448817, 0x003fc000}},
+  {{0x0044d29c, 0x0044d29c, 0x0044d29c, 0x003fc000}},
+  {{0x00451d21, 0x00451d21, 0x00451d21, 0x003fc000}},
+  {{0x004567a6, 0x004567a6, 0x004567a6, 0x003fc000}},
+  {{0x0045b22b, 0x0045b22b, 0x0045b22b, 0x003fc000}}
+};
+
+static const VP8kCstSSE2 VP8kUtoRGBA[256] = {
+  {{0, 0x000c8980, 0xffbf7300, 0}}, {{0, 0x000c706d, 0xffbff41a, 0}},
+  {{0, 0x000c575a, 0xffc07534, 0}}, {{0, 0x000c3e47, 0xffc0f64e, 0}},
+  {{0, 0x000c2534, 0xffc17768, 0}}, {{0, 0x000c0c21, 0xffc1f882, 0}},
+  {{0, 0x000bf30e, 0xffc2799c, 0}}, {{0, 0x000bd9fb, 0xffc2fab6, 0}},
+  {{0, 0x000bc0e8, 0xffc37bd0, 0}}, {{0, 0x000ba7d5, 0xffc3fcea, 0}},
+  {{0, 0x000b8ec2, 0xffc47e04, 0}}, {{0, 0x000b75af, 0xffc4ff1e, 0}},
+  {{0, 0x000b5c9c, 0xffc58038, 0}}, {{0, 0x000b4389, 0xffc60152, 0}},
+  {{0, 0x000b2a76, 0xffc6826c, 0}}, {{0, 0x000b1163, 0xffc70386, 0}},
+  {{0, 0x000af850, 0xffc784a0, 0}}, {{0, 0x000adf3d, 0xffc805ba, 0}},
+  {{0, 0x000ac62a, 0xffc886d4, 0}}, {{0, 0x000aad17, 0xffc907ee, 0}},
+  {{0, 0x000a9404, 0xffc98908, 0}}, {{0, 0x000a7af1, 0xffca0a22, 0}},
+  {{0, 0x000a61de, 0xffca8b3c, 0}}, {{0, 0x000a48cb, 0xffcb0c56, 0}},
+  {{0, 0x000a2fb8, 0xffcb8d70, 0}}, {{0, 0x000a16a5, 0xffcc0e8a, 0}},
+  {{0, 0x0009fd92, 0xffcc8fa4, 0}}, {{0, 0x0009e47f, 0xffcd10be, 0}},
+  {{0, 0x0009cb6c, 0xffcd91d8, 0}}, {{0, 0x0009b259, 0xffce12f2, 0}},
+  {{0, 0x00099946, 0xffce940c, 0}}, {{0, 0x00098033, 0xffcf1526, 0}},
+  {{0, 0x00096720, 0xffcf9640, 0}}, {{0, 0x00094e0d, 0xffd0175a, 0}},
+  {{0, 0x000934fa, 0xffd09874, 0}}, {{0, 0x00091be7, 0xffd1198e, 0}},
+  {{0, 0x000902d4, 0xffd19aa8, 0}}, {{0, 0x0008e9c1, 0xffd21bc2, 0}},
+  {{0, 0x0008d0ae, 0xffd29cdc, 0}}, {{0, 0x0008b79b, 0xffd31df6, 0}},
+  {{0, 0x00089e88, 0xffd39f10, 0}}, {{0, 0x00088575, 0xffd4202a, 0}},
+  {{0, 0x00086c62, 0xffd4a144, 0}}, {{0, 0x0008534f, 0xffd5225e, 0}},
+  {{0, 0x00083a3c, 0xffd5a378, 0}}, {{0, 0x00082129, 0xffd62492, 0}},
+  {{0, 0x00080816, 0xffd6a5ac, 0}}, {{0, 0x0007ef03, 0xffd726c6, 0}},
+  {{0, 0x0007d5f0, 0xffd7a7e0, 0}}, {{0, 0x0007bcdd, 0xffd828fa, 0}},
+  {{0, 0x0007a3ca, 0xffd8aa14, 0}}, {{0, 0x00078ab7, 0xffd92b2e, 0}},
+  {{0, 0x000771a4, 0xffd9ac48, 0}}, {{0, 0x00075891, 0xffda2d62, 0}},
+  {{0, 0x00073f7e, 0xffdaae7c, 0}}, {{0, 0x0007266b, 0xffdb2f96, 0}},
+  {{0, 0x00070d58, 0xffdbb0b0, 0}}, {{0, 0x0006f445, 0xffdc31ca, 0}},
+  {{0, 0x0006db32, 0xffdcb2e4, 0}}, {{0, 0x0006c21f, 0xffdd33fe, 0}},
+  {{0, 0x0006a90c, 0xffddb518, 0}}, {{0, 0x00068ff9, 0xffde3632, 0}},
+  {{0, 0x000676e6, 0xffdeb74c, 0}}, {{0, 0x00065dd3, 0xffdf3866, 0}},
+  {{0, 0x000644c0, 0xffdfb980, 0}}, {{0, 0x00062bad, 0xffe03a9a, 0}},
+  {{0, 0x0006129a, 0xffe0bbb4, 0}}, {{0, 0x0005f987, 0xffe13cce, 0}},
+  {{0, 0x0005e074, 0xffe1bde8, 0}}, {{0, 0x0005c761, 0xffe23f02, 0}},
+  {{0, 0x0005ae4e, 0xffe2c01c, 0}}, {{0, 0x0005953b, 0xffe34136, 0}},
+  {{0, 0x00057c28, 0xffe3c250, 0}}, {{0, 0x00056315, 0xffe4436a, 0}},
+  {{0, 0x00054a02, 0xffe4c484, 0}}, {{0, 0x000530ef, 0xffe5459e, 0}},
+  {{0, 0x000517dc, 0xffe5c6b8, 0}}, {{0, 0x0004fec9, 0xffe647d2, 0}},
+  {{0, 0x0004e5b6, 0xffe6c8ec, 0}}, {{0, 0x0004cca3, 0xffe74a06, 0}},
+  {{0, 0x0004b390, 0xffe7cb20, 0}}, {{0, 0x00049a7d, 0xffe84c3a, 0}},
+  {{0, 0x0004816a, 0xffe8cd54, 0}}, {{0, 0x00046857, 0xffe94e6e, 0}},
+  {{0, 0x00044f44, 0xffe9cf88, 0}}, {{0, 0x00043631, 0xffea50a2, 0}},
+  {{0, 0x00041d1e, 0xffead1bc, 0}}, {{0, 0x0004040b, 0xffeb52d6, 0}},
+  {{0, 0x0003eaf8, 0xffebd3f0, 0}}, {{0, 0x0003d1e5, 0xffec550a, 0}},
+  {{0, 0x0003b8d2, 0xffecd624, 0}}, {{0, 0x00039fbf, 0xffed573e, 0}},
+  {{0, 0x000386ac, 0xffedd858, 0}}, {{0, 0x00036d99, 0xffee5972, 0}},
+  {{0, 0x00035486, 0xffeeda8c, 0}}, {{0, 0x00033b73, 0xffef5ba6, 0}},
+  {{0, 0x00032260, 0xffefdcc0, 0}}, {{0, 0x0003094d, 0xfff05dda, 0}},
+  {{0, 0x0002f03a, 0xfff0def4, 0}}, {{0, 0x0002d727, 0xfff1600e, 0}},
+  {{0, 0x0002be14, 0xfff1e128, 0}}, {{0, 0x0002a501, 0xfff26242, 0}},
+  {{0, 0x00028bee, 0xfff2e35c, 0}}, {{0, 0x000272db, 0xfff36476, 0}},
+  {{0, 0x000259c8, 0xfff3e590, 0}}, {{0, 0x000240b5, 0xfff466aa, 0}},
+  {{0, 0x000227a2, 0xfff4e7c4, 0}}, {{0, 0x00020e8f, 0xfff568de, 0}},
+  {{0, 0x0001f57c, 0xfff5e9f8, 0}}, {{0, 0x0001dc69, 0xfff66b12, 0}},
+  {{0, 0x0001c356, 0xfff6ec2c, 0}}, {{0, 0x0001aa43, 0xfff76d46, 0}},
+  {{0, 0x00019130, 0xfff7ee60, 0}}, {{0, 0x0001781d, 0xfff86f7a, 0}},
+  {{0, 0x00015f0a, 0xfff8f094, 0}}, {{0, 0x000145f7, 0xfff971ae, 0}},
+  {{0, 0x00012ce4, 0xfff9f2c8, 0}}, {{0, 0x000113d1, 0xfffa73e2, 0}},
+  {{0, 0x0000fabe, 0xfffaf4fc, 0}}, {{0, 0x0000e1ab, 0xfffb7616, 0}},
+  {{0, 0x0000c898, 0xfffbf730, 0}}, {{0, 0x0000af85, 0xfffc784a, 0}},
+  {{0, 0x00009672, 0xfffcf964, 0}}, {{0, 0x00007d5f, 0xfffd7a7e, 0}},
+  {{0, 0x0000644c, 0xfffdfb98, 0}}, {{0, 0x00004b39, 0xfffe7cb2, 0}},
+  {{0, 0x00003226, 0xfffefdcc, 0}}, {{0, 0x00001913, 0xffff7ee6, 0}},
+  {{0, 0x00000000, 0x00000000, 0}}, {{0, 0xffffe6ed, 0x0000811a, 0}},
+  {{0, 0xffffcdda, 0x00010234, 0}}, {{0, 0xffffb4c7, 0x0001834e, 0}},
+  {{0, 0xffff9bb4, 0x00020468, 0}}, {{0, 0xffff82a1, 0x00028582, 0}},
+  {{0, 0xffff698e, 0x0003069c, 0}}, {{0, 0xffff507b, 0x000387b6, 0}},
+  {{0, 0xffff3768, 0x000408d0, 0}}, {{0, 0xffff1e55, 0x000489ea, 0}},
+  {{0, 0xffff0542, 0x00050b04, 0}}, {{0, 0xfffeec2f, 0x00058c1e, 0}},
+  {{0, 0xfffed31c, 0x00060d38, 0}}, {{0, 0xfffeba09, 0x00068e52, 0}},
+  {{0, 0xfffea0f6, 0x00070f6c, 0}}, {{0, 0xfffe87e3, 0x00079086, 0}},
+  {{0, 0xfffe6ed0, 0x000811a0, 0}}, {{0, 0xfffe55bd, 0x000892ba, 0}},
+  {{0, 0xfffe3caa, 0x000913d4, 0}}, {{0, 0xfffe2397, 0x000994ee, 0}},
+  {{0, 0xfffe0a84, 0x000a1608, 0}}, {{0, 0xfffdf171, 0x000a9722, 0}},
+  {{0, 0xfffdd85e, 0x000b183c, 0}}, {{0, 0xfffdbf4b, 0x000b9956, 0}},
+  {{0, 0xfffda638, 0x000c1a70, 0}}, {{0, 0xfffd8d25, 0x000c9b8a, 0}},
+  {{0, 0xfffd7412, 0x000d1ca4, 0}}, {{0, 0xfffd5aff, 0x000d9dbe, 0}},
+  {{0, 0xfffd41ec, 0x000e1ed8, 0}}, {{0, 0xfffd28d9, 0x000e9ff2, 0}},
+  {{0, 0xfffd0fc6, 0x000f210c, 0}}, {{0, 0xfffcf6b3, 0x000fa226, 0}},
+  {{0, 0xfffcdda0, 0x00102340, 0}}, {{0, 0xfffcc48d, 0x0010a45a, 0}},
+  {{0, 0xfffcab7a, 0x00112574, 0}}, {{0, 0xfffc9267, 0x0011a68e, 0}},
+  {{0, 0xfffc7954, 0x001227a8, 0}}, {{0, 0xfffc6041, 0x0012a8c2, 0}},
+  {{0, 0xfffc472e, 0x001329dc, 0}}, {{0, 0xfffc2e1b, 0x0013aaf6, 0}},
+  {{0, 0xfffc1508, 0x00142c10, 0}}, {{0, 0xfffbfbf5, 0x0014ad2a, 0}},
+  {{0, 0xfffbe2e2, 0x00152e44, 0}}, {{0, 0xfffbc9cf, 0x0015af5e, 0}},
+  {{0, 0xfffbb0bc, 0x00163078, 0}}, {{0, 0xfffb97a9, 0x0016b192, 0}},
+  {{0, 0xfffb7e96, 0x001732ac, 0}}, {{0, 0xfffb6583, 0x0017b3c6, 0}},
+  {{0, 0xfffb4c70, 0x001834e0, 0}}, {{0, 0xfffb335d, 0x0018b5fa, 0}},
+  {{0, 0xfffb1a4a, 0x00193714, 0}}, {{0, 0xfffb0137, 0x0019b82e, 0}},
+  {{0, 0xfffae824, 0x001a3948, 0}}, {{0, 0xfffacf11, 0x001aba62, 0}},
+  {{0, 0xfffab5fe, 0x001b3b7c, 0}}, {{0, 0xfffa9ceb, 0x001bbc96, 0}},
+  {{0, 0xfffa83d8, 0x001c3db0, 0}}, {{0, 0xfffa6ac5, 0x001cbeca, 0}},
+  {{0, 0xfffa51b2, 0x001d3fe4, 0}}, {{0, 0xfffa389f, 0x001dc0fe, 0}},
+  {{0, 0xfffa1f8c, 0x001e4218, 0}}, {{0, 0xfffa0679, 0x001ec332, 0}},
+  {{0, 0xfff9ed66, 0x001f444c, 0}}, {{0, 0xfff9d453, 0x001fc566, 0}},
+  {{0, 0xfff9bb40, 0x00204680, 0}}, {{0, 0xfff9a22d, 0x0020c79a, 0}},
+  {{0, 0xfff9891a, 0x002148b4, 0}}, {{0, 0xfff97007, 0x0021c9ce, 0}},
+  {{0, 0xfff956f4, 0x00224ae8, 0}}, {{0, 0xfff93de1, 0x0022cc02, 0}},
+  {{0, 0xfff924ce, 0x00234d1c, 0}}, {{0, 0xfff90bbb, 0x0023ce36, 0}},
+  {{0, 0xfff8f2a8, 0x00244f50, 0}}, {{0, 0xfff8d995, 0x0024d06a, 0}},
+  {{0, 0xfff8c082, 0x00255184, 0}}, {{0, 0xfff8a76f, 0x0025d29e, 0}},
+  {{0, 0xfff88e5c, 0x002653b8, 0}}, {{0, 0xfff87549, 0x0026d4d2, 0}},
+  {{0, 0xfff85c36, 0x002755ec, 0}}, {{0, 0xfff84323, 0x0027d706, 0}},
+  {{0, 0xfff82a10, 0x00285820, 0}}, {{0, 0xfff810fd, 0x0028d93a, 0}},
+  {{0, 0xfff7f7ea, 0x00295a54, 0}}, {{0, 0xfff7ded7, 0x0029db6e, 0}},
+  {{0, 0xfff7c5c4, 0x002a5c88, 0}}, {{0, 0xfff7acb1, 0x002adda2, 0}},
+  {{0, 0xfff7939e, 0x002b5ebc, 0}}, {{0, 0xfff77a8b, 0x002bdfd6, 0}},
+  {{0, 0xfff76178, 0x002c60f0, 0}}, {{0, 0xfff74865, 0x002ce20a, 0}},
+  {{0, 0xfff72f52, 0x002d6324, 0}}, {{0, 0xfff7163f, 0x002de43e, 0}},
+  {{0, 0xfff6fd2c, 0x002e6558, 0}}, {{0, 0xfff6e419, 0x002ee672, 0}},
+  {{0, 0xfff6cb06, 0x002f678c, 0}}, {{0, 0xfff6b1f3, 0x002fe8a6, 0}},
+  {{0, 0xfff698e0, 0x003069c0, 0}}, {{0, 0xfff67fcd, 0x0030eada, 0}},
+  {{0, 0xfff666ba, 0x00316bf4, 0}}, {{0, 0xfff64da7, 0x0031ed0e, 0}},
+  {{0, 0xfff63494, 0x00326e28, 0}}, {{0, 0xfff61b81, 0x0032ef42, 0}},
+  {{0, 0xfff6026e, 0x0033705c, 0}}, {{0, 0xfff5e95b, 0x0033f176, 0}},
+  {{0, 0xfff5d048, 0x00347290, 0}}, {{0, 0xfff5b735, 0x0034f3aa, 0}},
+  {{0, 0xfff59e22, 0x003574c4, 0}}, {{0, 0xfff5850f, 0x0035f5de, 0}},
+  {{0, 0xfff56bfc, 0x003676f8, 0}}, {{0, 0xfff552e9, 0x0036f812, 0}},
+  {{0, 0xfff539d6, 0x0037792c, 0}}, {{0, 0xfff520c3, 0x0037fa46, 0}},
+  {{0, 0xfff507b0, 0x00387b60, 0}}, {{0, 0xfff4ee9d, 0x0038fc7a, 0}},
+  {{0, 0xfff4d58a, 0x00397d94, 0}}, {{0, 0xfff4bc77, 0x0039feae, 0}},
+  {{0, 0xfff4a364, 0x003a7fc8, 0}}, {{0, 0xfff48a51, 0x003b00e2, 0}},
+  {{0, 0xfff4713e, 0x003b81fc, 0}}, {{0, 0xfff4582b, 0x003c0316, 0}},
+  {{0, 0xfff43f18, 0x003c8430, 0}}, {{0, 0xfff42605, 0x003d054a, 0}},
+  {{0, 0xfff40cf2, 0x003d8664, 0}}, {{0, 0xfff3f3df, 0x003e077e, 0}},
+  {{0, 0xfff3dacc, 0x003e8898, 0}}, {{0, 0xfff3c1b9, 0x003f09b2, 0}},
+  {{0, 0xfff3a8a6, 0x003f8acc, 0}}, {{0, 0xfff38f93, 0x00400be6, 0}}
+};
+
+static VP8kCstSSE2 VP8kVtoRGBA[256] = {
+  {{0xffcced80, 0x001a0400, 0, 0}}, {{0xffcd53a5, 0x0019cff8, 0, 0}},
+  {{0xffcdb9ca, 0x00199bf0, 0, 0}}, {{0xffce1fef, 0x001967e8, 0, 0}},
+  {{0xffce8614, 0x001933e0, 0, 0}}, {{0xffceec39, 0x0018ffd8, 0, 0}},
+  {{0xffcf525e, 0x0018cbd0, 0, 0}}, {{0xffcfb883, 0x001897c8, 0, 0}},
+  {{0xffd01ea8, 0x001863c0, 0, 0}}, {{0xffd084cd, 0x00182fb8, 0, 0}},
+  {{0xffd0eaf2, 0x0017fbb0, 0, 0}}, {{0xffd15117, 0x0017c7a8, 0, 0}},
+  {{0xffd1b73c, 0x001793a0, 0, 0}}, {{0xffd21d61, 0x00175f98, 0, 0}},
+  {{0xffd28386, 0x00172b90, 0, 0}}, {{0xffd2e9ab, 0x0016f788, 0, 0}},
+  {{0xffd34fd0, 0x0016c380, 0, 0}}, {{0xffd3b5f5, 0x00168f78, 0, 0}},
+  {{0xffd41c1a, 0x00165b70, 0, 0}}, {{0xffd4823f, 0x00162768, 0, 0}},
+  {{0xffd4e864, 0x0015f360, 0, 0}}, {{0xffd54e89, 0x0015bf58, 0, 0}},
+  {{0xffd5b4ae, 0x00158b50, 0, 0}}, {{0xffd61ad3, 0x00155748, 0, 0}},
+  {{0xffd680f8, 0x00152340, 0, 0}}, {{0xffd6e71d, 0x0014ef38, 0, 0}},
+  {{0xffd74d42, 0x0014bb30, 0, 0}}, {{0xffd7b367, 0x00148728, 0, 0}},
+  {{0xffd8198c, 0x00145320, 0, 0}}, {{0xffd87fb1, 0x00141f18, 0, 0}},
+  {{0xffd8e5d6, 0x0013eb10, 0, 0}}, {{0xffd94bfb, 0x0013b708, 0, 0}},
+  {{0xffd9b220, 0x00138300, 0, 0}}, {{0xffda1845, 0x00134ef8, 0, 0}},
+  {{0xffda7e6a, 0x00131af0, 0, 0}}, {{0xffdae48f, 0x0012e6e8, 0, 0}},
+  {{0xffdb4ab4, 0x0012b2e0, 0, 0}}, {{0xffdbb0d9, 0x00127ed8, 0, 0}},
+  {{0xffdc16fe, 0x00124ad0, 0, 0}}, {{0xffdc7d23, 0x001216c8, 0, 0}},
+  {{0xffdce348, 0x0011e2c0, 0, 0}}, {{0xffdd496d, 0x0011aeb8, 0, 0}},
+  {{0xffddaf92, 0x00117ab0, 0, 0}}, {{0xffde15b7, 0x001146a8, 0, 0}},
+  {{0xffde7bdc, 0x001112a0, 0, 0}}, {{0xffdee201, 0x0010de98, 0, 0}},
+  {{0xffdf4826, 0x0010aa90, 0, 0}}, {{0xffdfae4b, 0x00107688, 0, 0}},
+  {{0xffe01470, 0x00104280, 0, 0}}, {{0xffe07a95, 0x00100e78, 0, 0}},
+  {{0xffe0e0ba, 0x000fda70, 0, 0}}, {{0xffe146df, 0x000fa668, 0, 0}},
+  {{0xffe1ad04, 0x000f7260, 0, 0}}, {{0xffe21329, 0x000f3e58, 0, 0}},
+  {{0xffe2794e, 0x000f0a50, 0, 0}}, {{0xffe2df73, 0x000ed648, 0, 0}},
+  {{0xffe34598, 0x000ea240, 0, 0}}, {{0xffe3abbd, 0x000e6e38, 0, 0}},
+  {{0xffe411e2, 0x000e3a30, 0, 0}}, {{0xffe47807, 0x000e0628, 0, 0}},
+  {{0xffe4de2c, 0x000dd220, 0, 0}}, {{0xffe54451, 0x000d9e18, 0, 0}},
+  {{0xffe5aa76, 0x000d6a10, 0, 0}}, {{0xffe6109b, 0x000d3608, 0, 0}},
+  {{0xffe676c0, 0x000d0200, 0, 0}}, {{0xffe6dce5, 0x000ccdf8, 0, 0}},
+  {{0xffe7430a, 0x000c99f0, 0, 0}}, {{0xffe7a92f, 0x000c65e8, 0, 0}},
+  {{0xffe80f54, 0x000c31e0, 0, 0}}, {{0xffe87579, 0x000bfdd8, 0, 0}},
+  {{0xffe8db9e, 0x000bc9d0, 0, 0}}, {{0xffe941c3, 0x000b95c8, 0, 0}},
+  {{0xffe9a7e8, 0x000b61c0, 0, 0}}, {{0xffea0e0d, 0x000b2db8, 0, 0}},
+  {{0xffea7432, 0x000af9b0, 0, 0}}, {{0xffeada57, 0x000ac5a8, 0, 0}},
+  {{0xffeb407c, 0x000a91a0, 0, 0}}, {{0xffeba6a1, 0x000a5d98, 0, 0}},
+  {{0xffec0cc6, 0x000a2990, 0, 0}}, {{0xffec72eb, 0x0009f588, 0, 0}},
+  {{0xffecd910, 0x0009c180, 0, 0}}, {{0xffed3f35, 0x00098d78, 0, 0}},
+  {{0xffeda55a, 0x00095970, 0, 0}}, {{0xffee0b7f, 0x00092568, 0, 0}},
+  {{0xffee71a4, 0x0008f160, 0, 0}}, {{0xffeed7c9, 0x0008bd58, 0, 0}},
+  {{0xffef3dee, 0x00088950, 0, 0}}, {{0xffefa413, 0x00085548, 0, 0}},
+  {{0xfff00a38, 0x00082140, 0, 0}}, {{0xfff0705d, 0x0007ed38, 0, 0}},
+  {{0xfff0d682, 0x0007b930, 0, 0}}, {{0xfff13ca7, 0x00078528, 0, 0}},
+  {{0xfff1a2cc, 0x00075120, 0, 0}}, {{0xfff208f1, 0x00071d18, 0, 0}},
+  {{0xfff26f16, 0x0006e910, 0, 0}}, {{0xfff2d53b, 0x0006b508, 0, 0}},
+  {{0xfff33b60, 0x00068100, 0, 0}}, {{0xfff3a185, 0x00064cf8, 0, 0}},
+  {{0xfff407aa, 0x000618f0, 0, 0}}, {{0xfff46dcf, 0x0005e4e8, 0, 0}},
+  {{0xfff4d3f4, 0x0005b0e0, 0, 0}}, {{0xfff53a19, 0x00057cd8, 0, 0}},
+  {{0xfff5a03e, 0x000548d0, 0, 0}}, {{0xfff60663, 0x000514c8, 0, 0}},
+  {{0xfff66c88, 0x0004e0c0, 0, 0}}, {{0xfff6d2ad, 0x0004acb8, 0, 0}},
+  {{0xfff738d2, 0x000478b0, 0, 0}}, {{0xfff79ef7, 0x000444a8, 0, 0}},
+  {{0xfff8051c, 0x000410a0, 0, 0}}, {{0xfff86b41, 0x0003dc98, 0, 0}},
+  {{0xfff8d166, 0x0003a890, 0, 0}}, {{0xfff9378b, 0x00037488, 0, 0}},
+  {{0xfff99db0, 0x00034080, 0, 0}}, {{0xfffa03d5, 0x00030c78, 0, 0}},
+  {{0xfffa69fa, 0x0002d870, 0, 0}}, {{0xfffad01f, 0x0002a468, 0, 0}},
+  {{0xfffb3644, 0x00027060, 0, 0}}, {{0xfffb9c69, 0x00023c58, 0, 0}},
+  {{0xfffc028e, 0x00020850, 0, 0}}, {{0xfffc68b3, 0x0001d448, 0, 0}},
+  {{0xfffcced8, 0x0001a040, 0, 0}}, {{0xfffd34fd, 0x00016c38, 0, 0}},
+  {{0xfffd9b22, 0x00013830, 0, 0}}, {{0xfffe0147, 0x00010428, 0, 0}},
+  {{0xfffe676c, 0x0000d020, 0, 0}}, {{0xfffecd91, 0x00009c18, 0, 0}},
+  {{0xffff33b6, 0x00006810, 0, 0}}, {{0xffff99db, 0x00003408, 0, 0}},
+  {{0x00000000, 0x00000000, 0, 0}}, {{0x00006625, 0xffffcbf8, 0, 0}},
+  {{0x0000cc4a, 0xffff97f0, 0, 0}}, {{0x0001326f, 0xffff63e8, 0, 0}},
+  {{0x00019894, 0xffff2fe0, 0, 0}}, {{0x0001feb9, 0xfffefbd8, 0, 0}},
+  {{0x000264de, 0xfffec7d0, 0, 0}}, {{0x0002cb03, 0xfffe93c8, 0, 0}},
+  {{0x00033128, 0xfffe5fc0, 0, 0}}, {{0x0003974d, 0xfffe2bb8, 0, 0}},
+  {{0x0003fd72, 0xfffdf7b0, 0, 0}}, {{0x00046397, 0xfffdc3a8, 0, 0}},
+  {{0x0004c9bc, 0xfffd8fa0, 0, 0}}, {{0x00052fe1, 0xfffd5b98, 0, 0}},
+  {{0x00059606, 0xfffd2790, 0, 0}}, {{0x0005fc2b, 0xfffcf388, 0, 0}},
+  {{0x00066250, 0xfffcbf80, 0, 0}}, {{0x0006c875, 0xfffc8b78, 0, 0}},
+  {{0x00072e9a, 0xfffc5770, 0, 0}}, {{0x000794bf, 0xfffc2368, 0, 0}},
+  {{0x0007fae4, 0xfffbef60, 0, 0}}, {{0x00086109, 0xfffbbb58, 0, 0}},
+  {{0x0008c72e, 0xfffb8750, 0, 0}}, {{0x00092d53, 0xfffb5348, 0, 0}},
+  {{0x00099378, 0xfffb1f40, 0, 0}}, {{0x0009f99d, 0xfffaeb38, 0, 0}},
+  {{0x000a5fc2, 0xfffab730, 0, 0}}, {{0x000ac5e7, 0xfffa8328, 0, 0}},
+  {{0x000b2c0c, 0xfffa4f20, 0, 0}}, {{0x000b9231, 0xfffa1b18, 0, 0}},
+  {{0x000bf856, 0xfff9e710, 0, 0}}, {{0x000c5e7b, 0xfff9b308, 0, 0}},
+  {{0x000cc4a0, 0xfff97f00, 0, 0}}, {{0x000d2ac5, 0xfff94af8, 0, 0}},
+  {{0x000d90ea, 0xfff916f0, 0, 0}}, {{0x000df70f, 0xfff8e2e8, 0, 0}},
+  {{0x000e5d34, 0xfff8aee0, 0, 0}}, {{0x000ec359, 0xfff87ad8, 0, 0}},
+  {{0x000f297e, 0xfff846d0, 0, 0}}, {{0x000f8fa3, 0xfff812c8, 0, 0}},
+  {{0x000ff5c8, 0xfff7dec0, 0, 0}}, {{0x00105bed, 0xfff7aab8, 0, 0}},
+  {{0x0010c212, 0xfff776b0, 0, 0}}, {{0x00112837, 0xfff742a8, 0, 0}},
+  {{0x00118e5c, 0xfff70ea0, 0, 0}}, {{0x0011f481, 0xfff6da98, 0, 0}},
+  {{0x00125aa6, 0xfff6a690, 0, 0}}, {{0x0012c0cb, 0xfff67288, 0, 0}},
+  {{0x001326f0, 0xfff63e80, 0, 0}}, {{0x00138d15, 0xfff60a78, 0, 0}},
+  {{0x0013f33a, 0xfff5d670, 0, 0}}, {{0x0014595f, 0xfff5a268, 0, 0}},
+  {{0x0014bf84, 0xfff56e60, 0, 0}}, {{0x001525a9, 0xfff53a58, 0, 0}},
+  {{0x00158bce, 0xfff50650, 0, 0}}, {{0x0015f1f3, 0xfff4d248, 0, 0}},
+  {{0x00165818, 0xfff49e40, 0, 0}}, {{0x0016be3d, 0xfff46a38, 0, 0}},
+  {{0x00172462, 0xfff43630, 0, 0}}, {{0x00178a87, 0xfff40228, 0, 0}},
+  {{0x0017f0ac, 0xfff3ce20, 0, 0}}, {{0x001856d1, 0xfff39a18, 0, 0}},
+  {{0x0018bcf6, 0xfff36610, 0, 0}}, {{0x0019231b, 0xfff33208, 0, 0}},
+  {{0x00198940, 0xfff2fe00, 0, 0}}, {{0x0019ef65, 0xfff2c9f8, 0, 0}},
+  {{0x001a558a, 0xfff295f0, 0, 0}}, {{0x001abbaf, 0xfff261e8, 0, 0}},
+  {{0x001b21d4, 0xfff22de0, 0, 0}}, {{0x001b87f9, 0xfff1f9d8, 0, 0}},
+  {{0x001bee1e, 0xfff1c5d0, 0, 0}}, {{0x001c5443, 0xfff191c8, 0, 0}},
+  {{0x001cba68, 0xfff15dc0, 0, 0}}, {{0x001d208d, 0xfff129b8, 0, 0}},
+  {{0x001d86b2, 0xfff0f5b0, 0, 0}}, {{0x001decd7, 0xfff0c1a8, 0, 0}},
+  {{0x001e52fc, 0xfff08da0, 0, 0}}, {{0x001eb921, 0xfff05998, 0, 0}},
+  {{0x001f1f46, 0xfff02590, 0, 0}}, {{0x001f856b, 0xffeff188, 0, 0}},
+  {{0x001feb90, 0xffefbd80, 0, 0}}, {{0x002051b5, 0xffef8978, 0, 0}},
+  {{0x0020b7da, 0xffef5570, 0, 0}}, {{0x00211dff, 0xffef2168, 0, 0}},
+  {{0x00218424, 0xffeeed60, 0, 0}}, {{0x0021ea49, 0xffeeb958, 0, 0}},
+  {{0x0022506e, 0xffee8550, 0, 0}}, {{0x0022b693, 0xffee5148, 0, 0}},
+  {{0x00231cb8, 0xffee1d40, 0, 0}}, {{0x002382dd, 0xffede938, 0, 0}},
+  {{0x0023e902, 0xffedb530, 0, 0}}, {{0x00244f27, 0xffed8128, 0, 0}},
+  {{0x0024b54c, 0xffed4d20, 0, 0}}, {{0x00251b71, 0xffed1918, 0, 0}},
+  {{0x00258196, 0xffece510, 0, 0}}, {{0x0025e7bb, 0xffecb108, 0, 0}},
+  {{0x00264de0, 0xffec7d00, 0, 0}}, {{0x0026b405, 0xffec48f8, 0, 0}},
+  {{0x00271a2a, 0xffec14f0, 0, 0}}, {{0x0027804f, 0xffebe0e8, 0, 0}},
+  {{0x0027e674, 0xffebace0, 0, 0}}, {{0x00284c99, 0xffeb78d8, 0, 0}},
+  {{0x0028b2be, 0xffeb44d0, 0, 0}}, {{0x002918e3, 0xffeb10c8, 0, 0}},
+  {{0x00297f08, 0xffeadcc0, 0, 0}}, {{0x0029e52d, 0xffeaa8b8, 0, 0}},
+  {{0x002a4b52, 0xffea74b0, 0, 0}}, {{0x002ab177, 0xffea40a8, 0, 0}},
+  {{0x002b179c, 0xffea0ca0, 0, 0}}, {{0x002b7dc1, 0xffe9d898, 0, 0}},
+  {{0x002be3e6, 0xffe9a490, 0, 0}}, {{0x002c4a0b, 0xffe97088, 0, 0}},
+  {{0x002cb030, 0xffe93c80, 0, 0}}, {{0x002d1655, 0xffe90878, 0, 0}},
+  {{0x002d7c7a, 0xffe8d470, 0, 0}}, {{0x002de29f, 0xffe8a068, 0, 0}},
+  {{0x002e48c4, 0xffe86c60, 0, 0}}, {{0x002eaee9, 0xffe83858, 0, 0}},
+  {{0x002f150e, 0xffe80450, 0, 0}}, {{0x002f7b33, 0xffe7d048, 0, 0}},
+  {{0x002fe158, 0xffe79c40, 0, 0}}, {{0x0030477d, 0xffe76838, 0, 0}},
+  {{0x0030ada2, 0xffe73430, 0, 0}}, {{0x003113c7, 0xffe70028, 0, 0}},
+  {{0x003179ec, 0xffe6cc20, 0, 0}}, {{0x0031e011, 0xffe69818, 0, 0}},
+  {{0x00324636, 0xffe66410, 0, 0}}, {{0x0032ac5b, 0xffe63008, 0, 0}}
+};

src/enc/Makefile.am

@@ -1,4 +1,3 @@
-AM_CPPFLAGS = -I$(top_srcdir)/src
 noinst_LTLIBRARIES = libwebpencode.la
 
 libwebpencode_la_SOURCES =
@@ -12,10 +11,14 @@ libwebpencode_la_SOURCES += filter.c
 libwebpencode_la_SOURCES += frame.c
 libwebpencode_la_SOURCES += histogram.c
 libwebpencode_la_SOURCES += iterator.c
-libwebpencode_la_SOURCES += layer.c
 libwebpencode_la_SOURCES += picture.c
+libwebpencode_la_SOURCES += picture_csp.c
+libwebpencode_la_SOURCES += picture_psnr.c
+libwebpencode_la_SOURCES += picture_rescale.c
+libwebpencode_la_SOURCES += picture_tools.c
 libwebpencode_la_SOURCES += quant.c
 libwebpencode_la_SOURCES += syntax.c
+libwebpencode_la_SOURCES += token.c
 libwebpencode_la_SOURCES += tree.c
 libwebpencode_la_SOURCES += vp8enci.h
 libwebpencode_la_SOURCES += vp8l.c

src/enc/alpha.c

@@ -1,8 +1,10 @@
 // Copyright 2011 Google Inc. All Rights Reserved.
 //
-// This code is licensed under the same terms as WebM:
-//  Software License Agreement:  http://www.webmproject.org/license/software/
-//  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Alpha-plane compression.
@@ -15,12 +17,9 @@
 #include "./vp8enci.h"
 #include "../utils/filters.h"
 #include "../utils/quant_levels.h"
+#include "../utils/utils.h"
 #include "../webp/format_constants.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 // -----------------------------------------------------------------------------
 // Encodes the given alpha data via specified compression method 'method'.
 // The pre-processing (quantization) is performed if 'quality' is less than 100.
@@ -36,7 +35,7 @@ extern "C" {
 //
 // 'output' corresponds to the buffer containing compressed alpha data.
 //          This buffer is allocated by this method and caller should call
-//          free(*output) when done.
+//          WebPSafeFree(*output) when done.
 // 'output_size' corresponds to size of this compressed alpha buffer.
 //
 // Returns 1 on successfully encoding the alpha and
@@ -48,12 +47,11 @@ extern "C" {
 
 static int EncodeLossless(const uint8_t* const data, int width, int height,
                           int effort_level,  // in [0..6] range
-                          VP8BitWriter* const bw,
+                          VP8LBitWriter* const bw,
                           WebPAuxStats* const stats) {
   int ok = 0;
   WebPConfig config;
   WebPPicture picture;
-  VP8LBitWriter tmp_bw;
 
   WebPPictureInit(&picture);
   picture.width = width;
@@ -69,7 +67,7 @@ static int EncodeLossless(const uint8_t* const data, int width, int height,
     const uint8_t* src = data;
     for (j = 0; j < picture.height; ++j) {
       for (i = 0; i < picture.width; ++i) {
-        dst[i] = (src[i] << 8) | 0xff000000u;
+        dst[i] = src[i] << 8;  // we leave A/R/B channels zero'd.
       }
       src += width;
       dst += picture.argb_stride;
@@ -79,37 +77,46 @@ static int EncodeLossless(const uint8_t* const data, int width, int height,
   WebPConfigInit(&config);
   config.lossless = 1;
   config.method = effort_level;  // impact is very small
-  // Set moderate default quality setting for alpha. Higher qualities (80 and
-  // above) could be very slow.
-  config.quality = 10.f + 15.f * effort_level;
-  if (config.quality > 100.f) config.quality = 100.f;
+  // Set a low default quality for encoding alpha. Ensure that Alpha quality at
+  // lower methods (3 and below) is less than the threshold for triggering
+  // costly 'BackwardReferencesTraceBackwards'.
+  config.quality = 8.f * effort_level;
+  assert(config.quality >= 0 && config.quality <= 100.f);
 
-  ok = VP8LBitWriterInit(&tmp_bw, (width * height) >> 3);
-  ok = ok && (VP8LEncodeStream(&config, &picture, &tmp_bw) == VP8_ENC_OK);
+  ok = (VP8LEncodeStream(&config, &picture, bw) == VP8_ENC_OK);
   WebPPictureFree(&picture);
-  if (ok) {
-    const uint8_t* const data = VP8LBitWriterFinish(&tmp_bw);
-    const size_t data_size = VP8LBitWriterNumBytes(&tmp_bw);
-    VP8BitWriterAppend(bw, data, data_size);
+  ok = ok && !bw->error_;
+  if (!ok) {
+    VP8LBitWriterDestroy(bw);
+    return 0;
   }
-  VP8LBitWriterDestroy(&tmp_bw);
-  return ok && !bw->error_;
+  return 1;
+
 }
 
 // -----------------------------------------------------------------------------
 
+// Small struct to hold the result of a filter mode compression attempt.
+typedef struct {
+  size_t score;
+  VP8BitWriter bw;
+  WebPAuxStats stats;
+} FilterTrial;
+
+// This function always returns an initialized 'bw' object, even upon error.
 static int EncodeAlphaInternal(const uint8_t* const data, int width, int height,
                                int method, int filter, int reduce_levels,
                                int effort_level,  // in [0..6] range
                                uint8_t* const tmp_alpha,
-                               VP8BitWriter* const bw,
-                               WebPAuxStats* const stats) {
+                               FilterTrial* result) {
   int ok = 0;
   const uint8_t* alpha_src;
   WebPFilterFunc filter_func;
   uint8_t header;
-  size_t expected_size;
   const size_t data_size = width * height;
+  const uint8_t* output = NULL;
+  size_t output_size = 0;
+  VP8LBitWriter tmp_bw;
 
   assert((uint64_t)data_size == (uint64_t)width * height);  // as per spec
   assert(filter >= 0 && filter < WEBP_FILTER_LAST);
@@ -118,30 +125,51 @@ static int EncodeAlphaInternal(const uint8_t* const data, int width, int height,
   assert(sizeof(header) == ALPHA_HEADER_LEN);
   // TODO(skal): have a common function and #define's to validate alpha params.
 
-  expected_size =
-      (method == ALPHA_NO_COMPRESSION) ? (ALPHA_HEADER_LEN + data_size)
-                                       : (data_size >> 5);
-  header = method | (filter << 2);
-  if (reduce_levels) header |= ALPHA_PREPROCESSED_LEVELS << 4;
-
-  VP8BitWriterInit(bw, expected_size);
-  VP8BitWriterAppend(bw, &header, ALPHA_HEADER_LEN);
-
   filter_func = WebPFilters[filter];
-  if (filter_func) {
-    filter_func(data, width, height, 1, width, tmp_alpha);
+  if (filter_func != NULL) {
+    filter_func(data, width, height, width, tmp_alpha);
     alpha_src = tmp_alpha;
   }  else {
     alpha_src = data;
   }
 
-  if (method == ALPHA_NO_COMPRESSION) {
-    ok = VP8BitWriterAppend(bw, alpha_src, width * height);
-    ok = ok && !bw->error_;
-  } else {
-    ok = EncodeLossless(alpha_src, width, height, effort_level, bw, stats);
-    VP8BitWriterFinish(bw);
+  if (method != ALPHA_NO_COMPRESSION) {
+    ok = VP8LBitWriterInit(&tmp_bw, data_size >> 3);
+    ok = ok && EncodeLossless(alpha_src, width, height, effort_level,
+                              &tmp_bw, &result->stats);
+    if (ok) {
+      output = VP8LBitWriterFinish(&tmp_bw);
+      output_size = VP8LBitWriterNumBytes(&tmp_bw);
+      if (output_size > data_size) {
+        // compressed size is larger than source! Revert to uncompressed mode.
+        method = ALPHA_NO_COMPRESSION;
+        VP8LBitWriterDestroy(&tmp_bw);
+      }
+    } else {
+      VP8LBitWriterDestroy(&tmp_bw);
+      return 0;
+    }
   }
+
+  if (method == ALPHA_NO_COMPRESSION) {
+    output = alpha_src;
+    output_size = data_size;
+    ok = 1;
+  }
+
+  // Emit final result.
+  header = method | (filter << 2);
+  if (reduce_levels) header |= ALPHA_PREPROCESSED_LEVELS << 4;
+
+  VP8BitWriterInit(&result->bw, ALPHA_HEADER_LEN + output_size);
+  ok = ok && VP8BitWriterAppend(&result->bw, &header, ALPHA_HEADER_LEN);
+  ok = ok && VP8BitWriterAppend(&result->bw, output, output_size);
+
+  if (method != ALPHA_NO_COMPRESSION) {
+    VP8LBitWriterDestroy(&tmp_bw);
+  }
+  ok = ok && !result->bw.error_;
+  result->score = VP8BitWriterSize(&result->bw);
   return ok;
 }
 
@@ -157,6 +185,104 @@ static void CopyPlane(const uint8_t* src, int src_stride,
   }
 }
 
+static int GetNumColors(const uint8_t* data, int width, int height,
+                        int stride) {
+  int j;
+  int colors = 0;
+  uint8_t color[256] = { 0 };
+
+  for (j = 0; j < height; ++j) {
+    int i;
+    const uint8_t* const p = data + j * stride;
+    for (i = 0; i < width; ++i) {
+      color[p[i]] = 1;
+    }
+  }
+  for (j = 0; j < 256; ++j) {
+    if (color[j] > 0) ++colors;
+  }
+  return colors;
+}
+
+#define FILTER_TRY_NONE (1 << WEBP_FILTER_NONE)
+#define FILTER_TRY_ALL ((1 << WEBP_FILTER_LAST) - 1)
+
+// Given the input 'filter' option, return an OR'd bit-set of filters to try.
+static uint32_t GetFilterMap(const uint8_t* alpha, int width, int height,
+                             int filter, int effort_level) {
+  uint32_t bit_map = 0U;
+  if (filter == WEBP_FILTER_FAST) {
+    // Quick estimate of the best candidate.
+    int try_filter_none = (effort_level > 3);
+    const int kMinColorsForFilterNone = 16;
+    const int kMaxColorsForFilterNone = 192;
+    const int num_colors = GetNumColors(alpha, width, height, width);
+    // For low number of colors, NONE yields better compression.
+    filter = (num_colors <= kMinColorsForFilterNone) ? WEBP_FILTER_NONE :
+             EstimateBestFilter(alpha, width, height, width);
+    bit_map |= 1 << filter;
+    // For large number of colors, try FILTER_NONE in addition to the best
+    // filter as well.
+    if (try_filter_none || num_colors > kMaxColorsForFilterNone) {
+      bit_map |= FILTER_TRY_NONE;
+    }
+  } else if (filter == WEBP_FILTER_NONE) {
+    bit_map = FILTER_TRY_NONE;
+  } else {  // WEBP_FILTER_BEST -> try all
+    bit_map = FILTER_TRY_ALL;
+  }
+  return bit_map;
+}
+
+static void InitFilterTrial(FilterTrial* const score) {
+  score->score = (size_t)~0U;
+  VP8BitWriterInit(&score->bw, 0);
+}
+
+static int ApplyFiltersAndEncode(const uint8_t* alpha, int width, int height,
+                                 size_t data_size, int method, int filter,
+                                 int reduce_levels, int effort_level,
+                                 uint8_t** const output,
+                                 size_t* const output_size,
+                                 WebPAuxStats* const stats) {
+  int ok = 1;
+  FilterTrial best;
+  uint32_t try_map =
+      GetFilterMap(alpha, width, height, filter, effort_level);
+  InitFilterTrial(&best);
+  if (try_map != FILTER_TRY_NONE) {
+    uint8_t* filtered_alpha =  (uint8_t*)WebPSafeMalloc(1ULL, data_size);
+    if (filtered_alpha == NULL) return 0;
+
+    for (filter = WEBP_FILTER_NONE; ok && try_map; ++filter, try_map >>= 1) {
+      if (try_map & 1) {
+        FilterTrial trial;
+        ok = EncodeAlphaInternal(alpha, width, height, method, filter,
+                                 reduce_levels, effort_level, filtered_alpha,
+                                 &trial);
+        if (ok && trial.score < best.score) {
+          VP8BitWriterWipeOut(&best.bw);
+          best = trial;
+        } else {
+          VP8BitWriterWipeOut(&trial.bw);
+        }
+      }
+    }
+    WebPSafeFree(filtered_alpha);
+  } else {
+    ok = EncodeAlphaInternal(alpha, width, height, method, WEBP_FILTER_NONE,
+                             reduce_levels, effort_level, NULL, &best);
+  }
+  if (ok) {
+    if (stats != NULL) *stats = best.stats;
+    *output_size = VP8BitWriterSize(&best.bw);
+    *output = VP8BitWriterBuf(&best.bw);
+  } else {
+    VP8BitWriterWipeOut(&best.bw);
+  }
+  return ok;
+}
+
 static int EncodeAlpha(VP8Encoder* const enc,
                        int quality, int method, int filter,
                        int effort_level,
@@ -187,7 +313,12 @@ static int EncodeAlpha(VP8Encoder* const enc,
     return 0;
   }
 
-  quant_alpha = (uint8_t*)malloc(data_size);
+  if (method == ALPHA_NO_COMPRESSION) {
+    // Don't filter, as filtering will make no impact on compressed size.
+    filter = WEBP_FILTER_NONE;
+  }
+
+  quant_alpha = (uint8_t*)WebPSafeMalloc(1ULL, data_size);
   if (quant_alpha == NULL) {
     return 0;
   }
@@ -205,126 +336,98 @@ static int EncodeAlpha(VP8Encoder* const enc,
   }
 
   if (ok) {
-    VP8BitWriter bw;
-    int test_filter;
-    uint8_t* filtered_alpha = NULL;
-
-    // We always test WEBP_FILTER_NONE first.
-    ok = EncodeAlphaInternal(quant_alpha, width, height,
-                             method, WEBP_FILTER_NONE, reduce_levels,
-                             effort_level, NULL, &bw, pic->stats);
-    if (!ok) {
-      VP8BitWriterWipeOut(&bw);
-      goto End;
+    ok = ApplyFiltersAndEncode(quant_alpha, width, height, data_size, method,
+                               filter, reduce_levels, effort_level, output,
+                               output_size, pic->stats);
+    if (pic->stats != NULL) {  // need stats?
+      pic->stats->coded_size += (int)(*output_size);
+      enc->sse_[3] = sse;
     }
-
-    if (filter == WEBP_FILTER_FAST) {  // Quick estimate of a second candidate?
-      filter = EstimateBestFilter(quant_alpha, width, height, width);
-    }
-    // Stop?
-    if (filter == WEBP_FILTER_NONE) {
-      goto Ok;
-    }
-
-    filtered_alpha = (uint8_t*)malloc(data_size);
-    ok = (filtered_alpha != NULL);
-    if (!ok) {
-      goto End;
-    }
-
-    // Try the other mode(s).
-    {
-      WebPAuxStats best_stats;
-      size_t best_score = VP8BitWriterSize(&bw);
-
-      memset(&best_stats, 0, sizeof(best_stats));  // prevent spurious warning
-      if (pic->stats != NULL) best_stats = *pic->stats;
-      for (test_filter = WEBP_FILTER_HORIZONTAL;
-           ok && (test_filter <= WEBP_FILTER_GRADIENT);
-           ++test_filter) {
-        VP8BitWriter tmp_bw;
-        if (filter != WEBP_FILTER_BEST && test_filter != filter) {
-          continue;
-        }
-        ok = EncodeAlphaInternal(quant_alpha, width, height,
-                                 method, test_filter, reduce_levels,
-                                 effort_level, filtered_alpha, &tmp_bw,
-                                 pic->stats);
-        if (ok) {
-          const size_t score = VP8BitWriterSize(&tmp_bw);
-          if (score < best_score) {
-            // swap bitwriter objects.
-            VP8BitWriter tmp = tmp_bw;
-            tmp_bw = bw;
-            bw = tmp;
-            best_score = score;
-            if (pic->stats != NULL) best_stats = *pic->stats;
-          }
-        } else {
-          VP8BitWriterWipeOut(&bw);
-        }
-        VP8BitWriterWipeOut(&tmp_bw);
-      }
-      if (pic->stats != NULL) *pic->stats = best_stats;
-    }
- Ok:
-    if (ok) {
-      *output_size = VP8BitWriterSize(&bw);
-      *output = VP8BitWriterBuf(&bw);
-      if (pic->stats != NULL) {         // need stats?
-        pic->stats->coded_size += (int)(*output_size);
-        enc->sse_[3] = sse;
-      }
-    }
-    free(filtered_alpha);
   }
- End:
-  free(quant_alpha);
+
+  WebPSafeFree(quant_alpha);
   return ok;
 }
 
-
 //------------------------------------------------------------------------------
 // Main calls
 
+static int CompressAlphaJob(VP8Encoder* const enc, void* dummy) {
+  const WebPConfig* config = enc->config_;
+  uint8_t* alpha_data = NULL;
+  size_t alpha_size = 0;
+  const int effort_level = config->method;  // maps to [0..6]
+  const WEBP_FILTER_TYPE filter =
+      (config->alpha_filtering == 0) ? WEBP_FILTER_NONE :
+      (config->alpha_filtering == 1) ? WEBP_FILTER_FAST :
+                                       WEBP_FILTER_BEST;
+  if (!EncodeAlpha(enc, config->alpha_quality, config->alpha_compression,
+                   filter, effort_level, &alpha_data, &alpha_size)) {
+    return 0;
+  }
+  if (alpha_size != (uint32_t)alpha_size) {  // Sanity check.
+    WebPSafeFree(alpha_data);
+    return 0;
+  }
+  enc->alpha_data_size_ = (uint32_t)alpha_size;
+  enc->alpha_data_ = alpha_data;
+  (void)dummy;
+  return 1;
+}
+
 void VP8EncInitAlpha(VP8Encoder* const enc) {
   enc->has_alpha_ = WebPPictureHasTransparency(enc->pic_);
   enc->alpha_data_ = NULL;
   enc->alpha_data_size_ = 0;
+  if (enc->thread_level_ > 0) {
+    WebPWorker* const worker = &enc->alpha_worker_;
+    WebPGetWorkerInterface()->Init(worker);
+    worker->data1 = enc;
+    worker->data2 = NULL;
+    worker->hook = (WebPWorkerHook)CompressAlphaJob;
+  }
+}
+
+int VP8EncStartAlpha(VP8Encoder* const enc) {
+  if (enc->has_alpha_) {
+    if (enc->thread_level_ > 0) {
+      WebPWorker* const worker = &enc->alpha_worker_;
+      // Makes sure worker is good to go.
+      if (!WebPGetWorkerInterface()->Reset(worker)) {
+        return 0;
+      }
+      WebPGetWorkerInterface()->Launch(worker);
+      return 1;
+    } else {
+      return CompressAlphaJob(enc, NULL);   // just do the job right away
+    }
+  }
+  return 1;
 }
 
 int VP8EncFinishAlpha(VP8Encoder* const enc) {
   if (enc->has_alpha_) {
-    const WebPConfig* config = enc->config_;
-    uint8_t* tmp_data = NULL;
-    size_t tmp_size = 0;
-    const int effort_level = config->method;  // maps to [0..6]
-    const WEBP_FILTER_TYPE filter =
-        (config->alpha_filtering == 0) ? WEBP_FILTER_NONE :
-        (config->alpha_filtering == 1) ? WEBP_FILTER_FAST :
-                                         WEBP_FILTER_BEST;
-
-    if (!EncodeAlpha(enc, config->alpha_quality, config->alpha_compression,
-                     filter, effort_level, &tmp_data, &tmp_size)) {
-      return 0;
+    if (enc->thread_level_ > 0) {
+      WebPWorker* const worker = &enc->alpha_worker_;
+      if (!WebPGetWorkerInterface()->Sync(worker)) return 0;  // error
     }
-    if (tmp_size != (uint32_t)tmp_size) {  // Sanity check.
-      free(tmp_data);
-      return 0;
-    }
-    enc->alpha_data_size_ = (uint32_t)tmp_size;
-    enc->alpha_data_ = tmp_data;
   }
   return WebPReportProgress(enc->pic_, enc->percent_ + 20, &enc->percent_);
 }
 
-void VP8EncDeleteAlpha(VP8Encoder* const enc) {
-  free(enc->alpha_data_);
+int VP8EncDeleteAlpha(VP8Encoder* const enc) {
+  int ok = 1;
+  if (enc->thread_level_ > 0) {
+    WebPWorker* const worker = &enc->alpha_worker_;
+    // finish anything left in flight
+    ok = WebPGetWorkerInterface()->Sync(worker);
+    // still need to end the worker, even if !ok
+    WebPGetWorkerInterface()->End(worker);
+  }
+  WebPSafeFree(enc->alpha_data_);
   enc->alpha_data_ = NULL;
   enc->alpha_data_size_ = 0;
   enc->has_alpha_ = 0;
+  return ok;
 }
 
-#if defined(__cplusplus) || defined(c_plusplus)
-}    // extern "C"
-#endif