Fix OOB write in BuildHuffmanTable.

First, BuildHuffmanTable is called to check if the data is valid.
If it is and the table is not big enough, more memory is allocated.

This will make sure that valid (but unoptimized because of unbalanced
codes) streams are still decodable.

Bug: chromium:1479274
Change-Id: I31c36dbf3aa78d35ecf38706b50464fd3d375741
(cherry picked from commit 902bc91903)
(cherry picked from commit 2af26267cd)

.gitattributes

@@ -1,5 +1,6 @@
 .gitattributes export-ignore
 .gitignore export-ignore
 .mailmap export-ignore
+*.bat text eol=crlf
 *.pdf -text -diff
 *.ppm -text -diff

.gitignore

@@ -1,6 +1,7 @@
 *.l[ao]
 *.[ao]
 *.pc
+.DS_Store
 .deps
 .libs
 /aclocal.m4
@@ -20,9 +21,11 @@
 Makefile
 Makefile.in
 examples/anim_diff
+examples/anim_dump
 examples/[cdv]webp
 examples/gif2webp
 examples/img2webp
+examples/webpinfo
 examples/webpmux
 src/webp/config.h*
 src/webp/stamp-h1
@@ -31,11 +34,21 @@ src/webp/stamp-h1
 *.idb
 *.pdb
 /iosbuild
-/WebP.framework
+/xcframeworkbuild
+/WebP*.*framework
 CMakeCache.txt
 CMakeFiles/
 cmake_install.cmake
 .gradle
 /build
 extras/get_disto
+extras/vwebp_sdl
 extras/webp_quality
+tests/fuzzer/advanced_api_fuzzer
+tests/fuzzer/animation_api_fuzzer
+tests/fuzzer/animdecoder_fuzzer
+tests/fuzzer/animencoder_fuzzer
+tests/fuzzer/demux_api_fuzzer
+tests/fuzzer/enc_dec_fuzzer
+tests/fuzzer/mux_demux_api_fuzzer
+tests/fuzzer/simple_api_fuzzer

.mailmap

@@ -1,7 +1,8 @@
-<johann.koenig@duck.com> <johannkoenig@google.com>
+Johann Koenig <johann.koenig@duck.com>
+Johann Koenig <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>
+Pascal Massimino <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>

AUTHORS

@@ -1,13 +1,16 @@
 Contributors:
+- Aidan O'Loan (aidanol at gmail dot com)
 - Alan Browning (browning at google dot com)
 - Charles Munger (clm at google dot com)
+- Cheng Yi (cyi at google dot com)
 - Christian Duvivier (cduvivier at google dot com)
 - Djordje Pesut (djordje dot pesut at imgtec dot com)
 - Hui Su (huisu at google dot com)
+- Ingvar Stepanyan (rreverser at google dot com)
 - James Zern (jzern at google dot com)
 - Jan Engelhardt (jengelh at medozas dot de)
 - Jehan (jehan at girinstud dot io)
-- Johann (johann dot koenig at duck dot com)
+- Johann Koenig (johann dot koenig at duck dot com)
 - Jovan Zelincevic (jovan dot zelincevic at imgtec dot com)
 - Jyrki Alakuijala (jyrki at google dot com)
 - Konstantin Ivlev (tomskside at gmail dot com)
@@ -20,6 +23,7 @@ Contributors:
 - Mislav Bradac (mislavm at google dot com)
 - Nico Weber (thakis at chromium dot org)
 - Noel Chromium (noel at chromium dot org)
+- Oliver Wolff (oliver dot wolff at qt dot io)
 - Owen Rodley (orodley at google dot com)
 - Parag Salasakar (img dot mips1 at gmail dot com)
 - Pascal Massimino (pascal dot massimino at gmail dot com)
@@ -38,5 +42,6 @@ Contributors:
 - Vikas Arora (vikasa at google dot com)
 - Vincent Rabaud (vrabaud at google dot com)
 - Vlad Tsyrklevich (vtsyrklevich at chromium dot org)
+- Wan-Teh Chang (wtc at google dot com)
 - Yang Zhang (yang dot zhang at arm dot com)
 - Yannis Guyon (yguyon at google dot com)

Android.mk

@@ -97,6 +97,7 @@ dsp_enc_srcs := \
     src/dsp/cost.c \
     src/dsp/cost_mips32.c \
     src/dsp/cost_mips_dsp_r2.c \
+    src/dsp/cost_neon.$(NEON) \
     src/dsp/cost_sse2.c \
     src/dsp/enc.c \
     src/dsp/enc_mips32.c \

CMakeLists.txt

@@ -1,9 +1,24 @@
+##
+##  Copyright (c) 2020 Google LLC.
+##
+##  Use of this source code is governed by a BSD-style license
+##  that can be found in the LICENSE 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.
+
 cmake_minimum_required(VERSION 3.5)
 
 project(WebP C)
 
 # Options for coder / decoder executables.
-option(WEBP_ENABLE_SIMD "Enable any SIMD optimization." ON)
+if(NOT EMSCRIPTEN)
+  # Disable SIMD on Emscripten by default, as it's a new unstable Wasm feature.
+  # Users can still explicitly opt-in to make a SIMD-enabled build.
+  set(WEBP_ENABLE_SIMD_DEFAULT ON)
+endif()
+option(WEBP_ENABLE_SIMD "Enable any SIMD optimization."
+       ${WEBP_ENABLE_SIMD_DEFAULT})
 option(WEBP_BUILD_ANIM_UTILS "Build animation utilities." ON)
 option(WEBP_BUILD_CWEBP "Build the cwebp command line tool." ON)
 option(WEBP_BUILD_DWEBP "Build the dwebp command line tool." ON)
@@ -17,9 +32,15 @@ option(WEBP_BUILD_WEBP_JS "Emscripten build of webp.js." OFF)
 option(WEBP_NEAR_LOSSLESS "Enable near-lossless encoding" ON)
 option(WEBP_ENABLE_SWAP_16BIT_CSP "Enable byte swap for 16 bit colorspaces."
        OFF)
+set(WEBP_BITTRACE "0" CACHE STRING "Bit trace mode (0=none, 1=bit, 2=bytes)")
+set_property(CACHE WEBP_BITTRACE PROPERTY STRINGS 0 1 2)
+
+# Option needed for handling Unicode file names on Windows.
+if(WIN32)
+  option(WEBP_UNICODE "Build Unicode executables." ON)
+endif()
 
 if(WEBP_BUILD_WEBP_JS)
-  set(WEBP_ENABLE_SIMD OFF)
   set(WEBP_BUILD_ANIM_UTILS OFF)
   set(WEBP_BUILD_CWEBP OFF)
   set(WEBP_BUILD_DWEBP OFF)
@@ -29,6 +50,10 @@ if(WEBP_BUILD_WEBP_JS)
   set(WEBP_BUILD_WEBPINFO OFF)
   set(WEBP_BUILD_WEBPMUX OFF)
   set(WEBP_BUILD_EXTRAS OFF)
+
+  if(WEBP_ENABLE_SIMD)
+    message("wasm2js does not support SIMD, disabling webp.js generation.")
+  endif()
 endif()
 
 set(WEBP_DEP_LIBRARIES)
@@ -36,7 +61,7 @@ set(WEBP_DEP_INCLUDE_DIRS)
 
 if(NOT CMAKE_BUILD_TYPE)
   set(CMAKE_BUILD_TYPE "Release"
-      CACHE "Build type: Release, Debug, MinSizeRel or RelWithDebInfo" STRING
+      CACHE STRING "Build type: Release, Debug, MinSizeRel or RelWithDebInfo"
       FORCE)
 endif()
 
@@ -50,11 +75,21 @@ if(WEBP_ENABLE_SWAP_16BIT_CSP)
   add_definitions(-DWEBP_SWAP_16BIT_CSP=1)
 endif()
 
+if(NOT WEBP_BITTRACE STREQUAL "0")
+  add_definitions(-DBITTRACE=${WEBP_BITTRACE})
+endif()
+
+if(WEBP_UNICODE)
+  # Windows recommends setting both UNICODE and _UNICODE.
+  add_definitions(-DUNICODE -D_UNICODE)
+endif()
+
 set(prefix ${CMAKE_INSTALL_PREFIX})
 set(exec_prefix "\$\{prefix\}")
 set(libdir "\$\{prefix\}/lib")
 set(includedir "\$\{prefix\}/include")
 set(PTHREAD_LIBS ${CMAKE_THREAD_LIBS_INIT})
+set(INSTALLED_LIBRARIES)
 
 # ##############################################################################
 # Android only.
@@ -62,6 +97,7 @@ if(ANDROID)
   include_directories(${ANDROID_NDK}/sources/android/cpufeatures)
   add_library(cpufeatures STATIC
               ${ANDROID_NDK}/sources/android/cpufeatures/cpu-features.c)
+  list(APPEND INSTALLED_LIBRARIES cpufeatures)
   target_link_libraries(cpufeatures dl)
   set(WEBP_DEP_LIBRARIES ${WEBP_DEP_LIBRARIES} cpufeatures)
   set(WEBP_DEP_INCLUDE_DIRS ${WEBP_DEP_INCLUDE_DIRS}
@@ -146,6 +182,29 @@ add_definitions(-DHAVE_CONFIG_H)
 
 # ##############################################################################
 # Build the webpdecoder library.
+
+# Creates a source file with an unused stub function in $CMAKE_BINARY_DIR and
+# adds it to the specified target. Currently used only with Xcode.
+#
+# See also:
+# https://cmake.org/cmake/help/v3.18/command/add_library.html#object-libraries
+# "Some native build systems (such as Xcode) may not like targets that have
+# only object files, so consider adding at least one real source file to any
+# target that references $<TARGET_OBJECTS:objlib>."
+function(libwebp_add_stub_file TARGET)
+  set(stub_source_dir "${CMAKE_BINARY_DIR}")
+  set(stub_source_file
+      "${stub_source_dir}/libwebp_${TARGET}_stub.c")
+  set(stub_source_code
+      "// Generated file. DO NOT EDIT!\n"
+      "// C source file created for target ${TARGET}.\n"
+      "void libwebp_${TARGET}_stub_function(void)\;\n"
+      "void libwebp_${TARGET}_stub_function(void) {}\n")
+  file(WRITE "${stub_source_file}" ${stub_source_code})
+
+  target_sources(${TARGET} PRIVATE ${stub_source_file})
+endfunction()
+
 if(MSVC)
   # avoid security warnings for e.g., fopen() used in the examples.
   add_definitions(-D_CRT_SECURE_NO_WARNINGS)
@@ -172,6 +231,9 @@ add_library(webpdecoder
             $<TARGET_OBJECTS:webpdecode>
             $<TARGET_OBJECTS:webpdspdecode>
             $<TARGET_OBJECTS:webputilsdecode>)
+if(XCODE)
+  libwebp_add_stub_file(webpdecoder)
+endif()
 target_link_libraries(webpdecoder ${WEBP_DEP_LIBRARIES})
 target_include_directories(
   webpdecoder
@@ -211,6 +273,9 @@ add_library(webp
             $<TARGET_OBJECTS:webpdsp>
             $<TARGET_OBJECTS:webpencode>
             $<TARGET_OBJECTS:webputils>)
+if(XCODE)
+  libwebp_add_stub_file(webp)
+endif()
 target_link_libraries(webp ${WEBP_DEP_LIBRARIES})
 target_include_directories(webp
                            PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}
@@ -296,7 +361,7 @@ string(REGEX MATCH
              ${TMP})
 
 # Define the libraries to install.
-set(INSTALLED_LIBRARIES webpdecoder webp webpdemux)
+list(APPEND INSTALLED_LIBRARIES webpdecoder webp webpdemux)
 
 # Deal with SIMD. Change the compile flags for SIMD files we use.
 list(LENGTH WEBP_SIMD_FILES_TO_INCLUDE WEBP_SIMD_FILES_TO_INCLUDE_LENGTH)
@@ -331,20 +396,21 @@ if(WEBP_BUILD_ANIM_UTILS
   parse_makefile_am(${CMAKE_CURRENT_SOURCE_DIR}/examples "EXAMPLEUTIL_SRCS"
                     "example_util_[^ ]*")
   list(APPEND EXAMPLEUTIL_SRCS ${CMAKE_CURRENT_SOURCE_DIR}/examples/stopwatch.h)
-  add_library(exampleutil ${EXAMPLEUTIL_SRCS})
+  add_library(exampleutil STATIC ${EXAMPLEUTIL_SRCS})
+  target_link_libraries(exampleutil imageioutil)
   target_include_directories(
     exampleutil
     PUBLIC $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/src>)
 
   parse_makefile_am(${CMAKE_CURRENT_SOURCE_DIR}/imageio "IMAGEIOUTILS_SRCS"
                     "imageio_util_[^ ]*")
-  add_library(imageioutil ${IMAGEIOUTILS_SRCS})
+  add_library(imageioutil STATIC ${IMAGEIOUTILS_SRCS})
   target_link_libraries(imageioutil webp)
 
   # Image-decoding utility library.
   parse_makefile_am(${CMAKE_CURRENT_SOURCE_DIR}/imageio "IMAGEDEC_SRCS"
                     "imagedec_[^ ]*")
-  add_library(imagedec ${IMAGEDEC_SRCS})
+  add_library(imagedec STATIC ${IMAGEDEC_SRCS})
   target_link_libraries(imagedec
                         imageioutil
                         webpdemux
@@ -354,8 +420,8 @@ if(WEBP_BUILD_ANIM_UTILS
   # Image-encoding utility library.
   parse_makefile_am(${CMAKE_CURRENT_SOURCE_DIR}/imageio "IMAGEENC_SRCS"
                     "imageenc_[^ ]*")
-  add_library(imageenc ${IMAGEENC_SRCS})
-  target_link_libraries(imageenc webp)
+  add_library(imageenc STATIC ${IMAGEENC_SRCS})
+  target_link_libraries(imageenc imageioutil webp)
 
   set_property(TARGET exampleutil
                       imageioutil
@@ -448,12 +514,12 @@ if(WEBP_BUILD_VWEBP)
     target_link_libraries(vwebp
                           ${OPENGL_LIBRARIES}
                           exampleutil
-                          GLUT::GLUT
+                          ${GLUT_glut_LIBRARY}
                           imageioutil
                           webp
                           webpdemux)
     target_include_directories(vwebp
-                               PRIVATE GLUT::GLUT
+                               PRIVATE ${GLUT_INCLUDE_DIR}
                                        ${CMAKE_CURRENT_BINARY_DIR}/src
                                        ${OPENGL_INCLUDE_DIR})
     install(TARGETS vwebp RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
@@ -506,7 +572,9 @@ if(WEBP_BUILD_EXTRAS)
   # webp_quality
   add_executable(webp_quality ${WEBP_QUALITY_SRCS} ${WEBP_EXTRAS_SRCS})
   target_link_libraries(webp_quality exampleutil imagedec)
-  target_include_directories(webp_quality PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
+  target_include_directories(webp_quality
+                             PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}
+                                     ${CMAKE_CURRENT_BINARY_DIR})
   install(TARGETS webp_quality RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
 
   # vwebp_sdl
@@ -527,18 +595,21 @@ if(WEBP_BUILD_EXTRAS)
 endif()
 
 if(WEBP_BUILD_WEBP_JS)
-  # JavaScript version
-  add_executable(webp_js ${CMAKE_CURRENT_SOURCE_DIR}/extras/webp_to_sdl.c)
-  target_link_libraries(webp_js webpdecoder SDL)
-  target_include_directories(webp_js PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
-  set(WEBP_HAVE_SDL 1)
-  set_target_properties(
-    webp_js
-    PROPERTIES LINK_FLAGS
-               "-s EXPORTED_FUNCTIONS='[\"_WebpToSDL\"]' -s INVOKE_RUN=0 \
-       -s EXTRA_EXPORTED_RUNTIME_METHODS='[\"cwrap\"]'")
-  set_target_properties(webp_js PROPERTIES OUTPUT_NAME webp)
-  target_compile_definitions(webp_js PUBLIC EMSCRIPTEN WEBP_HAVE_SDL)
+  # wasm2js does not support SIMD.
+  if(NOT WEBP_ENABLE_SIMD)
+    # JavaScript version
+    add_executable(webp_js ${CMAKE_CURRENT_SOURCE_DIR}/extras/webp_to_sdl.c)
+    target_link_libraries(webp_js webpdecoder SDL)
+    target_include_directories(webp_js PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
+    set(WEBP_HAVE_SDL 1)
+    set_target_properties(
+      webp_js
+      PROPERTIES LINK_FLAGS "-s WASM=0 \
+         -s EXPORTED_FUNCTIONS='[\"_WebpToSDL\"]' -s INVOKE_RUN=0 \
+         -s EXTRA_EXPORTED_RUNTIME_METHODS='[\"cwrap\"]'")
+    set_target_properties(webp_js PROPERTIES OUTPUT_NAME webp)
+    target_compile_definitions(webp_js PUBLIC EMSCRIPTEN WEBP_HAVE_SDL)
+  endif()
 
   # WASM version
   add_executable(webp_wasm ${CMAKE_CURRENT_SOURCE_DIR}/extras/webp_to_sdl.c)
@@ -546,7 +617,7 @@ if(WEBP_BUILD_WEBP_JS)
   target_include_directories(webp_wasm PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
   set_target_properties(
     webp_wasm
-    PROPERTIES LINK_FLAGS "-s WASM=1 -s 'BINARYEN_METHOD=\"native-wasm\"' \
+    PROPERTIES LINK_FLAGS "-s WASM=1 \
        -s EXPORTED_FUNCTIONS='[\"_WebpToSDL\"]' -s INVOKE_RUN=0 \
        -s EXTRA_EXPORTED_RUNTIME_METHODS='[\"cwrap\"]'")
   target_compile_definitions(webp_wasm PUBLIC EMSCRIPTEN WEBP_HAVE_SDL)
@@ -589,7 +660,6 @@ if(WEBP_BUILD_ANIM_UTILS)
 endif()
 
 # Install the different headers and libraries.
-include(GNUInstallDirs)
 install(TARGETS ${INSTALLED_LIBRARIES}
         EXPORT ${PROJECT_NAME}Targets
         PUBLIC_HEADER DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/webp

ChangeLog

@@ -1,15 +1,202 @@
+170a8712 Fix check_c_source_compiles with pthread.
+85995719 disable CombinedShannonEntropy_SSE2 on x86
+8af7436f Merge "{ios,xcframework}build.sh: make min version(s) more visible" into 1.2.0
+e56c3c5b pngdec: raise memory limit if needed
+13b8e9fe {ios,xcframework}build.sh: make min version(s) more visible
+a9225410 animdecoder_fuzzer: fix memory leak
+d6c2285d update gradle to 6.1.1
+52ce6333 update NEWS
+28c49820 bump version to 1.2.0
+7363dff2 webp/encode.h: restore WEBP_ENCODER_ABI_VERSION to v1.1.0
+826aafa5 update AUTHORS
+63258823 animdecoder_fuzzer: validate canvas size
+9eb26381 CMake: remove duplicate "include(GNUInstallDirs)"
+2e7bed79 WebPPicture: clarify the ownership of user-owned data.
+cccf5e33 webpmux: add an '-set loop <value>' option
+c9a3f6a1 Merge changes Ie29f9867,I289c54c4
+319f56f1 iosbuild.sh: sync some aspects of xcframeworkbuild.sh
+e8e8db98 add xcframeworkbuild.sh
+ae545534 dsp.h: allow config.h to override MSVC SIMD autodetection
+fef789f3 Merge "cmake: fix per-file assembly flags"
+fc14fc03 Have C encoding predictors use decoding predictors.
+7656f0b3 README,cosmetics: fix a couple typos
+d2e245ea cmake: disable webp.js if WEBP_ENABLE_SIMD=1
+96099a79 cmake: fix per-file assembly flags
+5abb5582 Merge "cmake: fix compilation w/Xcode generator"
+8484a120 cmake: fix compilation w/Xcode generator
+d7bf01c9 Merge changes Ifcae0f38,Iee2d7401
+36c81ff6 WASM-SIMD: port 2 patches from rreverser@'s tree
+988b02ab Merge "Couple of fixes to allow SIMD on Emscripten"
+26faf770 wicdec: fail with animated images
+ab2d08a8 [cd]webp: document lack of animated webp support
+52273943 Couple of fixes to allow SIMD on Emscripten
+8870ba7f Fix skia bug #10952
+4b3c6953 Detect if StoreFrame read more than anmf_payload_size bytes
+17fd4ba8 webp/decode.h,cosmetics: normalize 'flip' comment
+411d3677 remove some unreachable break statements
+3700ffd7 WebPPictureHasTransparency: remove unreachable return
+83604bf3 {animencoder,enc_dec}_fuzzer: convert some abort()s to returns
+eb44119c Merge changes I8ae09473,I678c8b1e
+9f6055fc fuzz_utils.h: rename max() to Max()
+695788e7 fuzz_utils.h: make functions WEBP_INLINE
+906c1fcd make ImgIoUtilReadFile use WebPMalloc instead of malloc
+8cb7e536 rename demux_api_fuzzer.c -> mux_demux_api_fuzzer.c
+443db47d add animdecoder_fuzzer.cc
+36a6eea3 Merge "import fuzzers from oss-fuzz/chromium"
+ec5f12c1 Makefile.vc: remove deprecated /Gm option
+64425a08 picture_tools_enc: fix windows build warning
+bd94090a import fuzzers from oss-fuzz/chromium
+cf847cba use WEBP_DSP_INIT_FUNC for Init{GammaTables*,GetCoeffs}
+55a080e5 Add WebPReplaceTransparentPixels() in dsp
+84739717 GetBackgroundColorGIF: promote to uint32_t before << 24
+def64e92 cwebp: Fix -print_psnr for near_lossless
+cf2f88b3 Add palette and spatial for q >= 75 and -m 5
+f0110bae Add no-color cache configuration to the cruncher
+749a8b99 Better estimate of the cache cost.
+4f9f00cc Use spatial predictors on top of palette no matter what.
+7658c686 Add spatial prediction on top of palette in cruncher.
+133ff0e3 webp_js: force WASM=0 option explicitly
+e3c259a2 Fix integer overflow in EmitFancyRGB.
+b3ff0bde man/{gif2,img2}webp,webpmux: normalize some wording
+f9b30586 fix ABI breakage introduced by 6a0ff358
+1d58dcfc README.webp_js: update note about emscripten version
+44070266 README.webp_js: s/fastcomp/upstream/
+2565fa8f README.webp_js: update cmake command
+47309ef5 webp: WEBP_OFFSET_PTR()
+687ab00e DC{4,8,16}_NEON: replace vmovl w/vaddl
+1b92fe75 DC16_NEON,aarch64: use vaddlv
+53f3d8cf dec_neon,DC8_NEON: use vaddlv instead of movl+vaddv
+27d08240 Fix integer overflow in WebPAnimDecoderGetNext()
+69776e38 Merge "remove call to MBAnalyzeBestIntra4Mode for method >= 5"
+a99078c1 remove call to MBAnalyzeBestIntra4Mode for method >= 5
+22e404cc CMakeLists.txt: fix set(CACHE) argument order
+71690b52 fix MSVC warning
+6a0ff358 Enc: add a qmin / qmax range for quality factor
+0fa56f30 Merge tag 'v1.1.0'
+6cf504d0 PNM decoding: handle max_value != 255
+d7844e97 update ChangeLog (tag: v1.1.0-rc2, tag: v1.1.0, origin/1.1.0)
+7f006436 Makefile.vc: fix webp_quality.exe link
+cf047e83 Makefile.vc: fix webp_quality.exe link
+c074c653 update NEWS
+30f09551 bump version to 1.1.0
+a76694a1 update AUTHORS
+6e3ef7b3 extras: fix WEBP_SWAP_16BIT_CSP check
+47178dbd extras: add WebPUnmultiplyARGB() convenience function
+22cbae33 idec_dec: fix 0 offset of NULL pointer
+290dd0b4 muxread: fix 0 offset of NULL pointer
+0df474ac Merge "lossless_(enc_|)sse2: avoid offsetting a NULL pointer"
+c6b75a19 lossless_(enc_|)sse2: avoid offsetting a NULL pointer
+295e5e38 fix UBSAN warning
+e2575e05 DC8_NEON,aarch64: use vaddv
+b0e09e34 dec_neon: Fix build failure under some toolchains
+cf0e903c dsp/lossless: Fix non gcc ARM builds
+bb7bc40b Remove ubsan errors.
+78881b76 CMake: fix GLUT library link
+9f750f7a cmake: fix BUILD_SHARED_LIBS build on mac
+17850e74 libwebp: Remove char-subscripts warning in pnmdec.c
+2fa2552d Merge "Expose WebPMalloc() in addition to WebPFree()"
+a4df4aae Expose WebPMalloc() in addition to WebPFree()
+853ea3d8 imageio/tiff: Return error before allocating bad tile size
+af650c0b Fix a Wxor-used-as-pow false positive
+601ef17c libwebp.py: update to swig 3.0.12
+0e48d889 bugfix: last alpha rows were incorrectly decoded
+24d2ccb4 webp: Fix imageio ReadPNM() TUPLTYPE
+fab8f9cf cosmetics: normalize '*' association
+94138e0e update .gitignore
+0fe1a89d update ChangeLog (tag: v1.0.3-rc1, tag: v1.0.3)
+2ad0916d update NEWS
+1287362b bump version to 1.0.3
+7b968cc2 update AUTHORS
+9d6988f4 Fix the oscillating prediction problem at low quality
+312f74d0 makefile.unix: allow *_LIBS to be overridden w/EXTRA_LIBS
+92dbf237 filters_sse2,cosmetics: shorten some long lines
+a277d197 filters_sse2.c: quiet integer sanitizer warnings
+804540f1 Fix cpufeatures in CMake.
+bf00c15b Add CMake option for bittrace.
+a788b498 filters_sse2.c: quiet integer sanitizer warnings
+e6a92c5e filters.c: quiet integer sanitizer warnings
+ec1cc40a lossless.c: remove U32 -> S8 conversion warnings
+1106478f remove conversion U32 -> S8 warnings
+812a6b49 lossless_enc: fix some conversion warning
+4627c1c9 lossless_enc,TransformColorBlue: quiet uint32_t conv warning
+c84673a6 lossless_enc_sse{2,41}: quiet signed conv warnings
+776a7757 dec_sse2: quiet signed conv warnings
+bd39c063 Merge "thread_utils: release mutex before signaling"
+0550576f Merge "(alpha_processing,enc}_sse2: quiet signed conv warnings"
+6682f2c4 thread_utils: release mutex before signaling
+e78dea75 (alpha_processing,enc}_sse2: quiet signed conv warnings
+9acf18ba iosbuild.sh: add WebP{Demux,Mux}.framework
+b9be7e65 vwebp: remove the -fit option (and make it default)
+1394a2bb Merge "README.webp_js: update Emscripten.cmake note"
+dd3e7f8a README.webp_js: update Emscripten.cmake note
+32cf8801 predictor_enc,GetBestGreenRedToBlue: quiet implicit conv warnings
+e1c8acb5 Merge "vwebp: add a -fit option"
+cbd23dd5 vwebp: add a -fit option
+2e672351 bit_writer_utils,Flush: quiet implicit conversion warnings
+1326988d swig: update libwebp_python_wrap.c
+0e7f8548 update generated swig files
+17ed1438 Merge "PutLE{16,24}: quiet implicit conversion warnings"
+24686538 PutLE{16,24}: quiet implicit conversion warnings
+153bb3a0 fix some clang-7 warnings:
+ab2dc893 Rescaler: fix rounding error
+aa65f89a HistogramCombineStochastic: fix free of uninit value
+af0bac64 Merge "encode.h: mention 'exact' default in WebPEncodeLossless*"
+6d2e11ec encode.h: mention 'exact' default in WebPEncodeLossless*
+8c3f04fe AndroidCPUInfo: reorder terms in conditional
+fcfd9c71 BitTrace: if BITTRACE is > 0, record and print syntax bits used
+067031ea Speedups for unused Huffman groups.
+01ac46ba libwebp: Display "libjpeg error:" in imageio/jpegdec
+d9a662e1 WebPRescalerGetScaledDimensions: round scaled dimension up
+62eb3f08 libwebp: Fix missing '{' in README
+e05f785a Merge "unicode,INIT_WARGV: add missing cast"
+63c9a69f tag the VP8LHashPix() function for potential uint roll-over
+2b7214ab unicode,INIT_WARGV: add missing cast
+bf424b46 tag the GetPixPairHash64() function for potential uint roll-over
+7d05d6ca Have the color cache computation be u32-bit only.
+6bcf8769 Remove BINARYEN_METHOD in wasm settings.
+2b98df90 update ChangeLog (tag: v1.0.2-rc1, tag: v1.0.2)
+61e372b7 update NEWS
+7ae658a0 bump version to 1.0.2
+51c4907d update AUTHORS
+666bd6c6 man/cwebp.1: refine near-lossless text
+561cdce5 Clarify the doc about GetFeatures.
+aec2cf02 near_lossless: fix fuzzing-detected integer overflow
+928a75de webp: Fix VP8LBitWriterClone() bug
+5173d4ee neon IsFlat
+5b081219 IsFlat: inline when possible
+381b7b54 IsFlat: use int for thresh
+6ed15ea1 fix unprobable leak in webp_sdl.c
+22bbb24e Merge "IsFlat: return int"
+8b3fb238 Merge tag 'v1.0.1'
+f435de95 IsFlat: return int
+41521aed utils.h: only define WEBP_NEED_LOG_TABLE_8BIT when needed
+9f4d4a3f neon: GetResidualCost
+0fd7514b neon: SetResidualCoeffs
+f95a996c Simpler histogram clustering.
+e85d3313 update ChangeLog (tag: v1.0.1-rc2, tag: v1.0.1)
 fa8210e4 Fix pair update in stochastic entropy merging.
+fd198f73 add codereview.settings
 825389ac README.mux: add a reference to the AnimDecoder API
 3be698c3 CMake: fix webp_js compilation
+485ff86f Fix pair update in stochastic entropy merging.
+4cd0582d CMake: fix webp_js compilation
 4cbb4caf update NEWS
 f5a5918d bump version to 1.0.1
 d61385db Speed-up: Make sure we only initialize histograms when needed.
+6752904b Speed-up: Make sure we only initialize histograms when needed.
 0c570316 update AUTHORS
 301a2dda img2webp: add help note about arguments from a file
 f0abab92 Speedups for empty histograms.
 f2dfd925 Split HistogramAdd to only have the high level logic in C.
 06b7bc7d Fix compilation on windows and clang-cl+ninja.
+b6284d82 img2webp: add help note about arguments from a file
+decf6f6b Speedups for empty histograms.
+dea3e899 Split HistogramAdd to only have the high level logic in C.
+632798ae Merge "Fix compilation on windows and clang-cl+ninja."
+dc1a9518 Merge "libwebp: Unicode command tools on Windows"
+9cf9841b libwebp: Unicode command tools on Windows
 98179495 remove some minor TODOs
+a376e7b9 Fix compilation on windows and clang-cl+ninja.
 cbf82cc0 Remove AVX2 files.
 5030e902 Merge "TIFF decoder: remove unused KINV definition"
 ac543311 Remove a few more useless #defines
@@ -455,7 +642,7 @@ dcf9d82a imageio: add limited PNM support for reading
 6524fcd6 vwebp_sdl: simple viewer based on SDL
 6cf24a24 get_disto: fix reference file read
 43d472aa Merge tag 'v0.6.0'
-50d1a848 update ChangeLog (tag: v0.6.0, origin/0.6.0, 0.6.0)
+50d1a848 update ChangeLog (tag: v0.6.0, origin/0.6.0)
 20a7fea0 extras/Makefile.am: fix libwebpextras.la reference
 415f3ffe update ChangeLog (tag: v0.6.0-rc3)
 3c6d1224 update NEWS
@@ -532,7 +719,7 @@ b016cb91 NEON: faster fancy upsampling
 f04eb376 Merge tag 'v0.5.2'
 341d711c NEON: 5% faster conversion to RGB565 and RGBA4444
 abb54827 remove Clang warnings with unused arch arguments.
-ece9684f update ChangeLog (tag: v0.5.2-rc2, tag: v0.5.2, origin/0.5.2, 0.5.2)
+ece9684f update ChangeLog (tag: v0.5.2-rc2, tag: v0.5.2, origin/0.5.2)
 aa7744ca anim_util: quiet implicit conv warnings in 32-bit
 d9120271 jpegdec: correct ContextFill signature
 24eb3940 Remove some errors when compiling the code as C++.
@@ -819,7 +1006,7 @@ bbb6ecd9 Merge "Add MSA optimized distortion functions"
 c0991a14 io,EmitRescaledAlphaYUV: factor out a common expr
 48bf5ed1 build.gradle: remove tab
 bfef6c9f Merge tag 'v0.5.1'
-3d97bb75 update ChangeLog (tag: v0.5.1, origin/0.5.1, 0.5.1)
+3d97bb75 update ChangeLog (tag: v0.5.1, origin/0.5.1)
 deb54d91 Clarify the expected 'config' lifespan in WebPIDecode()
 435308e0 Add MSA optimized encoder transform functions
 dce64bfa Add MSA optimized alpha filter functions
@@ -1013,7 +1200,7 @@ b74657fb configure: fix builtin detection w/-Werror
 6c1d7631 avoid Yoda style for comparison
 8ce975ac SSE optimization for vector mismatch.
 7db53831 Merge tag 'v0.5.0'
-37f04949 update ChangeLog (tag: v0.5.0-rc1, tag: v0.5.0, origin/0.5.0, 0.5.0)
+37f04949 update ChangeLog (tag: v0.5.0-rc1, tag: v0.5.0, origin/0.5.0)
 7e7b6ccc faster rgb565/rgb4444/argb output
 4c7f565f update NEWS
 1f62b6b2 update AUTHORS
@@ -1797,7 +1984,7 @@ b5a36cc9 add -near_lossless [0..100] experimental option
 0524d9e5 dsp: detect mips64 & disable mips32 code
 d3485d96 cwebp.1: fix quality description placement
 29a9fe22 Merge tag 'v0.4.1'
-8af27718 update ChangeLog (tag: v0.4.1, origin/0.4.1, 0.4.1)
+8af27718 update ChangeLog (tag: v0.4.1, origin/0.4.1)
 e09e9ff6 Record & log the image pre-processing time.
 f59c0b4b iosbuild.sh: specify optimization flags
 8d34ea3e update ChangeLog (tag: v0.4.1-rc1)
@@ -2182,7 +2369,7 @@ ea59a8e9 Merge "Merge tag 'v0.4.0'"
 effcb0fd Merge tag 'v0.4.0'
 7c76255d autoconf: update ax_pthread.m4
 fff2a11b make -short work with -print_ssim, -print_psnr, etc.
-68e7901d update ChangeLog (tag: v0.4.0-rc1, tag: v0.4.0, origin/0.4.0, 0.4.0)
+68e7901d update ChangeLog (tag: v0.4.0-rc1, tag: v0.4.0, origin/0.4.0)
 256e4333 update NEWS description with new general features
 29625340 Merge "gif2webp: don't use C99 %zu" into 0.4.0
 3b9f9dd0 gif2webp: don't use C99 %zu
@@ -2958,7 +3145,7 @@ a61a824b Merge "Add NULL check in chunk APIs"
 a0770727 mux struct naming
 6c66dde8 Merge "Tune Lossless encoder"
 ab5ea217 Tune Lossless encoder
-74fefc8c Update ChangeLog (tag: v0.2.1, origin/0.2.0, 0.2.0)
+74fefc8c Update ChangeLog (tag: v0.2.1, origin/0.2.0)
 92f8059c Rename some chunks:
 3bb4bbeb Merge "Mux API change:"
 d0c79f05 Mux API change:

Makefile.vc

@@ -28,7 +28,7 @@ PLATFORM_LDFLAGS = /SAFESEH
 
 NOLOGO     = /nologo
 CCNODBG    = cl.exe $(NOLOGO) /O2 /DNDEBUG
-CCDEBUG    = cl.exe $(NOLOGO) /Od /Gm /Zi /D_DEBUG /RTC1
+CCDEBUG    = cl.exe $(NOLOGO) /Od /Zi /D_DEBUG /RTC1
 CFLAGS     = /I. /Isrc $(NOLOGO) /W3 /EHsc /c
 CFLAGS     = $(CFLAGS) /DWIN32 /D_CRT_SECURE_NO_WARNINGS /DWIN32_LEAN_AND_MEAN
 LDFLAGS    = /LARGEADDRESSAWARE /MANIFEST /NXCOMPAT /DYNAMICBASE
@@ -129,12 +129,16 @@ LIBWEBP_PDBNAME = $(DIROBJ)\$(LIBWEBP_BASENAME)_dll.pdb
 CFGSET = TRUE
 !ENDIF
 
+!IF "$(UNICODE)" == "1"
+CFLAGS = $(CFLAGS) /D_UNICODE /DUNICODE
+!ENDIF
+
 #######################
 # Usage
 #
 !IF "$(CFGSET)" == "FALSE"
 !MESSAGE Usage: nmake /f Makefile.vc [CFG=<config>]
-!MESSAGE .          [OBJDIR=<path>] [RTLIBCFG=<rtlib>] [<target>]
+!MESSAGE .          [OBJDIR=<path>] [RTLIBCFG=<rtlib>] [UNICODE=1] [<target>]
 !MESSAGE
 !MESSAGE where <config> is one of:
 !MESSAGE -  release-static                - release static library
@@ -234,6 +238,7 @@ DSP_ENC_OBJS = \
     $(DIROBJ)\dsp\cost.obj \
     $(DIROBJ)\dsp\cost_mips32.obj \
     $(DIROBJ)\dsp\cost_mips_dsp_r2.obj \
+    $(DIROBJ)\dsp\cost_neon.obj \
     $(DIROBJ)\dsp\cost_sse2.obj \
     $(DIROBJ)\dsp\enc.obj \
     $(DIROBJ)\dsp\enc_mips32.obj \
@@ -386,7 +391,13 @@ $(DIRBIN)\get_disto.exe: $(IMAGEIO_DEC_OBJS) $(IMAGEIO_UTIL_OBJS)
 $(DIRBIN)\get_disto.exe: $(LIBWEBPDEMUX) $(LIBWEBP)
 $(DIRBIN)\webp_quality.exe: $(DIROBJ)\extras\webp_quality.obj
 $(DIRBIN)\webp_quality.exe: $(IMAGEIO_UTIL_OBJS)
-$(DIRBIN)\webp_quality.exe: $(EXTRAS_OBJS) $(LIBWEBP)
+$(DIRBIN)\webp_quality.exe: $(EXTRAS_OBJS)
+# EXTRA_OBJS requires private symbols from dsp. Explicitly add those when
+# building libwebp as a dll.
+!IF "$(DLLBUILD)" == "TRUE"
+$(DIRBIN)\webp_quality.exe: $(DSP_DEC_OBJS)
+!ENDIF
+$(DIRBIN)\webp_quality.exe: $(LIBWEBP)
 $(DIRBIN)\webpinfo.exe: $(DIROBJ)\examples\webpinfo.obj
 $(DIRBIN)\webpinfo.exe: $(IMAGEIO_DEC_OBJS)
 $(DIRBIN)\webpinfo.exe: $(EX_UTIL_OBJS) $(IMAGEIO_UTIL_OBJS)
@@ -483,15 +494,18 @@ $(DIROBJ)\examples\gifdec.obj: examples\gifdec.c
 {src\utils}.c{$(DIROBJ)\utils}.obj::
 	$(CC) $(CFLAGS) /Fd$(LIBWEBP_PDBNAME) /Fo$(DIROBJ)\utils\ $<
 
+LNKLIBS     = ole32.lib windowscodecs.lib shlwapi.lib
+!IF "$(UNICODE)" == "1"
+LNKLIBS     = $(LNKLIBS) Shell32.lib
+!ENDIF
+
 {$(DIROBJ)\examples}.obj{$(DIRBIN)}.exe:
-	$(LNKEXE) $(LDFLAGS) /OUT:$@ $** \
-	    ole32.lib windowscodecs.lib shlwapi.lib
+	$(LNKEXE) $(LDFLAGS) /OUT:$@ $** $(LNKLIBS)
 	$(MT) -manifest $@.manifest -outputresource:$@;1
 	del $@.manifest
 
 {$(DIROBJ)\extras}.obj{$(DIRBIN)}.exe:
-	$(LNKEXE) $(LDFLAGS) /OUT:$@ $** \
-	    ole32.lib windowscodecs.lib shlwapi.lib
+	$(LNKEXE) $(LDFLAGS) /OUT:$@ $** $(LNKLIBS)
 	$(MT) -manifest $@.manifest -outputresource:$@;1
 	del $@.manifest
 

NEWS

@@ -1,3 +1,46 @@
+- 12/23/2020: version 1.2.0
+  * API changes:
+    - libwebp:
+      encode.h: add a qmin / qmax range for quality factor (cwebp adds -qrange)
+  * lossless encoder improvements
+  * SIMD support for Wasm builds
+  * add xcframeworkbuild.sh, supports Mac Catalyst builds
+  * import fuzzers from oss-fuzz & chromium (#409)
+  * webpmux: add an '-set loop <value>' option (#494)
+  * toolchain updates and bug fixes (#449, #463, #470, #475, #477, #478, #479,
+    #488, #491)
+
+- 12/18/2019: version 1.1.0
+  * API changes:
+    - libwebp:
+      WebPMalloc (issue #442)
+    - extras:
+      WebPUnmultiplyARGB
+  * alpha decode fix (issue #439)
+  * toolchain updates and bug fixes
+    (chromium: #1026858, #1027136, #1027409, #1028620, #1028716, #995200)
+    (oss-fuzz: #19430, #19447)
+
+- 7/4/2019: version 1.0.3
+  This is a binary compatible release.
+  * resize fixes for Nx1 sizes and the addition of non-opaque alpha values for
+    odd sizes (issues #418, #434)
+  * lossless encode/decode performance improvements
+  * lossy compression performance improvement at low quality levels with flat
+    content (issue #432)
+  * python swig files updated to support python 3
+  Tool updates:
+    vwebp will now preserve the aspect ratio of images that exceed monitor
+    resolution by scaling the image to fit (issue #433)
+
+- 1/14/2019: version 1.0.2
+  This is a binary compatible release.
+  * (Windows) unicode file support in the tools (linux and mac already had
+    support, issue #398)
+  * lossless encoder speedups
+  * lossy encoder speedup on ARM
+  * lossless multi-threaded security fix (chromium:917029)
+
 - 11/2/2018: version 1.0.1
   This is a binary compatible release.
   * lossless encoder speedups

README

@@ -4,7 +4,7 @@
           \__\__/\____/\_____/__/ ____  ___
                 / _/ /    \    \ /  _ \/ _/
                /  \_/   / /   \ \   __/  \__
-               \____/____/\_____/_____/____/v1.0.1
+               \____/____/\_____/_____/____/v1.2.0
 
 Description:
 ============
@@ -113,7 +113,7 @@ make install
 
 CMake:
 ------
-With CMake, you can compile libwebp, cwebp, dwebp, gif2web, img2webp, webpinfo
+With CMake, you can compile libwebp, cwebp, dwebp, gif2webp, img2webp, webpinfo
 and the JS bindings.
 
 Prerequisites:
@@ -136,6 +136,8 @@ cmake -DWEBP_BUILD_CWEBP=ON -DWEBP_BUILD_DWEBP=ON ../
 
 or through your favorite interface (like ccmake or cmake-qt-gui).
 
+Use option -DWEBP_UNICODE=ON for Unicode support on Windows (with chcp 65001).
+
 Finally, once installed, you can also use WebP in your CMake project by doing:
 
 find_package(WebP)
@@ -223,6 +225,7 @@ Usage:
 
 If input size (-s) for an image is not specified, it is
 assumed to be a PNG, JPEG, TIFF or WebP file.
+Note: Animated PNG and WebP files are not supported.
 
 Options:
   -h / -help ............. short help
@@ -252,6 +255,8 @@ Options:
   -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)
+  -qrange <min> <max> .... specifies the permissible quality range
+                           (default: 0 100)
   -crop <x> <y> <w> <h> .. crop picture with the given rectangle
   -resize <w> <h> ........ resize picture (after any cropping)
   -mt .................... use multi-threading if available
@@ -292,6 +297,7 @@ Experimental Options:
   -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
@@ -339,7 +345,9 @@ The full list of options is available using -h:
 > dwebp -h
 Usage: dwebp in_file [options] [-o out_file]
 
-Decodes the WebP image file to PNG format [Default]
+Decodes the WebP image file to PNG format [Default].
+Note: Animated WebP files are not supported.
+
 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
@@ -421,7 +429,7 @@ Prerequisites:
 1) OpenGL & OpenGL Utility Toolkit (GLUT)
   Linux:
     $ sudo apt-get install freeglut3-dev mesa-common-dev
-  Mac + XCode:
+  Mac + Xcode:
     - These libraries should be available in the OpenGL / GLUT frameworks.
   Windows:
     http://freeglut.sourceforge.net/index.php#download
@@ -595,7 +603,7 @@ The encoding flow looks like:
   // Setup a config, starting form a preset and tuning some additional
   // parameters
   WebPConfig config;
-  if (!WebPConfigPreset(&config, WEBP_PRESET_PHOTO, quality_factor))
+  if (!WebPConfigPreset(&config, WEBP_PRESET_PHOTO, quality_factor)) {
     return 0;   // version error
   }
   // ... additional tuning

README.mux

@@ -1,7 +1,7 @@
           __   __  ____  ____  ____  __ __  _     __ __
          /  \\/  \/  _ \/  _ \/  _ \/  \  \/ \___/_ / _\
          \       /   __/  _  \   __/      /  /  (_/  /__
-          \__\__/\_____/_____/__/  \__//_/\_____/__/___/v1.0.1
+          \__\__/\_____/_____/__/  \__//_/\_____/__/___/v1.2.0
 
 
 Description:
@@ -44,6 +44,7 @@ GET_OPTIONS:
 
 SET_OPTIONS:
  Set color profile/metadata:
+   loop LOOP_COUNT   set the loop count
    icc  file.icc     set ICC profile
    exif file.exif    set EXIF metadata
    xmp  file.xmp     set XMP metadata

README.webp_js

@@ -17,19 +17,22 @@ using Emscripten and CMake.
 
  - make sure the file $EMSCRIPTEN/cmake/Modules/Platform/Emscripten.cmake is
    accessible. This is the toolchain file used by CMake to invoke Emscripten.
+   If $EMSCRIPTEN is unset search for Emscripten.cmake under $EMSDK and set
+   $EMSCRIPTEN accordingly, for example:
+   unix-like environments: export EMSCRIPTEN=$EMSDK/upstream/emscripten
+   windows: set EMSCRIPTEN=%EMSDK%\upstream\emscripten
 
  - configure the project 'WEBP_JS' with CMake using:
 
  cd webp_js && \
  cmake -DWEBP_BUILD_WEBP_JS=ON \
-       -DEMSCRIPTEN_GENERATE_BITCODE_STATIC_LIBRARIES=1 \
        -DCMAKE_TOOLCHAIN_FILE=$EMSCRIPTEN/cmake/Modules/Platform/Emscripten.cmake \
        ../
 
  - compile webp.js using 'make'.
 
  - that's it! Upon completion, you should have the webp.js and
-   webp.js.mem files generated.
+   webp.wasm files generated.
 
 The callable JavaScript function is WebPToSDL(), which decodes a raw WebP
 bitstream into a canvas. See webp_js/index.html for a simple usage sample
@@ -55,8 +58,9 @@ Web-Assembly (WASM) version:
   See webp_js/index_wasm.html for a simple demo page using the WASM version
   of the library.
 
-  You will need a fairly recent version of Emscripten (at least 1.37.8) and of
-  your WASM-enabled browser to run this version. Consider it very experimental!
+  You will need a fairly recent version of Emscripten (at least 1.37.8,
+  latest-upstream is recommended) and of your WASM-enabled browser to run this
+  version.
 
 Caveat:
 =======
@@ -74,3 +78,6 @@ Caveat:
     https://github.com/kripken/emscripten/issues/3788
 
     Therefore, SSE2 optimization is currently disabled in CMakeLists.txt.
+
+  - If WEBP_ENABLE_SIMD is set to 1 the JavaScript version (webp.js) will be
+    disabled as wasm2js does not support SIMD.

build.gradle

@@ -173,6 +173,7 @@ model {
             include "cost.c"
             include "cost_mips32.c"
             include "cost_mips_dsp_r2.c"
+            include "cost_neon.$NEON"
             include "cost_sse2.c"
             include "enc.c"
             include "enc_mips32.c"
@@ -431,8 +432,3 @@ model {
     }
   }
 }
-
-// Task to generate the wrapper.
-task wrapper(type: Wrapper) {
-  gradleVersion = '2.13'
-}

cmake/config.h.in

@@ -93,10 +93,6 @@
 /* Define to the version of this package. */
 #cmakedefine PACKAGE_VERSION "@PACKAGE_VERSION@"
 
-/* Define to necessary symbol if this constant uses a non-standard name on
-   your system. */
-#cmakedefine PTHREAD_CREATE_JOINABLE 1
-
 /* Define to 1 if you have the ANSI C header files. */
 #cmakedefine STDC_HEADERS 1
 

cmake/cpu.cmake

@@ -31,9 +31,17 @@ endfunction()
 set(WEBP_SIMD_FLAGS "SSE41;SSE2;MIPS32;MIPS_DSP_R2;NEON;MSA")
 set(WEBP_SIMD_FILE_EXTENSIONS
     "_sse41.c;_sse2.c;_mips32.c;_mips_dsp_r2.c;_neon.c;_msa.c")
-if(MSVC)
-  # MSVC does not have a SSE4 flag but AVX support implies SSE4 support.
-  set(SIMD_ENABLE_FLAGS "/arch:AVX;/arch:SSE2;;;;")
+if(MSVC AND CMAKE_C_COMPILER_ID STREQUAL "MSVC")
+  # With at least Visual Studio 12 (2013)+ /arch is not necessary to build SSE2
+  # or SSE4 code unless a lesser /arch is forced. MSVC does not have a SSE4
+  # flag, but an AVX one. Using that with SSE4 code risks generating illegal
+  # instructions when used on machines with SSE4 only. The flags are left for
+  # older (untested) versions to avoid any potential compatibility issues.
+  if(MSVC_VERSION GREATER_EQUAL 1800 AND NOT CMAKE_C_FLAGS MATCHES "/arch:")
+    set(SIMD_ENABLE_FLAGS)
+  else()
+    set(SIMD_ENABLE_FLAGS "/arch:AVX;/arch:SSE2;;;;")
+  endif()
   set(SIMD_DISABLE_FLAGS)
 else()
   set(SIMD_ENABLE_FLAGS
@@ -57,9 +65,14 @@ endif()
 
 list(LENGTH WEBP_SIMD_FLAGS WEBP_SIMD_FLAGS_LENGTH)
 math(EXPR WEBP_SIMD_FLAGS_RANGE "${WEBP_SIMD_FLAGS_LENGTH} - 1")
-unset(HIGHEST_SSE_FLAG)
 
 foreach(I_SIMD RANGE ${WEBP_SIMD_FLAGS_RANGE})
+  # With Emscripten 2.0.9 -msimd128 -mfpu=neon will enable NEON, but the
+  # source will fail to compile.
+  if(EMSCRIPTEN AND ${I_SIMD} GREATER_EQUAL 2)
+    break()
+  endif()
+
   list(GET WEBP_SIMD_FLAGS ${I_SIMD} WEBP_SIMD_FLAG)
 
   # First try with no extra flag added as the compiler might have default flags
@@ -70,10 +83,15 @@ foreach(I_SIMD RANGE ${WEBP_SIMD_FLAGS_RANGE})
   webp_check_compiler_flag(${WEBP_SIMD_FLAG} ${WEBP_ENABLE_SIMD})
   if(NOT WEBP_HAVE_${WEBP_SIMD_FLAG})
     list(GET SIMD_ENABLE_FLAGS ${I_SIMD} SIMD_COMPILE_FLAG)
+    if(EMSCRIPTEN)
+      set(SIMD_COMPILE_FLAG "-msimd128 ${SIMD_COMPILE_FLAG}")
+    endif()
     set(CMAKE_REQUIRED_FLAGS ${SIMD_COMPILE_FLAG})
     webp_check_compiler_flag(${WEBP_SIMD_FLAG} ${WEBP_ENABLE_SIMD})
   else()
-    if(MSVC)
+    if(MSVC AND SIMD_ENABLE_FLAGS)
+      # The detection for SSE2/SSE4 support under MSVC is based on the compiler
+      # version so e.g., clang-cl will require flags to enable the assembly.
       list(GET SIMD_ENABLE_FLAGS ${I_SIMD} SIMD_COMPILE_FLAG)
     else()
       set(SIMD_COMPILE_FLAG " ")
@@ -84,17 +102,10 @@ foreach(I_SIMD RANGE ${WEBP_SIMD_FLAGS_RANGE})
   file(GLOB SIMD_FILES "${CMAKE_CURRENT_LIST_DIR}/../"
             "src/dsp/*${WEBP_SIMD_FILE_EXTENSION}")
   if(WEBP_HAVE_${WEBP_SIMD_FLAG})
-    if(${I_SIMD} LESS 2 AND NOT HIGHEST_SSE_FLAG)
-      set(HIGHEST_SSE_FLAG ${SIMD_COMPILE_FLAG})
-    endif()
     # Memorize the file and flags.
     foreach(FILE ${SIMD_FILES})
       list(APPEND WEBP_SIMD_FILES_TO_INCLUDE ${FILE})
-      if(${I_SIMD} LESS 2)
-        list(APPEND WEBP_SIMD_FLAGS_TO_INCLUDE ${HIGHEST_SSE_FLAG})
-      else()
-        list(APPEND WEBP_SIMD_FLAGS_TO_INCLUDE ${SIMD_COMPILE_FLAG})
-      endif()
+      list(APPEND WEBP_SIMD_FLAGS_TO_INCLUDE ${SIMD_COMPILE_FLAG})
     endforeach()
   else()
     # Remove the file from the list.

cmake/deps.cmake

@@ -22,20 +22,23 @@ check_c_source_compiles("
   " HAVE_BUILTIN_BSWAP64)
 
 # Check for libraries.
-find_package(Threads)
+if(NOT WEBP_BUILD_WEBP_JS)
+  # Disable pThreads for WASM.
+  find_package(Threads)
+endif()
 if(Threads_FOUND)
-  if(CMAKE_USE_PTHREADS_INIT)
+  # work around cmake bug on QNX (https://cmake.org/Bug/view.php?id=11333)
+  if(CMAKE_USE_PTHREADS_INIT AND NOT CMAKE_SYSTEM_NAME STREQUAL "QNX")
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -pthread")
   endif()
-  foreach(PTHREAD_TEST HAVE_PTHREAD_PRIO_INHERIT PTHREAD_CREATE_UNDETACHED)
-    check_c_source_compiles("
-        #include <pthread.h>
-        int main (void) {
-          int attr = ${PTHREAD_TEST};
-          return attr;
-        }
-      " ${PTHREAD_TEST})
-  endforeach()
+  check_c_source_compiles("
+      #include <pthread.h>
+      int main (void) {
+        int attr = PTHREAD_PRIO_INHERIT;
+        return attr;
+      }
+    " FLAG_HAVE_PTHREAD_PRIO_INHERIT)
+  set(HAVE_PTHREAD_PRIO_INHERIT ${FLAG_HAVE_PTHREAD_PRIO_INHERIT})
   list(APPEND WEBP_DEP_LIBRARIES ${CMAKE_THREAD_LIBS_INIT})
 endif()
 set(WEBP_USE_THREAD ${Threads_FOUND})

codereview.settings

@@ -0,0 +1,4 @@
+# This file is used by git cl to get repository specific information.
+GERRIT_HOST: True
+CODE_REVIEW_SERVER: chromium-review.googlesource.com
+GERRIT_SQUASH_UPLOADS: False

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([libwebp], [1.0.1],
+AC_INIT([libwebp], [1.2.0],
         [https://bugs.chromium.org/p/webp],,
         [http://developers.google.com/speed/webp])
 AC_CANONICAL_HOST
@@ -80,6 +80,7 @@ TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wparentheses-equality])
 TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wshadow])
 TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wshorten-64-to-32])
 TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wundef])
+TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wunreachable-code-aggressive])
 TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wunreachable-code])
 TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wunused-but-set-variable])
 TEST_AND_ADD_CFLAGS([AM_CFLAGS], [-Wunused])

examples/anim_diff.c

@@ -21,6 +21,7 @@
 
 #include "./anim_util.h"
 #include "./example_util.h"
+#include "./unicode.h"
 
 #if defined(_MSC_VER) && _MSC_VER < 1900
 #define snprintf _snprintf
@@ -218,12 +219,14 @@ int main(int argc, const char* argv[]) {
   const char* files[2] = { NULL, NULL };
   AnimatedImage images[2];
 
+  INIT_WARGV(argc, argv);
+
   for (c = 1; c < argc; ++c) {
     int parse_error = 0;
     if (!strcmp(argv[c], "-dump_frames")) {
       if (c < argc - 1) {
         dump_frames = 1;
-        dump_folder = argv[++c];
+        dump_folder = (const char*)GET_WARGV(argv, ++c);
       } else {
         parse_error = 1;
       }
@@ -243,7 +246,7 @@ int main(int argc, const char* argv[]) {
       }
     } else if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       Help();
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-version")) {
       int dec_version, demux_version;
       GetAnimatedImageVersions(&dec_version, &demux_version);
@@ -252,13 +255,13 @@ int main(int argc, const char* argv[]) {
              (dec_version >> 0) & 0xff,
              (demux_version >> 16) & 0xff, (demux_version >> 8) & 0xff,
              (demux_version >> 0) & 0xff);
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else {
       if (!got_input1) {
-        files[0] = argv[c];
+        files[0] = (const char*)GET_WARGV(argv, c);
         got_input1 = 1;
       } else if (!got_input2) {
-        files[1] = argv[c];
+        files[1] = (const char*)GET_WARGV(argv, c);
         got_input2 = 1;
       } else {
         parse_error = 1;
@@ -266,29 +269,30 @@ int main(int argc, const char* argv[]) {
     }
     if (parse_error) {
       Help();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     }
   }
   if (argc < 3) {
     Help();
-    return -1;
+    FREE_WARGV_AND_RETURN(-1);
   }
 
 
   if (!got_input2) {
     Help();
-    return -1;
+    FREE_WARGV_AND_RETURN(-1);
   }
 
   if (dump_frames) {
-    printf("Dumping decoded frames in: %s\n", dump_folder);
+    WPRINTF("Dumping decoded frames in: %s\n", (const W_CHAR*)dump_folder);
   }
 
   memset(images, 0, sizeof(images));
   for (i = 0; i < 2; ++i) {
-    printf("Decoding file: %s\n", files[i]);
+    WPRINTF("Decoding file: %s\n", (const W_CHAR*)files[i]);
     if (!ReadAnimatedImage(files[i], &images[i], dump_frames, dump_folder)) {
-      fprintf(stderr, "Error decoding file: %s\n Aborting.\n", files[i]);
+      WFPRINTF(stderr, "Error decoding file: %s\n Aborting.\n",
+               (const W_CHAR*)files[i]);
       return_code = -2;
       goto End;
     } else {
@@ -298,14 +302,16 @@ int main(int argc, const char* argv[]) {
 
   if (!CompareAnimatedImagePair(&images[0], &images[1],
                                 premultiply, min_psnr)) {
-    fprintf(stderr, "\nFiles %s and %s differ.\n", files[0], files[1]);
+    WFPRINTF(stderr, "\nFiles %s and %s differ.\n", (const W_CHAR*)files[0],
+             (const W_CHAR*)files[1]);
     return_code = -3;
   } else {
-    printf("\nFiles %s and %s are identical.\n", files[0], files[1]);
+    WPRINTF("\nFiles %s and %s are identical.\n", (const W_CHAR*)files[0],
+            (const W_CHAR*)files[1]);
     return_code = 0;
   }
  End:
   ClearAnimatedImage(&images[0]);
   ClearAnimatedImage(&images[1]);
-  return return_code;
+  FREE_WARGV_AND_RETURN(return_code);
 }

examples/anim_dump.c

@@ -17,6 +17,7 @@
 #include "./anim_util.h"
 #include "webp/decode.h"
 #include "../imageio/image_enc.h"
+#include "./unicode.h"
 
 #if defined(_MSC_VER) && _MSC_VER < 1900
 #define snprintf _snprintf
@@ -36,15 +37,17 @@ static void Help(void) {
 
 int main(int argc, const char* argv[]) {
   int error = 0;
-  const char* dump_folder = ".";
-  const char* prefix = "dump_";
-  const char* suffix = "png";
+  const W_CHAR* dump_folder = TO_W_CHAR(".");
+  const W_CHAR* prefix = TO_W_CHAR("dump_");
+  const W_CHAR* suffix = TO_W_CHAR("png");
   WebPOutputFileFormat format = PNG;
   int c;
 
+  INIT_WARGV(argc, argv);
+
   if (argc < 2) {
     Help();
-    return -1;
+    FREE_WARGV_AND_RETURN(-1);
   }
 
   for (c = 1; !error && c < argc; ++c) {
@@ -54,23 +57,23 @@ int main(int argc, const char* argv[]) {
         error = 1;
         break;
       }
-      dump_folder = argv[++c];
+      dump_folder = GET_WARGV(argv, ++c);
     } else if (!strcmp(argv[c], "-prefix")) {
       if (c + 1 == argc) {
         fprintf(stderr, "missing argument after option '%s'\n", argv[c]);
         error = 1;
         break;
       }
-      prefix = argv[++c];
+      prefix = GET_WARGV(argv, ++c);
     } else if (!strcmp(argv[c], "-tiff")) {
       format = TIFF;
-      suffix = "tiff";
+      suffix = TO_W_CHAR("tiff");
     } else if (!strcmp(argv[c], "-pam")) {
       format = PAM;
-      suffix = "pam";
+      suffix = TO_W_CHAR("pam");
     } else if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       Help();
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-version")) {
       int dec_version, demux_version;
       GetAnimatedImageVersions(&dec_version, &demux_version);
@@ -79,21 +82,21 @@ int main(int argc, const char* argv[]) {
              (dec_version >> 0) & 0xff,
              (demux_version >> 16) & 0xff, (demux_version >> 8) & 0xff,
              (demux_version >> 0) & 0xff);
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else {
       uint32_t i;
       AnimatedImage image;
-      const char* const file = argv[c];
+      const W_CHAR* const file = GET_WARGV(argv, c);
       memset(&image, 0, sizeof(image));
-      printf("Decoding file: %s as %s/%sxxxx.%s\n",
-             file, dump_folder, prefix, suffix);
-      if (!ReadAnimatedImage(file, &image, 0, NULL)) {
-        fprintf(stderr, "Error decoding file: %s\n Aborting.\n", file);
+      WPRINTF("Decoding file: %s as %s/%sxxxx.%s\n",
+              file, dump_folder, prefix, suffix);
+      if (!ReadAnimatedImage((const char*)file, &image, 0, NULL)) {
+        WFPRINTF(stderr, "Error decoding file: %s\n Aborting.\n", file);
         error = 1;
         break;
       }
       for (i = 0; !error && i < image.num_frames; ++i) {
-        char out_file[1024];
+        W_CHAR out_file[1024];
         WebPDecBuffer buffer;
         WebPInitDecBuffer(&buffer);
         buffer.colorspace = MODE_RGBA;
@@ -103,10 +106,10 @@ int main(int argc, const char* argv[]) {
         buffer.u.RGBA.rgba = image.frames[i].rgba;
         buffer.u.RGBA.stride = buffer.width * sizeof(uint32_t);
         buffer.u.RGBA.size = buffer.u.RGBA.stride * buffer.height;
-        snprintf(out_file, sizeof(out_file), "%s/%s%.4d.%s",
-                 dump_folder, prefix, i, suffix);
-        if (!WebPSaveImage(&buffer, format, out_file)) {
-          fprintf(stderr, "Error while saving image '%s'\n", out_file);
+        WSNPRINTF(out_file, sizeof(out_file), "%s/%s%.4d.%s",
+                  dump_folder, prefix, i, suffix);
+        if (!WebPSaveImage(&buffer, format, (const char*)out_file)) {
+          WFPRINTF(stderr, "Error while saving image '%s'\n", out_file);
           error = 1;
         }
         WebPFreeDecBuffer(&buffer);
@@ -114,5 +117,5 @@ int main(int argc, const char* argv[]) {
       ClearAnimatedImage(&image);
     }
   }
-  return error ? 1 : 0;
+  FREE_WARGV_AND_RETURN(error ? 1 : 0);
 }

examples/anim_util.c

@@ -24,6 +24,8 @@
 #include "webp/demux.h"
 #include "../imageio/imageio_util.h"
 #include "./gifdec.h"
+#include "./unicode.h"
+#include "./unicode_gif.h"
 
 #if defined(_MSC_VER) && _MSC_VER < 1900
 #define snprintf _snprintf
@@ -58,15 +60,15 @@ static int AllocateFrames(AnimatedImage* const image, uint32_t num_frames) {
       !CheckSizeForOverflow(total_frame_size)) {
     return 0;
   }
-  mem = (uint8_t*)malloc((size_t)total_size);
-  frames = (DecodedFrame*)malloc((size_t)total_frame_size);
+  mem = (uint8_t*)WebPMalloc((size_t)total_size);
+  frames = (DecodedFrame*)WebPMalloc((size_t)total_frame_size);
 
   if (mem == NULL || frames == NULL) {
-    free(mem);
-    free(frames);
+    WebPFree(mem);
+    WebPFree(frames);
     return 0;
   }
-  free(image->raw_mem);
+  WebPFree(image->raw_mem);
   image->num_frames = num_frames;
   image->frames = frames;
   for (i = 0; i < num_frames; ++i) {
@@ -80,8 +82,8 @@ static int AllocateFrames(AnimatedImage* const image, uint32_t num_frames) {
 
 void ClearAnimatedImage(AnimatedImage* const image) {
   if (image != NULL) {
-    free(image->raw_mem);
-    free(image->frames);
+    WebPFree(image->raw_mem);
+    WebPFree(image->frames);
     image->num_frames = 0;
     image->frames = NULL;
     image->raw_mem = NULL;
@@ -152,42 +154,42 @@ static int DumpFrame(const char filename[], const char dump_folder[],
   int ok = 0;
   size_t max_len;
   int y;
-  const char* base_name = NULL;
-  char* file_name = NULL;
+  const W_CHAR* base_name = NULL;
+  W_CHAR* file_name = NULL;
   FILE* f = NULL;
   const char* row;
 
-  if (dump_folder == NULL) dump_folder = ".";
+  if (dump_folder == NULL) dump_folder = (const char*)TO_W_CHAR(".");
 
-  base_name = strrchr(filename, '/');
-  base_name = (base_name == NULL) ? filename : base_name + 1;
-  max_len = strlen(dump_folder) + 1 + strlen(base_name)
+  base_name = WSTRRCHR(filename, '/');
+  base_name = (base_name == NULL) ? (const W_CHAR*)filename : base_name + 1;
+  max_len = WSTRLEN(dump_folder) + 1 + WSTRLEN(base_name)
           + strlen("_frame_") + strlen(".pam") + 8;
-  file_name = (char*)malloc(max_len * sizeof(*file_name));
+  file_name = (W_CHAR*)WebPMalloc(max_len * sizeof(*file_name));
   if (file_name == NULL) goto End;
 
-  if (snprintf(file_name, max_len, "%s/%s_frame_%d.pam",
-               dump_folder, base_name, frame_num) < 0) {
+  if (WSNPRINTF(file_name, max_len, "%s/%s_frame_%d.pam",
+                (const W_CHAR*)dump_folder, base_name, frame_num) < 0) {
     fprintf(stderr, "Error while generating file name\n");
     goto End;
   }
 
-  f = fopen(file_name, "wb");
+  f = WFOPEN(file_name, "wb");
   if (f == NULL) {
-    fprintf(stderr, "Error opening file for writing: %s\n", file_name);
+    WFPRINTF(stderr, "Error opening file for writing: %s\n", file_name);
     ok = 0;
     goto End;
   }
   if (fprintf(f, "P7\nWIDTH %d\nHEIGHT %d\n"
               "DEPTH 4\nMAXVAL 255\nTUPLTYPE RGB_ALPHA\nENDHDR\n",
               canvas_width, canvas_height) < 0) {
-    fprintf(stderr, "Write error for file %s\n", file_name);
+    WFPRINTF(stderr, "Write error for file %s\n", file_name);
     goto End;
   }
   row = (const char*)rgba;
   for (y = 0; y < canvas_height; ++y) {
     if (fwrite(row, canvas_width * kNumChannels, 1, f) != 1) {
-      fprintf(stderr, "Error writing to file: %s\n", file_name);
+      WFPRINTF(stderr, "Error writing to file: %s\n", file_name);
       goto End;
     }
     row += canvas_width * kNumChannels;
@@ -195,7 +197,7 @@ static int DumpFrame(const char filename[], const char dump_folder[],
   ok = 1;
  End:
   if (f != NULL) fclose(f);
-  free(file_name);
+  WebPFree(file_name);
   return ok;
 }
 
@@ -223,7 +225,7 @@ static int ReadAnimatedWebP(const char filename[],
 
   dec = WebPAnimDecoderNew(webp_data, NULL);
   if (dec == NULL) {
-    fprintf(stderr, "Error parsing image: %s\n", filename);
+    WFPRINTF(stderr, "Error parsing image: %s\n", (const W_CHAR*)filename);
     goto End;
   }
 
@@ -403,7 +405,7 @@ static uint32_t GetBackgroundColorGIF(GifFileType* gif) {
     return 0xffffffff;  // Invalid: assume white.
   } else {
     const GifColorType color = color_map->Colors[gif->SBackGroundColor];
-    return (0xff << 24) |
+    return (0xffu << 24) |
            (color.Red << 16) |
            (color.Green << 8) |
            (color.Blue << 0);
@@ -511,15 +513,15 @@ static int ReadAnimatedGIF(const char filename[], AnimatedImage* const image,
   int gif_error;
   GifFileType* gif;
 
-  gif = DGifOpenFileName(filename, NULL);
+  gif = DGifOpenFileUnicode((const W_CHAR*)filename, NULL);
   if (gif == NULL) {
-    fprintf(stderr, "Could not read file: %s.\n", filename);
+    WFPRINTF(stderr, "Could not read file: %s.\n", (const W_CHAR*)filename);
     return 0;
   }
 
   gif_error = DGifSlurp(gif);
   if (gif_error != GIF_OK) {
-    fprintf(stderr, "Could not parse image: %s.\n", filename);
+    WFPRINTF(stderr, "Could not parse image: %s.\n", (const W_CHAR*)filename);
     GIFDisplayError(gif, gif_error);
     DGifCloseFile(gif, NULL);
     return 0;
@@ -705,7 +707,7 @@ int ReadAnimatedImage(const char filename[], AnimatedImage* const image,
   memset(image, 0, sizeof(*image));
 
   if (!ImgIoUtilReadFile(filename, &webp_data.bytes, &webp_data.size)) {
-    fprintf(stderr, "Error reading file: %s\n", filename);
+    WFPRINTF(stderr, "Error reading file: %s\n", (const W_CHAR*)filename);
     return 0;
   }
 
@@ -715,9 +717,9 @@ int ReadAnimatedImage(const char filename[], AnimatedImage* const image,
   } else if (IsGIF(&webp_data)) {
     ok = ReadAnimatedGIF(filename, image, dump_frames, dump_folder);
   } else {
-    fprintf(stderr,
-            "Unknown file type: %s. Supported file types are WebP and GIF\n",
-            filename);
+    WFPRINTF(stderr,
+             "Unknown file type: %s. Supported file types are WebP and GIF\n",
+             (const W_CHAR*)filename);
     ok = 0;
   }
   if (!ok) ClearAnimatedImage(image);

examples/cwebp.c

@@ -12,6 +12,7 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
+#include <assert.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -23,7 +24,9 @@
 #include "../examples/example_util.h"
 #include "../imageio/image_dec.h"
 #include "../imageio/imageio_util.h"
+#include "../imageio/webpdec.h"
 #include "./stopwatch.h"
+#include "./unicode.h"
 #include "webp/encode.h"
 
 #ifndef WEBP_DLL
@@ -88,9 +91,10 @@ static int ReadPicture(const char* const filename, WebPPicture* const pic,
     }
   }
   if (!ok) {
-    fprintf(stderr, "Error! Could not process file %s\n", filename);
+    WFPRINTF(stderr, "Error! Could not process file %s\n",
+             (const W_CHAR*)filename);
   }
-  free((void*)data);
+  WebPFree((void*)data);
   return ok;
 }
 
@@ -114,9 +118,10 @@ static int ReadPicture(const char* const filename, WebPPicture* const pic,
   }
  End:
   if (!ok) {
-    fprintf(stderr, "Error! Could not process file %s\n", filename);
+    WFPRINTF(stderr, "Error! Could not process file %s\n",
+             (const W_CHAR*)filename);
   }
-  free((void*)data);
+  WebPFree((void*)data);
   return ok;
 }
 
@@ -125,7 +130,8 @@ static int ReadPicture(const char* const filename, WebPPicture* const pic,
 static void AllocExtraInfo(WebPPicture* const pic) {
   const int mb_w = (pic->width + 15) / 16;
   const int mb_h = (pic->height + 15) / 16;
-  pic->extra_info = (uint8_t*)malloc(mb_w * mb_h * sizeof(*pic->extra_info));
+  pic->extra_info =
+      (uint8_t*)WebPMalloc(mb_w * mb_h * sizeof(*pic->extra_info));
 }
 
 static void PrintByteCount(const int bytes[4], int total_size,
@@ -144,7 +150,7 @@ static void PrintPercents(const int counts[4]) {
   int s;
   const int total = counts[0] + counts[1] + counts[2] + counts[3];
   for (s = 0; s < 4; ++s) {
-    fprintf(stderr, "|      %2d%%", (int)(100. * counts[s] / total + .5));
+    fprintf(stderr, "|     %3d%%", (int)(100. * counts[s] / total + .5));
   }
   fprintf(stderr, "| %7d\n", total);
 }
@@ -185,7 +191,7 @@ static void PrintExtraInfoLossless(const WebPPicture* const pic,
   if (short_output) {
     fprintf(stderr, "%7d %2.2f\n", stats->coded_size, stats->PSNR[3]);
   } else {
-    fprintf(stderr, "File:      %s\n", file_name);
+    WFPRINTF(stderr, "File:      %s\n", (const W_CHAR*)file_name);
     fprintf(stderr, "Dimension: %d x %d\n", pic->width, pic->height);
     fprintf(stderr, "Output:    %d bytes (%.2f bpp)\n", stats->coded_size,
             8.f * stats->coded_size / pic->width / pic->height);
@@ -204,7 +210,7 @@ static void PrintExtraInfoLossy(const WebPPicture* const pic, int short_output,
     const int num_i16 = stats->block_count[1];
     const int num_skip = stats->block_count[2];
     const int total = num_i4 + num_i16;
-    fprintf(stderr, "File:      %s\n", file_name);
+    WFPRINTF(stderr, "File:      %s\n", (const W_CHAR*)file_name);
     fprintf(stderr, "Dimension: %d x %d%s\n",
             pic->width, pic->height,
             stats->alpha_data_size ? " (with alpha)" : "");
@@ -309,7 +315,7 @@ static int DumpPicture(const WebPPicture* const picture, const char* PGM_name) {
   const int alpha_height =
       WebPPictureHasTransparency(picture) ? picture->height : 0;
   const int height = picture->height + uv_height + alpha_height;
-  FILE* const f = fopen(PGM_name, "wb");
+  FILE* const f = WFOPEN(PGM_name, "wb");
   if (f == NULL) return 0;
   fprintf(f, "P5\n%d %d\n255\n", stride, height);
   for (y = 0; y < picture->height; ++y) {
@@ -521,6 +527,7 @@ static void HelpLong(void) {
   printf(" cwebp [-preset <...>] [options] in_file [-o out_file]\n\n");
   printf("If input size (-s) for an image is not specified, it is\n"
          "assumed to be a PNG, JPEG, TIFF or WebP file.\n");
+  printf("Note: Animated PNG and WebP files are not supported.\n");
 #ifdef HAVE_WINCODEC_H
   printf("Windows builds can take as input any of the files handled by WIC.\n");
 #endif
@@ -561,6 +568,8 @@ static void HelpLong(void) {
   printf("                           "
          "the first partition (0=no degradation ... 100=full)\n");
   printf("  -pass <int> ............ analysis pass number (1..10)\n");
+  printf("  -qrange <min> <max> .... specifies the permissible quality range\n"
+         "                           (default: 0 100)\n");
   printf("  -crop <x> <y> <w> <h> .. crop picture with the given rectangle\n");
   printf("  -resize <w> <h> ........ resize picture (after any cropping)\n");
   printf("  -mt .................... use multi-threading if available\n");
@@ -637,10 +646,10 @@ static const char* const kErrorMessages[VP8_ENC_ERROR_LAST] = {
 
 //------------------------------------------------------------------------------
 
-int main(int argc, const char *argv[]) {
+int main(int argc, const char* argv[]) {
   int return_value = -1;
-  const char *in_file = NULL, *out_file = NULL, *dump_file = NULL;
-  FILE *out = NULL;
+  const char* in_file = NULL, *out_file = NULL, *dump_file = NULL;
+  FILE* out = NULL;
   int c;
   int short_output = 0;
   int quiet = 0;
@@ -660,35 +669,38 @@ int main(int argc, const char *argv[]) {
   WebPConfig config;
   WebPAuxStats stats;
   WebPMemoryWriter memory_writer;
+  int use_memory_writer;
   Metadata metadata;
   Stopwatch stop_watch;
 
+  INIT_WARGV(argc, argv);
+
   MetadataInit(&metadata);
   WebPMemoryWriterInit(&memory_writer);
   if (!WebPPictureInit(&picture) ||
       !WebPPictureInit(&original_picture) ||
       !WebPConfigInit(&config)) {
     fprintf(stderr, "Error! Version mismatch!\n");
-    return -1;
+    FREE_WARGV_AND_RETURN(-1);
   }
 
   if (argc == 1) {
     HelpShort();
-    return 0;
+    FREE_WARGV_AND_RETURN(0);
   }
 
   for (c = 1; c < argc; ++c) {
     int parse_error = 0;
     if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       HelpShort();
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-H") || !strcmp(argv[c], "-longhelp")) {
       HelpLong();
-      return 0;
-    } else if (!strcmp(argv[c], "-o") && c < argc - 1) {
-      out_file = argv[++c];
-    } else if (!strcmp(argv[c], "-d") && c < argc - 1) {
-      dump_file = argv[++c];
+      FREE_WARGV_AND_RETURN(0);
+    } else if (!strcmp(argv[c], "-o") && c + 1 < argc) {
+      out_file = (const char*)GET_WARGV(argv, ++c);
+    } else if (!strcmp(argv[c], "-d") && c + 1 < argc) {
+      dump_file = (const char*)GET_WARGV(argv, ++c);
       config.show_compressed = 1;
     } else if (!strcmp(argv[c], "-print_psnr")) {
       config.show_compressed = 1;
@@ -701,7 +713,7 @@ int main(int argc, const char *argv[]) {
       print_distortion = 2;
     } else if (!strcmp(argv[c], "-short")) {
       ++short_output;
-    } else if (!strcmp(argv[c], "-s") && c < argc - 2) {
+    } else if (!strcmp(argv[c], "-s") && c + 2 < argc) {
       picture.width = ExUtilGetInt(argv[++c], 0, &parse_error);
       picture.height = ExUtilGetInt(argv[++c], 0, &parse_error);
       if (picture.width > WEBP_MAX_DIMENSION || picture.width < 0 ||
@@ -711,30 +723,30 @@ int main(int argc, const char *argv[]) {
                 picture.width, picture.height);
         goto Error;
       }
-    } else if (!strcmp(argv[c], "-m") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-m") && c + 1 < argc) {
       config.method = ExUtilGetInt(argv[++c], 0, &parse_error);
       use_lossless_preset = 0;   // disable -z option
-    } else if (!strcmp(argv[c], "-q") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-q") && c + 1 < argc) {
       config.quality = ExUtilGetFloat(argv[++c], &parse_error);
       use_lossless_preset = 0;   // disable -z option
-    } else if (!strcmp(argv[c], "-z") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-z") && c + 1 < argc) {
       lossless_preset = ExUtilGetInt(argv[++c], 0, &parse_error);
       if (use_lossless_preset != 0) use_lossless_preset = 1;
-    } else if (!strcmp(argv[c], "-alpha_q") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-alpha_q") && c + 1 < argc) {
       config.alpha_quality = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-alpha_method") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-alpha_method") && c + 1 < argc) {
       config.alpha_compression = ExUtilGetInt(argv[++c], 0, &parse_error);
     } else if (!strcmp(argv[c], "-alpha_cleanup")) {
       // This flag is obsolete, does opposite of -exact.
       config.exact = 0;
     } else if (!strcmp(argv[c], "-exact")) {
       config.exact = 1;
-    } else if (!strcmp(argv[c], "-blend_alpha") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-blend_alpha") && c + 1 < argc) {
       blend_alpha = 1;
       // background color is given in hex with an optional '0x' prefix
       background_color = ExUtilGetInt(argv[++c], 16, &parse_error);
       background_color = background_color & 0x00ffffffu;
-    } else if (!strcmp(argv[c], "-alpha_filter") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-alpha_filter") && c + 1 < argc) {
       ++c;
       if (!strcmp(argv[c], "none")) {
         config.alpha_filtering = 0;
@@ -750,10 +762,10 @@ int main(int argc, const char *argv[]) {
       keep_alpha = 0;
     } else if (!strcmp(argv[c], "-lossless")) {
       config.lossless = 1;
-    } else if (!strcmp(argv[c], "-near_lossless") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-near_lossless") && c + 1 < argc) {
       config.near_lossless = ExUtilGetInt(argv[++c], 0, &parse_error);
       config.lossless = 1;  // use near-lossless only with lossless
-    } else if (!strcmp(argv[c], "-hint") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-hint") && c + 1 < argc) {
       ++c;
       if (!strcmp(argv[c], "photo")) {
         config.image_hint = WEBP_HINT_PHOTO;
@@ -765,13 +777,13 @@ int main(int argc, const char *argv[]) {
         fprintf(stderr, "Error! Unrecognized image hint: %s\n", argv[c]);
         goto Error;
       }
-    } else if (!strcmp(argv[c], "-size") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-size") && c + 1 < argc) {
       config.target_size = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-psnr") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-psnr") && c + 1 < argc) {
       config.target_PSNR = ExUtilGetFloat(argv[++c], &parse_error);
-    } else if (!strcmp(argv[c], "-sns") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-sns") && c + 1 < argc) {
       config.sns_strength = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-f") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-f") && c + 1 < argc) {
       config.filter_strength = ExUtilGetInt(argv[++c], 0, &parse_error);
     } else if (!strcmp(argv[c], "-af")) {
       config.autofilter = 1;
@@ -785,27 +797,32 @@ int main(int argc, const char *argv[]) {
       config.filter_type = 1;
     } else if (!strcmp(argv[c], "-nostrong")) {
       config.filter_type = 0;
-    } else if (!strcmp(argv[c], "-sharpness") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-sharpness") && c + 1 < argc) {
       config.filter_sharpness = ExUtilGetInt(argv[++c], 0, &parse_error);
     } else if (!strcmp(argv[c], "-sharp_yuv")) {
       config.use_sharp_yuv = 1;
-    } else if (!strcmp(argv[c], "-pass") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-pass") && c + 1 < argc) {
       config.pass = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-pre") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-qrange") && c + 2 < argc) {
+      config.qmin = ExUtilGetInt(argv[++c], 0, &parse_error);
+      config.qmax = ExUtilGetInt(argv[++c], 0, &parse_error);
+      if (config.qmin < 0) config.qmin = 0;
+      if (config.qmax > 100) config.qmax = 100;
+    } else if (!strcmp(argv[c], "-pre") && c + 1 < argc) {
       config.preprocessing = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-segments") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-segments") && c + 1 < argc) {
       config.segments = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-partition_limit") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-partition_limit") && c + 1 < argc) {
       config.partition_limit = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-map") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-map") && c + 1 < argc) {
       picture.extra_info_type = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-crop") && c < argc - 4) {
+    } else if (!strcmp(argv[c], "-crop") && c + 4 < argc) {
       crop = 1;
       crop_x = ExUtilGetInt(argv[++c], 0, &parse_error);
       crop_y = ExUtilGetInt(argv[++c], 0, &parse_error);
       crop_w = ExUtilGetInt(argv[++c], 0, &parse_error);
       crop_h = ExUtilGetInt(argv[++c], 0, &parse_error);
-    } else if (!strcmp(argv[c], "-resize") && c < argc - 2) {
+    } else if (!strcmp(argv[c], "-resize") && c + 2 < argc) {
       resize_w = ExUtilGetInt(argv[++c], 0, &parse_error);
       resize_h = ExUtilGetInt(argv[++c], 0, &parse_error);
 #ifndef WEBP_DLL
@@ -816,12 +833,12 @@ int main(int argc, const char *argv[]) {
       const int version = WebPGetEncoderVersion();
       printf("%d.%d.%d\n",
              (version >> 16) & 0xff, (version >> 8) & 0xff, version & 0xff);
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-progress")) {
       show_progress = 1;
     } else if (!strcmp(argv[c], "-quiet")) {
       quiet = 1;
-    } else if (!strcmp(argv[c], "-preset") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-preset") && c + 1 < argc) {
       WebPPreset preset;
       ++c;
       if (!strcmp(argv[c], "default")) {
@@ -844,7 +861,7 @@ int main(int argc, const char *argv[]) {
         fprintf(stderr, "Error! Could initialize configuration with preset.\n");
         goto Error;
       }
-    } else if (!strcmp(argv[c], "-metadata") && c < argc - 1) {
+    } else if (!strcmp(argv[c], "-metadata") && c + 1 < argc) {
       static const struct {
         const char* option;
         int flag;
@@ -878,8 +895,7 @@ int main(int argc, const char *argv[]) {
         if (i == kNumTokens) {
           fprintf(stderr, "Error! Unknown metadata type '%.*s'\n",
                   (int)(token - start), start);
-          HelpLong();
-          return -1;
+          FREE_WARGV_AND_RETURN(-1);
         }
         start = token + 1;
       }
@@ -893,19 +909,19 @@ int main(int argc, const char *argv[]) {
     } else if (!strcmp(argv[c], "-v")) {
       verbose = 1;
     } else if (!strcmp(argv[c], "--")) {
-      if (c < argc - 1) in_file = argv[++c];
+      if (c + 1 < argc) in_file = (const char*)GET_WARGV(argv, ++c);
       break;
     } else if (argv[c][0] == '-') {
       fprintf(stderr, "Error! Unknown option '%s'\n", argv[c]);
       HelpLong();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     } else {
-      in_file = argv[c];
+      in_file = (const char*)GET_WARGV(argv, c);
     }
 
     if (parse_error) {
       HelpLong();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     }
   }
   if (in_file == NULL) {
@@ -955,7 +971,8 @@ int main(int argc, const char *argv[]) {
   }
   if (!ReadPicture(in_file, &picture, keep_alpha,
                    (keep_metadata == 0) ? NULL : &metadata)) {
-    fprintf(stderr, "Error! Cannot read input picture file '%s'\n", in_file);
+    WFPRINTF(stderr, "Error! Cannot read input picture file '%s'\n",
+             (const W_CHAR*)in_file);
     goto Error;
   }
   picture.progress_hook = (show_progress && !quiet) ? ProgressReport : NULL;
@@ -968,28 +985,41 @@ int main(int argc, const char *argv[]) {
     const double read_time = StopwatchReadAndReset(&stop_watch);
     fprintf(stderr, "Time to read input: %.3fs\n", read_time);
   }
+  // The bitstream should be kept in memory when metadata must be appended
+  // before writing it to a file/stream, and/or when the near-losslessly encoded
+  // bitstream must be decoded for distortion computation (lossy will modify the
+  // 'picture' but not the lossless pipeline).
+  // Otherwise directly write the bitstream to a file.
+  use_memory_writer = (out_file != NULL && keep_metadata) ||
+                      (!quiet && print_distortion >= 0 && config.lossless &&
+                       config.near_lossless < 100);
 
   // Open the output
   if (out_file != NULL) {
-    const int use_stdout = !strcmp(out_file, "-");
-    out = use_stdout ? ImgIoUtilSetBinaryMode(stdout) : fopen(out_file, "wb");
+    const int use_stdout = !WSTRCMP(out_file, "-");
+    out = use_stdout ? ImgIoUtilSetBinaryMode(stdout) : WFOPEN(out_file, "wb");
     if (out == NULL) {
-      fprintf(stderr, "Error! Cannot open output file '%s'\n", out_file);
+      WFPRINTF(stderr, "Error! Cannot open output file '%s'\n",
+               (const W_CHAR*)out_file);
       goto Error;
     } else {
       if (!short_output && !quiet) {
-        fprintf(stderr, "Saving file '%s'\n", out_file);
+        WFPRINTF(stderr, "Saving file '%s'\n", (const W_CHAR*)out_file);
       }
     }
-    if (keep_metadata == 0) {
-      picture.writer = MyWriter;
-      picture.custom_ptr = (void*)out;
-    } else {
+    if (use_memory_writer) {
       picture.writer = WebPMemoryWrite;
       picture.custom_ptr = (void*)&memory_writer;
+    } else {
+      picture.writer = MyWriter;
+      picture.custom_ptr = (void*)out;
     }
   } else {
     out = NULL;
+    if (use_memory_writer) {
+      picture.writer = WebPMemoryWrite;
+      picture.custom_ptr = (void*)&memory_writer;
+    }
     if (!quiet && !short_output) {
       fprintf(stderr, "No output file specified (no -o flag). Encoding will\n");
       fprintf(stderr, "be performed, but its results discarded.\n\n");
@@ -1068,8 +1098,12 @@ int main(int argc, const char *argv[]) {
   if (picture.extra_info_type > 0) {
     AllocExtraInfo(&picture);
   }
-  if (print_distortion >= 0) {  // Save original picture for later comparison
-    WebPPictureCopy(&picture, &original_picture);
+  // Save original picture for later comparison. Only for lossy as lossless does
+  // not modify 'picture' (even near-lossless).
+  if (print_distortion >= 0 && !config.lossless &&
+      !WebPPictureCopy(&picture, &original_picture)) {
+    fprintf(stderr, "Error! Cannot copy temporary picture\n");
+    goto Error;
   }
 
   // Compress.
@@ -1087,41 +1121,71 @@ int main(int argc, const char *argv[]) {
     fprintf(stderr, "Time to encode picture: %.3fs\n", encode_time);
   }
 
-  // Write info
+  // Get the decompressed image for the lossless pipeline.
+  if (!quiet && print_distortion >= 0 && config.lossless) {
+    if (config.near_lossless == 100) {
+      // Pure lossless: image was not modified, make 'original_picture' a view
+      // of 'picture' by copying all members except the freeable pointers.
+      original_picture = picture;
+      original_picture.memory_ = original_picture.memory_argb_ = NULL;
+    } else {
+      // Decode the bitstream stored in 'memory_writer' to get the altered image
+      // to 'picture'; save the 'original_picture' beforehand.
+      assert(use_memory_writer);
+      original_picture = picture;
+      if (!WebPPictureInit(&picture)) {  // Do not free 'picture'.
+        fprintf(stderr, "Error! Version mismatch!\n");
+        goto Error;
+      }
+
+      picture.use_argb = 1;
+      if (!ReadWebP(memory_writer.mem, memory_writer.size, &picture,
+                    /*keep_alpha=*/WebPPictureHasTransparency(&picture),
+                    /*metadata=*/NULL)) {
+        fprintf(stderr, "Error! Cannot decode encoded WebP bitstream\n");
+        fprintf(stderr, "Error code: %d (%s)\n", picture.error_code,
+                kErrorMessages[picture.error_code]);
+        goto Error;
+      }
+      picture.stats = original_picture.stats;
+    }
+    original_picture.stats = NULL;
+  }
+
+  // Write the YUV planes to a PGM file. Only available for lossy.
   if (dump_file) {
     if (picture.use_argb) {
       fprintf(stderr, "Warning: can't dump file (-d option) "
                       "in lossless mode.\n");
     } else if (!DumpPicture(&picture, dump_file)) {
-      fprintf(stderr, "Warning, couldn't dump picture %s\n", dump_file);
+      WFPRINTF(stderr, "Warning, couldn't dump picture %s\n",
+               (const W_CHAR*)dump_file);
     }
   }
 
-  if (keep_metadata != 0) {
-    if (out != NULL) {
-      if (!WriteWebPWithMetadata(out, &picture, &memory_writer,
-                                 &metadata, keep_metadata, &metadata_written)) {
-        fprintf(stderr, "Error writing WebP file with metadata!\n");
-        goto Error;
-      }
-    } else {  // output is disabled, just display the metadata stats.
-      const struct {
-        const MetadataPayload* const payload;
-        int flag;
-      } *iter, info[] = {
-        { &metadata.exif, METADATA_EXIF },
-        { &metadata.iccp, METADATA_ICC },
-        { &metadata.xmp, METADATA_XMP },
-        { NULL, 0 }
-      };
-      uint32_t unused1 = 0;
-      uint64_t unused2 = 0;
+  if (use_memory_writer && out != NULL &&
+      !WriteWebPWithMetadata(out, &picture, &memory_writer, &metadata,
+                             keep_metadata, &metadata_written)) {
+    fprintf(stderr, "Error writing WebP file!\n");
+    goto Error;
+  }
 
-      for (iter = info; iter->payload != NULL; ++iter) {
-        if (UpdateFlagsAndSize(iter->payload, !!(keep_metadata & iter->flag),
-                               0, &unused1, &unused2)) {
-          metadata_written |= iter->flag;
-        }
+  if (out == NULL && keep_metadata) {
+    // output is disabled, just display the metadata stats.
+    const struct {
+      const MetadataPayload* const payload;
+      int flag;
+    } *iter, info[] = {{&metadata.exif, METADATA_EXIF},
+                       {&metadata.iccp, METADATA_ICC},
+                       {&metadata.xmp, METADATA_XMP},
+                       {NULL, 0}};
+    uint32_t unused1 = 0;
+    uint64_t unused2 = 0;
+
+    for (iter = info; iter->payload != NULL; ++iter) {
+      if (UpdateFlagsAndSize(iter->payload, !!(keep_metadata & iter->flag),
+                             /*flag=*/0, &unused1, &unused2)) {
+        metadata_written |= iter->flag;
       }
     }
   }
@@ -1161,7 +1225,7 @@ int main(int argc, const char *argv[]) {
 
  Error:
   WebPMemoryWriterClear(&memory_writer);
-  free(picture.extra_info);
+  WebPFree(picture.extra_info);
   MetadataFree(&metadata);
   WebPPictureFree(&picture);
   WebPPictureFree(&original_picture);
@@ -1169,7 +1233,7 @@ int main(int argc, const char *argv[]) {
     fclose(out);
   }
 
-  return return_value;
+  FREE_WARGV_AND_RETURN(return_value);
 }
 
 //------------------------------------------------------------------------------

examples/dwebp.c

@@ -24,6 +24,7 @@
 #include "../imageio/image_enc.h"
 #include "../imageio/webpdec.h"
 #include "./stopwatch.h"
+#include "./unicode.h"
 
 static int verbose = 0;
 static int quiet = 0;
@@ -42,7 +43,7 @@ extern void* VP8GetCPUInfo;   // opaque forward declaration.
 
 static int SaveOutput(const WebPDecBuffer* const buffer,
                       WebPOutputFileFormat format, const char* const out_file) {
-  const int use_stdout = (out_file != NULL) && !strcmp(out_file, "-");
+  const int use_stdout = (out_file != NULL) && !WSTRCMP(out_file, "-");
   int ok = 1;
   Stopwatch stop_watch;
 
@@ -56,7 +57,7 @@ static int SaveOutput(const WebPDecBuffer* const buffer,
       if (use_stdout) {
         fprintf(stderr, "Saved to stdout\n");
       } else {
-        fprintf(stderr, "Saved file %s\n", out_file);
+        WFPRINTF(stderr, "Saved file %s\n", (const W_CHAR*)out_file);
       }
     }
     if (verbose) {
@@ -67,7 +68,7 @@ static int SaveOutput(const WebPDecBuffer* const buffer,
     if (use_stdout) {
       fprintf(stderr, "Error writing to stdout !!\n");
     } else {
-      fprintf(stderr, "Error writing file %s !!\n", out_file);
+      WFPRINTF(stderr, "Error writing file %s !!\n", (const W_CHAR*)out_file);
     }
   }
   return ok;
@@ -75,7 +76,8 @@ static int SaveOutput(const WebPDecBuffer* const buffer,
 
 static void Help(void) {
   printf("Usage: dwebp in_file [options] [-o out_file]\n\n"
-         "Decodes the WebP image file to PNG format [Default]\n"
+         "Decodes the WebP image file to PNG format [Default].\n"
+         "Note: Animated WebP files are not supported.\n\n"
          "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"
@@ -131,7 +133,7 @@ static uint8_t* AllocateExternalBuffer(WebPDecoderConfig* config,
                      format == RGB_565) ? 2
                   : 4;
     uint32_t stride = bpp * w + 7;   // <- just for exercising
-    external_buffer = (uint8_t*)malloc(stride * h);
+    external_buffer = (uint8_t*)WebPMalloc(stride * h);
     if (external_buffer == NULL) return NULL;
     output_buffer->u.RGBA.stride = stride;
     output_buffer->u.RGBA.size = stride * h;
@@ -144,7 +146,7 @@ static uint8_t* AllocateExternalBuffer(WebPDecoderConfig* config,
     uint32_t total_size = stride * h * (has_alpha ? 2 : 1)
                         + 2 * uv_stride * (h + 1) / 2;
     assert(format >= YUV && format <= YUVA);
-    external_buffer = (uint8_t*)malloc(total_size);
+    external_buffer = (uint8_t*)WebPMalloc(total_size);
     if (external_buffer == NULL) return NULL;
     tmp = external_buffer;
     output_buffer->u.YUVA.y = tmp;
@@ -175,10 +177,10 @@ static uint8_t* AllocateExternalBuffer(WebPDecoderConfig* config,
   return external_buffer;
 }
 
-int main(int argc, const char *argv[]) {
+int main(int argc, const char* argv[]) {
   int ok = 0;
-  const char *in_file = NULL;
-  const char *out_file = NULL;
+  const char* in_file = NULL;
+  const char* out_file = NULL;
 
   WebPDecoderConfig config;
   WebPDecBuffer* const output_buffer = &config.output;
@@ -191,18 +193,20 @@ int main(int argc, const char *argv[]) {
   int incremental = 0;
   int c;
 
+  INIT_WARGV(argc, argv);
+
   if (!WebPInitDecoderConfig(&config)) {
     fprintf(stderr, "Library version mismatch!\n");
-    return -1;
+    FREE_WARGV_AND_RETURN(-1);
   }
 
   for (c = 1; c < argc; ++c) {
     int parse_error = 0;
     if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       Help();
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-o") && c < argc - 1) {
-      out_file = argv[++c];
+      out_file = (const char*)GET_WARGV(argv, ++c);
     } else if (!strcmp(argv[c], "-alpha")) {
       format = ALPHA_PLANE_ONLY;
     } else if (!strcmp(argv[c], "-nofancy")) {
@@ -223,7 +227,7 @@ int main(int argc, const char *argv[]) {
       const int version = WebPGetDecoderVersion();
       printf("%d.%d.%d\n",
              (version >> 16) & 0xff, (version >> 8) & 0xff, version & 0xff);
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-pgm")) {
       format = PGM;
     } else if (!strcmp(argv[c], "-yuv")) {
@@ -284,26 +288,26 @@ int main(int argc, const char *argv[]) {
     } else if (!strcmp(argv[c], "-incremental")) {
       incremental = 1;
     } else if (!strcmp(argv[c], "--")) {
-      if (c < argc - 1) in_file = argv[++c];
+      if (c < argc - 1) in_file = (const char*)GET_WARGV(argv, ++c);
       break;
     } else if (argv[c][0] == '-') {
       fprintf(stderr, "Unknown option '%s'\n", argv[c]);
       Help();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     } else {
-      in_file = argv[c];
+      in_file = (const char*)GET_WARGV(argv, c);
     }
 
     if (parse_error) {
       Help();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     }
   }
 
   if (in_file == NULL) {
     fprintf(stderr, "missing input file!!\n");
     Help();
-    return -1;
+    FREE_WARGV_AND_RETURN(-1);
   }
 
   if (quiet) verbose = 0;
@@ -312,7 +316,7 @@ int main(int argc, const char *argv[]) {
     VP8StatusCode status = VP8_STATUS_OK;
     size_t data_size = 0;
     if (!LoadWebP(in_file, &data, &data_size, bitstream)) {
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     }
 
     switch (format) {
@@ -389,18 +393,18 @@ int main(int argc, const char *argv[]) {
 
   if (out_file != NULL) {
     if (!quiet) {
-      fprintf(stderr, "Decoded %s. Dimensions: %d x %d %s. Format: %s. "
-                      "Now saving...\n",
-              in_file, output_buffer->width, output_buffer->height,
+      WFPRINTF(stderr, "Decoded %s.", (const W_CHAR*)in_file);
+      fprintf(stderr, " Dimensions: %d x %d %s. Format: %s. Now saving...\n",
+              output_buffer->width, output_buffer->height,
               bitstream->has_alpha ? " (with alpha)" : "",
               kFormatType[bitstream->format]);
     }
     ok = SaveOutput(output_buffer, format, out_file);
   } else {
     if (!quiet) {
-      fprintf(stderr, "File %s can be decoded "
-                      "(dimensions: %d x %d %s. Format: %s).\n",
-              in_file, output_buffer->width, output_buffer->height,
+      WFPRINTF(stderr, "File %s can be decoded ", (const W_CHAR*)in_file);
+      fprintf(stderr, "(dimensions: %d x %d %s. Format: %s).\n",
+              output_buffer->width, output_buffer->height,
               bitstream->has_alpha ? " (with alpha)" : "",
               kFormatType[bitstream->format]);
       fprintf(stderr, "Nothing written; "
@@ -409,9 +413,9 @@ int main(int argc, const char *argv[]) {
   }
  Exit:
   WebPFreeDecBuffer(output_buffer);
-  free((void*)external_buffer);
-  free((void*)data);
-  return ok ? 0 : -1;
+  WebPFree((void*)external_buffer);
+  WebPFree((void*)data);
+  FREE_WARGV_AND_RETURN(ok ? 0 : -1);
 }
 
 //------------------------------------------------------------------------------

examples/example_util.c

@@ -75,7 +75,7 @@ static void ResetCommandLineArguments(int argc, const char* argv[],
 void ExUtilDeleteCommandLineArguments(CommandLineArguments* const args) {
   if (args != NULL) {
     if (args->own_argv_) {
-      free((void*)args->argv_);
+      WebPFree((void*)args->argv_);
       WebPDataClear(&args->argv_data_);
     }
     ResetCommandLineArguments(0, NULL, args);
@@ -90,11 +90,19 @@ int ExUtilInitCommandLineArguments(int argc, const char* argv[],
   if (argc == 1 && argv[0][0] != '-') {
     char* cur;
     const char sep[] = " \t\r\n\f\v";
+
+#if defined(_WIN32) && defined(_UNICODE)
+    fprintf(stderr,
+            "Error: Reading arguments from a file is a feature unavailable "
+            "with Unicode binaries.\n");
+    return 0;
+#endif
+
     if (!ExUtilReadFileToWebPData(argv[0], &args->argv_data_)) {
       return 0;
     }
     args->own_argv_ = 1;
-    args->argv_ = (const char**)malloc(MAX_ARGC * sizeof(*args->argv_));
+    args->argv_ = (const char**)WebPMalloc(MAX_ARGC * sizeof(*args->argv_));
     if (args->argv_ == NULL) return 0;
 
     argc = 0;

examples/gif2webp.c

@@ -33,6 +33,8 @@
 #include "../examples/example_util.h"
 #include "../imageio/imageio_util.h"
 #include "./gifdec.h"
+#include "./unicode.h"
+#include "./unicode_gif.h"
 
 #if !defined(STDIN_FILENO)
 #define STDIN_FILENO 0
@@ -94,12 +96,12 @@ static void Help(void) {
 
 //------------------------------------------------------------------------------
 
-int main(int argc, const char *argv[]) {
+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;
+  const W_CHAR* in_file = NULL, *out_file = NULL;
   GifFileType* gif = NULL;
   int frame_duration = 0;
   int frame_timestamp = 0;
@@ -132,11 +134,13 @@ int main(int argc, const char *argv[]) {
   int default_kmin = 1;  // Whether to use default kmin value.
   int default_kmax = 1;
 
+  INIT_WARGV(argc, argv);
+
   if (!WebPConfigInit(&config) || !WebPAnimEncoderOptionsInit(&enc_options) ||
       !WebPPictureInit(&frame) || !WebPPictureInit(&curr_canvas) ||
       !WebPPictureInit(&prev_canvas)) {
     fprintf(stderr, "Error! Version mismatch!\n");
-    return -1;
+    FREE_WARGV_AND_RETURN(-1);
   }
   config.lossless = 1;  // Use lossless compression by default.
 
@@ -146,16 +150,16 @@ int main(int argc, const char *argv[]) {
 
   if (argc == 1) {
     Help();
-    return 0;
+    FREE_WARGV_AND_RETURN(0);
   }
 
   for (c = 1; c < argc; ++c) {
     int parse_error = 0;
     if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       Help();
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-o") && c < argc - 1) {
-      out_file = argv[++c];
+      out_file = GET_WARGV(argv, ++c);
     } else if (!strcmp(argv[c], "-lossy")) {
       config.lossless = 0;
     } else if (!strcmp(argv[c], "-mixed")) {
@@ -212,7 +216,7 @@ int main(int argc, const char *argv[]) {
           fprintf(stderr, "Error! Unknown metadata type '%.*s'\n",
                   (int)(token - start), start);
           Help();
-          return -1;
+          FREE_WARGV_AND_RETURN(-1);
         }
         start = token + 1;
       }
@@ -225,7 +229,7 @@ int main(int argc, const char *argv[]) {
              (enc_version >> 16) & 0xff, (enc_version >> 8) & 0xff,
              enc_version & 0xff, (mux_version >> 16) & 0xff,
              (mux_version >> 8) & 0xff, mux_version & 0xff);
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-quiet")) {
       quiet = 1;
       enc_options.verbose = 0;
@@ -233,19 +237,19 @@ int main(int argc, const char *argv[]) {
       verbose = 1;
       enc_options.verbose = 1;
     } else if (!strcmp(argv[c], "--")) {
-      if (c < argc - 1) in_file = argv[++c];
+      if (c < argc - 1) in_file = GET_WARGV(argv, ++c);
       break;
     } else if (argv[c][0] == '-') {
       fprintf(stderr, "Error! Unknown option '%s'\n", argv[c]);
       Help();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     } else {
-      in_file = argv[c];
+      in_file = GET_WARGV(argv, c);
     }
 
     if (parse_error) {
       Help();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     }
   }
 
@@ -269,13 +273,7 @@ int main(int argc, const char *argv[]) {
   }
 
   // Start the decoder object
-#if LOCAL_GIF_PREREQ(5,0)
-  gif = !strcmp(in_file, "-") ? DGifOpenFileHandle(STDIN_FILENO, &gif_error)
-                              : DGifOpenFileName(in_file, &gif_error);
-#else
-  gif = !strcmp(in_file, "-") ? DGifOpenFileHandle(STDIN_FILENO)
-                              : DGifOpenFileName(in_file);
-#endif
+  gif = DGifOpenFileUnicode(in_file, &gif_error);
   if (gif == NULL) goto End;
 
   // Loop over GIF images
@@ -381,7 +379,7 @@ int main(int argc, const char *argv[]) {
       }
       case EXTENSION_RECORD_TYPE: {
         int extension;
-        GifByteType *data = NULL;
+        GifByteType* data = NULL;
         if (DGifGetExtension(gif, &extension, &data) == GIF_ERROR) {
           goto End;
         }
@@ -544,17 +542,18 @@ int main(int argc, const char *argv[]) {
   }
 
   if (out_file != NULL) {
-    if (!ImgIoUtilWriteFile(out_file, webp_data.bytes, webp_data.size)) {
-      fprintf(stderr, "Error writing output file: %s\n", out_file);
+    if (!ImgIoUtilWriteFile((const char*)out_file, webp_data.bytes,
+                            webp_data.size)) {
+      WFPRINTF(stderr, "Error writing output file: %s\n", out_file);
       goto End;
     }
     if (!quiet) {
-      if (!strcmp(out_file, "-")) {
+      if (!WSTRCMP(out_file, "-")) {
         fprintf(stderr, "Saved %d bytes to STDIO\n",
                 (int)webp_data.size);
       } else {
-        fprintf(stderr, "Saved output file (%d bytes): %s\n",
-                (int)webp_data.size, out_file);
+        WFPRINTF(stderr, "Saved output file (%d bytes): %s\n",
+                 (int)webp_data.size, out_file);
       }
     }
   } else {
@@ -589,12 +588,12 @@ int main(int argc, const char *argv[]) {
 #endif
   }
 
-  return !ok;
+  FREE_WARGV_AND_RETURN(!ok);
 }
 
 #else  // !WEBP_HAVE_GIF
 
-int main(int argc, const char *argv[]) {
+int main(int argc, const char* argv[]) {
   fprintf(stderr, "GIF support not enabled in %s.\n", argv[0]);
   (void)argc;
   return 0;

examples/gifdec.c

@@ -137,7 +137,7 @@ int GIFReadFrame(GifFileType* const gif, int transparent_index,
   }
   dst = sub_image.argb;
 
-  tmp = (uint8_t*)malloc(rect.width * sizeof(*tmp));
+  tmp = (uint8_t*)WebPMalloc(rect.width * sizeof(*tmp));
   if (tmp == NULL) goto End;
 
   if (image_desc->Interlace) {  // Interlaced image.
@@ -168,7 +168,7 @@ int GIFReadFrame(GifFileType* const gif, int transparent_index,
  End:
   if (!ok) picture->error_code = sub_image.error_code;
   WebPPictureFree(&sub_image);
-  free(tmp);
+  WebPFree(tmp);
   return ok;
 }
 

examples/img2webp.c

@@ -27,6 +27,7 @@
 #include "../imageio/image_dec.h"
 #include "../imageio/imageio_util.h"
 #include "./stopwatch.h"
+#include "./unicode.h"
 #include "webp/encode.h"
 #include "webp/mux.h"
 
@@ -83,7 +84,7 @@ static int ReadImage(const char filename[], WebPPicture* const pic) {
   if (!ImgIoUtilReadFile(filename, &data, &data_size)) return 0;
   reader = WebPGuessImageReader(data, data_size);
   ok = reader(data, data_size, pic, 1, NULL);
-  free((void*)data);
+  WebPFree((void*)data);
   return ok;
 }
 
@@ -138,8 +139,13 @@ int main(int argc, const char* argv[]) {
   int c;
   int have_input = 0;
   CommandLineArguments cmd_args;
-  int ok = ExUtilInitCommandLineArguments(argc - 1, argv + 1, &cmd_args);
-  if (!ok) return 1;
+  int ok;
+
+  INIT_WARGV(argc, argv);
+
+  ok = ExUtilInitCommandLineArguments(argc - 1, argv + 1, &cmd_args);
+  if (!ok) FREE_WARGV_AND_RETURN(1);
+
   argc = cmd_args.argc_;
   argv = cmd_args.argv_;
 
@@ -158,7 +164,7 @@ int main(int argc, const char* argv[]) {
       int parse_error = 0;
       if (!strcmp(argv[c], "-o") && c + 1 < argc) {
         argv[c] = NULL;
-        output = argv[++c];
+        output = (const char*)GET_WARGV_SHIFTED(argv, ++c);
       } else if (!strcmp(argv[c], "-kmin") && c + 1 < argc) {
         argv[c] = NULL;
         anim_config.kmin = ExUtilGetInt(argv[++c], 0, &parse_error);
@@ -245,7 +251,7 @@ int main(int argc, const char* argv[]) {
 
     // read next input image
     pic.use_argb = 1;
-    ok = ReadImage(argv[c], &pic);
+    ok = ReadImage((const char*)GET_WARGV_SHIFTED(argv, c), &pic);
     if (!ok) goto End;
 
     if (enc == NULL) {
@@ -277,8 +283,8 @@ int main(int argc, const char* argv[]) {
     if (!ok) goto End;
 
     if (verbose) {
-      fprintf(stderr, "Added frame #%3d at time %4d (file: %s)\n",
-              pic_num, timestamp_ms, argv[c]);
+      WFPRINTF(stderr, "Added frame #%3d at time %4d (file: %s)\n",
+               pic_num, timestamp_ms, GET_WARGV_SHIFTED(argv, c));
     }
     timestamp_ms += duration;
     ++pic_num;
@@ -302,7 +308,7 @@ int main(int argc, const char* argv[]) {
   if (ok) {
     if (output != NULL) {
       ok = ImgIoUtilWriteFile(output, webp_data.bytes, webp_data.size);
-      if (ok) fprintf(stderr, "output file: %s     ", output);
+      if (ok) WFPRINTF(stderr, "output file: %s     ", (const W_CHAR*)output);
     } else {
       fprintf(stderr, "[no output file specified]   ");
     }
@@ -314,5 +320,5 @@ int main(int argc, const char* argv[]) {
   }
   WebPDataClear(&webp_data);
   ExUtilDeleteCommandLineArguments(&cmd_args);
-  return ok ? 0 : 1;
+  FREE_WARGV_AND_RETURN(ok ? 0 : 1);
 }

examples/unicode.h

@@ -0,0 +1,102 @@
+// Copyright 2018 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.
+// -----------------------------------------------------------------------------
+//
+// Unicode support for Windows. The main idea is to maintain an array of Unicode
+// arguments (wargv) and use it only for file paths. The regular argv is used
+// for everything else.
+//
+// Author: Yannis Guyon (yguyon@google.com)
+
+#ifndef WEBP_EXAMPLES_UNICODE_H_
+#define WEBP_EXAMPLES_UNICODE_H_
+
+#if defined(_WIN32) && defined(_UNICODE)
+
+// wchar_t is used instead of TCHAR because we only perform additional work when
+// Unicode is enabled and because the output of CommandLineToArgvW() is wchar_t.
+
+#include <wchar.h>
+#include <windows.h>
+#include <shellapi.h>
+
+// Create a wchar_t array containing Unicode parameters.
+#define INIT_WARGV(ARGC, ARGV)                                                \
+  int wargc;                                                                  \
+  const W_CHAR** const wargv =                                                \
+      (const W_CHAR**)CommandLineToArgvW(GetCommandLineW(), &wargc);          \
+  do {                                                                        \
+    if (wargv == NULL || wargc != (ARGC)) {                                   \
+      fprintf(stderr, "Error: Unable to get Unicode arguments.\n");           \
+      FREE_WARGV_AND_RETURN(-1);                                              \
+    }                                                                         \
+  } while (0)
+
+// Use this to get a Unicode argument (e.g. file path).
+#define GET_WARGV(UNUSED, C) wargv[C]
+// For cases where argv is shifted by one compared to wargv.
+#define GET_WARGV_SHIFTED(UNUSED, C) wargv[(C) + 1]
+#define GET_WARGV_OR_NULL() wargv
+
+// Release resources. LocalFree() is needed after CommandLineToArgvW().
+#define FREE_WARGV() LOCAL_FREE((W_CHAR** const)wargv)
+#define LOCAL_FREE(WARGV)                  \
+  do {                                     \
+    if ((WARGV) != NULL) LocalFree(WARGV); \
+  } while (0)
+
+#define W_CHAR wchar_t  // WCHAR without underscore might already be defined.
+#define TO_W_CHAR(STR) (L##STR)
+
+#define WFOPEN(ARG, OPT) _wfopen((const W_CHAR*)ARG, TO_W_CHAR(OPT))
+
+#define WPRINTF(STR, ...) wprintf(TO_W_CHAR(STR), __VA_ARGS__)
+#define WFPRINTF(STDERR, STR, ...) fwprintf(STDERR, TO_W_CHAR(STR), __VA_ARGS__)
+
+#define WSTRLEN(FILENAME) wcslen((const W_CHAR*)FILENAME)
+#define WSTRCMP(FILENAME, STR) wcscmp((const W_CHAR*)FILENAME, TO_W_CHAR(STR))
+#define WSTRRCHR(FILENAME, STR) wcsrchr((const W_CHAR*)FILENAME, TO_W_CHAR(STR))
+#define WSNPRINTF(A, B, STR, ...) _snwprintf(A, B, TO_W_CHAR(STR), __VA_ARGS__)
+
+#else
+
+// Unicode file paths work as is on Unix platforms, and no extra work is done on
+// Windows either if Unicode is disabled.
+
+#define INIT_WARGV(ARGC, ARGV)
+
+#define GET_WARGV(ARGV, C) (ARGV)[C]
+#define GET_WARGV_SHIFTED(ARGV, C) (ARGV)[C]
+#define GET_WARGV_OR_NULL() NULL
+
+#define FREE_WARGV()
+#define LOCAL_FREE(WARGV)
+
+#define W_CHAR char
+#define TO_W_CHAR(STR) (STR)
+
+#define WFOPEN(ARG, OPT) fopen(ARG, OPT)
+
+#define WPRINTF(STR, ...) printf(STR, __VA_ARGS__)
+#define WFPRINTF(STDERR, STR, ...) fprintf(STDERR, STR, __VA_ARGS__)
+
+#define WSTRLEN(FILENAME) strlen(FILENAME)
+#define WSTRCMP(FILENAME, STR) strcmp(FILENAME, STR)
+#define WSTRRCHR(FILENAME, STR) strrchr(FILENAME, STR)
+#define WSNPRINTF(A, B, STR, ...) snprintf(A, B, STR, __VA_ARGS__)
+
+#endif  // defined(_WIN32) && defined(_UNICODE)
+
+// Don't forget to free wargv before returning (e.g. from main).
+#define FREE_WARGV_AND_RETURN(VALUE) \
+  do {                               \
+    FREE_WARGV();                    \
+    return (VALUE);                  \
+  } while (0)
+
+#endif  // WEBP_EXAMPLES_UNICODE_H_

examples/unicode_gif.h

@@ -0,0 +1,75 @@
+// Copyright 2018 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.
+// -----------------------------------------------------------------------------
+//
+// giflib doesn't have a Unicode DGifOpenFileName(). Let's make one.
+//
+// Author: Yannis Guyon (yguyon@google.com)
+
+#ifndef WEBP_EXAMPLES_UNICODE_GIF_H_
+#define WEBP_EXAMPLES_UNICODE_GIF_H_
+
+#include "./unicode.h"
+#ifdef HAVE_CONFIG_H
+#include "webp/config.h"  // For WEBP_HAVE_GIF
+#endif
+
+#if defined(WEBP_HAVE_GIF)
+
+#ifdef _WIN32
+#include <fcntl.h>  // Not standard, needed for _topen and flags.
+#include <io.h>
+#endif
+
+#include <gif_lib.h>
+#include <string.h>
+#include "./gifdec.h"
+
+#if !defined(STDIN_FILENO)
+#define STDIN_FILENO 0
+#endif
+
+static GifFileType* DGifOpenFileUnicode(const W_CHAR* file_name, int* error) {
+  if (!WSTRCMP(file_name, "-")) {
+#if LOCAL_GIF_PREREQ(5, 0)
+    return DGifOpenFileHandle(STDIN_FILENO, error);
+#else
+    (void)error;
+    return DGifOpenFileHandle(STDIN_FILENO);
+#endif
+  }
+
+#if defined(_WIN32) && defined(_UNICODE)
+
+  int file_handle = _wopen(file_name, _O_RDONLY | _O_BINARY);
+  if (file_handle == -1) {
+    if (error != NULL) *error = D_GIF_ERR_OPEN_FAILED;
+    return NULL;
+  }
+
+#if LOCAL_GIF_PREREQ(5, 0)
+  return DGifOpenFileHandle(file_handle, error);
+#else
+  return DGifOpenFileHandle(file_handle);
+#endif
+
+#else
+
+#if LOCAL_GIF_PREREQ(5, 0)
+  return DGifOpenFileName(file_name, error);
+#else
+  return DGifOpenFileName(file_name);
+#endif
+
+#endif  // defined(_WIN32) && defined(_UNICODE)
+  // DGifCloseFile() is called later.
+}
+
+#endif  // defined(WEBP_HAVE_GIF)
+
+#endif  // WEBP_EXAMPLES_UNICODE_GIF_H_

examples/vwebp.c

@@ -42,6 +42,7 @@
 
 #include "../examples/example_util.h"
 #include "../imageio/imageio_util.h"
+#include "./unicode.h"
 
 #if defined(_MSC_VER) && _MSC_VER < 1900
 #define snprintf _snprintf
@@ -417,8 +418,9 @@ static void HandleDisplay(void) {
 }
 
 static void StartDisplay(void) {
-  const int width = kParams.canvas_width;
-  const int height = kParams.canvas_height;
+  int width = kParams.canvas_width;
+  int height = kParams.canvas_height;
+  int screen_width, screen_height;
   // TODO(webp:365) GLUT_DOUBLE results in flickering / old frames to be
   // partially displayed with animated webp + alpha.
 #if defined(__APPLE__) || defined(_WIN32)
@@ -426,6 +428,18 @@ static void StartDisplay(void) {
 #else
   glutInitDisplayMode(GLUT_DOUBLE | GLUT_RGBA);
 #endif
+  screen_width = glutGet(GLUT_SCREEN_WIDTH);
+  screen_height = glutGet(GLUT_SCREEN_HEIGHT);
+  if (width > screen_width || height > screen_height) {
+    if (width > screen_width) {
+      height = (height * screen_width + width - 1) / width;
+      width = screen_width;
+    }
+    if (height > screen_height) {
+      width = (width * screen_height + height - 1) / height;
+      height = screen_height;
+    }
+  }
   glutInitWindowSize(width, height);
   glutCreateWindow("WebP viewer");
   glutDisplayFunc(HandleDisplay);
@@ -465,14 +479,16 @@ static void Help(void) {
       "  'q' / 'Q' / ESC .... quit\n");
 }
 
-int main(int argc, char *argv[]) {
+int main(int argc, char* argv[]) {
   int c;
   WebPDecoderConfig* const config = &kParams.config;
   WebPIterator* const curr = &kParams.curr_frame;
 
+  INIT_WARGV(argc, argv);
+
   if (!WebPInitDecoderConfig(config)) {
     fprintf(stderr, "Library version mismatch!\n");
-    return -1;
+    FREE_WARGV_AND_RETURN(-1);
   }
   config->options.dithering_strength = 50;
   config->options.alpha_dithering_strength = 100;
@@ -484,7 +500,7 @@ int main(int argc, char *argv[]) {
     int parse_error = 0;
     if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       Help();
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-noicc")) {
       kParams.use_color_profile = 0;
     } else if (!strcmp(argv[c], "-nofancy")) {
@@ -507,30 +523,30 @@ int main(int argc, char *argv[]) {
              (dec_version >> 16) & 0xff, (dec_version >> 8) & 0xff,
              dec_version & 0xff, (dmux_version >> 16) & 0xff,
              (dmux_version >> 8) & 0xff, dmux_version & 0xff);
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else if (!strcmp(argv[c], "-mt")) {
       config->options.use_threads = 1;
     } else if (!strcmp(argv[c], "--")) {
-      if (c < argc - 1) kParams.file_name = argv[++c];
+      if (c < argc - 1) kParams.file_name = (const char*)GET_WARGV(argv, ++c);
       break;
     } else if (argv[c][0] == '-') {
       printf("Unknown option '%s'\n", argv[c]);
       Help();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     } else {
-      kParams.file_name = argv[c];
+      kParams.file_name = (const char*)GET_WARGV(argv, c);
     }
 
     if (parse_error) {
       Help();
-      return -1;
+      FREE_WARGV_AND_RETURN(-1);
     }
   }
 
   if (kParams.file_name == NULL) {
     printf("missing input file!!\n");
     Help();
-    return 0;
+    FREE_WARGV_AND_RETURN(0);
   }
 
   if (!ImgIoUtilReadFile(kParams.file_name,
@@ -605,16 +621,16 @@ int main(int argc, char *argv[]) {
 
   // Should only be reached when using FREEGLUT:
   ClearParams();
-  return 0;
+  FREE_WARGV_AND_RETURN(0);
 
  Error:
   ClearParams();
-  return -1;
+  FREE_WARGV_AND_RETURN(-1);
 }
 
 #else   // !WEBP_HAVE_GL
 
-int main(int argc, const char *argv[]) {
+int main(int argc, const char* argv[]) {
   fprintf(stderr, "OpenGL support not enabled in %s.\n", argv[0]);
   (void)argc;
   return 0;

examples/webpinfo.c

@@ -20,6 +20,7 @@
 #endif
 
 #include "../imageio/imageio_util.h"
+#include "./unicode.h"
 #include "webp/decode.h"
 #include "webp/format_constants.h"
 #include "webp/mux_types.h"
@@ -1119,19 +1120,21 @@ int main(int argc, const char* argv[]) {
   WebPInfoStatus webp_info_status = WEBP_INFO_OK;
   WebPInfo webp_info;
 
+  INIT_WARGV(argc, argv);
+
   if (argc == 1) {
     HelpShort();
-    return WEBP_INFO_OK;
+    FREE_WARGV_AND_RETURN(WEBP_INFO_OK);
   }
 
   // Parse command-line input.
   for (c = 1; c < argc; ++c) {
     if (!strcmp(argv[c], "-h") || !strcmp(argv[c], "-help")) {
       HelpShort();
-      return WEBP_INFO_OK;
+      FREE_WARGV_AND_RETURN(WEBP_INFO_OK);
     } else if (!strcmp(argv[c], "-H") || !strcmp(argv[c], "-longhelp")) {
       HelpLong();
-      return WEBP_INFO_OK;
+      FREE_WARGV_AND_RETURN(WEBP_INFO_OK);
     } else if (!strcmp(argv[c], "-quiet")) {
       quiet = 1;
     } else if (!strcmp(argv[c], "-diag")) {
@@ -1144,7 +1147,7 @@ int main(int argc, const char* argv[]) {
       const int version = WebPGetDecoderVersion();
       printf("WebP Decoder version: %d.%d.%d\n",
              (version >> 16) & 0xff, (version >> 8) & 0xff, version & 0xff);
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else {  // Assume the remaining are all input files.
       break;
     }
@@ -1152,27 +1155,28 @@ int main(int argc, const char* argv[]) {
 
   if (c == argc) {
     HelpShort();
-    return WEBP_INFO_INVALID_COMMAND;
+    FREE_WARGV_AND_RETURN(WEBP_INFO_INVALID_COMMAND);
   }
 
   // Process input files one by one.
   for (; c < argc; ++c) {
     WebPData webp_data;
-    const char* in_file = NULL;
+    const W_CHAR* in_file = NULL;
     WebPInfoInit(&webp_info);
     webp_info.quiet_ = quiet;
     webp_info.show_diagnosis_ = show_diag;
     webp_info.show_summary_ = show_summary;
     webp_info.parse_bitstream_ = parse_bitstream;
-    in_file = argv[c];
-    if (in_file == NULL || !ReadFileToWebPData(in_file, &webp_data)) {
+    in_file = GET_WARGV(argv, c);
+    if (in_file == NULL ||
+        !ReadFileToWebPData((const char*)in_file, &webp_data)) {
       webp_info_status = WEBP_INFO_INVALID_COMMAND;
-      fprintf(stderr, "Failed to open input file %s.\n", in_file);
+      WFPRINTF(stderr, "Failed to open input file %s.\n", in_file);
       continue;
     }
-    if (!webp_info.quiet_) printf("File: %s\n", in_file);
+    if (!webp_info.quiet_) WPRINTF("File: %s\n", in_file);
     webp_info_status = AnalyzeWebP(&webp_info, &webp_data);
     WebPDataClear(&webp_data);
   }
-  return webp_info_status;
+  FREE_WARGV_AND_RETURN(webp_info_status);
 }

examples/webpmux.c

@@ -26,6 +26,7 @@
     webpmux -set icc image_profile.icc in.webp -o out_icc_container.webp
     webpmux -set exif image_metadata.exif in.webp -o out_exif_container.webp
     webpmux -set xmp image_metadata.xmp in.webp -o out_xmp_container.webp
+    webpmux -set loop 1 in.webp -o out_looped.webp
 
   Extract relevant data from WebP container file:
     webpmux -get frame n in.webp -o out_frame.webp
@@ -62,6 +63,7 @@
 #include "webp/mux.h"
 #include "../examples/example_util.h"
 #include "../imageio/imageio_util.h"
+#include "./unicode.h"
 
 //------------------------------------------------------------------------------
 // Config object to parse command-line arguments.
@@ -96,6 +98,7 @@ typedef enum {
   FEATURE_ICCP,
   FEATURE_ANMF,
   FEATURE_DURATION,
+  FEATURE_LOOP,
   LAST_FEATURE
 } FeatureType;
 
@@ -313,6 +316,7 @@ static void PrintHelp(void) {
   printf("\n");
   printf("SET_OPTIONS:\n");
   printf(" Set color profile/metadata:\n");
+  printf("   loop LOOP_COUNT   set the loop count\n");
   printf("   icc  file.icc     set ICC profile\n");
   printf("   exif file.exif    set EXIF metadata\n");
   printf("   xmp  file.xmp     set XMP metadata\n");
@@ -390,23 +394,25 @@ static int CreateMux(const char* const filename, WebPMux** mux) {
   *mux = WebPMuxCreate(&bitstream, 1);
   WebPDataClear(&bitstream);
   if (*mux != NULL) return 1;
-  fprintf(stderr, "Failed to create mux object from file %s.\n", filename);
+  WFPRINTF(stderr, "Failed to create mux object from file %s.\n",
+           (const W_CHAR*)filename);
   return 0;
 }
 
 static int WriteData(const char* filename, const WebPData* const webpdata) {
   int ok = 0;
-  FILE* fout = strcmp(filename, "-") ? fopen(filename, "wb")
-                                     : ImgIoUtilSetBinaryMode(stdout);
+  FILE* fout = WSTRCMP(filename, "-") ? WFOPEN(filename, "wb")
+                                      : ImgIoUtilSetBinaryMode(stdout);
   if (fout == NULL) {
-    fprintf(stderr, "Error opening output WebP file %s!\n", filename);
+    WFPRINTF(stderr, "Error opening output WebP file %s!\n",
+             (const W_CHAR*)filename);
     return 0;
   }
   if (fwrite(webpdata->bytes, webpdata->size, 1, fout) != 1) {
-    fprintf(stderr, "Error writing file %s!\n", filename);
+    WFPRINTF(stderr, "Error writing file %s!\n", (const W_CHAR*)filename);
   } else {
-    fprintf(stderr, "Saved file %s (%d bytes)\n",
-            filename, (int)webpdata->size);
+    WFPRINTF(stderr, "Saved file %s (%d bytes)\n",
+             (const W_CHAR*)filename, (int)webpdata->size);
     ok = 1;
   }
   if (fout != stdout) fclose(fout);
@@ -612,13 +618,16 @@ static int ValidateCommandLine(const CommandLineArguments* const cmd_args,
   CHECK_NUM_ARGS_AT_MOST(NUM, LABEL);
 
 // Parses command-line arguments to fill up config object. Also performs some
-// semantic checks.
-static int ParseCommandLine(Config* config) {
+// semantic checks. unicode_argv contains wchar_t arguments or is null.
+static int ParseCommandLine(Config* config, const W_CHAR** const unicode_argv) {
   int i = 0;
   int feature_arg_index = 0;
   int ok = 1;
   int argc = config->cmd_args_.argc_;
   const char* const* argv = config->cmd_args_.argv_;
+  // Unicode file paths will be used if available.
+  const char* const* wargv =
+      (unicode_argv != NULL) ? (const char**)(unicode_argv + 1) : argv;
 
   while (i < argc) {
     FeatureArg* const arg = &config->args_[feature_arg_index];
@@ -696,7 +705,7 @@ static int ParseCommandLine(Config* config) {
         i += 2;
       } else if (!strcmp(argv[i], "-o")) {
         CHECK_NUM_ARGS_AT_LEAST(2, ErrParse);
-        config->output_ = argv[i + 1];
+        config->output_ = wargv[i + 1];
         i += 2;
       } else if (!strcmp(argv[i], "-info")) {
         CHECK_NUM_ARGS_EXACTLY(2, ErrParse);
@@ -705,24 +714,26 @@ static int ParseCommandLine(Config* config) {
         } else {
           config->action_type_ = ACTION_INFO;
           config->arg_count_ = 0;
-          config->input_ = argv[i + 1];
+          config->input_ = wargv[i + 1];
         }
         i += 2;
       } else if (!strcmp(argv[i], "-h") || !strcmp(argv[i], "-help")) {
         PrintHelp();
         DeleteConfig(config);
+        LOCAL_FREE((W_CHAR** const)unicode_argv);
         exit(0);
       } else if (!strcmp(argv[i], "-version")) {
         const int version = WebPGetMuxVersion();
         printf("%d.%d.%d\n",
                (version >> 16) & 0xff, (version >> 8) & 0xff, version & 0xff);
         DeleteConfig(config);
+        LOCAL_FREE((W_CHAR** const)unicode_argv);
         exit(0);
       } else if (!strcmp(argv[i], "--")) {
         if (i < argc - 1) {
           ++i;
           if (config->input_ == NULL) {
-            config->input_ = argv[i];
+            config->input_ = wargv[i];
           } else {
             ERROR_GOTO2("ERROR at '%s': Multiple input files specified.\n",
                         argv[i], ErrParse);
@@ -747,7 +758,7 @@ static int ParseCommandLine(Config* config) {
         }
         if (config->action_type_ == ACTION_SET) {
           CHECK_NUM_ARGS_AT_LEAST(2, ErrParse);
-          arg->filename_ = argv[i + 1];
+          arg->filename_ = wargv[i + 1];
           ++feature_arg_index;
           i += 2;
         } else {
@@ -760,9 +771,16 @@ static int ParseCommandLine(Config* config) {
         arg->params_ = argv[i + 1];
         ++feature_arg_index;
         i += 2;
+      } else if (!strcmp(argv[i], "loop") &&
+                 (config->action_type_ == ACTION_SET)) {
+        CHECK_NUM_ARGS_AT_LEAST(2, ErrParse);
+        config->type_ = FEATURE_LOOP;
+        arg->params_ = argv[i + 1];
+        ++feature_arg_index;
+        i += 2;
       } else {  // Assume input file.
         if (config->input_ == NULL) {
-          config->input_ = argv[i];
+          config->input_ = wargv[i];
         } else {
           ERROR_GOTO2("ERROR at '%s': Multiple input files specified.\n",
                       argv[i], ErrParse);
@@ -808,8 +826,8 @@ static int ValidateConfig(Config* const config) {
 }
 
 // Create config object from command-line arguments.
-static int InitializeConfig(int argc, const char* argv[],
-                            Config* const config) {
+static int InitializeConfig(int argc, const char* argv[], Config* const config,
+                            const W_CHAR** const unicode_argv) {
   int num_feature_args = 0;
   int ok;
 
@@ -830,7 +848,7 @@ static int InitializeConfig(int argc, const char* argv[],
   }
 
   // Parse command-line.
-  if (!ParseCommandLine(config) || !ValidateConfig(config)) {
+  if (!ParseCommandLine(config, unicode_argv) || !ValidateConfig(config)) {
     ERROR_GOTO1("Exiting due to command-line parsing error.\n", Err1);
   }
 
@@ -1003,13 +1021,38 @@ static int Process(const Config* config) {
           ok = ExUtilReadFileToWebPData(config->args_[0].filename_, &chunk);
           if (!ok) goto Err2;
           err = WebPMuxSetChunk(mux, kFourccList[config->type_], &chunk, 1);
-          free((void*)chunk.bytes);
+          WebPDataClear(&chunk);
           if (err != WEBP_MUX_OK) {
             ERROR_GOTO3("ERROR (%s): Could not set the %s.\n",
                         ErrorString(err), kDescriptions[config->type_], Err2);
           }
           break;
         }
+        case FEATURE_LOOP: {
+          WebPMuxAnimParams params = { 0xFFFFFFFF, 0 };
+          int parse_error = 0;
+          const int loop_count =
+              ExUtilGetInt(config->args_[0].params_, 10, &parse_error);
+          if (loop_count < 0 || loop_count > 65535 || parse_error) {
+            ERROR_GOTO1("ERROR: Loop count must be in the range 0 to 65535.\n",
+                        Err2);
+          }
+          ok = CreateMux(config->input_, &mux);
+          if (!ok) goto Err2;
+          ok = (WebPMuxGetAnimationParams(mux, &params) == WEBP_MUX_OK);
+          if (!ok) {
+            ERROR_GOTO1("ERROR: input file does not seem to be an animation.\n",
+                        Err2);
+          }
+          params.loop_count = loop_count;
+          err = WebPMuxSetAnimationParams(mux, &params);
+          ok = (err == WEBP_MUX_OK);
+          if (!ok) {
+            ERROR_GOTO2("ERROR (%s): Could not set animation parameters.\n",
+                        ErrorString(err), Err2);
+          }
+          break;
+        }
         default: {
           ERROR_GOTO1("ERROR: Invalid feature for action 'set'.\n", Err2);
           break;
@@ -1037,7 +1080,7 @@ static int Process(const Config* config) {
         int* durations = NULL;
         WebPMux* new_mux = DuplicateMuxHeader(mux);
         if (new_mux == NULL) goto Err2;
-        durations = (int*)malloc((size_t)num_frames * sizeof(*durations));
+        durations = (int*)WebPMalloc((size_t)num_frames * sizeof(*durations));
         if (durations == NULL) goto Err2;
         for (i = 0; i < num_frames; ++i) durations[i] = -1;
 
@@ -1095,7 +1138,7 @@ static int Process(const Config* config) {
         new_mux = NULL;
 
  Err3:
-        free(durations);
+        WebPFree(durations);
         WebPMuxDelete(new_mux);
         if (!ok) goto Err2;
       }
@@ -1140,14 +1183,18 @@ static int Process(const Config* config) {
 
 int main(int argc, const char* argv[]) {
   Config config;
-  int ok = InitializeConfig(argc - 1, argv + 1, &config);
+  int ok;
+
+  INIT_WARGV(argc, argv);
+
+  ok = InitializeConfig(argc - 1, argv + 1, &config, GET_WARGV_OR_NULL());
   if (ok) {
     ok = Process(&config);
   } else {
     PrintHelp();
   }
   DeleteConfig(&config);
-  return !ok;
+  FREE_WARGV_AND_RETURN(!ok);
 }
 
 //------------------------------------------------------------------------------

extras/extras.c

@@ -12,13 +12,14 @@
 
 #include "extras/extras.h"
 #include "webp/format_constants.h"
+#include "src/dsp/dsp.h"
 
 #include <assert.h>
 #include <string.h>
 
 #define XTRA_MAJ_VERSION 1
-#define XTRA_MIN_VERSION 0
-#define XTRA_REV_VERSION 1
+#define XTRA_MIN_VERSION 2
+#define XTRA_REV_VERSION 0
 
 //------------------------------------------------------------------------------
 
@@ -57,7 +58,7 @@ int WebPImportRGB565(const uint8_t* rgb565, WebPPicture* pic) {
   for (y = 0; y < pic->height; ++y) {
     const int width = pic->width;
     for (x = 0; x < width; ++x) {
-#ifdef WEBP_SWAP_16BIT_CSP
+#if defined(WEBP_SWAP_16BIT_CSP) && (WEBP_SWAP_16BIT_CSP == 1)
       const uint32_t rg = rgb565[2 * x + 1];
       const uint32_t gb = rgb565[2 * x + 0];
 #else
@@ -90,7 +91,7 @@ int WebPImportRGB4444(const uint8_t* rgb4444, WebPPicture* pic) {
   for (y = 0; y < pic->height; ++y) {
     const int width = pic->width;
     for (x = 0; x < width; ++x) {
-#ifdef WEBP_SWAP_16BIT_CSP
+#if defined(WEBP_SWAP_16BIT_CSP) && (WEBP_SWAP_16BIT_CSP == 1)
       const uint32_t rg = rgb4444[2 * x + 1];
       const uint32_t ba = rgb4444[2 * x + 0];
 #else
@@ -144,3 +145,18 @@ int WebPImportColorMappedARGB(const uint8_t* indexed, int indexed_stride,
 }
 
 //------------------------------------------------------------------------------
+
+int WebPUnmultiplyARGB(WebPPicture* pic) {
+  int y;
+  uint32_t* dst;
+  if (pic == NULL || pic->use_argb != 1 || pic->argb == NULL) return 0;
+  WebPInitAlphaProcessing();
+  dst = pic->argb;
+  for (y = 0; y < pic->height; ++y) {
+    WebPMultARGBRow(dst, pic->width, /*inverse=*/1);
+    dst += pic->argb_stride;
+  }
+  return 1;
+}
+
+//------------------------------------------------------------------------------

extras/extras.h

@@ -19,7 +19,7 @@ extern "C" {
 
 #include "webp/encode.h"
 
-#define WEBP_EXTRAS_ABI_VERSION 0x0001    // MAJOR(8b) + MINOR(8b)
+#define WEBP_EXTRAS_ABI_VERSION 0x0002    // MAJOR(8b) + MINOR(8b)
 
 //------------------------------------------------------------------------------
 
@@ -51,6 +51,13 @@ WebPImportColorMappedARGB(const uint8_t* indexed, int indexed_stride,
                           const uint32_t palette[], int palette_size,
                           WebPPicture* pic);
 
+// Convert the ARGB content of 'pic' from associated to unassociated.
+// 'pic' can be for instance the result of calling of some WebPPictureImportXXX
+// functions, with pic->use_argb set to 'true'. It is assumed (and not checked)
+// that the pre-multiplied r/g/b values as less or equal than the alpha value.
+// Return false in case of error (invalid parameter, ...).
+WEBP_EXTERN int WebPUnmultiplyARGB(WebPPicture* pic);
+
 //------------------------------------------------------------------------------
 
 // Parse a bitstream, search for VP8 (lossy) header and report a

extras/get_disto.c

@@ -26,6 +26,7 @@
 #include "webp/encode.h"
 #include "imageio/image_dec.h"
 #include "imageio/imageio_util.h"
+#include "../examples/unicode.h"
 
 static size_t ReadPicture(const char* const filename, WebPPicture* const pic,
                           int keep_alpha) {
@@ -48,7 +49,8 @@ static size_t ReadPicture(const char* const filename, WebPPicture* const pic,
 
  End:
   if (!ok) {
-    fprintf(stderr, "Error! Could not process file %s\n", filename);
+    WFPRINTF(stderr, "Error! Could not process file %s\n",
+             (const W_CHAR*)filename);
   }
   free((void*)data);
   return ok ? data_size : 0;
@@ -224,7 +226,7 @@ static void Help(void) {
           " Also handles PNG, JPG and TIFF files, in addition to WebP.\n");
 }
 
-int main(int argc, const char *argv[]) {
+int main(int argc, const char* argv[]) {
   WebPPicture pic1, pic2;
   size_t size1 = 0, size2 = 0;
   int ret = 1;
@@ -239,9 +241,11 @@ int main(int argc, const char *argv[]) {
   const char* name2 = NULL;
   const char* output = NULL;
 
+  INIT_WARGV(argc, argv);
+
   if (!WebPPictureInit(&pic1) || !WebPPictureInit(&pic2)) {
     fprintf(stderr, "Can't init pictures\n");
-    return 1;
+    FREE_WARGV_AND_RETURN(1);
   }
 
   for (c = 1; c < argc; ++c) {
@@ -263,11 +267,11 @@ int main(int argc, const char *argv[]) {
         fprintf(stderr, "missing file name after %s option.\n", argv[c - 1]);
         goto End;
       }
-      output = argv[c];
+      output = (const char*)GET_WARGV(argv, c);
     } else if (name1 == NULL) {
-      name1 = argv[c];
+      name1 = (const char*)GET_WARGV(argv, c);
     } else {
-      name2 = argv[c];
+      name2 = (const char*)GET_WARGV(argv, c);
     }
   }
   if (help || name1 == NULL || name2 == NULL) {
@@ -347,5 +351,5 @@ int main(int argc, const char *argv[]) {
  End:
   WebPPictureFree(&pic1);
   WebPPictureFree(&pic2);
-  return ret;
+  FREE_WARGV_AND_RETURN(ret);
 }

extras/vwebp_sdl.c

@@ -25,6 +25,7 @@
 #include "webp_to_sdl.h"
 #include "webp/decode.h"
 #include "imageio/imageio_util.h"
+#include "../examples/unicode.h"
 
 #if defined(WEBP_HAVE_JUST_SDL_H)
 #include <SDL.h>
@@ -51,29 +52,33 @@ static void ProcessEvents(void) {
 int main(int argc, char* argv[]) {
   int c;
   int ok = 0;
+
+  INIT_WARGV(argc, argv);
+
   for (c = 1; c < argc; ++c) {
     const char* file = NULL;
     const uint8_t* webp = NULL;
     size_t webp_size = 0;
     if (!strcmp(argv[c], "-h")) {
       printf("Usage: %s [-h] image.webp [more_files.webp...]\n", argv[0]);
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else {
-      file = argv[c];
+      file = (const char*)GET_WARGV(argv, c);
     }
     if (file == NULL) continue;
     if (!ImgIoUtilReadFile(file, &webp, &webp_size)) {
-      fprintf(stderr, "Error opening file: %s\n", file);
+      WFPRINTF(stderr, "Error opening file: %s\n", (const W_CHAR*)file);
       goto Error;
     }
     if (webp_size != (size_t)(int)webp_size) {
+      free((void*)webp);
       fprintf(stderr, "File too large.\n");
       goto Error;
     }
     ok = WebpToSDL((const char*)webp, (int)webp_size);
     free((void*)webp);
     if (!ok) {
-      fprintf(stderr, "Error decoding file %s\n", file);
+      WFPRINTF(stderr, "Error decoding file %s\n", (const W_CHAR*)file);
       goto Error;
     }
     ProcessEvents();
@@ -82,12 +87,12 @@ int main(int argc, char* argv[]) {
 
  Error:
   SDL_Quit();
-  return ok ? 0 : 1;
+  FREE_WARGV_AND_RETURN(ok ? 0 : 1);
 }
 
 #else  // !WEBP_HAVE_SDL
 
-int main(int argc, const char *argv[]) {
+int main(int argc, const char* argv[]) {
   fprintf(stderr, "SDL support not enabled in %s.\n", argv[0]);
   (void)argc;
   return 0;

extras/webp_quality.c

@@ -13,26 +13,30 @@
 
 #include "extras/extras.h"
 #include "imageio/imageio_util.h"
+#include "../examples/unicode.h"
 
-int main(int argc, const char *argv[]) {
+int main(int argc, const char* argv[]) {
   int c;
   int quiet = 0;
   int ok = 1;
+
+  INIT_WARGV(argc, argv);
+
   for (c = 1; ok && c < argc; ++c) {
     if (!strcmp(argv[c], "-quiet")) {
       quiet = 1;
     } else if (!strcmp(argv[c], "-help") || !strcmp(argv[c], "-h")) {
       printf("webp_quality [-h][-quiet] webp_files...\n");
-      return 0;
+      FREE_WARGV_AND_RETURN(0);
     } else {
-      const char* const filename = argv[c];
+      const char* const filename = (const char*)GET_WARGV(argv, c);
       const uint8_t* data = NULL;
       size_t data_size = 0;
       int q;
       ok = ImgIoUtilReadFile(filename, &data, &data_size);
       if (!ok) break;
       q = VP8EstimateQuality(data, data_size);
-      if (!quiet) printf("[%s] ", filename);
+      if (!quiet) WPRINTF("[%s] ", (const W_CHAR*)filename);
       if (q < 0) {
         fprintf(stderr, "Not a WebP file, or not a lossy WebP file.\n");
         ok = 0;
@@ -46,5 +50,5 @@ int main(int argc, const char *argv[]) {
       free((void*)data);
     }
   }
-  return ok ? 0 : 1;
+  FREE_WARGV_AND_RETURN(ok ? 0 : 1);
 }

gradle/wrapper/gradle-wrapper.properties

@@ -1,6 +1,5 @@
-#Thu May 12 17:06:25 CEST 2016
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
+distributionUrl=https\://services.gradle.org/distributions/gradle-6.1.1-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-2.13-bin.zip

gradlew

@@ -1,4 +1,20 @@
-#!/usr/bin/env bash
+#!/usr/bin/env sh
+
+#
+# Copyright 2015 the original author or authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
 
 ##############################################################################
 ##
@@ -28,16 +44,16 @@ APP_NAME="Gradle"
 APP_BASE_NAME=`basename "$0"`
 
 # Add default JVM options here. You can also use JAVA_OPTS and GRADLE_OPTS to pass JVM options to this script.
-DEFAULT_JVM_OPTS=""
+DEFAULT_JVM_OPTS='"-Xmx64m" "-Xms64m"'
 
 # Use the maximum available, or set MAX_FD != -1 to use that value.
 MAX_FD="maximum"
 
-warn ( ) {
+warn () {
     echo "$*"
 }
 
-die ( ) {
+die () {
     echo
     echo "$*"
     echo
@@ -109,8 +125,8 @@ if $darwin; then
     GRADLE_OPTS="$GRADLE_OPTS \"-Xdock:name=$APP_NAME\" \"-Xdock:icon=$APP_HOME/media/gradle.icns\""
 fi
 
-# For Cygwin, switch paths to Windows format before running java
-if $cygwin ; then
+# For Cygwin or MSYS, switch paths to Windows format before running java
+if [ "$cygwin" = "true" -o "$msys" = "true" ] ; then
     APP_HOME=`cygpath --path --mixed "$APP_HOME"`
     CLASSPATH=`cygpath --path --mixed "$CLASSPATH"`
     JAVACMD=`cygpath --unix "$JAVACMD"`
@@ -138,27 +154,30 @@ if $cygwin ; then
         else
             eval `echo args$i`="\"$arg\""
         fi
-        i=$((i+1))
+        i=`expr $i + 1`
     done
     case $i in
-        (0) set -- ;;
-        (1) set -- "$args0" ;;
-        (2) set -- "$args0" "$args1" ;;
-        (3) set -- "$args0" "$args1" "$args2" ;;
-        (4) set -- "$args0" "$args1" "$args2" "$args3" ;;
-        (5) set -- "$args0" "$args1" "$args2" "$args3" "$args4" ;;
-        (6) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" ;;
-        (7) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" ;;
-        (8) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" "$args7" ;;
-        (9) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" "$args7" "$args8" ;;
+        0) set -- ;;
+        1) set -- "$args0" ;;
+        2) set -- "$args0" "$args1" ;;
+        3) set -- "$args0" "$args1" "$args2" ;;
+        4) set -- "$args0" "$args1" "$args2" "$args3" ;;
+        5) set -- "$args0" "$args1" "$args2" "$args3" "$args4" ;;
+        6) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" ;;
+        7) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" ;;
+        8) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" "$args7" ;;
+        9) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" "$args7" "$args8" ;;
     esac
 fi
 
-# Split up the JVM_OPTS And GRADLE_OPTS values into an array, following the shell quoting and substitution rules
-function splitJvmOpts() {
-    JVM_OPTS=("$@")
+# Escape application args
+save () {
+    for i do printf %s\\n "$i" | sed "s/'/'\\\\''/g;1s/^/'/;\$s/\$/' \\\\/" ; done
+    echo " "
 }
-eval splitJvmOpts $DEFAULT_JVM_OPTS $JAVA_OPTS $GRADLE_OPTS
-JVM_OPTS[${#JVM_OPTS[*]}]="-Dorg.gradle.appname=$APP_BASE_NAME"
+APP_ARGS=`save "$@"`
 
-exec "$JAVACMD" "${JVM_OPTS[@]}" -classpath "$CLASSPATH" org.gradle.wrapper.GradleWrapperMain "$@"
+# Collect all arguments for the java command, following the shell quoting and substitution rules
+eval set -- $DEFAULT_JVM_OPTS $JAVA_OPTS $GRADLE_OPTS "\"-Dorg.gradle.appname=$APP_BASE_NAME\"" -classpath "\"$CLASSPATH\"" org.gradle.wrapper.GradleWrapperMain "$APP_ARGS"
+
+exec "$JAVACMD" "$@"

gradlew.bat

@@ -1,90 +1,100 @@
-@if "%DEBUG%" == "" @echo off
-@rem ##########################################################################
-@rem
-@rem  Gradle startup script for Windows
-@rem
-@rem ##########################################################################
-
-@rem Set local scope for the variables with windows NT shell
-if "%OS%"=="Windows_NT" setlocal
-
-set DIRNAME=%~dp0
-if "%DIRNAME%" == "" set DIRNAME=.
-set APP_BASE_NAME=%~n0
-set APP_HOME=%DIRNAME%
-
-@rem Add default JVM options here. You can also use JAVA_OPTS and GRADLE_OPTS to pass JVM options to this script.
-set DEFAULT_JVM_OPTS=
-
-@rem Find java.exe
-if defined JAVA_HOME goto findJavaFromJavaHome
-
-set JAVA_EXE=java.exe
-%JAVA_EXE% -version >NUL 2>&1
-if "%ERRORLEVEL%" == "0" goto init
-
-echo.
-echo ERROR: JAVA_HOME is not set and no 'java' command could be found in your PATH.
-echo.
-echo Please set the JAVA_HOME variable in your environment to match the
-echo location of your Java installation.
-
-goto fail
-
-:findJavaFromJavaHome
-set JAVA_HOME=%JAVA_HOME:"=%
-set JAVA_EXE=%JAVA_HOME%/bin/java.exe
-
-if exist "%JAVA_EXE%" goto init
-
-echo.
-echo ERROR: JAVA_HOME is set to an invalid directory: %JAVA_HOME%
-echo.
-echo Please set the JAVA_HOME variable in your environment to match the
-echo location of your Java installation.
-
-goto fail
-
-:init
-@rem Get command-line arguments, handling Windows variants
-
-if not "%OS%" == "Windows_NT" goto win9xME_args
-if "%@eval[2+2]" == "4" goto 4NT_args
-
-:win9xME_args
-@rem Slurp the command line arguments.
-set CMD_LINE_ARGS=
-set _SKIP=2
-
-:win9xME_args_slurp
-if "x%~1" == "x" goto execute
-
-set CMD_LINE_ARGS=%*
-goto execute
-
-:4NT_args
-@rem Get arguments from the 4NT Shell from JP Software
-set CMD_LINE_ARGS=%$
-
-:execute
-@rem Setup the command line
-
-set CLASSPATH=%APP_HOME%\gradle\wrapper\gradle-wrapper.jar
-
-@rem Execute Gradle
-"%JAVA_EXE%" %DEFAULT_JVM_OPTS% %JAVA_OPTS% %GRADLE_OPTS% "-Dorg.gradle.appname=%APP_BASE_NAME%" -classpath "%CLASSPATH%" org.gradle.wrapper.GradleWrapperMain %CMD_LINE_ARGS%
-
-:end
-@rem End local scope for the variables with windows NT shell
-if "%ERRORLEVEL%"=="0" goto mainEnd
-
-:fail
-rem Set variable GRADLE_EXIT_CONSOLE if you need the _script_ return code instead of
-rem the _cmd.exe /c_ return code!
-if  not "" == "%GRADLE_EXIT_CONSOLE%" exit 1
-exit /b 1
-
-:mainEnd
-if "%OS%"=="Windows_NT" endlocal
-
-:omega
+@rem
+@rem Copyright 2015 the original author or authors.
+@rem
+@rem Licensed under the Apache License, Version 2.0 (the "License");
+@rem you may not use this file except in compliance with the License.
+@rem You may obtain a copy of the License at
+@rem
+@rem      https://www.apache.org/licenses/LICENSE-2.0
+@rem
+@rem Unless required by applicable law or agreed to in writing, software
+@rem distributed under the License is distributed on an "AS IS" BASIS,
+@rem WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+@rem See the License for the specific language governing permissions and
+@rem limitations under the License.
+@rem
+
+@if "%DEBUG%" == "" @echo off
+@rem ##########################################################################
+@rem
+@rem  Gradle startup script for Windows
+@rem
+@rem ##########################################################################
+
+@rem Set local scope for the variables with windows NT shell
+if "%OS%"=="Windows_NT" setlocal
+
+set DIRNAME=%~dp0
+if "%DIRNAME%" == "" set DIRNAME=.
+set APP_BASE_NAME=%~n0
+set APP_HOME=%DIRNAME%
+
+@rem Add default JVM options here. You can also use JAVA_OPTS and GRADLE_OPTS to pass JVM options to this script.
+set DEFAULT_JVM_OPTS="-Xmx64m" "-Xms64m"
+
+@rem Find java.exe
+if defined JAVA_HOME goto findJavaFromJavaHome
+
+set JAVA_EXE=java.exe
+%JAVA_EXE% -version >NUL 2>&1
+if "%ERRORLEVEL%" == "0" goto init
+
+echo.
+echo ERROR: JAVA_HOME is not set and no 'java' command could be found in your PATH.
+echo.
+echo Please set the JAVA_HOME variable in your environment to match the
+echo location of your Java installation.
+
+goto fail
+
+:findJavaFromJavaHome
+set JAVA_HOME=%JAVA_HOME:"=%
+set JAVA_EXE=%JAVA_HOME%/bin/java.exe
+
+if exist "%JAVA_EXE%" goto init
+
+echo.
+echo ERROR: JAVA_HOME is set to an invalid directory: %JAVA_HOME%
+echo.
+echo Please set the JAVA_HOME variable in your environment to match the
+echo location of your Java installation.
+
+goto fail
+
+:init
+@rem Get command-line arguments, handling Windows variants
+
+if not "%OS%" == "Windows_NT" goto win9xME_args
+
+:win9xME_args
+@rem Slurp the command line arguments.
+set CMD_LINE_ARGS=
+set _SKIP=2
+
+:win9xME_args_slurp
+if "x%~1" == "x" goto execute
+
+set CMD_LINE_ARGS=%*
+
+:execute
+@rem Setup the command line
+
+set CLASSPATH=%APP_HOME%\gradle\wrapper\gradle-wrapper.jar
+
+@rem Execute Gradle
+"%JAVA_EXE%" %DEFAULT_JVM_OPTS% %JAVA_OPTS% %GRADLE_OPTS% "-Dorg.gradle.appname=%APP_BASE_NAME%" -classpath "%CLASSPATH%" org.gradle.wrapper.GradleWrapperMain %CMD_LINE_ARGS%
+
+:end
+@rem End local scope for the variables with windows NT shell
+if "%ERRORLEVEL%"=="0" goto mainEnd
+
+:fail
+rem Set variable GRADLE_EXIT_CONSOLE if you need the _script_ return code instead of
+rem the _cmd.exe /c_ return code!
+if  not "" == "%GRADLE_EXIT_CONSOLE%" exit 1
+exit /b 1
+
+:mainEnd
+if "%OS%"=="Windows_NT" endlocal
+
+:omega

imageio/image_enc.c

@@ -29,11 +29,13 @@
                          // code with COBJMACROS.
 #include <ole2.h>  // CreateStreamOnHGlobal()
 #include <shlwapi.h>
+#include <tchar.h>
 #include <windows.h>
 #include <wincodec.h>
 #endif
 
 #include "./imageio_util.h"
+#include "../examples/unicode.h"
 
 //------------------------------------------------------------------------------
 // PNG
@@ -61,11 +63,12 @@ static HRESULT CreateOutputStream(const char* out_file_name,
     // 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));
+    IFS(SHCreateStreamOnFile((const LPTSTR)out_file_name,
+                             STGM_WRITE | STGM_CREATE, stream));
   }
   if (FAILED(hr)) {
-    fprintf(stderr, "Error opening output file %s (%08lx)\n",
-            out_file_name, hr);
+    _ftprintf(stderr, _T("Error opening output file %s (%08lx)\n"),
+              (const LPTSTR)out_file_name, hr);
   }
   return hr;
 }
@@ -549,7 +552,8 @@ int WebPSaveImage(const WebPDecBuffer* const buffer,
                   const char* const out_file_name) {
   FILE* fout = NULL;
   int needs_open_file = 1;
-  const int use_stdout = (out_file_name != NULL) && !strcmp(out_file_name, "-");
+  const int use_stdout =
+      (out_file_name != NULL) && !WSTRCMP(out_file_name, "-");
   int ok = 1;
 
   if (buffer == NULL || out_file_name == NULL) return 0;
@@ -560,9 +564,10 @@ int WebPSaveImage(const WebPDecBuffer* const buffer,
 
   if (needs_open_file) {
     fout = use_stdout ? ImgIoUtilSetBinaryMode(stdout)
-                      : fopen(out_file_name, "wb");
+                      : WFOPEN(out_file_name, "wb");
     if (fout == NULL) {
-      fprintf(stderr, "Error opening output file %s\n", out_file_name);
+      WFPRINTF(stderr, "Error opening output file %s\n",
+               (const W_CHAR*)out_file_name);
       return 0;
     }
   }

imageio/imageio_util.c

@@ -18,6 +18,7 @@
 #endif
 #include <stdlib.h>
 #include <string.h>
+#include "../examples/unicode.h"
 
 // -----------------------------------------------------------------------------
 // File I/O
@@ -73,7 +74,7 @@ int ImgIoUtilReadFile(const char* const file_name,
   uint8_t* file_data;
   size_t file_size;
   FILE* in;
-  const int from_stdin = (file_name == NULL) || !strcmp(file_name, "-");
+  const int from_stdin = (file_name == NULL) || !WSTRCMP(file_name, "-");
 
   if (from_stdin) return ImgIoUtilReadFromStdin(data, data_size);
 
@@ -81,29 +82,29 @@ int ImgIoUtilReadFile(const char* const file_name,
   *data = NULL;
   *data_size = 0;
 
-  in = fopen(file_name, "rb");
+  in = WFOPEN(file_name, "rb");
   if (in == NULL) {
-    fprintf(stderr, "cannot open input file '%s'\n", file_name);
+    WFPRINTF(stderr, "cannot open input file '%s'\n", (const W_CHAR*)file_name);
     return 0;
   }
   fseek(in, 0, SEEK_END);
   file_size = ftell(in);
   fseek(in, 0, SEEK_SET);
   // we allocate one extra byte for the \0 terminator
-  file_data = (uint8_t*)malloc(file_size + 1);
+  file_data = (uint8_t*)WebPMalloc(file_size + 1);
   if (file_data == NULL) {
     fclose(in);
-    fprintf(stderr, "memory allocation failure when reading file %s\n",
-            file_name);
+    WFPRINTF(stderr, "memory allocation failure when reading file %s\n",
+             (const W_CHAR*)file_name);
     return 0;
   }
   ok = (fread(file_data, file_size, 1, in) == 1);
   fclose(in);
 
   if (!ok) {
-    fprintf(stderr, "Could not read %d bytes of data from file %s\n",
-            (int)file_size, file_name);
-    free(file_data);
+    WFPRINTF(stderr, "Could not read %d bytes of data from file %s\n",
+             (int)file_size, (const W_CHAR*)file_name);
+    WebPFree(file_data);
     return 0;
   }
   file_data[file_size] = '\0';  // convenient 0-terminator
@@ -118,14 +119,15 @@ int ImgIoUtilWriteFile(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, "-");
+  const int to_stdout = (file_name == NULL) || !WSTRCMP(file_name, "-");
 
   if (data == NULL) {
     return 0;
   }
-  out = to_stdout ? ImgIoUtilSetBinaryMode(stdout) : fopen(file_name, "wb");
+  out = to_stdout ? ImgIoUtilSetBinaryMode(stdout) : WFOPEN(file_name, "wb");
   if (out == NULL) {
-    fprintf(stderr, "Error! Cannot open output file '%s'\n", file_name);
+    WFPRINTF(stderr, "Error! Cannot open output file '%s'\n",
+             (const W_CHAR*)file_name);
     return 0;
   }
   ok = (fwrite(data, data_size, 1, out) == 1);

imageio/imageio_util.h

@@ -29,7 +29,7 @@ FILE* ImgIoUtilSetBinaryMode(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().
+// be deleted using WebPFree().
 // Note: for convenience, the data will be null-terminated with an extra byte
 // (not accounted for in *data_size), in case the file is text and intended
 // to be used as a C-string.

imageio/jpegdec.c

@@ -206,6 +206,7 @@ struct my_error_mgr {
 
 static void my_error_exit(j_common_ptr dinfo) {
   struct my_error_mgr* myerr = (struct my_error_mgr*)dinfo->err;
+  fprintf(stderr, "libjpeg error: ");
   dinfo->err->output_message(dinfo);
   longjmp(myerr->setjmp_buffer, 1);
 }

imageio/pngdec.c

@@ -259,6 +259,15 @@ int ReadPNG(const uint8_t* const data, size_t data_size,
     goto End;
   }
 
+#if LOCAL_PNG_PREREQ(1,5) || \
+    (LOCAL_PNG_PREREQ(1,4) && PNG_LIBPNG_VER_RELEASE >= 1)
+  // If it looks like the bitstream is going to need more memory than libpng's
+  // internal limit (default: 8M), try to (reasonably) raise it.
+  if (data_size > png_get_chunk_malloc_max(png) && data_size < (1u << 24)) {
+    png_set_chunk_malloc_max(png, data_size);
+  }
+#endif
+
   info = png_create_info_struct(png);
   if (info == NULL) goto Error;
   end_info = png_create_info_struct(png);

imageio/pnmdec.c

@@ -26,15 +26,15 @@ typedef enum {
   DEPTH_FLAG      = 1 << 2,
   MAXVAL_FLAG     = 1 << 3,
   TUPLE_FLAG      = 1 << 4,
-  ALL_NEEDED_FLAGS = 0x1f
+  ALL_NEEDED_FLAGS = WIDTH_FLAG | HEIGHT_FLAG | DEPTH_FLAG | MAXVAL_FLAG
 } PNMFlags;
 
 typedef struct {
   const uint8_t* data;
   size_t data_size;
   int width, height;
-  int bytes_per_px;   // 1, 3, 4
-  int depth;
+  int bytes_per_px;
+  int depth;          // 1 (grayscale), 2 (grayscale + alpha), 3 (rgb), 4 (rgba)
   int max_value;
   int type;           // 5, 6 or 7
   int seen_flags;
@@ -74,6 +74,7 @@ static size_t ReadPAMFields(PNMInfo* const info, size_t off) {
   char out[MAX_LINE_SIZE + 1];
   size_t out_size;
   int tmp;
+  int expected_depth = -1;
   assert(info != NULL);
   while (1) {
     off = ReadLine(info->data, off, info->data_size, out, &out_size);
@@ -95,13 +96,16 @@ static size_t ReadPAMFields(PNMInfo* const info, size_t off) {
       info->seen_flags |= MAXVAL_FLAG;
       info->max_value = tmp;
     } else if (!strcmp(out, "TUPLTYPE RGB_ALPHA")) {
-      info->bytes_per_px = 4;
+      expected_depth = 4;
       info->seen_flags |= TUPLE_FLAG;
     } else if (!strcmp(out, "TUPLTYPE RGB")) {
-      info->bytes_per_px = 3;
+      expected_depth = 3;
+      info->seen_flags |= TUPLE_FLAG;
+    } else if (!strcmp(out, "TUPLTYPE GRAYSCALE_ALPHA")) {
+      expected_depth = 2;
       info->seen_flags |= TUPLE_FLAG;
     } else if (!strcmp(out, "TUPLTYPE GRAYSCALE")) {
-      info->bytes_per_px = 1;
+      expected_depth = 1;
       info->seen_flags |= TUPLE_FLAG;
     } else if (!strcmp(out, "ENDHDR")) {
       break;
@@ -110,23 +114,24 @@ static size_t ReadPAMFields(PNMInfo* const info, size_t off) {
       int i;
       if (out_size > 20) sprintf(out + 20 - strlen(kEllipsis), kEllipsis);
       for (i = 0; i < (int)strlen(out); ++i) {
-        if (!isprint(out[i])) out[i] = ' ';
+        // isprint() might trigger a "char-subscripts" warning if given a char.
+        if (!isprint((int)out[i])) out[i] = ' ';
       }
       fprintf(stderr, "PAM header error: unrecognized entry [%s]\n", out);
       return 0;
     }
   }
-  if (!(info->seen_flags & TUPLE_FLAG)) {
-    if (info->depth > 0 && info->depth <= 4 && info->depth != 2) {
-      info->seen_flags |= TUPLE_FLAG;
-      info->bytes_per_px = info->depth * (info->max_value > 255 ? 2 : 1);
-    } else {
-      fprintf(stderr, "PAM: invalid bitdepth (%d).\n", info->depth);
-      return 0;
-    }
+  if (!(info->seen_flags & ALL_NEEDED_FLAGS)) {
+    fprintf(stderr, "PAM header error: missing tags%s%s%s%s\n",
+            (info->seen_flags & WIDTH_FLAG) ? "" : " WIDTH",
+            (info->seen_flags & HEIGHT_FLAG) ? "" : " HEIGHT",
+            (info->seen_flags & DEPTH_FLAG) ? "" : " DEPTH",
+            (info->seen_flags & MAXVAL_FLAG) ? "" : " MAXVAL");
+    return 0;
   }
-  if (info->seen_flags != ALL_NEEDED_FLAGS) {
-    fprintf(stderr, "PAM: incomplete header.\n");
+  if (expected_depth != -1 && info->depth != expected_depth) {
+    fprintf(stderr, "PAM header error: expected DEPTH %d but got DEPTH %d\n",
+            expected_depth, info->depth);
     return 0;
   }
   return off;
@@ -160,16 +165,15 @@ static size_t ReadHeader(PNMInfo* const info) {
 
     // finish initializing missing fields
     info->depth = (info->type == 5) ? 1 : 3;
-    info->bytes_per_px = info->depth * (info->max_value > 255 ? 2 : 1);
   }
   // perform some basic numerical validation
   if (info->width <= 0 || info->height <= 0 ||
       info->type <= 0 || info->type >= 9 ||
-      info->depth <= 0 || info->depth == 2 || info->depth > 4 ||
-      info->bytes_per_px < info->depth ||
+      info->depth <= 0 || info->depth > 4 ||
       info->max_value <= 0 || info->max_value >= 65536) {
     return 0;
   }
+  info->bytes_per_px = info->depth * (info->max_value > 255 ? 2 : 1);
   return off;
 }
 
@@ -178,7 +182,7 @@ int ReadPNM(const uint8_t* const data, size_t data_size,
             struct Metadata* const metadata) {
   int ok = 0;
   int i, j;
-  uint64_t stride, pixel_bytes;
+  uint64_t stride, pixel_bytes, sample_size, depth;
   uint8_t* rgb = NULL, *tmp_rgb;
   size_t offset;
   PNMInfo info;
@@ -209,8 +213,10 @@ int ReadPNM(const uint8_t* const data, size_t data_size,
     fprintf(stderr, "Truncated PNM file (P%d).\n", info.type);
     goto End;
   }
-  stride =
-      (uint64_t)(info.bytes_per_px < 3 ? 3 : info.bytes_per_px) * info.width;
+  sample_size = (info.max_value > 255) ? 2 : 1;
+  // final depth
+  depth = (info.depth == 1 || info.depth == 3 || !keep_alpha) ? 3 : 4;
+  stride = depth * info.width;
   if (stride != (size_t)stride ||
       !ImgIoUtilCheckSizeArgumentsOverflow(stride, info.height)) {
     goto End;
@@ -219,30 +225,63 @@ int ReadPNM(const uint8_t* const data, size_t data_size,
   rgb = (uint8_t*)malloc((size_t)stride * info.height);
   if (rgb == NULL) goto End;
 
-  // Convert input
+  // Convert input.
+  // We only optimize for the sample_size=1, max_value=255, depth=1 case.
   tmp_rgb = rgb;
   for (j = 0; j < info.height; ++j) {
-    assert(offset + info.bytes_per_px * info.width <= data_size);
-    if (info.depth == 1) {
-      // convert grayscale -> RGB
-      for (i = 0; i < info.width; ++i) {
-        const uint8_t v = data[offset + i];
-        tmp_rgb[3 * i + 0] = tmp_rgb[3 * i + 1] = tmp_rgb[3 * i + 2] = v;
-      }
-    } else if (info.depth == 3) {   // RGB
-      memcpy(tmp_rgb, data + offset, 3 * info.width * sizeof(*data));
-    } else if (info.depth == 4) {   // RGBA
-      memcpy(tmp_rgb, data + offset, 4 * info.width * sizeof(*data));
-    }
+    const uint8_t* in = data + offset;
     offset += info.bytes_per_px * info.width;
+    assert(offset <= data_size);
+    if (info.max_value == 255 && info.depth >= 3) {
+      // RGB or RGBA
+      if (info.depth == 3 || keep_alpha) {
+        memcpy(tmp_rgb, in, info.depth * info.width * sizeof(*in));
+      } else {
+        assert(info.depth == 4 && !keep_alpha);
+        for (i = 0; i < info.width; ++i) {
+          tmp_rgb[3 * i + 0] = in[4 * i + 0];
+          tmp_rgb[3 * i + 1] = in[4 * i + 1];
+          tmp_rgb[3 * i + 2] = in[4 * i + 2];
+        }
+      }
+    } else {
+      // Unoptimized case, we need to handle non-trivial operations:
+      //   * convert 16b to 8b (if max_value > 255)
+      //   * rescale to [0..255] range (if max_value != 255)
+      //   * drop the alpha channel (if keep_alpha is false)
+      const uint32_t round = info.max_value / 2;
+      int k = 0;
+      for (i = 0; i < info.width * info.depth; ++i) {
+        uint32_t v = (sample_size == 2) ? 256u * in[2 * i + 0] + in[2 * i + 1]
+                   : in[i];
+        if (info.max_value != 255) v = (v * 255u + round) / info.max_value;
+        if (v > 255u) v = 255u;
+        if (info.depth > 2) {
+          if (!keep_alpha && info.depth == 4 && (i % 4) == 3) {
+            // skip alpha
+          } else {
+            tmp_rgb[k] = v;
+            k += 1;
+          }
+        } else if (info.depth == 1 || (i % 2) == 0) {
+          tmp_rgb[k + 0] = tmp_rgb[k + 1] = tmp_rgb[k + 2] = v;
+          k += 3;
+        } else if (keep_alpha && info.depth == 2) {
+          tmp_rgb[k] = v;
+          k += 1;
+        } else {
+          // skip alpha
+        }
+      }
+    }
     tmp_rgb += stride;
   }
 
   // WebP conversion.
   pic->width = info.width;
   pic->height = info.height;
-  ok = (info.depth == 4) ? WebPPictureImportRGBA(pic, rgb, (int)stride)
-                         : WebPPictureImportRGB(pic, rgb, (int)stride);
+  ok = (depth == 4) ? WebPPictureImportRGBA(pic, rgb, (int)stride)
+                    : WebPPictureImportRGB(pic, rgb, (int)stride);
   if (!ok) goto End;
 
   ok = 1;

imageio/tiffdec.c

@@ -156,7 +156,7 @@ int ReadTIFF(const uint8_t* const data, size_t data_size,
              Metadata* const metadata) {
   MyData my_data = { data, (toff_t)data_size, 0 };
   TIFF* tif;
-  uint32_t width, height;
+  uint32_t image_width, image_height, tile_width, tile_height;
   uint16_t samples_per_px = 0;
   uint16_t extra_samples = 0;
   uint16_t* extra_samples_ptr = NULL;
@@ -189,15 +189,25 @@ int ReadTIFF(const uint8_t* const data, size_t data_size,
   }
   if (samples_per_px < 3 || samples_per_px > 4) goto End;  // not supported
 
-  if (!(TIFFGetField(tif, TIFFTAG_IMAGEWIDTH, &width) &&
-        TIFFGetField(tif, TIFFTAG_IMAGELENGTH, &height))) {
+  if (!(TIFFGetField(tif, TIFFTAG_IMAGEWIDTH, &image_width) &&
+        TIFFGetField(tif, TIFFTAG_IMAGELENGTH, &image_height))) {
     fprintf(stderr, "Error! Cannot retrieve TIFF image dimensions.\n");
     goto End;
   }
-  if (!ImgIoUtilCheckSizeArgumentsOverflow((uint64_t)width * height,
+  if (!ImgIoUtilCheckSizeArgumentsOverflow((uint64_t)image_width * image_height,
                                            sizeof(*raster))) {
     goto End;
   }
+  // According to spec, a tile can be bigger than the image. However it should
+  // be a multiple of 16 and not way too large, so check that it's not more than
+  // twice the image size, for dimensions above some arbitrary minimum 32.
+  if ((TIFFGetField(tif, TIFFTAG_TILEWIDTH, &tile_width) &&
+       tile_width > 32 && tile_width / 2 > image_width) ||
+      (TIFFGetField(tif, TIFFTAG_TILELENGTH, &tile_height) &&
+       tile_height > 32 && tile_height / 2 > image_height)) {
+    fprintf(stderr, "Error! TIFF tile dimensions are too big.\n");
+    goto End;
+  }
   if (samples_per_px > 3 && !TIFFGetField(tif, TIFFTAG_EXTRASAMPLES,
                                           &extra_samples, &extra_samples_ptr)) {
     fprintf(stderr, "Error! Cannot retrieve TIFF ExtraSamples info.\n");
@@ -205,27 +215,28 @@ int ReadTIFF(const uint8_t* const data, size_t data_size,
   }
 
   // _Tiffmalloc uses a signed type for size.
-  alloc_size = (int64_t)((uint64_t)width * height * sizeof(*raster));
+  alloc_size =
+      (int64_t)((uint64_t)image_width * image_height * sizeof(*raster));
   if (alloc_size < 0 || alloc_size != (tsize_t)alloc_size) goto End;
 
   raster = (uint32*)_TIFFmalloc((tsize_t)alloc_size);
   if (raster != NULL) {
-    if (TIFFReadRGBAImageOriented(tif, width, height, raster,
+    if (TIFFReadRGBAImageOriented(tif, image_width, image_height, raster,
                                   ORIENTATION_TOPLEFT, 1)) {
-      const int stride = width * sizeof(*raster);
-      pic->width = width;
-      pic->height = height;
+      const int stride = image_width * sizeof(*raster);
+      pic->width = image_width;
+      pic->height = image_height;
       // TIFF data is ABGR
 #ifdef WORDS_BIGENDIAN
-      TIFFSwabArrayOfLong(raster, width * height);
+      TIFFSwabArrayOfLong(raster, image_width * image_height);
 #endif
       // if we have an alpha channel, we must un-multiply from rgbA to RGBA
       if (extra_samples == 1 && extra_samples_ptr != NULL &&
           extra_samples_ptr[0] == EXTRASAMPLE_ASSOCALPHA) {
         uint32_t y;
         uint8_t* tmp = (uint8_t*)raster;
-        for (y = 0; y < height; ++y) {
-          MultARGBRow(tmp, width);
+        for (y = 0; y < image_height; ++y) {
+          MultARGBRow(tmp, image_width);
           tmp += stride;
         }
       }

imageio/webpdec.c

@@ -22,6 +22,7 @@
 #include "webp/decode.h"
 #include "webp/demux.h"
 #include "webp/encode.h"
+#include "../examples/unicode.h"
 #include "./imageio_util.h"
 #include "./metadata.h"
 
@@ -43,7 +44,7 @@ static void PrintAnimationWarning(const WebPDecoderConfig* const config) {
 }
 
 void PrintWebPError(const char* const in_file, int status) {
-  fprintf(stderr, "Decoding of %s failed.\n", in_file);
+  WFPRINTF(stderr, "Decoding of %s failed.\n", (const W_CHAR*)in_file);
   fprintf(stderr, "Status: %d", status);
   if (status >= VP8_STATUS_OK && status <= VP8_STATUS_NOT_ENOUGH_DATA) {
     fprintf(stderr, "(%s)", kStatusMessages[status]);
@@ -64,7 +65,7 @@ int LoadWebP(const char* const in_file,
 
   status = WebPGetFeatures(*data, *data_size, bitstream);
   if (status != VP8_STATUS_OK) {
-    free((void*)*data);
+    WebPFree((void*)*data);
     *data = NULL;
     *data_size = 0;
     PrintWebPError(in_file, status);

imageio/wicdec.c

@@ -29,12 +29,14 @@
                          // code with COBJMACROS.
 #include <ole2.h>  // CreateStreamOnHGlobal()
 #include <shlwapi.h>
+#include <tchar.h>
 #include <windows.h>
 #include <wincodec.h>
 
-#include "webp/encode.h"
+#include "../examples/unicode.h"
 #include "./imageio_util.h"
 #include "./metadata.h"
+#include "webp/encode.h"
 
 #define IFS(fn)                                                     \
   do {                                                              \
@@ -85,7 +87,7 @@ WEBP_DEFINE_GUID(GUID_WICPixelFormat64bppRGBA_,
 
 static HRESULT OpenInputStream(const char* filename, IStream** stream) {
   HRESULT hr = S_OK;
-  if (!strcmp(filename, "-")) {
+  if (!WSTRCMP(filename, "-")) {
     const uint8_t* data = NULL;
     size_t data_size = 0;
     const int ok = ImgIoUtilReadFile(filename, &data, &data_size);
@@ -108,11 +110,12 @@ static HRESULT OpenInputStream(const char* filename, IStream** stream) {
       hr = E_FAIL;
     }
   } else {
-    IFS(SHCreateStreamOnFileA(filename, STGM_READ, stream));
+    IFS(SHCreateStreamOnFile((const LPTSTR)filename, STGM_READ, stream));
   }
 
   if (FAILED(hr)) {
-    fprintf(stderr, "Error opening input file %s (%08lx)\n", filename, hr);
+    _ftprintf(stderr, _T("Error opening input file %s (%08lx)\n"),
+              (const LPTSTR)filename, hr);
   }
   return hr;
 }
@@ -295,9 +298,15 @@ int ReadPictureWithWIC(const char* const filename,
           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;
+  if (SUCCEEDED(hr)) {
+    if (frame_count == 0) {
+      fprintf(stderr, "No frame found in input file.\n");
+      hr = E_FAIL;
+    } else if (frame_count > 1) {
+      // WIC will be tried before native WebP decoding so avoid duplicating the
+      // error message.
+      hr = E_FAIL;
+    }
   }
   IFS(IWICBitmapDecoder_GetFrame(decoder, 0, &frame));
   IFS(IWICBitmapFrameDecode_GetPixelFormat(frame, &src_pixel_format));

iosbuild.sh

@@ -1,8 +1,9 @@
 #!/bin/bash
 #
-# This script generates 'WebP.framework' and 'WebPDecoder.framework'. An iOS
-# app can decode WebP images by including 'WebPDecoder.framework' and both
-# encode and decode WebP images by including 'WebP.framework'.
+# This script generates 'WebP.framework' and 'WebPDecoder.framework',
+# 'WebPDemux.framework' and 'WebPMux.framework'.
+# An iOS app can decode WebP images by including 'WebPDecoder.framework' and
+# both encode and decode WebP images by including 'WebP.framework'.
 #
 # Run ./iosbuild.sh to generate the frameworks under the current directory
 # (the previous build will be erased if it exists).
@@ -12,6 +13,9 @@
 
 set -e
 
+# Set this variable based on the desired minimum deployment target.
+readonly IOS_MIN_VERSION=6.0
+
 # 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)}'
@@ -35,35 +39,55 @@ readonly TOPDIR=$(pwd)
 readonly BUILDDIR="${TOPDIR}/iosbuild"
 readonly TARGETDIR="${TOPDIR}/WebP.framework"
 readonly DECTARGETDIR="${TOPDIR}/WebPDecoder.framework"
+readonly MUXTARGETDIR="${TOPDIR}/WebPMux.framework"
+readonly DEMUXTARGETDIR="${TOPDIR}/WebPDemux.framework"
 readonly DEVELOPER=$(xcode-select --print-path)
 readonly PLATFORMSROOT="${DEVELOPER}/Platforms"
 readonly LIPO=$(xcrun -sdk iphoneos${SDK} -find lipo)
 LIBLIST=''
 DECLIBLIST=''
+MUXLIBLIST=''
+DEMUXLIBLIST=''
 
 if [[ -z "${SDK}" ]]; then
   echo "iOS SDK not available"
   exit 1
 elif [[ ${SDK%%.*} -gt 8 ]]; then
   EXTRA_CFLAGS="-fembed-bitcode"
-elif [[ ${SDK} < 6.0 ]]; then
+elif [[ ${SDK%%.*} -le 6 ]]; then
   echo "You need iOS SDK version 6.0 or above"
   exit 1
-else
-  echo "iOS SDK Version ${SDK}"
 fi
 
-rm -rf ${BUILDDIR} ${TARGETDIR} ${DECTARGETDIR}
-mkdir -p ${BUILDDIR} ${TARGETDIR}/Headers/ ${DECTARGETDIR}/Headers/
+echo "Xcode Version: ${XCODE}"
+echo "iOS SDK Version: ${SDK}"
+
+if [[ -e "${BUILDDIR}" || -e "${TARGETDIR}" || -e "${DECTARGETDIR}" \
+      || -e "${MUXTARGETDIR}" || -e "${DEMUXTARGETDIR}" ]]; then
+  cat << EOF
+WARNING: The following directories will be deleted:
+WARNING:   ${BUILDDIR}
+WARNING:   ${TARGETDIR}
+WARNING:   ${DECTARGETDIR}
+WARNING:   ${MUXTARGETDIR}
+WARNING:   ${DEMUXTARGETDIR}
+WARNING: The build will continue in 5 seconds...
+EOF
+  sleep 5
+fi
+rm -rf ${BUILDDIR} ${TARGETDIR} ${DECTARGETDIR} \
+    ${MUXTARGETDIR} ${DEMUXTARGETDIR}
+mkdir -p ${BUILDDIR} ${TARGETDIR}/Headers/ ${DECTARGETDIR}/Headers/ \
+    ${MUXTARGETDIR}/Headers/ ${DEMUXTARGETDIR}/Headers/
 
 if [[ ! -e ${SRCDIR}/configure ]]; then
   if ! (cd ${SRCDIR} && sh autogen.sh); then
-    cat <<EOT
+    cat << EOF
 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
+EOF
     exit 1
   fi
 fi
@@ -97,7 +121,7 @@ for PLATFORM in ${PLATFORMS}; do
   SDKROOT="${PLATFORMSROOT}/"
   SDKROOT+="${PLATFORM}.platform/Developer/SDKs/${PLATFORM}${SDK}.sdk/"
   CFLAGS="-arch ${ARCH2:-${ARCH}} -pipe -isysroot ${SDKROOT} -O3 -DNDEBUG"
-  CFLAGS+=" -miphoneos-version-min=6.0 ${EXTRA_CFLAGS}"
+  CFLAGS+=" -miphoneos-version-min=${IOS_MIN_VERSION} ${EXTRA_CFLAGS}"
 
   set -x
   export PATH="${DEVROOT}/usr/bin:${OLDPATH}"
@@ -105,6 +129,7 @@ for PLATFORM in ${PLATFORMS}; do
     --build=$(${SRCDIR}/config.guess) \
     --disable-shared --enable-static \
     --enable-libwebpdecoder --enable-swap-16bit-csp \
+    --enable-libwebpmux \
     CFLAGS="${CFLAGS}"
   set +x
 
@@ -115,6 +140,8 @@ for PLATFORM in ${PLATFORMS}; do
 
   LIBLIST+=" ${ROOTDIR}/lib/libwebp.a"
   DECLIBLIST+=" ${ROOTDIR}/lib/libwebpdecoder.a"
+  MUXLIBLIST+=" ${ROOTDIR}/lib/libwebpmux.a"
+  DEMUXLIBLIST+=" ${ROOTDIR}/lib/libwebpdemux.a"
 
   make clean
   cd ..
@@ -122,8 +149,22 @@ for PLATFORM in ${PLATFORMS}; do
   export PATH=${OLDPATH}
 done
 
+echo "LIBLIST = ${LIBLIST}"
 cp -a ${SRCDIR}/src/webp/{decode,encode,types}.h ${TARGETDIR}/Headers/
 ${LIPO} -create ${LIBLIST} -output ${TARGETDIR}/WebP
 
+echo "DECLIBLIST = ${DECLIBLIST}"
 cp -a ${SRCDIR}/src/webp/{decode,types}.h ${DECTARGETDIR}/Headers/
 ${LIPO} -create ${DECLIBLIST} -output ${DECTARGETDIR}/WebPDecoder
+
+echo "MUXLIBLIST = ${MUXLIBLIST}"
+cp -a ${SRCDIR}/src/webp/{types,mux,mux_types}.h \
+    ${MUXTARGETDIR}/Headers/
+${LIPO} -create ${MUXLIBLIST} -output ${MUXTARGETDIR}/WebPMux
+
+echo "DEMUXLIBLIST = ${DEMUXLIBLIST}"
+cp -a ${SRCDIR}/src/webp/{decode,types,mux_types,demux}.h \
+    ${DEMUXTARGETDIR}/Headers/
+${LIPO} -create ${DEMUXLIBLIST} -output ${DEMUXTARGETDIR}/WebPDemux
+
+echo  "SUCCESS"

makefile.unix

@@ -118,6 +118,10 @@ GROFF = /usr/bin/groff
 COL = /usr/bin/col
 LDFLAGS = $(EXTRA_LIBS) $(EXTRA_FLAGS) -lm
 
+ifdef BITTRACE
+CFLAGS += -DBITTRACE=$(BITTRACE)
+endif
+
 ANIM_UTIL_OBJS = \
     examples/anim_util.o \
 
@@ -185,6 +189,7 @@ DSP_ENC_OBJS = \
     src/dsp/cost.o \
     src/dsp/cost_mips32.o \
     src/dsp/cost_mips_dsp_r2.o \
+    src/dsp/cost_neon.o \
     src/dsp/cost_sse2.o \
     src/dsp/enc.o \
     src/dsp/enc_mips32.o \
@@ -397,7 +402,7 @@ examples/webpinfo: examples/webpinfo.o
 examples/anim_diff: examples/libanim_util.a examples/libgifdec.a
 examples/anim_diff: src/demux/libwebpdemux.a examples/libexample_util.a
 examples/anim_diff: imageio/libimageio_util.a src/libwebp.a
-examples/anim_diff: EXTRA_LIBS += $(GIF_LIBS)
+examples/anim_diff: override EXTRA_LIBS += $(GIF_LIBS)
 examples/anim_diff: EXTRA_FLAGS += -DWEBP_HAVE_GIF
 examples/anim_dump: examples/libanim_util.a examples/libgifdec.a
 examples/anim_dump: src/demux/libwebpdemux.a
@@ -405,27 +410,27 @@ examples/anim_dump: examples/libexample_util.a
 examples/anim_dump: imageio/libimageio_util.a
 examples/anim_dump: imageio/libimageenc.a
 examples/anim_dump: src/libwebp.a
-examples/anim_dump: EXTRA_LIBS += $(GIF_LIBS) $(DWEBP_LIBS)
+examples/anim_dump: override EXTRA_LIBS += $(GIF_LIBS) $(DWEBP_LIBS)
 examples/cwebp: examples/libexample_util.a
 examples/cwebp: imageio/libimagedec.a
 examples/cwebp: src/demux/libwebpdemux.a
 examples/cwebp: imageio/libimageio_util.a
 examples/cwebp: src/libwebp.a
-examples/cwebp: EXTRA_LIBS += $(CWEBP_LIBS)
+examples/cwebp: override EXTRA_LIBS += $(CWEBP_LIBS)
 examples/dwebp: examples/libexample_util.a
 examples/dwebp: imageio/libimagedec.a
 examples/dwebp: src/demux/libwebpdemux.a
 examples/dwebp: imageio/libimageenc.a
 examples/dwebp: imageio/libimageio_util.a
 examples/dwebp: src/libwebp.a
-examples/dwebp: EXTRA_LIBS += $(DWEBP_LIBS)
+examples/dwebp: override EXTRA_LIBS += $(DWEBP_LIBS)
 examples/gif2webp: examples/libexample_util.a imageio/libimageio_util.a
 examples/gif2webp: examples/libgifdec.a src/mux/libwebpmux.a src/libwebp.a
-examples/gif2webp: EXTRA_LIBS += $(GIF_LIBS)
+examples/gif2webp: override EXTRA_LIBS += $(GIF_LIBS)
 examples/gif2webp: EXTRA_FLAGS += -DWEBP_HAVE_GIF
 examples/vwebp: examples/libexample_util.a src/demux/libwebpdemux.a
 examples/vwebp: imageio/libimageio_util.a src/libwebp.a
-examples/vwebp: EXTRA_LIBS += $(GL_LIBS)
+examples/vwebp: override EXTRA_LIBS += $(GL_LIBS)
 examples/vwebp: EXTRA_FLAGS += -DWEBP_HAVE_GL
 examples/webpmux: examples/libexample_util.a imageio/libimageio_util.a
 examples/webpmux: src/mux/libwebpmux.a src/libwebpdecoder.a
@@ -433,7 +438,7 @@ examples/img2webp: examples/libexample_util.a imageio/libimageio_util.a
 examples/img2webp: imageio/libimagedec.a
 examples/img2webp: src/demux/libwebpdemux.a
 examples/img2webp: src/mux/libwebpmux.a src/libwebp.a
-examples/img2webp: EXTRA_LIBS += $(CWEBP_LIBS)
+examples/img2webp: override EXTRA_LIBS += $(CWEBP_LIBS)
 examples/webpinfo: examples/libexample_util.a imageio/libimageio_util.a
 examples/webpinfo: src/libwebpdecoder.a
 
@@ -442,7 +447,7 @@ extras/get_disto: imageio/libimagedec.a
 extras/get_disto: src/demux/libwebpdemux.a
 extras/get_disto: imageio/libimageio_util.a
 extras/get_disto: src/libwebp.a
-extras/get_disto: EXTRA_LIBS += $(CWEBP_LIBS)
+extras/get_disto: override EXTRA_LIBS += $(CWEBP_LIBS)
 
 extras/webp_quality: extras/webp_quality.o
 extras/webp_quality: imageio/libimageio_util.a
@@ -453,7 +458,7 @@ extras/vwebp_sdl: extras/webp_to_sdl.o
 extras/vwebp_sdl: imageio/libimageio_util.a
 extras/vwebp_sdl: src/libwebp.a
 extras/vwebp_sdl: EXTRA_FLAGS += -DWEBP_HAVE_SDL $(SDL_FLAGS)
-extras/vwebp_sdl: EXTRA_LIBS += $(SDL_LIBS)
+extras/vwebp_sdl: override EXTRA_LIBS += $(SDL_LIBS)
 
 $(OUT_EXAMPLES) $(EXTRA_EXAMPLES) $(OTHER_EXAMPLES):
 	$(CC) -o $@ $^ $(LDFLAGS)

man/cwebp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH CWEBP 1 "January 20, 2017"
+.TH CWEBP 1 "November 19, 2020"
 .SH NAME
 cwebp \- compress an image file to a WebP file
 .SH SYNOPSIS
@@ -13,6 +13,7 @@ command.
 .PP
 \fBcwebp\fP compresses an image using the WebP format.
 Input format can be either PNG, JPEG, TIFF, WebP or raw Y'CbCr samples.
+Note: Animated PNG and WebP files are not supported.
 .SH OPTIONS
 The basic options are:
 .TP
@@ -41,10 +42,12 @@ the invisible pixel values (R/G/B or Y/U/V) will be preserved only if the
 \-exact option is used.
 .TP
 .BI \-near_lossless " int
-Use near\-lossless image preprocessing. This option adjusts pixel values
-to help compressibility, but has minimal impact on the visual quality.
-It triggers lossless compression mode automatically.
-Range is 0 (maximum preprocessing) to 100 (no preprocessing, the default).
+Specify the level of near\-lossless image preprocessing. This option adjusts
+pixel values to help compressibility, but has minimal impact on the visual
+quality. It triggers lossless compression mode automatically. The range is 0
+(maximum preprocessing) to 100 (no preprocessing, the default). The typical
+value is around 60. Note that lossy with \fB\-q 100\fP can at times yield
+better results.
 .TP
 .BI \-q " float
 Specify the compression factor for RGB channels between 0 and 100. The default
@@ -132,6 +135,13 @@ options \fB\-size\fP or \fB\-psnr\fP. Maximum value is 10, default is 1.
 If options \fB\-size\fP or \fB\-psnr\fP were used, but \fB\-pass\fP wasn't
 specified, a default value of '6' passes will be used.
 .TP
+.BI \-qrange " int int
+Specifies the permissible interval for the quality factor. This is particularly
+useful when using multi-pass (\fB\-size\fP or \fB\-psnr\fP options).
+Default is 0 100.
+If the quality factor is outside this range, it will be clamped.
+If the minimum value must be less or equal to the maximum one.
+.TP
 .B \-af
 Turns auto\-filter on. This algorithm will spend additional time optimizing
 the filtering strength to reach a well\-balanced quality.

man/dwebp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH DWEBP 1 "June 23, 2016"
+.TH DWEBP 1 "November 19, 2020"
 .SH NAME
 dwebp \- decompress a WebP file to an image file
 .SH SYNOPSIS
@@ -12,6 +12,7 @@ This manual page documents the
 command.
 .PP
 \fBdwebp\fP decompresses WebP files into PNG, PAM, PPM or PGM images.
+Note: Animated WebP files are not supported.
 .SH OPTIONS
 The basic options are:
 .TP

man/gif2webp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH GIF2WEBP 1 "January 25, 2018"
+.TH GIF2WEBP 1 "May 1, 2020"
 .SH NAME
 gif2webp \- Convert a GIF image to WebP
 .SH SYNOPSIS
@@ -62,9 +62,9 @@ larger file size and lower compression quality.
 .TP
 .BI \-min_size
 Encode image to achieve smallest size. This disables key frame insertion and
-picks the dispose method resulting in smallest output for each frame. It uses
-lossless compression by default, but can be combined with \-q, \-m, \-lossy or
-\-mixed options.
+picks the dispose method resulting in the smallest output for each frame. It
+uses lossless compression by default, but can be combined with \-q, \-m,
+\-lossy or \-mixed options.
 .TP
 .BI \-kmin " int
 .TP

man/img2webp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH IMG2WEBP 1 "April 3, 2018"
+.TH IMG2WEBP 1 "May 1, 2020"
 .SH NAME
 img2webp \- create animated WebP file from a sequence of input images.
 .SH SYNOPSIS
@@ -16,8 +16,8 @@ command.
 \fBimg2webp\fP compresses a sequence of images using the animated WebP format.
 Input images can either be PNG, JPEG, TIFF or WebP.
 If a single file name (not starting with the character '\-') is supplied as
-the argument, the command line argument are actually tokenized from this file.
-This allows for easy scripting or using large number of arguments.
+the argument, the command line arguments are actually tokenized from this file.
+This allows for easy scripting or using a large number of arguments.
 .SH FILE-LEVEL OPTIONS
 The file-level options are applied at the beginning of the compression process,
 before the input frames are read.
@@ -27,7 +27,7 @@ Specify the name of the output WebP file.
 .TP
 .BI \-min_size
 Encode images to achieve smallest size. This disables key frame insertion and
-picks the parameters resulting in smallest output for each frame. It uses
+picks the parameters resulting in the smallest output for each frame. It uses
 lossless compression by default, but can be combined with \-q, \-m, \-lossy or
 \-mixed options.
 .TP

man/vwebp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH VWEBP 1 "July 20, 2018"
+.TH VWEBP 1 "June 5, 2019"
 .SH NAME
 vwebp \- decompress a WebP file and display it in a window
 .SH SYNOPSIS

man/webpmux.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH WEBPMUX 1 "April 23, 2018"
+.TH WEBPMUX 1 "December 12, 2020"
 .SH NAME
 webpmux \- create animated WebP files from non\-animated WebP images, extract
 frames from animated WebP images, and manage XMP/EXIF metadata and ICC profile.
@@ -59,7 +59,7 @@ command.
 to add/extract/strip XMP/EXIF metadata and ICC profile.
 If a single file name (not starting with the character '\-') is supplied as
 the argument, the command line arguments are actually tokenized from this file.
-This allows for easy scripting or using large number of arguments.
+This allows for easy scripting or using a large number of arguments.
 .SH OPTIONS
 .SS GET_OPTIONS (\-get):
 .TP
@@ -77,6 +77,11 @@ Get nth frame from an animated image. (n = 0 has a special meaning: last frame).
 
 .SS SET_OPTIONS (\-set)
 .TP
+.BI loop " loop_count
+Set loop count on an animated file.
+.P
+Where: 'loop_count' must be in range [0, 65535].
+.TP
 .BI icc " file.icc
 Set ICC profile.
 .P

src/Makefile.am

@@ -36,7 +36,7 @@ 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 7:3:0
+libwebp_la_LDFLAGS = -no-undefined -version-info 8:1:1
 libwebpincludedir = $(includedir)/webp
 pkgconfig_DATA = libwebp.pc
 
@@ -48,7 +48,7 @@ if BUILD_LIBWEBPDECODER
   libwebpdecoder_la_LIBADD += dsp/libwebpdspdecode.la
   libwebpdecoder_la_LIBADD += utils/libwebputilsdecode.la
 
-  libwebpdecoder_la_LDFLAGS = -no-undefined -version-info 3:3:0
+  libwebpdecoder_la_LDFLAGS = -no-undefined -version-info 4:1:1
   pkgconfig_DATA += libwebpdecoder.pc
 endif
 

src/dec/frame_dec.c

@@ -732,7 +732,7 @@ 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->mt_method_ > 0) {
+  if (dec->filter_type_ > 0 && 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.

src/dec/idec_dec.c

@@ -166,9 +166,11 @@ static int AppendToMemBuffer(WebPIDecoder* const idec,
   VP8Decoder* const dec = (VP8Decoder*)idec->dec_;
   MemBuffer* const mem = &idec->mem_;
   const int need_compressed_alpha = NeedCompressedAlpha(idec);
-  const uint8_t* const old_start = mem->buf_ + mem->start_;
+  const uint8_t* const old_start =
+      (mem->buf_ == NULL) ? NULL : mem->buf_ + mem->start_;
   const uint8_t* const old_base =
       need_compressed_alpha ? dec->alpha_data_ : old_start;
+  assert(mem->buf_ != NULL || mem->start_ == 0);
   assert(mem->mode_ == MEM_MODE_APPEND);
   if (data_size > MAX_CHUNK_PAYLOAD) {
     // security safeguard: trying to allocate more than what the format
@@ -184,7 +186,7 @@ static int AppendToMemBuffer(WebPIDecoder* const idec,
     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);
+    if (old_base != NULL) memcpy(new_buf, old_base, current_size);
     WebPSafeFree(mem->buf_);
     mem->buf_ = new_buf;
     mem->buf_size_ = (size_t)extra_size;
@@ -192,6 +194,7 @@ static int AppendToMemBuffer(WebPIDecoder* const idec,
     mem->end_ = current_size;
   }
 
+  assert(mem->buf_ != NULL);
   memcpy(mem->buf_ + mem->end_, data, data_size);
   mem->end_ += data_size;
   assert(mem->end_ <= mem->buf_size_);
@@ -204,7 +207,9 @@ 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_buf = mem->buf_;
-  const uint8_t* const old_start = old_buf + mem->start_;
+  const uint8_t* const old_start =
+      (old_buf == NULL) ? NULL : old_buf + mem->start_;
+  assert(old_buf != NULL || mem->start_ == 0);
   assert(mem->mode_ == MEM_MODE_MAP);
 
   if (data_size < mem->buf_size_) return 0;  // can't remap to a shorter buffer!

src/dec/io_dec.c

@@ -25,21 +25,16 @@
 static int EmitYUV(const VP8Io* const io, WebPDecParams* const p) {
   WebPDecBuffer* output = p->output;
   const WebPYUVABuffer* const buf = &output->u.YUVA;
-  uint8_t* const y_dst = buf->y + io->mb_y * buf->y_stride;
-  uint8_t* const u_dst = buf->u + (io->mb_y >> 1) * buf->u_stride;
-  uint8_t* const v_dst = buf->v + (io->mb_y >> 1) * buf->v_stride;
+  uint8_t* const y_dst = buf->y + (size_t)io->mb_y * buf->y_stride;
+  uint8_t* const u_dst = buf->u + (size_t)(io->mb_y >> 1) * buf->u_stride;
+  uint8_t* const v_dst = buf->v + (size_t)(io->mb_y >> 1) * buf->v_stride;
   const int mb_w = io->mb_w;
   const int mb_h = io->mb_h;
   const int uv_w = (mb_w + 1) / 2;
   const int uv_h = (mb_h + 1) / 2;
-  int j;
-  for (j = 0; j < mb_h; ++j) {
-    memcpy(y_dst + j * buf->y_stride, io->y + j * io->y_stride, mb_w);
-  }
-  for (j = 0; j < uv_h; ++j) {
-    memcpy(u_dst + j * buf->u_stride, io->u + j * io->uv_stride, uv_w);
-    memcpy(v_dst + j * buf->v_stride, io->v + j * io->uv_stride, uv_w);
-  }
+  WebPCopyPlane(io->y, io->y_stride, y_dst, buf->y_stride, mb_w, mb_h);
+  WebPCopyPlane(io->u, io->uv_stride, u_dst, buf->u_stride, uv_w, uv_h);
+  WebPCopyPlane(io->v, io->uv_stride, v_dst, buf->v_stride, uv_w, uv_h);
   return io->mb_h;
 }
 
@@ -47,7 +42,7 @@ static int EmitYUV(const VP8Io* const io, WebPDecParams* const p) {
 static int EmitSampledRGB(const VP8Io* const io, WebPDecParams* const p) {
   WebPDecBuffer* const output = p->output;
   WebPRGBABuffer* const buf = &output->u.RGBA;
-  uint8_t* const dst = buf->rgba + io->mb_y * buf->stride;
+  uint8_t* const dst = buf->rgba + (size_t)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,
@@ -62,7 +57,7 @@ static int EmitSampledRGB(const VP8Io* const io, WebPDecParams* const p) {
 static int EmitFancyRGB(const VP8Io* const io, WebPDecParams* const p) {
   int num_lines_out = io->mb_h;   // a priori guess
   const WebPRGBABuffer* const buf = &p->output->u.RGBA;
-  uint8_t* dst = buf->rgba + io->mb_y * buf->stride;
+  uint8_t* dst = buf->rgba + (size_t)io->mb_y * buf->stride;
   WebPUpsampleLinePairFunc upsample = WebPUpsamplers[p->output->colorspace];
   const uint8_t* cur_y = io->y;
   const uint8_t* cur_u = io->u;
@@ -133,7 +128,7 @@ static int EmitAlphaYUV(const VP8Io* const io, WebPDecParams* const p,
   const WebPYUVABuffer* const buf = &p->output->u.YUVA;
   const int mb_w = io->mb_w;
   const int mb_h = io->mb_h;
-  uint8_t* dst = buf->a + io->mb_y * buf->a_stride;
+  uint8_t* dst = buf->a + (size_t)io->mb_y * buf->a_stride;
   int j;
   (void)expected_num_lines_out;
   assert(expected_num_lines_out == mb_h);
@@ -186,7 +181,7 @@ static int EmitAlphaRGB(const VP8Io* const io, WebPDecParams* const p,
         (colorspace == MODE_ARGB || colorspace == MODE_Argb);
     const WebPRGBABuffer* const buf = &p->output->u.RGBA;
     int num_rows;
-    const int start_y = GetAlphaSourceRow(io, &alpha, &num_rows);
+    const size_t start_y = GetAlphaSourceRow(io, &alpha, &num_rows);
     uint8_t* const base_rgba = buf->rgba + start_y * buf->stride;
     uint8_t* const dst = base_rgba + (alpha_first ? 0 : 3);
     const int has_alpha = WebPDispatchAlpha(alpha, io->width, mb_w,
@@ -210,7 +205,7 @@ static int EmitAlphaRGBA4444(const VP8Io* const io, WebPDecParams* const p,
     const WEBP_CSP_MODE colorspace = p->output->colorspace;
     const WebPRGBABuffer* const buf = &p->output->u.RGBA;
     int num_rows;
-    const int start_y = GetAlphaSourceRow(io, &alpha, &num_rows);
+    const size_t start_y = GetAlphaSourceRow(io, &alpha, &num_rows);
     uint8_t* const base_rgba = buf->rgba + start_y * buf->stride;
 #if (WEBP_SWAP_16BIT_CSP == 1)
     uint8_t* alpha_dst = base_rgba;
@@ -276,9 +271,9 @@ static int EmitRescaledYUV(const VP8Io* const io, WebPDecParams* const p) {
 static int EmitRescaledAlphaYUV(const VP8Io* const io, WebPDecParams* const p,
                                 int expected_num_lines_out) {
   const WebPYUVABuffer* const buf = &p->output->u.YUVA;
-  uint8_t* const dst_a = buf->a + p->last_y * buf->a_stride;
+  uint8_t* const dst_a = buf->a + (size_t)p->last_y * buf->a_stride;
   if (io->a != NULL) {
-    uint8_t* const dst_y = buf->y + p->last_y * buf->y_stride;
+    uint8_t* const dst_y = buf->y + (size_t)p->last_y * buf->y_stride;
     const int num_lines_out = Rescale(io->a, io->width, io->mb_h, p->scaler_a);
     assert(expected_num_lines_out == num_lines_out);
     if (num_lines_out > 0) {   // unmultiply the Y
@@ -356,7 +351,7 @@ static int ExportRGB(WebPDecParams* const p, int y_pos) {
   const WebPYUV444Converter convert =
       WebPYUV444Converters[p->output->colorspace];
   const WebPRGBABuffer* const buf = &p->output->u.RGBA;
-  uint8_t* dst = buf->rgba + y_pos * buf->stride;
+  uint8_t* dst = buf->rgba + (size_t)y_pos * buf->stride;
   int num_lines_out = 0;
   // For RGB rescaling, because of the YUV420, current scan position
   // U/V can be +1/-1 line from the Y one.  Hence the double test.
@@ -383,15 +378,15 @@ static int EmitRescaledRGB(const VP8Io* const io, WebPDecParams* const p) {
   while (j < mb_h) {
     const int y_lines_in =
         WebPRescalerImport(p->scaler_y, mb_h - j,
-                           io->y + j * io->y_stride, io->y_stride);
+                           io->y + (size_t)j * io->y_stride, io->y_stride);
     j += y_lines_in;
     if (WebPRescaleNeededLines(p->scaler_u, uv_mb_h - uv_j)) {
-      const int u_lines_in =
-          WebPRescalerImport(p->scaler_u, uv_mb_h - uv_j,
-                             io->u + uv_j * io->uv_stride, io->uv_stride);
-      const int v_lines_in =
-          WebPRescalerImport(p->scaler_v, uv_mb_h - uv_j,
-                             io->v + uv_j * io->uv_stride, io->uv_stride);
+      const int u_lines_in = WebPRescalerImport(
+          p->scaler_u, uv_mb_h - uv_j, io->u + (size_t)uv_j * io->uv_stride,
+          io->uv_stride);
+      const int v_lines_in = WebPRescalerImport(
+          p->scaler_v, uv_mb_h - uv_j, io->v + (size_t)uv_j * io->uv_stride,
+          io->uv_stride);
       (void)v_lines_in;   // remove a gcc warning
       assert(u_lines_in == v_lines_in);
       uv_j += u_lines_in;
@@ -403,7 +398,7 @@ static int EmitRescaledRGB(const VP8Io* const io, WebPDecParams* const p) {
 
 static int ExportAlpha(WebPDecParams* const p, int y_pos, int max_lines_out) {
   const WebPRGBABuffer* const buf = &p->output->u.RGBA;
-  uint8_t* const base_rgba = buf->rgba + y_pos * buf->stride;
+  uint8_t* const base_rgba = buf->rgba + (size_t)y_pos * buf->stride;
   const WEBP_CSP_MODE colorspace = p->output->colorspace;
   const int alpha_first =
       (colorspace == MODE_ARGB || colorspace == MODE_Argb);
@@ -431,7 +426,7 @@ static int ExportAlpha(WebPDecParams* const p, int y_pos, int max_lines_out) {
 static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos,
                                int max_lines_out) {
   const WebPRGBABuffer* const buf = &p->output->u.RGBA;
-  uint8_t* const base_rgba = buf->rgba + y_pos * buf->stride;
+  uint8_t* const base_rgba = buf->rgba + (size_t)y_pos * buf->stride;
 #if (WEBP_SWAP_16BIT_CSP == 1)
   uint8_t* alpha_dst = base_rgba;
 #else
@@ -470,7 +465,7 @@ static int EmitRescaledAlphaRGB(const VP8Io* const io, WebPDecParams* const p,
     int lines_left = expected_num_out_lines;
     const int y_end = p->last_y + lines_left;
     while (lines_left > 0) {
-      const int row_offset = scaler->src_y - io->mb_y;
+      const int64_t row_offset = (int64_t)scaler->src_y - io->mb_y;
       WebPRescalerImport(scaler, io->mb_h + io->mb_y - scaler->src_y,
                          io->a + row_offset * io->width, io->width);
       lines_left -= p->emit_alpha_row(p, y_end - lines_left, lines_left);

src/dec/quant_dec.c

@@ -61,12 +61,17 @@ static const uint16_t kAcTable[128] = {
 
 void VP8ParseQuant(VP8Decoder* const dec) {
   VP8BitReader* const br = &dec->br_;
-  const int base_q0 = VP8GetValue(br, 7);
-  const int dqy1_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dqy2_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dqy2_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dquv_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
-  const int dquv_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
+  const int base_q0 = VP8GetValue(br, 7, "global-header");
+  const int dqy1_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dqy2_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dqy2_ac = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dquv_dc = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
+  const int dquv_ac = VP8Get(br, "global-header") ?
+       VP8GetSignedValue(br, 4, "global-header") : 0;
 
   const VP8SegmentHeader* const hdr = &dec->segment_hdr_;
   int i;

src/dec/tree_dec.c

@@ -296,20 +296,21 @@ static void ParseIntraMode(VP8BitReader* const br,
   // 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]);
+    block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0], "segments")
+                    ?  VP8GetBit(br, dec->proba_.segments_[1], "segments")
+                    :  VP8GetBit(br, dec->proba_.segments_[2], "segments") + 2;
   } else {
     block->segment_ = 0;  // default for intra
   }
-  if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_);
+  if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_, "skip");
 
-  block->is_i4x4_ = !VP8GetBit(br, 145);   // decide for B_PRED first
+  block->is_i4x4_ = !VP8GetBit(br, 145, "block-size");
   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);
+        VP8GetBit(br, 156, "pred-modes") ?
+            (VP8GetBit(br, 128, "pred-modes") ? TM_PRED : H_PRED) :
+            (VP8GetBit(br, 163, "pred-modes") ? V_PRED : DC_PRED);
     block->imodes_[0] = ymode;
     memset(top, ymode, 4 * sizeof(*top));
     memset(left, ymode, 4 * sizeof(*left));
@@ -323,22 +324,25 @@ static void ParseIntraMode(VP8BitReader* const br,
         const uint8_t* const prob = kBModesProba[top[x]][ymode];
 #if (USE_GENERIC_TREE == 1)
         // Generic tree-parsing
-        int i = kYModesIntra4[VP8GetBit(br, prob[0])];
+        int i = kYModesIntra4[VP8GetBit(br, prob[0], "pred-modes")];
         while (i > 0) {
-          i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i])];
+          i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i], "pred-modes")];
         }
         ymode = -i;
 #else
         // Hardcoded tree parsing
-        ymode = !VP8GetBit(br, prob[0]) ? B_DC_PRED :
-                  !VP8GetBit(br, prob[1]) ? B_TM_PRED :
-                    !VP8GetBit(br, prob[2]) ? B_VE_PRED :
-                      !VP8GetBit(br, prob[3]) ?
-                        (!VP8GetBit(br, prob[4]) ? B_HE_PRED :
-                          (!VP8GetBit(br, prob[5]) ? B_RD_PRED : B_VR_PRED)) :
-                        (!VP8GetBit(br, prob[6]) ? B_LD_PRED :
-                          (!VP8GetBit(br, prob[7]) ? B_VL_PRED :
-                            (!VP8GetBit(br, prob[8]) ? B_HD_PRED : B_HU_PRED)));
+        ymode = !VP8GetBit(br, prob[0], "pred-modes") ? B_DC_PRED :
+                  !VP8GetBit(br, prob[1], "pred-modes") ? B_TM_PRED :
+                    !VP8GetBit(br, prob[2], "pred-modes") ? B_VE_PRED :
+                      !VP8GetBit(br, prob[3], "pred-modes") ?
+                        (!VP8GetBit(br, prob[4], "pred-modes") ? B_HE_PRED :
+                          (!VP8GetBit(br, prob[5], "pred-modes") ? B_RD_PRED
+                                                                 : B_VR_PRED)) :
+                        (!VP8GetBit(br, prob[6], "pred-modes") ? B_LD_PRED :
+                          (!VP8GetBit(br, prob[7], "pred-modes") ? B_VL_PRED :
+                            (!VP8GetBit(br, prob[8], "pred-modes") ? B_HD_PRED
+                                                                   : B_HU_PRED))
+                        );
 #endif  // USE_GENERIC_TREE
         top[x] = ymode;
       }
@@ -348,9 +352,9 @@ static void ParseIntraMode(VP8BitReader* const br,
     }
   }
   // Hardcoded UVMode decision tree
-  block->uvmode_ = !VP8GetBit(br, 142) ? DC_PRED
-                 : !VP8GetBit(br, 114) ? V_PRED
-                 : VP8GetBit(br, 183) ? TM_PRED : H_PRED;
+  block->uvmode_ = !VP8GetBit(br, 142, "pred-modes-uv") ? DC_PRED
+                 : !VP8GetBit(br, 114, "pred-modes-uv") ? V_PRED
+                 : VP8GetBit(br, 183, "pred-modes-uv") ? TM_PRED : H_PRED;
 }
 
 int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec) {
@@ -514,8 +518,10 @@ 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) {
-          const int v = VP8GetBit(br, CoeffsUpdateProba[t][b][c][p]) ?
-                        VP8GetValue(br, 8) : CoeffsProba0[t][b][c][p];
+          const int v =
+              VP8GetBit(br, CoeffsUpdateProba[t][b][c][p], "global-header") ?
+                        VP8GetValue(br, 8, "global-header") :
+                        CoeffsProba0[t][b][c][p];
           proba->bands_[t][b].probas_[c][p] = v;
         }
       }
@@ -524,9 +530,8 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
       proba->bands_ptr_[t][b] = &proba->bands_[t][kBands[b]];
     }
   }
-  dec->use_skip_proba_ = VP8Get(br);
+  dec->use_skip_proba_ = VP8Get(br, "global-header");
   if (dec->use_skip_proba_) {
-    dec->skip_p_ = VP8GetValue(br, 8);
+    dec->skip_p_ = VP8GetValue(br, 8, "global-header");
   }
 }
-

src/dec/vp8_dec.c

@@ -161,23 +161,26 @@ static int ParseSegmentHeader(VP8BitReader* br,
                               VP8SegmentHeader* hdr, VP8Proba* proba) {
   assert(br != NULL);
   assert(hdr != NULL);
-  hdr->use_segment_ = VP8Get(br);
+  hdr->use_segment_ = VP8Get(br, "global-header");
   if (hdr->use_segment_) {
-    hdr->update_map_ = VP8Get(br);
-    if (VP8Get(br)) {   // update data
+    hdr->update_map_ = VP8Get(br, "global-header");
+    if (VP8Get(br, "global-header")) {   // update data
       int s;
-      hdr->absolute_delta_ = VP8Get(br);
+      hdr->absolute_delta_ = VP8Get(br, "global-header");
       for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
-        hdr->quantizer_[s] = VP8Get(br) ? VP8GetSignedValue(br, 7) : 0;
+        hdr->quantizer_[s] = VP8Get(br, "global-header") ?
+            VP8GetSignedValue(br, 7, "global-header") : 0;
       }
       for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
-        hdr->filter_strength_[s] = VP8Get(br) ? VP8GetSignedValue(br, 6) : 0;
+        hdr->filter_strength_[s] = VP8Get(br, "global-header") ?
+            VP8GetSignedValue(br, 6, "global-header") : 0;
       }
     }
     if (hdr->update_map_) {
       int s;
       for (s = 0; s < MB_FEATURE_TREE_PROBS; ++s) {
-        proba->segments_[s] = VP8Get(br) ? VP8GetValue(br, 8) : 255u;
+        proba->segments_[s] = VP8Get(br, "global-header") ?
+            VP8GetValue(br, 8, "global-header") : 255u;
       }
     }
   } else {
@@ -205,7 +208,7 @@ static VP8StatusCode ParsePartitions(VP8Decoder* const dec,
   size_t last_part;
   size_t p;
 
-  dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2)) - 1;
+  dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2, "global-header")) - 1;
   last_part = dec->num_parts_minus_one_;
   if (size < 3 * last_part) {
     // we can't even read the sizes with sz[]! That's a failure.
@@ -229,21 +232,21 @@ static VP8StatusCode ParsePartitions(VP8Decoder* const dec,
 // Paragraph 9.4
 static int ParseFilterHeader(VP8BitReader* br, VP8Decoder* const dec) {
   VP8FilterHeader* const hdr = &dec->filter_hdr_;
-  hdr->simple_    = VP8Get(br);
-  hdr->level_     = VP8GetValue(br, 6);
-  hdr->sharpness_ = VP8GetValue(br, 3);
-  hdr->use_lf_delta_ = VP8Get(br);
+  hdr->simple_    = VP8Get(br, "global-header");
+  hdr->level_     = VP8GetValue(br, 6, "global-header");
+  hdr->sharpness_ = VP8GetValue(br, 3, "global-header");
+  hdr->use_lf_delta_ = VP8Get(br, "global-header");
   if (hdr->use_lf_delta_) {
-    if (VP8Get(br)) {   // update lf-delta?
+    if (VP8Get(br, "global-header")) {   // update lf-delta?
       int i;
       for (i = 0; i < NUM_REF_LF_DELTAS; ++i) {
-        if (VP8Get(br)) {
-          hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6);
+        if (VP8Get(br, "global-header")) {
+          hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6, "global-header");
         }
       }
       for (i = 0; i < NUM_MODE_LF_DELTAS; ++i) {
-        if (VP8Get(br)) {
-          hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6);
+        if (VP8Get(br, "global-header")) {
+          hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6, "global-header");
         }
       }
     }
@@ -352,8 +355,8 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
   buf_size -= frm_hdr->partition_length_;
 
   if (frm_hdr->key_frame_) {
-    pic_hdr->colorspace_ = VP8Get(br);
-    pic_hdr->clamp_type_ = VP8Get(br);
+    pic_hdr->colorspace_ = VP8Get(br, "global-header");
+    pic_hdr->clamp_type_ = VP8Get(br, "global-header");
   }
   if (!ParseSegmentHeader(br, &dec->segment_hdr_, &dec->proba_)) {
     return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
@@ -378,7 +381,7 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
                        "Not a key frame.");
   }
 
-  VP8Get(br);   // ignore the value of update_proba_
+  VP8Get(br, "global-header");   // ignore the value of update_proba_
 
   VP8ParseProba(br, dec);
 
@@ -403,28 +406,28 @@ static const uint8_t kZigzag[16] = {
 // 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])) {
+  if (!VP8GetBit(br, p[3], "coeffs")) {
+    if (!VP8GetBit(br, p[4], "coeffs")) {
       v = 2;
     } else {
-      v = 3 + VP8GetBit(br, p[5]);
+      v = 3 + VP8GetBit(br, p[5], "coeffs");
     }
   } else {
-    if (!VP8GetBit(br, p[6])) {
-      if (!VP8GetBit(br, p[7])) {
-        v = 5 + VP8GetBit(br, 159);
+    if (!VP8GetBit(br, p[6], "coeffs")) {
+      if (!VP8GetBit(br, p[7], "coeffs")) {
+        v = 5 + VP8GetBit(br, 159, "coeffs");
       } else {
-        v = 7 + 2 * VP8GetBit(br, 165);
-        v += VP8GetBit(br, 145);
+        v = 7 + 2 * VP8GetBit(br, 165, "coeffs");
+        v += VP8GetBit(br, 145, "coeffs");
       }
     } else {
       const uint8_t* tab;
-      const int bit1 = VP8GetBit(br, p[8]);
-      const int bit0 = VP8GetBit(br, p[9 + bit1]);
+      const int bit1 = VP8GetBit(br, p[8], "coeffs");
+      const int bit0 = VP8GetBit(br, p[9 + bit1], "coeffs");
       const int cat = 2 * bit1 + bit0;
       v = 0;
       for (tab = kCat3456[cat]; *tab; ++tab) {
-        v += v + VP8GetBit(br, *tab);
+        v += v + VP8GetBit(br, *tab, "coeffs");
       }
       v += 3 + (8 << cat);
     }
@@ -438,24 +441,24 @@ static int GetCoeffsFast(VP8BitReader* const br,
                          int ctx, const quant_t dq, int n, int16_t* out) {
   const uint8_t* p = prob[n]->probas_[ctx];
   for (; n < 16; ++n) {
-    if (!VP8GetBit(br, p[0])) {
+    if (!VP8GetBit(br, p[0], "coeffs")) {
       return n;  // previous coeff was last non-zero coeff
     }
-    while (!VP8GetBit(br, p[1])) {       // sequence of zero coeffs
+    while (!VP8GetBit(br, p[1], "coeffs")) {       // sequence of zero coeffs
       p = prob[++n]->probas_[0];
       if (n == 16) return 16;
     }
     {        // non zero coeff
       const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
       int v;
-      if (!VP8GetBit(br, p[2])) {
+      if (!VP8GetBit(br, p[2], "coeffs")) {
         v = 1;
         p = p_ctx[1];
       } else {
         v = GetLargeValue(br, p);
         p = p_ctx[2];
       }
-      out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
+      out[kZigzag[n]] = VP8GetSigned(br, v, "coeffs") * dq[n > 0];
     }
   }
   return 16;
@@ -468,36 +471,34 @@ static int GetCoeffsAlt(VP8BitReader* const br,
                         int ctx, const quant_t dq, int n, int16_t* out) {
   const uint8_t* p = prob[n]->probas_[ctx];
   for (; n < 16; ++n) {
-    if (!VP8GetBitAlt(br, p[0])) {
+    if (!VP8GetBitAlt(br, p[0], "coeffs")) {
       return n;  // previous coeff was last non-zero coeff
     }
-    while (!VP8GetBitAlt(br, p[1])) {       // sequence of zero coeffs
+    while (!VP8GetBitAlt(br, p[1], "coeffs")) {       // sequence of zero coeffs
       p = prob[++n]->probas_[0];
       if (n == 16) return 16;
     }
     {        // non zero coeff
       const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
       int v;
-      if (!VP8GetBitAlt(br, p[2])) {
+      if (!VP8GetBitAlt(br, p[2], "coeffs")) {
         v = 1;
         p = p_ctx[1];
       } else {
         v = GetLargeValue(br, p);
         p = p_ctx[2];
       }
-      out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
+      out[kZigzag[n]] = VP8GetSigned(br, v, "coeffs") * dq[n > 0];
     }
   }
   return 16;
 }
 
-static WEBP_TSAN_IGNORE_FUNCTION void InitGetCoeffs(void) {
-  if (GetCoeffs == NULL) {
-    if (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kSlowSSSE3)) {
-      GetCoeffs = GetCoeffsAlt;
-    } else {
-      GetCoeffs = GetCoeffsFast;
-    }
+WEBP_DSP_INIT_FUNC(InitGetCoeffs) {
+  if (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kSlowSSSE3)) {
+    GetCoeffs = GetCoeffsAlt;
+  } else {
+    GetCoeffs = GetCoeffsFast;
   }
 }
 

src/dec/vp8i_dec.h

@@ -31,8 +31,8 @@ extern "C" {
 
 // version numbers
 #define DEC_MAJ_VERSION 1
-#define DEC_MIN_VERSION 0
-#define DEC_REV_VERSION 1
+#define DEC_MIN_VERSION 2
+#define DEC_REV_VERSION 0
 
 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
 // Constraints are: We need to store one 16x16 block of luma samples (y),

src/dec/vp8l_dec.c

@@ -253,11 +253,11 @@ static int ReadHuffmanCodeLengths(
   int symbol;
   int max_symbol;
   int prev_code_len = DEFAULT_CODE_LENGTH;
-  HuffmanCode table[1 << LENGTHS_TABLE_BITS];
+  HuffmanTables tables;
 
-  if (!VP8LBuildHuffmanTable(table, LENGTHS_TABLE_BITS,
-                             code_length_code_lengths,
-                             NUM_CODE_LENGTH_CODES)) {
+  if (!VP8LHuffmanTablesAllocate(1 << LENGTHS_TABLE_BITS, &tables) ||
+      !VP8LBuildHuffmanTable(&tables, LENGTHS_TABLE_BITS,
+                             code_length_code_lengths, NUM_CODE_LENGTH_CODES)) {
     goto End;
   }
 
@@ -277,7 +277,7 @@ static int ReadHuffmanCodeLengths(
     int code_len;
     if (max_symbol-- == 0) break;
     VP8LFillBitWindow(br);
-    p = &table[VP8LPrefetchBits(br) & LENGTHS_TABLE_MASK];
+    p = &tables.curr_segment->start[VP8LPrefetchBits(br) & LENGTHS_TABLE_MASK];
     VP8LSetBitPos(br, br->bit_pos_ + p->bits);
     code_len = p->value;
     if (code_len < kCodeLengthLiterals) {
@@ -300,6 +300,7 @@ static int ReadHuffmanCodeLengths(
   ok = 1;
 
  End:
+  VP8LHuffmanTablesDeallocate(&tables);
   if (!ok) dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
   return ok;
 }
@@ -307,7 +308,8 @@ static int ReadHuffmanCodeLengths(
 // 'code_lengths' is pre-allocated temporary buffer, used for creating Huffman
 // tree.
 static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec,
-                           int* const code_lengths, HuffmanCode* const table) {
+                           int* const code_lengths,
+                           HuffmanTables* const table) {
   int ok = 0;
   int size = 0;
   VP8LBitReader* const br = &dec->br_;
@@ -362,12 +364,7 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
   VP8LMetadata* const hdr = &dec->hdr_;
   uint32_t* huffman_image = NULL;
   HTreeGroup* htree_groups = NULL;
-  // When reading htrees, some might be unused, as the format allows it.
-  // We will still read them but put them in this htree_group_bogus.
-  HTreeGroup htree_group_bogus;
-  HuffmanCode* huffman_tables = NULL;
-  HuffmanCode* huffman_tables_bogus = NULL;
-  HuffmanCode* next = NULL;
+  HuffmanTables* huffman_tables = &hdr->huffman_tables_;
   int num_htree_groups = 1;
   int num_htree_groups_max = 1;
   int max_alphabet_size = 0;
@@ -376,6 +373,10 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
   int* mapping = NULL;
   int ok = 0;
 
+  // Check the table has been 0 initialized (through InitMetadata).
+  assert(huffman_tables->root.start == NULL);
+  assert(huffman_tables->curr_segment == NULL);
+
   if (allow_recursion && VP8LReadBits(br, 1)) {
     // use meta Huffman codes.
     const int huffman_precision = VP8LReadBits(br, 3) + 2;
@@ -418,12 +419,6 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
         if (*mapped_group == -1) *mapped_group = num_htree_groups++;
         huffman_image[i] = *mapped_group;
       }
-      huffman_tables_bogus = (HuffmanCode*)WebPSafeMalloc(
-          table_size, sizeof(*huffman_tables_bogus));
-      if (huffman_tables_bogus == NULL) {
-        dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
-        goto Error;
-      }
     } else {
       num_htree_groups = num_htree_groups_max;
     }
@@ -444,72 +439,80 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
 
   code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size,
                                       sizeof(*code_lengths));
-  huffman_tables = (HuffmanCode*)WebPSafeMalloc(num_htree_groups * table_size,
-                                                sizeof(*huffman_tables));
   htree_groups = VP8LHtreeGroupsNew(num_htree_groups);
 
-  if (htree_groups == NULL || code_lengths == NULL || huffman_tables == NULL) {
+  if (htree_groups == NULL || code_lengths == NULL ||
+      !VP8LHuffmanTablesAllocate(num_htree_groups * table_size,
+                                 huffman_tables)) {
     dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
     goto Error;
   }
 
-  next = huffman_tables;
   for (i = 0; i < num_htree_groups_max; ++i) {
-    // If the index "i" is unused in the Huffman image, read the coefficients
-    // but store them to a bogus htree_group.
-    const int is_bogus = (mapping != NULL && mapping[i] == -1);
-    HTreeGroup* const htree_group =
-        is_bogus ? &htree_group_bogus :
-        &htree_groups[(mapping == NULL) ? i : mapping[i]];
-    HuffmanCode** const htrees = htree_group->htrees;
-    HuffmanCode* huffman_tables_i = is_bogus ? huffman_tables_bogus : next;
-    int size;
-    int total_size = 0;
-    int is_trivial_literal = 1;
-    int max_bits = 0;
-    for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
-      int alphabet_size = kAlphabetSize[j];
-      htrees[j] = huffman_tables_i;
-      if (j == 0 && color_cache_bits > 0) {
-        alphabet_size += 1 << color_cache_bits;
-      }
-      size =
-          ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_tables_i);
-      if (size == 0) {
-        goto Error;
-      }
-      if (is_trivial_literal && kLiteralMap[j] == 1) {
-        is_trivial_literal = (huffman_tables_i->bits == 0);
-      }
-      total_size += huffman_tables_i->bits;
-      huffman_tables_i += size;
-      if (j <= ALPHA) {
-        int local_max_bits = code_lengths[0];
-        int k;
-        for (k = 1; k < alphabet_size; ++k) {
-          if (code_lengths[k] > local_max_bits) {
-            local_max_bits = code_lengths[k];
-          }
+    // If the index "i" is unused in the Huffman image, just make sure the
+    // coefficients are valid but do not store them.
+    if (mapping != NULL && mapping[i] == -1) {
+      for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
+        int alphabet_size = kAlphabetSize[j];
+        if (j == 0 && color_cache_bits > 0) {
+          alphabet_size += (1 << color_cache_bits);
+        }
+        // Passing in NULL so that nothing gets filled.
+        if (!ReadHuffmanCode(alphabet_size, dec, code_lengths, NULL)) {
+          goto Error;
         }
-        max_bits += local_max_bits;
       }
-    }
-    if (!is_bogus) next = huffman_tables_i;
-    htree_group->is_trivial_literal = is_trivial_literal;
-    htree_group->is_trivial_code = 0;
-    if (is_trivial_literal) {
-      const int red = htrees[RED][0].value;
-      const int blue = htrees[BLUE][0].value;
-      const int alpha = htrees[ALPHA][0].value;
-      htree_group->literal_arb = ((uint32_t)alpha << 24) | (red << 16) | blue;
-      if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) {
-        htree_group->is_trivial_code = 1;
-        htree_group->literal_arb |= htrees[GREEN][0].value << 8;
+    } else {
+      HTreeGroup* const htree_group =
+          &htree_groups[(mapping == NULL) ? i : mapping[i]];
+      HuffmanCode** const htrees = htree_group->htrees;
+      int size;
+      int total_size = 0;
+      int is_trivial_literal = 1;
+      int max_bits = 0;
+      for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
+        int alphabet_size = kAlphabetSize[j];
+        if (j == 0 && color_cache_bits > 0) {
+          alphabet_size += (1 << color_cache_bits);
+        }
+        size =
+            ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_tables);
+        htrees[j] = huffman_tables->curr_segment->curr_table;
+        if (size == 0) {
+          goto Error;
+        }
+        if (is_trivial_literal && kLiteralMap[j] == 1) {
+          is_trivial_literal = (htrees[j]->bits == 0);
+        }
+        total_size += htrees[j]->bits;
+        huffman_tables->curr_segment->curr_table += size;
+        if (j <= ALPHA) {
+          int local_max_bits = code_lengths[0];
+          int k;
+          for (k = 1; k < alphabet_size; ++k) {
+            if (code_lengths[k] > local_max_bits) {
+              local_max_bits = code_lengths[k];
+            }
+          }
+          max_bits += local_max_bits;
+        }
       }
+      htree_group->is_trivial_literal = is_trivial_literal;
+      htree_group->is_trivial_code = 0;
+      if (is_trivial_literal) {
+        const int red = htrees[RED][0].value;
+        const int blue = htrees[BLUE][0].value;
+        const int alpha = htrees[ALPHA][0].value;
+        htree_group->literal_arb = ((uint32_t)alpha << 24) | (red << 16) | blue;
+        if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) {
+          htree_group->is_trivial_code = 1;
+          htree_group->literal_arb |= htrees[GREEN][0].value << 8;
+        }
+      }
+      htree_group->use_packed_table =
+          !htree_group->is_trivial_code && (max_bits < HUFFMAN_PACKED_BITS);
+      if (htree_group->use_packed_table) BuildPackedTable(htree_group);
     }
-    htree_group->use_packed_table =
-        !htree_group->is_trivial_code && (max_bits < HUFFMAN_PACKED_BITS);
-    if (htree_group->use_packed_table) BuildPackedTable(htree_group);
   }
   ok = 1;
 
@@ -517,15 +520,13 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
   hdr->huffman_image_ = huffman_image;
   hdr->num_htree_groups_ = num_htree_groups;
   hdr->htree_groups_ = htree_groups;
-  hdr->huffman_tables_ = huffman_tables;
 
  Error:
   WebPSafeFree(code_lengths);
-  WebPSafeFree(huffman_tables_bogus);
   WebPSafeFree(mapping);
   if (!ok) {
     WebPSafeFree(huffman_image);
-    WebPSafeFree(huffman_tables);
+    VP8LHuffmanTablesDeallocate(huffman_tables);
     VP8LHtreeGroupsFree(htree_groups);
   }
   return ok;
@@ -757,11 +758,11 @@ static WEBP_INLINE HTreeGroup* GetHtreeGroupForPos(VP8LMetadata* const hdr,
 
 typedef void (*ProcessRowsFunc)(VP8LDecoder* const dec, int row);
 
-static void ApplyInverseTransforms(VP8LDecoder* const dec, int num_rows,
+static void ApplyInverseTransforms(VP8LDecoder* const dec,
+                                   int start_row, int num_rows,
                                    const uint32_t* const rows) {
   int n = dec->next_transform_;
   const int cache_pixs = dec->width_ * num_rows;
-  const int start_row = dec->last_row_;
   const int end_row = start_row + num_rows;
   const uint32_t* rows_in = rows;
   uint32_t* const rows_out = dec->argb_cache_;
@@ -792,8 +793,7 @@ static void ProcessRows(VP8LDecoder* const dec, int row) {
     VP8Io* const io = dec->io_;
     uint8_t* rows_data = (uint8_t*)dec->argb_cache_;
     const int in_stride = io->width * sizeof(uint32_t);  // in unit of RGBA
-
-    ApplyInverseTransforms(dec, num_rows, rows);
+    ApplyInverseTransforms(dec, dec->last_row_, num_rows, rows);
     if (!SetCropWindow(io, dec->last_row_, row, &rows_data, in_stride)) {
       // Nothing to output (this time).
     } else {
@@ -951,7 +951,6 @@ static WEBP_INLINE void CopyBlock8b(uint8_t* const dst, int dist, int length) {
         break;
       default:
         goto Copy;
-        break;
     }
     CopySmallPattern8b(src, dst, length, pattern);
     return;
@@ -1196,6 +1195,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
       VP8LFillBitWindow(br);
       dist_code = GetCopyDistance(dist_symbol, br);
       dist = PlaneCodeToDistance(width, dist_code);
+
       if (VP8LIsEndOfStream(br)) break;
       if (src - data < (ptrdiff_t)dist || src_end - src < (ptrdiff_t)length) {
         goto Error;
@@ -1357,7 +1357,7 @@ static void ClearMetadata(VP8LMetadata* const hdr) {
   assert(hdr != NULL);
 
   WebPSafeFree(hdr->huffman_image_);
-  WebPSafeFree(hdr->huffman_tables_);
+  VP8LHuffmanTablesDeallocate(&hdr->huffman_tables_);
   VP8LHtreeGroupsFree(hdr->htree_groups_);
   VP8LColorCacheClear(&hdr->color_cache_);
   VP8LColorCacheClear(&hdr->saved_color_cache_);
@@ -1556,7 +1556,7 @@ static void ExtractAlphaRows(VP8LDecoder* const dec, int last_row) {
     const int cache_pixs = width * num_rows_to_process;
     uint8_t* const dst = output + width * cur_row;
     const uint32_t* const src = dec->argb_cache_;
-    ApplyInverseTransforms(dec, num_rows_to_process, in);
+    ApplyInverseTransforms(dec, cur_row, num_rows_to_process, in);
     WebPExtractGreen(src, dst, cache_pixs);
     AlphaApplyFilter(alph_dec,
                      cur_row, cur_row + num_rows_to_process, dst, width);
@@ -1673,7 +1673,7 @@ int VP8LDecodeImage(VP8LDecoder* const dec) {
   // Sanity checks.
   if (dec == NULL) return 0;
 
-  assert(dec->hdr_.huffman_tables_ != NULL);
+  assert(dec->hdr_.huffman_tables_.root.start != NULL);
   assert(dec->hdr_.htree_groups_ != NULL);
   assert(dec->hdr_.num_htree_groups_ > 0);
 

src/dec/vp8li_dec.h

@@ -37,7 +37,7 @@ struct VP8LTransform {
   int                    bits_;   // subsampling bits defining transform window.
   int                    xsize_;  // transform window X index.
   int                    ysize_;  // transform window Y index.
-  uint32_t              *data_;   // transform data.
+  uint32_t*              data_;   // transform data.
 };
 
 typedef struct {
@@ -48,23 +48,23 @@ typedef struct {
   int             huffman_mask_;
   int             huffman_subsample_bits_;
   int             huffman_xsize_;
-  uint32_t       *huffman_image_;
+  uint32_t*       huffman_image_;
   int             num_htree_groups_;
-  HTreeGroup     *htree_groups_;
-  HuffmanCode    *huffman_tables_;
+  HTreeGroup*     htree_groups_;
+  HuffmanTables   huffman_tables_;
 } VP8LMetadata;
 
 typedef struct VP8LDecoder VP8LDecoder;
 struct VP8LDecoder {
   VP8StatusCode    status_;
   VP8LDecodeState  state_;
-  VP8Io           *io_;
+  VP8Io*           io_;
 
-  const WebPDecBuffer *output_;    // shortcut to io->opaque->output
+  const WebPDecBuffer* output_;    // shortcut to io->opaque->output
 
-  uint32_t        *pixels_;        // Internal data: either uint8_t* for alpha
+  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.
+  uint32_t*        argb_cache_;    // Scratch buffer for temporary BGRA storage.
 
   VP8LBitReader    br_;
   int              incremental_;   // if true, incremental decoding is expected
@@ -86,8 +86,8 @@ struct VP8LDecoder {
   // or'd bitset storing the transforms types.
   uint32_t         transforms_seen_;
 
-  uint8_t         *rescaler_memory;  // Working memory for rescaling work.
-  WebPRescaler    *rescaler;         // Common rescaler for all channels.
+  uint8_t*         rescaler_memory;  // Working memory for rescaling work.
+  WebPRescaler*    rescaler;         // Common rescaler for all channels.
 };
 
 //------------------------------------------------------------------------------

src/demux/Makefile.am

@@ -13,6 +13,6 @@ noinst_HEADERS =
 noinst_HEADERS += ../webp/format_constants.h
 
 libwebpdemux_la_LIBADD = ../libwebp.la
-libwebpdemux_la_LDFLAGS = -no-undefined -version-info 2:5:0
+libwebpdemux_la_LDFLAGS = -no-undefined -version-info 2:7:0
 libwebpdemuxincludedir = $(includedir)/webp
 pkgconfig_DATA = libwebpdemux.pc

src/demux/anim_decode.c

@@ -346,12 +346,15 @@ int WebPAnimDecoderGetNext(WebPAnimDecoder* dec,
   {
     const uint8_t* in = iter.fragment.bytes;
     const size_t in_size = iter.fragment.size;
-    const size_t out_offset =
-        (iter.y_offset * width + iter.x_offset) * NUM_CHANNELS;
+    const uint32_t stride = width * NUM_CHANNELS;  // at most 25 + 2 bits
+    const uint64_t out_offset = (uint64_t)iter.y_offset * stride +
+                                (uint64_t)iter.x_offset * NUM_CHANNELS;  // 53b
+    const uint64_t size = (uint64_t)iter.height * stride;  // at most 25 + 27b
     WebPDecoderConfig* const config = &dec->config_;
     WebPRGBABuffer* const buf = &config->output.u.RGBA;
-    buf->stride = NUM_CHANNELS * width;
-    buf->size = buf->stride * iter.height;
+    if ((size_t)size != size) goto Error;
+    buf->stride = (int)stride;
+    buf->size = (size_t)size;
     buf->rgba = dec->curr_frame_ + out_offset;
 
     if (WebPDecode(in, in_size, config) != VP8_STATUS_OK) {

src/demux/demux.c

@@ -24,8 +24,8 @@
 #include "src/webp/format_constants.h"
 
 #define DMUX_MAJ_VERSION 1
-#define DMUX_MIN_VERSION 0
-#define DMUX_REV_VERSION 1
+#define DMUX_MIN_VERSION 2
+#define DMUX_REV_VERSION 0
 
 typedef struct {
   size_t start_;        // start location of the data
@@ -312,6 +312,7 @@ static ParseStatus ParseAnimationFrame(
   int bits;
   MemBuffer* const mem = &dmux->mem_;
   Frame* frame;
+  size_t start_offset;
   ParseStatus status =
       NewFrame(mem, ANMF_CHUNK_SIZE, frame_chunk_size, &frame);
   if (status != PARSE_OK) return status;
@@ -332,7 +333,11 @@ static ParseStatus ParseAnimationFrame(
 
   // Store a frame only if the animation flag is set there is some data for
   // this frame is available.
+  start_offset = mem->start_;
   status = StoreFrame(dmux->num_frames_ + 1, anmf_payload_size, mem, frame);
+  if (status != PARSE_ERROR && mem->start_ - start_offset > anmf_payload_size) {
+    status = PARSE_ERROR;
+  }
   if (status != PARSE_ERROR && is_animation && frame->frame_num_ > 0) {
     added_frame = AddFrame(dmux, frame);
     if (added_frame) {

src/demux/libwebpdemux.rc

@@ -6,8 +6,8 @@
 LANGUAGE LANG_ENGLISH, SUBLANG_ENGLISH_US
 
 VS_VERSION_INFO VERSIONINFO
- FILEVERSION 1,0,0,1
- PRODUCTVERSION 1,0,0,1
+ FILEVERSION 1,0,2,0
+ PRODUCTVERSION 1,0,2,0
  FILEFLAGSMASK 0x3fL
 #ifdef _DEBUG
  FILEFLAGS 0x1L
@@ -24,12 +24,12 @@ BEGIN
         BEGIN
             VALUE "CompanyName", "Google, Inc."
             VALUE "FileDescription", "libwebpdemux DLL"
-            VALUE "FileVersion", "1.0.1"
+            VALUE "FileVersion", "1.2.0"
             VALUE "InternalName", "libwebpdemux.dll"
-            VALUE "LegalCopyright", "Copyright (C) 2018"
+            VALUE "LegalCopyright", "Copyright (C) 2019"
             VALUE "OriginalFilename", "libwebpdemux.dll"
             VALUE "ProductName", "WebP Image Demuxer"
-            VALUE "ProductVersion", "1.0.1"
+            VALUE "ProductVersion", "1.2.0"
         END
     END
     BLOCK "VarFileInfo"

src/dsp/Makefile.am

@@ -40,6 +40,7 @@ ENC_SOURCES =
 ENC_SOURCES += cost.c
 ENC_SOURCES += enc.c
 ENC_SOURCES += lossless_enc.c
+ENC_SOURCES += quant.h
 ENC_SOURCES += ssim.c
 
 libwebpdspdecode_sse41_la_SOURCES =
@@ -121,6 +122,7 @@ libwebpdsp_sse41_la_CFLAGS = $(AM_CFLAGS) $(SSE41_FLAGS)
 libwebpdsp_sse41_la_LIBADD = libwebpdspdecode_sse41.la
 
 libwebpdsp_neon_la_SOURCES =
+libwebpdsp_neon_la_SOURCES += cost_neon.c
 libwebpdsp_neon_la_SOURCES += enc_neon.c
 libwebpdsp_neon_la_SOURCES += lossless_enc_neon.c
 libwebpdsp_neon_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)

src/dsp/alpha_processing.c

@@ -359,6 +359,11 @@ static int HasAlpha32b_C(const uint8_t* src, int length) {
   return 0;
 }
 
+static void AlphaReplace_C(uint32_t* src, int length, uint32_t color) {
+  int x;
+  for (x = 0; x < length; ++x) if ((src[x] >> 24) == 0) src[x] = color;
+}
+
 //------------------------------------------------------------------------------
 // Simple channel manipulations.
 
@@ -400,6 +405,7 @@ void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
 
 int (*WebPHasAlpha8b)(const uint8_t* src, int length);
 int (*WebPHasAlpha32b)(const uint8_t* src, int length);
+void (*WebPAlphaReplace)(uint32_t* src, int length, uint32_t color);
 
 //------------------------------------------------------------------------------
 // Init function
@@ -428,6 +434,7 @@ WEBP_DSP_INIT_FUNC(WebPInitAlphaProcessing) {
 
   WebPHasAlpha8b = HasAlpha8b_C;
   WebPHasAlpha32b = HasAlpha32b_C;
+  WebPAlphaReplace = AlphaReplace_C;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
@@ -469,4 +476,5 @@ WEBP_DSP_INIT_FUNC(WebPInitAlphaProcessing) {
   assert(WebPPackRGB != NULL);
   assert(WebPHasAlpha8b != NULL);
   assert(WebPHasAlpha32b != NULL);
+  assert(WebPAlphaReplace != NULL);
 }

src/dsp/alpha_processing_sse2.c

@@ -214,7 +214,7 @@ static void ApplyAlphaMultiply_SSE2(uint8_t* rgba, int alpha_first,
 // Alpha detection
 
 static int HasAlpha8b_SSE2(const uint8_t* src, int length) {
-  const __m128i all_0xff = _mm_set1_epi8(0xff);
+  const __m128i all_0xff = _mm_set1_epi8((char)0xff);
   int i = 0;
   for (; i + 16 <= length; i += 16) {
     const __m128i v = _mm_loadu_si128((const __m128i*)(src + i));
@@ -228,7 +228,7 @@ static int HasAlpha8b_SSE2(const uint8_t* src, int length) {
 
 static int HasAlpha32b_SSE2(const uint8_t* src, int length) {
   const __m128i alpha_mask = _mm_set1_epi32(0xff);
-  const __m128i all_0xff = _mm_set1_epi8(0xff);
+  const __m128i all_0xff = _mm_set1_epi8((char)0xff);
   int i = 0;
   // We don't know if we can access the last 3 bytes after the last alpha
   // value 'src[4 * length - 4]' (because we don't know if alpha is the first
@@ -265,6 +265,27 @@ static int HasAlpha32b_SSE2(const uint8_t* src, int length) {
   return 0;
 }
 
+static void AlphaReplace_SSE2(uint32_t* src, int length, uint32_t color) {
+  const __m128i m_color = _mm_set1_epi32(color);
+  const __m128i zero = _mm_setzero_si128();
+  int i = 0;
+  for (; i + 8 <= length; i += 8) {
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)(src + i + 0));
+    const __m128i a1 = _mm_loadu_si128((const __m128i*)(src + i + 4));
+    const __m128i b0 = _mm_srai_epi32(a0, 24);
+    const __m128i b1 = _mm_srai_epi32(a1, 24);
+    const __m128i c0 = _mm_cmpeq_epi32(b0, zero);
+    const __m128i c1 = _mm_cmpeq_epi32(b1, zero);
+    const __m128i d0 = _mm_and_si128(c0, m_color);
+    const __m128i d1 = _mm_and_si128(c1, m_color);
+    const __m128i e0 = _mm_andnot_si128(c0, a0);
+    const __m128i e1 = _mm_andnot_si128(c1, a1);
+    _mm_storeu_si128((__m128i*)(src + i + 0), _mm_or_si128(d0, e0));
+    _mm_storeu_si128((__m128i*)(src + i + 4), _mm_or_si128(d1, e1));
+  }
+  for (; i < length; ++i) if ((src[i] >> 24) == 0) src[i] = color;
+}
+
 // -----------------------------------------------------------------------------
 // Apply alpha value to rows
 
@@ -334,6 +355,7 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessingSSE2(void) {
 
   WebPHasAlpha8b = HasAlpha8b_SSE2;
   WebPHasAlpha32b = HasAlpha32b_SSE2;
+  WebPAlphaReplace = AlphaReplace_SSE2;
 }
 
 #else  // !WEBP_USE_SSE2

src/dsp/cost.c

@@ -377,6 +377,7 @@ VP8SetResidualCoeffsFunc VP8SetResidualCoeffs;
 extern void VP8EncDspCostInitMIPS32(void);
 extern void VP8EncDspCostInitMIPSdspR2(void);
 extern void VP8EncDspCostInitSSE2(void);
+extern void VP8EncDspCostInitNEON(void);
 
 WEBP_DSP_INIT_FUNC(VP8EncDspCostInit) {
   VP8GetResidualCost = GetResidualCost_C;
@@ -398,6 +399,11 @@ WEBP_DSP_INIT_FUNC(VP8EncDspCostInit) {
     if (VP8GetCPUInfo(kSSE2)) {
       VP8EncDspCostInitSSE2();
     }
+#endif
+#if defined(WEBP_USE_NEON)
+    if (VP8GetCPUInfo(kNEON)) {
+      VP8EncDspCostInitNEON();
+    }
 #endif
   }
 }

src/dsp/cost_neon.c

@@ -0,0 +1,122 @@
+// Copyright 2018 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.
+// -----------------------------------------------------------------------------
+//
+// ARM NEON version of cost functions
+
+#include "src/dsp/dsp.h"
+
+#if defined(WEBP_USE_NEON)
+
+#include "src/dsp/neon.h"
+#include "src/enc/cost_enc.h"
+
+static const uint8_t position[16] = { 1, 2,  3,  4,  5,  6,  7,  8,
+                                      9, 10, 11, 12, 13, 14, 15, 16 };
+
+static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
+                                   VP8Residual* const res) {
+  const int16x8_t minus_one = vdupq_n_s16(-1);
+  const int16x8_t coeffs_0 = vld1q_s16(coeffs);
+  const int16x8_t coeffs_1 = vld1q_s16(coeffs + 8);
+  const uint16x8_t eob_0 = vtstq_s16(coeffs_0, minus_one);
+  const uint16x8_t eob_1 = vtstq_s16(coeffs_1, minus_one);
+  const uint8x16_t eob = vcombine_u8(vqmovn_u16(eob_0), vqmovn_u16(eob_1));
+  const uint8x16_t masked = vandq_u8(eob, vld1q_u8(position));
+
+#ifdef __aarch64__
+  res->last = vmaxvq_u8(masked) - 1;
+#else
+  const uint8x8_t eob_8x8 = vmax_u8(vget_low_u8(masked), vget_high_u8(masked));
+  const uint16x8_t eob_16x8 = vmovl_u8(eob_8x8);
+  const uint16x4_t eob_16x4 =
+      vmax_u16(vget_low_u16(eob_16x8), vget_high_u16(eob_16x8));
+  const uint32x4_t eob_32x4 = vmovl_u16(eob_16x4);
+  uint32x2_t eob_32x2 =
+      vmax_u32(vget_low_u32(eob_32x4), vget_high_u32(eob_32x4));
+  eob_32x2 = vpmax_u32(eob_32x2, eob_32x2);
+
+  vst1_lane_s32(&res->last, vreinterpret_s32_u32(eob_32x2), 0);
+  --res->last;
+#endif  // __aarch64__
+
+  res->coeffs = coeffs;
+}
+
+static int GetResidualCost_NEON(int ctx0, const VP8Residual* const res) {
+  uint8_t levels[16], ctxs[16];
+  uint16_t abs_levels[16];
+  int n = res->first;
+  // should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1
+  const int p0 = res->prob[n][ctx0][0];
+  CostArrayPtr const costs = res->costs;
+  const uint16_t* t = costs[n][ctx0];
+  // bit_cost(1, p0) is already incorporated in t[] tables, but only if ctx != 0
+  // (as required by the syntax). For ctx0 == 0, we need to add it here or it'll
+  // be missing during the loop.
+  int cost = (ctx0 == 0) ? VP8BitCost(1, p0) : 0;
+
+  if (res->last < 0) {
+    return VP8BitCost(0, p0);
+  }
+
+  {   // precompute clamped levels and contexts, packed to 8b.
+    const uint8x16_t kCst2 = vdupq_n_u8(2);
+    const uint8x16_t kCst67 = vdupq_n_u8(MAX_VARIABLE_LEVEL);
+    const int16x8_t c0 = vld1q_s16(res->coeffs);
+    const int16x8_t c1 = vld1q_s16(res->coeffs + 8);
+    const uint16x8_t E0 = vreinterpretq_u16_s16(vabsq_s16(c0));
+    const uint16x8_t E1 = vreinterpretq_u16_s16(vabsq_s16(c1));
+    const uint8x16_t F = vcombine_u8(vqmovn_u16(E0), vqmovn_u16(E1));
+    const uint8x16_t G = vminq_u8(F, kCst2);   // context = 0,1,2
+    const uint8x16_t H = vminq_u8(F, kCst67);  // clamp_level in [0..67]
+
+    vst1q_u8(ctxs, G);
+    vst1q_u8(levels, H);
+
+    vst1q_u16(abs_levels, E0);
+    vst1q_u16(abs_levels + 8, E1);
+  }
+  for (; n < res->last; ++n) {
+    const int ctx = ctxs[n];
+    const int level = levels[n];
+    const int flevel = abs_levels[n];   // full level
+    cost += VP8LevelFixedCosts[flevel] + t[level];  // simplified VP8LevelCost()
+    t = costs[n + 1][ctx];
+  }
+  // Last coefficient is always non-zero
+  {
+    const int level = levels[n];
+    const int flevel = abs_levels[n];
+    assert(flevel != 0);
+    cost += VP8LevelFixedCosts[flevel] + t[level];
+    if (n < 15) {
+      const int b = VP8EncBands[n + 1];
+      const int ctx = ctxs[n];
+      const int last_p0 = res->prob[b][ctx][0];
+      cost += VP8BitCost(0, last_p0);
+    }
+  }
+  return cost;
+}
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void VP8EncDspCostInitNEON(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspCostInitNEON(void) {
+  VP8SetResidualCoeffs = SetResidualCoeffs_NEON;
+  VP8GetResidualCost = GetResidualCost_NEON;
+}
+
+#else  // !WEBP_USE_NEON
+
+WEBP_DSP_INIT_STUB(VP8EncDspCostInitNEON)
+
+#endif  // WEBP_USE_NEON

src/dsp/cpu.c

@@ -55,12 +55,18 @@ 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), "c"(0));
 }
-#elif (defined(_M_X64) || defined(_M_IX86)) && \
-      defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 150030729  // >= VS2008 SP1
+#elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
+
+#if defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 150030729  // >= VS2008 SP1
 #include <intrin.h>
 #define GetCPUInfo(info, type) __cpuidex(info, type, 0)  // set ecx=0
-#elif defined(WEBP_MSC_SSE2)
+#define WEBP_HAVE_MSC_CPUID
+#elif _MSC_VER > 1310
+#include <intrin.h>
 #define GetCPUInfo __cpuid
+#define WEBP_HAVE_MSC_CPUID
+#endif
+
 #endif
 
 // NaCl has no support for xgetbv or the raw opcode.
@@ -94,7 +100,7 @@ static WEBP_INLINE uint64_t xgetbv(void) {
 #define xgetbv() 0U  // no AVX for older x64 or unrecognized toolchains.
 #endif
 
-#if defined(__i386__) || defined(__x86_64__) || defined(WEBP_MSC_SSE2)
+#if defined(__i386__) || defined(__x86_64__) || defined(WEBP_HAVE_MSC_CPUID)
 
 // helper function for run-time detection of slow SSSE3 platforms
 static int CheckSlowModel(int info) {
@@ -173,12 +179,36 @@ static int AndroidCPUInfo(CPUFeature feature) {
   const AndroidCpuFamily cpu_family = android_getCpuFamily();
   const uint64_t cpu_features = android_getCpuFeatures();
   if (feature == kNEON) {
-    return (cpu_family == ANDROID_CPU_FAMILY_ARM &&
-            0 != (cpu_features & ANDROID_CPU_ARM_FEATURE_NEON));
+    return cpu_family == ANDROID_CPU_FAMILY_ARM &&
+           (cpu_features & ANDROID_CPU_ARM_FEATURE_NEON) != 0;
   }
   return 0;
 }
 VP8CPUInfo VP8GetCPUInfo = AndroidCPUInfo;
+#elif defined(EMSCRIPTEN) // also needs to be before generic NEON test
+// Use compile flags as an indicator of SIMD support instead of a runtime check.
+static int wasmCPUInfo(CPUFeature feature) {
+  switch (feature) {
+#ifdef WEBP_USE_SSE2
+    case kSSE2:
+      return 1;
+#endif
+#ifdef WEBP_USE_SSE41
+    case kSSE3:
+    case kSlowSSSE3:
+    case kSSE4_1:
+      return 1;
+#endif
+#ifdef WEBP_USE_NEON
+    case kNEON:
+      return 1;
+#endif
+    default:
+      break;
+  }
+  return 0;
+}
+VP8CPUInfo VP8GetCPUInfo = wasmCPUInfo;
 #elif defined(WEBP_USE_NEON)
 // define a dummy function to enable turning off NEON at runtime by setting
 // VP8DecGetCPUInfo = NULL

src/dsp/dec_neon.c

@@ -1283,12 +1283,12 @@ static void DC4_NEON(uint8_t* dst) {    // DC
   const uint8x8_t A = vld1_u8(dst - BPS);  // top row
   const uint16x4_t p0 = vpaddl_u8(A);  // cascading summation of the top
   const uint16x4_t p1 = vpadd_u16(p0, p0);
-  const uint16x8_t L0 = vmovl_u8(vld1_u8(dst + 0 * BPS - 1));
-  const uint16x8_t L1 = vmovl_u8(vld1_u8(dst + 1 * BPS - 1));
-  const uint16x8_t L2 = vmovl_u8(vld1_u8(dst + 2 * BPS - 1));
-  const uint16x8_t L3 = vmovl_u8(vld1_u8(dst + 3 * BPS - 1));
-  const uint16x8_t s0 = vaddq_u16(L0, L1);
-  const uint16x8_t s1 = vaddq_u16(L2, L3);
+  const uint8x8_t L0 = vld1_u8(dst + 0 * BPS - 1);
+  const uint8x8_t L1 = vld1_u8(dst + 1 * BPS - 1);
+  const uint8x8_t L2 = vld1_u8(dst + 2 * BPS - 1);
+  const uint8x8_t L3 = vld1_u8(dst + 3 * BPS - 1);
+  const uint16x8_t s0 = vaddl_u8(L0, L1);
+  const uint16x8_t s1 = vaddl_u8(L2, L3);
   const uint16x8_t s01 = vaddq_u16(s0, s1);
   const uint16x8_t sum = vaddq_u16(s01, vcombine_u16(p1, p1));
   const uint8x8_t dc0 = vrshrn_n_u16(sum, 3);  // (sum + 4) >> 3
@@ -1361,7 +1361,8 @@ static void RD4_NEON(uint8_t* dst) {   // Down-right
   const uint32_t J = dst[-1 + 1 * BPS];
   const uint32_t K = dst[-1 + 2 * BPS];
   const uint32_t L = dst[-1 + 3 * BPS];
-  const uint64x1_t LKJI____ = vcreate_u64(L | (K << 8) | (J << 16) | (I << 24));
+  const uint64x1_t LKJI____ =
+      vcreate_u64((uint64_t)L | (K << 8) | (J << 16) | (I << 24));
   const uint64x1_t LKJIXABC = vorr_u64(LKJI____, ____XABC);
   const uint8x8_t KJIXABC_ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 8));
   const uint8x8_t JIXABC__ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 16));
@@ -1427,25 +1428,30 @@ static WEBP_INLINE void DC8_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x8_t A = vld1_u8(dst - BPS);  // top row
+#if defined(__aarch64__)
+    const uint16_t p2 = vaddlv_u8(A);
+    sum_top = vdupq_n_u16(p2);
+#else
     const uint16x4_t p0 = vpaddl_u8(A);  // cascading summation of the top
     const uint16x4_t p1 = vpadd_u16(p0, p0);
     const uint16x4_t p2 = vpadd_u16(p1, p1);
     sum_top = vcombine_u16(p2, p2);
+#endif
   }
 
   if (do_left) {
-    const uint16x8_t L0 = vmovl_u8(vld1_u8(dst + 0 * BPS - 1));
-    const uint16x8_t L1 = vmovl_u8(vld1_u8(dst + 1 * BPS - 1));
-    const uint16x8_t L2 = vmovl_u8(vld1_u8(dst + 2 * BPS - 1));
-    const uint16x8_t L3 = vmovl_u8(vld1_u8(dst + 3 * BPS - 1));
-    const uint16x8_t L4 = vmovl_u8(vld1_u8(dst + 4 * BPS - 1));
-    const uint16x8_t L5 = vmovl_u8(vld1_u8(dst + 5 * BPS - 1));
-    const uint16x8_t L6 = vmovl_u8(vld1_u8(dst + 6 * BPS - 1));
-    const uint16x8_t L7 = vmovl_u8(vld1_u8(dst + 7 * BPS - 1));
-    const uint16x8_t s0 = vaddq_u16(L0, L1);
-    const uint16x8_t s1 = vaddq_u16(L2, L3);
-    const uint16x8_t s2 = vaddq_u16(L4, L5);
-    const uint16x8_t s3 = vaddq_u16(L6, L7);
+    const uint8x8_t L0 = vld1_u8(dst + 0 * BPS - 1);
+    const uint8x8_t L1 = vld1_u8(dst + 1 * BPS - 1);
+    const uint8x8_t L2 = vld1_u8(dst + 2 * BPS - 1);
+    const uint8x8_t L3 = vld1_u8(dst + 3 * BPS - 1);
+    const uint8x8_t L4 = vld1_u8(dst + 4 * BPS - 1);
+    const uint8x8_t L5 = vld1_u8(dst + 5 * BPS - 1);
+    const uint8x8_t L6 = vld1_u8(dst + 6 * BPS - 1);
+    const uint8x8_t L7 = vld1_u8(dst + 7 * BPS - 1);
+    const uint16x8_t s0 = vaddl_u8(L0, L1);
+    const uint16x8_t s1 = vaddl_u8(L2, L3);
+    const uint16x8_t s2 = vaddl_u8(L4, L5);
+    const uint16x8_t s3 = vaddl_u8(L6, L7);
     const uint16x8_t s01 = vaddq_u16(s0, s1);
     const uint16x8_t s23 = vaddq_u16(s2, s3);
     sum_left = vaddq_u16(s01, s23);
@@ -1505,29 +1511,34 @@ static WEBP_INLINE void DC16_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x16_t A = vld1q_u8(dst - BPS);  // top row
+#if defined(__aarch64__)
+    const uint16_t p3 = vaddlvq_u8(A);
+    sum_top = vdupq_n_u16(p3);
+#else
     const uint16x8_t p0 = vpaddlq_u8(A);  // cascading summation of the top
     const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
     const uint16x4_t p2 = vpadd_u16(p1, p1);
     const uint16x4_t p3 = vpadd_u16(p2, p2);
     sum_top = vcombine_u16(p3, p3);
+#endif
   }
 
   if (do_left) {
     int i;
     sum_left = vdupq_n_u16(0);
     for (i = 0; i < 16; i += 8) {
-      const uint16x8_t L0 = vmovl_u8(vld1_u8(dst + (i + 0) * BPS - 1));
-      const uint16x8_t L1 = vmovl_u8(vld1_u8(dst + (i + 1) * BPS - 1));
-      const uint16x8_t L2 = vmovl_u8(vld1_u8(dst + (i + 2) * BPS - 1));
-      const uint16x8_t L3 = vmovl_u8(vld1_u8(dst + (i + 3) * BPS - 1));
-      const uint16x8_t L4 = vmovl_u8(vld1_u8(dst + (i + 4) * BPS - 1));
-      const uint16x8_t L5 = vmovl_u8(vld1_u8(dst + (i + 5) * BPS - 1));
-      const uint16x8_t L6 = vmovl_u8(vld1_u8(dst + (i + 6) * BPS - 1));
-      const uint16x8_t L7 = vmovl_u8(vld1_u8(dst + (i + 7) * BPS - 1));
-      const uint16x8_t s0 = vaddq_u16(L0, L1);
-      const uint16x8_t s1 = vaddq_u16(L2, L3);
-      const uint16x8_t s2 = vaddq_u16(L4, L5);
-      const uint16x8_t s3 = vaddq_u16(L6, L7);
+      const uint8x8_t L0 = vld1_u8(dst + (i + 0) * BPS - 1);
+      const uint8x8_t L1 = vld1_u8(dst + (i + 1) * BPS - 1);
+      const uint8x8_t L2 = vld1_u8(dst + (i + 2) * BPS - 1);
+      const uint8x8_t L3 = vld1_u8(dst + (i + 3) * BPS - 1);
+      const uint8x8_t L4 = vld1_u8(dst + (i + 4) * BPS - 1);
+      const uint8x8_t L5 = vld1_u8(dst + (i + 5) * BPS - 1);
+      const uint8x8_t L6 = vld1_u8(dst + (i + 6) * BPS - 1);
+      const uint8x8_t L7 = vld1_u8(dst + (i + 7) * BPS - 1);
+      const uint16x8_t s0 = vaddl_u8(L0, L1);
+      const uint16x8_t s1 = vaddl_u8(L2, L3);
+      const uint16x8_t s2 = vaddl_u8(L4, L5);
+      const uint16x8_t s3 = vaddl_u8(L6, L7);
       const uint16x8_t s01 = vaddq_u16(s0, s1);
       const uint16x8_t s23 = vaddq_u16(s2, s3);
       const uint16x8_t sum = vaddq_u16(s01, s23);

src/dsp/dec_sse2.c

@@ -326,7 +326,7 @@ static WEBP_INLINE void Update2Pixels_SSE2(__m128i* const pi, __m128i* const qi,
   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);
+  const __m128i sign_bit = _mm_set1_epi8((char)0x80);
   *pi = _mm_adds_epi8(*pi, delta);
   *qi = _mm_subs_epi8(*qi, delta);
   FLIP_SIGN_BIT2(*pi, *qi);
@@ -338,9 +338,9 @@ static WEBP_INLINE void NeedsFilter_SSE2(const __m128i* const p1,
                                          const __m128i* const q0,
                                          const __m128i* const q1,
                                          int thresh, __m128i* const mask) {
-  const __m128i m_thresh = _mm_set1_epi8(thresh);
+  const __m128i m_thresh = _mm_set1_epi8((char)thresh);
   const __m128i t1 = MM_ABS(*p1, *q1);        // abs(p1 - q1)
-  const __m128i kFE = _mm_set1_epi8(0xFE);
+  const __m128i kFE = _mm_set1_epi8((char)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
 
@@ -360,7 +360,7 @@ static WEBP_INLINE void DoFilter2_SSE2(__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);
+  const __m128i sign_bit = _mm_set1_epi8((char)0x80);
   // convert p1/q1 to int8_t (for GetBaseDelta_SSE2)
   const __m128i p1s = _mm_xor_si128(*p1, sign_bit);
   const __m128i q1s = _mm_xor_si128(*q1, sign_bit);
@@ -380,7 +380,7 @@ static WEBP_INLINE void DoFilter4_SSE2(__m128i* const p1, __m128i* const p0,
                                        const __m128i* const mask,
                                        int hev_thresh) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i sign_bit = _mm_set1_epi8(0x80);
+  const __m128i sign_bit = _mm_set1_epi8((char)0x80);
   const __m128i k64 = _mm_set1_epi8(64);
   const __m128i k3 = _mm_set1_epi8(3);
   const __m128i k4 = _mm_set1_epi8(4);
@@ -427,7 +427,7 @@ static WEBP_INLINE void DoFilter6_SSE2(__m128i* const p2, __m128i* const p1,
                                        const __m128i* const mask,
                                        int hev_thresh) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i sign_bit = _mm_set1_epi8(0x80);
+  const __m128i sign_bit = _mm_set1_epi8((char)0x80);
   __m128i a, not_hev;
 
   // compute hev mask
@@ -941,7 +941,7 @@ static void VR4_SSE2(uint8_t* dst) {   // Vertical-Right
   const __m128i ABCD0 = _mm_srli_si128(XABCD, 1);
   const __m128i abcd = _mm_avg_epu8(XABCD, ABCD0);
   const __m128i _XABCD = _mm_slli_si128(XABCD, 1);
-  const __m128i IXABCD = _mm_insert_epi16(_XABCD, I | (X << 8), 0);
+  const __m128i IXABCD = _mm_insert_epi16(_XABCD, (short)(I | (X << 8)), 0);
   const __m128i avg1 = _mm_avg_epu8(IXABCD, ABCD0);
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
   const __m128i avg2 = _mm_subs_epu8(avg1, lsb);

src/dsp/dsp.h

@@ -51,9 +51,7 @@ extern "C" {
 # define __has_builtin(x) 0
 #endif
 
-// for now, none of the optimizations below are available in emscripten
-#if !defined(EMSCRIPTEN)
-
+#if !defined(HAVE_CONFIG_H)
 #if defined(_MSC_VER) && _MSC_VER > 1310 && \
     (defined(_M_X64) || defined(_M_IX86))
 #define WEBP_MSC_SSE2  // Visual C++ SSE2 targets
@@ -63,6 +61,7 @@ extern "C" {
     (defined(_M_X64) || defined(_M_IX86))
 #define WEBP_MSC_SSE41  // Visual C++ SSE4.1 targets
 #endif
+#endif
 
 // WEBP_HAVE_* are used to indicate the presence of the instruction set in dsp
 // files without intrinsics, allowing the corresponding Init() to be called.
@@ -76,6 +75,9 @@ extern "C" {
 #define WEBP_USE_SSE41
 #endif
 
+#undef WEBP_MSC_SSE41
+#undef WEBP_MSC_SSE2
+
 // The intrinsics currently cause compiler errors with arm-nacl-gcc and the
 // inline assembly would need to be modified for use with Native Client.
 #if (defined(__ARM_NEON__) || \
@@ -110,8 +112,6 @@ extern "C" {
 #define WEBP_USE_MSA
 #endif
 
-#endif  /* EMSCRIPTEN */
-
 #ifndef WEBP_DSP_OMIT_C_CODE
 #define WEBP_DSP_OMIT_C_CODE 1
 #endif
@@ -193,6 +193,12 @@ extern "C" {
 #endif
 #endif
 
+// If 'ptr' is NULL, returns NULL. Otherwise returns 'ptr + off'.
+// Prevents undefined behavior sanitizer nullptr-with-nonzero-offset warning.
+#if !defined(WEBP_OFFSET_PTR)
+#define WEBP_OFFSET_PTR(ptr, off) (((ptr) == NULL) ? NULL : ((ptr) + (off)))
+#endif
+
 // Regularize the definition of WEBP_SWAP_16BIT_CSP (backward compatibility)
 #if !defined(WEBP_SWAP_16BIT_CSP)
 #define WEBP_SWAP_16BIT_CSP 0
@@ -246,9 +252,9 @@ extern VP8Fdct VP8FTransform2;   // performs two transforms at a time
 extern VP8WHT VP8FTransformWHT;
 // Predictions
 // *dst is the destination block. *top and *left can be NULL.
-typedef void (*VP8IntraPreds)(uint8_t *dst, const uint8_t* left,
+typedef void (*VP8IntraPreds)(uint8_t* dst, const uint8_t* left,
                               const uint8_t* top);
-typedef void (*VP8Intra4Preds)(uint8_t *dst, const uint8_t* top);
+typedef void (*VP8Intra4Preds)(uint8_t* dst, const uint8_t* top);
 extern VP8Intra4Preds VP8EncPredLuma4;
 extern VP8IntraPreds VP8EncPredLuma16;
 extern VP8IntraPreds VP8EncPredChroma8;
@@ -632,6 +638,8 @@ extern void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
 extern int (*WebPHasAlpha8b)(const uint8_t* src, int length);
 // This function returns true if src[4*i] contains a value different from 0xff.
 extern int (*WebPHasAlpha32b)(const uint8_t* src, int length);
+// replaces transparent values in src[] by 'color'.
+extern void (*WebPAlphaReplace)(uint32_t* src, int length, uint32_t color);
 
 // To be called first before using the above.
 void WebPInitAlphaProcessing(void);

src/dsp/enc_sse2.c

@@ -777,7 +777,7 @@ static WEBP_INLINE void VR4_SSE2(uint8_t* dst,
   const __m128i ABCD0 = _mm_srli_si128(XABCD, 1);
   const __m128i abcd = _mm_avg_epu8(XABCD, ABCD0);
   const __m128i _XABCD = _mm_slli_si128(XABCD, 1);
-  const __m128i IXABCD = _mm_insert_epi16(_XABCD, I | (X << 8), 0);
+  const __m128i IXABCD = _mm_insert_epi16(_XABCD, (short)(I | (X << 8)), 0);
   const __m128i avg1 = _mm_avg_epu8(IXABCD, ABCD0);
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
   const __m128i avg2 = _mm_subs_epu8(avg1, lsb);

src/dsp/filters.c

@@ -33,9 +33,9 @@ static WEBP_INLINE void PredictLine_C(const uint8_t* src, const uint8_t* pred,
                                       uint8_t* dst, int length, int inverse) {
   int i;
   if (inverse) {
-    for (i = 0; i < length; ++i) dst[i] = src[i] + pred[i];
+    for (i = 0; i < length; ++i) dst[i] = (uint8_t)(src[i] + pred[i]);
   } else {
-    for (i = 0; i < length; ++i) dst[i] = src[i] - pred[i];
+    for (i = 0; i < length; ++i) dst[i] = (uint8_t)(src[i] - pred[i]);
   }
 }
 
@@ -155,7 +155,7 @@ static WEBP_INLINE void DoGradientFilter_C(const uint8_t* in,
       const int pred = GradientPredictor_C(preds[w - 1],
                                            preds[w - stride],
                                            preds[w - stride - 1]);
-      out[w] = in[w] + (inverse ? pred : -pred);
+      out[w] = (uint8_t)(in[w] + (inverse ? pred : -pred));
     }
     ++row;
     preds += stride;
@@ -194,7 +194,7 @@ static void HorizontalUnfilter_C(const uint8_t* prev, const uint8_t* in,
   uint8_t pred = (prev == NULL) ? 0 : prev[0];
   int i;
   for (i = 0; i < width; ++i) {
-    out[i] = pred + in[i];
+    out[i] = (uint8_t)(pred + in[i]);
     pred = out[i];
   }
 }
@@ -206,7 +206,7 @@ static void VerticalUnfilter_C(const uint8_t* prev, const uint8_t* in,
     HorizontalUnfilter_C(NULL, in, out, width);
   } else {
     int i;
-    for (i = 0; i < width; ++i) out[i] = prev[i] + in[i];
+    for (i = 0; i < width; ++i) out[i] = (uint8_t)(prev[i] + in[i]);
   }
 }
 #endif  // !WEBP_NEON_OMIT_C_CODE
@@ -220,7 +220,7 @@ static void GradientUnfilter_C(const uint8_t* prev, const uint8_t* in,
     int i;
     for (i = 0; i < width; ++i) {
       top = prev[i];  // need to read this first, in case prev==out
-      left = in[i] + GradientPredictor_C(left, top, top_left);
+      left = (uint8_t)(in[i] + GradientPredictor_C(left, top, top_left));
       top_left = top;
       out[i] = left;
     }

src/dsp/filters_sse2.c

@@ -163,7 +163,8 @@ static void GradientPredictDirect_SSE2(const uint8_t* const row,
     _mm_storel_epi64((__m128i*)(out + i), H);
   }
   for (; i < length; ++i) {
-    out[i] = row[i] - GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
+    const int delta = GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
+    out[i] = (uint8_t)(row[i] - delta);
   }
 }
 
@@ -188,7 +189,7 @@ static WEBP_INLINE void DoGradientFilter_SSE2(const uint8_t* in,
 
   // Filter line-by-line.
   while (row < last_row) {
-    out[0] = in[0] - in[-stride];
+    out[0] = (uint8_t)(in[0] - in[-stride]);
     GradientPredictDirect_SSE2(in + 1, in + 1 - stride, out + 1, width - 1);
     ++row;
     in += stride;
@@ -223,7 +224,7 @@ static void HorizontalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
                                     uint8_t* out, int width) {
   int i;
   __m128i last;
-  out[0] = in[0] + (prev == NULL ? 0 : prev[0]);
+  out[0] = (uint8_t)(in[0] + (prev == NULL ? 0 : prev[0]));
   if (width <= 1) return;
   last = _mm_set_epi32(0, 0, 0, out[0]);
   for (i = 1; i + 8 <= width; i += 8) {
@@ -238,7 +239,7 @@ static void HorizontalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
     _mm_storel_epi64((__m128i*)(out + i), A7);
     last = _mm_srli_epi64(A7, 56);
   }
-  for (; i < width; ++i) out[i] = in[i] + out[i - 1];
+  for (; i < width; ++i) out[i] = (uint8_t)(in[i] + out[i - 1]);
 }
 
 static void VerticalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
@@ -259,7 +260,7 @@ static void VerticalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
       _mm_storeu_si128((__m128i*)&out[i +  0], C0);
       _mm_storeu_si128((__m128i*)&out[i + 16], C1);
     }
-    for (; i < width; ++i) out[i] = in[i] + prev[i];
+    for (; i < width; ++i) out[i] = (uint8_t)(in[i] + prev[i]);
   }
 }
 
@@ -296,7 +297,8 @@ static void GradientPredictInverse_SSE2(const uint8_t* const in,
       _mm_storel_epi64((__m128i*)&row[i], out);
     }
     for (; i < length; ++i) {
-      row[i] = in[i] + GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
+      const int delta = GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
+      row[i] = (uint8_t)(in[i] + delta);
     }
   }
 }
@@ -306,7 +308,7 @@ static void GradientUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
   if (prev == NULL) {
     HorizontalUnfilter_SSE2(NULL, in, out, width);
   } else {
-    out[0] = in[0] + prev[0];  // predict from above
+    out[0] = (uint8_t)(in[0] + prev[0]);  // predict from above
     GradientPredictInverse_SSE2(in + 1, prev + 1, out + 1, width - 1);
   }
 }

src/dsp/lossless.c

@@ -81,7 +81,7 @@ static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
 
 // gcc <= 4.9 on ARM generates incorrect code in Select() when Sub3() is
 // inlined.
-#if defined(__arm__) && LOCAL_GCC_VERSION <= 0x409
+#if defined(__arm__) && defined(__GNUC__) && LOCAL_GCC_VERSION <= 0x409
 # define LOCAL_INLINE __attribute__ ((noinline))
 #else
 # define LOCAL_INLINE WEBP_INLINE
@@ -107,88 +107,93 @@ static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
 //------------------------------------------------------------------------------
 // Predictors
 
-static uint32_t Predictor0_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor0_C(uint32_t left, const uint32_t* const top) {
   (void)top;
   (void)left;
   return ARGB_BLACK;
 }
-static uint32_t Predictor1_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor1_C(uint32_t left, const uint32_t* const top) {
   (void)top;
   return left;
 }
-static uint32_t Predictor2_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor2_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[0];
 }
-static uint32_t Predictor3_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor3_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[1];
 }
-static uint32_t Predictor4_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor4_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[-1];
 }
-static uint32_t Predictor5_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor5_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average3(left, top[0], top[1]);
   return pred;
 }
-static uint32_t Predictor6_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor6_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(left, top[-1]);
   return pred;
 }
-static uint32_t Predictor7_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor7_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(left, top[0]);
   return pred;
 }
-static uint32_t Predictor8_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor8_C(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_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor9_C(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_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor10_C(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_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor11_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Select(top[0], left, top[-1]);
   return pred;
 }
-static uint32_t Predictor12_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor12_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]);
   return pred;
 }
-static uint32_t Predictor13_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor13_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]);
   return pred;
 }
 
-GENERATE_PREDICTOR_ADD(Predictor0_C, PredictorAdd0_C)
+static void PredictorAdd0_C(const uint32_t* in, const uint32_t* upper,
+                            int num_pixels, uint32_t* out) {
+  int x;
+  (void)upper;
+  for (x = 0; x < num_pixels; ++x) out[x] = VP8LAddPixels(in[x], ARGB_BLACK);
+}
 static void PredictorAdd1_C(const uint32_t* in, const uint32_t* upper,
                             int num_pixels, uint32_t* out) {
   int i;
   uint32_t left = out[-1];
+  (void)upper;
   for (i = 0; i < num_pixels; ++i) {
     out[i] = left = VP8LAddPixels(in[i], left);
   }
-  (void)upper;
 }
-GENERATE_PREDICTOR_ADD(Predictor2_C, PredictorAdd2_C)
-GENERATE_PREDICTOR_ADD(Predictor3_C, PredictorAdd3_C)
-GENERATE_PREDICTOR_ADD(Predictor4_C, PredictorAdd4_C)
-GENERATE_PREDICTOR_ADD(Predictor5_C, PredictorAdd5_C)
-GENERATE_PREDICTOR_ADD(Predictor6_C, PredictorAdd6_C)
-GENERATE_PREDICTOR_ADD(Predictor7_C, PredictorAdd7_C)
-GENERATE_PREDICTOR_ADD(Predictor8_C, PredictorAdd8_C)
-GENERATE_PREDICTOR_ADD(Predictor9_C, PredictorAdd9_C)
-GENERATE_PREDICTOR_ADD(Predictor10_C, PredictorAdd10_C)
-GENERATE_PREDICTOR_ADD(Predictor11_C, PredictorAdd11_C)
-GENERATE_PREDICTOR_ADD(Predictor12_C, PredictorAdd12_C)
-GENERATE_PREDICTOR_ADD(Predictor13_C, PredictorAdd13_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor2_C, PredictorAdd2_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor3_C, PredictorAdd3_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor4_C, PredictorAdd4_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor5_C, PredictorAdd5_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor6_C, PredictorAdd6_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor7_C, PredictorAdd7_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor8_C, PredictorAdd8_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor9_C, PredictorAdd9_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor10_C, PredictorAdd10_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor11_C, PredictorAdd11_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor12_C, PredictorAdd12_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor13_C, PredictorAdd13_C)
 
 //------------------------------------------------------------------------------
 
@@ -270,14 +275,14 @@ void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
   int i;
   for (i = 0; i < num_pixels; ++i) {
     const uint32_t argb = src[i];
-    const uint32_t green = argb >> 8;
+    const int8_t green = (int8_t)(argb >> 8);
     const uint32_t red = argb >> 16;
     int new_red = red & 0xff;
     int new_blue = argb & 0xff;
     new_red += ColorTransformDelta(m->green_to_red_, green);
     new_red &= 0xff;
     new_blue += ColorTransformDelta(m->green_to_blue_, green);
-    new_blue += ColorTransformDelta(m->red_to_blue_, new_red);
+    new_blue += ColorTransformDelta(m->red_to_blue_, (int8_t)new_red);
     new_blue &= 0xff;
     dst[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue);
   }
@@ -557,7 +562,6 @@ VP8LPredictorFunc VP8LPredictors[16];
 
 // exposed plain-C implementations
 VP8LPredictorAddSubFunc VP8LPredictorsAdd_C[16];
-VP8LPredictorFunc VP8LPredictors_C[16];
 
 VP8LTransformColorInverseFunc VP8LTransformColorInverse;
 
@@ -595,8 +599,7 @@ extern void VP8LDspInitMSA(void);
 } while (0);
 
 WEBP_DSP_INIT_FUNC(VP8LDspInit) {
-  COPY_PREDICTOR_ARRAY(Predictor, VP8LPredictors)
-  COPY_PREDICTOR_ARRAY(Predictor, VP8LPredictors_C)
+  COPY_PREDICTOR_ARRAY(VP8LPredictor, VP8LPredictors)
   COPY_PREDICTOR_ARRAY(PredictorAdd, VP8LPredictorsAdd)
   COPY_PREDICTOR_ARRAY(PredictorAdd, VP8LPredictorsAdd_C)
 

src/dsp/lossless.h

@@ -30,7 +30,22 @@ extern "C" {
 
 typedef uint32_t (*VP8LPredictorFunc)(uint32_t left, const uint32_t* const top);
 extern VP8LPredictorFunc VP8LPredictors[16];
-extern VP8LPredictorFunc VP8LPredictors_C[16];
+
+uint32_t VP8LPredictor0_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor1_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor2_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor3_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor4_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor5_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor6_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor7_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor8_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor9_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor10_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor11_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor12_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor13_C(uint32_t left, const uint32_t* const top);
+
 // These Add/Sub function expects upper[-1] and out[-1] to be readable.
 typedef void (*VP8LPredictorAddSubFunc)(const uint32_t* in,
                                         const uint32_t* upper, int num_pixels,

src/dsp/lossless_common.h

@@ -177,24 +177,13 @@ uint32_t VP8LSubPixels(uint32_t a, uint32_t b) {
 static void PREDICTOR_ADD(const uint32_t* in, const uint32_t* upper, \
                           int num_pixels, uint32_t* out) {           \
   int x;                                                             \
+  assert(upper != NULL);                                             \
   for (x = 0; x < num_pixels; ++x) {                                 \
     const uint32_t pred = (PREDICTOR)(out[x - 1], upper + x);        \
     out[x] = VP8LAddPixels(in[x], pred);                             \
   }                                                                  \
 }
 
-// It subtracts the prediction from the input pixel and stores the residual
-// in the output pixel.
-#define GENERATE_PREDICTOR_SUB(PREDICTOR, PREDICTOR_SUB)             \
-static void PREDICTOR_SUB(const uint32_t* in, const uint32_t* upper, \
-                          int num_pixels, uint32_t* out) {           \
-  int x;                                                             \
-  for (x = 0; x < num_pixels; ++x) {                                 \
-    const uint32_t pred = (PREDICTOR)(in[x - 1], upper + x);         \
-    out[x] = VP8LSubPixels(in[x], pred);                             \
-  }                                                                  \
-}
-
 #ifdef __cplusplus
 }    // extern "C"
 #endif

src/dsp/lossless_enc.c

@@ -515,13 +515,17 @@ static WEBP_INLINE int ColorTransformDelta(int8_t color_pred, int8_t color) {
   return ((int)color_pred * color) >> 5;
 }
 
+static WEBP_INLINE int8_t U32ToS8(uint32_t v) {
+  return (int8_t)(v & 0xff);
+}
+
 void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
                           int num_pixels) {
   int i;
   for (i = 0; i < num_pixels; ++i) {
     const uint32_t argb = data[i];
-    const uint32_t green = argb >> 8;
-    const uint32_t red = argb >> 16;
+    const int8_t green = U32ToS8(argb >>  8);
+    const int8_t red   = U32ToS8(argb >> 16);
     int new_red = red & 0xff;
     int new_blue = argb & 0xff;
     new_red -= ColorTransformDelta(m->green_to_red_, green);
@@ -535,7 +539,7 @@ void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
 
 static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
                                              uint32_t argb) {
-  const uint32_t green = argb >> 8;
+  const int8_t green = U32ToS8(argb >> 8);
   int new_red = argb >> 16;
   new_red -= ColorTransformDelta(green_to_red, green);
   return (new_red & 0xff);
@@ -544,9 +548,9 @@ static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
 static WEBP_INLINE uint8_t TransformColorBlue(uint8_t green_to_blue,
                                               uint8_t red_to_blue,
                                               uint32_t argb) {
-  const uint32_t green = argb >> 8;
-  const uint32_t red = argb >> 16;
-  uint8_t new_blue = argb;
+  const int8_t green = U32ToS8(argb >>  8);
+  const int8_t red   = U32ToS8(argb >> 16);
+  uint8_t new_blue = argb & 0xff;
   new_blue -= ColorTransformDelta(green_to_blue, green);
   new_blue -= ColorTransformDelta(red_to_blue, red);
   return (new_blue & 0xff);
@@ -558,7 +562,7 @@ void VP8LCollectColorRedTransforms_C(const uint32_t* argb, int stride,
   while (tile_height-- > 0) {
     int x;
     for (x = 0; x < tile_width; ++x) {
-      ++histo[TransformColorRed(green_to_red, argb[x])];
+      ++histo[TransformColorRed((uint8_t)green_to_red, argb[x])];
     }
     argb += stride;
   }
@@ -571,7 +575,8 @@ void VP8LCollectColorBlueTransforms_C(const uint32_t* argb, int stride,
   while (tile_height-- > 0) {
     int x;
     for (x = 0; x < tile_width; ++x) {
-      ++histo[TransformColorBlue(green_to_blue, red_to_blue, argb[x])];
+      ++histo[TransformColorBlue((uint8_t)green_to_blue, (uint8_t)red_to_blue,
+                                 argb[x])];
     }
     argb += stride;
   }
@@ -697,140 +702,6 @@ void VP8LHistogramAdd(const VP8LHistogram* const a,
 //------------------------------------------------------------------------------
 // Image transforms.
 
-static WEBP_INLINE uint32_t Average2(uint32_t a0, uint32_t a1) {
-  return (((a0 ^ a1) & 0xfefefefeu) >> 1) + (a0 & a1);
-}
-
-static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) {
-  return Average2(Average2(a0, a2), a1);
-}
-
-static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
-                                     uint32_t a2, uint32_t a3) {
-  return Average2(Average2(a0, a1), Average2(a2, a3));
-}
-
-static WEBP_INLINE uint32_t Clip255(uint32_t a) {
-  if (a < 256) {
-    return a;
-  }
-  // return 0, when a is a negative integer.
-  // return 255, when a is positive.
-  return ~a >> 24;
-}
-
-static WEBP_INLINE int AddSubtractComponentFull(int a, int b, int c) {
-  return Clip255(a + b - c);
-}
-
-static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
-                                                   uint32_t c2) {
-  const int a = AddSubtractComponentFull(c0 >> 24, c1 >> 24, c2 >> 24);
-  const int r = AddSubtractComponentFull((c0 >> 16) & 0xff,
-                                         (c1 >> 16) & 0xff,
-                                         (c2 >> 16) & 0xff);
-  const int g = AddSubtractComponentFull((c0 >> 8) & 0xff,
-                                         (c1 >> 8) & 0xff,
-                                         (c2 >> 8) & 0xff);
-  const int b = AddSubtractComponentFull(c0 & 0xff, c1 & 0xff, c2 & 0xff);
-  return ((uint32_t)a << 24) | (r << 16) | (g << 8) | b;
-}
-
-static WEBP_INLINE int AddSubtractComponentHalf(int a, int b) {
-  return Clip255(a + (a - b) / 2);
-}
-
-static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
-                                                   uint32_t c2) {
-  const uint32_t ave = Average2(c0, c1);
-  const int a = AddSubtractComponentHalf(ave >> 24, c2 >> 24);
-  const int r = AddSubtractComponentHalf((ave >> 16) & 0xff, (c2 >> 16) & 0xff);
-  const int g = AddSubtractComponentHalf((ave >> 8) & 0xff, (c2 >> 8) & 0xff);
-  const int b = AddSubtractComponentHalf((ave >> 0) & 0xff, (c2 >> 0) & 0xff);
-  return ((uint32_t)a << 24) | (r << 16) | (g << 8) | b;
-}
-
-// gcc-4.9 on ARM generates incorrect code in Select() when Sub3() is inlined.
-#if defined(__arm__) && \
-    (LOCAL_GCC_VERSION == 0x409 || LOCAL_GCC_VERSION == 0x408)
-# define LOCAL_INLINE __attribute__ ((noinline))
-#else
-# define LOCAL_INLINE WEBP_INLINE
-#endif
-
-static LOCAL_INLINE int Sub3(int a, int b, int c) {
-  const int pb = b - c;
-  const int pa = a - c;
-  return abs(pb) - abs(pa);
-}
-
-#undef LOCAL_INLINE
-
-static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
-  const int pa_minus_pb =
-      Sub3((a >> 24)       , (b >> 24)       , (c >> 24)       ) +
-      Sub3((a >> 16) & 0xff, (b >> 16) & 0xff, (c >> 16) & 0xff) +
-      Sub3((a >>  8) & 0xff, (b >>  8) & 0xff, (c >>  8) & 0xff) +
-      Sub3((a      ) & 0xff, (b      ) & 0xff, (c      ) & 0xff);
-  return (pa_minus_pb <= 0) ? a : b;
-}
-
-//------------------------------------------------------------------------------
-// Predictors
-
-static uint32_t Predictor2(uint32_t left, const uint32_t* const top) {
-  (void)left;
-  return top[0];
-}
-static uint32_t Predictor3(uint32_t left, const uint32_t* const top) {
-  (void)left;
-  return top[1];
-}
-static uint32_t Predictor4(uint32_t left, const uint32_t* const top) {
-  (void)left;
-  return top[-1];
-}
-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;
-}
-
-//------------------------------------------------------------------------------
-
 static void PredictorSub0_C(const uint32_t* in, const uint32_t* upper,
                             int num_pixels, uint32_t* out) {
   int i;
@@ -845,18 +716,33 @@ static void PredictorSub1_C(const uint32_t* in, const uint32_t* upper,
   (void)upper;
 }
 
-GENERATE_PREDICTOR_SUB(Predictor2, PredictorSub2_C)
-GENERATE_PREDICTOR_SUB(Predictor3, PredictorSub3_C)
-GENERATE_PREDICTOR_SUB(Predictor4, PredictorSub4_C)
-GENERATE_PREDICTOR_SUB(Predictor5, PredictorSub5_C)
-GENERATE_PREDICTOR_SUB(Predictor6, PredictorSub6_C)
-GENERATE_PREDICTOR_SUB(Predictor7, PredictorSub7_C)
-GENERATE_PREDICTOR_SUB(Predictor8, PredictorSub8_C)
-GENERATE_PREDICTOR_SUB(Predictor9, PredictorSub9_C)
-GENERATE_PREDICTOR_SUB(Predictor10, PredictorSub10_C)
-GENERATE_PREDICTOR_SUB(Predictor11, PredictorSub11_C)
-GENERATE_PREDICTOR_SUB(Predictor12, PredictorSub12_C)
-GENERATE_PREDICTOR_SUB(Predictor13, PredictorSub13_C)
+// It subtracts the prediction from the input pixel and stores the residual
+// in the output pixel.
+#define GENERATE_PREDICTOR_SUB(PREDICTOR_I)                                \
+static void PredictorSub##PREDICTOR_I##_C(const uint32_t* in,              \
+                                          const uint32_t* upper,           \
+                                          int num_pixels, uint32_t* out) { \
+  int x;                                                                   \
+  assert(upper != NULL);                                                   \
+  for (x = 0; x < num_pixels; ++x) {                                       \
+    const uint32_t pred =                                                  \
+        VP8LPredictor##PREDICTOR_I##_C(in[x - 1], upper + x);              \
+    out[x] = VP8LSubPixels(in[x], pred);                                   \
+  }                                                                        \
+}
+
+GENERATE_PREDICTOR_SUB(2)
+GENERATE_PREDICTOR_SUB(3)
+GENERATE_PREDICTOR_SUB(4)
+GENERATE_PREDICTOR_SUB(5)
+GENERATE_PREDICTOR_SUB(6)
+GENERATE_PREDICTOR_SUB(7)
+GENERATE_PREDICTOR_SUB(8)
+GENERATE_PREDICTOR_SUB(9)
+GENERATE_PREDICTOR_SUB(10)
+GENERATE_PREDICTOR_SUB(11)
+GENERATE_PREDICTOR_SUB(12)
+GENERATE_PREDICTOR_SUB(13)
 
 //------------------------------------------------------------------------------
 

src/dsp/lossless_enc_sse2.c

@@ -249,6 +249,7 @@ static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
     }                                  \
   } while (0)
 
+#if !(defined(__i386__) || defined(_M_IX86))
 static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
   int i;
   double retval = 0.;
@@ -300,6 +301,8 @@ static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
   retval += VP8LFastSLog2(sumX) + VP8LFastSLog2(sumXY);
   return (float)retval;
 }
+#endif  // !(defined(__i386__) || defined(_M_IX86))
+
 #undef ANALYZE_X_OR_Y
 #undef ANALYZE_XY
 
@@ -363,7 +366,7 @@ static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
   assert(xbits <= 3);
   switch (xbits) {
     case 0: {
-      const __m128i ff = _mm_set1_epi16(0xff00);
+      const __m128i ff = _mm_set1_epi16((short)0xff00);
       const __m128i zero = _mm_setzero_si128();
       // Store 0xff000000 | (row[x] << 8).
       for (x = 0; x + 16 <= width; x += 16, dst += 16) {
@@ -382,7 +385,7 @@ static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
       break;
     }
     case 1: {
-      const __m128i ff = _mm_set1_epi16(0xff00);
+      const __m128i ff = _mm_set1_epi16((short)0xff00);
       const __m128i mul = _mm_set1_epi16(0x110);
       for (x = 0; x + 16 <= width; x += 16, dst += 8) {
         // 0a0b | (where a/b are 4 bits).
@@ -455,24 +458,27 @@ static void PredictorSub0_SSE2(const uint32_t* in, const uint32_t* upper,
     _mm_storeu_si128((__m128i*)&out[i], res);
   }
   if (i != num_pixels) {
-    VP8LPredictorsSub_C[0](in + i, upper + i, num_pixels - i, out + i);
+    VP8LPredictorsSub_C[0](in + i, NULL, num_pixels - i, out + i);
   }
+  (void)upper;
 }
 
-#define GENERATE_PREDICTOR_1(X, IN)                                           \
-static void PredictorSub##X##_SSE2(const uint32_t* in, const uint32_t* upper, \
-                                   int num_pixels, uint32_t* out) {           \
-  int i;                                                                      \
-  for (i = 0; i + 4 <= num_pixels; i += 4) {                                  \
-    const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);              \
-    const __m128i pred = _mm_loadu_si128((const __m128i*)&(IN));              \
-    const __m128i res = _mm_sub_epi8(src, pred);                              \
-    _mm_storeu_si128((__m128i*)&out[i], res);                                 \
-  }                                                                           \
-  if (i != num_pixels) {                                                      \
-    VP8LPredictorsSub_C[(X)](in + i, upper + i, num_pixels - i, out + i);     \
-  }                                                                           \
-}
+#define GENERATE_PREDICTOR_1(X, IN)                                         \
+  static void PredictorSub##X##_SSE2(const uint32_t* const in,              \
+                                     const uint32_t* const upper,           \
+                                     int num_pixels, uint32_t* const out) { \
+    int i;                                                                  \
+    for (i = 0; i + 4 <= num_pixels; i += 4) {                              \
+      const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);          \
+      const __m128i pred = _mm_loadu_si128((const __m128i*)&(IN));          \
+      const __m128i res = _mm_sub_epi8(src, pred);                          \
+      _mm_storeu_si128((__m128i*)&out[i], res);                             \
+    }                                                                       \
+    if (i != num_pixels) {                                                  \
+      VP8LPredictorsSub_C[(X)](in + i, WEBP_OFFSET_PTR(upper, i),           \
+                               num_pixels - i, out + i);                    \
+    }                                                                       \
+  }
 
 GENERATE_PREDICTOR_1(1, in[i - 1])       // Predictor1: L
 GENERATE_PREDICTOR_1(2, upper[i])        // Predictor2: T
@@ -656,7 +662,12 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE2(void) {
   VP8LCollectColorRedTransforms = CollectColorRedTransforms_SSE2;
   VP8LAddVector = AddVector_SSE2;
   VP8LAddVectorEq = AddVectorEq_SSE2;
+  // TODO(https://crbug.com/webp/499): this function produces different results
+  // from the C code due to use of double/float resulting in output differences
+  // when compared to -noasm.
+#if !(defined(__i386__) || defined(_M_IX86))
   VP8LCombinedShannonEntropy = CombinedShannonEntropy_SSE2;
+#endif
   VP8LVectorMismatch = VectorMismatch_SSE2;
   VP8LBundleColorMap = BundleColorMap_SSE2;
 

src/dsp/lossless_enc_sse41.c

@@ -51,9 +51,9 @@ static void CollectColorBlueTransforms_SSE41(const uint32_t* argb, int stride,
                                              int histo[]) {
   const __m128i mults_r = _mm_set1_epi16(CST_5b(red_to_blue));
   const __m128i mults_g = _mm_set1_epi16(CST_5b(green_to_blue));
-  const __m128i mask_g = _mm_set1_epi16(0xff00);   // green mask
-  const __m128i mask_gb = _mm_set1_epi32(0xffff);  // green/blue mask
-  const __m128i mask_b = _mm_set1_epi16(0x00ff);   // blue mask
+  const __m128i mask_g = _mm_set1_epi16((short)0xff00);   // green mask
+  const __m128i mask_gb = _mm_set1_epi32(0xffff);         // green/blue mask
+  const __m128i mask_b = _mm_set1_epi16(0x00ff);          // blue mask
   const __m128i shuffler_lo = _mm_setr_epi8(-1, 2, -1, 6, -1, 10, -1, 14, -1,
                                             -1, -1, -1, -1, -1, -1, -1);
   const __m128i shuffler_hi = _mm_setr_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1,

src/dsp/lossless_sse2.c

@@ -191,8 +191,9 @@ static void PredictorAdd0_SSE2(const uint32_t* in, const uint32_t* upper,
     _mm_storeu_si128((__m128i*)&out[i], res);
   }
   if (i != num_pixels) {
-    VP8LPredictorsAdd_C[0](in + i, upper + i, num_pixels - i, out + i);
+    VP8LPredictorsAdd_C[0](in + i, NULL, num_pixels - i, out + i);
   }
+  (void)upper;
 }
 
 // Predictor1: left.

src/dsp/quant.h

@@ -0,0 +1,85 @@
+// Copyright 2018 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.
+// -----------------------------------------------------------------------------
+
+#ifndef WEBP_DSP_QUANT_H_
+#define WEBP_DSP_QUANT_H_
+
+#include <string.h>
+
+#include "src/dsp/dsp.h"
+#include "src/webp/types.h"
+
+#if defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) && \
+    !defined(WEBP_HAVE_NEON_RTCD)
+#include <arm_neon.h>
+
+#define IsFlat IsFlat_NEON
+
+static uint32x2_t horizontal_add_uint32x4(const uint32x4_t a) {
+  const uint64x2_t b = vpaddlq_u32(a);
+  return vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
+                  vreinterpret_u32_u64(vget_high_u64(b)));
+}
+
+static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
+                              int thresh) {
+  const int16x8_t tst_ones = vdupq_n_s16(-1);
+  uint32x4_t sum = vdupq_n_u32(0);
+
+  for (int i = 0; i < num_blocks; ++i) {
+    // Set DC to zero.
+    const int16x8_t a_0 = vsetq_lane_s16(0, vld1q_s16(levels), 0);
+    const int16x8_t a_1 = vld1q_s16(levels + 8);
+
+    const uint16x8_t b_0 = vshrq_n_u16(vtstq_s16(a_0, tst_ones), 15);
+    const uint16x8_t b_1 = vshrq_n_u16(vtstq_s16(a_1, tst_ones), 15);
+
+    sum = vpadalq_u16(sum, b_0);
+    sum = vpadalq_u16(sum, b_1);
+
+    levels += 16;
+  }
+  return thresh >= (int32_t)vget_lane_u32(horizontal_add_uint32x4(sum), 0);
+}
+
+#else
+
+#define IsFlat IsFlat_C
+
+static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
+                              int thresh) {
+  int score = 0;
+  while (num_blocks-- > 0) {      // TODO(skal): refine positional scoring?
+    int i;
+    for (i = 1; i < 16; ++i) {    // omit DC, we're only interested in AC
+      score += (levels[i] != 0);
+      if (score > thresh) return 0;
+    }
+    levels += 16;
+  }
+  return 1;
+}
+
+#endif  // defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) &&
+        // !defined(WEBP_HAVE_NEON_RTCD)
+
+static WEBP_INLINE int IsFlatSource16(const uint8_t* src) {
+  const uint32_t v = src[0] * 0x01010101u;
+  int i;
+  for (i = 0; i < 16; ++i) {
+    if (memcmp(src + 0, &v, 4) || memcmp(src +  4, &v, 4) ||
+        memcmp(src + 8, &v, 4) || memcmp(src + 12, &v, 4)) {
+      return 0;
+    }
+    src += BPS;
+  }
+  return 1;
+}
+
+#endif  // WEBP_DSP_QUANT_H_

src/dsp/upsampling_msa.c

@@ -576,9 +576,9 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y,        \
   const uint32_t l_uv = ((cur_u[0]) | ((cur_v[0]) << 16));               \
   const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2;            \
   const uint8_t* ptop_y = &top_y[1];                                     \
-  uint8_t *ptop_dst = top_dst + XSTEP;                                   \
+  uint8_t* ptop_dst = top_dst + XSTEP;                                   \
   const uint8_t* pbot_y = &bot_y[1];                                     \
-  uint8_t *pbot_dst = bot_dst + XSTEP;                                   \
+  uint8_t* pbot_dst = bot_dst + XSTEP;                                   \
                                                                          \
   FUNC(top_y[0], uv0 & 0xff, (uv0 >> 16), top_dst);                      \
   if (bot_y != NULL) {                                                   \

src/dsp/upsampling_neon.c

@@ -58,8 +58,8 @@
 } while (0)
 
 // Turn the macro into a function for reducing code-size when non-critical
-static void Upsample16Pixels_NEON(const uint8_t *r1, const uint8_t *r2,
-                                  uint8_t *out) {
+static void Upsample16Pixels_NEON(const uint8_t* r1, const uint8_t* r2,
+                                  uint8_t* out) {
   UPSAMPLE_16PIXELS(r1, r2, out);
 }
 
@@ -190,14 +190,14 @@ static const int16_t kCoeffs1[4] = { 19077, 26149, 6419, 13320 };
 }
 
 #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) { \
+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);     \
+  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;                             \

src/enc/analysis_enc.c

@@ -126,16 +126,6 @@ static void InitHistogram(VP8Histogram* const histo) {
   histo->last_non_zero = 1;
 }
 
-static void MergeHistograms(const VP8Histogram* const in,
-                            VP8Histogram* const out) {
-  if (in->max_value > out->max_value) {
-    out->max_value = in->max_value;
-  }
-  if (in->last_non_zero > out->last_non_zero) {
-    out->last_non_zero = in->last_non_zero;
-  }
-}
-
 //------------------------------------------------------------------------------
 // Simplified k-Means, to assign Nb segments based on alpha-histogram
 
@@ -285,49 +275,6 @@ static int FastMBAnalyze(VP8EncIterator* const it) {
   return 0;
 }
 
-static int MBAnalyzeBestIntra4Mode(VP8EncIterator* const it,
-                                   int best_alpha) {
-  uint8_t modes[16];
-  const int max_mode = MAX_INTRA4_MODE;
-  int i4_alpha;
-  VP8Histogram total_histo;
-  int cur_histo = 0;
-  InitHistogram(&total_histo);
-
-  VP8IteratorStartI4(it);
-  do {
-    int mode;
-    int best_mode_alpha = DEFAULT_ALPHA;
-    VP8Histogram histos[2];
-    const uint8_t* const src = it->yuv_in_ + Y_OFF_ENC + VP8Scan[it->i4_];
-
-    VP8MakeIntra4Preds(it);
-    for (mode = 0; mode < max_mode; ++mode) {
-      int alpha;
-
-      InitHistogram(&histos[cur_histo]);
-      VP8CollectHistogram(src, it->yuv_p_ + VP8I4ModeOffsets[mode],
-                          0, 1, &histos[cur_histo]);
-      alpha = GetAlpha(&histos[cur_histo]);
-      if (IS_BETTER_ALPHA(alpha, best_mode_alpha)) {
-        best_mode_alpha = alpha;
-        modes[it->i4_] = mode;
-        cur_histo ^= 1;   // keep track of best histo so far.
-      }
-    }
-    // accumulate best histogram
-    MergeHistograms(&histos[cur_histo ^ 1], &total_histo);
-    // Note: we reuse the original samples for predictors
-  } while (VP8IteratorRotateI4(it, it->yuv_in_ + Y_OFF_ENC));
-
-  i4_alpha = GetAlpha(&total_histo);
-  if (IS_BETTER_ALPHA(i4_alpha, best_alpha)) {
-    VP8SetIntra4Mode(it, modes);
-    best_alpha = i4_alpha;
-  }
-  return best_alpha;
-}
-
 static int MBAnalyzeBestUVMode(VP8EncIterator* const it) {
   int best_alpha = DEFAULT_ALPHA;
   int smallest_alpha = 0;
@@ -371,13 +318,6 @@ static void MBAnalyze(VP8EncIterator* const it,
     best_alpha = FastMBAnalyze(it);
   } else {
     best_alpha = MBAnalyzeBestIntra16Mode(it);
-    if (enc->method_ >= 5) {
-      // We go and make a fast decision for intra4/intra16.
-      // It's usually not a good and definitive pick, but helps seeding the
-      // stats about level bit-cost.
-      // TODO(skal): improve criterion.
-      best_alpha = MBAnalyzeBestIntra4Mode(it, best_alpha);
-    }
   }
   best_uv_alpha = MBAnalyzeBestUVMode(it);
 

src/enc/backward_references_enc.c

@@ -11,13 +11,14 @@
 //
 
 #include <assert.h>
+#include <float.h>
 #include <math.h>
 
-#include "src/enc/backward_references_enc.h"
-#include "src/enc/histogram_enc.h"
+#include "src/dsp/dsp.h"
 #include "src/dsp/lossless.h"
 #include "src/dsp/lossless_common.h"
-#include "src/dsp/dsp.h"
+#include "src/enc/backward_references_enc.h"
+#include "src/enc/histogram_enc.h"
 #include "src/utils/color_cache_utils.h"
 #include "src/utils/utils.h"
 
@@ -103,6 +104,20 @@ void VP8LBackwardRefsClear(VP8LBackwardRefs* const refs) {
   }
 }
 
+// Swaps the content of two VP8LBackwardRefs.
+static void BackwardRefsSwap(VP8LBackwardRefs* const refs1,
+                             VP8LBackwardRefs* const refs2) {
+  const int point_to_refs1 =
+      (refs1->tail_ != NULL && refs1->tail_ == &refs1->refs_);
+  const int point_to_refs2 =
+      (refs2->tail_ != NULL && refs2->tail_ == &refs2->refs_);
+  const VP8LBackwardRefs tmp = *refs1;
+  *refs1 = *refs2;
+  *refs2 = tmp;
+  if (point_to_refs2) refs1->tail_ = &refs1->refs_;
+  if (point_to_refs1) refs2->tail_ = &refs2->refs_;
+}
+
 void VP8LBackwardRefsInit(VP8LBackwardRefs* const refs, int block_size) {
   assert(refs != NULL);
   memset(refs, 0, sizeof(*refs));
@@ -154,6 +169,22 @@ static PixOrCopyBlock* BackwardRefsNewBlock(VP8LBackwardRefs* const refs) {
   return b;
 }
 
+// Return 1 on success, 0 on error.
+static int BackwardRefsClone(const VP8LBackwardRefs* const from,
+                             VP8LBackwardRefs* const to) {
+  const PixOrCopyBlock* block_from = from->refs_;
+  VP8LClearBackwardRefs(to);
+  while (block_from != NULL) {
+    PixOrCopyBlock* const block_to = BackwardRefsNewBlock(to);
+    if (block_to == NULL) return 0;
+    memcpy(block_to->start_, block_from->start_,
+           block_from->size_ * sizeof(PixOrCopy));
+    block_to->size_ = block_from->size_;
+    block_from = block_from->next_;
+  }
+  return 1;
+}
+
 extern void VP8LBackwardRefsCursorAdd(VP8LBackwardRefs* const refs,
                                       const PixOrCopy v);
 void VP8LBackwardRefsCursorAdd(VP8LBackwardRefs* const refs,
@@ -191,13 +222,14 @@ void VP8LHashChainClear(VP8LHashChain* const p) {
 
 // -----------------------------------------------------------------------------
 
-#define HASH_MULTIPLIER_HI (0xc6a4a793ULL)
-#define HASH_MULTIPLIER_LO (0x5bd1e996ULL)
+static const uint32_t kHashMultiplierHi = 0xc6a4a793u;
+static const uint32_t kHashMultiplierLo = 0x5bd1e996u;
 
-static WEBP_INLINE uint32_t GetPixPairHash64(const uint32_t* const argb) {
+static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
+uint32_t GetPixPairHash64(const uint32_t* const argb) {
   uint32_t key;
-  key  = (argb[1] * HASH_MULTIPLIER_HI) & 0xffffffffu;
-  key += (argb[0] * HASH_MULTIPLIER_LO) & 0xffffffffu;
+  key  = argb[1] * kHashMultiplierHi;
+  key += argb[0] * kHashMultiplierLo;
   key = key >> (32 - HASH_BITS);
   return key;
 }
@@ -752,12 +784,18 @@ static int CalculateBestCacheSize(const uint32_t* argb, int quality,
         }
       }
     } else {
+      int code, extra_bits, extra_bits_value;
       // We should compute the contribution of the (distance,length)
       // histograms but those are the same independently from the cache size.
       // As those constant contributions are in the end added to the other
-      // histogram contributions, we can safely ignore them.
+      // histogram contributions, we can ignore them, except for the length
+      // prefix that is part of the literal_ histogram.
       int len = PixOrCopyLength(v);
       uint32_t argb_prev = *argb ^ 0xffffffffu;
+      VP8LPrefixEncode(len, &code, &extra_bits, &extra_bits_value);
+      for (i = 0; i <= cache_bits_max; ++i) {
+        ++histos[i]->literal_[NUM_LITERAL_CODES + code];
+      }
       // Update the color caches.
       do {
         if (*argb != argb_prev) {
@@ -841,16 +879,21 @@ extern int VP8LBackwardReferencesTraceBackwards(
     int xsize, int ysize, const uint32_t* const argb, int cache_bits,
     const VP8LHashChain* const hash_chain,
     const VP8LBackwardRefs* const refs_src, VP8LBackwardRefs* const refs_dst);
-static VP8LBackwardRefs* GetBackwardReferences(
-    int width, int height, const uint32_t* const argb, int quality,
-    int lz77_types_to_try, int* const cache_bits,
-    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* best,
-    VP8LBackwardRefs* worst) {
-  const int cache_bits_initial = *cache_bits;
-  double bit_cost_best = -1;
+static int GetBackwardReferences(int width, int height,
+                                 const uint32_t* const argb, int quality,
+                                 int lz77_types_to_try, int cache_bits_max,
+                                 int do_no_cache,
+                                 const VP8LHashChain* const hash_chain,
+                                 VP8LBackwardRefs* const refs,
+                                 int* const cache_bits_best) {
   VP8LHistogram* histo = NULL;
-  int lz77_type, lz77_type_best = 0;
+  int i, lz77_type;
+  // Index 0 is for a color cache, index 1 for no cache (if needed).
+  int lz77_types_best[2] = {0, 0};
+  double bit_costs_best[2] = {DBL_MAX, DBL_MAX};
   VP8LHashChain hash_chain_box;
+  VP8LBackwardRefs* const refs_tmp = &refs[do_no_cache ? 2 : 1];
+  int status = 0;
   memset(&hash_chain_box, 0, sizeof(hash_chain_box));
 
   histo = VP8LAllocateHistogram(MAX_COLOR_CACHE_BITS);
@@ -859,87 +902,130 @@ static VP8LBackwardRefs* GetBackwardReferences(
   for (lz77_type = 1; lz77_types_to_try;
        lz77_types_to_try &= ~lz77_type, lz77_type <<= 1) {
     int res = 0;
-    double bit_cost;
-    int cache_bits_tmp = cache_bits_initial;
+    double bit_cost = 0.;
     if ((lz77_types_to_try & lz77_type) == 0) continue;
     switch (lz77_type) {
       case kLZ77RLE:
-        res = BackwardReferencesRle(width, height, argb, 0, worst);
+        res = BackwardReferencesRle(width, height, argb, 0, refs_tmp);
         break;
       case kLZ77Standard:
         // Compute LZ77 with no cache (0 bits), as the ideal LZ77 with a color
         // cache is not that different in practice.
-        res = BackwardReferencesLz77(width, height, argb, 0, hash_chain, worst);
+        res = BackwardReferencesLz77(width, height, argb, 0, hash_chain,
+                                     refs_tmp);
         break;
       case kLZ77Box:
         if (!VP8LHashChainInit(&hash_chain_box, width * height)) goto Error;
         res = BackwardReferencesLz77Box(width, height, argb, 0, hash_chain,
-                                        &hash_chain_box, worst);
+                                        &hash_chain_box, refs_tmp);
         break;
       default:
         assert(0);
     }
     if (!res) goto Error;
 
-    // Next, try with a color cache and update the references.
-    if (!CalculateBestCacheSize(argb, quality, worst, &cache_bits_tmp)) {
-      goto Error;
-    }
-    if (cache_bits_tmp > 0) {
-      if (!BackwardRefsWithLocalCache(argb, cache_bits_tmp, worst)) {
-        goto Error;
+    // Start with the no color cache case.
+    for (i = 1; i >= 0; --i) {
+      int cache_bits = (i == 1) ? 0 : cache_bits_max;
+
+      if (i == 1 && !do_no_cache) continue;
+
+      if (i == 0) {
+        // Try with a color cache.
+        if (!CalculateBestCacheSize(argb, quality, refs_tmp, &cache_bits)) {
+          goto Error;
+        }
+        if (cache_bits > 0) {
+          if (!BackwardRefsWithLocalCache(argb, cache_bits, refs_tmp)) {
+            goto Error;
+          }
+        }
+      }
+
+      if (i == 0 && do_no_cache && cache_bits == 0) {
+        // No need to re-compute bit_cost as it was computed at i == 1.
+      } else {
+        VP8LHistogramCreate(histo, refs_tmp, cache_bits);
+        bit_cost = VP8LHistogramEstimateBits(histo);
+      }
+
+      if (bit_cost < bit_costs_best[i]) {
+        if (i == 1) {
+          // Do not swap as the full cache analysis would have the wrong
+          // VP8LBackwardRefs to start with.
+          if (!BackwardRefsClone(refs_tmp, &refs[1])) goto Error;
+        } else {
+          BackwardRefsSwap(refs_tmp, &refs[0]);
+        }
+        bit_costs_best[i] = bit_cost;
+        lz77_types_best[i] = lz77_type;
+        if (i == 0) *cache_bits_best = cache_bits;
       }
     }
-
-    // Keep the best backward references.
-    VP8LHistogramCreate(histo, worst, cache_bits_tmp);
-    bit_cost = VP8LHistogramEstimateBits(histo);
-    if (lz77_type_best == 0 || bit_cost < bit_cost_best) {
-      VP8LBackwardRefs* const tmp = worst;
-      worst = best;
-      best = tmp;
-      bit_cost_best = bit_cost;
-      *cache_bits = cache_bits_tmp;
-      lz77_type_best = lz77_type;
-    }
   }
-  assert(lz77_type_best > 0);
+  assert(lz77_types_best[0] > 0);
+  assert(!do_no_cache || lz77_types_best[1] > 0);
 
   // Improve on simple LZ77 but only for high quality (TraceBackwards is
   // costly).
-  if ((lz77_type_best == kLZ77Standard || lz77_type_best == kLZ77Box) &&
-      quality >= 25) {
-    const VP8LHashChain* const hash_chain_tmp =
-        (lz77_type_best == kLZ77Standard) ? hash_chain : &hash_chain_box;
-    double bit_cost_trace;
-    if (!VP8LBackwardReferencesTraceBackwards(width, height, argb, *cache_bits,
-                                              hash_chain_tmp, best, worst)) {
-      goto Error;
+  for (i = 1; i >= 0; --i) {
+    if (i == 1 && !do_no_cache) continue;
+    if ((lz77_types_best[i] == kLZ77Standard ||
+         lz77_types_best[i] == kLZ77Box) &&
+        quality >= 25) {
+      const VP8LHashChain* const hash_chain_tmp =
+          (lz77_types_best[i] == kLZ77Standard) ? hash_chain : &hash_chain_box;
+      const int cache_bits = (i == 1) ? 0 : *cache_bits_best;
+      double bit_cost_trace;
+      if (!VP8LBackwardReferencesTraceBackwards(width, height, argb, cache_bits,
+                                                hash_chain_tmp, &refs[i],
+                                                refs_tmp)) {
+        goto Error;
+      }
+      VP8LHistogramCreate(histo, refs_tmp, cache_bits);
+      bit_cost_trace = VP8LHistogramEstimateBits(histo);
+      if (bit_cost_trace < bit_costs_best[i]) {
+        BackwardRefsSwap(refs_tmp, &refs[i]);
+      }
     }
-    VP8LHistogramCreate(histo, worst, *cache_bits);
-    bit_cost_trace = VP8LHistogramEstimateBits(histo);
-    if (bit_cost_trace < bit_cost_best) best = worst;
-  }
 
-  BackwardReferences2DLocality(width, best);
+    BackwardReferences2DLocality(width, &refs[i]);
+
+    if (i == 1 && lz77_types_best[0] == lz77_types_best[1] &&
+        *cache_bits_best == 0) {
+      // If the best cache size is 0 and we have the same best LZ77, just copy
+      // the data over and stop here.
+      if (!BackwardRefsClone(&refs[1], &refs[0])) goto Error;
+      break;
+    }
+  }
+  status = 1;
 
 Error:
   VP8LHashChainClear(&hash_chain_box);
   VP8LFreeHistogram(histo);
-  return best;
+  return status;
 }
 
-VP8LBackwardRefs* VP8LGetBackwardReferences(
+WebPEncodingError VP8LGetBackwardReferences(
     int width, int height, const uint32_t* const argb, int quality,
-    int low_effort, int lz77_types_to_try, int* const cache_bits,
-    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs_tmp1,
-    VP8LBackwardRefs* const refs_tmp2) {
+    int low_effort, int lz77_types_to_try, int cache_bits_max, int do_no_cache,
+    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs,
+    int* const cache_bits_best) {
   if (low_effort) {
-    return GetBackwardReferencesLowEffort(width, height, argb, cache_bits,
-                                          hash_chain, refs_tmp1);
+    VP8LBackwardRefs* refs_best;
+    *cache_bits_best = cache_bits_max;
+    refs_best = GetBackwardReferencesLowEffort(
+        width, height, argb, cache_bits_best, hash_chain, refs);
+    if (refs_best == NULL) return VP8_ENC_ERROR_OUT_OF_MEMORY;
+    // Set it in first position.
+    BackwardRefsSwap(refs_best, &refs[0]);
   } else {
-    return GetBackwardReferences(width, height, argb, quality,
-                                 lz77_types_to_try, cache_bits, hash_chain,
-                                 refs_tmp1, refs_tmp2);
+    if (!GetBackwardReferences(width, height, argb, quality, lz77_types_to_try,
+                               cache_bits_max, do_no_cache, hash_chain, refs,
+                               cache_bits_best)) {
+      return VP8_ENC_ERROR_OUT_OF_MEMORY;
+    }
   }
+  return VP8_ENC_OK;
 }

src/enc/backward_references_enc.h

@@ -16,6 +16,7 @@
 #include <assert.h>
 #include <stdlib.h>
 #include "src/webp/types.h"
+#include "src/webp/encode.h"
 #include "src/webp/format_constants.h"
 
 #ifdef __cplusplus
@@ -218,14 +219,19 @@ enum VP8LLZ77Type {
 // Evaluates best possible backward references for specified quality.
 // The input cache_bits to 'VP8LGetBackwardReferences' sets the maximum cache
 // bits to use (passing 0 implies disabling the local color cache).
-// The optimal cache bits is evaluated and set for the *cache_bits parameter.
-// The return value is the pointer to the best of the two backward refs viz,
-// refs[0] or refs[1].
-VP8LBackwardRefs* VP8LGetBackwardReferences(
+// The optimal cache bits is evaluated and set for the *cache_bits_best
+// parameter with the matching refs_best.
+// If do_no_cache == 0, refs is an array of 2 values and the best
+// VP8LBackwardRefs is put in the first element.
+// If do_no_cache != 0, refs is an array of 3 values and the best
+// VP8LBackwardRefs is put in the first element, the best value with no-cache in
+// the second element.
+// In both cases, the last element is used as temporary internally.
+WebPEncodingError VP8LGetBackwardReferences(
     int width, int height, const uint32_t* const argb, int quality,
-    int low_effort, int lz77_types_to_try, int* const cache_bits,
-    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs_tmp1,
-    VP8LBackwardRefs* const refs_tmp2);
+    int low_effort, int lz77_types_to_try, int cache_bits_max, int do_no_cache,
+    const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs,
+    int* const cache_bits_best);
 
 #ifdef __cplusplus
 }

src/enc/config_enc.c

@@ -39,6 +39,8 @@ int WebPConfigInitInternal(WebPConfig* config,
   config->partitions = 0;
   config->segments = 4;
   config->pass = 1;
+  config->qmin = 0;
+  config->qmax = 100;
   config->show_compressed = 0;
   config->preprocessing = 0;
   config->autofilter = 0;
@@ -106,6 +108,9 @@ int WebPValidateConfig(const WebPConfig* config) {
   if (config->filter_type < 0 || config->filter_type > 1) return 0;
   if (config->autofilter < 0 || config->autofilter > 1) return 0;
   if (config->pass < 1 || config->pass > 10) return 0;
+  if (config->qmin < 0 || config->qmax > 100 || config->qmin > config->qmax) {
+    return 0;
+  }
   if (config->show_compressed < 0 || config->show_compressed > 1) return 0;
   if (config->preprocessing < 0 || config->preprocessing > 7) return 0;
   if (config->partitions < 0 || config->partitions > 3) return 0;

src/enc/frame_enc.c

@@ -31,10 +31,15 @@
 // we allow 2k of extra head-room in PARTITION0 limit.
 #define PARTITION0_SIZE_LIMIT ((VP8_MAX_PARTITION0_SIZE - 2048ULL) << 11)
 
+static float Clamp(float v, float min, float max) {
+  return (v < min) ? min : (v > max) ? max : v;
+}
+
 typedef struct {  // struct for organizing convergence in either size or PSNR
   int is_first;
   float dq;
   float q, last_q;
+  float qmin, qmax;
   double value, last_value;   // PSNR or size
   double target;
   int do_size_search;
@@ -47,7 +52,9 @@ static int InitPassStats(const VP8Encoder* const enc, PassStats* const s) {
 
   s->is_first = 1;
   s->dq = 10.f;
-  s->q = s->last_q = enc->config_->quality;
+  s->qmin = 1.f * enc->config_->qmin;
+  s->qmax = 1.f * enc->config_->qmax;
+  s->q = s->last_q = Clamp(enc->config_->quality, s->qmin, s->qmax);
   s->target = do_size_search ? (double)target_size
             : (target_PSNR > 0.) ? target_PSNR
             : 40.;   // default, just in case
@@ -56,10 +63,6 @@ static int InitPassStats(const VP8Encoder* const enc, PassStats* const s) {
   return do_size_search;
 }
 
-static float Clamp(float v, float min, float max) {
-  return (v < min) ? min : (v > max) ? max : v;
-}
-
 static float ComputeNextQ(PassStats* const s) {
   float dq;
   if (s->is_first) {
@@ -75,7 +78,7 @@ static float ComputeNextQ(PassStats* const s) {
   s->dq = Clamp(dq, -30.f, 30.f);
   s->last_q = s->q;
   s->last_value = s->value;
-  s->q = Clamp(s->q + s->dq, 0.f, 100.f);
+  s->q = Clamp(s->q + s->dq, s->qmin, s->qmax);
   return s->q;
 }
 
@@ -848,9 +851,10 @@ int VP8EncTokenLoop(VP8Encoder* const enc) {
     }
 
 #if (DEBUG_SEARCH > 0)
-    printf("#%2d metric:%.1lf -> %.1lf   last_q=%.2lf q=%.2lf dq=%.2lf\n",
+    printf("#%2d metric:%.1lf -> %.1lf   last_q=%.2lf q=%.2lf dq=%.2lf "
+           " range:[%.1f, %.1f]\n",
            num_pass_left, stats.last_value, stats.value,
-           stats.last_q, stats.q, stats.dq);
+           stats.last_q, stats.q, stats.dq, stats.qmin, stats.qmax);
 #endif
     if (enc->max_i4_header_bits_ > 0 && size_p0 > PARTITION0_SIZE_LIMIT) {
       ++num_pass_left;

src/enc/histogram_enc.c

@@ -165,7 +165,7 @@ VP8LHistogramSet* VP8LAllocateHistogramSet(int size, int cache_bits) {
 void VP8LHistogramSetClear(VP8LHistogramSet* const set) {
   int i;
   const int cache_bits = set->histograms[0]->palette_code_bits_;
-  const int size = set->size;
+  const int size = set->max_size;
   const size_t total_size = HistogramSetTotalSize(size, cache_bits);
   uint8_t* memory = (uint8_t*)set;
 
@@ -180,6 +180,20 @@ void VP8LHistogramSetClear(VP8LHistogramSet* const set) {
   }
 }
 
+// Removes the histogram 'i' from 'set' by setting it to NULL.
+static void HistogramSetRemoveHistogram(VP8LHistogramSet* const set, int i,
+                                        int* const num_used) {
+  assert(set->histograms[i] != NULL);
+  set->histograms[i] = NULL;
+  --*num_used;
+  // If we remove the last valid one, shrink until the next valid one.
+  if (i == set->size - 1) {
+    while (set->size >= 1 && set->histograms[set->size - 1] == NULL) {
+      --set->size;
+    }
+  }
+}
+
 // -----------------------------------------------------------------------------
 
 void VP8LHistogramAddSinglePixOrCopy(VP8LHistogram* const histo,
@@ -194,6 +208,7 @@ void VP8LHistogramAddSinglePixOrCopy(VP8LHistogram* const histo,
   } else if (PixOrCopyIsCacheIdx(v)) {
     const int literal_ix =
         NUM_LITERAL_CODES + NUM_LENGTH_CODES + PixOrCopyCacheIdx(v);
+    assert(histo->palette_code_bits_ != 0);
     ++histo->literal_[literal_ix];
   } else {
     int code, extra_bits;
@@ -447,7 +462,9 @@ static double HistogramAddEval(const VP8LHistogram* const a,
 static double HistogramAddThresh(const VP8LHistogram* const a,
                                  const VP8LHistogram* const b,
                                  double cost_threshold) {
-  double cost = -a->bit_cost_;
+  double cost;
+  assert(a != NULL && b != NULL);
+  cost = -a->bit_cost_;
   GetCombinedHistogramEntropy(a, b, cost_threshold, &cost);
   return cost;
 }
@@ -561,14 +578,17 @@ static void HistogramBuild(
 }
 
 // Copies the histograms and computes its bit_cost.
-static void HistogramCopyAndAnalyze(
-    VP8LHistogramSet* const orig_histo, VP8LHistogramSet* const image_histo) {
-  int i;
-  const int histo_size = orig_histo->size;
+static const uint16_t kInvalidHistogramSymbol = (uint16_t)(-1);
+static void HistogramCopyAndAnalyze(VP8LHistogramSet* const orig_histo,
+                                    VP8LHistogramSet* const image_histo,
+                                    int* const num_used,
+                                    uint16_t* const histogram_symbols) {
+  int i, cluster_id;
+  int num_used_orig = *num_used;
   VP8LHistogram** const orig_histograms = orig_histo->histograms;
   VP8LHistogram** const histograms = image_histo->histograms;
-  image_histo->size = 0;
-  for (i = 0; i < histo_size; ++i) {
+  assert(image_histo->max_size == orig_histo->max_size);
+  for (cluster_id = 0, i = 0; i < orig_histo->max_size; ++i) {
     VP8LHistogram* const histo = orig_histograms[i];
     UpdateHistogramCost(histo);
 
@@ -576,10 +596,19 @@ static void HistogramCopyAndAnalyze(
     // with no information (when they are skipped because of LZ77).
     if (!histo->is_used_[0] && !histo->is_used_[1] && !histo->is_used_[2]
         && !histo->is_used_[3] && !histo->is_used_[4]) {
-      continue;
+      // The first histogram is always used. If an histogram is empty, we set
+      // its id to be the same as the previous one: this will improve
+      // compressibility for later LZ77.
+      assert(i > 0);
+      HistogramSetRemoveHistogram(image_histo, i, num_used);
+      HistogramSetRemoveHistogram(orig_histo, i, &num_used_orig);
+      histogram_symbols[i] = kInvalidHistogramSymbol;
+    } else {
+      // Copy histograms from orig_histo[] to image_histo[].
+      HistogramCopy(histo, histograms[i]);
+      histogram_symbols[i] = cluster_id++;
+      assert(cluster_id <= image_histo->max_size);
     }
-    // Copy histograms from orig_histo[] to image_histo[].
-    HistogramCopy(histo, histograms[image_histo->size++]);
   }
 }
 
@@ -596,29 +625,33 @@ static void HistogramAnalyzeEntropyBin(VP8LHistogramSet* const image_histo,
 
   // Analyze the dominant (literal, red and blue) entropy costs.
   for (i = 0; i < histo_size; ++i) {
+    if (histograms[i] == NULL) continue;
     UpdateDominantCostRange(histograms[i], &cost_range);
   }
 
   // bin-hash histograms on three of the dominant (literal, red and blue)
   // symbol costs and store the resulting bin_id for each histogram.
   for (i = 0; i < histo_size; ++i) {
+    // bin_map[i] is not set to a special value as its use will later be guarded
+    // by another (histograms[i] == NULL).
+    if (histograms[i] == NULL) continue;
     bin_map[i] = GetHistoBinIndex(histograms[i], &cost_range, low_effort);
   }
 }
 
-// Compact image_histo[] by merging some histograms with same bin_id together if
-// it's advantageous.
+// Merges some histograms with same bin_id together if it's advantageous.
+// Sets the remaining histograms to NULL.
 static void HistogramCombineEntropyBin(VP8LHistogramSet* const image_histo,
+                                       int* num_used,
+                                       const uint16_t* const clusters,
+                                       uint16_t* const cluster_mappings,
                                        VP8LHistogram* cur_combo,
                                        const uint16_t* const bin_map,
-                                       int bin_map_size, int num_bins,
+                                       int num_bins,
                                        double combine_cost_factor,
                                        int low_effort) {
   VP8LHistogram** const histograms = image_histo->histograms;
   int idx;
-  // Work in-place: processed histograms are put at the beginning of
-  // image_histo[]. At the end, we just have to truncate the array.
-  int size = 0;
   struct {
     int16_t first;    // position of the histogram that accumulates all
                       // histograms with the same bin_id
@@ -631,16 +664,19 @@ static void HistogramCombineEntropyBin(VP8LHistogramSet* const image_histo,
     bin_info[idx].num_combine_failures = 0;
   }
 
-  for (idx = 0; idx < bin_map_size; ++idx) {
-    const int bin_id = bin_map[idx];
-    const int first = bin_info[bin_id].first;
-    assert(size <= idx);
+  // By default, a cluster matches itself.
+  for (idx = 0; idx < *num_used; ++idx) cluster_mappings[idx] = idx;
+  for (idx = 0; idx < image_histo->size; ++idx) {
+    int bin_id, first;
+    if (histograms[idx] == NULL) continue;
+    bin_id = bin_map[idx];
+    first = bin_info[bin_id].first;
     if (first == -1) {
-      // just move histogram #idx to its final position
-      histograms[size] = histograms[idx];
-      bin_info[bin_id].first = size++;
+      bin_info[bin_id].first = idx;
     } else if (low_effort) {
       HistogramAdd(histograms[idx], histograms[first], histograms[first]);
+      HistogramSetRemoveHistogram(image_histo, idx, num_used);
+      cluster_mappings[clusters[idx]] = clusters[first];
     } else {
       // try to merge #idx into #first (both share the same bin_id)
       const double bit_cost = histograms[idx]->bit_cost_;
@@ -663,19 +699,18 @@ static void HistogramCombineEntropyBin(VP8LHistogramSet* const image_histo,
             bin_info[bin_id].num_combine_failures >= max_combine_failures) {
           // move the (better) merged histogram to its final slot
           HistogramSwap(&cur_combo, &histograms[first]);
+          HistogramSetRemoveHistogram(image_histo, idx, num_used);
+          cluster_mappings[clusters[idx]] = clusters[first];
         } else {
-          histograms[size++] = histograms[idx];
           ++bin_info[bin_id].num_combine_failures;
         }
-      } else {
-        histograms[size++] = histograms[idx];
       }
     }
   }
-  image_histo->size = size;
   if (low_effort) {
     // for low_effort case, update the final cost when everything is merged
-    for (idx = 0; idx < size; ++idx) {
+    for (idx = 0; idx < image_histo->size; ++idx) {
+      if (histograms[idx] == NULL) continue;
       UpdateHistogramCost(histograms[idx]);
     }
   }
@@ -706,16 +741,9 @@ typedef struct {
   int max_size;
 } HistoQueue;
 
-static int HistoQueueInit(HistoQueue* const histo_queue, const int max_index) {
+static int HistoQueueInit(HistoQueue* const histo_queue, const int max_size) {
   histo_queue->size = 0;
-  // max_index^2 for the queue size is safe. If you look at
-  // HistogramCombineGreedy, and imagine that UpdateQueueFront always pushes
-  // data to the queue, you insert at most:
-  // - max_index*(max_index-1)/2 (the first two for loops)
-  // - max_index - 1 in the last for loop at the first iteration of the while
-  //   loop, max_index - 2 at the second iteration ... therefore
-  //   max_index*(max_index-1)/2 overall too
-  histo_queue->max_size = max_index * max_index;
+  histo_queue->max_size = max_size;
   // We allocate max_size + 1 because the last element at index "size" is
   // used as temporary data (and it could be up to max_size).
   histo_queue->queue = (HistogramPair*)WebPSafeMalloc(
@@ -778,6 +806,8 @@ static double HistoQueuePush(HistoQueue* const histo_queue,
   const VP8LHistogram* h2;
   HistogramPair pair;
 
+  // Stop here if the queue is full.
+  if (histo_queue->size == histo_queue->max_size) return 0.;
   assert(threshold <= 0.);
   if (idx1 > idx2) {
     const int tmp = idx2;
@@ -794,8 +824,6 @@ static double HistoQueuePush(HistoQueue* const histo_queue,
   // Do not even consider the pair if it does not improve the entropy.
   if (pair.cost_diff >= threshold) return 0.;
 
-  // We cannot add more elements than the capacity.
-  assert(histo_queue->size < histo_queue->max_size);
   histo_queue->queue[histo_queue->size++] = pair;
   HistoQueueUpdateHead(histo_queue, &histo_queue->queue[histo_queue->size - 1]);
 
@@ -806,42 +834,43 @@ static double HistoQueuePush(HistoQueue* const histo_queue,
 
 // Combines histograms by continuously choosing the one with the highest cost
 // reduction.
-static int HistogramCombineGreedy(VP8LHistogramSet* const image_histo) {
+static int HistogramCombineGreedy(VP8LHistogramSet* const image_histo,
+                                  int* const num_used) {
   int ok = 0;
-  int image_histo_size = image_histo->size;
+  const int image_histo_size = image_histo->size;
   int i, j;
   VP8LHistogram** const histograms = image_histo->histograms;
-  // Indexes of remaining histograms.
-  int* const clusters =
-      (int*)WebPSafeMalloc(image_histo_size, sizeof(*clusters));
   // Priority queue of histogram pairs.
   HistoQueue histo_queue;
 
-  if (!HistoQueueInit(&histo_queue, image_histo_size) || clusters == NULL) {
+  // image_histo_size^2 for the queue size is safe. If you look at
+  // HistogramCombineGreedy, and imagine that UpdateQueueFront always pushes
+  // data to the queue, you insert at most:
+  // - image_histo_size*(image_histo_size-1)/2 (the first two for loops)
+  // - image_histo_size - 1 in the last for loop at the first iteration of
+  //   the while loop, image_histo_size - 2 at the second iteration ...
+  //   therefore image_histo_size*(image_histo_size-1)/2 overall too
+  if (!HistoQueueInit(&histo_queue, image_histo_size * image_histo_size)) {
     goto End;
   }
 
   for (i = 0; i < image_histo_size; ++i) {
-    // Initialize clusters indexes.
-    clusters[i] = i;
+    if (image_histo->histograms[i] == NULL) continue;
     for (j = i + 1; j < image_histo_size; ++j) {
-      // Initialize positions array.
+      // Initialize queue.
+      if (image_histo->histograms[j] == NULL) continue;
       HistoQueuePush(&histo_queue, histograms, i, j, 0.);
     }
   }
 
-  while (image_histo_size > 1 && histo_queue.size > 0) {
+  while (histo_queue.size > 0) {
     const int idx1 = histo_queue.queue[0].idx1;
     const int idx2 = histo_queue.queue[0].idx2;
     HistogramAdd(histograms[idx2], histograms[idx1], histograms[idx1]);
     histograms[idx1]->bit_cost_ = histo_queue.queue[0].cost_combo;
+
     // Remove merged histogram.
-    for (i = 0; i + 1 < image_histo_size; ++i) {
-      if (clusters[i] >= idx2) {
-        clusters[i] = clusters[i + 1];
-      }
-    }
-    --image_histo_size;
+    HistogramSetRemoveHistogram(image_histo, idx2, num_used);
 
     // Remove pairs intersecting the just combined best pair.
     for (i = 0; i < histo_queue.size;) {
@@ -856,24 +885,15 @@ static int HistogramCombineGreedy(VP8LHistogramSet* const image_histo) {
     }
 
     // Push new pairs formed with combined histogram to the queue.
-    for (i = 0; i < image_histo_size; ++i) {
-      if (clusters[i] != idx1) {
-        HistoQueuePush(&histo_queue, histograms, idx1, clusters[i], 0.);
-      }
-    }
-  }
-  // Move remaining histograms to the beginning of the array.
-  for (i = 0; i < image_histo_size; ++i) {
-    if (i != clusters[i]) {  // swap the two histograms
-      HistogramSwap(&histograms[i], &histograms[clusters[i]]);
+    for (i = 0; i < image_histo->size; ++i) {
+      if (i == idx1 || image_histo->histograms[i] == NULL) continue;
+      HistoQueuePush(&histo_queue, image_histo->histograms, idx1, i, 0.);
     }
   }
 
-  image_histo->size = image_histo_size;
   ok = 1;
 
  End:
-  WebPSafeFree(clusters);
   HistoQueueClear(&histo_queue);
   return ok;
 }
@@ -881,47 +901,68 @@ static int HistogramCombineGreedy(VP8LHistogramSet* const image_histo) {
 // Perform histogram aggregation using a stochastic approach.
 // 'do_greedy' is set to 1 if a greedy approach needs to be performed
 // afterwards, 0 otherwise.
+static int PairComparison(const void* idx1, const void* idx2) {
+  // To be used with bsearch: <0 when *idx1<*idx2, >0 if >, 0 when ==.
+  return (*(int*) idx1 - *(int*) idx2);
+}
 static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
-                                      int min_cluster_size,
+                                      int* const num_used, int min_cluster_size,
                                       int* const do_greedy) {
-  int iter;
+  int j, iter;
   uint32_t seed = 1;
   int tries_with_no_success = 0;
-  int image_histo_size = image_histo->size;
-  const int outer_iters = image_histo_size;
+  const int outer_iters = *num_used;
   const int num_tries_no_success = outer_iters / 2;
   VP8LHistogram** const histograms = image_histo->histograms;
-  // Priority queue of histogram pairs. Its size of "kCostHeapSizeSqrt"^2
+  // Priority queue of histogram pairs. Its size of 'kHistoQueueSize'
   // impacts the quality of the compression and the speed: the smaller the
   // faster but the worse for the compression.
   HistoQueue histo_queue;
-  const int kHistoQueueSizeSqrt = 3;
+  const int kHistoQueueSize = 9;
   int ok = 0;
+  // mapping from an index in image_histo with no NULL histogram to the full
+  // blown image_histo.
+  int* mappings;
 
-  if (!HistoQueueInit(&histo_queue, kHistoQueueSizeSqrt)) {
-    goto End;
+  if (*num_used < min_cluster_size) {
+    *do_greedy = 1;
+    return 1;
   }
+
+  mappings = (int*) WebPSafeMalloc(*num_used, sizeof(*mappings));
+  if (mappings == NULL) return 0;
+  if (!HistoQueueInit(&histo_queue, kHistoQueueSize)) goto End;
+  // Fill the initial mapping.
+  for (j = 0, iter = 0; iter < image_histo->size; ++iter) {
+    if (histograms[iter] == NULL) continue;
+    mappings[j++] = iter;
+  }
+  assert(j == *num_used);
+
   // Collapse similar histograms in 'image_histo'.
-  ++min_cluster_size;
-  for (iter = 0; iter < outer_iters && image_histo_size >= min_cluster_size &&
-                 ++tries_with_no_success < num_tries_no_success;
+  for (iter = 0;
+       iter < outer_iters && *num_used >= min_cluster_size &&
+           ++tries_with_no_success < num_tries_no_success;
        ++iter) {
+    int* mapping_index;
     double best_cost =
         (histo_queue.size == 0) ? 0. : histo_queue.queue[0].cost_diff;
     int best_idx1 = -1, best_idx2 = 1;
-    int j;
-    const uint32_t rand_range = (image_histo_size - 1) * image_histo_size;
-    // image_histo_size / 2 was chosen empirically. Less means faster but worse
+    const uint32_t rand_range = (*num_used - 1) * (*num_used);
+    // (*num_used) / 2 was chosen empirically. Less means faster but worse
     // compression.
-    const int num_tries = image_histo_size / 2;
+    const int num_tries = (*num_used) / 2;
 
-    for (j = 0; j < num_tries; ++j) {
+    // Pick random samples.
+    for (j = 0; *num_used >= 2 && j < num_tries; ++j) {
       double curr_cost;
       // Choose two different histograms at random and try to combine them.
       const uint32_t tmp = MyRand(&seed) % rand_range;
-      const uint32_t idx1 = tmp / (image_histo_size - 1);
-      uint32_t idx2 = tmp % (image_histo_size - 1);
+      uint32_t idx1 = tmp / (*num_used - 1);
+      uint32_t idx2 = tmp % (*num_used - 1);
       if (idx2 >= idx1) ++idx2;
+      idx1 = mappings[idx1];
+      idx2 = mappings[idx2];
 
       // Calculate cost reduction on combination.
       curr_cost =
@@ -934,18 +975,21 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
     }
     if (histo_queue.size == 0) continue;
 
-    // Merge the two best histograms.
+    // Get the best histograms.
     best_idx1 = histo_queue.queue[0].idx1;
     best_idx2 = histo_queue.queue[0].idx2;
     assert(best_idx1 < best_idx2);
-    HistogramAddEval(histograms[best_idx1], histograms[best_idx2],
-                     histograms[best_idx1], 0);
-    // Swap the best_idx2 histogram with the last one (which is now unused).
-    --image_histo_size;
-    if (best_idx2 != image_histo_size) {
-      HistogramSwap(&histograms[image_histo_size], &histograms[best_idx2]);
-    }
-    histograms[image_histo_size] = NULL;
+    // Pop best_idx2 from mappings.
+    mapping_index = (int*) bsearch(&best_idx2, mappings, *num_used,
+                                   sizeof(best_idx2), &PairComparison);
+    assert(mapping_index != NULL);
+    memmove(mapping_index, mapping_index + 1, sizeof(*mapping_index) *
+        ((*num_used) - (mapping_index - mappings) - 1));
+    // Merge the histograms and remove best_idx2 from the queue.
+    HistogramAdd(histograms[best_idx2], histograms[best_idx1],
+                 histograms[best_idx1]);
+    histograms[best_idx1]->bit_cost_ = histo_queue.queue[0].cost_combo;
+    HistogramSetRemoveHistogram(image_histo, best_idx2, num_used);
     // Parse the queue and update each pair that deals with best_idx1,
     // best_idx2 or image_histo_size.
     for (j = 0; j < histo_queue.size;) {
@@ -968,12 +1012,6 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
         p->idx2 = best_idx1;
         do_eval = 1;
       }
-      if (p->idx2 == image_histo_size) {
-        // No need to re-evaluate here as it does not involve a pair
-        // containing best_idx1 or best_idx2.
-        p->idx2 = best_idx2;
-      }
-      assert(p->idx2 < image_histo_size);
       // Make sure the index order is respected.
       if (p->idx1 > p->idx2) {
         const int tmp = p->idx2;
@@ -991,15 +1029,14 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
       HistoQueueUpdateHead(&histo_queue, p);
       ++j;
     }
-
     tries_with_no_success = 0;
   }
-  image_histo->size = image_histo_size;
-  *do_greedy = (image_histo->size <= min_cluster_size);
+  *do_greedy = (*num_used <= min_cluster_size);
   ok = 1;
 
 End:
   HistoQueueClear(&histo_queue);
+  WebPSafeFree(mappings);
   return ok;
 }
 
@@ -1007,23 +1044,29 @@ End:
 // Histogram refinement
 
 // Find the best 'out' histogram for each of the 'in' histograms.
+// At call-time, 'out' contains the histograms of the clusters.
 // Note: we assume that out[]->bit_cost_ is already up-to-date.
 static void HistogramRemap(const VP8LHistogramSet* const in,
-                           const VP8LHistogramSet* const out,
+                           VP8LHistogramSet* const out,
                            uint16_t* const symbols) {
   int i;
   VP8LHistogram** const in_histo = in->histograms;
   VP8LHistogram** const out_histo = out->histograms;
-  const int in_size = in->size;
+  const int in_size = out->max_size;
   const int out_size = out->size;
   if (out_size > 1) {
     for (i = 0; i < in_size; ++i) {
       int best_out = 0;
       double best_bits = MAX_COST;
       int k;
+      if (in_histo[i] == NULL) {
+        // Arbitrarily set to the previous value if unused to help future LZ77.
+        symbols[i] = symbols[i - 1];
+        continue;
+      }
       for (k = 0; k < out_size; ++k) {
-        const double cur_bits =
-            HistogramAddThresh(out_histo[k], in_histo[i], best_bits);
+        double cur_bits;
+        cur_bits = HistogramAddThresh(out_histo[k], in_histo[i], best_bits);
         if (k == 0 || cur_bits < best_bits) {
           best_bits = cur_bits;
           best_out = k;
@@ -1039,12 +1082,13 @@ static void HistogramRemap(const VP8LHistogramSet* const in,
   }
 
   // Recompute each out based on raw and symbols.
-  for (i = 0; i < out_size; ++i) {
-    HistogramClear(out_histo[i]);
-  }
+  VP8LHistogramSetClear(out);
+  out->size = out_size;
 
   for (i = 0; i < in_size; ++i) {
-    const int idx = symbols[i];
+    int idx;
+    if (in_histo[i] == NULL) continue;
+    idx = symbols[i];
     HistogramAdd(in_histo[i], out_histo[idx], out_histo[idx]);
   }
 }
@@ -1060,6 +1104,70 @@ static double GetCombineCostFactor(int histo_size, int quality) {
   return combine_cost_factor;
 }
 
+// Given a HistogramSet 'set', the mapping of clusters 'cluster_mapping' and the
+// current assignment of the cells in 'symbols', merge the clusters and
+// assign the smallest possible clusters values.
+static void OptimizeHistogramSymbols(const VP8LHistogramSet* const set,
+                                     uint16_t* const cluster_mappings,
+                                     int num_clusters,
+                                     uint16_t* const cluster_mappings_tmp,
+                                     uint16_t* const symbols) {
+  int i, cluster_max;
+  int do_continue = 1;
+  // First, assign the lowest cluster to each pixel.
+  while (do_continue) {
+    do_continue = 0;
+    for (i = 0; i < num_clusters; ++i) {
+      int k;
+      k = cluster_mappings[i];
+      while (k != cluster_mappings[k]) {
+        cluster_mappings[k] = cluster_mappings[cluster_mappings[k]];
+        k = cluster_mappings[k];
+      }
+      if (k != cluster_mappings[i]) {
+        do_continue = 1;
+        cluster_mappings[i] = k;
+      }
+    }
+  }
+  // Create a mapping from a cluster id to its minimal version.
+  cluster_max = 0;
+  memset(cluster_mappings_tmp, 0,
+         set->max_size * sizeof(*cluster_mappings_tmp));
+  assert(cluster_mappings[0] == 0);
+  // Re-map the ids.
+  for (i = 0; i < set->max_size; ++i) {
+    int cluster;
+    if (symbols[i] == kInvalidHistogramSymbol) continue;
+    cluster = cluster_mappings[symbols[i]];
+    assert(symbols[i] < num_clusters);
+    if (cluster > 0 && cluster_mappings_tmp[cluster] == 0) {
+      ++cluster_max;
+      cluster_mappings_tmp[cluster] = cluster_max;
+    }
+    symbols[i] = cluster_mappings_tmp[cluster];
+  }
+
+  // Make sure all cluster values are used.
+  cluster_max = 0;
+  for (i = 0; i < set->max_size; ++i) {
+    if (symbols[i] == kInvalidHistogramSymbol) continue;
+    if (symbols[i] <= cluster_max) continue;
+    ++cluster_max;
+    assert(symbols[i] == cluster_max);
+  }
+}
+
+static void RemoveEmptyHistograms(VP8LHistogramSet* const image_histo) {
+  uint32_t size;
+  int i;
+  for (i = 0, size = 0; i < image_histo->size; ++i) {
+    if (image_histo->histograms[i] == NULL) continue;
+    image_histo->histograms[size++] = image_histo->histograms[i];
+  }
+  image_histo->size = size;
+}
+
 int VP8LGetHistoImageSymbols(int xsize, int ysize,
                              const VP8LBackwardRefs* const refs,
                              int quality, int low_effort,
@@ -1078,27 +1186,36 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
   // maximum quality q==100 (to preserve the compression gains at that level).
   const int entropy_combine_num_bins = low_effort ? NUM_PARTITIONS : BIN_SIZE;
   int entropy_combine;
-
-  if (orig_histo == NULL) goto Error;
+  uint16_t* const map_tmp =
+      WebPSafeMalloc(2 * image_histo_raw_size, sizeof(map_tmp));
+  uint16_t* const cluster_mappings = map_tmp + image_histo_raw_size;
+  int num_used = image_histo_raw_size;
+  if (orig_histo == NULL || map_tmp == NULL) goto Error;
 
   // Construct the histograms from backward references.
   HistogramBuild(xsize, histo_bits, refs, orig_histo);
   // Copies the histograms and computes its bit_cost.
-  HistogramCopyAndAnalyze(orig_histo, image_histo);
+  // histogram_symbols is optimized
+  HistogramCopyAndAnalyze(orig_histo, image_histo, &num_used,
+                          histogram_symbols);
+
   entropy_combine =
-      (image_histo->size > entropy_combine_num_bins * 2) && (quality < 100);
+      (num_used > entropy_combine_num_bins * 2) && (quality < 100);
+
   if (entropy_combine) {
-    const int bin_map_size = image_histo->size;
-    // Reuse histogram_symbols storage. By definition, it's guaranteed to be ok.
-    uint16_t* const bin_map = histogram_symbols;
+    uint16_t* const bin_map = map_tmp;
     const double combine_cost_factor =
         GetCombineCostFactor(image_histo_raw_size, quality);
+    const uint32_t num_clusters = num_used;
 
     HistogramAnalyzeEntropyBin(image_histo, bin_map, low_effort);
     // Collapse histograms with similar entropy.
-    HistogramCombineEntropyBin(image_histo, tmp_histo, bin_map, bin_map_size,
+    HistogramCombineEntropyBin(image_histo, &num_used, histogram_symbols,
+                               cluster_mappings, tmp_histo, bin_map,
                                entropy_combine_num_bins, combine_cost_factor,
                                low_effort);
+    OptimizeHistogramSymbols(image_histo, cluster_mappings, num_clusters,
+                             map_tmp, histogram_symbols);
   }
 
   // Don't combine the histograms using stochastic and greedy heuristics for
@@ -1108,21 +1225,26 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
     // cubic ramp between 1 and MAX_HISTO_GREEDY:
     const int threshold_size = (int)(1 + (x * x * x) * (MAX_HISTO_GREEDY - 1));
     int do_greedy;
-    if (!HistogramCombineStochastic(image_histo, threshold_size, &do_greedy)) {
+    if (!HistogramCombineStochastic(image_histo, &num_used, threshold_size,
+                                    &do_greedy)) {
       goto Error;
     }
-    if (do_greedy && !HistogramCombineGreedy(image_histo)) {
-      goto Error;
+    if (do_greedy) {
+      RemoveEmptyHistograms(image_histo);
+      if (!HistogramCombineGreedy(image_histo, &num_used)) {
+        goto Error;
+      }
     }
   }
 
-  // TODO(vrabaud): Optimize HistogramRemap for low-effort compression mode.
   // Find the optimal map from original histograms to the final ones.
+  RemoveEmptyHistograms(image_histo);
   HistogramRemap(orig_histo, image_histo, histogram_symbols);
 
   ok = 1;
 
  Error:
   VP8LFreeHistogramSet(orig_histo);
+  WebPSafeFree(map_tmp);
   return ok;
 }