Merge tag 'v0.5.1'

libwebp-0.5.1
- 6/14/2016: version 0.5.1
  This is a binary compatible release.
  * miscellaneous bug fixes (issues #280, #289)
  * reverted alpha plane encoding with color cache for compatibility with
    libwebp 0.4.0->0.4.3 (issues #291, #298)
  * lossless encoding performance improvements
  * memory reduction in both lossless encoding and decoding
  * force mux output to be in the extended format (VP8X) when undefined chunks
    are present (issue #294)
  * gradle, cmake build support
  * workaround for compiler bug causing 64-bit decode failures on android
    devices using clang-3.8 in the r11c NDK
  * various WebPAnimEncoder improvements

* tag 'v0.5.1': (30 commits)
  update ChangeLog
  Clarify the expected 'config' lifespan in WebPIDecode()
  update ChangeLog
  Fix corner case in CostManagerInit.
  gif2webp: normalize the number of .'s in the help message
  vwebp: normalize the number of .'s in the help message
  cwebp: normalize the number of .'s in the help message
  fix rescaling bug: alpha plane wasn't filled with 0xff
  Improve lossless compression.
  'our bug tracker' -> 'the bug tracker'
  normalize the number of .'s in the help message
  pngdec,ReadFunc: throw an error on invalid read
  decode.h,WebPGetInfo: normalize function comment
  Inline GetResidual for speed.
  Speed-up uniform-region processing.
  free -> WebPSafeFree()
  DecodeImageData(): change the incorrect assert
  Fix a boundary case in BackwardReferencesHashChainDistanceOnly.
  Make sure to consider small distances in LZ77.
  add some asserts to delimit the perimeter of CostManager's operation
  ...

Change-Id: I44cee79fddd43527062ea9d83be67da42484ebfc

.mailmap

@@ -8,3 +8,4 @@ Vikas Arora <vikasa@google.com>
 <slobodan.prijic@imgtec.com> <Slobodan.Prijic@imgtec.com>
 <vrabaud@google.com> <vincent.rabaud@gmail.com>
 Tamar Levy <tamar.levy@intel.com>
+<qrczak@google.com> <qrczak>

AUTHORS

@@ -10,10 +10,13 @@ Contributors:
 - Lode Vandevenne (lode at google dot com)
 - Lou Quillio (louquillio at google dot com)
 - Mans Rullgard (mans at mansr dot com)
+- Marcin Kowalczyk (qrczak at google dot com)
 - Martin Olsson (mnemo at minimum dot se)
 - Mikołaj Zalewski (mikolajz at google dot com)
 - Mislav Bradac (mislavm at google dot com)
+- Nico Weber (thakis at chromium dot org)
 - Noel Chromium (noel at chromium dot org)
+- Parag Salasakar (img dot mips1 at gmail dot com)
 - Pascal Massimino (pascal dot massimino at gmail dot com)
 - Paweł Hajdan, Jr (phajdan dot jr at chromium dot org)
 - Pierre Joye (pierre dot php at gmail dot com)

ChangeLog

@@ -1,3 +1,179 @@
+deb54d9 Clarify the expected 'config' lifespan in WebPIDecode()
+c7e2d24 update ChangeLog (tag: v0.5.1-rc5)
+c7eb06f Fix corner case in CostManagerInit.
+ab7937a gif2webp: normalize the number of .'s in the help message
+3cdec84 vwebp: normalize the number of .'s in the help message
+bdf6241 cwebp: normalize the number of .'s in the help message
+06a38c7 fix rescaling bug: alpha plane wasn't filled with 0xff
+319e37b Improve lossless compression.
+447adbc 'our bug tracker' -> 'the bug tracker'
+97b9e64 normalize the number of .'s in the help message
+bb50bf4 pngdec,ReadFunc: throw an error on invalid read
+38063af decode.h,WebPGetInfo: normalize function comment
+9e8e1b7 Inline GetResidual for speed.
+7d58d1b Speed-up uniform-region processing.
+23e29cb Merge "Fix a boundary case in BackwardReferencesHashChainDistanceOnly." into 0.5.1
+0bb23b2 free -> WebPSafeFree()
+e7b9177 Merge "DecodeImageData(): change the incorrect assert" into 0.5.1
+2abfa54 DecodeImageData(): change the incorrect assert
+5a48fcd Merge "configure: test for -Wfloat-conversion"
+0174d18 Fix a boundary case in BackwardReferencesHashChainDistanceOnly.
+6a9c262 Merge "Added MSA optimized transform functions"
+cfbcc5e Make sure to consider small distances in LZ77.
+5e60c42 Added MSA optimized transform functions
+3dc28d7 configure: test for -Wfloat-conversion
+f2a0946 add some asserts to delimit the perimeter of CostManager's operation
+9a583c6 fix invalid-write bug for alpha-decoding
+f66512d make gradlew executable
+6fda58f backward_references: quiet double->int warning
+a48cc9d Merge "Fix a compression regression for images with long uniform regions." into 0.5.1
+cc2720c Merge "Revert an LZ77 boundary constant." into 0.5.1
+059aab4 Fix a compression regression for images with long uniform regions.
+b0c7e49 Check more backward matches with higher quality.
+a361151 Revert an LZ77 boundary constant.
+8190374 README: fix typo
+7551db4 update NEWS
+0fb2269 bump version to 0.5.1
+f453761 update AUTHORS & .mailmap
+3259571 Refactor GetColorPalette method.
+1df5e26 avoid using tmp histogram in PreparePair()
+7685123 fix comment typos
+a246b92 Speedup backward references.
+76d73f1 Merge "CostManager: introduce a free-list of ~10 intervals"
+eab39d8 CostManager: introduce a free-list of ~10 intervals
+4c59aac Merge "mips msa webp configuration"
+043c33f Merge "Improve speed and compression in backward reference for lossless."
+71be9b8 Merge "clarify variable names in HistogramRemap()"
+0ba7fd7 Improve speed and compression in backward reference for lossless.
+0481d42 CostManager: cache one interval and re-use it when possible
+41b7e6b Merge "histogram: fix bin calculation"
+96c3d62 histogram: fix bin calculation
+fe9e31e clarify variable names in HistogramRemap()
+ce3c824 disable near-lossless quantization if palette is used
+e11da08 mips msa webp configuration
+5f8f998 mux: Presence of unknown chunks should trigger VP8X chunk output.
+cadec0b Merge "Sync mips32 and dsp_r2 YUV->RGB code with C verison"
+d963775 Compute the hash chain once and for all for lossless compression.
+50a4866 Sync mips32 and dsp_r2 YUV->RGB code with C verison
+eee788e Merge "introduce a common signature for all image reader function"
+d77b877 introduce a common signature for all image reader function
+ca8d951 remove some obsolete TODOs
+ae2a722 collect all decoding utilities from examples/ in libexampledec.a
+0b8ae85 Merge "Move DitherCombine8x8 to dsp/dec.c"
+77cad88 Merge "ReadWebP: avoid conversion to ARGB if final format is YUVA"
+ab8d669 ReadWebP: avoid conversion to ARGB if final format is YUVA
+f8b7ce9 Merge "test pointer to NULL explicitly"
+5df6f21 test pointer to NULL explicitly
+77f21c9 Move DitherCombine8x8 to dsp/dec.c
+c9e6d86 Add gradle support
+c65f41e Revert "Add gradle support"
+bf731ed Add gradle support
+08333b8 WebPAnimEncoder: Detect when canvas is modified, restore only when needed.
+0209d7e Merge "speed-up MapToPalette() with binary search"
+fdd29a3 speed-up MapToPalette() with binary search
+cf4a651 Revert "Refactor GetColorPalette method."
+0a27aca Merge changes Idfa8ce83,I19adc9c4
+f25c440 WebPAnimEncoder: Restore original canvas between multiple encodes.
+169004b Refactor GetColorPalette method.
+576362a VP8LDoFillBitWindow: support big-endian in fast path
+ac49e4e bit_reader.c: s/VP8L_USE_UNALIGNED_LOAD/VP8L_USE_FAST_LOAD/
+d39ceb5 VP8LDoFillBitWindow: remove stale TODO
+2ec2de1 Merge "Speed-up BackwardReferencesHashChainDistanceOnly."
+3e023c1 Speed-up BackwardReferencesHashChainDistanceOnly.
+f2e1efb Improve near lossless compression when a prediction filter is used.
+e15afbc dsp.h: fix ubsan macro name
+e53c9cc dsp.h: add WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
+af81fdb utils.h: quiet -fsanitize=undefined warnings
+ea0be35 dsp.h: remove utils.h include
+cd276ae utils/*.c: ../utils/utils.h -> ./utils.h
+c892713 utils/Makefile.am: add some missing headers
+ea24e02 Merge "dsp.h: add WEBP_UBSAN_IGNORE_UNDEF"
+369e264 dsp.h: add WEBP_UBSAN_IGNORE_UNDEF
+0d020a7 Merge "add runtime NEON detection"
+5ee2136 Merge "add VP8LAddPixels() to lossless.h"
+47435a6 add VP8LAddPixels() to lossless.h
+8fa6ac6 remove two ubsan warnings
+74fb56f add runtime NEON detection
+4154a83 MIPS update to new Unfilter API
+c80b9fc Merge "cherry-pick decoder fix for 64-bit android devices"
+6235147 cherry-pick decoder fix for 64-bit android devices
+d41b8c4 configure: test for -Wformat-* w/-Wformat present
+5f95589 Fix WEBP_ALIGN in case the argument is a pointer to a type larger than a byte.
+2309fd5 replace num_parts_ by num_parts_minus_one_ (unsigned)
+9629f4b SimplifySegments: quiet -Warray-bounds warning
+de47492 Merge "update the Unfilter API in dsp to process one row independently"
+2102ccd update the Unfilter API in dsp to process one row independently
+e3912d5 WebPAnimEncoder: Restore canvas before evaluating blending possibility.
+6e12e1e WebPAnimEncoder: Fix for single-frame optimization.
+602f344 Merge changes I1d03acac,Ifcb64219
+95ecccf only apply color-mapping for alpha on the cropped area
+47dd070 anim_diff: Add an experimental option for max inter-frame diff.
+aa809cf only allocate alpha_plane_ up to crop_bottom row
+31f2b8d WebPAnimEncoder: FlattenSimilarPixels(): look for similar
+774dfbd perform alpha filtering within the decoding loop
+a4cae68 lossless decoding: only process decoded row up to last_row
+238cdcd Only call WebPDequantizeLevels() on cropped area
+cf6c713 alpha: preparatory cleanup
+b95ac0a Merge "VP8GetHeaders(): initialize VP8Io with sane value for crop/scale dimensions"
+8923139 VP8GetHeaders(): initialize VP8Io with sane value for crop/scale dimensions
+5828e19 use_8b_decode -> use_8b_decode_
+8dca024 fix bug in alpha.c that was triggering a memory error in incremental mode
+9a950c5 WebPAnimEncoder: Disable filtering when blending is used with lossy encoding.
+eb42390 WebPAnimEncoder: choose max diff for framerect based on quality.
+ff0a94b WebPAnimEncoder lossy: ignore small pixel differences for frame rectangles.
+f804008 gif2webp: Remove the 'prev_to_prev_canvas' buffer.
+6d8c07d Merge "WebPDequantizeLevels(): use stride in CountLevels()"
+d96fe5e WebPDequantizeLevels(): use stride in CountLevels()
+ec1b240 WebPPictureImport*: check output pointer
+c076876 Merge "Revert "Re-enable encoding of alpha plane with color cache for next release.""
+41f14bc WebPPictureImport*: check src pointer
+64eed38 Pass stride parameter to WebPDequantizeLevels()
+97934e2 Revert "Re-enable encoding of alpha plane with color cache for next release."
+e88c4ca fix -m 2 mode-cost evaluation (causing partition0 overflow)
+4562e83 Merge "add extra meaning to WebPDecBuffer::is_external_memory"
+abdb109 add extra meaning to WebPDecBuffer::is_external_memory
+875aec7 enc_neon,cosmetics: break long comment
+71e856c GetMBSSIM,cosmetics: fix alignment
+a90edff fix missing 'extern' for SSIM function in dsp/
+423ecaf move some SSIM-accumulation function for dsp/
+f08e662 Merge "Fix FindClosestDiscretized in near lossless:"
+0d40cc5 enc_neon,Disto4x4: remove an unnecessary transpose
+e8feb20 Fix FindClosestDiscretized in near lossless:
+8200643 anim_util: quiet static analysis warning
+a6f23c4 Merge "AnimEncoder: Support progress hook and user data."
+a519377 Merge "Near lossless feature: fix some comments."
+da98d31 AnimEncoder: Support progress hook and user data.
+3335713 Near lossless feature: fix some comments.
+0beed01 cosmetics: fix indent after 2f5e898
+6753f35 Merge "FTransformWHT optimization."
+6583bb1 Improve SSE4.1 implementation of TTransform.
+7561d0c FTransformWHT optimization.
+7ccdb73 fix indentation after patch #328220
+6ec0d2a clarify the logic of the error path when decoding fails.
+8aa352b Merge "Remove an unnecessary transposition in TTransform."
+db86088 Merge "remove useless #include"
+9960c31 Remove an unnecessary transposition in TTransform.
+6e36b51 Small speedup in FTransform.
+9dbd4aa Merge "fix C and SIMD flags completion."
+e60853e Add missing common_sse2.h file to makefile.unix
+696eb2b fix C and SIMD flags completion.
+2b4fe33 Merge "fix multiple allocation for transform buffer"
+2f5e898 fix multiple allocation for transform buffer
+bf2b4f1 Regroup common SSE code + optimization.
+4ed650a force "-pass 6" if -psnr or -size is used but -pass isn't.
+3ef1ce9 yuv_sse2: fix -Wconstant-conversion warning
+a7a03e9 Merge changes I4852d18f,I51ccb85d
+5e122bd gif2webp: set enc_options.verbose = 0 w/-quiet
+ab3c258 anim_encode,DefaultEncoderOptions: init verbose
+8f0dee7 Merge "configure: fix builtin detection w/-Werror"
+4a7b85a cmake: fix builtin detection w/-Werror
+b74657f configure: fix builtin detection w/-Werror
+3661b98 Add a CMakeLists.txt
+75f4af4 remove useless #include
+6c1d763 avoid Yoda style for comparison
+8ce975a SSE optimization for vector mismatch.
+7db5383 Merge tag 'v0.5.0'
+37f0494 update ChangeLog (tag: v0.5.0-rc1, tag: v0.5.0, origin/0.5.0, 0.5.0)
 7e7b6cc faster rgb565/rgb4444/argb output
 4c7f565 update NEWS
 1f62b6b update AUTHORS
@@ -140,7 +316,7 @@ f7c507a Merge "remove unnecessary #include "yuv.h""
 7861578 for ReadXXXX() image-readers, use the value of pic->use_argb
 14e4043 remove unnecessary #include "yuv.h"
 469ba2c vwebp: fix incorrect clipping w/NO_BLEND
-4b9186b update issue tracker url (master)
+4b9186b update issue tracker url
 d64d376 change WEBP_ALIGN_CST value to 31
 f717b82 vp8l.c, cosmetics: fix indent after 95509f9
 927ccdc Merge "fix alignment of allocated memory in AllocateTransformBuffer"

NEWS

@@ -1,3 +1,17 @@
+- 6/14/2016: version 0.5.1
+  This is a binary compatible release.
+  * miscellaneous bug fixes (issues #280, #289)
+  * reverted alpha plane encoding with color cache for compatibility with
+    libwebp 0.4.0->0.4.3 (issues #291, #298)
+  * lossless encoding performance improvements
+  * memory reduction in both lossless encoding and decoding
+  * force mux output to be in the extended format (VP8X) when undefined chunks
+    are present (issue #294)
+  * gradle, cmake build support
+  * workaround for compiler bug causing 64-bit decode failures on android
+    devices using clang-3.8 in the r11c NDK
+  * various WebPAnimEncoder improvements
+
 - 12/17/2015: version 0.5.0
   * miscellaneous bug & build fixes (issues #234, #258, #274, #275, #278)
   * encoder & decoder speed-ups on x86/ARM/MIPS for lossy & lossless

README

@@ -4,7 +4,7 @@
           \__\__/\____/\_____/__/ ____  ___
                 / _/ /    \    \ /  _ \/ _/
                /  \_/   / /   \ \   __/  \__
-               \____/____/\_____/_____/____/v0.5.0
+               \____/____/\_____/_____/____/v0.5.1
 
 Description:
 ============
@@ -20,7 +20,7 @@ https://chromium.googlesource.com/webm/libwebp
 
 It is released under the same license as the WebM project.
 See http://www.webmproject.org/license/software/ or the
-file "COPYING" file for details. An additional intellectual
+"COPYING" file for details. An additional intellectual
 property rights grant can be found in the file PATENTS.
 
 Building:
@@ -218,8 +218,8 @@ If input size (-s) for an image is not specified, it is
 assumed to be a PNG, JPEG, TIFF or WebP file.
 
 Options:
-  -h / -help  ............ short help
+  -h / -help ............. short help
-  -H / -longhelp  ........ long help
+  -H / -longhelp ......... long help
   -q <float> ............. quality factor (0:small..100:big)
   -alpha_q <int> ......... transparency-compression quality (0..100)
   -preset <string> ....... preset setting, one of:
@@ -341,7 +341,7 @@ Use following options to convert into alternate image formats:
   -yuv ......... save the raw YUV samples in flat layout
 
  Other options are:
-  -version  .... print version number and exit
+  -version ..... print version number and exit
   -nofancy ..... don't use the fancy YUV420 upscaler
   -nofilter .... disable in-loop filtering
   -nodither .... disable dithering
@@ -353,8 +353,8 @@ Use following options to convert into alternate image formats:
   -flip ........ flip the output vertically
   -alpha ....... only save the alpha plane
   -incremental . use incremental decoding (useful for tests)
-  -h     ....... this help message
+  -h ........... this help message
-  -v     ....... verbose (e.g. print encoding/decoding times)
+  -v ........... verbose (e.g. print encoding/decoding times)
   -quiet ....... quiet mode, don't print anything
   -noasm ....... disable all assembly optimizations
 
@@ -370,7 +370,7 @@ Usage: vwebp in_file [options]
 
 Decodes the WebP image file and visualize it using OpenGL
 Options are:
-  -version  .... print version number and exit
+  -version ..... print version number and exit
   -noicc ....... don't use the icc profile if present
   -nofancy ..... don't use the fancy YUV420 upscaler
   -nofilter .... disable in-loop filtering
@@ -378,7 +378,7 @@ Options are:
   -noalphadither disable alpha plane dithering
   -mt .......... use multi-threading
   -info ........ print info
-  -h     ....... this help message
+  -h ........... this help message
 
 Keyboard shortcuts:
   'c' ................ toggle use of color profile
@@ -420,7 +420,7 @@ vwebp.
 Usage:
  gif2webp [options] gif_file -o webp_file
 Options:
-  -h / -help  ............ this help
+  -h / -help ............. this help
   -lossy ................. encode image using lossy compression
   -mixed ................. for each frame in the image, pick lossy
                            or lossless compression heuristically
@@ -704,7 +704,7 @@ an otherwise too-large picture. Some CPU can be saved too, incidentally.
 Bugs:
 =====
 
-Please report all bugs to our issue tracker:
+Please report all bugs to the issue tracker:
     https://bugs.chromium.org/p/webp
 Patches welcome! See this page to get started:
     http://www.webmproject.org/code/contribute/submitting-patches/

README.mux

@@ -1,7 +1,7 @@
           __   __  ____  ____  ____  __ __  _     __ __
          /  \\/  \/  _ \/  _ \/  _ \/  \  \/ \___/_ / _\
          \       /   __/  _  \   __/      /  /  (_/  /__
-          \__\__/\_____/_____/__/  \__//_/\_____/__/___/v0.3.0
+          \__\__/\_____/_____/__/  \__//_/\_____/__/___/v0.3.1
 
 
 Description:
@@ -198,7 +198,7 @@ For a detailed AnimEncoder API reference, please refer to the header file
 Bugs:
 =====
 
-Please report all bugs to our issue tracker:
+Please report all bugs to the issue tracker:
     https://bugs.chromium.org/p/webp
 Patches welcome! See this page to get started:
     http://www.webmproject.org/code/contribute/submitting-patches/

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([libwebp], [0.5.0],
+AC_INIT([libwebp], [0.5.1],
         [https://bugs.chromium.org/p/webp],,
         [http://developers.google.com/speed/webp])
 AC_CANONICAL_HOST

examples/cwebp.c

@@ -514,8 +514,8 @@ static void HelpLong(void) {
   printf("Windows builds can take as input any of the files handled by WIC.\n");
 #endif
   printf("\nOptions:\n");
-  printf("  -h / -help  ............ short help\n");
+  printf("  -h / -help ............. short help\n");
-  printf("  -H / -longhelp  ........ long help\n");
+  printf("  -H / -longhelp ......... long help\n");
   printf("  -q <float> ............. quality factor (0:small..100:big)\n");
   printf("  -alpha_q <int> ......... transparency-compression quality "
          "(0..100)\n");

examples/dwebp.c

@@ -572,7 +572,7 @@ static void Help(void) {
          "  -yuv ......... save the raw YUV samples in flat layout\n"
          "\n"
          " Other options are:\n"
-         "  -version  .... print version number and exit\n"
+         "  -version ..... print version number and exit\n"
          "  -nofancy ..... don't use the fancy YUV420 upscaler\n"
          "  -nofilter .... disable in-loop filtering\n"
          "  -nodither .... disable dithering\n"
@@ -584,8 +584,8 @@ static void Help(void) {
          "  -flip ........ flip the output vertically\n"
          "  -alpha ....... only save the alpha plane\n"
          "  -incremental . use incremental decoding (useful for tests)\n"
-         "  -h     ....... this help message\n"
+         "  -h ........... this help message\n"
-         "  -v     ....... verbose (e.g. print encoding/decoding times)\n"
+         "  -v ........... verbose (e.g. print encoding/decoding times)\n"
          "  -quiet ....... quiet mode, don't print anything\n"
 #ifndef WEBP_DLL
          "  -noasm ....... disable all assembly optimizations\n"

examples/gif2webp.c

@@ -55,7 +55,7 @@ static void Help(void) {
   printf("Usage:\n");
   printf(" gif2webp [options] gif_file -o webp_file\n");
   printf("Options:\n");
-  printf("  -h / -help  ............ this help\n");
+  printf("  -h / -help ............. this help\n");
   printf("  -lossy ................. encode image using lossy compression\n");
   printf("  -mixed ................. for each frame in the image, pick lossy\n"
          "                           or lossless compression heuristically\n");

examples/pngdec.c

@@ -18,7 +18,6 @@
 #include <stdio.h>
 
 #ifdef WEBP_HAVE_PNG
-#include <assert.h>
 #include <png.h>
 #include <setjmp.h>   // note: this must be included *after* png.h
 #include <stdlib.h>
@@ -198,7 +197,9 @@ typedef struct {
 
 static void ReadFunc(png_structp png_ptr, png_bytep data, png_size_t length) {
   PNGReadContext* const ctx = (PNGReadContext*)png_get_io_ptr(png_ptr);
-  assert(ctx->offset + length <= ctx->data_size);
+  if (ctx->data_size - ctx->offset < length) {
+    png_error(png_ptr, "ReadFunc: invalid read length (overflow)!");
+  }
   memcpy(data, ctx->data + ctx->offset, length);
   ctx->offset += length;
 }

examples/vwebp.c

@@ -387,7 +387,7 @@ static void Help(void) {
   printf("Usage: vwebp in_file [options]\n\n"
          "Decodes the WebP image file and visualize it using OpenGL\n"
          "Options are:\n"
-         "  -version  .... print version number and exit\n"
+         "  -version ..... print version number and exit\n"
          "  -noicc ....... don't use the icc profile if present\n"
          "  -nofancy ..... don't use the fancy YUV420 upscaler\n"
          "  -nofilter .... disable in-loop filtering\n"
@@ -395,7 +395,7 @@ static void Help(void) {
          "  -noalphadither disable alpha plane dithering\n"
          "  -mt .......... use multi-threading\n"
          "  -info ........ print info\n"
-         "  -h     ....... this help message\n"
+         "  -h ........... this help message\n"
          "\n"
          "Keyboard shortcuts:\n"
          "  'c' ................ toggle use of color profile\n"

man/cwebp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH CWEBP 1 "February 11, 2016"
+.TH CWEBP 1 "June 23, 2016"
 .SH NAME
 cwebp \- compress an image file to a WebP file
 .SH SYNOPSIS
@@ -281,7 +281,7 @@ Note: each input format may not support all combinations.
 Disable all assembly optimizations.
 
 .SH BUGS
-Please report all bugs to our issue tracker:
+Please report all bugs to the issue tracker:
 https://bugs.chromium.org/p/webp
 .br
 Patches welcome! See this page to get started:

man/dwebp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH DWEBP 1 "December 11, 2015"
+.TH DWEBP 1 "June 23, 2016"
 .SH NAME
 dwebp \- decompress a WebP file to an image file
 .SH SYNOPSIS
@@ -108,7 +108,7 @@ Print extra information (decoding time in particular).
 Disable all assembly optimizations.
 
 .SH BUGS
-Please report all bugs to our issue tracker:
+Please report all bugs to the issue tracker:
 https://bugs.chromium.org/p/webp
 .br
 Patches welcome! See this page to get started:

man/gif2webp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH GIF2WEBP 1 "December 11, 2015"
+.TH GIF2WEBP 1 "June 23, 2016"
 .SH NAME
 gif2webp \- Convert a GIF image to WebP
 .SH SYNOPSIS
@@ -117,7 +117,7 @@ Print extra information.
 Do not print anything.
 
 .SH BUGS
-Please report all bugs to our issue tracker:
+Please report all bugs to the issue tracker:
 https://bugs.chromium.org/p/webp
 .br
 Patches welcome! See this page to get started:

man/vwebp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH VWEBP 1 "December 11, 2015"
+.TH VWEBP 1 "June 23, 2016"
 .SH NAME
 vwebp \- decompress a WebP file and display it in a window
 .SH SYNOPSIS
@@ -63,7 +63,7 @@ Overlay file information.
 Quit.
 
 .SH BUGS
-Please report all bugs to our issue tracker:
+Please report all bugs to the issue tracker:
 https://bugs.chromium.org/p/webp
 .br
 Patches welcome! See this page to get started:

man/webpmux.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH WEBPMUX 1 "December 11, 2015"
+.TH WEBPMUX 1 "June 23, 2016"
 .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.
@@ -128,7 +128,7 @@ Output file in WebP format.
 The nature of EXIF, XMP and ICC data is not checked and is assumed to be valid.
 
 .SH BUGS
-Please report all bugs to our issue tracker:
+Please report all bugs to the issue tracker:
 https://bugs.chromium.org/p/webp
 .br
 Patches welcome! See this page to get started:

src/Makefile.am

@@ -38,7 +38,7 @@ 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 6:0:0
+libwebp_la_LDFLAGS = -no-undefined -version-info 6:1:0
 libwebpincludedir = $(includedir)/webp
 pkgconfig_DATA = libwebp.pc
 
@@ -50,7 +50,7 @@ if BUILD_LIBWEBPDECODER
   libwebpdecoder_la_LIBADD += dsp/libwebpdspdecode.la
   libwebpdecoder_la_LIBADD += utils/libwebputilsdecode.la
 
-  libwebpdecoder_la_LDFLAGS = -no-undefined -version-info 2:0:0
+  libwebpdecoder_la_LDFLAGS = -no-undefined -version-info 2:1:0
   pkgconfig_DATA += libwebpdecoder.pc
 endif
 

src/dec/vp8l.c

@@ -721,6 +721,9 @@ static void ProcessRows(VP8LDecoder* const dec, int row) {
   const int num_rows = row - dec->last_row_;
 
   assert(row <= dec->io_->crop_bottom);
+  // We can't process more than NUM_ARGB_CACHE_ROWS at a time (that's the size
+  // of argb_cache_), but we currently don't need more than that.
+  assert(num_rows <= NUM_ARGB_CACHE_ROWS);
   if (num_rows > 0) {    // Emit output.
     VP8Io* const io = dec->io_;
     uint8_t* rows_data = (uint8_t*)dec->argb_cache_;
@@ -1055,7 +1058,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
   const int mask = hdr->huffman_mask_;
   const HTreeGroup* htree_group =
       (src < src_last) ? GetHtreeGroupForPos(hdr, col, row) : NULL;
-  assert(src < src_end);
+  assert(dec->last_row_ < last_row);
   assert(src_last <= src_end);
 
   while (src < src_last) {
@@ -1464,23 +1467,31 @@ static int AllocateInternalBuffers8b(VP8LDecoder* const dec) {
 
 // Special row-processing that only stores the alpha data.
 static void ExtractAlphaRows(VP8LDecoder* const dec, int last_row) {
-  const int num_rows = last_row - dec->last_row_;
+  int cur_row = dec->last_row_;
-  const uint32_t* const in = dec->pixels_ + dec->width_ * dec->last_row_;
+  int num_rows = last_row - cur_row;
+  const uint32_t* in = dec->pixels_ + dec->width_ * cur_row;
 
   assert(last_row <= dec->io_->crop_bottom);
-  if (num_rows > 0) {
+  while (num_rows > 0) {
+    const int num_rows_to_process =
+        (num_rows > NUM_ARGB_CACHE_ROWS) ? NUM_ARGB_CACHE_ROWS : num_rows;
     // Extract alpha (which is stored in the green plane).
     ALPHDecoder* const alph_dec = (ALPHDecoder*)dec->io_->opaque;
     uint8_t* const output = alph_dec->output_;
     const int width = dec->io_->width;      // the final width (!= dec->width_)
-    const int cache_pixs = width * num_rows;
+    const int cache_pixs = width * num_rows_to_process;
-    uint8_t* dst = output + width * dec->last_row_;
+    uint8_t* const dst = output + width * cur_row;
     const uint32_t* const src = dec->argb_cache_;
     int i;
-    ApplyInverseTransforms(dec, num_rows, in);
+    ApplyInverseTransforms(dec, num_rows_to_process, in);
     for (i = 0; i < cache_pixs; ++i) dst[i] = (src[i] >> 8) & 0xff;
-    AlphaApplyFilter(alph_dec, dec->last_row_, last_row, dst, width);
+    AlphaApplyFilter(alph_dec,
+                     cur_row, cur_row + num_rows_to_process, dst, width);
+    num_rows -= num_rows_to_process;
+    in += num_rows_to_process * dec->width_;
+    cur_row += num_rows_to_process;
   }
+  assert(cur_row == last_row);
   dec->last_row_ = dec->last_out_row_ = last_row;
 }
 

src/dsp/lossless_enc.c

@@ -684,14 +684,13 @@ static uint32_t NearLossless(uint32_t value, uint32_t predict,
 // Returns the difference between the pixel and its prediction. In case of a
 // lossy encoding, updates the source image to avoid propagating the deviation
 // further to pixels which depend on the current pixel for their predictions.
-static uint32_t GetResidual(int width, int height,
+static WEBP_INLINE uint32_t GetResidual(int width, int height,
-                            uint32_t* const upper_row,
+                                        uint32_t* const upper_row,
-                            uint32_t* const current_row,
+                                        uint32_t* const current_row,
-                            const uint8_t* const max_diffs,
+                                        const uint8_t* const max_diffs,
-                            int mode, VP8LPredictorFunc pred_func,
+                                        int mode, VP8LPredictorFunc pred_func,
-                            int x, int y,
+                                        int x, int y, int max_quantization,
-                            int max_quantization,
+                                        int exact, int used_subtract_green) {
-                            int exact, int used_subtract_green) {
   const uint32_t predict = Predict(pred_func, x, y, current_row, upper_row);
   uint32_t residual;
   if (max_quantization == 1 || mode == 0 || y == 0 || y == height - 1 ||

src/enc/backward_references.c

@@ -245,6 +245,7 @@ int VP8LHashChainFill(VP8LHashChain* const p, int quality,
                       const uint32_t* const argb, int xsize, int ysize) {
   const int size = xsize * ysize;
   const int iter_max = GetMaxItersForQuality(quality);
+  const int iter_min = iter_max - quality / 10;
   const uint32_t window_size = GetWindowSizeForHashChain(quality, xsize);
   int pos;
   uint32_t base_position;
@@ -296,7 +297,11 @@ int VP8LHashChainFill(VP8LHashChain* const p, int quality,
       if (best_length < curr_length) {
         best_length = curr_length;
         best_distance = base_position - pos;
-        if (curr_length >= length_max) {
+        // Stop if we have reached the maximum length. Otherwise, make sure
+        // we have executed a minimum number of iterations depending on the
+        // quality.
+        if ((best_length == MAX_LENGTH) ||
+            (curr_length >= length_max && iter < iter_min)) {
           break;
         }
       }
@@ -437,7 +442,6 @@ static int BackwardReferencesLz77(int xsize, int ysize,
     int len = 0;
     int j;
     HashChainFindCopy(hash_chain, i, &offset, &len);
-    // MIN_LENGTH+1 is empirically better than MIN_LENGTH.
     if (len > MIN_LENGTH + 1) {
       const int len_ini = len;
       int max_reach = 0;
@@ -648,9 +652,8 @@ typedef struct {
   CostCacheInterval* cache_intervals_;
   size_t cache_intervals_size_;
   double cost_cache_[MAX_LENGTH];  // Contains the GetLengthCost(cost_model, k).
-  // cost_cache_min_dist_[i] contains the index of the minimum in
+  double min_cost_cache_;          // The minimum value in cost_cache_[1:].
-  // cost_cache_[1:i]. cost_cache_min_dist_[0] is unused.
+  double max_cost_cache_;          // The maximum value in cost_cache_[1:].
-  uint16_t cost_cache_min_dist_[MAX_LENGTH];
   float* costs_;
   uint16_t* dist_array_;
   // Most of the time, we only need few intervals -> use a free-list, to avoid
@@ -660,6 +663,10 @@ typedef struct {
   // These are regularly malloc'd remains. This list can't grow larger than than
   // size COST_CACHE_INTERVAL_SIZE_MAX - COST_MANAGER_MAX_FREE_LIST, note.
   CostInterval* recycled_intervals_;
+  // Buffer used in BackwardReferencesHashChainDistanceOnly to store the ends
+  // of the intervals that can have impacted the cost at a pixel.
+  int* interval_ends_;
+  int interval_ends_size_;
 } CostManager;
 
 static int IsCostCacheIntervalWritable(int start, int end) {
@@ -708,6 +715,7 @@ static void CostManagerClear(CostManager* const manager) {
 
   WebPSafeFree(manager->costs_);
   WebPSafeFree(manager->cache_intervals_);
+  WebPSafeFree(manager->interval_ends_);
 
   // Clear the interval lists.
   DeleteIntervalList(manager, manager->head_);
@@ -720,22 +728,6 @@ static void CostManagerClear(CostManager* const manager) {
   CostManagerInitFreeList(manager);
 }
 
-static void CostManagerMinDistCacheInit(CostManager* const manager) {
-  int min_idx = 1;
-  double min = manager->cost_cache_[1];
-  int i;
-
-  // cost_cache_min_dist_[i] contains the index of the minimum in
-  // cost_cache_[1:i].
-  for (i = 1; i < MAX_LENGTH; ++i) {
-    manager->cost_cache_min_dist_[i] = min_idx;
-    if (manager->cost_cache_[i] < min) {
-      min_idx = i;
-      min = manager->cost_cache_[i];
-    }
-  }
-}
-
 static int CostManagerInit(CostManager* const manager,
                            uint16_t* const dist_array, int pix_count,
                            const CostModel* const cost_model) {
@@ -745,6 +737,9 @@ static int CostManagerInit(CostManager* const manager,
   // Empirically, differences between intervals is usually of more than 1.
   const double min_cost_diff = 0.1;
 
+  manager->costs_ = NULL;
+  manager->cache_intervals_ = NULL;
+  manager->interval_ends_ = NULL;
   manager->head_ = NULL;
   manager->recycled_intervals_ = NULL;
   manager->count_ = 0;
@@ -761,10 +756,17 @@ static int CostManagerInit(CostManager* const manager,
         min_cost_diff) {
       ++manager->cache_intervals_size_;
     }
+    // Compute the minimum of cost_cache_.
+    if (i == 1) {
+      manager->min_cost_cache_ = manager->cost_cache_[1];
+      manager->max_cost_cache_ = manager->cost_cache_[1];
+    } else if (manager->cost_cache_[i] < manager->min_cost_cache_) {
+      manager->min_cost_cache_ = manager->cost_cache_[i];
+    } else if (manager->cost_cache_[i] > manager->max_cost_cache_) {
+      manager->max_cost_cache_ = manager->cost_cache_[i];
+    }
   }
 
-  CostManagerMinDistCacheInit(manager);
-
   // With the current cost models, we have 15 intervals, so we are safe by
   // setting a maximum of COST_CACHE_INTERVAL_SIZE_MAX.
   if (manager->cache_intervals_size_ > COST_CACHE_INTERVAL_SIZE_MAX) {
@@ -789,7 +791,8 @@ static int CostManagerInit(CostManager* const manager,
     // difference is found, a new interval is created and bounded.
     for (i = 0; i < cost_cache_size; ++i) {
       const double cost_val = manager->cost_cache_[i];
-      if (fabs(cost_val - cost_prev) > min_cost_diff && cur + 1 < end) {
+      if (i == 0 ||
+          (fabs(cost_val - cost_prev) > min_cost_diff && cur + 1 < end)) {
         if (i > 1) {
           const int is_writable =
               IsCostCacheIntervalWritable(cur->start_, cur->end_);
@@ -833,25 +836,67 @@ static int CostManagerInit(CostManager* const manager,
     CostManagerClear(manager);
     return 0;
   }
-  // Set the initial costs_ high for every pixel as we wil lkeep the minimum.
+  // Set the initial costs_ high for every pixel as we will keep the minimum.
   for (i = 0; i < pix_count; ++i) manager->costs_[i] = 1e38f;
 
+  // The cost at pixel is influenced by the cost intervals from previous pixels.
+  // Let us take the specific case where the offset is the same (which actually
+  // happens a lot in case of uniform regions).
+  // pixel i contributes to j>i a cost of: offset cost + cost_cache_[j-i]
+  // pixel i+1 contributes to j>i a cost of: 2*offset cost + cost_cache_[j-i-1]
+  // pixel i+2 contributes to j>i a cost of: 3*offset cost + cost_cache_[j-i-2]
+  // and so on.
+  // A pixel i influences the following length(j) < MAX_LENGTH pixels. What is
+  // the value of j such that pixel i + j cannot influence any of those pixels?
+  // This value is such that:
+  //               max of cost_cache_ < j*offset cost + min of cost_cache_
+  // (pixel i + j 's cost cannot beat the worst cost given by pixel i).
+  // This value will be used to optimize the cost computation in
+  // BackwardReferencesHashChainDistanceOnly.
+  {
+    // The offset cost is computed in GetDistanceCost and has a minimum value of
+    // the minimum in cost_model->distance_. The case where the offset cost is 0
+    // will be dealt with differently later so we are only interested in the
+    // minimum non-zero offset cost.
+    double offset_cost_min = 0.;
+    int size;
+    for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
+      if (cost_model->distance_[i] != 0) {
+        if (offset_cost_min == 0.) {
+          offset_cost_min = cost_model->distance_[i];
+        } else if (cost_model->distance_[i] < offset_cost_min) {
+          offset_cost_min = cost_model->distance_[i];
+        }
+      }
+    }
+    // In case all the cost_model->distance_ is 0, the next non-zero cost we
+    // can have is from the extra bit in GetDistanceCost, hence 1.
+    if (offset_cost_min < 1.) offset_cost_min = 1.;
+
+    size = 1 + (int)ceil((manager->max_cost_cache_ - manager->min_cost_cache_) /
+                         offset_cost_min);
+    // Empirically, we usually end up with a value below 100.
+    if (size > MAX_LENGTH) size = MAX_LENGTH;
+
+    manager->interval_ends_ =
+        (int*)WebPSafeMalloc(size, sizeof(*manager->interval_ends_));
+    if (manager->interval_ends_ == NULL) {
+      CostManagerClear(manager);
+      return 0;
+    }
+    manager->interval_ends_size_ = size;
+  }
+
   return 1;
 }
 
 // Given the distance_cost for pixel 'index', update the cost at pixel 'i' if it
 // is smaller than the previously computed value.
-// If index is negative, the distance is computed using
-// manager->cost_cache_min_dist_[-index] instead.
 static WEBP_INLINE void UpdateCost(CostManager* const manager, int i, int index,
                                    double distance_cost) {
-  int k;
+  int k = i - index;
   double cost_tmp;
-  if (index < 0) {
+  assert(k >= 0 && k < MAX_LENGTH);
-    k = manager->cost_cache_min_dist_[-index];
-  } else {
-    k = i - index;
-  }
   cost_tmp = distance_cost + manager->cost_cache_[k];
 
   if (manager->costs_[i] > cost_tmp) {
@@ -1173,41 +1218,63 @@ static int BackwardReferencesHashChainDistanceOnly(
     if (len >= MIN_LENGTH) {
       const int code = DistanceToPlaneCode(xsize, offset);
       const double offset_cost = GetDistanceCost(cost_model, code);
-      double distance_cost = prev_cost + offset_cost;
       const int first_i = i;
+      int j_max = 0, interval_ends_index = 0;
+      const int is_offset_zero = (offset_cost == 0.);
 
-      PushInterval(cost_manager, distance_cost, i, len);
+      if (!is_offset_zero) {
-
+        j_max = (int)ceil(
-      // The cost to get to pixel i+k is:
+            (cost_manager->max_cost_cache_ - cost_manager->min_cost_cache_) /
-      //   prev_cost + best weight of path going from i to l.
+            offset_cost);
-      // For the last cost, if the path were composed of several segments
+        if (j_max < 1) {
-      // (like [i,j), [j,k), [k,l) ), its cost would be:
+          j_max = 1;
-      //   offset cost + manager->cost_cache_[j-i] +
+        } else if (j_max > cost_manager->interval_ends_size_ - 1) {
-      //   offset cost + manager->cost_cache_[k-j] +
+          // This could only happen in the case of MAX_LENGTH.
-      //   offset cost + manager->cost_cache_[l-j]
+          j_max = cost_manager->interval_ends_size_ - 1;
-      // As the offset_cost is close to the values in cost_cache_ (both are the
+        }
-      // cost to store a number, whether it's an offset or a length), the best
+      }  // else j_max is unused anyway.
-      // paths are therefore the ones involving one hop.
-      for (; i + 1 < pix_count - 1; ++i) {
-        int offset_next, len_next, diff;
 
+      // Instead of considering all contributions from a pixel i by calling:
+      //         PushInterval(cost_manager, prev_cost + offset_cost, i, len);
+      // we optimize these contributions in case offset_cost stays the same for
+      // consecutive pixels. This describes a set of pixels similar to a
+      // previous set (e.g. constant color regions).
+      for (; i < pix_count - 1; ++i) {
+        int offset_next, len_next;
         prev_cost = cost_manager->costs_[i - 1];
-        distance_cost = prev_cost + offset_cost;
+
+        if (is_offset_zero) {
+          // No optimization can be made so we just push all of the
+          // contributions from i.
+          PushInterval(cost_manager, prev_cost, i, len);
+        } else {
+          // j_max is chosen as the smallest j such that:
+          //       max of cost_cache_ < j*offset cost + min of cost_cache_
+          // Therefore, the pixel influenced by i-j_max, cannot be influenced
+          // by i. Only the costs after the end of what i contributed need to be
+          // updated. cost_manager->interval_ends_ is a circular buffer that
+          // stores those ends.
+          const double distance_cost = prev_cost + offset_cost;
+          int j = cost_manager->interval_ends_[interval_ends_index];
+          if (i - first_i <= j_max ||
+              !IsCostCacheIntervalWritable(j, i + len)) {
+            PushInterval(cost_manager, distance_cost, i, len);
+          } else {
+            for (; j < i + len; ++j) {
+              UpdateCost(cost_manager, j, i, distance_cost);
+            }
+          }
+          // Store the new end in the circular buffer.
+          assert(interval_ends_index < cost_manager->interval_ends_size_);
+          cost_manager->interval_ends_[interval_ends_index] = i + len;
+          if (++interval_ends_index > j_max) interval_ends_index = 0;
+        }
 
         // Check whether i is the last pixel to consider, as it is handled
         // differently.
+        if (i + 1 >= pix_count - 1) break;
         HashChainFindCopy(hash_chain, i + 1, &offset_next, &len_next);
         if (offset_next != offset) break;
-
-        diff = i - first_i;
-        if (diff > 0) {
-          assert(len > 1);
-          if (diff >= len) diff = len - 1;
-          // The best path going to pixel i is the one starting in a pixel
-          // within diff hence at the pixel:
-          //     i - manager->cost_cache_min_dist_[diff].
-          UpdateCost(cost_manager, i, -diff, distance_cost);
-        }
         len = len_next;
         UpdateCostPerIndex(cost_manager, i);
         AddSingleLiteralWithCostModel(argb + i, &hashers, cost_model, i,
@@ -1215,7 +1282,6 @@ static int BackwardReferencesHashChainDistanceOnly(
                                       cost_manager->costs_, dist_array);
       }
       // Submit the last pixel.
-      if (first_i != i) PushInterval(cost_manager, distance_cost, i, len);
       UpdateCostPerIndex(cost_manager, i + 1);
 
       // This if is for speedup only. It roughly doubles the speed, and
@@ -1238,7 +1304,7 @@ static int BackwardReferencesHashChainDistanceOnly(
         }
         goto next_symbol;
       }
-      if (len != MIN_LENGTH) {
+      if (len > MIN_LENGTH) {
         int code_min_length;
         double cost_total;
         offset = HashChainFindOffset(hash_chain, i);

src/enc/vp8enci.h

@@ -32,7 +32,7 @@ extern "C" {
 // version numbers
 #define ENC_MAJ_VERSION 0
 #define ENC_MIN_VERSION 5
-#define ENC_REV_VERSION 0
+#define ENC_REV_VERSION 1
 
 enum { MAX_LF_LEVELS = 64,       // Maximum loop filter level
        MAX_VARIABLE_LEVEL = 67,  // last (inclusive) level with variable cost

src/enc/vp8l.c

@@ -290,7 +290,7 @@ static int AnalyzeEntropy(const uint32_t* argb,
         }
       }
     }
-    free(histo);
+    WebPSafeFree(histo);
     return 1;
   } else {
     return 0;

src/mux/Makefile.am

@@ -13,6 +13,6 @@ libwebpmuxinclude_HEADERS += ../webp/mux_types.h
 libwebpmuxinclude_HEADERS += ../webp/types.h
 
 libwebpmux_la_LIBADD = ../libwebp.la
-libwebpmux_la_LDFLAGS = -no-undefined -version-info 2:0:0
+libwebpmux_la_LDFLAGS = -no-undefined -version-info 2:1:0
 libwebpmuxincludedir = $(includedir)/webp
 pkgconfig_DATA = libwebpmux.pc

src/mux/muxi.h

@@ -28,7 +28,7 @@ extern "C" {
 
 #define MUX_MAJ_VERSION 0
 #define MUX_MIN_VERSION 3
-#define MUX_REV_VERSION 0
+#define MUX_REV_VERSION 1
 
 // Chunk object.
 typedef struct WebPChunk WebPChunk;

src/webp/decode.h

@@ -39,8 +39,8 @@ typedef struct WebPDecoderConfig WebPDecoderConfig;
 WEBP_EXTERN(int) WebPGetDecoderVersion(void);
 
 // Retrieve basic header information: width, height.
-// This function will also validate the header and return 0 in
+// This function will also validate the header, returning true on success,
-// case of formatting error.
+// false otherwise. '*width' and '*height' are only valid on successful return.
 // Pointers 'width' and 'height' can be passed NULL if deemed irrelevant.
 WEBP_EXTERN(int) WebPGetInfo(const uint8_t* data, size_t data_size,
                              int* width, int* height);
@@ -471,16 +471,18 @@ static WEBP_INLINE int WebPInitDecoderConfig(WebPDecoderConfig* config) {
 // parameter, in which case the features will be parsed and stored into
 // config->input. Otherwise, 'data' can be NULL and no parsing will occur.
 // Note that 'config' can be NULL too, in which case a default configuration
-// is used.
+// is used. If 'config' is not NULL, it must outlive the WebPIDecoder object
+// as some references to its fields will be used. No internal copy of 'config'
+// is made.
 // The return WebPIDecoder object must always be deleted calling WebPIDelete().
 // Returns NULL in case of error (and config->status will then reflect
-// the error condition).
+// the error condition, if available).
 WEBP_EXTERN(WebPIDecoder*) WebPIDecode(const uint8_t* data, size_t data_size,
                                        WebPDecoderConfig* config);
 
 // Non-incremental version. This version decodes the full data at once, taking
 // 'config' into account. Returns decoding status (which should be VP8_STATUS_OK
-// if the decoding was successful).
+// if the decoding was successful). Note that 'config' cannot be NULL.
 WEBP_EXTERN(VP8StatusCode) WebPDecode(const uint8_t* data, size_t data_size,
                                       WebPDecoderConfig* config);