muxread: fix reading of buffers > riff size

After:
  2c70ad76 muxread,CreateInternal: fix riff size checks (cl/200674839)

`SizeWithPadding()` adds `CHUNK_HEADER_SIZE` (plus additional 1 byte
padding if needed). A later check included `CHUNK_HEADER_SIZE` before
capping the value of the size passed to `WebPMuxCreateInternal()`,
missing cases with a small amount of extra data after the RIFF chunk
(like a newline when the file is opened and saved in a text editor) and
setting size to an incorrect value, so larger sizes would also fail.

Another check of `riff_size < CHUNK_HEADER_SIZE` after the call to
`SizeWithPadding()` is removed because 1) it could not fail given
`SizeWithPadding()` adds `CHUNK_HEADER_SIZE` to the value; and 2) it is
redundant as `size < RIFF_HEADER_SIZE + CHUNK_HEADER_SIZE` is checked
earlier in the function.

Bug: webp:42340561
Change-Id: I58dc4f071b27c2841001b4012aabdb1869f64f97

CMakeLists.txt

@@ -567,7 +567,8 @@ if(WEBP_BUILD_GIF2WEBP)
   add_executable(gif2webp ${GIF2WEBP_SRCS})
   target_link_libraries(gif2webp exampleutil imageioutil webp libwebpmux
                         ${WEBP_DEP_GIF_LIBRARIES})
-  target_include_directories(gif2webp PRIVATE ${CMAKE_CURRENT_BINARY_DIR}/src)
+  target_include_directories(gif2webp PRIVATE ${CMAKE_CURRENT_BINARY_DIR}/src
+                                              ${CMAKE_CURRENT_SOURCE_DIR})
   install(TARGETS gif2webp RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
 endif()
 

Makefile.vc

@@ -393,7 +393,7 @@ $(DIRBIN)\dwebp.exe: $(IMAGEIO_UTIL_OBJS)
 $(DIRBIN)\dwebp.exe: $(LIBWEBPDEMUX)
 $(DIRBIN)\gif2webp.exe: $(DIROBJ)\examples\gif2webp.obj $(EX_GIF_DEC_OBJS)
 $(DIRBIN)\gif2webp.exe: $(EX_UTIL_OBJS) $(IMAGEIO_UTIL_OBJS) $(LIBWEBPMUX)
-$(DIRBIN)\gif2webp.exe: $(LIBWEBP)
+$(DIRBIN)\gif2webp.exe: $(LIBWEBP) $(LIBSHARPYUV)
 $(DIRBIN)\vwebp.exe: $(DIROBJ)\examples\vwebp.obj $(EX_UTIL_OBJS)
 $(DIRBIN)\vwebp.exe: $(IMAGEIO_UTIL_OBJS) $(LIBWEBPDEMUX) $(LIBWEBP)
 $(DIRBIN)\vwebp_sdl.exe: $(DIROBJ)\extras\vwebp_sdl.obj

doc/tools.md

@@ -354,6 +354,10 @@ Options:
 -lossy ................. encode image using lossy compression
 -mixed ................. for each frame in the image, pick lossy
                          or lossless compression heuristically
+-near_lossless <int> ... use near-lossless image preprocessing
+                         (0..100=off), default=100
+-sharp_yuv ............. use sharper (and slower) RGB->YUV conversion
+                         (lossy only)
 -q <float> ............. quality factor (0:small..100:big)
 -m <int> ............... compression method (0=fast, 6=slowest)
 -min_size .............. minimize output size (default:off)

examples/Makefile.am

@@ -67,7 +67,7 @@ dwebp_LDADD += ../src/libwebp.la
 dwebp_LDADD +=$(PNG_LIBS) $(JPEG_LIBS)
 
 gif2webp_SOURCES = gif2webp.c gifdec.c gifdec.h
-gif2webp_CPPFLAGS = $(AM_CPPFLAGS) $(GIF_INCLUDES)
+gif2webp_CPPFLAGS = $(AM_CPPFLAGS) $(GIF_INCLUDES) -I$(top_srcdir)
 gif2webp_LDADD  =
 gif2webp_LDADD += libexample_util.la
 gif2webp_LDADD += ../imageio/libimageio_util.la

examples/gif2webp.c

@@ -28,6 +28,7 @@
 #endif
 
 #include <gif_lib.h>
+#include "sharpyuv/sharpyuv.h"
 #include "webp/encode.h"
 #include "webp/mux.h"
 #include "../examples/example_util.h"
@@ -70,6 +71,11 @@ static void Help(void) {
   printf("  -lossy ................. encode image using lossy compression\n");
   printf("  -mixed ................. for each frame in the image, pick lossy\n"
          "                           or lossless compression heuristically\n");
+  printf("  -near_lossless <int> ... use near-lossless image preprocessing\n"
+         "                           (0..100=off), default=100\n");
+  printf("  -sharp_yuv ............. use sharper (and slower) RGB->YUV "
+                                    "conversion\n"
+         "                           (lossy only)\n");
   printf("  -q <float> ............. quality factor (0:small..100:big)\n");
   printf("  -m <int> ............... compression method (0=fast, 6=slowest)\n");
   printf("  -min_size .............. minimize output size (default:off)\n"
@@ -166,6 +172,10 @@ int main(int argc, const char* argv[]) {
     } else if (!strcmp(argv[c], "-mixed")) {
       enc_options.allow_mixed = 1;
       config.lossless = 0;
+    } else if (!strcmp(argv[c], "-near_lossless") && c < argc - 1) {
+      config.near_lossless = ExUtilGetInt(argv[++c], 0, &parse_error);
+    } else if (!strcmp(argv[c], "-sharp_yuv")) {
+      config.use_sharp_yuv = 1;
     } else if (!strcmp(argv[c], "-loop_compatibility")) {
       loop_compatibility = 1;
     } else if (!strcmp(argv[c], "-q") && c < argc - 1) {
@@ -226,10 +236,13 @@ int main(int argc, const char* argv[]) {
     } else if (!strcmp(argv[c], "-version")) {
       const int enc_version = WebPGetEncoderVersion();
       const int mux_version = WebPGetMuxVersion();
+      const int sharpyuv_version = SharpYuvGetVersion();
       printf("WebP Encoder version: %d.%d.%d\nWebP Mux version: %d.%d.%d\n",
              (enc_version >> 16) & 0xff, (enc_version >> 8) & 0xff,
              enc_version & 0xff, (mux_version >> 16) & 0xff,
              (mux_version >> 8) & 0xff, mux_version & 0xff);
+      printf("libsharpyuv: %d.%d.%d\n", (sharpyuv_version >> 24) & 0xff,
+             (sharpyuv_version >> 16) & 0xffff, sharpyuv_version & 0xff);
       FREE_WARGV_AND_RETURN(EXIT_SUCCESS);
     } else if (!strcmp(argv[c], "-quiet")) {
       quiet = 1;

iosbuild.sh

@@ -53,7 +53,7 @@ DEMUXLIBLIST=''
 if [[ -z "${SDK}" ]]; then
   echo "iOS SDK not available"
   exit 1
-elif [[ ${SDK%%.*} -gt 8 ]]; then
+elif [[ ${SDK%%.*} -gt 8 && "${XCODE%%.*}" -lt 16 ]]; then
   EXTRA_CFLAGS="-fembed-bitcode"
 elif [[ ${SDK%%.*} -le 6 ]]; then
   echo "You need iOS SDK version 6.0 or above"

man/gif2webp.1

@@ -1,5 +1,5 @@
 .\"                                      Hey, EMACS: -*- nroff -*-
-.TH GIF2WEBP 1 "July 18, 2024"
+.TH GIF2WEBP 1 "November 4, 2024"
 .SH NAME
 gif2webp \- Convert a GIF image to WebP
 .SH SYNOPSIS
@@ -39,6 +39,18 @@ Encode the image using lossy compression.
 Mixed compression mode: optimize compression of the image by picking either
 lossy or lossless compression for each frame heuristically.
 .TP
+.BI \-near_lossless " int
+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
+.B \-sharp_yuv
+Use more accurate and sharper RGB->YUV conversion. Note that this process is
+slower than the default 'fast' RGB->YUV conversion.
+.TP
 .BI \-q " float
 Specify the compression factor for RGB channels between 0 and 100. The default
 is 75.

sharpyuv/sharpyuv_csp.c

@@ -22,16 +22,16 @@ void SharpYuvComputeConversionMatrix(const SharpYuvColorSpace* yuv_color_space,
   const float kr = yuv_color_space->kr;
   const float kb = yuv_color_space->kb;
   const float kg = 1.0f - kr - kb;
-  const float cr = 0.5f / (1.0f - kb);
-  const float cb = 0.5f / (1.0f - kr);
+  const float cb = 0.5f / (1.0f - kb);
+  const float cr = 0.5f / (1.0f - kr);
 
   const int shift = yuv_color_space->bit_depth - 8;
 
   const float denom = (float)((1 << yuv_color_space->bit_depth) - 1);
   float scale_y = 1.0f;
   float add_y = 0.0f;
-  float scale_u = cr;
-  float scale_v = cb;
+  float scale_u = cb;
+  float scale_v = cr;
   float add_uv = (float)(128 << shift);
   assert(yuv_color_space->bit_depth >= 8);
 
@@ -60,6 +60,10 @@ void SharpYuvComputeConversionMatrix(const SharpYuvColorSpace* yuv_color_space,
 
 // Matrices are in YUV_FIX fixed point precision.
 // WebP's matrix, similar but not identical to kRec601LimitedMatrix
+// Derived using the following formulas:
+// Y = 0.2569 * R + 0.5044 * G + 0.0979 * B + 16
+// U = -0.1483 * R - 0.2911 * G + 0.4394 * B + 128
+// V = 0.4394 * R - 0.3679 * G - 0.0715 * B + 128
 static const SharpYuvConversionMatrix kWebpMatrix = {
   {16839, 33059, 6420, 16 << 16},
   {-9719, -19081, 28800, 128 << 16},

src/dsp/lossless_enc_sse2.c

@@ -175,64 +175,102 @@ static void CollectColorRedTransforms_SSE2(const uint32_t* WEBP_RESTRICT argb,
 
 // Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
 // that's ok since the histogram values are less than 1<<28 (max picture size).
-#define LINE_SIZE 16    // 8 or 16
 static void AddVector_SSE2(const uint32_t* WEBP_RESTRICT a,
                            const uint32_t* WEBP_RESTRICT b,
                            uint32_t* WEBP_RESTRICT out, int size) {
-  int i;
-  for (i = 0; i + LINE_SIZE <= size; i += LINE_SIZE) {
+  int i = 0;
+  int aligned_size = size & ~15;
+  // Size is, at minimum, NUM_DISTANCE_CODES (40) and may be as large as
+  // NUM_LITERAL_CODES (256) + NUM_LENGTH_CODES (24) + (0 or a non-zero power of
+  // 2). See the usage in VP8LHistogramAdd().
+  assert(size >= 16);
+  assert(size % 2 == 0);
+
+  do {
     const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i +  0]);
     const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i +  4]);
-#if (LINE_SIZE == 16)
     const __m128i a2 = _mm_loadu_si128((const __m128i*)&a[i +  8]);
     const __m128i a3 = _mm_loadu_si128((const __m128i*)&a[i + 12]);
-#endif
     const __m128i b0 = _mm_loadu_si128((const __m128i*)&b[i +  0]);
     const __m128i b1 = _mm_loadu_si128((const __m128i*)&b[i +  4]);
-#if (LINE_SIZE == 16)
     const __m128i b2 = _mm_loadu_si128((const __m128i*)&b[i +  8]);
     const __m128i b3 = _mm_loadu_si128((const __m128i*)&b[i + 12]);
-#endif
     _mm_storeu_si128((__m128i*)&out[i +  0], _mm_add_epi32(a0, b0));
     _mm_storeu_si128((__m128i*)&out[i +  4], _mm_add_epi32(a1, b1));
-#if (LINE_SIZE == 16)
     _mm_storeu_si128((__m128i*)&out[i +  8], _mm_add_epi32(a2, b2));
     _mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
-#endif
+    i += 16;
+  } while (i != aligned_size);
+
+  if ((size & 8) != 0) {
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i + 0]);
+    const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i + 4]);
+    const __m128i b0 = _mm_loadu_si128((const __m128i*)&b[i + 0]);
+    const __m128i b1 = _mm_loadu_si128((const __m128i*)&b[i + 4]);
+    _mm_storeu_si128((__m128i*)&out[i + 0], _mm_add_epi32(a0, b0));
+    _mm_storeu_si128((__m128i*)&out[i + 4], _mm_add_epi32(a1, b1));
+    i += 8;
   }
-  for (; i < size; ++i) {
-    out[i] = a[i] + b[i];
+
+  size &= 7;
+  if (size == 4) {
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i]);
+    const __m128i b0 = _mm_loadu_si128((const __m128i*)&b[i]);
+    _mm_storeu_si128((__m128i*)&out[i], _mm_add_epi32(a0, b0));
+  } else if (size == 2) {
+    const __m128i a0 = _mm_loadl_epi64((const __m128i*)&a[i]);
+    const __m128i b0 = _mm_loadl_epi64((const __m128i*)&b[i]);
+    _mm_storel_epi64((__m128i*)&out[i], _mm_add_epi32(a0, b0));
   }
 }
 
 static void AddVectorEq_SSE2(const uint32_t* WEBP_RESTRICT a,
                              uint32_t* WEBP_RESTRICT out, int size) {
-  int i;
-  for (i = 0; i + LINE_SIZE <= size; i += LINE_SIZE) {
+  int i = 0;
+  int aligned_size = size & ~15;
+  // Size is, at minimum, NUM_DISTANCE_CODES (40) and may be as large as
+  // NUM_LITERAL_CODES (256) + NUM_LENGTH_CODES (24) + (0 or a non-zero power of
+  // 2). See the usage in VP8LHistogramAdd().
+  assert(size >= 16);
+  assert(size % 2 == 0);
+
+  do {
     const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i +  0]);
     const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i +  4]);
-#if (LINE_SIZE == 16)
     const __m128i a2 = _mm_loadu_si128((const __m128i*)&a[i +  8]);
     const __m128i a3 = _mm_loadu_si128((const __m128i*)&a[i + 12]);
-#endif
     const __m128i b0 = _mm_loadu_si128((const __m128i*)&out[i +  0]);
     const __m128i b1 = _mm_loadu_si128((const __m128i*)&out[i +  4]);
-#if (LINE_SIZE == 16)
     const __m128i b2 = _mm_loadu_si128((const __m128i*)&out[i +  8]);
     const __m128i b3 = _mm_loadu_si128((const __m128i*)&out[i + 12]);
-#endif
     _mm_storeu_si128((__m128i*)&out[i +  0], _mm_add_epi32(a0, b0));
     _mm_storeu_si128((__m128i*)&out[i +  4], _mm_add_epi32(a1, b1));
-#if (LINE_SIZE == 16)
     _mm_storeu_si128((__m128i*)&out[i +  8], _mm_add_epi32(a2, b2));
     _mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
-#endif
+    i += 16;
+  } while (i != aligned_size);
+
+  if ((size & 8) != 0) {
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i + 0]);
+    const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i + 4]);
+    const __m128i b0 = _mm_loadu_si128((const __m128i*)&out[i + 0]);
+    const __m128i b1 = _mm_loadu_si128((const __m128i*)&out[i + 4]);
+    _mm_storeu_si128((__m128i*)&out[i + 0], _mm_add_epi32(a0, b0));
+    _mm_storeu_si128((__m128i*)&out[i + 4], _mm_add_epi32(a1, b1));
+    i += 8;
   }
-  for (; i < size; ++i) {
-    out[i] += a[i];
+
+  size &= 7;
+  if (size == 4) {
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i]);
+    const __m128i b0 = _mm_loadu_si128((const __m128i*)&out[i]);
+    _mm_storeu_si128((__m128i*)&out[i], _mm_add_epi32(a0, b0));
+  } else if (size == 2) {
+    const __m128i a0 = _mm_loadl_epi64((const __m128i*)&a[i]);
+    const __m128i b0 = _mm_loadl_epi64((const __m128i*)&out[i]);
+    _mm_storel_epi64((__m128i*)&out[i], _mm_add_epi32(a0, b0));
   }
 }
-#undef LINE_SIZE
 
 //------------------------------------------------------------------------------
 // Entropy

src/dsp/yuv.h

@@ -11,15 +11,15 @@
 //
 // The exact naming is Y'CbCr, following the ITU-R BT.601 standard.
 // More information at: https://en.wikipedia.org/wiki/YCbCr
-// Y = 0.2569 * R + 0.5044 * G + 0.0979 * B + 16
-// U = -0.1483 * R - 0.2911 * G + 0.4394 * B + 128
-// V = 0.4394 * R - 0.3679 * G - 0.0715 * B + 128
+// Y = 0.2568 * R + 0.5041 * G + 0.0979 * B + 16
+// U = -0.1482 * R - 0.2910 * G + 0.4392 * B + 128
+// V = 0.4392 * R - 0.3678 * G - 0.0714 * B + 128
 // We use 16bit fixed point operations for RGB->YUV conversion (YUV_FIX).
 //
 // For the Y'CbCr to RGB conversion, the BT.601 specification reads:
 //   R = 1.164 * (Y-16) + 1.596 * (V-128)
-//   G = 1.164 * (Y-16) - 0.813 * (V-128) - 0.391 * (U-128)
-//   B = 1.164 * (Y-16)                   + 2.018 * (U-128)
+//   G = 1.164 * (Y-16) - 0.813 * (V-128) - 0.392 * (U-128)
+//   B = 1.164 * (Y-16)                   + 2.017 * (U-128)
 // where Y is in the [16,235] range, and U/V in the [16,240] range.
 //
 // The fixed-point implementation used here is:

src/mux/muxread.c

@@ -223,11 +223,11 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
   // Note this padding is historical and differs from demux.c which does not
   // pad the file size.
   riff_size = SizeWithPadding(riff_size);
-  if (riff_size < CHUNK_HEADER_SIZE) goto Err;
   if (riff_size > size) goto Err;
-  // There's no point in reading past the end of the RIFF chunk.
-  if (size > riff_size + CHUNK_HEADER_SIZE) {
-    size = riff_size + CHUNK_HEADER_SIZE;
+  // There's no point in reading past the end of the RIFF chunk. Note riff_size
+  // includes CHUNK_HEADER_SIZE after SizeWithPadding().
+  if (size > riff_size) {
+    size = riff_size;
   }
 
   end = data + size;

xcframeworkbuild.sh

@@ -172,7 +172,9 @@ for (( i = 0; i < $NUM_PLATFORMS; ++i )); do
     CFLAGS="-pipe -isysroot ${SDKROOT} -O3 -DNDEBUG"
     case "${PLATFORM}" in
       iPhone*)
-        CFLAGS+=" -fembed-bitcode"
+        if [[ "${XCODE%%.*}" -lt 16 ]]; then
+          CFLAGS+=" -fembed-bitcode"
+        fi
         CFLAGS+=" -target ${ARCH}-apple-ios${IOS_MIN_VERSION}"
         [[ "${PLATFORM}" == *Simulator* ]] && CFLAGS+="-simulator"
         ;;