Update ChangeLog

Change-Id: I89a9c8cf6793c38e2b13d6e0557414fb05620eec

ChangeLog

@@ -1,3 +1,32 @@
+abc0604 Merge "update NEWS" into 0.2.0
+57cf313 update NEWS
+25f585c bump version to 0.2.1
+fed7c04 libwebp: validate chunk size in ParseOptionalChunks
+552cd9b cwebp (windows): fix alpha image import on XP
+b14fea9 autoconf/libwebp: enable dll builds for mingw
+4a8fb27 [cd]webp: always output windows errors
+d662158 fix double to float conversion warning
+72b96a6 cwebp: fix jpg encodes on XP
+734f762 VP8LAllocateHistogramSet: fix overflow in size calculation
+f9cb58f GetHistoBits: fix integer overflow
+b30add2 EncodeImageInternal: fix uninitialized free
+3de58d7 fix the -g/O3 discrepancy for 32bit compile
+77aa7d5 fix the BITS=8 case
+e5970bd Make *InitSSE2() functions be empty on non-SSE2 platform
+ef5cc47 make *InitSSE2() functions be empty on non-SSE2 platform
+c4ea259 make VP8DspInitNEON() public
+4238bc0 Update ChangeLog (v0.2.0)
+c655380 dec/io.c: cosmetics
+fe1958f RGBA4444: harmonize lossless/lossy alpha values
+681cb30 fix RGBA4444 output w/fancy upsampling
+f06c1d8 Merge "Alignment fix" into 0.2.0
+f56e98f Alignment fix
+6fe843b avoid rgb-premultiply if there's only trivial alpha values
+528a11a fix the ARGB4444 premultiply arithmetic
+a0a4885 Lossless decoder fix for a special transform order
+62dd9bb Update encoding heuristic w.r.t palette colors.
+6f4272b remove unused ApplyInverseTransform()
+93bf0fa Update ChangeLog (v0.2.0-rc1)
 5934fc5 update AUTHORS
 014a711 update NEWS
 43b0d61 add support for ARGB -> YUVA conversion for lossless decoder

NEWS

@@ -1,3 +1,8 @@
+- 10/30/12: version 0.2.1
+  * Various security related fixes
+  * cwebp.exe: fix import errors on Windows XP
+  * enable DLL builds for mingw targets
+
 - 8/3/12: version 0.2.0
   * Add support for ARGB -> YUVA conversion for lossless decoder
     New functions: WebPINewYUVA, WebPIDecGetYUVA

README

@@ -4,7 +4,7 @@
           \__\__/\____/\_____/__/ ____  ___
                 / _/ /    \    \ /  _ \/ _/
                /  \_/   / /   \ \   __/  \__
-               \____/____/\_____/_____/____/v0.2.0
+               \____/____/\_____/_____/____/v0.2.1
 
 Description:
 ============

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([libwebp], [0.2.0],
+AC_INIT([libwebp], [0.2.1],
         [http://code.google.com/p/webp/issues],,
         [http://developers.google.com/speed/webp])
 AC_CANONICAL_TARGET

examples/cwebp.c

@@ -95,14 +95,12 @@ static int ReadYUV(FILE* in_file, WebPPicture* const pic) {
 
 #ifdef HAVE_WINCODEC_H
 
-#define IFS(fn)                \
+#define IFS(fn)                                                     \
-  do {                         \
+  do {                                                              \
-     if (SUCCEEDED(hr))        \
+     if (SUCCEEDED(hr)) {                                           \
-     {                         \
+        hr = (fn);                                                  \
-        hr = (fn);             \
+        if (FAILED(hr)) fprintf(stderr, #fn " failed %08x\n", hr);  \
-        if (FAILED(hr) && verbose)           \
+     }                                                              \
-          fprintf(stderr, #fn " failed %08x\n", hr);  \
-     }                         \
   } while (0)
 
 // modified version of DEFINE_GUID from guiddef.h.
@@ -115,6 +113,12 @@ static int ReadYUV(FILE* in_file, WebPPicture* const pic) {
 #define MAKE_REFGUID(x) &(x)
 #endif
 
+typedef struct WICFormatImporter {
+  const GUID* pixel_format;
+  int bytes_per_pixel;
+  int (*import)(WebPPicture* const, const uint8_t* const, int);
+} WICFormatImporter;
+
 static HRESULT OpenInputStream(const char* filename, IStream** ppStream) {
   HRESULT hr = S_OK;
   IFS(SHCreateStreamOnFileA(filename, STGM_READ, ppStream));
@@ -125,6 +129,31 @@ static HRESULT OpenInputStream(const char* filename, IStream** ppStream) {
 
 static HRESULT ReadPictureWithWIC(const char* filename,
                                   WebPPicture* const pic, int keep_alpha) {
+  // From Microsoft SDK 7.0a -- wincodec.h
+  // Create local copies for compatibility when building against earlier
+  // versions of the SDK.
+  WEBP_DEFINE_GUID(GUID_WICPixelFormat24bppBGR_,
+                   0x6fddc324, 0x4e03, 0x4bfe,
+                   0xb1, 0x85, 0x3d, 0x77, 0x76, 0x8d, 0xc9, 0x0c);
+  WEBP_DEFINE_GUID(GUID_WICPixelFormat24bppRGB_,
+                   0x6fddc324, 0x4e03, 0x4bfe,
+                   0xb1, 0x85, 0x3d, 0x77, 0x76, 0x8d, 0xc9, 0x0d);
+  WEBP_DEFINE_GUID(GUID_WICPixelFormat32bppBGRA_,
+                   0x6fddc324, 0x4e03, 0x4bfe,
+                   0xb1, 0x85, 0x3d, 0x77, 0x76, 0x8d, 0xc9, 0x0f);
+  WEBP_DEFINE_GUID(GUID_WICPixelFormat32bppRGBA_,
+                   0xf5c7ad2d, 0x6a8d, 0x43dd,
+                   0xa7, 0xa8, 0xa2, 0x99, 0x35, 0x26, 0x1a, 0xe9);
+  const WICFormatImporter alphaFormatImporters[] = {
+    { &GUID_WICPixelFormat32bppBGRA_, 4, WebPPictureImportBGRA },
+    { &GUID_WICPixelFormat32bppRGBA_, 4, WebPPictureImportRGBA },
+    { NULL, 0, NULL },
+  };
+  const WICFormatImporter nonAlphaFormatImporters[] = {
+    { &GUID_WICPixelFormat24bppBGR_, 3, WebPPictureImportBGR },
+    { &GUID_WICPixelFormat24bppRGB_, 3, WebPPictureImportRGB },
+    { NULL, 0, NULL },
+  };
   HRESULT hr = S_OK;
   IWICBitmapFrameDecode* pFrame = NULL;
   IWICFormatConverter* pConverter = NULL;
@@ -135,6 +164,7 @@ static HRESULT ReadPictureWithWIC(const char* filename,
   UINT width = 0, height = 0;
   BYTE* rgb = NULL;
   WICPixelFormatGUID srcPixelFormat = { 0 };
+  const WICFormatImporter* importer = NULL;
   GUID srcContainerFormat = { 0 };
   const GUID* alphaContainers[] = {
     &GUID_ContainerFormatBmp,
@@ -143,18 +173,6 @@ static HRESULT ReadPictureWithWIC(const char* filename,
   };
   int has_alpha = 0;
   int i, stride;
-  // From Microsoft SDK 7.0a
-  // Create local copies for compatibility when building against earlier
-  // versions of the SDK.
-  WEBP_DEFINE_GUID(GUID_WICPixelFormat24bppRGB_,
-                   0x6fddc324, 0x4e03, 0x4bfe,
-                   0xb1, 0x85, 0x3d, 0x77, 0x76, 0x8d, 0xc9, 0x0d);
-  WEBP_DEFINE_GUID(GUID_WICPixelFormat32bppRGBA_,
-                   0xf5c7ad2d, 0x6a8d, 0x43dd,
-                   0xa7, 0xa8, 0xa2, 0x99, 0x35, 0x26, 0x1a, 0xe9);
-  WEBP_DEFINE_GUID(GUID_WICPixelFormat32bppBGRA_,
-                   0x6fddc324, 0x4e03, 0x4bfe,
-                   0xb1, 0x85, 0x3d, 0x77, 0x76, 0x8d, 0xc9, 0x0f);
 
   IFS(CoInitialize(NULL));
   IFS(CoCreateInstance(MAKE_REFGUID(CLSID_WICImagingFactory), NULL,
@@ -179,32 +197,45 @@ static HRESULT ReadPictureWithWIC(const char* filename,
   IFS(IWICBitmapFrameDecode_GetPixelFormat(pFrame, &srcPixelFormat));
   IFS(IWICBitmapDecoder_GetContainerFormat(pDecoder, &srcContainerFormat));
 
-  has_alpha = keep_alpha;
+  if (keep_alpha) {
-  for (i = 0;
+    for (i = 0;
-       has_alpha && i < sizeof(alphaContainers)/sizeof(alphaContainers[0]);
+         i < sizeof(alphaContainers) / sizeof(alphaContainers[0]);
-       ++i) {
+         ++i) {
-    if (IsEqualGUID(MAKE_REFGUID(srcContainerFormat),
+      if (IsEqualGUID(MAKE_REFGUID(srcContainerFormat),
-                    MAKE_REFGUID(*alphaContainers[i]))) {
+                      MAKE_REFGUID(*alphaContainers[i]))) {
-      has_alpha =
+        has_alpha =
-          IsEqualGUID(MAKE_REFGUID(srcPixelFormat),
+            IsEqualGUID(MAKE_REFGUID(srcPixelFormat),
-                      MAKE_REFGUID(GUID_WICPixelFormat32bppRGBA_)) ||
+                        MAKE_REFGUID(GUID_WICPixelFormat32bppRGBA_)) ||
-          IsEqualGUID(MAKE_REFGUID(srcPixelFormat),
+            IsEqualGUID(MAKE_REFGUID(srcPixelFormat),
-                      MAKE_REFGUID(GUID_WICPixelFormat32bppBGRA_));
+                        MAKE_REFGUID(GUID_WICPixelFormat32bppBGRA_));
-      break;
+        break;
+      }
     }
   }
 
   // Prepare for pixel format conversion (if necessary).
   IFS(IWICImagingFactory_CreateFormatConverter(pFactory, &pConverter));
+
+  for (importer = has_alpha ? alphaFormatImporters : nonAlphaFormatImporters;
+       hr == S_OK && importer->import != NULL; ++importer) {
+    BOOL canConvert;
+    const HRESULT cchr = IWICFormatConverter_CanConvert(
+        pConverter,
+        MAKE_REFGUID(srcPixelFormat),
+        MAKE_REFGUID(*importer->pixel_format),
+        &canConvert);
+    if (SUCCEEDED(cchr) && canConvert) break;
+  }
+  if (importer->import == NULL) hr = E_FAIL;
+
   IFS(IWICFormatConverter_Initialize(pConverter, (IWICBitmapSource*)pFrame,
-          has_alpha ? MAKE_REFGUID(GUID_WICPixelFormat32bppRGBA_)
+          importer->pixel_format,
-                    : MAKE_REFGUID(GUID_WICPixelFormat24bppRGB_),
           WICBitmapDitherTypeNone,
           NULL, 0.0, WICBitmapPaletteTypeCustom));
 
   // Decode.
   IFS(IWICFormatConverter_GetSize(pConverter, &width, &height));
-  stride = (has_alpha ? 4 : 3) * width * sizeof(*rgb);
+  stride = importer->bytes_per_pixel * width * sizeof(*rgb);
   if (SUCCEEDED(hr)) {
     rgb = (BYTE*)malloc(stride * height);
     if (rgb == NULL)
@@ -218,8 +249,7 @@ static HRESULT ReadPictureWithWIC(const char* filename,
     int ok;
     pic->width = width;
     pic->height = height;
-    ok = has_alpha ? WebPPictureImportRGBA(pic, rgb, stride)
+    ok = importer->import(pic, rgb, stride);
-                   : WebPPictureImportRGB(pic, rgb, stride);
     if (!ok)
       hr = E_FAIL;
   }

examples/dwebp.c

@@ -67,14 +67,12 @@ typedef enum {
 
 #ifdef HAVE_WINCODEC_H
 
-#define IFS(fn)                \
+#define IFS(fn)                                                     \
-  do {                         \
+  do {                                                              \
-     if (SUCCEEDED(hr))        \
+     if (SUCCEEDED(hr)) {                                           \
-     {                         \
+        hr = (fn);                                                  \
-        hr = (fn);             \
+        if (FAILED(hr)) fprintf(stderr, #fn " failed %08x\n", hr);  \
-        if (FAILED(hr) && verbose)           \
+     }                                                              \
-          fprintf(stderr, #fn " failed %08x\n", hr);  \
-     }                         \
   } while (0)
 
 #ifdef __cplusplus

src/Makefile.am

@@ -20,7 +20,11 @@ libwebp_la_LIBADD += dsp/libwebpdsp.la
 libwebp_la_LIBADD += enc/libwebpencode.la
 libwebp_la_LIBADD += utils/libwebputils.la
 
-libwebp_la_LDFLAGS = -version-info 4:0:0
+# Use '-no-undefined' to declare that libwebp does not depend on any libraries
+# other than the ones listed on the command line, i.e., after linking, it will
+# not have unresolved symbols. Some platforms (Windows among them) require all
+# symbols in shared libraries to be resolved at library creation.
+libwebp_la_LDFLAGS = -no-undefined -version-info 4:1:0
 libwebpincludedir = $(includedir)/webp
 
 pkgconfig_DATA = libwebp.pc

src/dec/io.c

@@ -111,7 +111,7 @@ static int EmitFancyRGB(const VP8Io* const io, WebPDecParams* const p) {
   const uint8_t* top_u = p->tmp_u;
   const uint8_t* top_v = p->tmp_v;
   int y = io->mb_y;
-  int y_end = io->mb_y + io->mb_h;
+  const int y_end = io->mb_y + io->mb_h;
   const int mb_w = io->mb_w;
   const int uv_w = (mb_w + 1) / 2;
 
@@ -150,7 +150,7 @@ static int EmitFancyRGB(const VP8Io* const io, WebPDecParams* const p) {
     // Process the very last row of even-sized picture
     if (!(y_end & 1)) {
       upsample(cur_y, NULL, cur_u, cur_v, cur_u, cur_v,
-              dst + buf->stride, NULL, mb_w);
+               dst + buf->stride, NULL, mb_w);
     }
   }
   return num_lines_out;
@@ -184,49 +184,60 @@ static int EmitAlphaYUV(const VP8Io* const io, WebPDecParams* const p) {
   return 0;
 }
 
+static int GetAlphaSourceRow(const VP8Io* const io,
+                             const uint8_t** alpha, int* const num_rows) {
+  int start_y = io->mb_y;
+  *num_rows = io->mb_h;
+
+  // Compensate for the 1-line delay of the fancy upscaler.
+  // This is similar to EmitFancyRGB().
+  if (io->fancy_upsampling) {
+    if (start_y == 0) {
+      // We don't process the last row yet. It'll be done during the next call.
+      --*num_rows;
+    } else {
+      --start_y;
+      // Fortunately, *alpha data is persistent, so we can go back
+      // one row and finish alpha blending, now that the fancy upscaler
+      // completed the YUV->RGB interpolation.
+      *alpha -= io->width;
+    }
+    if (io->crop_top + io->mb_y + io->mb_h == io->crop_bottom) {
+      // If it's the very last call, we process all the remaining rows!
+      *num_rows = io->crop_bottom - io->crop_top - start_y;
+    }
+  }
+  return start_y;
+}
+
 static int EmitAlphaRGB(const VP8Io* const io, WebPDecParams* const p) {
   const uint8_t* alpha = io->a;
   if (alpha != NULL) {
     const int mb_w = io->mb_w;
-    const int mb_h = io->mb_h;
-    int i, j;
     const WEBP_CSP_MODE colorspace = p->output->colorspace;
     const int alpha_first =
         (colorspace == MODE_ARGB || colorspace == MODE_Argb);
     const WebPRGBABuffer* const buf = &p->output->u.RGBA;
-    int start_y = io->mb_y;
+    int num_rows;
-    int num_rows = mb_h;
+    const int start_y = GetAlphaSourceRow(io, &alpha, &num_rows);
+    uint8_t* const base_rgba = buf->rgba + start_y * buf->stride;
+    uint8_t* dst = base_rgba + (alpha_first ? 0 : 3);
+    uint32_t alpha_mask = 0xff;
+    int i, j;
 
-    // We compensate for the 1-line delay of fancy upscaler.
+    for (j = 0; j < num_rows; ++j) {
-    // This is similar to EmitFancyRGB().
+      for (i = 0; i < mb_w; ++i) {
-    if (io->fancy_upsampling) {
+        const uint32_t alpha_value = alpha[i];
-      if (start_y == 0) {
+        dst[4 * i] = alpha_value;
-        // We don't process the last row yet. It'll be done during next call.
+        alpha_mask &= alpha_value;
-        --num_rows;
-      } else {
-        --start_y;
-        // Fortunately, *alpha data is persistent, so we can go back
-        // one row and finish alpha blending, now that the fancy upscaler
-        // completed the YUV->RGB interpolation.
-        alpha -= io->width;
-      }
-      if (io->crop_top + io->mb_y + mb_h == io->crop_bottom) {
-        // If it's the very last call, we process all the remaing rows!
-        num_rows = io->crop_bottom - io->crop_top - start_y;
       }
+      alpha += io->width;
+      dst += buf->stride;
     }
-    {
+    // alpha_mask is < 0xff if there's non-trivial alpha to premultiply with.
-      uint8_t* const base_rgba = buf->rgba + start_y * buf->stride;
+    if (alpha_mask != 0xff && WebPIsPremultipliedMode(colorspace)) {
-      uint8_t* dst = base_rgba + (alpha_first ? 0 : 3);
+      WebPApplyAlphaMultiply(base_rgba, alpha_first,
-      for (j = 0; j < num_rows; ++j) {
+                             mb_w, num_rows, buf->stride);
-        for (i = 0; i < mb_w; ++i) dst[4 * i] = alpha[i];
-        alpha += io->width;
-        dst += buf->stride;
-      }
-      if (WebPIsPremultipliedMode(colorspace)) {
-        WebPApplyAlphaMultiply(base_rgba, alpha_first,
-                               mb_w, num_rows, buf->stride);
-      }
     }
   }
   return 0;
@@ -236,22 +247,27 @@ static int EmitAlphaRGBA4444(const VP8Io* const io, WebPDecParams* const p) {
   const uint8_t* alpha = io->a;
   if (alpha != NULL) {
     const int mb_w = io->mb_w;
-    const int mb_h = io->mb_h;
+    const WEBP_CSP_MODE colorspace = p->output->colorspace;
-    int i, j;
     const WebPRGBABuffer* const buf = &p->output->u.RGBA;
-    uint8_t* const base_rgba = buf->rgba + io->mb_y * buf->stride;
+    int num_rows;
+    const int start_y = GetAlphaSourceRow(io, &alpha, &num_rows);
+    uint8_t* const base_rgba = buf->rgba + start_y * buf->stride;
     uint8_t* alpha_dst = base_rgba + 1;
-    for (j = 0; j < mb_h; ++j) {
+    uint32_t alpha_mask = 0x0f;
+    int i, j;
+
+    for (j = 0; j < num_rows; ++j) {
       for (i = 0; i < mb_w; ++i) {
         // Fill in the alpha value (converted to 4 bits).
-        const uint32_t alpha_val = VP8Clip4Bits(alpha[i]);
+        const uint32_t alpha_value = alpha[i] >> 4;
-        alpha_dst[2 * i] = (alpha_dst[2 * i] & 0xf0) | alpha_val;
+        alpha_dst[2 * i] = (alpha_dst[2 * i] & 0xf0) | alpha_value;
+        alpha_mask &= alpha_value;
       }
       alpha += io->width;
       alpha_dst += buf->stride;
     }
-    if (p->output->colorspace == MODE_rgbA_4444) {
+    if (alpha_mask != 0x0f && WebPIsPremultipliedMode(colorspace)) {
-      WebPApplyAlphaMultiply4444(base_rgba, mb_w, mb_h, buf->stride);
+      WebPApplyAlphaMultiply4444(base_rgba, mb_w, num_rows, buf->stride);
     }
   }
   return 0;
@@ -396,17 +412,22 @@ static int ExportAlpha(WebPDecParams* const p, int y_pos) {
   uint8_t* dst = base_rgba + (alpha_first ? 0 : 3);
   int num_lines_out = 0;
   const int is_premult_alpha = WebPIsPremultipliedMode(colorspace);
+  uint32_t alpha_mask = 0xff;
   const int width = p->scaler_a.dst_width;
 
   while (WebPRescalerHasPendingOutput(&p->scaler_a)) {
     int i;
     assert(p->last_y + y_pos + num_lines_out < p->output->height);
     WebPRescalerExportRow(&p->scaler_a);
-    for (i = 0; i < width; ++i) dst[4 * i] = p->scaler_a.dst[i];
+    for (i = 0; i < width; ++i) {
+      const uint32_t alpha_value = p->scaler_a.dst[i];
+      dst[4 * i] = alpha_value;
+      alpha_mask &= alpha_value;
+    }
     dst += buf->stride;
     ++num_lines_out;
   }
-  if (is_premult_alpha) {
+  if (is_premult_alpha && alpha_mask != 0xff) {
     WebPApplyAlphaMultiply(base_rgba, alpha_first,
                            width, num_lines_out, buf->stride);
   }
@@ -421,6 +442,7 @@ static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos) {
   const WEBP_CSP_MODE colorspace = p->output->colorspace;
   const int width = p->scaler_a.dst_width;
   const int is_premult_alpha = WebPIsPremultipliedMode(colorspace);
+  uint32_t alpha_mask = 0x0f;
 
   while (WebPRescalerHasPendingOutput(&p->scaler_a)) {
     int i;
@@ -428,13 +450,14 @@ static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos) {
     WebPRescalerExportRow(&p->scaler_a);
     for (i = 0; i < width; ++i) {
       // Fill in the alpha value (converted to 4 bits).
-      const uint32_t alpha_val = VP8Clip4Bits(p->scaler_a.dst[i]);
+      const uint32_t alpha_value = p->scaler_a.dst[i] >> 4;
-      alpha_dst[2 * i] = (alpha_dst[2 * i] & 0xf0) | alpha_val;
+      alpha_dst[2 * i] = (alpha_dst[2 * i] & 0xf0) | alpha_value;
+      alpha_mask &= alpha_value;
     }
     alpha_dst += buf->stride;
     ++num_lines_out;
   }
-  if (is_premult_alpha) {
+  if (is_premult_alpha && alpha_mask != 0x0f) {
     WebPApplyAlphaMultiply4444(base_rgba, width, num_lines_out, buf->stride);
   }
   return num_lines_out;
@@ -471,8 +494,7 @@ static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) {
     tmp_size1 += work_size;
     tmp_size2 += out_width;
   }
-  p->memory =
+  p->memory = calloc(1, tmp_size1 * sizeof(*work) + tmp_size2 * sizeof(*tmp));
-      calloc(1, tmp_size1 * sizeof(*work) + tmp_size2 * sizeof(*tmp));
   if (p->memory == NULL) {
     return 0;   // memory error
   }
@@ -569,7 +591,7 @@ static int CustomSetup(VP8Io* io) {
 //------------------------------------------------------------------------------
 
 static int CustomPut(const VP8Io* io) {
-  WebPDecParams* p = (WebPDecParams*)io->opaque;
+  WebPDecParams* const p = (WebPDecParams*)io->opaque;
   const int mb_w = io->mb_w;
   const int mb_h = io->mb_h;
   int num_lines_out;

src/dec/vp8i.h

@@ -28,7 +28,7 @@ extern "C" {
 // version numbers
 #define DEC_MAJ_VERSION 0
 #define DEC_MIN_VERSION 2
-#define DEC_REV_VERSION 0
+#define DEC_REV_VERSION 1
 
 #define ONLY_KEYFRAME_CODE      // to remove any code related to P-Frames
 

src/dec/vp8l.c

@@ -615,20 +615,22 @@ static WEBP_INLINE HTreeGroup* GetHtreeGroupForPos(VP8LMetadata* const hdr,
 
 typedef void (*ProcessRowsFunc)(VP8LDecoder* const dec, int row);
 
-static void ApplyTransforms(VP8LDecoder* const dec, int num_rows,
+static void ApplyInverseTransforms(VP8LDecoder* const dec, int num_rows,
-                            const uint32_t* const rows) {
+                                   const uint32_t* const rows) {
   int n = dec->next_transform_;
   const int cache_pixs = dec->width_ * num_rows;
-  uint32_t* rows_data = dec->argb_cache_;
   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_;
 
   // Inverse transforms.
   // TODO: most transforms only need to operate on the cropped region only.
-  memcpy(rows_data, rows, cache_pixs * sizeof(*rows_data));
+  memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out));
   while (n-- > 0) {
     VP8LTransform* const transform = &dec->transforms_[n];
-    VP8LInverseTransform(transform, start_row, end_row, rows, rows_data);
+    VP8LInverseTransform(transform, start_row, end_row, rows_in, rows_out);
+    rows_in = rows_out;
   }
 }
 
@@ -639,7 +641,7 @@ static void ProcessRows(VP8LDecoder* const dec, int row) {
   const int num_rows = row - dec->last_row_;
 
   if (num_rows <= 0) return;  // Nothing to be done.
-  ApplyTransforms(dec, num_rows, rows);
+  ApplyInverseTransforms(dec, num_rows, rows);
 
   // Emit output.
   {
@@ -797,19 +799,6 @@ static void ClearTransform(VP8LTransform* const transform) {
   transform->data_ = NULL;
 }
 
-static void ApplyInverseTransforms(VP8LDecoder* const dec, int start_idx,
-                                   uint32_t* const decoded_data) {
-  int n = dec->next_transform_;
-  assert(start_idx >= 0);
-  while (n-- > start_idx) {
-    VP8LTransform* const transform = &dec->transforms_[n];
-    VP8LInverseTransform(transform, 0, transform->ysize_,
-                         decoded_data, decoded_data);
-    ClearTransform(transform);
-  }
-  dec->next_transform_ = start_idx;
-}
-
 // For security reason, we need to remap the color map to span
 // the total possible bundled values, and not just the num_colors.
 static int ExpandColorMap(int num_colors, VP8LTransform* const transform) {
@@ -964,7 +953,6 @@ static int DecodeImageStream(int xsize, int ysize,
   VP8LBitReader* const br = &dec->br_;
   VP8LMetadata* const hdr = &dec->hdr_;
   uint32_t* data = NULL;
-  const int transform_start_idx = dec->next_transform_;
   int color_cache_bits = 0;
 
   // Read the transforms (may recurse).
@@ -1024,9 +1012,6 @@ static int DecodeImageStream(int xsize, int ysize,
   ok = DecodeImageData(dec, data, transform_xsize, transform_ysize, NULL);
   ok = ok && !br->error_;
 
-  // Apply transforms on the decoded data.
-  if (ok) ApplyInverseTransforms(dec, transform_start_idx, data);
-
  End:
 
   if (!ok) {
@@ -1083,7 +1068,7 @@ static void ExtractAlphaRows(VP8LDecoder* const dec, int row) {
   const uint32_t* const in = dec->argb_ + dec->width_ * dec->last_row_;
 
   if (num_rows <= 0) return;  // Nothing to be done.
-  ApplyTransforms(dec, num_rows, in);
+  ApplyInverseTransforms(dec, num_rows, in);
 
   // Extract alpha (which is stored in the green plane).
   {

src/dec/webp.c

@@ -76,6 +76,9 @@ static VP8StatusCode ParseRIFF(const uint8_t** const data,
       if (size < TAG_SIZE + CHUNK_HEADER_SIZE) {
         return VP8_STATUS_BITSTREAM_ERROR;
       }
+      if (size > MAX_CHUNK_PAYLOAD) {
+        return VP8_STATUS_BITSTREAM_ERROR;
+      }
       // We have a RIFF container. Skip it.
       *riff_size = size;
       *data += RIFF_HEADER_SIZE;
@@ -177,6 +180,9 @@ static VP8StatusCode ParseOptionalChunks(const uint8_t** const data,
     }
 
     chunk_size = get_le32(buf + TAG_SIZE);
+    if (chunk_size > MAX_CHUNK_PAYLOAD) {
+      return VP8_STATUS_BITSTREAM_ERROR;          // Not a valid chunk size.
+    }
     // For odd-sized chunk-payload, there's one byte padding at the end.
     disk_chunk_size = (CHUNK_HEADER_SIZE + chunk_size + 1) & ~1;
     total_size += disk_chunk_size;

src/dsp/dec_neon.c

@@ -12,14 +12,14 @@
 
 #include "./dsp.h"
 
-#if defined(WEBP_USE_NEON)
-
-#include "../dec/vp8i.h"
-
 #if defined(__cplusplus) || defined(c_plusplus)
 extern "C" {
 #endif
 
+#if defined(WEBP_USE_NEON)
+
+#include "../dec/vp8i.h"
+
 #define QRegs "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7",                  \
               "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
 
@@ -311,19 +311,24 @@ static void TransformTwoNEON(const int16_t* in, uint8_t* dst, int do_two) {
   }
 }
 
+#endif   // WEBP_USE_NEON
+
+//------------------------------------------------------------------------------
+// Entry point
+
 extern void VP8DspInitNEON(void);
 
 void VP8DspInitNEON(void) {
+#if defined(WEBP_USE_NEON)
   VP8Transform = TransformTwoNEON;
 
   VP8SimpleVFilter16 = SimpleVFilter16NEON;
   VP8SimpleHFilter16 = SimpleHFilter16NEON;
   VP8SimpleVFilter16i = SimpleVFilter16iNEON;
   VP8SimpleHFilter16i = SimpleHFilter16iNEON;
+#endif   // WEBP_USE_NEON
 }
 
 #if defined(__cplusplus) || defined(c_plusplus)
 }    // extern "C"
 #endif
-
-#endif   // WEBP_USE_NEON

src/dsp/dec_sse2.c

@@ -12,15 +12,15 @@
 
 #include "./dsp.h"
 
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
 #if defined(WEBP_USE_SSE2)
 
 #include <emmintrin.h>
 #include "../dec/vp8i.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
@@ -876,9 +876,15 @@ static void HFilter8iSSE2(uint8_t* u, uint8_t* v, int stride,
   Store16x4(u, v, stride, &p1, &p0, &q0, &q1);
 }
 
+#endif   // WEBP_USE_SSE2
+
+//------------------------------------------------------------------------------
+// Entry point
+
 extern void VP8DspInitSSE2(void);
 
 void VP8DspInitSSE2(void) {
+#if defined(WEBP_USE_SSE2)
   VP8Transform = TransformSSE2;
 
   VP8VFilter16 = VFilter16SSE2;
@@ -894,10 +900,9 @@ void VP8DspInitSSE2(void) {
   VP8SimpleHFilter16 = SimpleHFilter16SSE2;
   VP8SimpleVFilter16i = SimpleVFilter16iSSE2;
   VP8SimpleHFilter16i = SimpleHFilter16iSSE2;
+#endif   // WEBP_USE_SSE2
 }
 
 #if defined(__cplusplus) || defined(c_plusplus)
 }    // extern "C"
 #endif
-
-#endif   // WEBP_USE_SSE2

src/dsp/enc_sse2.c

@@ -11,16 +11,16 @@
 
 #include "./dsp.h"
 
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
 #if defined(WEBP_USE_SSE2)
 #include <stdlib.h>  // for abs()
 #include <emmintrin.h>
 
 #include "../enc/vp8enci.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 //------------------------------------------------------------------------------
 // Compute susceptibility based on DCT-coeff histograms:
 // the higher, the "easier" the macroblock is to compress.
@@ -819,8 +819,15 @@ static int QuantizeBlockSSE2(int16_t in[16], int16_t out[16],
   }
 }
 
+#endif   // WEBP_USE_SSE2
+
+//------------------------------------------------------------------------------
+// Entry point
+
 extern void VP8EncDspInitSSE2(void);
+
 void VP8EncDspInitSSE2(void) {
+#if defined(WEBP_USE_SSE2)
   VP8CollectHistogram = CollectHistogramSSE2;
   VP8EncQuantizeBlock = QuantizeBlockSSE2;
   VP8ITransform = ITransformSSE2;
@@ -828,10 +835,9 @@ void VP8EncDspInitSSE2(void) {
   VP8SSE4x4 = SSE4x4SSE2;
   VP8TDisto4x4 = Disto4x4SSE2;
   VP8TDisto16x16 = Disto16x16SSE2;
+#endif   // WEBP_USE_SSE2
 }
 
 #if defined(__cplusplus) || defined(c_plusplus)
 }    // extern "C"
 #endif
-
-#endif   // WEBP_USE_SSE2

src/dsp/lossless.c

@@ -935,7 +935,7 @@ static void ColorIndexInverseTransform(
       uint32_t packed_pixels = 0;
       int x;
       for (x = 0; x < width; ++x) {
-        // We need to load fresh 'packed_pixels' once every 'bytes_per_pixels'
+        // We need to load fresh 'packed_pixels' once every 'pixels_per_byte'
         // increments of x. Fortunately, pixels_per_byte is a power of 2, so
         // can just use a mask for that, instead of decrementing a counter.
         if ((x & count_mask) == 0) packed_pixels = ((*src++) >> 8) & 0xff;
@@ -976,7 +976,21 @@ void VP8LInverseTransform(const VP8LTransform* const transform,
       ColorSpaceInverseTransform(transform, row_start, row_end, out);
       break;
     case COLOR_INDEXING_TRANSFORM:
-      ColorIndexInverseTransform(transform, row_start, row_end, in, out);
+      if (in == out && transform->bits_ > 0) {
+        // Move packed pixels to the end of unpacked region, so that unpacking
+        // can occur seamlessly.
+        // Also, note that this is the only transform that applies on
+        // the effective width of VP8LSubSampleSize(xsize_, bits_). All other
+        // transforms work on effective width of xsize_.
+        const int out_stride = (row_end - row_start) * transform->xsize_;
+        const int in_stride = (row_end - row_start) *
+            VP8LSubSampleSize(transform->xsize_, transform->bits_);
+        uint32_t* const src = out + out_stride - in_stride;
+        memmove(src, out, in_stride * sizeof(*src));
+        ColorIndexInverseTransform(transform, row_start, row_end, src, out);
+      } else {
+        ColorIndexInverseTransform(transform, row_start, row_end, in, out);
+      }
       break;
   }
 }

src/dsp/upsampling.c

@@ -271,8 +271,7 @@ static void ApplyAlphaMultiply(uint8_t* rgba, int alpha_first,
 
 // rgbA4444
 
-#define MULTIPLIER(a)  ((a) * 0x11)
+#define MULTIPLIER(a)  ((a) * 0x1111)    // 0x1111 ~= (1 << 16) / 15
-#define PREMULTIPLY(x, m) (((x) * (m)) >> 12)
 
 static WEBP_INLINE uint8_t dither_hi(uint8_t x) {
   return (x & 0xf0) | (x >> 4);
@@ -282,24 +281,27 @@ static WEBP_INLINE uint8_t dither_lo(uint8_t x) {
   return (x & 0x0f) | (x << 4);
 }
 
+static WEBP_INLINE uint8_t multiply(uint8_t x, uint32_t m) {
+  return (x * m) >> 16;
+}
+
 static void ApplyAlphaMultiply4444(uint8_t* rgba4444,
                                    int w, int h, int stride) {
   while (h-- > 0) {
     int i;
     for (i = 0; i < w; ++i) {
-      const uint8_t a = dither_lo(rgba4444[2 * i + 1]);
+      const uint8_t a = (rgba4444[2 * i + 1] & 0x0f);
       const uint32_t mult = MULTIPLIER(a);
-      const uint8_t r = PREMULTIPLY(dither_hi(rgba4444[2 * i + 0]), mult);
+      const uint8_t r = multiply(dither_hi(rgba4444[2 * i + 0]), mult);
-      const uint8_t g = PREMULTIPLY(dither_lo(rgba4444[2 * i + 0]), mult);
+      const uint8_t g = multiply(dither_lo(rgba4444[2 * i + 0]), mult);
-      const uint8_t b = PREMULTIPLY(dither_hi(rgba4444[2 * i + 1]), mult);
+      const uint8_t b = multiply(dither_hi(rgba4444[2 * i + 1]), mult);
-      rgba4444[2 * i + 0] = (r & 0xf0) | (g & 0x0f);
+      rgba4444[2 * i + 0] = (r & 0xf0) | ((g >> 4) & 0x0f);
       rgba4444[2 * i + 1] = (b & 0xf0) | a;
     }
     rgba4444 += stride;
   }
 }
 #undef MULTIPLIER
-#undef PREMULTIPLY
 
 void (*WebPApplyAlphaMultiply)(uint8_t*, int, int, int, int)
     = ApplyAlphaMultiply;

src/dsp/upsampling_sse2.c

@@ -11,6 +11,10 @@
 
 #include "./dsp.h"
 
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
 #if defined(WEBP_USE_SSE2)
 
 #include <assert.h>
@@ -18,10 +22,6 @@
 #include <string.h>
 #include "./yuv.h"
 
-#if defined(__cplusplus) || defined(c_plusplus)
-extern "C" {
-#endif
-
 #ifdef FANCY_UPSAMPLING
 
 // We compute (9*a + 3*b + 3*c + d + 8) / 16 as follows
@@ -184,26 +184,32 @@ SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePairSSE2, VP8YuvToBgra, 4)
 #undef CONVERT2RGB
 #undef SSE2_UPSAMPLE_FUNC
 
+#endif  // FANCY_UPSAMPLING
+
+#endif   // WEBP_USE_SSE2
+
 //------------------------------------------------------------------------------
 
 extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
 
 void WebPInitUpsamplersSSE2(void) {
+#if defined(WEBP_USE_SSE2)
   WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePairSSE2;
   WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePairSSE2;
   WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePairSSE2;
   WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePairSSE2;
+#endif   // WEBP_USE_SSE2
 }
 
 void WebPInitPremultiplySSE2(void) {
+#if defined(WEBP_USE_SSE2)
   WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePairSSE2;
   WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePairSSE2;
+#endif   // WEBP_USE_SSE2
 }
 
-#endif  // FANCY_UPSAMPLING
-
 #if defined(__cplusplus) || defined(c_plusplus)
 }    // extern "C"
 #endif
 
-#endif   // WEBP_USE_SSE2
+

src/dsp/yuv.h

@@ -90,11 +90,6 @@ static WEBP_INLINE void VP8YuvToRgba(uint8_t y, uint8_t u, uint8_t v,
   rgba[3] = 0xff;
 }
 
-static WEBP_INLINE uint32_t VP8Clip4Bits(uint8_t c) {
-  const uint32_t v = (c + 8) >> 4;
-  return (v > 15) ? 15 : v;
-}
-
 // Must be called before everything, to initialize the tables.
 void VP8YUVInit(void);
 

src/enc/backward_references.c

@@ -459,8 +459,8 @@ static int BackwardReferencesHashChainDistanceOnly(
   const int quality = 100;
   const int pix_count = xsize * ysize;
   const int use_color_cache = (cache_bits > 0);
-  double* const cost =
+  float* const cost =
-      (double*)WebPSafeMalloc((uint64_t)pix_count, sizeof(*cost));
+      (float*)WebPSafeMalloc((uint64_t)pix_count, sizeof(*cost));
   CostModel* cost_model = (CostModel*)malloc(sizeof(*cost_model));
   HashChain* hash_chain = (HashChain*)malloc(sizeof(*hash_chain));
   VP8LColorCache hashers;
@@ -481,7 +481,7 @@ static int BackwardReferencesHashChainDistanceOnly(
     goto Error;
   }
 
-  for (i = 0; i < pix_count; ++i) cost[i] = 1e100;
+  for (i = 0; i < pix_count; ++i) cost[i] = 1e38f;
 
   // We loop one pixel at a time, but store all currently best points to
   // non-processed locations from this point.
@@ -509,10 +509,9 @@ static int BackwardReferencesHashChainDistanceOnly(
             prev_cost + GetDistanceCost(cost_model, code);
         int k;
         for (k = 1; k < len; ++k) {
-          const double cost_val =
+          const double cost_val = distance_cost + GetLengthCost(cost_model, k);
-              distance_cost + GetLengthCost(cost_model, k);
           if (cost[i + k] > cost_val) {
-            cost[i + k] = cost_val;
+            cost[i + k] = (float)cost_val;
             dist_array[i + k] = k + 1;
           }
         }
@@ -554,7 +553,7 @@ static int BackwardReferencesHashChainDistanceOnly(
         cost_val += GetLiteralCost(cost_model, argb[i]) * mul1;
       }
       if (cost[i] > cost_val) {
-        cost[i] = cost_val;
+        cost[i] = (float)cost_val;
         dist_array[i] = 1;  // only one is inserted.
       }
       if (use_color_cache) VP8LColorCacheInsert(&hashers, argb[i]);

src/enc/histogram.c

@@ -55,9 +55,9 @@ VP8LHistogramSet* VP8LAllocateHistogramSet(int size, int cache_bits) {
   int i;
   VP8LHistogramSet* set;
   VP8LHistogram* bulk;
-  const uint64_t total_size = (uint64_t)sizeof(*set)
+  const uint64_t total_size = sizeof(*set)
-                            + size * sizeof(*set->histograms)
+                            + (uint64_t)size * sizeof(*set->histograms)
-                            + size * sizeof(**set->histograms);
+                            + (uint64_t)size * sizeof(**set->histograms);
   uint8_t* memory = (uint8_t*)WebPSafeMalloc(total_size, sizeof(*memory));
   if (memory == NULL) return NULL;
 

src/enc/vp8enci.h

@@ -27,7 +27,7 @@ extern "C" {
 // version numbers
 #define ENC_MAJ_VERSION 0
 #define ENC_MIN_VERSION 2
-#define ENC_REV_VERSION 0
+#define ENC_REV_VERSION 1
 
 // size of histogram used by CollectHistogram.
 #define MAX_COEFF_THRESH   64

src/enc/vp8l.c

@@ -29,6 +29,7 @@ extern "C" {
 
 #define PALETTE_KEY_RIGHT_SHIFT   22  // Key for 1K buffer.
 #define MAX_HUFF_IMAGE_SIZE       (16 * 1024 * 1024)
+#define MAX_COLORS_FOR_GRAPH      64
 
 // -----------------------------------------------------------------------------
 // Palette
@@ -98,11 +99,11 @@ static int AnalyzeAndCreatePalette(const WebPPicture* const pic,
   return 1;
 }
 
-static int AnalyzeEntropy(const WebPPicture* const pic,
+static int AnalyzeEntropy(const uint32_t* argb,
+                          int width, int height, int argb_stride,
                           double* const nonpredicted_bits,
                           double* const predicted_bits) {
   int x, y;
-  const uint32_t* argb = pic->argb;
   const uint32_t* last_line = NULL;
   uint32_t last_pix = argb[0];    // so we're sure that pix_diff == 0
 
@@ -114,8 +115,8 @@ static int AnalyzeEntropy(const WebPPicture* const pic,
 
   VP8LHistogramInit(predicted, 0);
   VP8LHistogramInit(nonpredicted, 0);
-  for (y = 0; y < pic->height; ++y) {
+  for (y = 0; y < height; ++y) {
-    for (x = 0; x < pic->width; ++x) {
+    for (x = 0; x < width; ++x) {
       const uint32_t pix = argb[x];
       const uint32_t pix_diff = VP8LSubPixels(pix, last_pix);
       if (pix_diff == 0) continue;
@@ -131,7 +132,7 @@ static int AnalyzeEntropy(const WebPPicture* const pic,
       }
     }
     last_line = argb;
-    argb += pic->argb_stride;
+    argb += argb_stride;
   }
   *nonpredicted_bits = VP8LHistogramEstimateBitsBulk(nonpredicted);
   *predicted_bits = VP8LHistogramEstimateBitsBulk(predicted);
@@ -143,24 +144,35 @@ static int VP8LEncAnalyze(VP8LEncoder* const enc, WebPImageHint image_hint) {
   const WebPPicture* const pic = enc->pic_;
   assert(pic != NULL && pic->argb != NULL);
 
-  enc->use_palette_ = (image_hint == WEBP_HINT_GRAPH) ? 0 :
+  enc->use_palette_ =
       AnalyzeAndCreatePalette(pic, enc->palette_, &enc->palette_size_);
-  if (!enc->use_palette_) {
-    if (image_hint == WEBP_HINT_DEFAULT) {
-      double non_pred_entropy, pred_entropy;
-      if (!AnalyzeEntropy(pic, &non_pred_entropy, &pred_entropy)) {
-        return 0;
-      }
 
-      if (pred_entropy < 0.95 * non_pred_entropy) {
+  if (image_hint == WEBP_HINT_GRAPH) {
-        enc->use_predict_ = 1;
+    if (enc->use_palette_ && enc->palette_size_ < MAX_COLORS_FOR_GRAPH) {
-        enc->use_cross_color_ = 1;
+      enc->use_palette_ = 0;
-      }
-    } else if (image_hint == WEBP_HINT_PHOTO) {
-      enc->use_predict_ = 1;
-      enc->use_cross_color_ = 1;
     }
   }
+
+  if (!enc->use_palette_) {
+    if (image_hint == WEBP_HINT_PHOTO) {
+      enc->use_predict_ = 1;
+      enc->use_cross_color_ = 1;
+    } else {
+      double non_pred_entropy, pred_entropy;
+      if (!AnalyzeEntropy(pic->argb, pic->width, pic->height, pic->argb_stride,
+                          &non_pred_entropy, &pred_entropy)) {
+        return 0;
+      }
+      if (pred_entropy < 0.95 * non_pred_entropy) {
+        enc->use_predict_ = 1;
+        // TODO(vikasa): Observed some correlation of cross_color transform with
+        // predict. Need to investigate this further and add separate heuristic
+        // for setting use_cross_color flag.
+        enc->use_cross_color_ = 1;
+      }
+    }
+  }
+
   return 1;
 }
 
@@ -517,7 +529,12 @@ static int EncodeImageInternal(VP8LBitWriter* const bw,
                                 sizeof(*histogram_symbols));
   assert(histogram_bits >= MIN_HUFFMAN_BITS);
   assert(histogram_bits <= MAX_HUFFMAN_BITS);
-  if (histogram_image == NULL || histogram_symbols == NULL) goto Error;
+
+  if (histogram_image == NULL || histogram_symbols == NULL) {
+    free(histogram_image);
+    free(histogram_symbols);
+    return 0;
+  }
 
   // Calculate backward references from ARGB image.
   if (!VP8LGetBackwardReferences(width, height, argb, quality, cache_bits,
@@ -886,13 +903,13 @@ static int GetHistoBits(const WebPConfig* const config,
                         const WebPPicture* const pic) {
   const int width = pic->width;
   const int height = pic->height;
-  const size_t hist_size = sizeof(VP8LHistogram);
+  const uint64_t hist_size = sizeof(VP8LHistogram);
   // Make tile size a function of encoding method (Range: 0 to 6).
   int histo_bits = 7 - config->method;
   while (1) {
-    const size_t huff_image_size = VP8LSubSampleSize(width, histo_bits) *
+    const uint64_t huff_image_size = VP8LSubSampleSize(width, histo_bits) *
-                                   VP8LSubSampleSize(height, histo_bits) *
+                                     VP8LSubSampleSize(height, histo_bits) *
-                                   hist_size;
+                                     hist_size;
     if (huff_image_size <= MAX_HUFF_IMAGE_SIZE) break;
     ++histo_bits;
   }
@@ -961,6 +978,7 @@ WebPEncodingError VP8LEncodeStream(const WebPConfig* const config,
   if (enc->use_palette_) {
     err = ApplyPalette(bw, enc, quality);
     if (err != VP8_ENC_OK) goto Error;
+    // Color cache is disabled for palette.
     enc->cache_bits_ = 0;
   }
 

src/utils/bit_reader.h

@@ -93,6 +93,8 @@ static WEBP_INLINE void VP8LoadNewBytes(VP8BitReader* const br) {
 #elif (BITS == 16)
     // gcc will recognize a 'rorw $8, ...' here:
     bits = (bit_t)(in_bits >> 8) | ((in_bits & 0xff) << 8);
+#else   // BITS == 8
+    bits = (bit_t)in_bits;
 #endif
 #else    // LITTLE_ENDIAN
     bits = (bit_t)in_bits;