Merge "update to v1.2.0" am: 6ea96a06ef am: 7bb22bfa35

Original change: https://android-review.googlesource.com/c/platform/external/webp/+/1566404

MUST ONLY BE SUBMITTED BY AUTOMERGER

Change-Id: I5e6ffc59e8f938df4e4cfca73a9811d4c1329ab4

24 files changed, 164 insertions, 232 deletions

diff --git a/README b/README
index 0baba6ab..bbe2c81b 100644
--- a/README
+++ b/README
@@ -4,7 +4,7 @@
           \__\__/\____/\_____/__/ ____  ___
                 / _/ /    \    \ /  _ \/ _/
                /  \_/   / /   \ \   __/  \__
-               \____/____/\_____/_____/____/v1.1.0
+               \____/____/\_____/_____/____/v1.2.0
 
 Description:
 ============
@@ -113,7 +113,7 @@ make install
 
 CMake:
 ------
-With CMake, you can compile libwebp, cwebp, dwebp, gif2web, img2webp, webpinfo
+With CMake, you can compile libwebp, cwebp, dwebp, gif2webp, img2webp, webpinfo
 and the JS bindings.
 
 Prerequisites:
@@ -225,6 +225,7 @@ Usage:
 
 If input size (-s) for an image is not specified, it is
 assumed to be a PNG, JPEG, TIFF or WebP file.
+Note: Animated PNG and WebP files are not supported.
 
 Options:
   -h / -help ............. short help
@@ -344,7 +345,9 @@ The full list of options is available using -h:
 > dwebp -h
 Usage: dwebp in_file [options] [-o out_file]
 
-Decodes the WebP image file to PNG format [Default]
+Decodes the WebP image file to PNG format [Default].
+Note: Animated WebP files are not supported.
+
 Use following options to convert into alternate image formats:
   -pam ......... save the raw RGBA samples as a color PAM
   -ppm ......... save the raw RGB samples as a color PPM
@@ -426,7 +429,7 @@ Prerequisites:
 1) OpenGL & OpenGL Utility Toolkit (GLUT)
   Linux:
     $ sudo apt-get install freeglut3-dev mesa-common-dev
-  Mac + XCode:
+  Mac + Xcode:
     - These libraries should be available in the OpenGL / GLUT frameworks.
   Windows:
     http://freeglut.sourceforge.net/index.php#download
diff --git a/README.android b/README.android
index 4c9394c3..eb8a65ef 100644
--- a/README.android
+++ b/README.android
@@ -1,5 +1,5 @@
 URL: https://chromium.googlesource.com/webm/libwebp
-Version: v1.1.0-28-g55a080e5
+Version: v1.2.0
 License: Google BSD like
 
 Local modifications:
diff --git a/README.version b/README.version
index 32ba7bd5..fa87b874 100644
--- a/README.version
+++ b/README.version
@@ -1,3 +1,3 @@
-URL: https://chromium.googlesource.com/webm/libwebp/+archive/v1.1.0-28-g55a080e5.tar.gz
-Version: v1.1.0-28-g55a080e5
+URL: https://chromium.googlesource.com/webm/libwebp/+archive/v1.2.0.tar.gz
+Version: v1.2.0
 BugComponent: 20174
diff --git a/include/webp/decode.h b/include/webp/decode.h
index 80dd0ef0..44fcd64a 100644
--- a/include/webp/decode.h
+++ b/include/webp/decode.h
@@ -453,7 +453,7 @@ struct WebPDecoderOptions {
   int scaled_width, scaled_height;    // final resolution
   int use_threads;                    // if true, use multi-threaded decoding
   int dithering_strength;             // dithering strength (0=Off, 100=full)
-  int flip;                           // flip output vertically
+  int flip;                           // if true, flip output vertically
   int alpha_dithering_strength;       // alpha dithering strength in [0..100]
 
   uint32_t pad[5];                    // padding for later use
diff --git a/include/webp/encode.h b/include/webp/encode.h
index 54938dfe..b4c599df 100644
--- a/include/webp/encode.h
+++ b/include/webp/encode.h
@@ -20,7 +20,7 @@
 extern "C" {
 #endif
 
-#define WEBP_ENCODER_ABI_VERSION 0x0210    // MAJOR(8b) + MINOR(8b)
+#define WEBP_ENCODER_ABI_VERSION 0x020f    // MAJOR(8b) + MINOR(8b)
 
 // Note: forward declaring enumerations is not allowed in (strict) C and C++,
 // the types are left here for reference.
@@ -292,6 +292,11 @@ typedef enum WebPEncodingError {
 #define WEBP_MAX_DIMENSION 16383
 
 // Main exchange structure (input samples, output bytes, statistics)
+//
+// Once WebPPictureInit() has been called, it's ok to make all the INPUT fields
+// (use_argb, y/u/v, argb, ...) point to user-owned data, even if
+// WebPPictureAlloc() has been called. Depending on the value use_argb,
+// it's guaranteed that either *argb or *y/*u/*v content will be kept untouched.
 struct WebPPicture {
   //   INPUT
   //////////////
diff --git a/src/dec/io_dec.c b/src/dec/io_dec.c
index 727cce7b..29dc6345 100644
--- a/src/dec/io_dec.c
+++ b/src/dec/io_dec.c
@@ -465,7 +465,7 @@ static int EmitRescaledAlphaRGB(const VP8Io* const io, WebPDecParams* const p,
     int lines_left = expected_num_out_lines;
     const int y_end = p->last_y + lines_left;
     while (lines_left > 0) {
-      const size_t row_offset = scaler->src_y - io->mb_y;
+      const int64_t row_offset = (int64_t)scaler->src_y - io->mb_y;
       WebPRescalerImport(scaler, io->mb_h + io->mb_y - scaler->src_y,
                          io->a + row_offset * io->width, io->width);
       lines_left -= p->emit_alpha_row(p, y_end - lines_left, lines_left);
diff --git a/src/dec/vp8_dec.c b/src/dec/vp8_dec.c
index 57efb690..8f736974 100644
--- a/src/dec/vp8_dec.c
+++ b/src/dec/vp8_dec.c
@@ -494,13 +494,11 @@ static int GetCoeffsAlt(VP8BitReader* const br,
   return 16;
 }
 
-static WEBP_TSAN_IGNORE_FUNCTION void InitGetCoeffs(void) {
-  if (GetCoeffs == NULL) {
-    if (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kSlowSSSE3)) {
-      GetCoeffs = GetCoeffsAlt;
-    } else {
-      GetCoeffs = GetCoeffsFast;
-    }
+WEBP_DSP_INIT_FUNC(InitGetCoeffs) {
+  if (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kSlowSSSE3)) {
+    GetCoeffs = GetCoeffsAlt;
+  } else {
+    GetCoeffs = GetCoeffsFast;
   }
 }
 
diff --git a/src/dec/vp8i_dec.h b/src/dec/vp8i_dec.h
index 600a6844..a0c0af15 100644
--- a/src/dec/vp8i_dec.h
+++ b/src/dec/vp8i_dec.h
@@ -31,7 +31,7 @@ extern "C" {
 
 // version numbers
 #define DEC_MAJ_VERSION 1
-#define DEC_MIN_VERSION 1
+#define DEC_MIN_VERSION 2
 #define DEC_REV_VERSION 0
 
 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
diff --git a/src/dec/vp8l_dec.c b/src/dec/vp8l_dec.c
index 93615d4e..2d603b43 100644
--- a/src/dec/vp8l_dec.c
+++ b/src/dec/vp8l_dec.c
@@ -947,7 +947,6 @@ static WEBP_INLINE void CopyBlock8b(uint8_t* const dst, int dist, int length) {
         break;
       default:
         goto Copy;
-        break;
     }
     CopySmallPattern8b(src, dst, length, pattern);
     return;
diff --git a/src/demux/demux.c b/src/demux/demux.c
index 1b3cc2e0..860e2ce7 100644
--- a/src/demux/demux.c
+++ b/src/demux/demux.c
@@ -24,7 +24,7 @@
 #include "src/webp/format_constants.h"
 
 #define DMUX_MAJ_VERSION 1
-#define DMUX_MIN_VERSION 1
+#define DMUX_MIN_VERSION 2
 #define DMUX_REV_VERSION 0
 
 typedef struct {
@@ -312,6 +312,7 @@ static ParseStatus ParseAnimationFrame(
   int bits;
   MemBuffer* const mem = &dmux->mem_;
   Frame* frame;
+  size_t start_offset;
   ParseStatus status =
       NewFrame(mem, ANMF_CHUNK_SIZE, frame_chunk_size, &frame);
   if (status != PARSE_OK) return status;
@@ -332,7 +333,11 @@ static ParseStatus ParseAnimationFrame(
 
   // Store a frame only if the animation flag is set there is some data for
   // this frame is available.
+  start_offset = mem->start_;
   status = StoreFrame(dmux->num_frames_ + 1, anmf_payload_size, mem, frame);
+  if (status != PARSE_ERROR && mem->start_ - start_offset > anmf_payload_size) {
+    status = PARSE_ERROR;
+  }
   if (status != PARSE_ERROR && is_animation && frame->frame_num_ > 0) {
     added_frame = AddFrame(dmux, frame);
     if (added_frame) {
diff --git a/src/dsp/cpu.c b/src/dsp/cpu.c
index 0fa5b6a5..4ca90d88 100644
--- a/src/dsp/cpu.c
+++ b/src/dsp/cpu.c
@@ -55,12 +55,18 @@ static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
     : "=a"(cpu_info[0]), "=b"(cpu_info[1]), "=c"(cpu_info[2]), "=d"(cpu_info[3])
     : "a"(info_type), "c"(0));
 }
-#elif (defined(_M_X64) || defined(_M_IX86)) && \
-      defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 150030729  // >= VS2008 SP1
+#elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
+
+#if defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 150030729  // >= VS2008 SP1
 #include <intrin.h>
 #define GetCPUInfo(info, type) __cpuidex(info, type, 0)  // set ecx=0
-#elif defined(WEBP_MSC_SSE2)
+#define WEBP_HAVE_MSC_CPUID
+#elif _MSC_VER > 1310
+#include <intrin.h>
 #define GetCPUInfo __cpuid
+#define WEBP_HAVE_MSC_CPUID
+#endif
+
 #endif
 
 // NaCl has no support for xgetbv or the raw opcode.
@@ -94,7 +100,7 @@ static WEBP_INLINE uint64_t xgetbv(void) {
 #define xgetbv() 0U  // no AVX for older x64 or unrecognized toolchains.
 #endif
 
-#if defined(__i386__) || defined(__x86_64__) || defined(WEBP_MSC_SSE2)
+#if defined(__i386__) || defined(__x86_64__) || defined(WEBP_HAVE_MSC_CPUID)
 
 // helper function for run-time detection of slow SSSE3 platforms
 static int CheckSlowModel(int info) {
@@ -179,6 +185,30 @@ static int AndroidCPUInfo(CPUFeature feature) {
   return 0;
 }
 VP8CPUInfo VP8GetCPUInfo = AndroidCPUInfo;
+#elif defined(EMSCRIPTEN) // also needs to be before generic NEON test
+// Use compile flags as an indicator of SIMD support instead of a runtime check.
+static int wasmCPUInfo(CPUFeature feature) {
+  switch (feature) {
+#ifdef WEBP_USE_SSE2
+    case kSSE2:
+      return 1;
+#endif
+#ifdef WEBP_USE_SSE41
+    case kSSE3:
+    case kSlowSSSE3:
+    case kSSE4_1:
+      return 1;
+#endif
+#ifdef WEBP_USE_NEON
+    case kNEON:
+      return 1;
+#endif
+    default:
+      break;
+  }
+  return 0;
+}
+VP8CPUInfo VP8GetCPUInfo = wasmCPUInfo;
 #elif defined(WEBP_USE_NEON)
 // define a dummy function to enable turning off NEON at runtime by setting
 // VP8DecGetCPUInfo = NULL
diff --git a/src/dsp/dsp.h b/src/dsp/dsp.h
index cf50bb13..298c721a 100644
--- a/src/dsp/dsp.h
+++ b/src/dsp/dsp.h
@@ -51,9 +51,7 @@ extern "C" {
 # define __has_builtin(x) 0
 #endif
 
-// for now, none of the optimizations below are available in emscripten
-#if !defined(EMSCRIPTEN)
-
+#if !defined(HAVE_CONFIG_H)
 #if defined(_MSC_VER) && _MSC_VER > 1310 && \
     (defined(_M_X64) || defined(_M_IX86))
 #define WEBP_MSC_SSE2  // Visual C++ SSE2 targets
@@ -63,6 +61,7 @@ extern "C" {
     (defined(_M_X64) || defined(_M_IX86))
 #define WEBP_MSC_SSE41  // Visual C++ SSE4.1 targets
 #endif
+#endif
 
 // WEBP_HAVE_* are used to indicate the presence of the instruction set in dsp
 // files without intrinsics, allowing the corresponding Init() to be called.
@@ -76,6 +75,9 @@ extern "C" {
 #define WEBP_USE_SSE41
 #endif
 
+#undef WEBP_MSC_SSE41
+#undef WEBP_MSC_SSE2
+
 // The intrinsics currently cause compiler errors with arm-nacl-gcc and the
 // inline assembly would need to be modified for use with Native Client.
 #if (defined(__ARM_NEON__) || \
@@ -110,8 +112,6 @@ extern "C" {
 #define WEBP_USE_MSA
 #endif
 
-#endif  /* EMSCRIPTEN */
-
 #ifndef WEBP_DSP_OMIT_C_CODE
 #define WEBP_DSP_OMIT_C_CODE 1
 #endif
diff --git a/src/dsp/lossless.c b/src/dsp/lossless.c
index aad5f43e..46b220e2 100644
--- a/src/dsp/lossless.c
+++ b/src/dsp/lossless.c
@@ -107,62 +107,62 @@ static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
 //------------------------------------------------------------------------------
 // Predictors
 
-static uint32_t Predictor0_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor0_C(uint32_t left, const uint32_t* const top) {
   (void)top;
   (void)left;
   return ARGB_BLACK;
 }
-static uint32_t Predictor1_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor1_C(uint32_t left, const uint32_t* const top) {
   (void)top;
   return left;
 }
-static uint32_t Predictor2_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor2_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[0];
 }
-static uint32_t Predictor3_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor3_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[1];
 }
-static uint32_t Predictor4_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor4_C(uint32_t left, const uint32_t* const top) {
   (void)left;
   return top[-1];
 }
-static uint32_t Predictor5_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor5_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average3(left, top[0], top[1]);
   return pred;
 }
-static uint32_t Predictor6_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor6_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(left, top[-1]);
   return pred;
 }
-static uint32_t Predictor7_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor7_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(left, top[0]);
   return pred;
 }
-static uint32_t Predictor8_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor8_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(top[-1], top[0]);
   (void)left;
   return pred;
 }
-static uint32_t Predictor9_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor9_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average2(top[0], top[1]);
   (void)left;
   return pred;
 }
-static uint32_t Predictor10_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor10_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Average4(left, top[-1], top[0], top[1]);
   return pred;
 }
-static uint32_t Predictor11_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor11_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = Select(top[0], left, top[-1]);
   return pred;
 }
-static uint32_t Predictor12_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor12_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]);
   return pred;
 }
-static uint32_t Predictor13_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor13_C(uint32_t left, const uint32_t* const top) {
   const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]);
   return pred;
 }
@@ -182,18 +182,18 @@ static void PredictorAdd1_C(const uint32_t* in, const uint32_t* upper,
     out[i] = left = VP8LAddPixels(in[i], left);
   }
 }
-GENERATE_PREDICTOR_ADD(Predictor2_C, PredictorAdd2_C)
-GENERATE_PREDICTOR_ADD(Predictor3_C, PredictorAdd3_C)
-GENERATE_PREDICTOR_ADD(Predictor4_C, PredictorAdd4_C)
-GENERATE_PREDICTOR_ADD(Predictor5_C, PredictorAdd5_C)
-GENERATE_PREDICTOR_ADD(Predictor6_C, PredictorAdd6_C)
-GENERATE_PREDICTOR_ADD(Predictor7_C, PredictorAdd7_C)
-GENERATE_PREDICTOR_ADD(Predictor8_C, PredictorAdd8_C)
-GENERATE_PREDICTOR_ADD(Predictor9_C, PredictorAdd9_C)
-GENERATE_PREDICTOR_ADD(Predictor10_C, PredictorAdd10_C)
-GENERATE_PREDICTOR_ADD(Predictor11_C, PredictorAdd11_C)
-GENERATE_PREDICTOR_ADD(Predictor12_C, PredictorAdd12_C)
-GENERATE_PREDICTOR_ADD(Predictor13_C, PredictorAdd13_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor2_C, PredictorAdd2_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor3_C, PredictorAdd3_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor4_C, PredictorAdd4_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor5_C, PredictorAdd5_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor6_C, PredictorAdd6_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor7_C, PredictorAdd7_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor8_C, PredictorAdd8_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor9_C, PredictorAdd9_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor10_C, PredictorAdd10_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor11_C, PredictorAdd11_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor12_C, PredictorAdd12_C)
+GENERATE_PREDICTOR_ADD(VP8LPredictor13_C, PredictorAdd13_C)
 
 //------------------------------------------------------------------------------
 
@@ -562,7 +562,6 @@ VP8LPredictorFunc VP8LPredictors[16];
 
 // exposed plain-C implementations
 VP8LPredictorAddSubFunc VP8LPredictorsAdd_C[16];
-VP8LPredictorFunc VP8LPredictors_C[16];
 
 VP8LTransformColorInverseFunc VP8LTransformColorInverse;
 
@@ -600,8 +599,7 @@ extern void VP8LDspInitMSA(void);
 } while (0);
 
 WEBP_DSP_INIT_FUNC(VP8LDspInit) {
-  COPY_PREDICTOR_ARRAY(Predictor, VP8LPredictors)
-  COPY_PREDICTOR_ARRAY(Predictor, VP8LPredictors_C)
+  COPY_PREDICTOR_ARRAY(VP8LPredictor, VP8LPredictors)
   COPY_PREDICTOR_ARRAY(PredictorAdd, VP8LPredictorsAdd)
   COPY_PREDICTOR_ARRAY(PredictorAdd, VP8LPredictorsAdd_C)
 
diff --git a/src/dsp/lossless.h b/src/dsp/lossless.h
index f709cc86..ebd316d1 100644
--- a/src/dsp/lossless.h
+++ b/src/dsp/lossless.h
@@ -30,7 +30,22 @@ extern "C" {
 
 typedef uint32_t (*VP8LPredictorFunc)(uint32_t left, const uint32_t* const top);
 extern VP8LPredictorFunc VP8LPredictors[16];
-extern VP8LPredictorFunc VP8LPredictors_C[16];
+
+uint32_t VP8LPredictor0_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor1_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor2_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor3_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor4_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor5_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor6_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor7_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor8_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor9_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor10_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor11_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor12_C(uint32_t left, const uint32_t* const top);
+uint32_t VP8LPredictor13_C(uint32_t left, const uint32_t* const top);
+
 // These Add/Sub function expects upper[-1] and out[-1] to be readable.
 typedef void (*VP8LPredictorAddSubFunc)(const uint32_t* in,
                                         const uint32_t* upper, int num_pixels,
diff --git a/src/dsp/lossless_common.h b/src/dsp/lossless_common.h
index 9c2ebe68..96a106f9 100644
--- a/src/dsp/lossless_common.h
+++ b/src/dsp/lossless_common.h
@@ -184,19 +184,6 @@ static void PREDICTOR_ADD(const uint32_t* in, const uint32_t* upper, \
   }                                                                  \
 }
 
-// It subtracts the prediction from the input pixel and stores the residual
-// in the output pixel.
-#define GENERATE_PREDICTOR_SUB(PREDICTOR, PREDICTOR_SUB)             \
-static void PREDICTOR_SUB(const uint32_t* in, const uint32_t* upper, \
-                          int num_pixels, uint32_t* out) {           \
-  int x;                                                             \
-  assert(upper != NULL);                                             \
-  for (x = 0; x < num_pixels; ++x) {                                 \
-    const uint32_t pred = (PREDICTOR)(in[x - 1], upper + x);         \
-    out[x] = VP8LSubPixels(in[x], pred);                             \
-  }                                                                  \
-}
-
 #ifdef __cplusplus
 }    // extern "C"
 #endif
diff --git a/src/dsp/lossless_enc.c b/src/dsp/lossless_enc.c
index 9c36055a..a0c7ab91 100644
--- a/src/dsp/lossless_enc.c
+++ b/src/dsp/lossless_enc.c
@@ -702,140 +702,6 @@ void VP8LHistogramAdd(const VP8LHistogram* const a,
 //------------------------------------------------------------------------------
 // Image transforms.
 
-static WEBP_INLINE uint32_t Average2(uint32_t a0, uint32_t a1) {
-  return (((a0 ^ a1) & 0xfefefefeu) >> 1) + (a0 & a1);
-}
-
-static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) {
-  return Average2(Average2(a0, a2), a1);
-}
-
-static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
-                                     uint32_t a2, uint32_t a3) {
-  return Average2(Average2(a0, a1), Average2(a2, a3));
-}
-
-static WEBP_INLINE uint32_t Clip255(uint32_t a) {
-  if (a < 256) {
-    return a;
-  }
-  // return 0, when a is a negative integer.
-  // return 255, when a is positive.
-  return ~a >> 24;
-}
-
-static WEBP_INLINE int AddSubtractComponentFull(int a, int b, int c) {
-  return Clip255(a + b - c);
-}
-
-static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
-                                                   uint32_t c2) {
-  const int a = AddSubtractComponentFull(c0 >> 24, c1 >> 24, c2 >> 24);
-  const int r = AddSubtractComponentFull((c0 >> 16) & 0xff,
-                                         (c1 >> 16) & 0xff,
-                                         (c2 >> 16) & 0xff);
-  const int g = AddSubtractComponentFull((c0 >> 8) & 0xff,
-                                         (c1 >> 8) & 0xff,
-                                         (c2 >> 8) & 0xff);
-  const int b = AddSubtractComponentFull(c0 & 0xff, c1 & 0xff, c2 & 0xff);
-  return ((uint32_t)a << 24) | (r << 16) | (g << 8) | b;
-}
-
-static WEBP_INLINE int AddSubtractComponentHalf(int a, int b) {
-  return Clip255(a + (a - b) / 2);
-}
-
-static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
-                                                   uint32_t c2) {
-  const uint32_t ave = Average2(c0, c1);
-  const int a = AddSubtractComponentHalf(ave >> 24, c2 >> 24);
-  const int r = AddSubtractComponentHalf((ave >> 16) & 0xff, (c2 >> 16) & 0xff);
-  const int g = AddSubtractComponentHalf((ave >> 8) & 0xff, (c2 >> 8) & 0xff);
-  const int b = AddSubtractComponentHalf((ave >> 0) & 0xff, (c2 >> 0) & 0xff);
-  return ((uint32_t)a << 24) | (r << 16) | (g << 8) | b;
-}
-
-// gcc-4.9 on ARM generates incorrect code in Select() when Sub3() is inlined.
-#if defined(__arm__) && \
-    (LOCAL_GCC_VERSION == 0x409 || LOCAL_GCC_VERSION == 0x408)
-# define LOCAL_INLINE __attribute__ ((noinline))
-#else
-# define LOCAL_INLINE WEBP_INLINE
-#endif
-
-static LOCAL_INLINE int Sub3(int a, int b, int c) {
-  const int pb = b - c;
-  const int pa = a - c;
-  return abs(pb) - abs(pa);
-}
-
-#undef LOCAL_INLINE
-
-static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
-  const int pa_minus_pb =
-      Sub3((a >> 24)       , (b >> 24)       , (c >> 24)       ) +
-      Sub3((a >> 16) & 0xff, (b >> 16) & 0xff, (c >> 16) & 0xff) +
-      Sub3((a >>  8) & 0xff, (b >>  8) & 0xff, (c >>  8) & 0xff) +
-      Sub3((a      ) & 0xff, (b      ) & 0xff, (c      ) & 0xff);
-  return (pa_minus_pb <= 0) ? a : b;
-}
-
-//------------------------------------------------------------------------------
-// Predictors
-
-static uint32_t Predictor2(uint32_t left, const uint32_t* const top) {
-  (void)left;
-  return top[0];
-}
-static uint32_t Predictor3(uint32_t left, const uint32_t* const top) {
-  (void)left;
-  return top[1];
-}
-static uint32_t Predictor4(uint32_t left, const uint32_t* const top) {
-  (void)left;
-  return top[-1];
-}
-static uint32_t Predictor5(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average3(left, top[0], top[1]);
-  return pred;
-}
-static uint32_t Predictor6(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(left, top[-1]);
-  return pred;
-}
-static uint32_t Predictor7(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(left, top[0]);
-  return pred;
-}
-static uint32_t Predictor8(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(top[-1], top[0]);
-  (void)left;
-  return pred;
-}
-static uint32_t Predictor9(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(top[0], top[1]);
-  (void)left;
-  return pred;
-}
-static uint32_t Predictor10(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average4(left, top[-1], top[0], top[1]);
-  return pred;
-}
-static uint32_t Predictor11(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Select(top[0], left, top[-1]);
-  return pred;
-}
-static uint32_t Predictor12(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]);
-  return pred;
-}
-static uint32_t Predictor13(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]);
-  return pred;
-}
-
-//------------------------------------------------------------------------------
-
 static void PredictorSub0_C(const uint32_t* in, const uint32_t* upper,
                             int num_pixels, uint32_t* out) {
   int i;
@@ -850,18 +716,33 @@ static void PredictorSub1_C(const uint32_t* in, const uint32_t* upper,
   (void)upper;
 }
 
-GENERATE_PREDICTOR_SUB(Predictor2, PredictorSub2_C)
-GENERATE_PREDICTOR_SUB(Predictor3, PredictorSub3_C)
-GENERATE_PREDICTOR_SUB(Predictor4, PredictorSub4_C)
-GENERATE_PREDICTOR_SUB(Predictor5, PredictorSub5_C)
-GENERATE_PREDICTOR_SUB(Predictor6, PredictorSub6_C)
-GENERATE_PREDICTOR_SUB(Predictor7, PredictorSub7_C)
-GENERATE_PREDICTOR_SUB(Predictor8, PredictorSub8_C)
-GENERATE_PREDICTOR_SUB(Predictor9, PredictorSub9_C)
-GENERATE_PREDICTOR_SUB(Predictor10, PredictorSub10_C)
-GENERATE_PREDICTOR_SUB(Predictor11, PredictorSub11_C)
-GENERATE_PREDICTOR_SUB(Predictor12, PredictorSub12_C)
-GENERATE_PREDICTOR_SUB(Predictor13, PredictorSub13_C)
+// It subtracts the prediction from the input pixel and stores the residual
+// in the output pixel.
+#define GENERATE_PREDICTOR_SUB(PREDICTOR_I)                                \
+static void PredictorSub##PREDICTOR_I##_C(const uint32_t* in,              \
+                                          const uint32_t* upper,           \
+                                          int num_pixels, uint32_t* out) { \
+  int x;                                                                   \
+  assert(upper != NULL);                                                   \
+  for (x = 0; x < num_pixels; ++x) {                                       \
+    const uint32_t pred =                                                  \
+        VP8LPredictor##PREDICTOR_I##_C(in[x - 1], upper + x);              \
+    out[x] = VP8LSubPixels(in[x], pred);                                   \
+  }                                                                        \
+}
+
+GENERATE_PREDICTOR_SUB(2)
+GENERATE_PREDICTOR_SUB(3)
+GENERATE_PREDICTOR_SUB(4)
+GENERATE_PREDICTOR_SUB(5)
+GENERATE_PREDICTOR_SUB(6)
+GENERATE_PREDICTOR_SUB(7)
+GENERATE_PREDICTOR_SUB(8)
+GENERATE_PREDICTOR_SUB(9)
+GENERATE_PREDICTOR_SUB(10)
+GENERATE_PREDICTOR_SUB(11)
+GENERATE_PREDICTOR_SUB(12)
+GENERATE_PREDICTOR_SUB(13)
 
 //------------------------------------------------------------------------------
 
diff --git a/src/dsp/lossless_enc_sse2.c b/src/dsp/lossless_enc_sse2.c
index 68d1ac3c..90c26373 100644
--- a/src/dsp/lossless_enc_sse2.c
+++ b/src/dsp/lossless_enc_sse2.c
@@ -249,6 +249,7 @@ static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
     }                                  \
   } while (0)
 
+#if !(defined(__i386__) || defined(_M_IX86))
 static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
   int i;
   double retval = 0.;
@@ -300,6 +301,8 @@ static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
   retval += VP8LFastSLog2(sumX) + VP8LFastSLog2(sumXY);
   return (float)retval;
 }
+#endif  // !(defined(__i386__) || defined(_M_IX86))
+
 #undef ANALYZE_X_OR_Y
 #undef ANALYZE_XY
 
@@ -659,7 +662,12 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE2(void) {
   VP8LCollectColorRedTransforms = CollectColorRedTransforms_SSE2;
   VP8LAddVector = AddVector_SSE2;
   VP8LAddVectorEq = AddVectorEq_SSE2;
+  // TODO(https://crbug.com/webp/499): this function produces different results
+  // from the C code due to use of double/float resulting in output differences
+  // when compared to -noasm.
+#if !(defined(__i386__) || defined(_M_IX86))
   VP8LCombinedShannonEntropy = CombinedShannonEntropy_SSE2;
+#endif
   VP8LVectorMismatch = VectorMismatch_SSE2;
   VP8LBundleColorMap = BundleColorMap_SSE2;
 
diff --git a/src/enc/picture_csp_enc.c b/src/enc/picture_csp_enc.c
index 718e014e..35eede96 100644
--- a/src/enc/picture_csp_enc.c
+++ b/src/enc/picture_csp_enc.c
@@ -61,16 +61,14 @@ static int CheckNonOpaque(const uint8_t* alpha, int width, int height,
 // Checking for the presence of non-opaque alpha.
 int WebPPictureHasTransparency(const WebPPicture* picture) {
   if (picture == NULL) return 0;
-  if (!picture->use_argb) {
-    return CheckNonOpaque(picture->a, picture->width, picture->height,
-                          1, picture->a_stride);
-  } else {
+  if (picture->use_argb) {
     const int alpha_offset = ALPHA_OFFSET;
     return CheckNonOpaque((const uint8_t*)picture->argb + alpha_offset,
                           picture->width, picture->height,
                           4, picture->argb_stride * sizeof(*picture->argb));
   }
-  return 0;
+  return CheckNonOpaque(picture->a, picture->width, picture->height,
+                        1, picture->a_stride);
 }
 
 //------------------------------------------------------------------------------
@@ -90,8 +88,9 @@ int WebPPictureHasTransparency(const WebPPicture* picture) {
 static int kLinearToGammaTab[kGammaTabSize + 1];
 static uint16_t kGammaToLinearTab[256];
 static volatile int kGammaTablesOk = 0;
+static void InitGammaTables(void);
 
-static WEBP_TSAN_IGNORE_FUNCTION void InitGammaTables(void) {
+WEBP_DSP_INIT_FUNC(InitGammaTables) {
   if (!kGammaTablesOk) {
     int v;
     const double scale = (double)(1 << kGammaTabFix) / kGammaScale;
@@ -181,8 +180,9 @@ static uint32_t kLinearToGammaTabS[kGammaTabSize + 2];
 #define GAMMA_TO_LINEAR_BITS 14
 static uint32_t kGammaToLinearTabS[MAX_Y_T + 1];   // size scales with Y_FIX
 static volatile int kGammaTablesSOk = 0;
+static void InitGammaTablesS(void);
 
-static WEBP_TSAN_IGNORE_FUNCTION void InitGammaTablesS(void) {
+WEBP_DSP_INIT_FUNC(InitGammaTablesS) {
   assert(2 * GAMMA_TO_LINEAR_BITS < 32);  // we use uint32_t intermediate values
   if (!kGammaTablesSOk) {
     int v;
diff --git a/src/enc/picture_tools_enc.c b/src/enc/picture_tools_enc.c
index 5cc25c5f..38cb0153 100644
--- a/src/enc/picture_tools_enc.c
+++ b/src/enc/picture_tools_enc.c
@@ -83,7 +83,7 @@ static int SmoothenBlock(const uint8_t* a_ptr, int a_stride, uint8_t* y_ptr,
   return (count == 0);
 }
 
-void WebPReplaceTransparentPixels(WebPPicture* pic, uint32_t color) {
+void WebPReplaceTransparentPixels(WebPPicture* const pic, uint32_t color) {
   if (pic != NULL && pic->use_argb) {
     int y = pic->height;
     uint32_t* argb = pic->argb;
diff --git a/src/enc/vp8i_enc.h b/src/enc/vp8i_enc.h
index fc3b73be..0e35562a 100644
--- a/src/enc/vp8i_enc.h
+++ b/src/enc/vp8i_enc.h
@@ -31,7 +31,7 @@ extern "C" {
 
 // version numbers
 #define ENC_MAJ_VERSION 1
-#define ENC_MIN_VERSION 1
+#define ENC_MIN_VERSION 2
 #define ENC_REV_VERSION 0
 
 enum { MAX_LF_LEVELS = 64,       // Maximum loop filter level
diff --git a/src/mux/muxi.h b/src/mux/muxi.h
index ad3e1bdb..2289822e 100644
--- a/src/mux/muxi.h
+++ b/src/mux/muxi.h
@@ -28,7 +28,7 @@ extern "C" {
 // Defines and constants.
 
 #define MUX_MAJ_VERSION 1
-#define MUX_MIN_VERSION 1
+#define MUX_MIN_VERSION 2
 #define MUX_REV_VERSION 0
 
 // Chunk object.
diff --git a/src/mux/muxread.c b/src/mux/muxread.c
index ae3b876b..0101fde1 100644
--- a/src/mux/muxread.c
+++ b/src/mux/muxread.c
@@ -155,7 +155,6 @@ static int MuxImageParse(const WebPChunk* const chunk, int copy_data,
         break;
       default:
         goto Fail;
-        break;
     }
     subchunk_size = ChunkDiskSize(&subchunk);
     bytes += subchunk_size;
@@ -264,7 +263,6 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
         if (!MuxImageParse(&chunk, copy_data, wpi)) goto Err;
         ChunkRelease(&chunk);
         goto PushImage;
-        break;
       default:  // A non-image chunk.
         if (wpi->is_partial_) goto Err;  // Encountered a non-image chunk before
                                          // getting all chunks of an image.
diff --git a/src/webp/decode.h b/src/webp/decode.h
index 80dd0ef0..44fcd64a 100644
--- a/src/webp/decode.h
+++ b/src/webp/decode.h
@@ -453,7 +453,7 @@ struct WebPDecoderOptions {
   int scaled_width, scaled_height;    // final resolution
   int use_threads;                    // if true, use multi-threaded decoding
   int dithering_strength;             // dithering strength (0=Off, 100=full)
-  int flip;                           // flip output vertically
+  int flip;                           // if true, flip output vertically
   int alpha_dithering_strength;       // alpha dithering strength in [0..100]
 
   uint32_t pad[5];                    // padding for later use
diff --git a/src/webp/encode.h b/src/webp/encode.h
index 54938dfe..b4c599df 100644
--- a/src/webp/encode.h
+++ b/src/webp/encode.h
@@ -20,7 +20,7 @@
 extern "C" {
 #endif
 
-#define WEBP_ENCODER_ABI_VERSION 0x0210    // MAJOR(8b) + MINOR(8b)
+#define WEBP_ENCODER_ABI_VERSION 0x020f    // MAJOR(8b) + MINOR(8b)
 
 // Note: forward declaring enumerations is not allowed in (strict) C and C++,
 // the types are left here for reference.
@@ -292,6 +292,11 @@ typedef enum WebPEncodingError {
 #define WEBP_MAX_DIMENSION 16383
 
 // Main exchange structure (input samples, output bytes, statistics)
+//
+// Once WebPPictureInit() has been called, it's ok to make all the INPUT fields
+// (use_argb, y/u/v, argb, ...) point to user-owned data, even if
+// WebPPictureAlloc() has been called. Depending on the value use_argb,
+// it's guaranteed that either *argb or *y/*u/*v content will be kept untouched.
 struct WebPPicture {
   //   INPUT
   //////////////