Merge pull request #24274 from vrabaud:webp_1.3.2

Merge pull request #24274 from vrabaud:webp_1.3.2
 
Bump libwebp to 1.3.2 #24274

This is version [c1ffd9a](c1ffd9ac75)
It is 1.3.2 with a few patches that were made right after to help compilation.
No need for patches on the OpenCV side!

### Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch

3rdparty/libwebp/CMakeLists.txt

@@ -9,8 +9,8 @@ if(ANDROID)
   ocv_include_directories(${CPUFEATURES_INCLUDE_DIRS})
 endif()
 
-file(GLOB lib_srcs src/dec/*.c src/demux/*.c src/dsp/*.c src/enc/*.c src/mux/*.c src/utils/*.c src/webp/*.c)
-file(GLOB lib_hdrs src/dec/*.h src/demux/*.h src/dsp/*.h src/enc/*.h src/mux/*.h src/utils/*.h src/webp/*.h)
+file(GLOB lib_srcs sharpyuv/*.c src/dec/*.c src/demux/*.c src/dsp/*.c src/enc/*.c src/mux/*.c src/utils/*.c src/webp/*.c)
+file(GLOB lib_hdrs sharpyuv/*.h src/dec/*.h src/demux/*.h src/dsp/*.h src/enc/*.h src/mux/*.h src/utils/*.h src/webp/*.h)
 
 # FIXIT
 if(ANDROID AND ARMEABI_V7A AND NOT NEON)
@@ -21,12 +21,6 @@ if(ANDROID AND ARMEABI_V7A AND NOT NEON)
   endforeach()
 endif()
 
-# FIX for quant.h - requires C99 for() loops
-ocv_check_flag_support(C "-std=c99" _varname "${CMAKE_C_FLAGS}")
-if(${_varname})
-  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -std=c99")
-endif()
-
 
 # ----------------------------------------------------------------------------------
 #         Define the library target:

3rdparty/libwebp/patches/20190910-msa-asm-patch.diff

@@ -1,22 +0,0 @@
-diff --git a/3rdparty/libwebp/src/dsp/msa_macro.h b/3rdparty/libwebp/src/dsp/msa_macro.h
-index de026a1d9e..a16c0bb300 100644
---- a/3rdparty/libwebp/src/dsp/msa_macro.h
-+++ b/3rdparty/libwebp/src/dsp/msa_macro.h
-@@ -73,7 +73,7 @@
-   static inline TYPE FUNC_NAME(const void* const psrc) {  \
-     const uint8_t* const psrc_m = (const uint8_t*)psrc;   \
-     TYPE val_m;                                           \
--    asm volatile (                                        \
-+    __asm__ volatile (                                        \
-       "" #INSTR " %[val_m], %[psrc_m]  \n\t"              \
-       : [val_m] "=r" (val_m)                              \
-       : [psrc_m] "m" (*psrc_m));                          \
-@@ -86,7 +86,7 @@
-   static inline void FUNC_NAME(TYPE val, void* const pdst) { \
-     uint8_t* const pdst_m = (uint8_t*)pdst;                  \
-     TYPE val_m = val;                                        \
--    asm volatile (                                           \
-+    __asm__ volatile (                                           \
-       " " #INSTR "  %[val_m],  %[pdst_m]  \n\t"              \
-       : [pdst_m] "=m" (*pdst_m)                              \
-       : [val_m] "r" (val_m));                                \

3rdparty/libwebp/sharpyuv/sharpyuv.c

@@ -0,0 +1,565 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Sharp RGB to YUV conversion.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "sharpyuv/sharpyuv.h"
+
+#include <assert.h>
+#include <limits.h>
+#include <stddef.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "src/webp/types.h"
+#include "sharpyuv/sharpyuv_cpu.h"
+#include "sharpyuv/sharpyuv_dsp.h"
+#include "sharpyuv/sharpyuv_gamma.h"
+
+//------------------------------------------------------------------------------
+
+int SharpYuvGetVersion(void) {
+  return SHARPYUV_VERSION;
+}
+
+//------------------------------------------------------------------------------
+// Sharp RGB->YUV conversion
+
+static const int kNumIterations = 4;
+
+#define YUV_FIX 16  // fixed-point precision for RGB->YUV
+static const int kYuvHalf = 1 << (YUV_FIX - 1);
+
+// Max bit depth so that intermediate calculations fit in 16 bits.
+static const int kMaxBitDepth = 14;
+
+// Returns the precision shift to use based on the input rgb_bit_depth.
+static int GetPrecisionShift(int rgb_bit_depth) {
+  // Try to add 2 bits of precision if it fits in kMaxBitDepth. Otherwise remove
+  // bits if needed.
+  return ((rgb_bit_depth + 2) <= kMaxBitDepth) ? 2
+                                               : (kMaxBitDepth - rgb_bit_depth);
+}
+
+typedef int16_t fixed_t;      // signed type with extra precision for UV
+typedef uint16_t fixed_y_t;   // unsigned type with extra precision for W
+
+//------------------------------------------------------------------------------
+
+static uint8_t clip_8b(fixed_t v) {
+  return (!(v & ~0xff)) ? (uint8_t)v : (v < 0) ? 0u : 255u;
+}
+
+static uint16_t clip(fixed_t v, int max) {
+  return (v < 0) ? 0 : (v > max) ? max : (uint16_t)v;
+}
+
+static fixed_y_t clip_bit_depth(int y, int bit_depth) {
+  const int max = (1 << bit_depth) - 1;
+  return (!(y & ~max)) ? (fixed_y_t)y : (y < 0) ? 0 : max;
+}
+
+//------------------------------------------------------------------------------
+
+static int RGBToGray(int64_t r, int64_t g, int64_t b) {
+  const int64_t luma = 13933 * r + 46871 * g + 4732 * b + kYuvHalf;
+  return (int)(luma >> YUV_FIX);
+}
+
+static uint32_t ScaleDown(uint16_t a, uint16_t b, uint16_t c, uint16_t d,
+                          int rgb_bit_depth,
+                          SharpYuvTransferFunctionType transfer_type) {
+  const int bit_depth = rgb_bit_depth + GetPrecisionShift(rgb_bit_depth);
+  const uint32_t A = SharpYuvGammaToLinear(a, bit_depth, transfer_type);
+  const uint32_t B = SharpYuvGammaToLinear(b, bit_depth, transfer_type);
+  const uint32_t C = SharpYuvGammaToLinear(c, bit_depth, transfer_type);
+  const uint32_t D = SharpYuvGammaToLinear(d, bit_depth, transfer_type);
+  return SharpYuvLinearToGamma((A + B + C + D + 2) >> 2, bit_depth,
+                               transfer_type);
+}
+
+static WEBP_INLINE void UpdateW(const fixed_y_t* src, fixed_y_t* dst, int w,
+                                int rgb_bit_depth,
+                                SharpYuvTransferFunctionType transfer_type) {
+  const int bit_depth = rgb_bit_depth + GetPrecisionShift(rgb_bit_depth);
+  int i;
+  for (i = 0; i < w; ++i) {
+    const uint32_t R =
+        SharpYuvGammaToLinear(src[0 * w + i], bit_depth, transfer_type);
+    const uint32_t G =
+        SharpYuvGammaToLinear(src[1 * w + i], bit_depth, transfer_type);
+    const uint32_t B =
+        SharpYuvGammaToLinear(src[2 * w + i], bit_depth, transfer_type);
+    const uint32_t Y = RGBToGray(R, G, B);
+    dst[i] = (fixed_y_t)SharpYuvLinearToGamma(Y, bit_depth, transfer_type);
+  }
+}
+
+static void UpdateChroma(const fixed_y_t* src1, const fixed_y_t* src2,
+                         fixed_t* dst, int uv_w, int rgb_bit_depth,
+                         SharpYuvTransferFunctionType transfer_type) {
+  int i;
+  for (i = 0; i < uv_w; ++i) {
+    const int r =
+        ScaleDown(src1[0 * uv_w + 0], src1[0 * uv_w + 1], src2[0 * uv_w + 0],
+                  src2[0 * uv_w + 1], rgb_bit_depth, transfer_type);
+    const int g =
+        ScaleDown(src1[2 * uv_w + 0], src1[2 * uv_w + 1], src2[2 * uv_w + 0],
+                  src2[2 * uv_w + 1], rgb_bit_depth, transfer_type);
+    const int b =
+        ScaleDown(src1[4 * uv_w + 0], src1[4 * uv_w + 1], src2[4 * uv_w + 0],
+                  src2[4 * uv_w + 1], rgb_bit_depth, transfer_type);
+    const int W = RGBToGray(r, g, b);
+    dst[0 * uv_w] = (fixed_t)(r - W);
+    dst[1 * uv_w] = (fixed_t)(g - W);
+    dst[2 * uv_w] = (fixed_t)(b - W);
+    dst  += 1;
+    src1 += 2;
+    src2 += 2;
+  }
+}
+
+static void StoreGray(const fixed_y_t* rgb, fixed_y_t* y, int w) {
+  int i;
+  assert(w > 0);
+  for (i = 0; i < w; ++i) {
+    y[i] = RGBToGray(rgb[0 * w + i], rgb[1 * w + i], rgb[2 * w + i]);
+  }
+}
+
+//------------------------------------------------------------------------------
+
+static WEBP_INLINE fixed_y_t Filter2(int A, int B, int W0, int bit_depth) {
+  const int v0 = (A * 3 + B + 2) >> 2;
+  return clip_bit_depth(v0 + W0, bit_depth);
+}
+
+//------------------------------------------------------------------------------
+
+static WEBP_INLINE int Shift(int v, int shift) {
+  return (shift >= 0) ? (v << shift) : (v >> -shift);
+}
+
+static void ImportOneRow(const uint8_t* const r_ptr,
+                         const uint8_t* const g_ptr,
+                         const uint8_t* const b_ptr,
+                         int rgb_step,
+                         int rgb_bit_depth,
+                         int pic_width,
+                         fixed_y_t* const dst) {
+  // Convert the rgb_step from a number of bytes to a number of uint8_t or
+  // uint16_t values depending the bit depth.
+  const int step = (rgb_bit_depth > 8) ? rgb_step / 2 : rgb_step;
+  int i;
+  const int w = (pic_width + 1) & ~1;
+  for (i = 0; i < pic_width; ++i) {
+    const int off = i * step;
+    const int shift = GetPrecisionShift(rgb_bit_depth);
+    if (rgb_bit_depth == 8) {
+      dst[i + 0 * w] = Shift(r_ptr[off], shift);
+      dst[i + 1 * w] = Shift(g_ptr[off], shift);
+      dst[i + 2 * w] = Shift(b_ptr[off], shift);
+    } else {
+      dst[i + 0 * w] = Shift(((uint16_t*)r_ptr)[off], shift);
+      dst[i + 1 * w] = Shift(((uint16_t*)g_ptr)[off], shift);
+      dst[i + 2 * w] = Shift(((uint16_t*)b_ptr)[off], shift);
+    }
+  }
+  if (pic_width & 1) {  // replicate rightmost pixel
+    dst[pic_width + 0 * w] = dst[pic_width + 0 * w - 1];
+    dst[pic_width + 1 * w] = dst[pic_width + 1 * w - 1];
+    dst[pic_width + 2 * w] = dst[pic_width + 2 * w - 1];
+  }
+}
+
+static void InterpolateTwoRows(const fixed_y_t* const best_y,
+                               const fixed_t* prev_uv,
+                               const fixed_t* cur_uv,
+                               const fixed_t* next_uv,
+                               int w,
+                               fixed_y_t* out1,
+                               fixed_y_t* out2,
+                               int rgb_bit_depth) {
+  const int uv_w = w >> 1;
+  const int len = (w - 1) >> 1;   // length to filter
+  int k = 3;
+  const int bit_depth = rgb_bit_depth + GetPrecisionShift(rgb_bit_depth);
+  while (k-- > 0) {   // process each R/G/B segments in turn
+    // special boundary case for i==0
+    out1[0] = Filter2(cur_uv[0], prev_uv[0], best_y[0], bit_depth);
+    out2[0] = Filter2(cur_uv[0], next_uv[0], best_y[w], bit_depth);
+
+    SharpYuvFilterRow(cur_uv, prev_uv, len, best_y + 0 + 1, out1 + 1,
+                      bit_depth);
+    SharpYuvFilterRow(cur_uv, next_uv, len, best_y + w + 1, out2 + 1,
+                      bit_depth);
+
+    // special boundary case for i == w - 1 when w is even
+    if (!(w & 1)) {
+      out1[w - 1] = Filter2(cur_uv[uv_w - 1], prev_uv[uv_w - 1],
+                            best_y[w - 1 + 0], bit_depth);
+      out2[w - 1] = Filter2(cur_uv[uv_w - 1], next_uv[uv_w - 1],
+                            best_y[w - 1 + w], bit_depth);
+    }
+    out1 += w;
+    out2 += w;
+    prev_uv += uv_w;
+    cur_uv  += uv_w;
+    next_uv += uv_w;
+  }
+}
+
+static WEBP_INLINE int RGBToYUVComponent(int r, int g, int b,
+                                         const int coeffs[4], int sfix) {
+  const int srounder = 1 << (YUV_FIX + sfix - 1);
+  const int luma = coeffs[0] * r + coeffs[1] * g + coeffs[2] * b +
+                   coeffs[3] + srounder;
+  return (luma >> (YUV_FIX + sfix));
+}
+
+static int ConvertWRGBToYUV(const fixed_y_t* best_y, const fixed_t* best_uv,
+                            uint8_t* y_ptr, int y_stride, uint8_t* u_ptr,
+                            int u_stride, uint8_t* v_ptr, int v_stride,
+                            int rgb_bit_depth,
+                            int yuv_bit_depth, int width, int height,
+                            const SharpYuvConversionMatrix* yuv_matrix) {
+  int i, j;
+  const fixed_t* const best_uv_base = best_uv;
+  const int w = (width + 1) & ~1;
+  const int h = (height + 1) & ~1;
+  const int uv_w = w >> 1;
+  const int uv_h = h >> 1;
+  const int sfix = GetPrecisionShift(rgb_bit_depth);
+  const int yuv_max = (1 << yuv_bit_depth) - 1;
+
+  for (best_uv = best_uv_base, j = 0; j < height; ++j) {
+    for (i = 0; i < width; ++i) {
+      const int off = (i >> 1);
+      const int W = best_y[i];
+      const int r = best_uv[off + 0 * uv_w] + W;
+      const int g = best_uv[off + 1 * uv_w] + W;
+      const int b = best_uv[off + 2 * uv_w] + W;
+      const int y = RGBToYUVComponent(r, g, b, yuv_matrix->rgb_to_y, sfix);
+      if (yuv_bit_depth <= 8) {
+        y_ptr[i] = clip_8b(y);
+      } else {
+        ((uint16_t*)y_ptr)[i] = clip(y, yuv_max);
+      }
+    }
+    best_y += w;
+    best_uv += (j & 1) * 3 * uv_w;
+    y_ptr += y_stride;
+  }
+  for (best_uv = best_uv_base, j = 0; j < uv_h; ++j) {
+    for (i = 0; i < uv_w; ++i) {
+      const int off = i;
+      // Note r, g and b values here are off by W, but a constant offset on all
+      // 3 components doesn't change the value of u and v with a YCbCr matrix.
+      const int r = best_uv[off + 0 * uv_w];
+      const int g = best_uv[off + 1 * uv_w];
+      const int b = best_uv[off + 2 * uv_w];
+      const int u = RGBToYUVComponent(r, g, b, yuv_matrix->rgb_to_u, sfix);
+      const int v = RGBToYUVComponent(r, g, b, yuv_matrix->rgb_to_v, sfix);
+      if (yuv_bit_depth <= 8) {
+        u_ptr[i] = clip_8b(u);
+        v_ptr[i] = clip_8b(v);
+      } else {
+        ((uint16_t*)u_ptr)[i] = clip(u, yuv_max);
+        ((uint16_t*)v_ptr)[i] = clip(v, yuv_max);
+      }
+    }
+    best_uv += 3 * uv_w;
+    u_ptr += u_stride;
+    v_ptr += v_stride;
+  }
+  return 1;
+}
+
+//------------------------------------------------------------------------------
+// Main function
+
+static void* SafeMalloc(uint64_t nmemb, size_t size) {
+  const uint64_t total_size = nmemb * (uint64_t)size;
+  if (total_size != (size_t)total_size) return NULL;
+  return malloc((size_t)total_size);
+}
+
+#define SAFE_ALLOC(W, H, T) ((T*)SafeMalloc((W) * (H), sizeof(T)))
+
+static int DoSharpArgbToYuv(const uint8_t* r_ptr, const uint8_t* g_ptr,
+                            const uint8_t* b_ptr, int rgb_step, int rgb_stride,
+                            int rgb_bit_depth, uint8_t* y_ptr, int y_stride,
+                            uint8_t* u_ptr, int u_stride, uint8_t* v_ptr,
+                            int v_stride, int yuv_bit_depth, int width,
+                            int height,
+                            const SharpYuvConversionMatrix* yuv_matrix,
+                            SharpYuvTransferFunctionType transfer_type) {
+  // we expand the right/bottom border if needed
+  const int w = (width + 1) & ~1;
+  const int h = (height + 1) & ~1;
+  const int uv_w = w >> 1;
+  const int uv_h = h >> 1;
+  uint64_t prev_diff_y_sum = ~0;
+  int j, iter;
+
+  // TODO(skal): allocate one big memory chunk. But for now, it's easier
+  // for valgrind debugging to have several chunks.
+  fixed_y_t* const tmp_buffer = SAFE_ALLOC(w * 3, 2, fixed_y_t);   // scratch
+  fixed_y_t* const best_y_base = SAFE_ALLOC(w, h, fixed_y_t);
+  fixed_y_t* const target_y_base = SAFE_ALLOC(w, h, fixed_y_t);
+  fixed_y_t* const best_rgb_y = SAFE_ALLOC(w, 2, fixed_y_t);
+  fixed_t* const best_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t);
+  fixed_t* const target_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t);
+  fixed_t* const best_rgb_uv = SAFE_ALLOC(uv_w * 3, 1, fixed_t);
+  fixed_y_t* best_y = best_y_base;
+  fixed_y_t* target_y = target_y_base;
+  fixed_t* best_uv = best_uv_base;
+  fixed_t* target_uv = target_uv_base;
+  const uint64_t diff_y_threshold = (uint64_t)(3.0 * w * h);
+  int ok;
+  assert(w > 0);
+  assert(h > 0);
+
+  if (best_y_base == NULL || best_uv_base == NULL ||
+      target_y_base == NULL || target_uv_base == NULL ||
+      best_rgb_y == NULL || best_rgb_uv == NULL ||
+      tmp_buffer == NULL) {
+    ok = 0;
+    goto End;
+  }
+
+  // Import RGB samples to W/RGB representation.
+  for (j = 0; j < height; j += 2) {
+    const int is_last_row = (j == height - 1);
+    fixed_y_t* const src1 = tmp_buffer + 0 * w;
+    fixed_y_t* const src2 = tmp_buffer + 3 * w;
+
+    // prepare two rows of input
+    ImportOneRow(r_ptr, g_ptr, b_ptr, rgb_step, rgb_bit_depth, width,
+                 src1);
+    if (!is_last_row) {
+      ImportOneRow(r_ptr + rgb_stride, g_ptr + rgb_stride, b_ptr + rgb_stride,
+                   rgb_step, rgb_bit_depth, width, src2);
+    } else {
+      memcpy(src2, src1, 3 * w * sizeof(*src2));
+    }
+    StoreGray(src1, best_y + 0, w);
+    StoreGray(src2, best_y + w, w);
+
+    UpdateW(src1, target_y, w, rgb_bit_depth, transfer_type);
+    UpdateW(src2, target_y + w, w, rgb_bit_depth, transfer_type);
+    UpdateChroma(src1, src2, target_uv, uv_w, rgb_bit_depth, transfer_type);
+    memcpy(best_uv, target_uv, 3 * uv_w * sizeof(*best_uv));
+    best_y += 2 * w;
+    best_uv += 3 * uv_w;
+    target_y += 2 * w;
+    target_uv += 3 * uv_w;
+    r_ptr += 2 * rgb_stride;
+    g_ptr += 2 * rgb_stride;
+    b_ptr += 2 * rgb_stride;
+  }
+
+  // Iterate and resolve clipping conflicts.
+  for (iter = 0; iter < kNumIterations; ++iter) {
+    const fixed_t* cur_uv = best_uv_base;
+    const fixed_t* prev_uv = best_uv_base;
+    uint64_t diff_y_sum = 0;
+
+    best_y = best_y_base;
+    best_uv = best_uv_base;
+    target_y = target_y_base;
+    target_uv = target_uv_base;
+    for (j = 0; j < h; j += 2) {
+      fixed_y_t* const src1 = tmp_buffer + 0 * w;
+      fixed_y_t* const src2 = tmp_buffer + 3 * w;
+      {
+        const fixed_t* const next_uv = cur_uv + ((j < h - 2) ? 3 * uv_w : 0);
+        InterpolateTwoRows(best_y, prev_uv, cur_uv, next_uv, w,
+                           src1, src2, rgb_bit_depth);
+        prev_uv = cur_uv;
+        cur_uv = next_uv;
+      }
+
+      UpdateW(src1, best_rgb_y + 0 * w, w, rgb_bit_depth, transfer_type);
+      UpdateW(src2, best_rgb_y + 1 * w, w, rgb_bit_depth, transfer_type);
+      UpdateChroma(src1, src2, best_rgb_uv, uv_w, rgb_bit_depth, transfer_type);
+
+      // update two rows of Y and one row of RGB
+      diff_y_sum +=
+          SharpYuvUpdateY(target_y, best_rgb_y, best_y, 2 * w,
+                          rgb_bit_depth + GetPrecisionShift(rgb_bit_depth));
+      SharpYuvUpdateRGB(target_uv, best_rgb_uv, best_uv, 3 * uv_w);
+
+      best_y += 2 * w;
+      best_uv += 3 * uv_w;
+      target_y += 2 * w;
+      target_uv += 3 * uv_w;
+    }
+    // test exit condition
+    if (iter > 0) {
+      if (diff_y_sum < diff_y_threshold) break;
+      if (diff_y_sum > prev_diff_y_sum) break;
+    }
+    prev_diff_y_sum = diff_y_sum;
+  }
+
+  // final reconstruction
+  ok = ConvertWRGBToYUV(best_y_base, best_uv_base, y_ptr, y_stride, u_ptr,
+                        u_stride, v_ptr, v_stride, rgb_bit_depth, yuv_bit_depth,
+                        width, height, yuv_matrix);
+
+ End:
+  free(best_y_base);
+  free(best_uv_base);
+  free(target_y_base);
+  free(target_uv_base);
+  free(best_rgb_y);
+  free(best_rgb_uv);
+  free(tmp_buffer);
+  return ok;
+}
+#undef SAFE_ALLOC
+
+#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
+#include <pthread.h>  // NOLINT
+
+#define LOCK_ACCESS \
+    static pthread_mutex_t sharpyuv_lock = PTHREAD_MUTEX_INITIALIZER; \
+    if (pthread_mutex_lock(&sharpyuv_lock)) return
+#define UNLOCK_ACCESS_AND_RETURN                  \
+    do {                                          \
+      (void)pthread_mutex_unlock(&sharpyuv_lock); \
+      return;                                     \
+    } while (0)
+#else  // !(defined(WEBP_USE_THREAD) && !defined(_WIN32))
+#define LOCK_ACCESS do {} while (0)
+#define UNLOCK_ACCESS_AND_RETURN return
+#endif  // defined(WEBP_USE_THREAD) && !defined(_WIN32)
+
+// Hidden exported init function.
+// By default SharpYuvConvert calls it with SharpYuvGetCPUInfo. If needed,
+// users can declare it as extern and call it with an alternate VP8CPUInfo
+// function.
+extern VP8CPUInfo SharpYuvGetCPUInfo;
+SHARPYUV_EXTERN void SharpYuvInit(VP8CPUInfo cpu_info_func);
+void SharpYuvInit(VP8CPUInfo cpu_info_func) {
+  static volatile VP8CPUInfo sharpyuv_last_cpuinfo_used =
+      (VP8CPUInfo)&sharpyuv_last_cpuinfo_used;
+  LOCK_ACCESS;
+  // Only update SharpYuvGetCPUInfo when called from external code to avoid a
+  // race on reading the value in SharpYuvConvert().
+  if (cpu_info_func != (VP8CPUInfo)&SharpYuvGetCPUInfo) {
+    SharpYuvGetCPUInfo = cpu_info_func;
+  }
+  if (sharpyuv_last_cpuinfo_used == SharpYuvGetCPUInfo) {
+    UNLOCK_ACCESS_AND_RETURN;
+  }
+
+  SharpYuvInitDsp();
+  SharpYuvInitGammaTables();
+
+  sharpyuv_last_cpuinfo_used = SharpYuvGetCPUInfo;
+  UNLOCK_ACCESS_AND_RETURN;
+}
+
+int SharpYuvConvert(const void* r_ptr, const void* g_ptr, const void* b_ptr,
+                    int rgb_step, int rgb_stride, int rgb_bit_depth,
+                    void* y_ptr, int y_stride, void* u_ptr, int u_stride,
+                    void* v_ptr, int v_stride, int yuv_bit_depth, int width,
+                    int height, const SharpYuvConversionMatrix* yuv_matrix) {
+  SharpYuvOptions options;
+  options.yuv_matrix = yuv_matrix;
+  options.transfer_type = kSharpYuvTransferFunctionSrgb;
+  return SharpYuvConvertWithOptions(
+      r_ptr, g_ptr, b_ptr, rgb_step, rgb_stride, rgb_bit_depth, y_ptr, y_stride,
+      u_ptr, u_stride, v_ptr, v_stride, yuv_bit_depth, width, height, &options);
+}
+
+int SharpYuvOptionsInitInternal(const SharpYuvConversionMatrix* yuv_matrix,
+                                SharpYuvOptions* options, int version) {
+  const int major = (version >> 24);
+  const int minor = (version >> 16) & 0xff;
+  if (options == NULL || yuv_matrix == NULL ||
+      (major == SHARPYUV_VERSION_MAJOR && major == 0 &&
+       minor != SHARPYUV_VERSION_MINOR) ||
+      (major != SHARPYUV_VERSION_MAJOR)) {
+    return 0;
+  }
+  options->yuv_matrix = yuv_matrix;
+  options->transfer_type = kSharpYuvTransferFunctionSrgb;
+  return 1;
+}
+
+int SharpYuvConvertWithOptions(const void* r_ptr, const void* g_ptr,
+                               const void* b_ptr, int rgb_step, int rgb_stride,
+                               int rgb_bit_depth, void* y_ptr, int y_stride,
+                               void* u_ptr, int u_stride, void* v_ptr,
+                               int v_stride, int yuv_bit_depth, int width,
+                               int height, const SharpYuvOptions* options) {
+  const SharpYuvConversionMatrix* yuv_matrix = options->yuv_matrix;
+  SharpYuvTransferFunctionType transfer_type = options->transfer_type;
+  SharpYuvConversionMatrix scaled_matrix;
+  const int rgb_max = (1 << rgb_bit_depth) - 1;
+  const int rgb_round = 1 << (rgb_bit_depth - 1);
+  const int yuv_max = (1 << yuv_bit_depth) - 1;
+  const int sfix = GetPrecisionShift(rgb_bit_depth);
+
+  if (width < 1 || height < 1 || width == INT_MAX || height == INT_MAX ||
+      r_ptr == NULL || g_ptr == NULL || b_ptr == NULL || y_ptr == NULL ||
+      u_ptr == NULL || v_ptr == NULL) {
+    return 0;
+  }
+  if (rgb_bit_depth != 8 && rgb_bit_depth != 10 && rgb_bit_depth != 12 &&
+      rgb_bit_depth != 16) {
+    return 0;
+  }
+  if (yuv_bit_depth != 8 && yuv_bit_depth != 10 && yuv_bit_depth != 12) {
+    return 0;
+  }
+  if (rgb_bit_depth > 8 && (rgb_step % 2 != 0 || rgb_stride %2 != 0)) {
+    // Step/stride should be even for uint16_t buffers.
+    return 0;
+  }
+  if (yuv_bit_depth > 8 &&
+      (y_stride % 2 != 0 || u_stride % 2 != 0 || v_stride % 2 != 0)) {
+    // Stride should be even for uint16_t buffers.
+    return 0;
+  }
+  // The address of the function pointer is used to avoid a read race.
+  SharpYuvInit((VP8CPUInfo)&SharpYuvGetCPUInfo);
+
+  // Add scaling factor to go from rgb_bit_depth to yuv_bit_depth, to the
+  // rgb->yuv conversion matrix.
+  if (rgb_bit_depth == yuv_bit_depth) {
+    memcpy(&scaled_matrix, yuv_matrix, sizeof(scaled_matrix));
+  } else {
+    int i;
+    for (i = 0; i < 3; ++i) {
+      scaled_matrix.rgb_to_y[i] =
+          (yuv_matrix->rgb_to_y[i] * yuv_max + rgb_round) / rgb_max;
+      scaled_matrix.rgb_to_u[i] =
+          (yuv_matrix->rgb_to_u[i] * yuv_max + rgb_round) / rgb_max;
+      scaled_matrix.rgb_to_v[i] =
+          (yuv_matrix->rgb_to_v[i] * yuv_max + rgb_round) / rgb_max;
+    }
+  }
+  // Also incorporate precision change scaling.
+  scaled_matrix.rgb_to_y[3] = Shift(yuv_matrix->rgb_to_y[3], sfix);
+  scaled_matrix.rgb_to_u[3] = Shift(yuv_matrix->rgb_to_u[3], sfix);
+  scaled_matrix.rgb_to_v[3] = Shift(yuv_matrix->rgb_to_v[3], sfix);
+
+  return DoSharpArgbToYuv(r_ptr, g_ptr, b_ptr, rgb_step, rgb_stride,
+                          rgb_bit_depth, y_ptr, y_stride, u_ptr, u_stride,
+                          v_ptr, v_stride, yuv_bit_depth, width, height,
+                          &scaled_matrix, transfer_type);
+}
+
+//------------------------------------------------------------------------------

3rdparty/libwebp/sharpyuv/sharpyuv.h

@@ -0,0 +1,174 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Sharp RGB to YUV conversion.
+
+#ifndef WEBP_SHARPYUV_SHARPYUV_H_
+#define WEBP_SHARPYUV_SHARPYUV_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef SHARPYUV_EXTERN
+#ifdef WEBP_EXTERN
+#define SHARPYUV_EXTERN WEBP_EXTERN
+#else
+// This explicitly marks library functions and allows for changing the
+// signature for e.g., Windows DLL builds.
+#if defined(__GNUC__) && __GNUC__ >= 4
+#define SHARPYUV_EXTERN extern __attribute__((visibility("default")))
+#else
+#if defined(_MSC_VER) && defined(WEBP_DLL)
+#define SHARPYUV_EXTERN __declspec(dllexport)
+#else
+#define SHARPYUV_EXTERN extern
+#endif /* _MSC_VER && WEBP_DLL */
+#endif /* __GNUC__ >= 4 */
+#endif /* WEBP_EXTERN */
+#endif /* SHARPYUV_EXTERN */
+
+#ifndef SHARPYUV_INLINE
+#ifdef WEBP_INLINE
+#define SHARPYUV_INLINE WEBP_INLINE
+#else
+#ifndef _MSC_VER
+#if defined(__cplusplus) || !defined(__STRICT_ANSI__) || \
+    (defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L)
+#define SHARPYUV_INLINE inline
+#else
+#define SHARPYUV_INLINE
+#endif
+#else
+#define SHARPYUV_INLINE __forceinline
+#endif /* _MSC_VER */
+#endif /* WEBP_INLINE */
+#endif /* SHARPYUV_INLINE */
+
+// SharpYUV API version following the convention from semver.org
+#define SHARPYUV_VERSION_MAJOR 0
+#define SHARPYUV_VERSION_MINOR 4
+#define SHARPYUV_VERSION_PATCH 0
+// Version as a uint32_t. The major number is the high 8 bits.
+// The minor number is the middle 8 bits. The patch number is the low 16 bits.
+#define SHARPYUV_MAKE_VERSION(MAJOR, MINOR, PATCH) \
+  (((MAJOR) << 24) | ((MINOR) << 16) | (PATCH))
+#define SHARPYUV_VERSION                                                \
+  SHARPYUV_MAKE_VERSION(SHARPYUV_VERSION_MAJOR, SHARPYUV_VERSION_MINOR, \
+                        SHARPYUV_VERSION_PATCH)
+
+// Returns the library's version number, packed in hexadecimal. See
+// SHARPYUV_VERSION.
+SHARPYUV_EXTERN int SharpYuvGetVersion(void);
+
+// RGB to YUV conversion matrix, in 16 bit fixed point.
+// y = rgb_to_y[0] * r + rgb_to_y[1] * g + rgb_to_y[2] * b + rgb_to_y[3]
+// u = rgb_to_u[0] * r + rgb_to_u[1] * g + rgb_to_u[2] * b + rgb_to_u[3]
+// v = rgb_to_v[0] * r + rgb_to_v[1] * g + rgb_to_v[2] * b + rgb_to_v[3]
+// Then y, u and v values are divided by 1<<16 and rounded.
+typedef struct {
+  int rgb_to_y[4];
+  int rgb_to_u[4];
+  int rgb_to_v[4];
+} SharpYuvConversionMatrix;
+
+typedef struct SharpYuvOptions SharpYuvOptions;
+
+// Enums for transfer functions, as defined in H.273,
+// https://www.itu.int/rec/T-REC-H.273-202107-I/en
+typedef enum SharpYuvTransferFunctionType {
+  // 0 is reserved
+  kSharpYuvTransferFunctionBt709 = 1,
+  // 2 is unspecified
+  // 3 is reserved
+  kSharpYuvTransferFunctionBt470M = 4,
+  kSharpYuvTransferFunctionBt470Bg = 5,
+  kSharpYuvTransferFunctionBt601 = 6,
+  kSharpYuvTransferFunctionSmpte240 = 7,
+  kSharpYuvTransferFunctionLinear = 8,
+  kSharpYuvTransferFunctionLog100 = 9,
+  kSharpYuvTransferFunctionLog100_Sqrt10 = 10,
+  kSharpYuvTransferFunctionIec61966 = 11,
+  kSharpYuvTransferFunctionBt1361 = 12,
+  kSharpYuvTransferFunctionSrgb = 13,
+  kSharpYuvTransferFunctionBt2020_10Bit = 14,
+  kSharpYuvTransferFunctionBt2020_12Bit = 15,
+  kSharpYuvTransferFunctionSmpte2084 = 16,  // PQ
+  kSharpYuvTransferFunctionSmpte428 = 17,
+  kSharpYuvTransferFunctionHlg = 18,
+  kSharpYuvTransferFunctionNum
+} SharpYuvTransferFunctionType;
+
+// Converts RGB to YUV420 using a downsampling algorithm that minimizes
+// artefacts caused by chroma subsampling.
+// This is slower than standard downsampling (averaging of 4 UV values).
+// Assumes that the image will be upsampled using a bilinear filter. If nearest
+// neighbor is used instead, the upsampled image might look worse than with
+// standard downsampling.
+// r_ptr, g_ptr, b_ptr: pointers to the source r, g and b channels. Should point
+//     to uint8_t buffers if rgb_bit_depth is 8, or uint16_t buffers otherwise.
+// rgb_step: distance in bytes between two horizontally adjacent pixels on the
+//     r, g and b channels. If rgb_bit_depth is > 8, it should be a
+//     multiple of 2.
+// rgb_stride: distance in bytes between two vertically adjacent pixels on the
+//     r, g, and b channels. If rgb_bit_depth is > 8, it should be a
+//     multiple of 2.
+// rgb_bit_depth: number of bits for each r/g/b value. One of: 8, 10, 12, 16.
+//     Note: 16 bit input is truncated to 14 bits before conversion to yuv.
+// yuv_bit_depth: number of bits for each y/u/v value. One of: 8, 10, 12.
+// y_ptr, u_ptr, v_ptr: pointers to the destination y, u and v channels.  Should
+//     point to uint8_t buffers if yuv_bit_depth is 8, or uint16_t buffers
+//     otherwise.
+// y_stride, u_stride, v_stride: distance in bytes between two vertically
+//     adjacent pixels on the y, u and v channels. If yuv_bit_depth > 8, they
+//     should be multiples of 2.
+// width, height: width and height of the image in pixels
+// This function calls SharpYuvConvertWithOptions with a default transfer
+// function of kSharpYuvTransferFunctionSrgb.
+SHARPYUV_EXTERN int SharpYuvConvert(const void* r_ptr, const void* g_ptr,
+                                    const void* b_ptr, int rgb_step,
+                                    int rgb_stride, int rgb_bit_depth,
+                                    void* y_ptr, int y_stride, void* u_ptr,
+                                    int u_stride, void* v_ptr, int v_stride,
+                                    int yuv_bit_depth, int width, int height,
+                                    const SharpYuvConversionMatrix* yuv_matrix);
+
+struct SharpYuvOptions {
+  // This matrix cannot be NULL and can be initialized by
+  // SharpYuvComputeConversionMatrix.
+  const SharpYuvConversionMatrix* yuv_matrix;
+  SharpYuvTransferFunctionType transfer_type;
+};
+
+// Internal, version-checked, entry point
+SHARPYUV_EXTERN int SharpYuvOptionsInitInternal(const SharpYuvConversionMatrix*,
+                                                SharpYuvOptions*, int);
+
+// Should always be called, to initialize a fresh SharpYuvOptions
+// structure before modification. SharpYuvOptionsInit() must have succeeded
+// before using the 'options' object.
+static SHARPYUV_INLINE int SharpYuvOptionsInit(
+    const SharpYuvConversionMatrix* yuv_matrix, SharpYuvOptions* options) {
+  return SharpYuvOptionsInitInternal(yuv_matrix, options, SHARPYUV_VERSION);
+}
+
+SHARPYUV_EXTERN int SharpYuvConvertWithOptions(
+    const void* r_ptr, const void* g_ptr, const void* b_ptr, int rgb_step,
+    int rgb_stride, int rgb_bit_depth, void* y_ptr, int y_stride, void* u_ptr,
+    int u_stride, void* v_ptr, int v_stride, int yuv_bit_depth, int width,
+    int height, const SharpYuvOptions* options);
+
+// TODO(b/194336375): Add YUV444 to YUV420 conversion. Maybe also add 422
+// support (it's rarely used in practice, especially for images).
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // WEBP_SHARPYUV_SHARPYUV_H_

3rdparty/libwebp/sharpyuv/sharpyuv_cpu.c

@@ -0,0 +1,14 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+#include "sharpyuv/sharpyuv_cpu.h"
+
+// Include src/dsp/cpu.c to create SharpYuvGetCPUInfo from VP8GetCPUInfo. The
+// function pointer is renamed in sharpyuv_cpu.h.
+#include "src/dsp/cpu.c"

3rdparty/libwebp/sharpyuv/sharpyuv_cpu.h

@@ -0,0 +1,22 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+#ifndef WEBP_SHARPYUV_SHARPYUV_CPU_H_
+#define WEBP_SHARPYUV_SHARPYUV_CPU_H_
+
+#include "sharpyuv/sharpyuv.h"
+
+// Avoid exporting SharpYuvGetCPUInfo in shared object / DLL builds.
+// SharpYuvInit() replaces the use of the function pointer.
+#undef WEBP_EXTERN
+#define WEBP_EXTERN extern
+#define VP8GetCPUInfo SharpYuvGetCPUInfo
+#include "src/dsp/cpu.h"
+
+#endif  // WEBP_SHARPYUV_SHARPYUV_CPU_H_

3rdparty/libwebp/sharpyuv/sharpyuv_csp.c

@@ -0,0 +1,110 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Colorspace utilities.
+
+#include "sharpyuv/sharpyuv_csp.h"
+
+#include <assert.h>
+#include <math.h>
+#include <stddef.h>
+
+static int ToFixed16(float f) { return (int)floor(f * (1 << 16) + 0.5f); }
+
+void SharpYuvComputeConversionMatrix(const SharpYuvColorSpace* yuv_color_space,
+                                     SharpYuvConversionMatrix* matrix) {
+  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 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 add_uv = (float)(128 << shift);
+  assert(yuv_color_space->bit_depth >= 8);
+
+  if (yuv_color_space->range == kSharpYuvRangeLimited) {
+    scale_y *= (219 << shift) / denom;
+    scale_u *= (224 << shift) / denom;
+    scale_v *= (224 << shift) / denom;
+    add_y = (float)(16 << shift);
+  }
+
+  matrix->rgb_to_y[0] = ToFixed16(kr * scale_y);
+  matrix->rgb_to_y[1] = ToFixed16(kg * scale_y);
+  matrix->rgb_to_y[2] = ToFixed16(kb * scale_y);
+  matrix->rgb_to_y[3] = ToFixed16(add_y);
+
+  matrix->rgb_to_u[0] = ToFixed16(-kr * scale_u);
+  matrix->rgb_to_u[1] = ToFixed16(-kg * scale_u);
+  matrix->rgb_to_u[2] = ToFixed16((1 - kb) * scale_u);
+  matrix->rgb_to_u[3] = ToFixed16(add_uv);
+
+  matrix->rgb_to_v[0] = ToFixed16((1 - kr) * scale_v);
+  matrix->rgb_to_v[1] = ToFixed16(-kg * scale_v);
+  matrix->rgb_to_v[2] = ToFixed16(-kb * scale_v);
+  matrix->rgb_to_v[3] = ToFixed16(add_uv);
+}
+
+// Matrices are in YUV_FIX fixed point precision.
+// WebP's matrix, similar but not identical to kRec601LimitedMatrix.
+static const SharpYuvConversionMatrix kWebpMatrix = {
+  {16839, 33059, 6420, 16 << 16},
+  {-9719, -19081, 28800, 128 << 16},
+  {28800, -24116, -4684, 128 << 16},
+};
+// Kr=0.2990f Kb=0.1140f bits=8 range=kSharpYuvRangeLimited
+static const SharpYuvConversionMatrix kRec601LimitedMatrix = {
+  {16829, 33039, 6416, 16 << 16},
+  {-9714, -19071, 28784, 128 << 16},
+  {28784, -24103, -4681, 128 << 16},
+};
+// Kr=0.2990f Kb=0.1140f bits=8 range=kSharpYuvRangeFull
+static const SharpYuvConversionMatrix kRec601FullMatrix = {
+  {19595, 38470, 7471, 0},
+  {-11058, -21710, 32768, 128 << 16},
+  {32768, -27439, -5329, 128 << 16},
+};
+// Kr=0.2126f Kb=0.0722f bits=8 range=kSharpYuvRangeLimited
+static const SharpYuvConversionMatrix kRec709LimitedMatrix = {
+  {11966, 40254, 4064, 16 << 16},
+  {-6596, -22189, 28784, 128 << 16},
+  {28784, -26145, -2639, 128 << 16},
+};
+// Kr=0.2126f Kb=0.0722f bits=8 range=kSharpYuvRangeFull
+static const SharpYuvConversionMatrix kRec709FullMatrix = {
+  {13933, 46871, 4732, 0},
+  {-7509, -25259, 32768, 128 << 16},
+  {32768, -29763, -3005, 128 << 16},
+};
+
+const SharpYuvConversionMatrix* SharpYuvGetConversionMatrix(
+    SharpYuvMatrixType matrix_type) {
+  switch (matrix_type) {
+    case kSharpYuvMatrixWebp:
+      return &kWebpMatrix;
+    case kSharpYuvMatrixRec601Limited:
+      return &kRec601LimitedMatrix;
+    case kSharpYuvMatrixRec601Full:
+      return &kRec601FullMatrix;
+    case kSharpYuvMatrixRec709Limited:
+      return &kRec709LimitedMatrix;
+    case kSharpYuvMatrixRec709Full:
+      return &kRec709FullMatrix;
+    case kSharpYuvMatrixNum:
+      return NULL;
+  }
+  return NULL;
+}

3rdparty/libwebp/sharpyuv/sharpyuv_csp.h

@@ -0,0 +1,60 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Colorspace utilities.
+
+#ifndef WEBP_SHARPYUV_SHARPYUV_CSP_H_
+#define WEBP_SHARPYUV_SHARPYUV_CSP_H_
+
+#include "sharpyuv/sharpyuv.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Range of YUV values.
+typedef enum {
+  kSharpYuvRangeFull,     // YUV values between [0;255] (for 8 bit)
+  kSharpYuvRangeLimited   // Y in [16;235], YUV in [16;240] (for 8 bit)
+} SharpYuvRange;
+
+// Constants that define a YUV color space.
+typedef struct {
+  // Kr and Kb are defined such that:
+  // Y = Kr * r + Kg * g + Kb * b where Kg = 1 - Kr - Kb.
+  float kr;
+  float kb;
+  int bit_depth;  // 8, 10 or 12
+  SharpYuvRange range;
+} SharpYuvColorSpace;
+
+// Fills in 'matrix' for the given YUVColorSpace.
+SHARPYUV_EXTERN void SharpYuvComputeConversionMatrix(
+    const SharpYuvColorSpace* yuv_color_space,
+    SharpYuvConversionMatrix* matrix);
+
+// Enums for precomputed conversion matrices.
+typedef enum {
+  kSharpYuvMatrixWebp = 0,
+  kSharpYuvMatrixRec601Limited,
+  kSharpYuvMatrixRec601Full,
+  kSharpYuvMatrixRec709Limited,
+  kSharpYuvMatrixRec709Full,
+  kSharpYuvMatrixNum
+} SharpYuvMatrixType;
+
+// Returns a pointer to a matrix for one of the predefined colorspaces.
+SHARPYUV_EXTERN const SharpYuvConversionMatrix* SharpYuvGetConversionMatrix(
+    SharpYuvMatrixType matrix_type);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // WEBP_SHARPYUV_SHARPYUV_CSP_H_

3rdparty/libwebp/sharpyuv/sharpyuv_dsp.c

@@ -0,0 +1,104 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Speed-critical functions for Sharp YUV.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "sharpyuv/sharpyuv_dsp.h"
+
+#include <assert.h>
+#include <stdlib.h>
+
+#include "sharpyuv/sharpyuv_cpu.h"
+
+//-----------------------------------------------------------------------------
+
+#if !WEBP_NEON_OMIT_C_CODE
+static uint16_t clip(int v, int max) {
+  return (v < 0) ? 0 : (v > max) ? max : (uint16_t)v;
+}
+
+static uint64_t SharpYuvUpdateY_C(const uint16_t* ref, const uint16_t* src,
+                                  uint16_t* dst, int len, int bit_depth) {
+  uint64_t diff = 0;
+  int i;
+  const int max_y = (1 << bit_depth) - 1;
+  for (i = 0; i < len; ++i) {
+    const int diff_y = ref[i] - src[i];
+    const int new_y = (int)dst[i] + diff_y;
+    dst[i] = clip(new_y, max_y);
+    diff += (uint64_t)abs(diff_y);
+  }
+  return diff;
+}
+
+static void SharpYuvUpdateRGB_C(const int16_t* ref, const int16_t* src,
+                                int16_t* dst, int len) {
+  int i;
+  for (i = 0; i < len; ++i) {
+    const int diff_uv = ref[i] - src[i];
+    dst[i] += diff_uv;
+  }
+}
+
+static void SharpYuvFilterRow_C(const int16_t* A, const int16_t* B, int len,
+                                const uint16_t* best_y, uint16_t* out,
+                                int bit_depth) {
+  int i;
+  const int max_y = (1 << bit_depth) - 1;
+  for (i = 0; i < len; ++i, ++A, ++B) {
+    const int v0 = (A[0] * 9 + A[1] * 3 + B[0] * 3 + B[1] + 8) >> 4;
+    const int v1 = (A[1] * 9 + A[0] * 3 + B[1] * 3 + B[0] + 8) >> 4;
+    out[2 * i + 0] = clip(best_y[2 * i + 0] + v0, max_y);
+    out[2 * i + 1] = clip(best_y[2 * i + 1] + v1, max_y);
+  }
+}
+#endif  // !WEBP_NEON_OMIT_C_CODE
+
+//-----------------------------------------------------------------------------
+
+uint64_t (*SharpYuvUpdateY)(const uint16_t* src, const uint16_t* ref,
+                            uint16_t* dst, int len, int bit_depth);
+void (*SharpYuvUpdateRGB)(const int16_t* src, const int16_t* ref, int16_t* dst,
+                          int len);
+void (*SharpYuvFilterRow)(const int16_t* A, const int16_t* B, int len,
+                          const uint16_t* best_y, uint16_t* out,
+                          int bit_depth);
+
+extern VP8CPUInfo SharpYuvGetCPUInfo;
+extern void InitSharpYuvSSE2(void);
+extern void InitSharpYuvNEON(void);
+
+void SharpYuvInitDsp(void) {
+#if !WEBP_NEON_OMIT_C_CODE
+  SharpYuvUpdateY = SharpYuvUpdateY_C;
+  SharpYuvUpdateRGB = SharpYuvUpdateRGB_C;
+  SharpYuvFilterRow = SharpYuvFilterRow_C;
+#endif
+
+  if (SharpYuvGetCPUInfo != NULL) {
+#if defined(WEBP_HAVE_SSE2)
+    if (SharpYuvGetCPUInfo(kSSE2)) {
+      InitSharpYuvSSE2();
+    }
+#endif  // WEBP_HAVE_SSE2
+  }
+
+#if defined(WEBP_HAVE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE ||
+      (SharpYuvGetCPUInfo != NULL && SharpYuvGetCPUInfo(kNEON))) {
+    InitSharpYuvNEON();
+  }
+#endif  // WEBP_HAVE_NEON
+
+  assert(SharpYuvUpdateY != NULL);
+  assert(SharpYuvUpdateRGB != NULL);
+  assert(SharpYuvFilterRow != NULL);
+}

3rdparty/libwebp/sharpyuv/sharpyuv_dsp.h

@@ -0,0 +1,28 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Speed-critical functions for Sharp YUV.
+
+#ifndef WEBP_SHARPYUV_SHARPYUV_DSP_H_
+#define WEBP_SHARPYUV_SHARPYUV_DSP_H_
+
+#include "sharpyuv/sharpyuv_cpu.h"
+#include "src/webp/types.h"
+
+extern uint64_t (*SharpYuvUpdateY)(const uint16_t* src, const uint16_t* ref,
+                                   uint16_t* dst, int len, int bit_depth);
+extern void (*SharpYuvUpdateRGB)(const int16_t* src, const int16_t* ref,
+                                 int16_t* dst, int len);
+extern void (*SharpYuvFilterRow)(const int16_t* A, const int16_t* B, int len,
+                                 const uint16_t* best_y, uint16_t* out,
+                                 int bit_depth);
+
+void SharpYuvInitDsp(void);
+
+#endif  // WEBP_SHARPYUV_SHARPYUV_DSP_H_

3rdparty/libwebp/sharpyuv/sharpyuv_gamma.c

@@ -0,0 +1,419 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Gamma correction utilities.
+
+#include "sharpyuv/sharpyuv_gamma.h"
+
+#include <assert.h>
+#include <float.h>
+#include <math.h>
+
+#include "src/webp/types.h"
+
+// Gamma correction compensates loss of resolution during chroma subsampling.
+// Size of pre-computed table for converting from gamma to linear.
+#define GAMMA_TO_LINEAR_TAB_BITS 10
+#define GAMMA_TO_LINEAR_TAB_SIZE (1 << GAMMA_TO_LINEAR_TAB_BITS)
+static uint32_t kGammaToLinearTabS[GAMMA_TO_LINEAR_TAB_SIZE + 2];
+#define LINEAR_TO_GAMMA_TAB_BITS 9
+#define LINEAR_TO_GAMMA_TAB_SIZE (1 << LINEAR_TO_GAMMA_TAB_BITS)
+static uint32_t kLinearToGammaTabS[LINEAR_TO_GAMMA_TAB_SIZE + 2];
+
+static const double kGammaF = 1. / 0.45;
+#define GAMMA_TO_LINEAR_BITS 16
+
+static volatile int kGammaTablesSOk = 0;
+void SharpYuvInitGammaTables(void) {
+  assert(GAMMA_TO_LINEAR_BITS <= 16);
+  if (!kGammaTablesSOk) {
+    int v;
+    const double a = 0.09929682680944;
+    const double thresh = 0.018053968510807;
+    const double final_scale = 1 << GAMMA_TO_LINEAR_BITS;
+    // Precompute gamma to linear table.
+    {
+      const double norm = 1. / GAMMA_TO_LINEAR_TAB_SIZE;
+      const double a_rec = 1. / (1. + a);
+      for (v = 0; v <= GAMMA_TO_LINEAR_TAB_SIZE; ++v) {
+        const double g = norm * v;
+        double value;
+        if (g <= thresh * 4.5) {
+          value = g / 4.5;
+        } else {
+          value = pow(a_rec * (g + a), kGammaF);
+        }
+        kGammaToLinearTabS[v] = (uint32_t)(value * final_scale + .5);
+      }
+      // to prevent small rounding errors to cause read-overflow:
+      kGammaToLinearTabS[GAMMA_TO_LINEAR_TAB_SIZE + 1] =
+          kGammaToLinearTabS[GAMMA_TO_LINEAR_TAB_SIZE];
+    }
+    // Precompute linear to gamma table.
+    {
+      const double scale = 1. / LINEAR_TO_GAMMA_TAB_SIZE;
+      for (v = 0; v <= LINEAR_TO_GAMMA_TAB_SIZE; ++v) {
+        const double g = scale * v;
+        double value;
+        if (g <= thresh) {
+          value = 4.5 * g;
+        } else {
+          value = (1. + a) * pow(g, 1. / kGammaF) - a;
+        }
+        kLinearToGammaTabS[v] =
+            (uint32_t)(final_scale * value + 0.5);
+      }
+      // to prevent small rounding errors to cause read-overflow:
+      kLinearToGammaTabS[LINEAR_TO_GAMMA_TAB_SIZE + 1] =
+          kLinearToGammaTabS[LINEAR_TO_GAMMA_TAB_SIZE];
+    }
+    kGammaTablesSOk = 1;
+  }
+}
+
+static WEBP_INLINE int Shift(int v, int shift) {
+  return (shift >= 0) ? (v << shift) : (v >> -shift);
+}
+
+static WEBP_INLINE uint32_t FixedPointInterpolation(int v, uint32_t* tab,
+                                                    int tab_pos_shift_right,
+                                                    int tab_value_shift) {
+  const uint32_t tab_pos = Shift(v, -tab_pos_shift_right);
+  // fractional part, in 'tab_pos_shift' fixed-point precision
+  const uint32_t x = v - (tab_pos << tab_pos_shift_right);  // fractional part
+  // v0 / v1 are in kGammaToLinearBits fixed-point precision (range [0..1])
+  const uint32_t v0 = Shift(tab[tab_pos + 0], tab_value_shift);
+  const uint32_t v1 = Shift(tab[tab_pos + 1], tab_value_shift);
+  // Final interpolation.
+  const uint32_t v2 = (v1 - v0) * x;  // note: v1 >= v0.
+  const int half =
+      (tab_pos_shift_right > 0) ? 1 << (tab_pos_shift_right - 1) : 0;
+  const uint32_t result = v0 + ((v2 + half) >> tab_pos_shift_right);
+  return result;
+}
+
+static uint32_t ToLinearSrgb(uint16_t v, int bit_depth) {
+  const int shift = GAMMA_TO_LINEAR_TAB_BITS - bit_depth;
+  if (shift > 0) {
+    return kGammaToLinearTabS[v << shift];
+  }
+  return FixedPointInterpolation(v, kGammaToLinearTabS, -shift, 0);
+}
+
+static uint16_t FromLinearSrgb(uint32_t value, int bit_depth) {
+  return FixedPointInterpolation(
+      value, kLinearToGammaTabS,
+      (GAMMA_TO_LINEAR_BITS - LINEAR_TO_GAMMA_TAB_BITS),
+      bit_depth - GAMMA_TO_LINEAR_BITS);
+}
+
+////////////////////////////////////////////////////////////////////////////////
+
+#define CLAMP(x, low, high) \
+  (((x) < (low)) ? (low) : (((high) < (x)) ? (high) : (x)))
+#define MIN(a, b) (((a) < (b)) ? (a) : (b))
+#define MAX(a, b) (((a) > (b)) ? (a) : (b))
+
+static WEBP_INLINE float Roundf(float x) {
+  if (x < 0)
+    return (float)ceil((double)(x - 0.5f));
+  else
+    return (float)floor((double)(x + 0.5f));
+}
+
+static WEBP_INLINE float Powf(float base, float exp) {
+  return (float)pow((double)base, (double)exp);
+}
+
+static WEBP_INLINE float Log10f(float x) { return (float)log10((double)x); }
+
+static float ToLinear709(float gamma) {
+  if (gamma < 0.f) {
+    return 0.f;
+  } else if (gamma < 4.5f * 0.018053968510807f) {
+    return gamma / 4.5f;
+  } else if (gamma < 1.f) {
+    return Powf((gamma + 0.09929682680944f) / 1.09929682680944f, 1.f / 0.45f);
+  }
+  return 1.f;
+}
+
+static float FromLinear709(float linear) {
+  if (linear < 0.f) {
+    return 0.f;
+  } else if (linear < 0.018053968510807f) {
+    return linear * 4.5f;
+  } else if (linear < 1.f) {
+    return 1.09929682680944f * Powf(linear, 0.45f) - 0.09929682680944f;
+  }
+  return 1.f;
+}
+
+static float ToLinear470M(float gamma) {
+  return Powf(CLAMP(gamma, 0.f, 1.f), 1.f / 2.2f);
+}
+
+static float FromLinear470M(float linear) {
+  return Powf(CLAMP(linear, 0.f, 1.f), 2.2f);
+}
+
+static float ToLinear470Bg(float gamma) {
+  return Powf(CLAMP(gamma, 0.f, 1.f), 1.f / 2.8f);
+}
+
+static float FromLinear470Bg(float linear) {
+  return Powf(CLAMP(linear, 0.f, 1.f), 2.8f);
+}
+
+static float ToLinearSmpte240(float gamma) {
+  if (gamma < 0.f) {
+    return 0.f;
+  } else if (gamma < 4.f * 0.022821585529445f) {
+    return gamma / 4.f;
+  } else if (gamma < 1.f) {
+    return Powf((gamma + 0.111572195921731f) / 1.111572195921731f, 1.f / 0.45f);
+  }
+  return 1.f;
+}
+
+static float FromLinearSmpte240(float linear) {
+  if (linear < 0.f) {
+    return 0.f;
+  } else if (linear < 0.022821585529445f) {
+    return linear * 4.f;
+  } else if (linear < 1.f) {
+    return 1.111572195921731f * Powf(linear, 0.45f) - 0.111572195921731f;
+  }
+  return 1.f;
+}
+
+static float ToLinearLog100(float gamma) {
+  return (gamma < 0.01f) ? 0.0f : 1.0f + Log10f(MIN(gamma, 1.f)) / 2.0f;
+}
+
+static float FromLinearLog100(float linear) {
+  // The function is non-bijective so choose the middle of [0, 0.01].
+  const float mid_interval = 0.01f / 2.f;
+  return (linear <= 0.0f) ? mid_interval
+                          : Powf(10.0f, 2.f * (MIN(linear, 1.f) - 1.0f));
+}
+
+static float ToLinearLog100Sqrt10(float gamma) {
+  return (gamma < 0.00316227766f) ? 0.0f
+                                  : 1.0f + Log10f(MIN(gamma, 1.f)) / 2.5f;
+}
+
+static float FromLinearLog100Sqrt10(float linear) {
+  // The function is non-bijective so choose the middle of [0, 0.00316227766f[.
+  const float mid_interval = 0.00316227766f / 2.f;
+  return (linear < 0.0f) ? mid_interval
+                         : Powf(10.0f, 2.5f * (MIN(linear, 1.f) - 1.0f));
+}
+
+static float ToLinearIec61966(float gamma) {
+  if (gamma <= -4.5f * 0.018053968510807f) {
+    return Powf((-gamma + 0.09929682680944f) / -1.09929682680944f, 1.f / 0.45f);
+  } else if (gamma < 4.5f * 0.018053968510807f) {
+    return gamma / 4.5f;
+  }
+  return Powf((gamma + 0.09929682680944f) / 1.09929682680944f, 1.f / 0.45f);
+}
+
+static float FromLinearIec61966(float linear) {
+  if (linear <= -0.018053968510807f) {
+    return -1.09929682680944f * Powf(-linear, 0.45f) + 0.09929682680944f;
+  } else if (linear < 0.018053968510807f) {
+    return linear * 4.5f;
+  }
+  return 1.09929682680944f * Powf(linear, 0.45f) - 0.09929682680944f;
+}
+
+static float ToLinearBt1361(float gamma) {
+  if (gamma < -0.25f) {
+    return -0.25f;
+  } else if (gamma < 0.f) {
+    return Powf((gamma - 0.02482420670236f) / -0.27482420670236f, 1.f / 0.45f) /
+           -4.f;
+  } else if (gamma < 4.5f * 0.018053968510807f) {
+    return gamma / 4.5f;
+  } else if (gamma < 1.f) {
+    return Powf((gamma + 0.09929682680944f) / 1.09929682680944f, 1.f / 0.45f);
+  }
+  return 1.f;
+}
+
+static float FromLinearBt1361(float linear) {
+  if (linear < -0.25f) {
+    return -0.25f;
+  } else if (linear < 0.f) {
+    return -0.27482420670236f * Powf(-4.f * linear, 0.45f) + 0.02482420670236f;
+  } else if (linear < 0.018053968510807f) {
+    return linear * 4.5f;
+  } else if (linear < 1.f) {
+    return 1.09929682680944f * Powf(linear, 0.45f) - 0.09929682680944f;
+  }
+  return 1.f;
+}
+
+static float ToLinearPq(float gamma) {
+  if (gamma > 0.f) {
+    const float pow_gamma = Powf(gamma, 32.f / 2523.f);
+    const float num = MAX(pow_gamma - 107.f / 128.f, 0.0f);
+    const float den = MAX(2413.f / 128.f - 2392.f / 128.f * pow_gamma, FLT_MIN);
+    return Powf(num / den, 4096.f / 653.f);
+  }
+  return 0.f;
+}
+
+static float FromLinearPq(float linear) {
+  if (linear > 0.f) {
+    const float pow_linear = Powf(linear, 653.f / 4096.f);
+    const float num = 107.f / 128.f + 2413.f / 128.f * pow_linear;
+    const float den = 1.0f + 2392.f / 128.f * pow_linear;
+    return Powf(num / den, 2523.f / 32.f);
+  }
+  return 0.f;
+}
+
+static float ToLinearSmpte428(float gamma) {
+  return Powf(0.91655527974030934f * MAX(gamma, 0.f), 1.f / 2.6f);
+}
+
+static float FromLinearSmpte428(float linear) {
+  return Powf(MAX(linear, 0.f), 2.6f) / 0.91655527974030934f;
+}
+
+// Conversion in BT.2100 requires RGB info. Simplify to gamma correction here.
+static float ToLinearHlg(float gamma) {
+  if (gamma < 0.f) {
+    return 0.f;
+  } else if (gamma <= 0.5f) {
+    return Powf((gamma * gamma) * (1.f / 3.f), 1.2f);
+  }
+  return Powf((expf((gamma - 0.55991073f) / 0.17883277f) + 0.28466892f) / 12.0f,
+              1.2f);
+}
+
+static float FromLinearHlg(float linear) {
+  linear = Powf(linear, 1.f / 1.2f);
+  if (linear < 0.f) {
+    return 0.f;
+  } else if (linear <= (1.f / 12.f)) {
+    return sqrtf(3.f * linear);
+  }
+  return 0.17883277f * logf(12.f * linear - 0.28466892f) + 0.55991073f;
+}
+
+uint32_t SharpYuvGammaToLinear(uint16_t v, int bit_depth,
+                               SharpYuvTransferFunctionType transfer_type) {
+  float v_float, linear;
+  if (transfer_type == kSharpYuvTransferFunctionSrgb) {
+    return ToLinearSrgb(v, bit_depth);
+  }
+  v_float = (float)v / ((1 << bit_depth) - 1);
+  switch (transfer_type) {
+    case kSharpYuvTransferFunctionBt709:
+    case kSharpYuvTransferFunctionBt601:
+    case kSharpYuvTransferFunctionBt2020_10Bit:
+    case kSharpYuvTransferFunctionBt2020_12Bit:
+      linear = ToLinear709(v_float);
+      break;
+    case kSharpYuvTransferFunctionBt470M:
+      linear = ToLinear470M(v_float);
+      break;
+    case kSharpYuvTransferFunctionBt470Bg:
+      linear = ToLinear470Bg(v_float);
+      break;
+    case kSharpYuvTransferFunctionSmpte240:
+      linear = ToLinearSmpte240(v_float);
+      break;
+    case kSharpYuvTransferFunctionLinear:
+      return v;
+    case kSharpYuvTransferFunctionLog100:
+      linear = ToLinearLog100(v_float);
+      break;
+    case kSharpYuvTransferFunctionLog100_Sqrt10:
+      linear = ToLinearLog100Sqrt10(v_float);
+      break;
+    case kSharpYuvTransferFunctionIec61966:
+      linear = ToLinearIec61966(v_float);
+      break;
+    case kSharpYuvTransferFunctionBt1361:
+      linear = ToLinearBt1361(v_float);
+      break;
+    case kSharpYuvTransferFunctionSmpte2084:
+      linear = ToLinearPq(v_float);
+      break;
+    case kSharpYuvTransferFunctionSmpte428:
+      linear = ToLinearSmpte428(v_float);
+      break;
+    case kSharpYuvTransferFunctionHlg:
+      linear = ToLinearHlg(v_float);
+      break;
+    default:
+      assert(0);
+      linear = 0;
+      break;
+  }
+  return (uint32_t)Roundf(linear * ((1 << 16) - 1));
+}
+
+uint16_t SharpYuvLinearToGamma(uint32_t v, int bit_depth,
+                               SharpYuvTransferFunctionType transfer_type) {
+  float v_float, linear;
+  if (transfer_type == kSharpYuvTransferFunctionSrgb) {
+    return FromLinearSrgb(v, bit_depth);
+  }
+  v_float = (float)v / ((1 << 16) - 1);
+  switch (transfer_type) {
+    case kSharpYuvTransferFunctionBt709:
+    case kSharpYuvTransferFunctionBt601:
+    case kSharpYuvTransferFunctionBt2020_10Bit:
+    case kSharpYuvTransferFunctionBt2020_12Bit:
+      linear = FromLinear709(v_float);
+      break;
+    case kSharpYuvTransferFunctionBt470M:
+      linear = FromLinear470M(v_float);
+      break;
+    case kSharpYuvTransferFunctionBt470Bg:
+      linear = FromLinear470Bg(v_float);
+      break;
+    case kSharpYuvTransferFunctionSmpte240:
+      linear = FromLinearSmpte240(v_float);
+      break;
+    case kSharpYuvTransferFunctionLinear:
+      return v;
+    case kSharpYuvTransferFunctionLog100:
+      linear = FromLinearLog100(v_float);
+      break;
+    case kSharpYuvTransferFunctionLog100_Sqrt10:
+      linear = FromLinearLog100Sqrt10(v_float);
+      break;
+    case kSharpYuvTransferFunctionIec61966:
+      linear = FromLinearIec61966(v_float);
+      break;
+    case kSharpYuvTransferFunctionBt1361:
+      linear = FromLinearBt1361(v_float);
+      break;
+    case kSharpYuvTransferFunctionSmpte2084:
+      linear = FromLinearPq(v_float);
+      break;
+    case kSharpYuvTransferFunctionSmpte428:
+      linear = FromLinearSmpte428(v_float);
+      break;
+    case kSharpYuvTransferFunctionHlg:
+      linear = FromLinearHlg(v_float);
+      break;
+    default:
+      assert(0);
+      linear = 0;
+      break;
+  }
+  return (uint16_t)Roundf(linear * ((1 << bit_depth) - 1));
+}

3rdparty/libwebp/sharpyuv/sharpyuv_gamma.h

@@ -0,0 +1,38 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Gamma correction utilities.
+
+#ifndef WEBP_SHARPYUV_SHARPYUV_GAMMA_H_
+#define WEBP_SHARPYUV_SHARPYUV_GAMMA_H_
+
+#include "sharpyuv/sharpyuv.h"
+#include "src/webp/types.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Initializes precomputed tables. Must be called once before calling
+// SharpYuvGammaToLinear or SharpYuvLinearToGamma.
+void SharpYuvInitGammaTables(void);
+
+// Converts a 'bit_depth'-bit gamma color value to a 16-bit linear value.
+uint32_t SharpYuvGammaToLinear(uint16_t v, int bit_depth,
+                               SharpYuvTransferFunctionType transfer_type);
+
+// Converts a 16-bit linear color value to a 'bit_depth'-bit gamma value.
+uint16_t SharpYuvLinearToGamma(uint32_t value, int bit_depth,
+                               SharpYuvTransferFunctionType transfer_type);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // WEBP_SHARPYUV_SHARPYUV_GAMMA_H_

3rdparty/libwebp/sharpyuv/sharpyuv_neon.c

@@ -0,0 +1,181 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Speed-critical functions for Sharp YUV.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "sharpyuv/sharpyuv_dsp.h"
+
+#if defined(WEBP_USE_NEON)
+#include <assert.h>
+#include <stdlib.h>
+#include <arm_neon.h>
+
+static uint16_t clip_NEON(int v, int max) {
+  return (v < 0) ? 0 : (v > max) ? max : (uint16_t)v;
+}
+
+static uint64_t SharpYuvUpdateY_NEON(const uint16_t* ref, const uint16_t* src,
+                                     uint16_t* dst, int len, int bit_depth) {
+  const int max_y = (1 << bit_depth) - 1;
+  int i;
+  const int16x8_t zero = vdupq_n_s16(0);
+  const int16x8_t max = vdupq_n_s16(max_y);
+  uint64x2_t sum = vdupq_n_u64(0);
+  uint64_t diff;
+
+  for (i = 0; i + 8 <= len; i += 8) {
+    const int16x8_t A = vreinterpretq_s16_u16(vld1q_u16(ref + i));
+    const int16x8_t B = vreinterpretq_s16_u16(vld1q_u16(src + i));
+    const int16x8_t C = vreinterpretq_s16_u16(vld1q_u16(dst + i));
+    const int16x8_t D = vsubq_s16(A, B);       // diff_y
+    const int16x8_t F = vaddq_s16(C, D);       // new_y
+    const uint16x8_t H =
+        vreinterpretq_u16_s16(vmaxq_s16(vminq_s16(F, max), zero));
+    const int16x8_t I = vabsq_s16(D);          // abs(diff_y)
+    vst1q_u16(dst + i, H);
+    sum = vpadalq_u32(sum, vpaddlq_u16(vreinterpretq_u16_s16(I)));
+  }
+  diff = vgetq_lane_u64(sum, 0) + vgetq_lane_u64(sum, 1);
+  for (; i < len; ++i) {
+    const int diff_y = ref[i] - src[i];
+    const int new_y = (int)(dst[i]) + diff_y;
+    dst[i] = clip_NEON(new_y, max_y);
+    diff += (uint64_t)(abs(diff_y));
+  }
+  return diff;
+}
+
+static void SharpYuvUpdateRGB_NEON(const int16_t* ref, const int16_t* src,
+                                   int16_t* dst, int len) {
+  int i;
+  for (i = 0; i + 8 <= len; i += 8) {
+    const int16x8_t A = vld1q_s16(ref + i);
+    const int16x8_t B = vld1q_s16(src + i);
+    const int16x8_t C = vld1q_s16(dst + i);
+    const int16x8_t D = vsubq_s16(A, B);   // diff_uv
+    const int16x8_t E = vaddq_s16(C, D);   // new_uv
+    vst1q_s16(dst + i, E);
+  }
+  for (; i < len; ++i) {
+    const int diff_uv = ref[i] - src[i];
+    dst[i] += diff_uv;
+  }
+}
+
+static void SharpYuvFilterRow16_NEON(const int16_t* A, const int16_t* B,
+                                     int len, const uint16_t* best_y,
+                                     uint16_t* out, int bit_depth) {
+  const int max_y = (1 << bit_depth) - 1;
+  int i;
+  const int16x8_t max = vdupq_n_s16(max_y);
+  const int16x8_t zero = vdupq_n_s16(0);
+  for (i = 0; i + 8 <= len; i += 8) {
+    const int16x8_t a0 = vld1q_s16(A + i + 0);
+    const int16x8_t a1 = vld1q_s16(A + i + 1);
+    const int16x8_t b0 = vld1q_s16(B + i + 0);
+    const int16x8_t b1 = vld1q_s16(B + i + 1);
+    const int16x8_t a0b1 = vaddq_s16(a0, b1);
+    const int16x8_t a1b0 = vaddq_s16(a1, b0);
+    const int16x8_t a0a1b0b1 = vaddq_s16(a0b1, a1b0);  // A0+A1+B0+B1
+    const int16x8_t a0b1_2 = vaddq_s16(a0b1, a0b1);    // 2*(A0+B1)
+    const int16x8_t a1b0_2 = vaddq_s16(a1b0, a1b0);    // 2*(A1+B0)
+    const int16x8_t c0 = vshrq_n_s16(vaddq_s16(a0b1_2, a0a1b0b1), 3);
+    const int16x8_t c1 = vshrq_n_s16(vaddq_s16(a1b0_2, a0a1b0b1), 3);
+    const int16x8_t e0 = vrhaddq_s16(c1, a0);
+    const int16x8_t e1 = vrhaddq_s16(c0, a1);
+    const int16x8x2_t f = vzipq_s16(e0, e1);
+    const int16x8_t g0 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 0));
+    const int16x8_t g1 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 8));
+    const int16x8_t h0 = vaddq_s16(g0, f.val[0]);
+    const int16x8_t h1 = vaddq_s16(g1, f.val[1]);
+    const int16x8_t i0 = vmaxq_s16(vminq_s16(h0, max), zero);
+    const int16x8_t i1 = vmaxq_s16(vminq_s16(h1, max), zero);
+    vst1q_u16(out + 2 * i + 0, vreinterpretq_u16_s16(i0));
+    vst1q_u16(out + 2 * i + 8, vreinterpretq_u16_s16(i1));
+  }
+  for (; i < len; ++i) {
+    const int a0b1 = A[i + 0] + B[i + 1];
+    const int a1b0 = A[i + 1] + B[i + 0];
+    const int a0a1b0b1 = a0b1 + a1b0 + 8;
+    const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
+    const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
+    out[2 * i + 0] = clip_NEON(best_y[2 * i + 0] + v0, max_y);
+    out[2 * i + 1] = clip_NEON(best_y[2 * i + 1] + v1, max_y);
+  }
+}
+
+static void SharpYuvFilterRow32_NEON(const int16_t* A, const int16_t* B,
+                                     int len, const uint16_t* best_y,
+                                     uint16_t* out, int bit_depth) {
+  const int max_y = (1 << bit_depth) - 1;
+  int i;
+  const uint16x8_t max = vdupq_n_u16(max_y);
+  for (i = 0; i + 4 <= len; i += 4) {
+    const int16x4_t a0 = vld1_s16(A + i + 0);
+    const int16x4_t a1 = vld1_s16(A + i + 1);
+    const int16x4_t b0 = vld1_s16(B + i + 0);
+    const int16x4_t b1 = vld1_s16(B + i + 1);
+    const int32x4_t a0b1 = vaddl_s16(a0, b1);
+    const int32x4_t a1b0 = vaddl_s16(a1, b0);
+    const int32x4_t a0a1b0b1 = vaddq_s32(a0b1, a1b0);  // A0+A1+B0+B1
+    const int32x4_t a0b1_2 = vaddq_s32(a0b1, a0b1);    // 2*(A0+B1)
+    const int32x4_t a1b0_2 = vaddq_s32(a1b0, a1b0);    // 2*(A1+B0)
+    const int32x4_t c0 = vshrq_n_s32(vaddq_s32(a0b1_2, a0a1b0b1), 3);
+    const int32x4_t c1 = vshrq_n_s32(vaddq_s32(a1b0_2, a0a1b0b1), 3);
+    const int32x4_t e0 = vrhaddq_s32(c1, vmovl_s16(a0));
+    const int32x4_t e1 = vrhaddq_s32(c0, vmovl_s16(a1));
+    const int32x4x2_t f = vzipq_s32(e0, e1);
+
+    const int16x8_t g = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i));
+    const int32x4_t h0 = vaddw_s16(f.val[0], vget_low_s16(g));
+    const int32x4_t h1 = vaddw_s16(f.val[1], vget_high_s16(g));
+    const uint16x8_t i_16 = vcombine_u16(vqmovun_s32(h0), vqmovun_s32(h1));
+    const uint16x8_t i_clamped = vminq_u16(i_16, max);
+    vst1q_u16(out + 2 * i + 0, i_clamped);
+  }
+  for (; i < len; ++i) {
+    const int a0b1 = A[i + 0] + B[i + 1];
+    const int a1b0 = A[i + 1] + B[i + 0];
+    const int a0a1b0b1 = a0b1 + a1b0 + 8;
+    const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
+    const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
+    out[2 * i + 0] = clip_NEON(best_y[2 * i + 0] + v0, max_y);
+    out[2 * i + 1] = clip_NEON(best_y[2 * i + 1] + v1, max_y);
+  }
+}
+
+static void SharpYuvFilterRow_NEON(const int16_t* A, const int16_t* B, int len,
+                                   const uint16_t* best_y, uint16_t* out,
+                                   int bit_depth) {
+  if (bit_depth <= 10) {
+    SharpYuvFilterRow16_NEON(A, B, len, best_y, out, bit_depth);
+  } else {
+    SharpYuvFilterRow32_NEON(A, B, len, best_y, out, bit_depth);
+  }
+}
+
+//------------------------------------------------------------------------------
+
+extern void InitSharpYuvNEON(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void InitSharpYuvNEON(void) {
+  SharpYuvUpdateY = SharpYuvUpdateY_NEON;
+  SharpYuvUpdateRGB = SharpYuvUpdateRGB_NEON;
+  SharpYuvFilterRow = SharpYuvFilterRow_NEON;
+}
+
+#else  // !WEBP_USE_NEON
+
+extern void InitSharpYuvNEON(void);
+
+void InitSharpYuvNEON(void) {}
+
+#endif  // WEBP_USE_NEON

3rdparty/libwebp/sharpyuv/sharpyuv_sse2.c

@@ -0,0 +1,201 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Speed-critical functions for Sharp YUV.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "sharpyuv/sharpyuv_dsp.h"
+
+#if defined(WEBP_USE_SSE2)
+#include <stdlib.h>
+#include <emmintrin.h>
+
+static uint16_t clip_SSE2(int v, int max) {
+  return (v < 0) ? 0 : (v > max) ? max : (uint16_t)v;
+}
+
+static uint64_t SharpYuvUpdateY_SSE2(const uint16_t* ref, const uint16_t* src,
+                                     uint16_t* dst, int len, int bit_depth) {
+  const int max_y = (1 << bit_depth) - 1;
+  uint64_t diff = 0;
+  uint32_t tmp[4];
+  int i;
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i max = _mm_set1_epi16(max_y);
+  const __m128i one = _mm_set1_epi16(1);
+  __m128i sum = zero;
+
+  for (i = 0; i + 8 <= len; i += 8) {
+    const __m128i A = _mm_loadu_si128((const __m128i*)(ref + i));
+    const __m128i B = _mm_loadu_si128((const __m128i*)(src + i));
+    const __m128i C = _mm_loadu_si128((const __m128i*)(dst + i));
+    const __m128i D = _mm_sub_epi16(A, B);       // diff_y
+    const __m128i E = _mm_cmpgt_epi16(zero, D);  // sign (-1 or 0)
+    const __m128i F = _mm_add_epi16(C, D);       // new_y
+    const __m128i G = _mm_or_si128(E, one);      // -1 or 1
+    const __m128i H = _mm_max_epi16(_mm_min_epi16(F, max), zero);
+    const __m128i I = _mm_madd_epi16(D, G);      // sum(abs(...))
+    _mm_storeu_si128((__m128i*)(dst + i), H);
+    sum = _mm_add_epi32(sum, I);
+  }
+  _mm_storeu_si128((__m128i*)tmp, sum);
+  diff = tmp[3] + tmp[2] + tmp[1] + tmp[0];
+  for (; i < len; ++i) {
+    const int diff_y = ref[i] - src[i];
+    const int new_y = (int)dst[i] + diff_y;
+    dst[i] = clip_SSE2(new_y, max_y);
+    diff += (uint64_t)abs(diff_y);
+  }
+  return diff;
+}
+
+static void SharpYuvUpdateRGB_SSE2(const int16_t* ref, const int16_t* src,
+                                   int16_t* dst, int len) {
+  int i = 0;
+  for (i = 0; i + 8 <= len; i += 8) {
+    const __m128i A = _mm_loadu_si128((const __m128i*)(ref + i));
+    const __m128i B = _mm_loadu_si128((const __m128i*)(src + i));
+    const __m128i C = _mm_loadu_si128((const __m128i*)(dst + i));
+    const __m128i D = _mm_sub_epi16(A, B);   // diff_uv
+    const __m128i E = _mm_add_epi16(C, D);   // new_uv
+    _mm_storeu_si128((__m128i*)(dst + i), E);
+  }
+  for (; i < len; ++i) {
+    const int diff_uv = ref[i] - src[i];
+    dst[i] += diff_uv;
+  }
+}
+
+static void SharpYuvFilterRow16_SSE2(const int16_t* A, const int16_t* B,
+                                     int len, const uint16_t* best_y,
+                                     uint16_t* out, int bit_depth) {
+  const int max_y = (1 << bit_depth) - 1;
+  int i;
+  const __m128i kCst8 = _mm_set1_epi16(8);
+  const __m128i max = _mm_set1_epi16(max_y);
+  const __m128i zero = _mm_setzero_si128();
+  for (i = 0; i + 8 <= len; i += 8) {
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)(A + i + 0));
+    const __m128i a1 = _mm_loadu_si128((const __m128i*)(A + i + 1));
+    const __m128i b0 = _mm_loadu_si128((const __m128i*)(B + i + 0));
+    const __m128i b1 = _mm_loadu_si128((const __m128i*)(B + i + 1));
+    const __m128i a0b1 = _mm_add_epi16(a0, b1);
+    const __m128i a1b0 = _mm_add_epi16(a1, b0);
+    const __m128i a0a1b0b1 = _mm_add_epi16(a0b1, a1b0);  // A0+A1+B0+B1
+    const __m128i a0a1b0b1_8 = _mm_add_epi16(a0a1b0b1, kCst8);
+    const __m128i a0b1_2 = _mm_add_epi16(a0b1, a0b1);    // 2*(A0+B1)
+    const __m128i a1b0_2 = _mm_add_epi16(a1b0, a1b0);    // 2*(A1+B0)
+    const __m128i c0 = _mm_srai_epi16(_mm_add_epi16(a0b1_2, a0a1b0b1_8), 3);
+    const __m128i c1 = _mm_srai_epi16(_mm_add_epi16(a1b0_2, a0a1b0b1_8), 3);
+    const __m128i d0 = _mm_add_epi16(c1, a0);
+    const __m128i d1 = _mm_add_epi16(c0, a1);
+    const __m128i e0 = _mm_srai_epi16(d0, 1);
+    const __m128i e1 = _mm_srai_epi16(d1, 1);
+    const __m128i f0 = _mm_unpacklo_epi16(e0, e1);
+    const __m128i f1 = _mm_unpackhi_epi16(e0, e1);
+    const __m128i g0 = _mm_loadu_si128((const __m128i*)(best_y + 2 * i + 0));
+    const __m128i g1 = _mm_loadu_si128((const __m128i*)(best_y + 2 * i + 8));
+    const __m128i h0 = _mm_add_epi16(g0, f0);
+    const __m128i h1 = _mm_add_epi16(g1, f1);
+    const __m128i i0 = _mm_max_epi16(_mm_min_epi16(h0, max), zero);
+    const __m128i i1 = _mm_max_epi16(_mm_min_epi16(h1, max), zero);
+    _mm_storeu_si128((__m128i*)(out + 2 * i + 0), i0);
+    _mm_storeu_si128((__m128i*)(out + 2 * i + 8), i1);
+  }
+  for (; i < len; ++i) {
+    //   (9 * A0 + 3 * A1 + 3 * B0 + B1 + 8) >> 4 =
+    // = (8 * A0 + 2 * (A1 + B0) + (A0 + A1 + B0 + B1 + 8)) >> 4
+    // We reuse the common sub-expressions.
+    const int a0b1 = A[i + 0] + B[i + 1];
+    const int a1b0 = A[i + 1] + B[i + 0];
+    const int a0a1b0b1 = a0b1 + a1b0 + 8;
+    const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
+    const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
+    out[2 * i + 0] = clip_SSE2(best_y[2 * i + 0] + v0, max_y);
+    out[2 * i + 1] = clip_SSE2(best_y[2 * i + 1] + v1, max_y);
+  }
+}
+
+static WEBP_INLINE __m128i s16_to_s32(__m128i in) {
+  return _mm_srai_epi32(_mm_unpacklo_epi16(in, in), 16);
+}
+
+static void SharpYuvFilterRow32_SSE2(const int16_t* A, const int16_t* B,
+                                     int len, const uint16_t* best_y,
+                                     uint16_t* out, int bit_depth) {
+  const int max_y = (1 << bit_depth) - 1;
+  int i;
+  const __m128i kCst8 = _mm_set1_epi32(8);
+  const __m128i max = _mm_set1_epi16(max_y);
+  const __m128i zero = _mm_setzero_si128();
+  for (i = 0; i + 4 <= len; i += 4) {
+    const __m128i a0 = s16_to_s32(_mm_loadl_epi64((const __m128i*)(A + i + 0)));
+    const __m128i a1 = s16_to_s32(_mm_loadl_epi64((const __m128i*)(A + i + 1)));
+    const __m128i b0 = s16_to_s32(_mm_loadl_epi64((const __m128i*)(B + i + 0)));
+    const __m128i b1 = s16_to_s32(_mm_loadl_epi64((const __m128i*)(B + i + 1)));
+    const __m128i a0b1 = _mm_add_epi32(a0, b1);
+    const __m128i a1b0 = _mm_add_epi32(a1, b0);
+    const __m128i a0a1b0b1 = _mm_add_epi32(a0b1, a1b0);  // A0+A1+B0+B1
+    const __m128i a0a1b0b1_8 = _mm_add_epi32(a0a1b0b1, kCst8);
+    const __m128i a0b1_2 = _mm_add_epi32(a0b1, a0b1);  // 2*(A0+B1)
+    const __m128i a1b0_2 = _mm_add_epi32(a1b0, a1b0);  // 2*(A1+B0)
+    const __m128i c0 = _mm_srai_epi32(_mm_add_epi32(a0b1_2, a0a1b0b1_8), 3);
+    const __m128i c1 = _mm_srai_epi32(_mm_add_epi32(a1b0_2, a0a1b0b1_8), 3);
+    const __m128i d0 = _mm_add_epi32(c1, a0);
+    const __m128i d1 = _mm_add_epi32(c0, a1);
+    const __m128i e0 = _mm_srai_epi32(d0, 1);
+    const __m128i e1 = _mm_srai_epi32(d1, 1);
+    const __m128i f0 = _mm_unpacklo_epi32(e0, e1);
+    const __m128i f1 = _mm_unpackhi_epi32(e0, e1);
+    const __m128i g = _mm_loadu_si128((const __m128i*)(best_y + 2 * i + 0));
+    const __m128i h_16 = _mm_add_epi16(g, _mm_packs_epi32(f0, f1));
+    const __m128i final = _mm_max_epi16(_mm_min_epi16(h_16, max), zero);
+    _mm_storeu_si128((__m128i*)(out + 2 * i + 0), final);
+  }
+  for (; i < len; ++i) {
+    //   (9 * A0 + 3 * A1 + 3 * B0 + B1 + 8) >> 4 =
+    // = (8 * A0 + 2 * (A1 + B0) + (A0 + A1 + B0 + B1 + 8)) >> 4
+    // We reuse the common sub-expressions.
+    const int a0b1 = A[i + 0] + B[i + 1];
+    const int a1b0 = A[i + 1] + B[i + 0];
+    const int a0a1b0b1 = a0b1 + a1b0 + 8;
+    const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
+    const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
+    out[2 * i + 0] = clip_SSE2(best_y[2 * i + 0] + v0, max_y);
+    out[2 * i + 1] = clip_SSE2(best_y[2 * i + 1] + v1, max_y);
+  }
+}
+
+static void SharpYuvFilterRow_SSE2(const int16_t* A, const int16_t* B, int len,
+                                   const uint16_t* best_y, uint16_t* out,
+                                   int bit_depth) {
+  if (bit_depth <= 10) {
+    SharpYuvFilterRow16_SSE2(A, B, len, best_y, out, bit_depth);
+  } else {
+    SharpYuvFilterRow32_SSE2(A, B, len, best_y, out, bit_depth);
+  }
+}
+
+//------------------------------------------------------------------------------
+
+extern void InitSharpYuvSSE2(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void InitSharpYuvSSE2(void) {
+  SharpYuvUpdateY = SharpYuvUpdateY_SSE2;
+  SharpYuvUpdateRGB = SharpYuvUpdateRGB_SSE2;
+  SharpYuvFilterRow = SharpYuvFilterRow_SSE2;
+}
+#else  // !WEBP_USE_SSE2
+
+extern void InitSharpYuvSSE2(void);
+
+void InitSharpYuvSSE2(void) {}
+
+#endif  // WEBP_USE_SSE2

3rdparty/libwebp/src/dec/alpha_dec.c

@@ -117,21 +117,12 @@ static int ALPHDecode(VP8Decoder* const dec, int row, int num_rows) {
     const uint8_t* deltas = dec->alpha_data_ + ALPHA_HEADER_LEN + row * width;
     uint8_t* dst = dec->alpha_plane_ + row * width;
     assert(deltas <= &dec->alpha_data_[dec->alpha_data_size_]);
-    if (alph_dec->filter_ != WEBP_FILTER_NONE) {
-      assert(WebPUnfilters[alph_dec->filter_] != NULL);
-      for (y = 0; y < num_rows; ++y) {
-        WebPUnfilters[alph_dec->filter_](prev_line, deltas, dst, width);
-        prev_line = dst;
-        dst += width;
-        deltas += width;
-      }
-    } else {
-      for (y = 0; y < num_rows; ++y) {
-        memcpy(dst, deltas, width * sizeof(*dst));
-        prev_line = dst;
-        dst += width;
-        deltas += width;
-      }
+    assert(WebPUnfilters[alph_dec->filter_] != NULL);
+    for (y = 0; y < num_rows; ++y) {
+      WebPUnfilters[alph_dec->filter_](prev_line, deltas, dst, width);
+      prev_line = dst;
+      dst += width;
+      deltas += width;
     }
     dec->alpha_prev_line_ = prev_line;
   } else {  // alph_dec->method_ == ALPHA_LOSSLESS_COMPRESSION
@@ -155,7 +146,8 @@ static int AllocateAlphaPlane(VP8Decoder* const dec, const VP8Io* const io) {
   dec->alpha_plane_mem_ =
       (uint8_t*)WebPSafeMalloc(alpha_size, sizeof(*dec->alpha_plane_));
   if (dec->alpha_plane_mem_ == NULL) {
-    return 0;
+    return VP8SetError(dec, VP8_STATUS_OUT_OF_MEMORY,
+                       "Alpha decoder initialization failed.");
   }
   dec->alpha_plane_ = dec->alpha_plane_mem_;
   dec->alpha_prev_line_ = NULL;
@@ -183,16 +175,25 @@ const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec,
   assert(dec != NULL && io != NULL);
 
   if (row < 0 || num_rows <= 0 || row + num_rows > height) {
-    return NULL;    // sanity check.
+    return NULL;
   }
 
   if (!dec->is_alpha_decoded_) {
     if (dec->alph_dec_ == NULL) {    // Initialize decoder.
       dec->alph_dec_ = ALPHNew();
-      if (dec->alph_dec_ == NULL) return NULL;
+      if (dec->alph_dec_ == NULL) {
+        VP8SetError(dec, VP8_STATUS_OUT_OF_MEMORY,
+                    "Alpha decoder initialization failed.");
+        return NULL;
+      }
       if (!AllocateAlphaPlane(dec, io)) goto Error;
       if (!ALPHInit(dec->alph_dec_, dec->alpha_data_, dec->alpha_data_size_,
                     io, dec->alpha_plane_)) {
+        VP8LDecoder* const vp8l_dec = dec->alph_dec_->vp8l_dec_;
+        VP8SetError(dec,
+                    (vp8l_dec == NULL) ? VP8_STATUS_OUT_OF_MEMORY
+                                       : vp8l_dec->status_,
+                    "Alpha decoder initialization failed.");
         goto Error;
       }
       // if we allowed use of alpha dithering, check whether it's needed at all

3rdparty/libwebp/src/dec/buffer_dec.c

@@ -75,7 +75,7 @@ static VP8StatusCode CheckDecBuffer(const WebPDecBuffer* const buffer) {
     const WebPRGBABuffer* const buf = &buffer->u.RGBA;
     const int stride = abs(buf->stride);
     const uint64_t size =
-        MIN_BUFFER_SIZE(width * kModeBpp[mode], height, stride);
+        MIN_BUFFER_SIZE((uint64_t)width * kModeBpp[mode], height, stride);
     ok &= (size <= buf->size);
     ok &= (stride >= width * kModeBpp[mode]);
     ok &= (buf->rgba != NULL);
@@ -102,7 +102,7 @@ static VP8StatusCode AllocateBuffer(WebPDecBuffer* const buffer) {
     int stride;
     uint64_t size;
 
-    if ((uint64_t)w * kModeBpp[mode] >= (1ull << 32)) {
+    if ((uint64_t)w * kModeBpp[mode] >= (1ull << 31)) {
       return VP8_STATUS_INVALID_PARAM;
     }
     stride = w * kModeBpp[mode];
@@ -117,7 +117,6 @@ static VP8StatusCode AllocateBuffer(WebPDecBuffer* const buffer) {
     }
     total_size = size + 2 * uv_size + a_size;
 
-    // Security/sanity checks
     output = (uint8_t*)WebPSafeMalloc(total_size, sizeof(*output));
     if (output == NULL) {
       return VP8_STATUS_OUT_OF_MEMORY;
@@ -156,11 +155,11 @@ VP8StatusCode WebPFlipBuffer(WebPDecBuffer* const buffer) {
   }
   if (WebPIsRGBMode(buffer->colorspace)) {
     WebPRGBABuffer* const buf = &buffer->u.RGBA;
-    buf->rgba += (buffer->height - 1) * buf->stride;
+    buf->rgba += (int64_t)(buffer->height - 1) * buf->stride;
     buf->stride = -buf->stride;
   } else {
     WebPYUVABuffer* const buf = &buffer->u.YUVA;
-    const int H = buffer->height;
+    const int64_t H = buffer->height;
     buf->y += (H - 1) * buf->y_stride;
     buf->y_stride = -buf->y_stride;
     buf->u += ((H - 1) >> 1) * buf->u_stride;
@@ -188,8 +187,7 @@ VP8StatusCode WebPAllocateDecBuffer(int width, int height,
       const int ch = options->crop_height;
       const int x = options->crop_left & ~1;
       const int y = options->crop_top & ~1;
-      if (x < 0 || y < 0 || cw <= 0 || ch <= 0 ||
-          x + cw > width || y + ch > height) {
+      if (!WebPCheckCropDimensions(width, height, x, y, cw, ch)) {
         return VP8_STATUS_INVALID_PARAM;   // out of frame boundary.
       }
       width = cw;

3rdparty/libwebp/src/dec/frame_dec.c

@@ -705,7 +705,7 @@ static int AllocateMemory(VP8Decoder* const dec) {
                         + cache_size + alpha_size + WEBP_ALIGN_CST;
   uint8_t* mem;
 
-  if (needed != (size_t)needed) return 0;  // check for overflow
+  if (!CheckSizeOverflow(needed)) return 0;  // check for overflow
   if (needed > dec->mem_size_) {
     WebPSafeFree(dec->mem_);
     dec->mem_size_ = 0;

3rdparty/libwebp/src/dec/io_dec.c

@@ -298,46 +298,57 @@ static int InitYUVRescaler(const VP8Io* const io, WebPDecParams* const p) {
   const int uv_out_height = (out_height + 1) >> 1;
   const int uv_in_width  = (io->mb_w + 1) >> 1;
   const int uv_in_height = (io->mb_h + 1) >> 1;
-  const size_t work_size = 2 * out_width;   // scratch memory for luma rescaler
+  // scratch memory for luma rescaler
+  const size_t work_size = 2 * (size_t)out_width;
   const size_t uv_work_size = 2 * uv_out_width;  // and for each u/v ones
-  size_t tmp_size, rescaler_size;
+  uint64_t total_size;
+  size_t rescaler_size;
   rescaler_t* work;
   WebPRescaler* scalers;
   const int num_rescalers = has_alpha ? 4 : 3;
 
-  tmp_size = (work_size + 2 * uv_work_size) * sizeof(*work);
+  total_size = ((uint64_t)work_size + 2 * uv_work_size) * sizeof(*work);
   if (has_alpha) {
-    tmp_size += work_size * sizeof(*work);
+    total_size += (uint64_t)work_size * sizeof(*work);
   }
   rescaler_size = num_rescalers * sizeof(*p->scaler_y) + WEBP_ALIGN_CST;
+  total_size += rescaler_size;
+  if (!CheckSizeOverflow(total_size)) {
+    return 0;
+  }
 
-  p->memory = WebPSafeMalloc(1ULL, tmp_size + rescaler_size);
+  p->memory = WebPSafeMalloc(1ULL, (size_t)total_size);
   if (p->memory == NULL) {
     return 0;   // memory error
   }
   work = (rescaler_t*)p->memory;
 
-  scalers = (WebPRescaler*)WEBP_ALIGN((const uint8_t*)work + tmp_size);
+  scalers = (WebPRescaler*)WEBP_ALIGN(
+      (const uint8_t*)work + total_size - rescaler_size);
   p->scaler_y = &scalers[0];
   p->scaler_u = &scalers[1];
   p->scaler_v = &scalers[2];
   p->scaler_a = has_alpha ? &scalers[3] : NULL;
 
-  WebPRescalerInit(p->scaler_y, io->mb_w, io->mb_h,
-                   buf->y, out_width, out_height, buf->y_stride, 1,
-                   work);
-  WebPRescalerInit(p->scaler_u, uv_in_width, uv_in_height,
-                   buf->u, uv_out_width, uv_out_height, buf->u_stride, 1,
-                   work + work_size);
-  WebPRescalerInit(p->scaler_v, uv_in_width, uv_in_height,
-                   buf->v, uv_out_width, uv_out_height, buf->v_stride, 1,
-                   work + work_size + uv_work_size);
+  if (!WebPRescalerInit(p->scaler_y, io->mb_w, io->mb_h,
+                        buf->y, out_width, out_height, buf->y_stride, 1,
+                        work) ||
+      !WebPRescalerInit(p->scaler_u, uv_in_width, uv_in_height,
+                        buf->u, uv_out_width, uv_out_height, buf->u_stride, 1,
+                        work + work_size) ||
+      !WebPRescalerInit(p->scaler_v, uv_in_width, uv_in_height,
+                        buf->v, uv_out_width, uv_out_height, buf->v_stride, 1,
+                        work + work_size + uv_work_size)) {
+    return 0;
+  }
   p->emit = EmitRescaledYUV;
 
   if (has_alpha) {
-    WebPRescalerInit(p->scaler_a, io->mb_w, io->mb_h,
-                     buf->a, out_width, out_height, buf->a_stride, 1,
-                     work + work_size + 2 * uv_work_size);
+    if (!WebPRescalerInit(p->scaler_a, io->mb_w, io->mb_h,
+                          buf->a, out_width, out_height, buf->a_stride, 1,
+                          work + work_size + 2 * uv_work_size)) {
+      return 0;
+    }
     p->emit_alpha = EmitRescaledAlphaYUV;
     WebPInitAlphaProcessing();
   }
@@ -480,51 +491,58 @@ static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) {
   const int out_height = io->scaled_height;
   const int uv_in_width  = (io->mb_w + 1) >> 1;
   const int uv_in_height = (io->mb_h + 1) >> 1;
-  const size_t work_size = 2 * out_width;   // scratch memory for one rescaler
+  // scratch memory for one rescaler
+  const size_t work_size = 2 * (size_t)out_width;
   rescaler_t* work;  // rescalers work area
   uint8_t* tmp;   // tmp storage for scaled YUV444 samples before RGB conversion
-  size_t tmp_size1, tmp_size2, total_size, rescaler_size;
+  uint64_t tmp_size1, tmp_size2, total_size;
+  size_t rescaler_size;
   WebPRescaler* scalers;
   const int num_rescalers = has_alpha ? 4 : 3;
 
-  tmp_size1 = 3 * work_size;
-  tmp_size2 = 3 * out_width;
-  if (has_alpha) {
-    tmp_size1 += work_size;
-    tmp_size2 += out_width;
-  }
+  tmp_size1 = (uint64_t)num_rescalers * work_size;
+  tmp_size2 = (uint64_t)num_rescalers * out_width;
   total_size = tmp_size1 * sizeof(*work) + tmp_size2 * sizeof(*tmp);
   rescaler_size = num_rescalers * sizeof(*p->scaler_y) + WEBP_ALIGN_CST;
+  total_size += rescaler_size;
+  if (!CheckSizeOverflow(total_size)) {
+    return 0;
+  }
 
-  p->memory = WebPSafeMalloc(1ULL, total_size + rescaler_size);
+  p->memory = WebPSafeMalloc(1ULL, (size_t)total_size);
   if (p->memory == NULL) {
     return 0;   // memory error
   }
   work = (rescaler_t*)p->memory;
   tmp = (uint8_t*)(work + tmp_size1);
 
-  scalers = (WebPRescaler*)WEBP_ALIGN((const uint8_t*)work + total_size);
+  scalers = (WebPRescaler*)WEBP_ALIGN(
+      (const uint8_t*)work + total_size - rescaler_size);
   p->scaler_y = &scalers[0];
   p->scaler_u = &scalers[1];
   p->scaler_v = &scalers[2];
   p->scaler_a = has_alpha ? &scalers[3] : NULL;
 
-  WebPRescalerInit(p->scaler_y, io->mb_w, io->mb_h,
-                   tmp + 0 * out_width, out_width, out_height, 0, 1,
-                   work + 0 * work_size);
-  WebPRescalerInit(p->scaler_u, uv_in_width, uv_in_height,
-                   tmp + 1 * out_width, out_width, out_height, 0, 1,
-                   work + 1 * work_size);
-  WebPRescalerInit(p->scaler_v, uv_in_width, uv_in_height,
-                   tmp + 2 * out_width, out_width, out_height, 0, 1,
-                   work + 2 * work_size);
+  if (!WebPRescalerInit(p->scaler_y, io->mb_w, io->mb_h,
+                        tmp + 0 * out_width, out_width, out_height, 0, 1,
+                        work + 0 * work_size) ||
+      !WebPRescalerInit(p->scaler_u, uv_in_width, uv_in_height,
+                        tmp + 1 * out_width, out_width, out_height, 0, 1,
+                        work + 1 * work_size) ||
+      !WebPRescalerInit(p->scaler_v, uv_in_width, uv_in_height,
+                        tmp + 2 * out_width, out_width, out_height, 0, 1,
+                        work + 2 * work_size)) {
+    return 0;
+  }
   p->emit = EmitRescaledRGB;
   WebPInitYUV444Converters();
 
   if (has_alpha) {
-    WebPRescalerInit(p->scaler_a, io->mb_w, io->mb_h,
-                     tmp + 3 * out_width, out_width, out_height, 0, 1,
-                     work + 3 * work_size);
+    if (!WebPRescalerInit(p->scaler_a, io->mb_w, io->mb_h,
+                          tmp + 3 * out_width, out_width, out_height, 0, 1,
+                          work + 3 * work_size)) {
+      return 0;
+    }
     p->emit_alpha = EmitRescaledAlphaRGB;
     if (p->output->colorspace == MODE_RGBA_4444 ||
         p->output->colorspace == MODE_rgbA_4444) {

3rdparty/libwebp/src/dec/tree_dec.c

@@ -12,10 +12,11 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include "src/dec/vp8i_dec.h"
+#include "src/dsp/cpu.h"
 #include "src/utils/bit_reader_inl_utils.h"
 
 #if !defined(USE_GENERIC_TREE)
-#if !defined(__arm__) && !defined(_M_ARM) && !defined(__aarch64__)
+#if !defined(__arm__) && !defined(_M_ARM) && !WEBP_AARCH64
 // using a table is ~1-2% slower on ARM. Prefer the coded-tree approach then.
 #define USE_GENERIC_TREE 1   // ALTERNATE_CODE
 #else

3rdparty/libwebp/src/dec/vp8_dec.c

@@ -335,7 +335,7 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
     io->scaled_width = io->width;
     io->scaled_height = io->height;
 
-    io->mb_w = io->width;   // sanity check
+    io->mb_w = io->width;   // for soundness
     io->mb_h = io->height;  // ditto
 
     VP8ResetProba(&dec->proba_);
@@ -403,7 +403,7 @@ static const uint8_t kZigzag[16] = {
   0, 1, 4, 8,  5, 2, 3, 6,  9, 12, 13, 10,  7, 11, 14, 15
 };
 
-// See section 13-2: http://tools.ietf.org/html/rfc6386#section-13.2
+// See section 13-2: https://datatracker.ietf.org/doc/html/rfc6386#section-13.2
 static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) {
   int v;
   if (!VP8GetBit(br, p[3], "coeffs")) {
@@ -494,6 +494,8 @@ static int GetCoeffsAlt(VP8BitReader* const br,
   return 16;
 }
 
+extern VP8CPUInfo VP8GetCPUInfo;
+
 WEBP_DSP_INIT_FUNC(InitGetCoeffs) {
   if (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kSlowSSSE3)) {
     GetCoeffs = GetCoeffsAlt;

3rdparty/libwebp/src/dec/vp8i_dec.h

@@ -31,8 +31,8 @@ extern "C" {
 
 // version numbers
 #define DEC_MAJ_VERSION 1
-#define DEC_MIN_VERSION 2
-#define DEC_REV_VERSION 0
+#define DEC_MIN_VERSION 3
+#define DEC_REV_VERSION 1
 
 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
 // Constraints are: We need to store one 16x16 block of luma samples (y),

3rdparty/libwebp/src/dec/vp8l_dec.c

@@ -12,6 +12,7 @@
 // Authors: Vikas Arora (vikaas.arora@gmail.com)
 //          Jyrki Alakuijala (jyrki@google.com)
 
+#include <assert.h>
 #include <stdlib.h>
 
 #include "src/dec/alphai_dec.h"
@@ -84,7 +85,7 @@ static const uint8_t kCodeToPlane[CODE_TO_PLANE_CODES] = {
 // to 256 (green component values) + 24 (length prefix values)
 // + color_cache_size (between 0 and 2048).
 // All values computed for 8-bit first level lookup with Mark Adler's tool:
-// http://www.hdfgroup.org/ftp/lib-external/zlib/zlib-1.2.5/examples/enough.c
+// https://github.com/madler/zlib/blob/v1.2.5/examples/enough.c
 #define FIXED_TABLE_SIZE (630 * 3 + 410)
 static const uint16_t kTableSize[12] = {
   FIXED_TABLE_SIZE + 654,
@@ -101,6 +102,14 @@ static const uint16_t kTableSize[12] = {
   FIXED_TABLE_SIZE + 2704
 };
 
+static int VP8LSetError(VP8LDecoder* const dec, VP8StatusCode error) {
+  // The oldest error reported takes precedence over the new one.
+  if (dec->status_ == VP8_STATUS_OK || dec->status_ == VP8_STATUS_SUSPENDED) {
+    dec->status_ = error;
+  }
+  return 0;
+}
+
 static int DecodeImageStream(int xsize, int ysize,
                              int is_level0,
                              VP8LDecoder* const dec,
@@ -178,7 +187,7 @@ static WEBP_INLINE int PlaneCodeToDistance(int xsize, int plane_code) {
 
 //------------------------------------------------------------------------------
 // Decodes the next Huffman code from bit-stream.
-// FillBitWindow(br) needs to be called at minimum every second call
+// VP8LFillBitWindow(br) needs to be called at minimum every second call
 // to ReadSymbol, in order to pre-fetch enough bits.
 static WEBP_INLINE int ReadSymbol(const HuffmanCode* table,
                                   VP8LBitReader* const br) {
@@ -253,11 +262,11 @@ static int ReadHuffmanCodeLengths(
   int symbol;
   int max_symbol;
   int prev_code_len = DEFAULT_CODE_LENGTH;
-  HuffmanCode table[1 << LENGTHS_TABLE_BITS];
+  HuffmanTables tables;
 
-  if (!VP8LBuildHuffmanTable(table, LENGTHS_TABLE_BITS,
-                             code_length_code_lengths,
-                             NUM_CODE_LENGTH_CODES)) {
+  if (!VP8LHuffmanTablesAllocate(1 << LENGTHS_TABLE_BITS, &tables) ||
+      !VP8LBuildHuffmanTable(&tables, LENGTHS_TABLE_BITS,
+                             code_length_code_lengths, NUM_CODE_LENGTH_CODES)) {
     goto End;
   }
 
@@ -277,7 +286,7 @@ static int ReadHuffmanCodeLengths(
     int code_len;
     if (max_symbol-- == 0) break;
     VP8LFillBitWindow(br);
-    p = &table[VP8LPrefetchBits(br) & LENGTHS_TABLE_MASK];
+    p = &tables.curr_segment->start[VP8LPrefetchBits(br) & LENGTHS_TABLE_MASK];
     VP8LSetBitPos(br, br->bit_pos_ + p->bits);
     code_len = p->value;
     if (code_len < kCodeLengthLiterals) {
@@ -300,14 +309,16 @@ static int ReadHuffmanCodeLengths(
   ok = 1;
 
  End:
-  if (!ok) dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
+  VP8LHuffmanTablesDeallocate(&tables);
+  if (!ok) return VP8LSetError(dec, VP8_STATUS_BITSTREAM_ERROR);
   return ok;
 }
 
 // 'code_lengths' is pre-allocated temporary buffer, used for creating Huffman
 // tree.
 static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec,
-                           int* const code_lengths, HuffmanCode* const table) {
+                           int* const code_lengths,
+                           HuffmanTables* const table) {
   int ok = 0;
   int size = 0;
   VP8LBitReader* const br = &dec->br_;
@@ -321,7 +332,7 @@ static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec,
     // The first code is either 1 bit or 8 bit code.
     int symbol = VP8LReadBits(br, (first_symbol_len_code == 0) ? 1 : 8);
     code_lengths[symbol] = 1;
-    // The second code (if present), is always 8 bit long.
+    // The second code (if present), is always 8 bits long.
     if (num_symbols == 2) {
       symbol = VP8LReadBits(br, 8);
       code_lengths[symbol] = 1;
@@ -331,10 +342,7 @@ static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec,
     int i;
     int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 };
     const int num_codes = VP8LReadBits(br, 4) + 4;
-    if (num_codes > NUM_CODE_LENGTH_CODES) {
-      dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
-      return 0;
-    }
+    assert(num_codes <= NUM_CODE_LENGTH_CODES);
 
     for (i = 0; i < num_codes; ++i) {
       code_length_code_lengths[kCodeLengthCodeOrder[i]] = VP8LReadBits(br, 3);
@@ -349,36 +357,35 @@ static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec,
                                  code_lengths, alphabet_size);
   }
   if (!ok || size == 0) {
-    dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
-    return 0;
+    return VP8LSetError(dec, VP8_STATUS_BITSTREAM_ERROR);
   }
   return size;
 }
 
 static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
                             int color_cache_bits, int allow_recursion) {
-  int i, j;
+  int i;
   VP8LBitReader* const br = &dec->br_;
   VP8LMetadata* const hdr = &dec->hdr_;
   uint32_t* huffman_image = NULL;
   HTreeGroup* htree_groups = NULL;
-  HuffmanCode* huffman_tables = NULL;
-  HuffmanCode* huffman_table = NULL;
+  HuffmanTables* huffman_tables = &hdr->huffman_tables_;
   int num_htree_groups = 1;
   int num_htree_groups_max = 1;
-  int max_alphabet_size = 0;
-  int* code_lengths = NULL;
-  const int table_size = kTableSize[color_cache_bits];
   int* mapping = NULL;
   int ok = 0;
 
+  // Check the table has been 0 initialized (through InitMetadata).
+  assert(huffman_tables->root.start == NULL);
+  assert(huffman_tables->curr_segment == NULL);
+
   if (allow_recursion && VP8LReadBits(br, 1)) {
     // use meta Huffman codes.
     const int huffman_precision = VP8LReadBits(br, 3) + 2;
     const int huffman_xsize = VP8LSubSampleSize(xsize, huffman_precision);
     const int huffman_ysize = VP8LSubSampleSize(ysize, huffman_precision);
     const int huffman_pixs = huffman_xsize * huffman_ysize;
-    if (!DecodeImageStream(huffman_xsize, huffman_ysize, 0, dec,
+    if (!DecodeImageStream(huffman_xsize, huffman_ysize, /*is_level0=*/0, dec,
                            &huffman_image)) {
       goto Error;
     }
@@ -402,7 +409,7 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
       // values [0, num_htree_groups)
       mapping = (int*)WebPSafeMalloc(num_htree_groups_max, sizeof(*mapping));
       if (mapping == NULL) {
-        dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
+        VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
         goto Error;
       }
       // -1 means a value is unmapped, and therefore unused in the Huffman
@@ -421,29 +428,55 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
 
   if (br->eos_) goto Error;
 
-  // Find maximum alphabet size for the htree group.
-  for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
-    int alphabet_size = kAlphabetSize[j];
-    if (j == 0 && color_cache_bits > 0) {
-      alphabet_size += 1 << color_cache_bits;
-    }
-    if (max_alphabet_size < alphabet_size) {
-      max_alphabet_size = alphabet_size;
-    }
+  if (!ReadHuffmanCodesHelper(color_cache_bits, num_htree_groups,
+                              num_htree_groups_max, mapping, dec,
+                              huffman_tables, &htree_groups)) {
+    goto Error;
   }
+  ok = 1;
 
-  code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size,
-                                      sizeof(*code_lengths));
-  huffman_tables = (HuffmanCode*)WebPSafeMalloc(num_htree_groups * table_size,
-                                                sizeof(*huffman_tables));
-  htree_groups = VP8LHtreeGroupsNew(num_htree_groups);
+  // All OK. Finalize pointers.
+  hdr->huffman_image_ = huffman_image;
+  hdr->num_htree_groups_ = num_htree_groups;
+  hdr->htree_groups_ = htree_groups;
+
+ Error:
+  WebPSafeFree(mapping);
+  if (!ok) {
+    WebPSafeFree(huffman_image);
+    VP8LHuffmanTablesDeallocate(huffman_tables);
+    VP8LHtreeGroupsFree(htree_groups);
+  }
+  return ok;
+}
 
-  if (htree_groups == NULL || code_lengths == NULL || huffman_tables == NULL) {
-    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
+int ReadHuffmanCodesHelper(int color_cache_bits, int num_htree_groups,
+                           int num_htree_groups_max, const int* const mapping,
+                           VP8LDecoder* const dec,
+                           HuffmanTables* const huffman_tables,
+                           HTreeGroup** const htree_groups) {
+  int i, j, ok = 0;
+  const int max_alphabet_size =
+      kAlphabetSize[0] + ((color_cache_bits > 0) ? 1 << color_cache_bits : 0);
+  const int table_size = kTableSize[color_cache_bits];
+  int* code_lengths = NULL;
+
+  if ((mapping == NULL && num_htree_groups != num_htree_groups_max) ||
+      num_htree_groups > num_htree_groups_max) {
+    goto Error;
+  }
+
+  code_lengths =
+      (int*)WebPSafeCalloc((uint64_t)max_alphabet_size, sizeof(*code_lengths));
+  *htree_groups = VP8LHtreeGroupsNew(num_htree_groups);
+
+  if (*htree_groups == NULL || code_lengths == NULL ||
+      !VP8LHuffmanTablesAllocate(num_htree_groups * table_size,
+                                 huffman_tables)) {
+    VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
     goto Error;
   }
 
-  huffman_table = huffman_tables;
   for (i = 0; i < num_htree_groups_max; ++i) {
     // If the index "i" is unused in the Huffman image, just make sure the
     // coefficients are valid but do not store them.
@@ -460,7 +493,7 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
       }
     } else {
       HTreeGroup* const htree_group =
-          &htree_groups[(mapping == NULL) ? i : mapping[i]];
+          &(*htree_groups)[(mapping == NULL) ? i : mapping[i]];
       HuffmanCode** const htrees = htree_group->htrees;
       int size;
       int total_size = 0;
@@ -468,19 +501,20 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
       int max_bits = 0;
       for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
         int alphabet_size = kAlphabetSize[j];
-        htrees[j] = huffman_table;
         if (j == 0 && color_cache_bits > 0) {
           alphabet_size += (1 << color_cache_bits);
         }
-        size = ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_table);
+        size =
+            ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_tables);
+        htrees[j] = huffman_tables->curr_segment->curr_table;
         if (size == 0) {
           goto Error;
         }
         if (is_trivial_literal && kLiteralMap[j] == 1) {
-          is_trivial_literal = (huffman_table->bits == 0);
+          is_trivial_literal = (htrees[j]->bits == 0);
         }
-        total_size += huffman_table->bits;
-        huffman_table += size;
+        total_size += htrees[j]->bits;
+        huffman_tables->curr_segment->curr_table += size;
         if (j <= ALPHA) {
           int local_max_bits = code_lengths[0];
           int k;
@@ -511,19 +545,12 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
   }
   ok = 1;
 
-  // All OK. Finalize pointers.
-  hdr->huffman_image_ = huffman_image;
-  hdr->num_htree_groups_ = num_htree_groups;
-  hdr->htree_groups_ = htree_groups;
-  hdr->huffman_tables_ = huffman_tables;
-
  Error:
   WebPSafeFree(code_lengths);
-  WebPSafeFree(mapping);
   if (!ok) {
-    WebPSafeFree(huffman_image);
-    WebPSafeFree(huffman_tables);
-    VP8LHtreeGroupsFree(htree_groups);
+    VP8LHuffmanTablesDeallocate(huffman_tables);
+    VP8LHtreeGroupsFree(*htree_groups);
+    *htree_groups = NULL;
   }
   return ok;
 }
@@ -547,8 +574,7 @@ static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) {
                                scaled_data_size * sizeof(*scaled_data);
   uint8_t* memory = (uint8_t*)WebPSafeMalloc(memory_size, sizeof(*memory));
   if (memory == NULL) {
-    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
-    return 0;
+    return VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
   }
   assert(dec->rescaler_memory == NULL);
   dec->rescaler_memory = memory;
@@ -559,8 +585,11 @@ static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) {
   memory += work_size * sizeof(*work);
   scaled_data = (uint32_t*)memory;
 
-  WebPRescalerInit(dec->rescaler, in_width, in_height, (uint8_t*)scaled_data,
-                   out_width, out_height, 0, num_channels, work);
+  if (!WebPRescalerInit(dec->rescaler, in_width, in_height,
+                        (uint8_t*)scaled_data, out_width, out_height,
+                        0, num_channels, work)) {
+    return 0;
+  }
   return 1;
 }
 #endif   // WEBP_REDUCE_SIZE
@@ -574,13 +603,14 @@ static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) {
 static int Export(WebPRescaler* const rescaler, WEBP_CSP_MODE colorspace,
                   int rgba_stride, uint8_t* const rgba) {
   uint32_t* const src = (uint32_t*)rescaler->dst;
+  uint8_t* dst = rgba;
   const int dst_width = rescaler->dst_width;
   int num_lines_out = 0;
   while (WebPRescalerHasPendingOutput(rescaler)) {
-    uint8_t* const dst = rgba + num_lines_out * rgba_stride;
     WebPRescalerExportRow(rescaler);
     WebPMultARGBRow(src, dst_width, 1);
     VP8LConvertFromBGRA(src, dst_width, colorspace, dst);
+    dst += rgba_stride;
     ++num_lines_out;
   }
   return num_lines_out;
@@ -594,8 +624,8 @@ static int EmitRescaledRowsRGBA(const VP8LDecoder* const dec,
   int num_lines_in = 0;
   int num_lines_out = 0;
   while (num_lines_in < mb_h) {
-    uint8_t* const row_in = in + num_lines_in * in_stride;
-    uint8_t* const row_out = out + num_lines_out * out_stride;
+    uint8_t* const row_in = in + (uint64_t)num_lines_in * in_stride;
+    uint8_t* const row_out = out + (uint64_t)num_lines_out * out_stride;
     const int lines_left = mb_h - num_lines_in;
     const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left);
     int lines_imported;
@@ -796,7 +826,8 @@ static void ProcessRows(VP8LDecoder* const dec, int row) {
       const WebPDecBuffer* const output = dec->output_;
       if (WebPIsRGBMode(output->colorspace)) {  // convert to RGBA
         const WebPRGBABuffer* const buf = &output->u.RGBA;
-        uint8_t* const rgba = buf->rgba + dec->last_out_row_ * buf->stride;
+        uint8_t* const rgba =
+            buf->rgba + (int64_t)dec->last_out_row_ * buf->stride;
         const int num_rows_out =
 #if !defined(WEBP_REDUCE_SIZE)
          io->use_scaling ?
@@ -1077,12 +1108,10 @@ static int DecodeAlphaData(VP8LDecoder* const dec, uint8_t* const data,
  End:
   br->eos_ = VP8LIsEndOfStream(br);
   if (!ok || (br->eos_ && pos < end)) {
-    ok = 0;
-    dec->status_ = br->eos_ ? VP8_STATUS_SUSPENDED
-                            : VP8_STATUS_BITSTREAM_ERROR;
-  } else {
-    dec->last_pixel_ = pos;
+    return VP8LSetError(
+        dec, br->eos_ ? VP8_STATUS_SUSPENDED : VP8_STATUS_BITSTREAM_ERROR);
   }
+  dec->last_pixel_ = pos;
   return ok;
 }
 
@@ -1232,9 +1261,20 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
   }
 
   br->eos_ = VP8LIsEndOfStream(br);
-  if (dec->incremental_ && br->eos_ && src < src_end) {
+  // In incremental decoding:
+  // br->eos_ && src < src_last: if 'br' reached the end of the buffer and
+  // 'src_last' has not been reached yet, there is not enough data. 'dec' has to
+  // be reset until there is more data.
+  // !br->eos_ && src < src_last: this cannot happen as either the buffer is
+  // fully read, either enough has been read to reach 'src_last'.
+  // src >= src_last: 'src_last' is reached, all is fine. 'src' can actually go
+  // beyond 'src_last' in case the image is cropped and an LZ77 goes further.
+  // The buffer might have been enough or there is some left. 'br->eos_' does
+  // not matter.
+  assert(!dec->incremental_ || (br->eos_ && src < src_last) || src >= src_last);
+  if (dec->incremental_ && br->eos_ && src < src_last) {
     RestoreState(dec);
-  } else if (!br->eos_) {
+  } else if ((dec->incremental_ && src >= src_last) || !br->eos_) {
     // Process the remaining rows corresponding to last row-block.
     if (process_func != NULL) {
       process_func(dec, row > last_row ? last_row : row);
@@ -1249,8 +1289,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
   return 1;
 
  Error:
-  dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
-  return 0;
+  return VP8LSetError(dec, VP8_STATUS_BITSTREAM_ERROR);
 }
 
 // -----------------------------------------------------------------------------
@@ -1276,7 +1315,7 @@ static int ExpandColorMap(int num_colors, VP8LTransform* const transform) {
     uint8_t* const new_data = (uint8_t*)new_color_map;
     new_color_map[0] = transform->data_[0];
     for (i = 4; i < 4 * num_colors; ++i) {
-      // Equivalent to AddPixelEq(), on a byte-basis.
+      // Equivalent to VP8LAddPixels(), on a byte-basis.
       new_data[i] = (data[i] + new_data[i - 4]) & 0xff;
     }
     for (; i < 4 * final_num_colors; ++i) {
@@ -1317,7 +1356,7 @@ static int ReadTransform(int* const xsize, int const* ysize,
                                                transform->bits_),
                              VP8LSubSampleSize(transform->ysize_,
                                                transform->bits_),
-                             0, dec, &transform->data_);
+                             /*is_level0=*/0, dec, &transform->data_);
       break;
     case COLOR_INDEXING_TRANSFORM: {
        const int num_colors = VP8LReadBits(br, 8) + 1;
@@ -1327,11 +1366,14 @@ static int ReadTransform(int* const xsize, int const* ysize,
                       : 3;
        *xsize = VP8LSubSampleSize(transform->xsize_, bits);
        transform->bits_ = bits;
-       ok = DecodeImageStream(num_colors, 1, 0, dec, &transform->data_);
-       ok = ok && ExpandColorMap(num_colors, transform);
+       ok = DecodeImageStream(num_colors, /*ysize=*/1, /*is_level0=*/0, dec,
+                              &transform->data_);
+       if (ok && !ExpandColorMap(num_colors, transform)) {
+         return VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
+       }
       break;
     }
-    case SUBTRACT_GREEN:
+    case SUBTRACT_GREEN_TRANSFORM:
       break;
     default:
       assert(0);    // can't happen
@@ -1353,7 +1395,7 @@ static void ClearMetadata(VP8LMetadata* const hdr) {
   assert(hdr != NULL);
 
   WebPSafeFree(hdr->huffman_image_);
-  WebPSafeFree(hdr->huffman_tables_);
+  VP8LHuffmanTablesDeallocate(&hdr->huffman_tables_);
   VP8LHtreeGroupsFree(hdr->htree_groups_);
   VP8LColorCacheClear(&hdr->color_cache_);
   VP8LColorCacheClear(&hdr->saved_color_cache_);
@@ -1434,7 +1476,7 @@ static int DecodeImageStream(int xsize, int ysize,
     color_cache_bits = VP8LReadBits(br, 4);
     ok = (color_cache_bits >= 1 && color_cache_bits <= MAX_CACHE_BITS);
     if (!ok) {
-      dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
+      VP8LSetError(dec, VP8_STATUS_BITSTREAM_ERROR);
       goto End;
     }
   }
@@ -1443,7 +1485,7 @@ static int DecodeImageStream(int xsize, int ysize,
   ok = ok && ReadHuffmanCodes(dec, transform_xsize, transform_ysize,
                               color_cache_bits, is_level0);
   if (!ok) {
-    dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
+    VP8LSetError(dec, VP8_STATUS_BITSTREAM_ERROR);
     goto End;
   }
 
@@ -1451,8 +1493,7 @@ static int DecodeImageStream(int xsize, int ysize,
   if (color_cache_bits > 0) {
     hdr->color_cache_size_ = 1 << color_cache_bits;
     if (!VP8LColorCacheInit(&hdr->color_cache_, color_cache_bits)) {
-      dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
-      ok = 0;
+      ok = VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
       goto End;
     }
   } else {
@@ -1469,8 +1510,7 @@ static int DecodeImageStream(int xsize, int ysize,
     const uint64_t total_size = (uint64_t)transform_xsize * transform_ysize;
     data = (uint32_t*)WebPSafeMalloc(total_size, sizeof(*data));
     if (data == NULL) {
-      dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
-      ok = 0;
+      ok = VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
       goto End;
     }
   }
@@ -1514,9 +1554,8 @@ static int AllocateInternalBuffers32b(VP8LDecoder* const dec, int final_width) {
   assert(dec->width_ <= final_width);
   dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint32_t));
   if (dec->pixels_ == NULL) {
-    dec->argb_cache_ = NULL;    // for sanity check
-    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
-    return 0;
+    dec->argb_cache_ = NULL;    // for soundness
+    return VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
   }
   dec->argb_cache_ = dec->pixels_ + num_pixels + cache_top_pixels;
   return 1;
@@ -1524,11 +1563,10 @@ static int AllocateInternalBuffers32b(VP8LDecoder* const dec, int final_width) {
 
 static int AllocateInternalBuffers8b(VP8LDecoder* const dec) {
   const uint64_t total_num_pixels = (uint64_t)dec->width_ * dec->height_;
-  dec->argb_cache_ = NULL;    // for sanity check
+  dec->argb_cache_ = NULL;    // for soundness
   dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint8_t));
   if (dec->pixels_ == NULL) {
-    dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
-    return 0;
+    return VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
   }
   return 1;
 }
@@ -1583,7 +1621,8 @@ int VP8LDecodeAlphaHeader(ALPHDecoder* const alph_dec,
   dec->status_ = VP8_STATUS_OK;
   VP8LInitBitReader(&dec->br_, data, data_size);
 
-  if (!DecodeImageStream(alph_dec->width_, alph_dec->height_, 1, dec, NULL)) {
+  if (!DecodeImageStream(alph_dec->width_, alph_dec->height_, /*is_level0=*/1,
+                         dec, /*decoded_data=*/NULL)) {
     goto Err;
   }
 
@@ -1638,22 +1677,24 @@ int VP8LDecodeHeader(VP8LDecoder* const dec, VP8Io* const io) {
 
   if (dec == NULL) return 0;
   if (io == NULL) {
-    dec->status_ = VP8_STATUS_INVALID_PARAM;
-    return 0;
+    return VP8LSetError(dec, VP8_STATUS_INVALID_PARAM);
   }
 
   dec->io_ = io;
   dec->status_ = VP8_STATUS_OK;
   VP8LInitBitReader(&dec->br_, io->data, io->data_size);
   if (!ReadImageInfo(&dec->br_, &width, &height, &has_alpha)) {
-    dec->status_ = VP8_STATUS_BITSTREAM_ERROR;
+    VP8LSetError(dec, VP8_STATUS_BITSTREAM_ERROR);
     goto Error;
   }
   dec->state_ = READ_DIM;
   io->width = width;
   io->height = height;
 
-  if (!DecodeImageStream(width, height, 1, dec, NULL)) goto Error;
+  if (!DecodeImageStream(width, height, /*is_level0=*/1, dec,
+                         /*decoded_data=*/NULL)) {
+    goto Error;
+  }
   return 1;
 
  Error:
@@ -1666,10 +1707,9 @@ int VP8LDecodeImage(VP8LDecoder* const dec) {
   VP8Io* io = NULL;
   WebPDecParams* params = NULL;
 
-  // Sanity checks.
   if (dec == NULL) return 0;
 
-  assert(dec->hdr_.huffman_tables_ != NULL);
+  assert(dec->hdr_.huffman_tables_.root.start != NULL);
   assert(dec->hdr_.htree_groups_ != NULL);
   assert(dec->hdr_.num_htree_groups_ > 0);
 
@@ -1684,7 +1724,7 @@ int VP8LDecodeImage(VP8LDecoder* const dec) {
     assert(dec->output_ != NULL);
 
     if (!WebPIoInitFromOptions(params->options, io, MODE_BGRA)) {
-      dec->status_ = VP8_STATUS_INVALID_PARAM;
+      VP8LSetError(dec, VP8_STATUS_INVALID_PARAM);
       goto Err;
     }
 
@@ -1694,7 +1734,7 @@ int VP8LDecodeImage(VP8LDecoder* const dec) {
     if (io->use_scaling && !AllocateAndInitRescaler(dec, io)) goto Err;
 #else
     if (io->use_scaling) {
-      dec->status_ = VP8_STATUS_INVALID_PARAM;
+      VP8LSetError(dec, VP8_STATUS_INVALID_PARAM);
       goto Err;
     }
 #endif
@@ -1712,7 +1752,7 @@ int VP8LDecodeImage(VP8LDecoder* const dec) {
           dec->hdr_.saved_color_cache_.colors_ == NULL) {
         if (!VP8LColorCacheInit(&dec->hdr_.saved_color_cache_,
                                 dec->hdr_.color_cache_.hash_bits_)) {
-          dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
+          VP8LSetError(dec, VP8_STATUS_OUT_OF_MEMORY);
           goto Err;
         }
       }

3rdparty/libwebp/src/dec/vp8li_dec.h

@@ -51,7 +51,7 @@ typedef struct {
   uint32_t*       huffman_image_;
   int             num_htree_groups_;
   HTreeGroup*     htree_groups_;
-  HuffmanCode*    huffman_tables_;
+  HuffmanTables   huffman_tables_;
 } VP8LMetadata;
 
 typedef struct VP8LDecoder VP8LDecoder;
@@ -126,6 +126,19 @@ void VP8LClear(VP8LDecoder* const dec);
 // Clears and deallocate a lossless decoder instance.
 void VP8LDelete(VP8LDecoder* const dec);
 
+// Helper function for reading the different Huffman codes and storing them in
+// 'huffman_tables' and 'htree_groups'.
+// If mapping is NULL 'num_htree_groups_max' must equal 'num_htree_groups'.
+// If it is not NULL, it maps 'num_htree_groups_max' indices to the
+// 'num_htree_groups' groups. If 'num_htree_groups_max' > 'num_htree_groups',
+// some of those indices map to -1. This is used for non-balanced codes to
+// limit memory usage.
+int ReadHuffmanCodesHelper(int color_cache_bits, int num_htree_groups,
+                           int num_htree_groups_max, const int* const mapping,
+                           VP8LDecoder* const dec,
+                           HuffmanTables* const huffman_tables,
+                           HTreeGroup** const htree_groups);
+
 //------------------------------------------------------------------------------
 
 #ifdef __cplusplus

3rdparty/libwebp/src/dec/webp_dec.c

@@ -179,7 +179,7 @@ static VP8StatusCode ParseOptionalChunks(const uint8_t** const data,
       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;
+    disk_chunk_size = (CHUNK_HEADER_SIZE + chunk_size + 1) & ~1u;
     total_size += disk_chunk_size;
 
     // Check that total bytes skipped so far does not exceed riff_size.
@@ -658,19 +658,26 @@ uint8_t* WebPDecodeBGRA(const uint8_t* data, size_t data_size,
 uint8_t* WebPDecodeYUV(const uint8_t* data, size_t data_size,
                        int* width, int* height, uint8_t** u, uint8_t** v,
                        int* stride, int* uv_stride) {
-  WebPDecBuffer output;   // only to preserve the side-infos
-  uint8_t* const out = Decode(MODE_YUV, data, data_size,
-                              width, height, &output);
-
-  if (out != NULL) {
-    const WebPYUVABuffer* const buf = &output.u.YUVA;
-    *u = buf->u;
-    *v = buf->v;
-    *stride = buf->y_stride;
-    *uv_stride = buf->u_stride;
-    assert(buf->u_stride == buf->v_stride);
-  }
-  return out;
+  // data, width and height are checked by Decode().
+  if (u == NULL || v == NULL || stride == NULL || uv_stride == NULL) {
+    return NULL;
+  }
+
+  {
+    WebPDecBuffer output;   // only to preserve the side-infos
+    uint8_t* const out = Decode(MODE_YUV, data, data_size,
+                                width, height, &output);
+
+    if (out != NULL) {
+      const WebPYUVABuffer* const buf = &output.u.YUVA;
+      *u = buf->u;
+      *v = buf->v;
+      *stride = buf->y_stride;
+      *uv_stride = buf->u_stride;
+      assert(buf->u_stride == buf->v_stride);
+    }
+    return out;
+  }
 }
 
 static void DefaultFeatures(WebPBitstreamFeatures* const features) {
@@ -785,6 +792,13 @@ VP8StatusCode WebPDecode(const uint8_t* data, size_t data_size,
 //------------------------------------------------------------------------------
 // Cropping and rescaling.
 
+int WebPCheckCropDimensions(int image_width, int image_height,
+                            int x, int y, int w, int h) {
+  return !(x < 0 || y < 0 || w <= 0 || h <= 0 ||
+           x >= image_width || w > image_width || w > image_width - x ||
+           y >= image_height || h > image_height || h > image_height - y);
+}
+
 int WebPIoInitFromOptions(const WebPDecoderOptions* const options,
                           VP8Io* const io, WEBP_CSP_MODE src_colorspace) {
   const int W = io->width;
@@ -792,7 +806,7 @@ int WebPIoInitFromOptions(const WebPDecoderOptions* const options,
   int x = 0, y = 0, w = W, h = H;
 
   // Cropping
-  io->use_cropping = (options != NULL) && (options->use_cropping > 0);
+  io->use_cropping = (options != NULL) && options->use_cropping;
   if (io->use_cropping) {
     w = options->crop_width;
     h = options->crop_height;
@@ -802,7 +816,7 @@ int WebPIoInitFromOptions(const WebPDecoderOptions* const options,
       x &= ~1;
       y &= ~1;
     }
-    if (x < 0 || y < 0 || w <= 0 || h <= 0 || x + w > W || y + h > H) {
+    if (!WebPCheckCropDimensions(W, H, x, y, w, h)) {
       return 0;  // out of frame boundary error
     }
   }
@@ -814,7 +828,7 @@ int WebPIoInitFromOptions(const WebPDecoderOptions* const options,
   io->mb_h = h;
 
   // Scaling
-  io->use_scaling = (options != NULL) && (options->use_scaling > 0);
+  io->use_scaling = (options != NULL) && options->use_scaling;
   if (io->use_scaling) {
     int scaled_width = options->scaled_width;
     int scaled_height = options->scaled_height;
@@ -835,8 +849,8 @@ int WebPIoInitFromOptions(const WebPDecoderOptions* const options,
 
   if (io->use_scaling) {
     // disable filter (only for large downscaling ratio).
-    io->bypass_filtering = (io->scaled_width < W * 3 / 4) &&
-                           (io->scaled_height < H * 3 / 4);
+    io->bypass_filtering |= (io->scaled_width < W * 3 / 4) &&
+                            (io->scaled_height < H * 3 / 4);
     io->fancy_upsampling = 0;
   }
   return 1;

3rdparty/libwebp/src/dec/webpi_dec.h

@@ -77,6 +77,10 @@ VP8StatusCode WebPParseHeaders(WebPHeaderStructure* const headers);
 //------------------------------------------------------------------------------
 // Misc utils
 
+// Returns true if crop dimensions are within image bounds.
+int WebPCheckCropDimensions(int image_width, int image_height,
+                            int x, int y, int w, int h);
+
 // Initializes VP8Io with custom setup, io and teardown functions. The default
 // hooks will use the supplied 'params' as io->opaque handle.
 void WebPInitCustomIo(WebPDecParams* const params, VP8Io* const io);

3rdparty/libwebp/src/demux/anim_decode.c

@@ -23,6 +23,14 @@
 
 #define NUM_CHANNELS 4
 
+// Channel extraction from a uint32_t representation of a uint8_t RGBA/BGRA
+// buffer.
+#ifdef WORDS_BIGENDIAN
+#define CHANNEL_SHIFT(i) (24 - (i) * 8)
+#else
+#define CHANNEL_SHIFT(i) ((i) * 8)
+#endif
+
 typedef void (*BlendRowFunc)(uint32_t* const, const uint32_t* const, int);
 static void BlendPixelRowNonPremult(uint32_t* const src,
                                     const uint32_t* const dst, int num_pixels);
@@ -87,11 +95,19 @@ WebPAnimDecoder* WebPAnimDecoderNewInternal(
     int abi_version) {
   WebPAnimDecoderOptions options;
   WebPAnimDecoder* dec = NULL;
+  WebPBitstreamFeatures features;
   if (webp_data == NULL ||
       WEBP_ABI_IS_INCOMPATIBLE(abi_version, WEBP_DEMUX_ABI_VERSION)) {
     return NULL;
   }
 
+  // Validate the bitstream before doing expensive allocations. The demuxer may
+  // be more tolerant than the decoder.
+  if (WebPGetFeatures(webp_data->bytes, webp_data->size, &features) !=
+      VP8_STATUS_OK) {
+    return NULL;
+  }
+
   // Note: calloc() so that the pointer members are initialized to NULL.
   dec = (WebPAnimDecoder*)WebPSafeCalloc(1ULL, sizeof(*dec));
   if (dec == NULL) goto Error;
@@ -145,7 +161,7 @@ static int ZeroFillCanvas(uint8_t* buf, uint32_t canvas_width,
                           uint32_t canvas_height) {
   const uint64_t size =
       (uint64_t)canvas_width * canvas_height * NUM_CHANNELS * sizeof(*buf);
-  if (size != (size_t)size) return 0;
+  if (!CheckSizeOverflow(size)) return 0;
   memset(buf, 0, (size_t)size);
   return 1;
 }
@@ -166,7 +182,7 @@ static void ZeroFillFrameRect(uint8_t* buf, int buf_stride, int x_offset,
 static int CopyCanvas(const uint8_t* src, uint8_t* dst,
                       uint32_t width, uint32_t height) {
   const uint64_t size = (uint64_t)width * height * NUM_CHANNELS;
-  if (size != (size_t)size) return 0;
+  if (!CheckSizeOverflow(size)) return 0;
   assert(src != NULL && dst != NULL);
   memcpy(dst, src, (size_t)size);
   return 1;
@@ -201,35 +217,35 @@ static uint8_t BlendChannelNonPremult(uint32_t src, uint8_t src_a,
   const uint8_t dst_channel = (dst >> shift) & 0xff;
   const uint32_t blend_unscaled = src_channel * src_a + dst_channel * dst_a;
   assert(blend_unscaled < (1ULL << 32) / scale);
-  return (blend_unscaled * scale) >> 24;
+  return (blend_unscaled * scale) >> CHANNEL_SHIFT(3);
 }
 
 // Blend 'src' over 'dst' assuming they are NOT pre-multiplied by alpha.
 static uint32_t BlendPixelNonPremult(uint32_t src, uint32_t dst) {
-  const uint8_t src_a = (src >> 24) & 0xff;
+  const uint8_t src_a = (src >> CHANNEL_SHIFT(3)) & 0xff;
 
   if (src_a == 0) {
     return dst;
   } else {
-    const uint8_t dst_a = (dst >> 24) & 0xff;
+    const uint8_t dst_a = (dst >> CHANNEL_SHIFT(3)) & 0xff;
     // This is the approximate integer arithmetic for the actual formula:
     // dst_factor_a = (dst_a * (255 - src_a)) / 255.
     const uint8_t dst_factor_a = (dst_a * (256 - src_a)) >> 8;
     const uint8_t blend_a = src_a + dst_factor_a;
     const uint32_t scale = (1UL << 24) / blend_a;
 
-    const uint8_t blend_r =
-        BlendChannelNonPremult(src, src_a, dst, dst_factor_a, scale, 0);
-    const uint8_t blend_g =
-        BlendChannelNonPremult(src, src_a, dst, dst_factor_a, scale, 8);
-    const uint8_t blend_b =
-        BlendChannelNonPremult(src, src_a, dst, dst_factor_a, scale, 16);
+    const uint8_t blend_r = BlendChannelNonPremult(
+        src, src_a, dst, dst_factor_a, scale, CHANNEL_SHIFT(0));
+    const uint8_t blend_g = BlendChannelNonPremult(
+        src, src_a, dst, dst_factor_a, scale, CHANNEL_SHIFT(1));
+    const uint8_t blend_b = BlendChannelNonPremult(
+        src, src_a, dst, dst_factor_a, scale, CHANNEL_SHIFT(2));
     assert(src_a + dst_factor_a < 256);
 
-    return (blend_r << 0) |
-           (blend_g << 8) |
-           (blend_b << 16) |
-           ((uint32_t)blend_a << 24);
+    return ((uint32_t)blend_r << CHANNEL_SHIFT(0)) |
+           ((uint32_t)blend_g << CHANNEL_SHIFT(1)) |
+           ((uint32_t)blend_b << CHANNEL_SHIFT(2)) |
+           ((uint32_t)blend_a << CHANNEL_SHIFT(3));
   }
 }
 
@@ -239,7 +255,7 @@ static void BlendPixelRowNonPremult(uint32_t* const src,
                                     const uint32_t* const dst, int num_pixels) {
   int i;
   for (i = 0; i < num_pixels; ++i) {
-    const uint8_t src_alpha = (src[i] >> 24) & 0xff;
+    const uint8_t src_alpha = (src[i] >> CHANNEL_SHIFT(3)) & 0xff;
     if (src_alpha != 0xff) {
       src[i] = BlendPixelNonPremult(src[i], dst[i]);
     }
@@ -256,7 +272,7 @@ static WEBP_INLINE uint32_t ChannelwiseMultiply(uint32_t pix, uint32_t scale) {
 
 // Blend 'src' over 'dst' assuming they are pre-multiplied by alpha.
 static uint32_t BlendPixelPremult(uint32_t src, uint32_t dst) {
-  const uint8_t src_a = (src >> 24) & 0xff;
+  const uint8_t src_a = (src >> CHANNEL_SHIFT(3)) & 0xff;
   return src + ChannelwiseMultiply(dst, 256 - src_a);
 }
 
@@ -266,7 +282,7 @@ static void BlendPixelRowPremult(uint32_t* const src, const uint32_t* const dst,
                                  int num_pixels) {
   int i;
   for (i = 0; i < num_pixels; ++i) {
-    const uint8_t src_alpha = (src[i] >> 24) & 0xff;
+    const uint8_t src_alpha = (src[i] >> CHANNEL_SHIFT(3)) & 0xff;
     if (src_alpha != 0xff) {
       src[i] = BlendPixelPremult(src[i], dst[i]);
     }

3rdparty/libwebp/src/demux/demux.c

@@ -24,8 +24,8 @@
 #include "src/webp/format_constants.h"
 
 #define DMUX_MAJ_VERSION 1
-#define DMUX_MIN_VERSION 2
-#define DMUX_REV_VERSION 0
+#define DMUX_MIN_VERSION 3
+#define DMUX_REV_VERSION 1
 
 typedef struct {
   size_t start_;        // start location of the data
@@ -221,12 +221,16 @@ static ParseStatus StoreFrame(int frame_num, uint32_t min_size,
     const size_t chunk_start_offset = mem->start_;
     const uint32_t fourcc = ReadLE32(mem);
     const uint32_t payload_size = ReadLE32(mem);
-    const uint32_t payload_size_padded = payload_size + (payload_size & 1);
-    const size_t payload_available = (payload_size_padded > MemDataSize(mem))
-                                   ? MemDataSize(mem) : payload_size_padded;
-    const size_t chunk_size = CHUNK_HEADER_SIZE + payload_available;
+    uint32_t payload_size_padded;
+    size_t payload_available;
+    size_t chunk_size;
 
     if (payload_size > MAX_CHUNK_PAYLOAD) return PARSE_ERROR;
+
+    payload_size_padded = payload_size + (payload_size & 1);
+    payload_available = (payload_size_padded > MemDataSize(mem))
+                      ? MemDataSize(mem) : payload_size_padded;
+    chunk_size = CHUNK_HEADER_SIZE + payload_available;
     if (SizeIsInvalid(mem, payload_size_padded)) return PARSE_ERROR;
     if (payload_size_padded > MemDataSize(mem)) status = PARSE_NEED_MORE_DATA;
 
@@ -451,9 +455,11 @@ static ParseStatus ParseVP8XChunks(WebPDemuxer* const dmux) {
     const size_t chunk_start_offset = mem->start_;
     const uint32_t fourcc = ReadLE32(mem);
     const uint32_t chunk_size = ReadLE32(mem);
-    const uint32_t chunk_size_padded = chunk_size + (chunk_size & 1);
+    uint32_t chunk_size_padded;
 
     if (chunk_size > MAX_CHUNK_PAYLOAD) return PARSE_ERROR;
+
+    chunk_size_padded = chunk_size + (chunk_size & 1);
     if (SizeIsInvalid(mem, chunk_size_padded)) return PARSE_ERROR;
 
     switch (fourcc) {
@@ -608,7 +614,6 @@ static int IsValidExtendedFormat(const WebPDemuxer* const dmux) {
 
   while (f != NULL) {
     const int cur_frame_set = f->frame_num_;
-    int frame_count = 0;
 
     // Check frame properties.
     for (; f != NULL && f->frame_num_ == cur_frame_set; f = f->next_) {
@@ -643,8 +648,6 @@ static int IsValidExtendedFormat(const WebPDemuxer* const dmux) {
                             dmux->canvas_width_, dmux->canvas_height_)) {
         return 0;
       }
-
-      ++frame_count;
     }
   }
   return 1;

3rdparty/libwebp/src/dsp/alpha_processing.c

@@ -157,7 +157,8 @@ void WebPMultARGBRow_C(uint32_t* const ptr, int width, int inverse) {
   }
 }
 
-void WebPMultRow_C(uint8_t* const ptr, const uint8_t* const alpha,
+void WebPMultRow_C(uint8_t* WEBP_RESTRICT const ptr,
+                   const uint8_t* WEBP_RESTRICT const alpha,
                    int width, int inverse) {
   int x;
   for (x = 0; x < width; ++x) {
@@ -178,7 +179,8 @@ void WebPMultRow_C(uint8_t* const ptr, const uint8_t* const alpha,
 #undef MFIX
 
 void (*WebPMultARGBRow)(uint32_t* const ptr, int width, int inverse);
-void (*WebPMultRow)(uint8_t* const ptr, const uint8_t* const alpha,
+void (*WebPMultRow)(uint8_t* WEBP_RESTRICT const ptr,
+                    const uint8_t* WEBP_RESTRICT const alpha,
                     int width, int inverse);
 
 //------------------------------------------------------------------------------
@@ -193,8 +195,8 @@ void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows,
   }
 }
 
-void WebPMultRows(uint8_t* ptr, int stride,
-                  const uint8_t* alpha, int alpha_stride,
+void WebPMultRows(uint8_t* WEBP_RESTRICT ptr, int stride,
+                  const uint8_t* WEBP_RESTRICT alpha, int alpha_stride,
                   int width, int num_rows, int inverse) {
   int n;
   for (n = 0; n < num_rows; ++n) {
@@ -290,9 +292,9 @@ static void ApplyAlphaMultiply_16b_C(uint8_t* rgba4444,
 }
 
 #if !WEBP_NEON_OMIT_C_CODE
-static int DispatchAlpha_C(const uint8_t* alpha, int alpha_stride,
+static int DispatchAlpha_C(const uint8_t* WEBP_RESTRICT alpha, int alpha_stride,
                            int width, int height,
-                           uint8_t* dst, int dst_stride) {
+                           uint8_t* WEBP_RESTRICT dst, int dst_stride) {
   uint32_t alpha_mask = 0xff;
   int i, j;
 
@@ -309,9 +311,10 @@ static int DispatchAlpha_C(const uint8_t* alpha, int alpha_stride,
   return (alpha_mask != 0xff);
 }
 
-static void DispatchAlphaToGreen_C(const uint8_t* alpha, int alpha_stride,
-                                   int width, int height,
-                                   uint32_t* dst, int dst_stride) {
+static void DispatchAlphaToGreen_C(const uint8_t* WEBP_RESTRICT alpha,
+                                   int alpha_stride, int width, int height,
+                                   uint32_t* WEBP_RESTRICT dst,
+                                   int dst_stride) {
   int i, j;
   for (j = 0; j < height; ++j) {
     for (i = 0; i < width; ++i) {
@@ -322,9 +325,9 @@ static void DispatchAlphaToGreen_C(const uint8_t* alpha, int alpha_stride,
   }
 }
 
-static int ExtractAlpha_C(const uint8_t* argb, int argb_stride,
+static int ExtractAlpha_C(const uint8_t* WEBP_RESTRICT argb, int argb_stride,
                           int width, int height,
-                          uint8_t* alpha, int alpha_stride) {
+                          uint8_t* WEBP_RESTRICT alpha, int alpha_stride) {
   uint8_t alpha_mask = 0xff;
   int i, j;
 
@@ -340,7 +343,8 @@ static int ExtractAlpha_C(const uint8_t* argb, int argb_stride,
   return (alpha_mask == 0xff);
 }
 
-static void ExtractGreen_C(const uint32_t* argb, uint8_t* alpha, int size) {
+static void ExtractGreen_C(const uint32_t* WEBP_RESTRICT argb,
+                           uint8_t* WEBP_RESTRICT alpha, int size) {
   int i;
   for (i = 0; i < size; ++i) alpha[i] = argb[i] >> 8;
 }
@@ -372,8 +376,11 @@ static WEBP_INLINE uint32_t MakeARGB32(int a, int r, int g, int b) {
 }
 
 #ifdef WORDS_BIGENDIAN
-static void PackARGB_C(const uint8_t* a, const uint8_t* r, const uint8_t* g,
-                       const uint8_t* b, int len, uint32_t* out) {
+static void PackARGB_C(const uint8_t* WEBP_RESTRICT a,
+                       const uint8_t* WEBP_RESTRICT r,
+                       const uint8_t* WEBP_RESTRICT g,
+                       const uint8_t* WEBP_RESTRICT b,
+                       int len, uint32_t* WEBP_RESTRICT out) {
   int i;
   for (i = 0; i < len; ++i) {
     out[i] = MakeARGB32(a[4 * i], r[4 * i], g[4 * i], b[4 * i]);
@@ -381,8 +388,10 @@ static void PackARGB_C(const uint8_t* a, const uint8_t* r, const uint8_t* g,
 }
 #endif
 
-static void PackRGB_C(const uint8_t* r, const uint8_t* g, const uint8_t* b,
-                      int len, int step, uint32_t* out) {
+static void PackRGB_C(const uint8_t* WEBP_RESTRICT r,
+                      const uint8_t* WEBP_RESTRICT g,
+                      const uint8_t* WEBP_RESTRICT b,
+                      int len, int step, uint32_t* WEBP_RESTRICT out) {
   int i, offset = 0;
   for (i = 0; i < len; ++i) {
     out[i] = MakeARGB32(0xff, r[offset], g[offset], b[offset]);
@@ -392,16 +401,22 @@ static void PackRGB_C(const uint8_t* r, const uint8_t* g, const uint8_t* b,
 
 void (*WebPApplyAlphaMultiply)(uint8_t*, int, int, int, int);
 void (*WebPApplyAlphaMultiply4444)(uint8_t*, int, int, int);
-int (*WebPDispatchAlpha)(const uint8_t*, int, int, int, uint8_t*, int);
-void (*WebPDispatchAlphaToGreen)(const uint8_t*, int, int, int, uint32_t*, int);
-int (*WebPExtractAlpha)(const uint8_t*, int, int, int, uint8_t*, int);
-void (*WebPExtractGreen)(const uint32_t* argb, uint8_t* alpha, int size);
+int (*WebPDispatchAlpha)(const uint8_t* WEBP_RESTRICT, int, int, int,
+                         uint8_t* WEBP_RESTRICT, int);
+void (*WebPDispatchAlphaToGreen)(const uint8_t* WEBP_RESTRICT, int, int, int,
+                                 uint32_t* WEBP_RESTRICT, int);
+int (*WebPExtractAlpha)(const uint8_t* WEBP_RESTRICT, int, int, int,
+                        uint8_t* WEBP_RESTRICT, int);
+void (*WebPExtractGreen)(const uint32_t* WEBP_RESTRICT argb,
+                         uint8_t* WEBP_RESTRICT alpha, int size);
 #ifdef WORDS_BIGENDIAN
 void (*WebPPackARGB)(const uint8_t* a, const uint8_t* r, const uint8_t* g,
                      const uint8_t* b, int, uint32_t*);
 #endif
-void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
-                    int len, int step, uint32_t* out);
+void (*WebPPackRGB)(const uint8_t* WEBP_RESTRICT r,
+                    const uint8_t* WEBP_RESTRICT g,
+                    const uint8_t* WEBP_RESTRICT b,
+                    int len, int step, uint32_t* WEBP_RESTRICT out);
 
 int (*WebPHasAlpha8b)(const uint8_t* src, int length);
 int (*WebPHasAlpha32b)(const uint8_t* src, int length);
@@ -410,6 +425,7 @@ void (*WebPAlphaReplace)(uint32_t* src, int length, uint32_t color);
 //------------------------------------------------------------------------------
 // Init function
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void WebPInitAlphaProcessingMIPSdspR2(void);
 extern void WebPInitAlphaProcessingSSE2(void);
 extern void WebPInitAlphaProcessingSSE41(void);
@@ -438,10 +454,10 @@ WEBP_DSP_INIT_FUNC(WebPInitAlphaProcessing) {
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       WebPInitAlphaProcessingSSE2();
-#if defined(WEBP_USE_SSE41)
+#if defined(WEBP_HAVE_SSE41)
       if (VP8GetCPUInfo(kSSE4_1)) {
         WebPInitAlphaProcessingSSE41();
       }
@@ -455,7 +471,7 @@ WEBP_DSP_INIT_FUNC(WebPInitAlphaProcessing) {
 #endif
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     WebPInitAlphaProcessingNEON();

3rdparty/libwebp/src/dsp/alpha_processing_neon.c

@@ -80,10 +80,10 @@ static void ApplyAlphaMultiply_NEON(uint8_t* rgba, int alpha_first,
 
 //------------------------------------------------------------------------------
 
-static int DispatchAlpha_NEON(const uint8_t* alpha, int alpha_stride,
-                              int width, int height,
-                              uint8_t* dst, int dst_stride) {
-  uint32_t alpha_mask = 0xffffffffu;
+static int DispatchAlpha_NEON(const uint8_t* WEBP_RESTRICT alpha,
+                              int alpha_stride, int width, int height,
+                              uint8_t* WEBP_RESTRICT dst, int dst_stride) {
+  uint32_t alpha_mask = 0xffu;
   uint8x8_t mask8 = vdup_n_u8(0xff);
   uint32_t tmp[2];
   int i, j;
@@ -107,14 +107,16 @@ static int DispatchAlpha_NEON(const uint8_t* alpha, int alpha_stride,
     dst += dst_stride;
   }
   vst1_u8((uint8_t*)tmp, mask8);
+  alpha_mask *= 0x01010101;
   alpha_mask &= tmp[0];
   alpha_mask &= tmp[1];
   return (alpha_mask != 0xffffffffu);
 }
 
-static void DispatchAlphaToGreen_NEON(const uint8_t* alpha, int alpha_stride,
-                                      int width, int height,
-                                      uint32_t* dst, int dst_stride) {
+static void DispatchAlphaToGreen_NEON(const uint8_t* WEBP_RESTRICT alpha,
+                                      int alpha_stride, int width, int height,
+                                      uint32_t* WEBP_RESTRICT dst,
+                                      int dst_stride) {
   int i, j;
   uint8x8x4_t greens;   // leave A/R/B channels zero'd.
   greens.val[0] = vdup_n_u8(0);
@@ -131,10 +133,10 @@ static void DispatchAlphaToGreen_NEON(const uint8_t* alpha, int alpha_stride,
   }
 }
 
-static int ExtractAlpha_NEON(const uint8_t* argb, int argb_stride,
+static int ExtractAlpha_NEON(const uint8_t* WEBP_RESTRICT argb, int argb_stride,
                              int width, int height,
-                             uint8_t* alpha, int alpha_stride) {
-  uint32_t alpha_mask = 0xffffffffu;
+                             uint8_t* WEBP_RESTRICT alpha, int alpha_stride) {
+  uint32_t alpha_mask = 0xffu;
   uint8x8_t mask8 = vdup_n_u8(0xff);
   uint32_t tmp[2];
   int i, j;
@@ -156,13 +158,14 @@ static int ExtractAlpha_NEON(const uint8_t* argb, int argb_stride,
     alpha += alpha_stride;
   }
   vst1_u8((uint8_t*)tmp, mask8);
+  alpha_mask *= 0x01010101;
   alpha_mask &= tmp[0];
   alpha_mask &= tmp[1];
   return (alpha_mask == 0xffffffffu);
 }
 
-static void ExtractGreen_NEON(const uint32_t* argb,
-                              uint8_t* alpha, int size) {
+static void ExtractGreen_NEON(const uint32_t* WEBP_RESTRICT argb,
+                              uint8_t* WEBP_RESTRICT alpha, int size) {
   int i;
   for (i = 0; i + 16 <= size; i += 16) {
     const uint8x16x4_t rgbX = vld4q_u8((const uint8_t*)(argb + i));

3rdparty/libwebp/src/dsp/alpha_processing_sse2.c

@@ -18,16 +18,16 @@
 
 //------------------------------------------------------------------------------
 
-static int DispatchAlpha_SSE2(const uint8_t* alpha, int alpha_stride,
-                              int width, int height,
-                              uint8_t* dst, int dst_stride) {
+static int DispatchAlpha_SSE2(const uint8_t* WEBP_RESTRICT alpha,
+                              int alpha_stride, int width, int height,
+                              uint8_t* WEBP_RESTRICT dst, int dst_stride) {
   // alpha_and stores an 'and' operation of all the alpha[] values. The final
   // value is not 0xff if any of the alpha[] is not equal to 0xff.
   uint32_t alpha_and = 0xff;
   int i, j;
   const __m128i zero = _mm_setzero_si128();
-  const __m128i rgb_mask = _mm_set1_epi32(0xffffff00u);  // to preserve RGB
-  const __m128i all_0xff = _mm_set_epi32(0, 0, ~0u, ~0u);
+  const __m128i rgb_mask = _mm_set1_epi32((int)0xffffff00);  // to preserve RGB
+  const __m128i all_0xff = _mm_set_epi32(0, 0, ~0, ~0);
   __m128i all_alphas = all_0xff;
 
   // We must be able to access 3 extra bytes after the last written byte
@@ -72,9 +72,10 @@ static int DispatchAlpha_SSE2(const uint8_t* alpha, int alpha_stride,
   return (alpha_and != 0xff);
 }
 
-static void DispatchAlphaToGreen_SSE2(const uint8_t* alpha, int alpha_stride,
-                                      int width, int height,
-                                      uint32_t* dst, int dst_stride) {
+static void DispatchAlphaToGreen_SSE2(const uint8_t* WEBP_RESTRICT alpha,
+                                      int alpha_stride, int width, int height,
+                                      uint32_t* WEBP_RESTRICT dst,
+                                      int dst_stride) {
   int i, j;
   const __m128i zero = _mm_setzero_si128();
   const int limit = width & ~15;
@@ -98,15 +99,15 @@ static void DispatchAlphaToGreen_SSE2(const uint8_t* alpha, int alpha_stride,
   }
 }
 
-static int ExtractAlpha_SSE2(const uint8_t* argb, int argb_stride,
+static int ExtractAlpha_SSE2(const uint8_t* WEBP_RESTRICT argb, int argb_stride,
                              int width, int height,
-                             uint8_t* alpha, int alpha_stride) {
+                             uint8_t* WEBP_RESTRICT alpha, int alpha_stride) {
   // alpha_and stores an 'and' operation of all the alpha[] values. The final
   // value is not 0xff if any of the alpha[] is not equal to 0xff.
   uint32_t alpha_and = 0xff;
   int i, j;
-  const __m128i a_mask = _mm_set1_epi32(0xffu);  // to preserve alpha
-  const __m128i all_0xff = _mm_set_epi32(0, 0, ~0u, ~0u);
+  const __m128i a_mask = _mm_set1_epi32(0xff);  // to preserve alpha
+  const __m128i all_0xff = _mm_set_epi32(0, 0, ~0, ~0);
   __m128i all_alphas = all_0xff;
 
   // We must be able to access 3 extra bytes after the last written byte
@@ -143,6 +144,46 @@ static int ExtractAlpha_SSE2(const uint8_t* argb, int argb_stride,
   return (alpha_and == 0xff);
 }
 
+static void ExtractGreen_SSE2(const uint32_t* WEBP_RESTRICT argb,
+                              uint8_t* WEBP_RESTRICT alpha, int size) {
+  int i;
+  const __m128i mask = _mm_set1_epi32(0xff);
+  const __m128i* src = (const __m128i*)argb;
+
+  for (i = 0; i + 16 <= size; i += 16, src += 4) {
+    const __m128i a0 = _mm_loadu_si128(src + 0);
+    const __m128i a1 = _mm_loadu_si128(src + 1);
+    const __m128i a2 = _mm_loadu_si128(src + 2);
+    const __m128i a3 = _mm_loadu_si128(src + 3);
+    const __m128i b0 = _mm_srli_epi32(a0, 8);
+    const __m128i b1 = _mm_srli_epi32(a1, 8);
+    const __m128i b2 = _mm_srli_epi32(a2, 8);
+    const __m128i b3 = _mm_srli_epi32(a3, 8);
+    const __m128i c0 = _mm_and_si128(b0, mask);
+    const __m128i c1 = _mm_and_si128(b1, mask);
+    const __m128i c2 = _mm_and_si128(b2, mask);
+    const __m128i c3 = _mm_and_si128(b3, mask);
+    const __m128i d0 = _mm_packs_epi32(c0, c1);
+    const __m128i d1 = _mm_packs_epi32(c2, c3);
+    const __m128i e = _mm_packus_epi16(d0, d1);
+    // store
+    _mm_storeu_si128((__m128i*)&alpha[i], e);
+  }
+  if (i + 8 <= size) {
+    const __m128i a0 = _mm_loadu_si128(src + 0);
+    const __m128i a1 = _mm_loadu_si128(src + 1);
+    const __m128i b0 = _mm_srli_epi32(a0, 8);
+    const __m128i b1 = _mm_srli_epi32(a1, 8);
+    const __m128i c0 = _mm_and_si128(b0, mask);
+    const __m128i c1 = _mm_and_si128(b1, mask);
+    const __m128i d = _mm_packs_epi32(c0, c1);
+    const __m128i e = _mm_packus_epi16(d, d);
+    _mm_storel_epi64((__m128i*)&alpha[i], e);
+    i += 8;
+  }
+  for (; i < size; ++i) alpha[i] = argb[i] >> 8;
+}
+
 //------------------------------------------------------------------------------
 // Non-dither premultiplied modes
 
@@ -177,7 +218,7 @@ static int ExtractAlpha_SSE2(const uint8_t* argb, int argb_stride,
 static void ApplyAlphaMultiply_SSE2(uint8_t* rgba, int alpha_first,
                                     int w, int h, int stride) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i kMult = _mm_set1_epi16(0x8081u);
+  const __m128i kMult = _mm_set1_epi16((short)0x8081);
   const __m128i kMask = _mm_set_epi16(0, 0xff, 0xff, 0, 0, 0xff, 0xff, 0);
   const int kSpan = 4;
   while (h-- > 0) {
@@ -266,7 +307,7 @@ static int HasAlpha32b_SSE2(const uint8_t* src, int length) {
 }
 
 static void AlphaReplace_SSE2(uint32_t* src, int length, uint32_t color) {
-  const __m128i m_color = _mm_set1_epi32(color);
+  const __m128i m_color = _mm_set1_epi32((int)color);
   const __m128i zero = _mm_setzero_si128();
   int i = 0;
   for (; i + 8 <= length; i += 8) {
@@ -317,7 +358,8 @@ static void MultARGBRow_SSE2(uint32_t* const ptr, int width, int inverse) {
   if (width > 0) WebPMultARGBRow_C(ptr + x, width, inverse);
 }
 
-static void MultRow_SSE2(uint8_t* const ptr, const uint8_t* const alpha,
+static void MultRow_SSE2(uint8_t* WEBP_RESTRICT const ptr,
+                         const uint8_t* WEBP_RESTRICT const alpha,
                          int width, int inverse) {
   int x = 0;
   if (!inverse) {
@@ -352,6 +394,7 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessingSSE2(void) {
   WebPDispatchAlpha = DispatchAlpha_SSE2;
   WebPDispatchAlphaToGreen = DispatchAlphaToGreen_SSE2;
   WebPExtractAlpha = ExtractAlpha_SSE2;
+  WebPExtractGreen = ExtractGreen_SSE2;
 
   WebPHasAlpha8b = HasAlpha8b_SSE2;
   WebPHasAlpha32b = HasAlpha32b_SSE2;

3rdparty/libwebp/src/dsp/alpha_processing_sse41.c

@@ -19,14 +19,14 @@
 
 //------------------------------------------------------------------------------
 
-static int ExtractAlpha_SSE41(const uint8_t* argb, int argb_stride,
-                              int width, int height,
-                              uint8_t* alpha, int alpha_stride) {
+static int ExtractAlpha_SSE41(const uint8_t* WEBP_RESTRICT argb,
+                              int argb_stride, int width, int height,
+                              uint8_t* WEBP_RESTRICT alpha, int alpha_stride) {
   // alpha_and stores an 'and' operation of all the alpha[] values. The final
   // value is not 0xff if any of the alpha[] is not equal to 0xff.
   uint32_t alpha_and = 0xff;
   int i, j;
-  const __m128i all_0xff = _mm_set1_epi32(~0u);
+  const __m128i all_0xff = _mm_set1_epi32(~0);
   __m128i all_alphas = all_0xff;
 
   // We must be able to access 3 extra bytes after the last written byte

3rdparty/libwebp/src/dsp/cost.c

@@ -374,6 +374,7 @@ static void SetResidualCoeffs_C(const int16_t* const coeffs,
 VP8GetResidualCostFunc VP8GetResidualCost;
 VP8SetResidualCoeffsFunc VP8SetResidualCoeffs;
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8EncDspCostInitMIPS32(void);
 extern void VP8EncDspCostInitMIPSdspR2(void);
 extern void VP8EncDspCostInitSSE2(void);
@@ -395,12 +396,12 @@ WEBP_DSP_INIT_FUNC(VP8EncDspCostInit) {
       VP8EncDspCostInitMIPSdspR2();
     }
 #endif
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8EncDspCostInitSSE2();
     }
 #endif
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
     if (VP8GetCPUInfo(kNEON)) {
       VP8EncDspCostInitNEON();
     }

3rdparty/libwebp/src/dsp/cost_neon.c

@@ -29,7 +29,7 @@ static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
   const uint8x16_t eob = vcombine_u8(vqmovn_u16(eob_0), vqmovn_u16(eob_1));
   const uint8x16_t masked = vandq_u8(eob, vld1q_u8(position));
 
-#ifdef __aarch64__
+#if WEBP_AARCH64
   res->last = vmaxvq_u8(masked) - 1;
 #else
   const uint8x8_t eob_8x8 = vmax_u8(vget_low_u8(masked), vget_high_u8(masked));
@@ -43,7 +43,7 @@ static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
 
   vst1_lane_s32(&res->last, vreinterpret_s32_u32(eob_32x2), 0);
   --res->last;
-#endif  // __aarch64__
+#endif  // WEBP_AARCH64
 
   res->coeffs = coeffs;
 }

3rdparty/libwebp/src/dsp/cpu.c

@@ -11,7 +11,7 @@
 //
 // Author: Christian Duvivier (cduvivier@google.com)
 
-#include "src/dsp/dsp.h"
+#include "src/dsp/cpu.h"
 
 #if defined(WEBP_HAVE_NEON_RTCD)
 #include <stdio.h>
@@ -173,6 +173,7 @@ static int x86CPUInfo(CPUFeature feature) {
   }
   return 0;
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = x86CPUInfo;
 #elif defined(WEBP_ANDROID_NEON)  // NB: needs to be before generic NEON test.
 static int AndroidCPUInfo(CPUFeature feature) {
@@ -184,22 +185,23 @@ static int AndroidCPUInfo(CPUFeature feature) {
   }
   return 0;
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 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
+#ifdef WEBP_HAVE_SSE2
     case kSSE2:
       return 1;
 #endif
-#ifdef WEBP_USE_SSE41
+#ifdef WEBP_HAVE_SSE41
     case kSSE3:
     case kSlowSSSE3:
     case kSSE4_1:
       return 1;
 #endif
-#ifdef WEBP_USE_NEON
+#ifdef WEBP_HAVE_NEON
     case kNEON:
       return 1;
 #endif
@@ -208,10 +210,12 @@ static int wasmCPUInfo(CPUFeature feature) {
   }
   return 0;
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = wasmCPUInfo;
-#elif defined(WEBP_USE_NEON)
-// define a dummy function to enable turning off NEON at runtime by setting
-// VP8DecGetCPUInfo = NULL
+#elif defined(WEBP_HAVE_NEON)
+// In most cases this function doesn't check for NEON support (it's assumed by
+// the configuration), but enables turning off NEON at runtime, for testing
+// purposes, by setting VP8GetCPUInfo = NULL.
 static int armCPUInfo(CPUFeature feature) {
   if (feature != kNEON) return 0;
 #if defined(__linux__) && defined(WEBP_HAVE_NEON_RTCD)
@@ -235,6 +239,7 @@ static int armCPUInfo(CPUFeature feature) {
   return 1;
 #endif
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = armCPUInfo;
 #elif defined(WEBP_USE_MIPS32) || defined(WEBP_USE_MIPS_DSP_R2) || \
       defined(WEBP_USE_MSA)
@@ -246,7 +251,9 @@ static int mipsCPUInfo(CPUFeature feature) {
   }
 
 }
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = mipsCPUInfo;
 #else
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 VP8CPUInfo VP8GetCPUInfo = NULL;
 #endif

3rdparty/libwebp/src/dsp/cpu.h

@@ -0,0 +1,266 @@
+// Copyright 2022 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+//   CPU detection functions and macros.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#ifndef WEBP_DSP_CPU_H_
+#define WEBP_DSP_CPU_H_
+
+#include <stddef.h>
+
+#ifdef HAVE_CONFIG_H
+#include "src/webp/config.h"
+#endif
+
+#include "src/webp/types.h"
+
+#if defined(__GNUC__)
+#define LOCAL_GCC_VERSION ((__GNUC__ << 8) | __GNUC_MINOR__)
+#define LOCAL_GCC_PREREQ(maj, min) (LOCAL_GCC_VERSION >= (((maj) << 8) | (min)))
+#else
+#define LOCAL_GCC_VERSION 0
+#define LOCAL_GCC_PREREQ(maj, min) 0
+#endif
+
+#if defined(__clang__)
+#define LOCAL_CLANG_VERSION ((__clang_major__ << 8) | __clang_minor__)
+#define LOCAL_CLANG_PREREQ(maj, min) \
+  (LOCAL_CLANG_VERSION >= (((maj) << 8) | (min)))
+#else
+#define LOCAL_CLANG_VERSION 0
+#define LOCAL_CLANG_PREREQ(maj, min) 0
+#endif
+
+#ifndef __has_builtin
+#define __has_builtin(x) 0
+#endif
+
+//------------------------------------------------------------------------------
+// x86 defines.
+
+#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
+#endif
+
+#if defined(_MSC_VER) && _MSC_VER >= 1500 && \
+    (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.
+// Files containing intrinsics will need to be built targeting the instruction
+// set so should succeed on one of the earlier tests.
+#if (defined(__SSE2__) || defined(WEBP_MSC_SSE2)) && \
+    (!defined(HAVE_CONFIG_H) || defined(WEBP_HAVE_SSE2))
+#define WEBP_USE_SSE2
+#endif
+
+#if defined(WEBP_USE_SSE2) && !defined(WEBP_HAVE_SSE2)
+#define WEBP_HAVE_SSE2
+#endif
+
+#if (defined(__SSE4_1__) || defined(WEBP_MSC_SSE41)) && \
+    (!defined(HAVE_CONFIG_H) || defined(WEBP_HAVE_SSE41))
+#define WEBP_USE_SSE41
+#endif
+
+#if defined(WEBP_USE_SSE41) && !defined(WEBP_HAVE_SSE41)
+#define WEBP_HAVE_SSE41
+#endif
+
+#undef WEBP_MSC_SSE41
+#undef WEBP_MSC_SSE2
+
+//------------------------------------------------------------------------------
+// Arm defines.
+
+// 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__) || defined(__aarch64__)) &&       \
+     (!defined(HAVE_CONFIG_H) || defined(WEBP_HAVE_NEON))) && \
+    !defined(__native_client__)
+#define WEBP_USE_NEON
+#endif
+
+#if !defined(WEBP_USE_NEON) && defined(__ANDROID__) && \
+    defined(__ARM_ARCH_7A__) && defined(HAVE_CPU_FEATURES_H)
+#define WEBP_ANDROID_NEON  // Android targets that may have NEON
+#define WEBP_USE_NEON
+#endif
+
+// Note: ARM64 is supported in Visual Studio 2017, but requires the direct
+// inclusion of arm64_neon.h; Visual Studio 2019 includes this file in
+// arm_neon.h. Compile errors were seen with Visual Studio 2019 16.4 with
+// vtbl4_u8(); a fix was made in 16.6.
+#if defined(_MSC_VER) && \
+    ((_MSC_VER >= 1700 && defined(_M_ARM)) || \
+     (_MSC_VER >= 1926 && (defined(_M_ARM64) || defined(_M_ARM64EC))))
+#define WEBP_USE_NEON
+#define WEBP_USE_INTRINSICS
+#endif
+
+#if defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC)
+#define WEBP_AARCH64 1
+#else
+#define WEBP_AARCH64 0
+#endif
+
+#if defined(WEBP_USE_NEON) && !defined(WEBP_HAVE_NEON)
+#define WEBP_HAVE_NEON
+#endif
+
+//------------------------------------------------------------------------------
+// MIPS defines.
+
+#if defined(__mips__) && !defined(__mips64) && defined(__mips_isa_rev) && \
+    (__mips_isa_rev >= 1) && (__mips_isa_rev < 6)
+#define WEBP_USE_MIPS32
+#if (__mips_isa_rev >= 2)
+#define WEBP_USE_MIPS32_R2
+#if defined(__mips_dspr2) || (defined(__mips_dsp_rev) && __mips_dsp_rev >= 2)
+#define WEBP_USE_MIPS_DSP_R2
+#endif
+#endif
+#endif
+
+#if defined(__mips_msa) && defined(__mips_isa_rev) && (__mips_isa_rev >= 5)
+#define WEBP_USE_MSA
+#endif
+
+//------------------------------------------------------------------------------
+
+#ifndef WEBP_DSP_OMIT_C_CODE
+#define WEBP_DSP_OMIT_C_CODE 1
+#endif
+
+#if defined(WEBP_USE_NEON) && WEBP_DSP_OMIT_C_CODE
+#define WEBP_NEON_OMIT_C_CODE 1
+#else
+#define WEBP_NEON_OMIT_C_CODE 0
+#endif
+
+#if !(LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 8) || WEBP_AARCH64)
+#define WEBP_NEON_WORK_AROUND_GCC 1
+#else
+#define WEBP_NEON_WORK_AROUND_GCC 0
+#endif
+
+//------------------------------------------------------------------------------
+
+// This macro prevents thread_sanitizer from reporting known concurrent writes.
+#define WEBP_TSAN_IGNORE_FUNCTION
+#if defined(__has_feature)
+#if __has_feature(thread_sanitizer)
+#undef WEBP_TSAN_IGNORE_FUNCTION
+#define WEBP_TSAN_IGNORE_FUNCTION __attribute__((no_sanitize_thread))
+#endif
+#endif
+
+#if defined(__has_feature)
+#if __has_feature(memory_sanitizer)
+#define WEBP_MSAN
+#endif
+#endif
+
+#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
+#include <pthread.h>  // NOLINT
+
+#define WEBP_DSP_INIT(func)                                         \
+  do {                                                              \
+    static volatile VP8CPUInfo func##_last_cpuinfo_used =           \
+        (VP8CPUInfo)&func##_last_cpuinfo_used;                      \
+    static pthread_mutex_t func##_lock = PTHREAD_MUTEX_INITIALIZER; \
+    if (pthread_mutex_lock(&func##_lock)) break;                    \
+    if (func##_last_cpuinfo_used != VP8GetCPUInfo) func();          \
+    func##_last_cpuinfo_used = VP8GetCPUInfo;                       \
+    (void)pthread_mutex_unlock(&func##_lock);                       \
+  } while (0)
+#else  // !(defined(WEBP_USE_THREAD) && !defined(_WIN32))
+#define WEBP_DSP_INIT(func)                               \
+  do {                                                    \
+    static volatile VP8CPUInfo func##_last_cpuinfo_used = \
+        (VP8CPUInfo)&func##_last_cpuinfo_used;            \
+    if (func##_last_cpuinfo_used == VP8GetCPUInfo) break; \
+    func();                                               \
+    func##_last_cpuinfo_used = VP8GetCPUInfo;             \
+  } while (0)
+#endif  // defined(WEBP_USE_THREAD) && !defined(_WIN32)
+
+// Defines an Init + helper function that control multiple initialization of
+// function pointers / tables.
+/* Usage:
+   WEBP_DSP_INIT_FUNC(InitFunc) {
+     ...function body
+   }
+*/
+#define WEBP_DSP_INIT_FUNC(name)                                            \
+  static WEBP_TSAN_IGNORE_FUNCTION void name##_body(void);                  \
+  WEBP_TSAN_IGNORE_FUNCTION void name(void) { WEBP_DSP_INIT(name##_body); } \
+  static WEBP_TSAN_IGNORE_FUNCTION void name##_body(void)
+
+#define WEBP_UBSAN_IGNORE_UNDEF
+#define WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
+#if defined(__clang__) && defined(__has_attribute)
+#if __has_attribute(no_sanitize)
+// This macro prevents the undefined behavior sanitizer from reporting
+// failures. This is only meant to silence unaligned loads on platforms that
+// are known to support them.
+#undef WEBP_UBSAN_IGNORE_UNDEF
+#define WEBP_UBSAN_IGNORE_UNDEF __attribute__((no_sanitize("undefined")))
+
+// This macro prevents the undefined behavior sanitizer from reporting
+// failures related to unsigned integer overflows. This is only meant to
+// silence cases where this well defined behavior is expected.
+#undef WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
+#define WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW \
+  __attribute__((no_sanitize("unsigned-integer-overflow")))
+#endif
+#endif
+
+// If 'ptr' is NULL, returns NULL. Otherwise returns 'ptr + off'.
+// Prevents undefined behavior sanitizer nullptr-with-nonzero-offset warning.
+#if !defined(WEBP_OFFSET_PTR)
+#define WEBP_OFFSET_PTR(ptr, off) (((ptr) == NULL) ? NULL : ((ptr) + (off)))
+#endif
+
+// Regularize the definition of WEBP_SWAP_16BIT_CSP (backward compatibility)
+#if !defined(WEBP_SWAP_16BIT_CSP)
+#define WEBP_SWAP_16BIT_CSP 0
+#endif
+
+// some endian fix (e.g.: mips-gcc doesn't define __BIG_ENDIAN__)
+#if !defined(WORDS_BIGENDIAN) &&                   \
+    (defined(__BIG_ENDIAN__) || defined(_M_PPC) || \
+     (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)))
+#define WORDS_BIGENDIAN
+#endif
+
+typedef enum {
+  kSSE2,
+  kSSE3,
+  kSlowSSSE3,  // special feature for slow SSSE3 architectures
+  kSSE4_1,
+  kAVX,
+  kAVX2,
+  kNEON,
+  kMIPS32,
+  kMIPSdspR2,
+  kMSA
+} CPUFeature;
+
+// returns true if the CPU supports the feature.
+typedef int (*VP8CPUInfo)(CPUFeature feature);
+
+#endif  // WEBP_DSP_CPU_H_

3rdparty/libwebp/src/dsp/dec.c

@@ -734,6 +734,7 @@ VP8SimpleFilterFunc VP8SimpleHFilter16i;
 void (*VP8DitherCombine8x8)(const uint8_t* dither, uint8_t* dst,
                             int dst_stride);
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8DspInitSSE2(void);
 extern void VP8DspInitSSE41(void);
 extern void VP8DspInitNEON(void);
@@ -807,10 +808,10 @@ WEBP_DSP_INIT_FUNC(VP8DspInit) {
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8DspInitSSE2();
-#if defined(WEBP_USE_SSE41)
+#if defined(WEBP_HAVE_SSE41)
       if (VP8GetCPUInfo(kSSE4_1)) {
         VP8DspInitSSE41();
       }
@@ -834,7 +835,7 @@ WEBP_DSP_INIT_FUNC(VP8DspInit) {
 #endif
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     VP8DspInitNEON();

3rdparty/libwebp/src/dsp/dec_neon.c

@@ -1428,7 +1428,7 @@ static WEBP_INLINE void DC8_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x8_t A = vld1_u8(dst - BPS);  // top row
-#if defined(__aarch64__)
+#if WEBP_AARCH64
     const uint16_t p2 = vaddlv_u8(A);
     sum_top = vdupq_n_u16(p2);
 #else
@@ -1511,7 +1511,7 @@ static WEBP_INLINE void DC16_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x16_t A = vld1q_u8(dst - BPS);  // top row
-#if defined(__aarch64__)
+#if WEBP_AARCH64
     const uint16_t p3 = vaddlvq_u8(A);
     sum_top = vdupq_n_u16(p3);
 #else

3rdparty/libwebp/src/dsp/dec_sse2.c

@@ -158,10 +158,10 @@ static void Transform_SSE2(const int16_t* in, uint8_t* dst, int do_two) {
       dst3 = _mm_loadl_epi64((__m128i*)(dst + 3 * BPS));
     } else {
       // Load four bytes/pixels per line.
-      dst0 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 0 * BPS));
-      dst1 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 1 * BPS));
-      dst2 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 2 * BPS));
-      dst3 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 3 * BPS));
+      dst0 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 0 * BPS));
+      dst1 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 1 * BPS));
+      dst2 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 2 * BPS));
+      dst3 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 3 * BPS));
     }
     // Convert to 16b.
     dst0 = _mm_unpacklo_epi8(dst0, zero);
@@ -187,10 +187,10 @@ static void Transform_SSE2(const int16_t* in, uint8_t* dst, int do_two) {
       _mm_storel_epi64((__m128i*)(dst + 3 * BPS), dst3);
     } else {
       // Store four bytes/pixels per line.
-      WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(dst0));
-      WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(dst1));
-      WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
-      WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
+      WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(dst0));
+      WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(dst1));
+      WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
+      WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
     }
   }
 }
@@ -213,10 +213,10 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
   const __m128i m3 = _mm_subs_epi16(B, d4);
   const __m128i zero = _mm_setzero_si128();
   // Load the source pixels.
-  __m128i dst0 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 0 * BPS));
-  __m128i dst1 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 1 * BPS));
-  __m128i dst2 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 2 * BPS));
-  __m128i dst3 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 3 * BPS));
+  __m128i dst0 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 0 * BPS));
+  __m128i dst1 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 1 * BPS));
+  __m128i dst2 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 2 * BPS));
+  __m128i dst3 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 3 * BPS));
   // Convert to 16b.
   dst0 = _mm_unpacklo_epi8(dst0, zero);
   dst1 = _mm_unpacklo_epi8(dst1, zero);
@@ -233,10 +233,10 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
   dst2 = _mm_packus_epi16(dst2, dst2);
   dst3 = _mm_packus_epi16(dst3, dst3);
   // Store the results.
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(dst0));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(dst1));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(dst0));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(dst1));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
 }
 #undef MUL
 #endif   // USE_TRANSFORM_AC3
@@ -477,11 +477,11 @@ static WEBP_INLINE void Load8x4_SSE2(const uint8_t* const b, int stride,
   // A0 = 63 62 61 60 23 22 21 20 43 42 41 40 03 02 01 00
   // A1 = 73 72 71 70 33 32 31 30 53 52 51 50 13 12 11 10
   const __m128i A0 = _mm_set_epi32(
-      WebPMemToUint32(&b[6 * stride]), WebPMemToUint32(&b[2 * stride]),
-      WebPMemToUint32(&b[4 * stride]), WebPMemToUint32(&b[0 * stride]));
+      WebPMemToInt32(&b[6 * stride]), WebPMemToInt32(&b[2 * stride]),
+      WebPMemToInt32(&b[4 * stride]), WebPMemToInt32(&b[0 * stride]));
   const __m128i A1 = _mm_set_epi32(
-      WebPMemToUint32(&b[7 * stride]), WebPMemToUint32(&b[3 * stride]),
-      WebPMemToUint32(&b[5 * stride]), WebPMemToUint32(&b[1 * stride]));
+      WebPMemToInt32(&b[7 * stride]), WebPMemToInt32(&b[3 * stride]),
+      WebPMemToInt32(&b[5 * stride]), WebPMemToInt32(&b[1 * stride]));
 
   // B0 = 53 43 52 42 51 41 50 40 13 03 12 02 11 01 10 00
   // B1 = 73 63 72 62 71 61 70 60 33 23 32 22 31 21 30 20
@@ -540,7 +540,7 @@ static WEBP_INLINE void Store4x4_SSE2(__m128i* const x,
                                       uint8_t* dst, int stride) {
   int i;
   for (i = 0; i < 4; ++i, dst += stride) {
-    WebPUint32ToMem(dst, _mm_cvtsi128_si32(*x));
+    WebPInt32ToMem(dst, _mm_cvtsi128_si32(*x));
     *x = _mm_srli_si128(*x, 4);
   }
 }
@@ -908,10 +908,10 @@ static void VE4_SSE2(uint8_t* dst) {    // vertical
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(ABCDEFGH, CDEFGH00), one);
   const __m128i b = _mm_subs_epu8(a, lsb);
   const __m128i avg = _mm_avg_epu8(b, BCDEFGH0);
-  const uint32_t vals = _mm_cvtsi128_si32(avg);
+  const int vals = _mm_cvtsi128_si32(avg);
   int i;
   for (i = 0; i < 4; ++i) {
-    WebPUint32ToMem(dst + i * BPS, vals);
+    WebPInt32ToMem(dst + i * BPS, vals);
   }
 }
 
@@ -925,10 +925,10 @@ static void LD4_SSE2(uint8_t* dst) {   // Down-Left
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(ABCDEFGH, CDEFGHH0), one);
   const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
   const __m128i abcdefg = _mm_avg_epu8(avg2, BCDEFGH0);
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               abcdefg    ));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               abcdefg    ));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
 }
 
 static void VR4_SSE2(uint8_t* dst) {   // Vertical-Right
@@ -946,10 +946,10 @@ static void VR4_SSE2(uint8_t* dst) {   // Vertical-Right
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
   const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
   const __m128i efgh = _mm_avg_epu8(avg2, XABCD);
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               abcd    ));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(               efgh    ));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(abcd, 1)));
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(efgh, 1)));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               abcd    ));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(               efgh    ));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(abcd, 1)));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(efgh, 1)));
 
   // these two are hard to implement in SSE2, so we keep the C-version:
   DST(0, 2) = AVG3(J, I, X);
@@ -970,11 +970,12 @@ static void VL4_SSE2(uint8_t* dst) {   // Vertical-Left
   const __m128i abbc = _mm_or_si128(ab, bc);
   const __m128i lsb2 = _mm_and_si128(abbc, lsb1);
   const __m128i avg4 = _mm_subs_epu8(avg3, lsb2);
-  const uint32_t extra_out = _mm_cvtsi128_si32(_mm_srli_si128(avg4, 4));
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               avg1    ));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(               avg4    ));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg1, 1)));
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg4, 1)));
+  const uint32_t extra_out =
+      (uint32_t)_mm_cvtsi128_si32(_mm_srli_si128(avg4, 4));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               avg1    ));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(               avg4    ));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg1, 1)));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg4, 1)));
 
   // these two are hard to get and irregular
   DST(3, 2) = (extra_out >> 0) & 0xff;
@@ -990,7 +991,7 @@ static void RD4_SSE2(uint8_t* dst) {   // Down-right
   const uint32_t K = dst[-1 + 2 * BPS];
   const uint32_t L = dst[-1 + 3 * BPS];
   const __m128i LKJI_____ =
-      _mm_cvtsi32_si128(L | (K << 8) | (J << 16) | (I << 24));
+      _mm_cvtsi32_si128((int)(L | (K << 8) | (J << 16) | (I << 24)));
   const __m128i LKJIXABCD = _mm_or_si128(LKJI_____, ____XABCD);
   const __m128i KJIXABCD_ = _mm_srli_si128(LKJIXABCD, 1);
   const __m128i JIXABCD__ = _mm_srli_si128(LKJIXABCD, 2);
@@ -998,10 +999,10 @@ static void RD4_SSE2(uint8_t* dst) {   // Down-right
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(JIXABCD__, LKJIXABCD), one);
   const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
   const __m128i abcdefg = _mm_avg_epu8(avg2, KJIXABCD_);
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(               abcdefg    ));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(               abcdefg    ));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
 }
 
 #undef DST
@@ -1015,13 +1016,13 @@ static WEBP_INLINE void TrueMotion_SSE2(uint8_t* dst, int size) {
   const __m128i zero = _mm_setzero_si128();
   int y;
   if (size == 4) {
-    const __m128i top_values = _mm_cvtsi32_si128(WebPMemToUint32(top));
+    const __m128i top_values = _mm_cvtsi32_si128(WebPMemToInt32(top));
     const __m128i top_base = _mm_unpacklo_epi8(top_values, zero);
     for (y = 0; y < 4; ++y, dst += BPS) {
       const int val = dst[-1] - top[-1];
       const __m128i base = _mm_set1_epi16(val);
       const __m128i out = _mm_packus_epi16(_mm_add_epi16(base, top_base), zero);
-      WebPUint32ToMem(dst, _mm_cvtsi128_si32(out));
+      WebPInt32ToMem(dst, _mm_cvtsi128_si32(out));
     }
   } else if (size == 8) {
     const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
@@ -1062,7 +1063,7 @@ static void VE16_SSE2(uint8_t* dst) {
 static void HE16_SSE2(uint8_t* dst) {     // horizontal
   int j;
   for (j = 16; j > 0; --j) {
-    const __m128i values = _mm_set1_epi8(dst[-1]);
+    const __m128i values = _mm_set1_epi8((char)dst[-1]);
     _mm_storeu_si128((__m128i*)dst, values);
     dst += BPS;
   }
@@ -1070,7 +1071,7 @@ static void HE16_SSE2(uint8_t* dst) {     // horizontal
 
 static WEBP_INLINE void Put16_SSE2(uint8_t v, uint8_t* dst) {
   int j;
-  const __m128i values = _mm_set1_epi8(v);
+  const __m128i values = _mm_set1_epi8((char)v);
   for (j = 0; j < 16; ++j) {
     _mm_storeu_si128((__m128i*)(dst + j * BPS), values);
   }
@@ -1130,7 +1131,7 @@ static void VE8uv_SSE2(uint8_t* dst) {    // vertical
 // helper for chroma-DC predictions
 static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) {
   int j;
-  const __m128i values = _mm_set1_epi8(v);
+  const __m128i values = _mm_set1_epi8((char)v);
   for (j = 0; j < 8; ++j) {
     _mm_storel_epi64((__m128i*)(dst + j * BPS), values);
   }

3rdparty/libwebp/src/dsp/dec_sse41.c

@@ -23,7 +23,7 @@ static void HE16_SSE41(uint8_t* dst) {     // horizontal
   int j;
   const __m128i kShuffle3 = _mm_set1_epi8(3);
   for (j = 16; j > 0; --j) {
-    const __m128i in = _mm_cvtsi32_si128(WebPMemToUint32(dst - 4));
+    const __m128i in = _mm_cvtsi32_si128(WebPMemToInt32(dst - 4));
     const __m128i values = _mm_shuffle_epi8(in, kShuffle3);
     _mm_storeu_si128((__m128i*)dst, values);
     dst += BPS;

3rdparty/libwebp/src/dsp/dsp.h

@@ -18,6 +18,7 @@
 #include "src/webp/config.h"
 #endif
 
+#include "src/dsp/cpu.h"
 #include "src/webp/types.h"
 
 #ifdef __cplusplus
@@ -27,205 +28,22 @@ extern "C" {
 #define BPS 32   // this is the common stride for enc/dec
 
 //------------------------------------------------------------------------------
-// CPU detection
-
+// WEBP_RESTRICT
+
+// Declares a pointer with the restrict type qualifier if available.
+// This allows code to hint to the compiler that only this pointer references a
+// particular object or memory region within the scope of the block in which it
+// is declared. This may allow for improved optimizations due to the lack of
+// pointer aliasing. See also:
+// https://en.cppreference.com/w/c/language/restrict
 #if defined(__GNUC__)
-# define LOCAL_GCC_VERSION ((__GNUC__ << 8) | __GNUC_MINOR__)
-# define LOCAL_GCC_PREREQ(maj, min) \
-    (LOCAL_GCC_VERSION >= (((maj) << 8) | (min)))
-#else
-# define LOCAL_GCC_VERSION 0
-# define LOCAL_GCC_PREREQ(maj, min) 0
-#endif
-
-#if defined(__clang__)
-# define LOCAL_CLANG_VERSION ((__clang_major__ << 8) | __clang_minor__)
-# define LOCAL_CLANG_PREREQ(maj, min) \
-    (LOCAL_CLANG_VERSION >= (((maj) << 8) | (min)))
+#define WEBP_RESTRICT __restrict__
+#elif defined(_MSC_VER)
+#define WEBP_RESTRICT __restrict
 #else
-# define LOCAL_CLANG_VERSION 0
-# define LOCAL_CLANG_PREREQ(maj, min) 0
-#endif
-
-#ifndef __has_builtin
-# define __has_builtin(x) 0
-#endif
-
-#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
-#endif
-
-#if defined(_MSC_VER) && _MSC_VER >= 1500 && \
-    (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.
-// Files containing intrinsics will need to be built targeting the instruction
-// set so should succeed on one of the earlier tests.
-#if defined(__SSE2__) || defined(WEBP_MSC_SSE2) || defined(WEBP_HAVE_SSE2)
-#define WEBP_USE_SSE2
+#define WEBP_RESTRICT
 #endif
 
-#if defined(__SSE4_1__) || defined(WEBP_MSC_SSE41) || defined(WEBP_HAVE_SSE41)
-#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__) || \
-     defined(__aarch64__) || defined(WEBP_HAVE_NEON)) && \
-    !defined(__native_client__)
-#define WEBP_USE_NEON
-#endif
-
-#if !defined(WEBP_USE_NEON) && defined(__ANDROID__) && \
-    defined(__ARM_ARCH_7A__) && defined(HAVE_CPU_FEATURES_H)
-#define WEBP_ANDROID_NEON  // Android targets that may have NEON
-#define WEBP_USE_NEON
-#endif
-
-#if defined(_MSC_VER) && _MSC_VER >= 1700 && defined(_M_ARM)
-#define WEBP_USE_NEON
-#define WEBP_USE_INTRINSICS
-#endif
-
-#if defined(__mips__) && !defined(__mips64) && \
-    defined(__mips_isa_rev) && (__mips_isa_rev >= 1) && (__mips_isa_rev < 6)
-#define WEBP_USE_MIPS32
-#if (__mips_isa_rev >= 2)
-#define WEBP_USE_MIPS32_R2
-#if defined(__mips_dspr2) || (defined(__mips_dsp_rev) && __mips_dsp_rev >= 2)
-#define WEBP_USE_MIPS_DSP_R2
-#endif
-#endif
-#endif
-
-#if defined(__mips_msa) && defined(__mips_isa_rev) && (__mips_isa_rev >= 5)
-#define WEBP_USE_MSA
-#endif
-
-#ifndef WEBP_DSP_OMIT_C_CODE
-#define WEBP_DSP_OMIT_C_CODE 1
-#endif
-
-#if (defined(__aarch64__) || defined(__ARM_NEON__)) && WEBP_DSP_OMIT_C_CODE
-#define WEBP_NEON_OMIT_C_CODE 1
-#else
-#define WEBP_NEON_OMIT_C_CODE 0
-#endif
-
-#if !(LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
-#define WEBP_NEON_WORK_AROUND_GCC 1
-#else
-#define WEBP_NEON_WORK_AROUND_GCC 0
-#endif
-
-// This macro prevents thread_sanitizer from reporting known concurrent writes.
-#define WEBP_TSAN_IGNORE_FUNCTION
-#if defined(__has_feature)
-#if __has_feature(thread_sanitizer)
-#undef WEBP_TSAN_IGNORE_FUNCTION
-#define WEBP_TSAN_IGNORE_FUNCTION __attribute__((no_sanitize_thread))
-#endif
-#endif
-
-#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
-#include <pthread.h>  // NOLINT
-
-#define WEBP_DSP_INIT(func) do {                                    \
-  static volatile VP8CPUInfo func ## _last_cpuinfo_used =           \
-      (VP8CPUInfo)&func ## _last_cpuinfo_used;                      \
-  static pthread_mutex_t func ## _lock = PTHREAD_MUTEX_INITIALIZER; \
-  if (pthread_mutex_lock(&func ## _lock)) break;                    \
-  if (func ## _last_cpuinfo_used != VP8GetCPUInfo) func();          \
-  func ## _last_cpuinfo_used = VP8GetCPUInfo;                       \
-  (void)pthread_mutex_unlock(&func ## _lock);                       \
-} while (0)
-#else  // !(defined(WEBP_USE_THREAD) && !defined(_WIN32))
-#define WEBP_DSP_INIT(func) do {                                    \
-  static volatile VP8CPUInfo func ## _last_cpuinfo_used =           \
-      (VP8CPUInfo)&func ## _last_cpuinfo_used;                      \
-  if (func ## _last_cpuinfo_used == VP8GetCPUInfo) break;           \
-  func();                                                           \
-  func ## _last_cpuinfo_used = VP8GetCPUInfo;                       \
-} while (0)
-#endif  // defined(WEBP_USE_THREAD) && !defined(_WIN32)
-
-// Defines an Init + helper function that control multiple initialization of
-// function pointers / tables.
-/* Usage:
-   WEBP_DSP_INIT_FUNC(InitFunc) {
-     ...function body
-   }
-*/
-#define WEBP_DSP_INIT_FUNC(name)                             \
-  static WEBP_TSAN_IGNORE_FUNCTION void name ## _body(void); \
-  WEBP_TSAN_IGNORE_FUNCTION void name(void) {                \
-    WEBP_DSP_INIT(name ## _body);                            \
-  }                                                          \
-  static WEBP_TSAN_IGNORE_FUNCTION void name ## _body(void)
-
-#define WEBP_UBSAN_IGNORE_UNDEF
-#define WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
-#if defined(__clang__) && defined(__has_attribute)
-#if __has_attribute(no_sanitize)
-// This macro prevents the undefined behavior sanitizer from reporting
-// failures. This is only meant to silence unaligned loads on platforms that
-// are known to support them.
-#undef WEBP_UBSAN_IGNORE_UNDEF
-#define WEBP_UBSAN_IGNORE_UNDEF \
-  __attribute__((no_sanitize("undefined")))
-
-// This macro prevents the undefined behavior sanitizer from reporting
-// failures related to unsigned integer overflows. This is only meant to
-// silence cases where this well defined behavior is expected.
-#undef WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
-#define WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW \
-  __attribute__((no_sanitize("unsigned-integer-overflow")))
-#endif
-#endif
-
-// If 'ptr' is NULL, returns NULL. Otherwise returns 'ptr + off'.
-// Prevents undefined behavior sanitizer nullptr-with-nonzero-offset warning.
-#if !defined(WEBP_OFFSET_PTR)
-#define WEBP_OFFSET_PTR(ptr, off) (((ptr) == NULL) ? NULL : ((ptr) + (off)))
-#endif
-
-// Regularize the definition of WEBP_SWAP_16BIT_CSP (backward compatibility)
-#if !defined(WEBP_SWAP_16BIT_CSP)
-#define WEBP_SWAP_16BIT_CSP 0
-#endif
-
-// some endian fix (e.g.: mips-gcc doesn't define __BIG_ENDIAN__)
-#if !defined(WORDS_BIGENDIAN) && \
-    (defined(__BIG_ENDIAN__) || defined(_M_PPC) || \
-     (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)))
-#define WORDS_BIGENDIAN
-#endif
-
-typedef enum {
-  kSSE2,
-  kSSE3,
-  kSlowSSSE3,  // special feature for slow SSSE3 architectures
-  kSSE4_1,
-  kAVX,
-  kAVX2,
-  kNEON,
-  kMIPS32,
-  kMIPSdspR2,
-  kMSA
-} CPUFeature;
-// returns true if the CPU supports the feature.
-typedef int (*VP8CPUInfo)(CPUFeature feature);
-WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
 
 //------------------------------------------------------------------------------
 // Init stub generator
@@ -514,15 +332,6 @@ extern void WebPConvertARGBToUV_C(const uint32_t* argb, uint8_t* u, uint8_t* v,
 extern void WebPConvertRGBA32ToUV_C(const uint16_t* rgb,
                                     uint8_t* u, uint8_t* v, int width);
 
-// utilities for accurate RGB->YUV conversion
-extern uint64_t (*WebPSharpYUVUpdateY)(const uint16_t* src, const uint16_t* ref,
-                                       uint16_t* dst, int len);
-extern void (*WebPSharpYUVUpdateRGB)(const int16_t* src, const int16_t* ref,
-                                     int16_t* dst, int len);
-extern void (*WebPSharpYUVFilterRow)(const int16_t* A, const int16_t* B,
-                                     int len,
-                                     const uint16_t* best_y, uint16_t* out);
-
 // Must be called before using the above.
 void WebPInitConvertARGBToYUV(void);
 
@@ -578,26 +387,29 @@ extern void (*WebPApplyAlphaMultiply4444)(
 
 // Dispatch the values from alpha[] plane to the ARGB destination 'dst'.
 // Returns true if alpha[] plane has non-trivial values different from 0xff.
-extern int (*WebPDispatchAlpha)(const uint8_t* alpha, int alpha_stride,
-                                int width, int height,
-                                uint8_t* dst, int dst_stride);
+extern int (*WebPDispatchAlpha)(const uint8_t* WEBP_RESTRICT alpha,
+                                int alpha_stride, int width, int height,
+                                uint8_t* WEBP_RESTRICT dst, int dst_stride);
 
 // Transfer packed 8b alpha[] values to green channel in dst[], zero'ing the
 // A/R/B values. 'dst_stride' is the stride for dst[] in uint32_t units.
-extern void (*WebPDispatchAlphaToGreen)(const uint8_t* alpha, int alpha_stride,
-                                        int width, int height,
-                                        uint32_t* dst, int dst_stride);
+extern void (*WebPDispatchAlphaToGreen)(const uint8_t* WEBP_RESTRICT alpha,
+                                        int alpha_stride, int width, int height,
+                                        uint32_t* WEBP_RESTRICT dst,
+                                        int dst_stride);
 
 // Extract the alpha values from 32b values in argb[] and pack them into alpha[]
 // (this is the opposite of WebPDispatchAlpha).
 // Returns true if there's only trivial 0xff alpha values.
-extern int (*WebPExtractAlpha)(const uint8_t* argb, int argb_stride,
-                               int width, int height,
-                               uint8_t* alpha, int alpha_stride);
+extern int (*WebPExtractAlpha)(const uint8_t* WEBP_RESTRICT argb,
+                               int argb_stride, int width, int height,
+                               uint8_t* WEBP_RESTRICT alpha,
+                               int alpha_stride);
 
 // Extract the green values from 32b values in argb[] and pack them into alpha[]
 // (this is the opposite of WebPDispatchAlphaToGreen).
-extern void (*WebPExtractGreen)(const uint32_t* argb, uint8_t* alpha, int size);
+extern void (*WebPExtractGreen)(const uint32_t* WEBP_RESTRICT argb,
+                                uint8_t* WEBP_RESTRICT alpha, int size);
 
 // Pre-Multiply operation transforms x into x * A / 255  (where x=Y,R,G or B).
 // Un-Multiply operation transforms x into x * 255 / A.
@@ -610,29 +422,35 @@ void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows,
                       int inverse);
 
 // Same for a row of single values, with side alpha values.
-extern void (*WebPMultRow)(uint8_t* const ptr, const uint8_t* const alpha,
+extern void (*WebPMultRow)(uint8_t* WEBP_RESTRICT const ptr,
+                           const uint8_t* WEBP_RESTRICT const alpha,
                            int width, int inverse);
 
 // Same a WebPMultRow(), but for several 'num_rows' rows.
-void WebPMultRows(uint8_t* ptr, int stride,
-                  const uint8_t* alpha, int alpha_stride,
+void WebPMultRows(uint8_t* WEBP_RESTRICT ptr, int stride,
+                  const uint8_t* WEBP_RESTRICT alpha, int alpha_stride,
                   int width, int num_rows, int inverse);
 
 // Plain-C versions, used as fallback by some implementations.
-void WebPMultRow_C(uint8_t* const ptr, const uint8_t* const alpha,
+void WebPMultRow_C(uint8_t* WEBP_RESTRICT const ptr,
+                   const uint8_t* WEBP_RESTRICT const alpha,
                    int width, int inverse);
 void WebPMultARGBRow_C(uint32_t* const ptr, int width, int inverse);
 
 #ifdef WORDS_BIGENDIAN
 // ARGB packing function: a/r/g/b input is rgba or bgra order.
-extern void (*WebPPackARGB)(const uint8_t* a, const uint8_t* r,
-                            const uint8_t* g, const uint8_t* b, int len,
-                            uint32_t* out);
+extern void (*WebPPackARGB)(const uint8_t* WEBP_RESTRICT a,
+                            const uint8_t* WEBP_RESTRICT r,
+                            const uint8_t* WEBP_RESTRICT g,
+                            const uint8_t* WEBP_RESTRICT b,
+                            int len, uint32_t* WEBP_RESTRICT out);
 #endif
 
 // RGB packing function. 'step' can be 3 or 4. r/g/b input is rgb or bgr order.
-extern void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
-                           int len, int step, uint32_t* out);
+extern void (*WebPPackRGB)(const uint8_t* WEBP_RESTRICT r,
+                           const uint8_t* WEBP_RESTRICT g,
+                           const uint8_t* WEBP_RESTRICT b,
+                           int len, int step, uint32_t* WEBP_RESTRICT out);
 
 // This function returns true if src[i] contains a value different from 0xff.
 extern int (*WebPHasAlpha8b)(const uint8_t* src, int length);

3rdparty/libwebp/src/dsp/enc.c

@@ -732,6 +732,7 @@ VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT;
 VP8BlockCopy VP8Copy4x4;
 VP8BlockCopy VP8Copy16x8;
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8EncDspInitSSE2(void);
 extern void VP8EncDspInitSSE41(void);
 extern void VP8EncDspInitNEON(void);
@@ -773,10 +774,10 @@ WEBP_DSP_INIT_FUNC(VP8EncDspInit) {
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8EncDspInitSSE2();
-#if defined(WEBP_USE_SSE41)
+#if defined(WEBP_HAVE_SSE41)
       if (VP8GetCPUInfo(kSSE4_1)) {
         VP8EncDspInitSSE41();
       }
@@ -800,7 +801,7 @@ WEBP_DSP_INIT_FUNC(VP8EncDspInit) {
 #endif
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     VP8EncDspInitNEON();

3rdparty/libwebp/src/dsp/enc_neon.c

@@ -9,7 +9,7 @@
 //
 // ARM NEON version of speed-critical encoding functions.
 //
-// adapted from libvpx (http://www.webmproject.org/code/)
+// adapted from libvpx (https://www.webmproject.org/code/)
 
 #include "src/dsp/dsp.h"
 
@@ -764,9 +764,14 @@ static WEBP_INLINE void AccumulateSSE16_NEON(const uint8_t* const a,
 
 // Horizontal sum of all four uint32_t values in 'sum'.
 static int SumToInt_NEON(uint32x4_t sum) {
+#if WEBP_AARCH64
+  return (int)vaddvq_u32(sum);
+#else
   const uint64x2_t sum2 = vpaddlq_u32(sum);
-  const uint64_t sum3 = vgetq_lane_u64(sum2, 0) + vgetq_lane_u64(sum2, 1);
-  return (int)sum3;
+  const uint32x2_t sum3 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(sum2)),
+                                   vreinterpret_u32_u64(vget_high_u64(sum2)));
+  return (int)vget_lane_u32(sum3, 0);
+#endif
 }
 
 static int SSE16x16_NEON(const uint8_t* a, const uint8_t* b) {
@@ -860,7 +865,7 @@ static int QuantizeBlock_NEON(int16_t in[16], int16_t out[16],
   uint8x8x4_t shuffles;
   // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
   // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
   uint8x16x2_t all_out;
   INIT_VECTOR2(all_out, vreinterpretq_u8_s16(out0), vreinterpretq_u8_s16(out1));

3rdparty/libwebp/src/dsp/enc_sse2.c

@@ -25,9 +25,160 @@
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
-// Does one or two inverse transforms.
-static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
-                            int do_two) {
+// Does one inverse transform.
+static void ITransform_One_SSE2(const uint8_t* ref, const int16_t* in,
+                                uint8_t* dst) {
+  // This implementation makes use of 16-bit fixed point versions of two
+  // multiply constants:
+  //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
+  //    K2 = sqrt(2) * sin (pi/8) ~= 35468 / 2^16
+  //
+  // To be able to use signed 16-bit integers, we use the following trick to
+  // have constants within range:
+  // - Associated constants are obtained by subtracting the 16-bit fixed point
+  //   version of one:
+  //      k = K - (1 << 16)  =>  K = k + (1 << 16)
+  //      K1 = 85267  =>  k1 =  20091
+  //      K2 = 35468  =>  k2 = -30068
+  // - The multiplication of a variable by a constant become the sum of the
+  //   variable and the multiplication of that variable by the associated
+  //   constant:
+  //      (x * K) >> 16 = (x * (k + (1 << 16))) >> 16 = ((x * k ) >> 16) + x
+  const __m128i k1k2 = _mm_set_epi16(-30068, -30068, -30068, -30068,
+                                     20091, 20091, 20091, 20091);
+  const __m128i k2k1 = _mm_set_epi16(20091, 20091, 20091, 20091,
+                                     -30068, -30068, -30068, -30068);
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i zero_four = _mm_set_epi16(0, 0, 0, 0, 4, 4, 4, 4);
+  __m128i T01, T23;
+
+  // Load and concatenate the transform coefficients.
+  const __m128i in01 = _mm_loadu_si128((const __m128i*)&in[0]);
+  const __m128i in23 = _mm_loadu_si128((const __m128i*)&in[8]);
+  // a00 a10 a20 a30   a01 a11 a21 a31
+  // a02 a12 a22 a32   a03 a13 a23 a33
+
+  // Vertical pass and subsequent transpose.
+  {
+    const __m128i in1 = _mm_unpackhi_epi64(in01, in01);
+    const __m128i in3 = _mm_unpackhi_epi64(in23, in23);
+
+    // First pass, c and d calculations are longer because of the "trick"
+    // multiplications.
+    // c = MUL(in1, K2) - MUL(in3, K1) = MUL(in1, k2) - MUL(in3, k1) + in1 - in3
+    // d = MUL(in1, K1) + MUL(in3, K2) = MUL(in1, k1) + MUL(in3, k2) + in1 + in3
+    const __m128i a_d3 = _mm_add_epi16(in01, in23);
+    const __m128i b_c3 = _mm_sub_epi16(in01, in23);
+    const __m128i c1d1 = _mm_mulhi_epi16(in1, k2k1);
+    const __m128i c2d2 = _mm_mulhi_epi16(in3, k1k2);
+    const __m128i c3 = _mm_unpackhi_epi64(b_c3, b_c3);
+    const __m128i c4 = _mm_sub_epi16(c1d1, c2d2);
+    const __m128i c = _mm_add_epi16(c3, c4);
+    const __m128i d4u = _mm_add_epi16(c1d1, c2d2);
+    const __m128i du = _mm_add_epi16(a_d3, d4u);
+    const __m128i d = _mm_unpackhi_epi64(du, du);
+
+    // Second pass.
+    const __m128i comb_ab = _mm_unpacklo_epi64(a_d3, b_c3);
+    const __m128i comb_dc = _mm_unpacklo_epi64(d, c);
+
+    const __m128i tmp01 = _mm_add_epi16(comb_ab, comb_dc);
+    const __m128i tmp32 = _mm_sub_epi16(comb_ab, comb_dc);
+    const __m128i tmp23 = _mm_shuffle_epi32(tmp32, _MM_SHUFFLE(1, 0, 3, 2));
+
+    const __m128i transpose_0 = _mm_unpacklo_epi16(tmp01, tmp23);
+    const __m128i transpose_1 = _mm_unpackhi_epi16(tmp01, tmp23);
+    // a00 a20 a01 a21   a02 a22 a03 a23
+    // a10 a30 a11 a31   a12 a32 a13 a33
+
+    T01 = _mm_unpacklo_epi16(transpose_0, transpose_1);
+    T23 = _mm_unpackhi_epi16(transpose_0, transpose_1);
+    // a00 a10 a20 a30   a01 a11 a21 a31
+    // a02 a12 a22 a32   a03 a13 a23 a33
+  }
+
+  // Horizontal pass and subsequent transpose.
+  {
+    const __m128i T1 = _mm_unpackhi_epi64(T01, T01);
+    const __m128i T3 = _mm_unpackhi_epi64(T23, T23);
+
+    // First pass, c and d calculations are longer because of the "trick"
+    // multiplications.
+    const __m128i dc = _mm_add_epi16(T01, zero_four);
+
+    // c = MUL(T1, K2) - MUL(T3, K1) = MUL(T1, k2) - MUL(T3, k1) + T1 - T3
+    // d = MUL(T1, K1) + MUL(T3, K2) = MUL(T1, k1) + MUL(T3, k2) + T1 + T3
+    const __m128i a_d3 = _mm_add_epi16(dc, T23);
+    const __m128i b_c3 = _mm_sub_epi16(dc, T23);
+    const __m128i c1d1 = _mm_mulhi_epi16(T1, k2k1);
+    const __m128i c2d2 = _mm_mulhi_epi16(T3, k1k2);
+    const __m128i c3 = _mm_unpackhi_epi64(b_c3, b_c3);
+    const __m128i c4 = _mm_sub_epi16(c1d1, c2d2);
+    const __m128i c = _mm_add_epi16(c3, c4);
+    const __m128i d4u = _mm_add_epi16(c1d1, c2d2);
+    const __m128i du = _mm_add_epi16(a_d3, d4u);
+    const __m128i d = _mm_unpackhi_epi64(du, du);
+
+    // Second pass.
+    const __m128i comb_ab = _mm_unpacklo_epi64(a_d3, b_c3);
+    const __m128i comb_dc = _mm_unpacklo_epi64(d, c);
+
+    const __m128i tmp01 = _mm_add_epi16(comb_ab, comb_dc);
+    const __m128i tmp32 = _mm_sub_epi16(comb_ab, comb_dc);
+    const __m128i tmp23 = _mm_shuffle_epi32(tmp32, _MM_SHUFFLE(1, 0, 3, 2));
+
+    const __m128i shifted01 = _mm_srai_epi16(tmp01, 3);
+    const __m128i shifted23 = _mm_srai_epi16(tmp23, 3);
+    // a00 a01 a02 a03   a10 a11 a12 a13
+    // a20 a21 a22 a23   a30 a31 a32 a33
+
+    const __m128i transpose_0 = _mm_unpacklo_epi16(shifted01, shifted23);
+    const __m128i transpose_1 = _mm_unpackhi_epi16(shifted01, shifted23);
+    // a00 a20 a01 a21   a02 a22 a03 a23
+    // a10 a30 a11 a31   a12 a32 a13 a33
+
+    T01 = _mm_unpacklo_epi16(transpose_0, transpose_1);
+    T23 = _mm_unpackhi_epi16(transpose_0, transpose_1);
+    // a00 a10 a20 a30   a01 a11 a21 a31
+    // a02 a12 a22 a32   a03 a13 a23 a33
+  }
+
+  // Add inverse transform to 'ref' and store.
+  {
+    // Load the reference(s).
+    __m128i ref01, ref23, ref0123;
+    int32_t buf[4];
+
+    // Load four bytes/pixels per line.
+    const __m128i ref0 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[0 * BPS]));
+    const __m128i ref1 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[1 * BPS]));
+    const __m128i ref2 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[2 * BPS]));
+    const __m128i ref3 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[3 * BPS]));
+    ref01 = _mm_unpacklo_epi32(ref0, ref1);
+    ref23 = _mm_unpacklo_epi32(ref2, ref3);
+
+    // Convert to 16b.
+    ref01 = _mm_unpacklo_epi8(ref01, zero);
+    ref23 = _mm_unpacklo_epi8(ref23, zero);
+    // Add the inverse transform(s).
+    ref01 = _mm_add_epi16(ref01, T01);
+    ref23 = _mm_add_epi16(ref23, T23);
+    // Unsigned saturate to 8b.
+    ref0123 = _mm_packus_epi16(ref01, ref23);
+
+    _mm_storeu_si128((__m128i *)buf, ref0123);
+
+    // Store four bytes/pixels per line.
+    WebPInt32ToMem(&dst[0 * BPS], buf[0]);
+    WebPInt32ToMem(&dst[1 * BPS], buf[1]);
+    WebPInt32ToMem(&dst[2 * BPS], buf[2]);
+    WebPInt32ToMem(&dst[3 * BPS], buf[3]);
+  }
+}
+
+// Does two inverse transforms.
+static void ITransform_Two_SSE2(const uint8_t* ref, const int16_t* in,
+                                uint8_t* dst) {
   // This implementation makes use of 16-bit fixed point versions of two
   // multiply constants:
   //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
@@ -49,33 +200,21 @@ static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
   __m128i T0, T1, T2, T3;
 
   // Load and concatenate the transform coefficients (we'll do two inverse
-  // transforms in parallel). In the case of only one inverse transform, the
-  // second half of the vectors will just contain random value we'll never
-  // use nor store.
+  // transforms in parallel).
   __m128i in0, in1, in2, in3;
   {
-    in0 = _mm_loadl_epi64((const __m128i*)&in[0]);
-    in1 = _mm_loadl_epi64((const __m128i*)&in[4]);
-    in2 = _mm_loadl_epi64((const __m128i*)&in[8]);
-    in3 = _mm_loadl_epi64((const __m128i*)&in[12]);
-    // a00 a10 a20 a30   x x x x
-    // a01 a11 a21 a31   x x x x
-    // a02 a12 a22 a32   x x x x
-    // a03 a13 a23 a33   x x x x
-    if (do_two) {
-      const __m128i inB0 = _mm_loadl_epi64((const __m128i*)&in[16]);
-      const __m128i inB1 = _mm_loadl_epi64((const __m128i*)&in[20]);
-      const __m128i inB2 = _mm_loadl_epi64((const __m128i*)&in[24]);
-      const __m128i inB3 = _mm_loadl_epi64((const __m128i*)&in[28]);
-      in0 = _mm_unpacklo_epi64(in0, inB0);
-      in1 = _mm_unpacklo_epi64(in1, inB1);
-      in2 = _mm_unpacklo_epi64(in2, inB2);
-      in3 = _mm_unpacklo_epi64(in3, inB3);
-      // a00 a10 a20 a30   b00 b10 b20 b30
-      // a01 a11 a21 a31   b01 b11 b21 b31
-      // a02 a12 a22 a32   b02 b12 b22 b32
-      // a03 a13 a23 a33   b03 b13 b23 b33
-    }
+    const __m128i tmp0 = _mm_loadu_si128((const __m128i*)&in[0]);
+    const __m128i tmp1 = _mm_loadu_si128((const __m128i*)&in[8]);
+    const __m128i tmp2 = _mm_loadu_si128((const __m128i*)&in[16]);
+    const __m128i tmp3 = _mm_loadu_si128((const __m128i*)&in[24]);
+    in0 = _mm_unpacklo_epi64(tmp0, tmp2);
+    in1 = _mm_unpackhi_epi64(tmp0, tmp2);
+    in2 = _mm_unpacklo_epi64(tmp1, tmp3);
+    in3 = _mm_unpackhi_epi64(tmp1, tmp3);
+    // a00 a10 a20 a30   b00 b10 b20 b30
+    // a01 a11 a21 a31   b01 b11 b21 b31
+    // a02 a12 a22 a32   b02 b12 b22 b32
+    // a03 a13 a23 a33   b03 b13 b23 b33
   }
 
   // Vertical pass and subsequent transpose.
@@ -148,19 +287,11 @@ static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
     const __m128i zero = _mm_setzero_si128();
     // Load the reference(s).
     __m128i ref0, ref1, ref2, ref3;
-    if (do_two) {
-      // Load eight bytes/pixels per line.
-      ref0 = _mm_loadl_epi64((const __m128i*)&ref[0 * BPS]);
-      ref1 = _mm_loadl_epi64((const __m128i*)&ref[1 * BPS]);
-      ref2 = _mm_loadl_epi64((const __m128i*)&ref[2 * BPS]);
-      ref3 = _mm_loadl_epi64((const __m128i*)&ref[3 * BPS]);
-    } else {
-      // Load four bytes/pixels per line.
-      ref0 = _mm_cvtsi32_si128(WebPMemToUint32(&ref[0 * BPS]));
-      ref1 = _mm_cvtsi32_si128(WebPMemToUint32(&ref[1 * BPS]));
-      ref2 = _mm_cvtsi32_si128(WebPMemToUint32(&ref[2 * BPS]));
-      ref3 = _mm_cvtsi32_si128(WebPMemToUint32(&ref[3 * BPS]));
-    }
+    // Load eight bytes/pixels per line.
+    ref0 = _mm_loadl_epi64((const __m128i*)&ref[0 * BPS]);
+    ref1 = _mm_loadl_epi64((const __m128i*)&ref[1 * BPS]);
+    ref2 = _mm_loadl_epi64((const __m128i*)&ref[2 * BPS]);
+    ref3 = _mm_loadl_epi64((const __m128i*)&ref[3 * BPS]);
     // Convert to 16b.
     ref0 = _mm_unpacklo_epi8(ref0, zero);
     ref1 = _mm_unpacklo_epi8(ref1, zero);
@@ -176,20 +307,21 @@ static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
     ref1 = _mm_packus_epi16(ref1, ref1);
     ref2 = _mm_packus_epi16(ref2, ref2);
     ref3 = _mm_packus_epi16(ref3, ref3);
-    // Store the results.
-    if (do_two) {
-      // Store eight bytes/pixels per line.
-      _mm_storel_epi64((__m128i*)&dst[0 * BPS], ref0);
-      _mm_storel_epi64((__m128i*)&dst[1 * BPS], ref1);
-      _mm_storel_epi64((__m128i*)&dst[2 * BPS], ref2);
-      _mm_storel_epi64((__m128i*)&dst[3 * BPS], ref3);
-    } else {
-      // Store four bytes/pixels per line.
-      WebPUint32ToMem(&dst[0 * BPS], _mm_cvtsi128_si32(ref0));
-      WebPUint32ToMem(&dst[1 * BPS], _mm_cvtsi128_si32(ref1));
-      WebPUint32ToMem(&dst[2 * BPS], _mm_cvtsi128_si32(ref2));
-      WebPUint32ToMem(&dst[3 * BPS], _mm_cvtsi128_si32(ref3));
-    }
+    // Store eight bytes/pixels per line.
+    _mm_storel_epi64((__m128i*)&dst[0 * BPS], ref0);
+    _mm_storel_epi64((__m128i*)&dst[1 * BPS], ref1);
+    _mm_storel_epi64((__m128i*)&dst[2 * BPS], ref2);
+    _mm_storel_epi64((__m128i*)&dst[3 * BPS], ref3);
+  }
+}
+
+// Does one or two inverse transforms.
+static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
+                            int do_two) {
+  if (do_two) {
+    ITransform_Two_SSE2(ref, in, dst);
+  } else {
+    ITransform_One_SSE2(ref, in, dst);
   }
 }
 
@@ -481,7 +613,7 @@ static void CollectHistogram_SSE2(const uint8_t* ref, const uint8_t* pred,
 // helper for chroma-DC predictions
 static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) {
   int j;
-  const __m128i values = _mm_set1_epi8(v);
+  const __m128i values = _mm_set1_epi8((char)v);
   for (j = 0; j < 8; ++j) {
     _mm_storel_epi64((__m128i*)(dst + j * BPS), values);
   }
@@ -489,7 +621,7 @@ static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) {
 
 static WEBP_INLINE void Put16_SSE2(uint8_t v, uint8_t* dst) {
   int j;
-  const __m128i values = _mm_set1_epi8(v);
+  const __m128i values = _mm_set1_epi8((char)v);
   for (j = 0; j < 16; ++j) {
     _mm_store_si128((__m128i*)(dst + j * BPS), values);
   }
@@ -540,7 +672,7 @@ static WEBP_INLINE void VerticalPred_SSE2(uint8_t* dst,
 static WEBP_INLINE void HE8uv_SSE2(uint8_t* dst, const uint8_t* left) {
   int j;
   for (j = 0; j < 8; ++j) {
-    const __m128i values = _mm_set1_epi8(left[j]);
+    const __m128i values = _mm_set1_epi8((char)left[j]);
     _mm_storel_epi64((__m128i*)dst, values);
     dst += BPS;
   }
@@ -549,7 +681,7 @@ static WEBP_INLINE void HE8uv_SSE2(uint8_t* dst, const uint8_t* left) {
 static WEBP_INLINE void HE16_SSE2(uint8_t* dst, const uint8_t* left) {
   int j;
   for (j = 0; j < 16; ++j) {
-    const __m128i values = _mm_set1_epi8(left[j]);
+    const __m128i values = _mm_set1_epi8((char)left[j]);
     _mm_store_si128((__m128i*)dst, values);
     dst += BPS;
   }
@@ -722,10 +854,10 @@ static WEBP_INLINE void VE4_SSE2(uint8_t* dst,
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(ABCDEFGH, CDEFGH00), one);
   const __m128i b = _mm_subs_epu8(a, lsb);
   const __m128i avg = _mm_avg_epu8(b, BCDEFGH0);
-  const uint32_t vals = _mm_cvtsi128_si32(avg);
+  const int vals = _mm_cvtsi128_si32(avg);
   int i;
   for (i = 0; i < 4; ++i) {
-    WebPUint32ToMem(dst + i * BPS, vals);
+    WebPInt32ToMem(dst + i * BPS, vals);
   }
 }
 
@@ -760,10 +892,10 @@ static WEBP_INLINE void LD4_SSE2(uint8_t* dst,
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(ABCDEFGH, CDEFGHH0), one);
   const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
   const __m128i abcdefg = _mm_avg_epu8(avg2, BCDEFGH0);
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               abcdefg    ));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               abcdefg    ));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
 }
 
 static WEBP_INLINE void VR4_SSE2(uint8_t* dst,
@@ -782,10 +914,10 @@ static WEBP_INLINE void VR4_SSE2(uint8_t* dst,
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
   const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
   const __m128i efgh = _mm_avg_epu8(avg2, XABCD);
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               abcd    ));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(               efgh    ));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(abcd, 1)));
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(efgh, 1)));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               abcd    ));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(               efgh    ));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(abcd, 1)));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(efgh, 1)));
 
   // these two are hard to implement in SSE2, so we keep the C-version:
   DST(0, 2) = AVG3(J, I, X);
@@ -807,11 +939,12 @@ static WEBP_INLINE void VL4_SSE2(uint8_t* dst,
   const __m128i abbc = _mm_or_si128(ab, bc);
   const __m128i lsb2 = _mm_and_si128(abbc, lsb1);
   const __m128i avg4 = _mm_subs_epu8(avg3, lsb2);
-  const uint32_t extra_out = _mm_cvtsi128_si32(_mm_srli_si128(avg4, 4));
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               avg1    ));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(               avg4    ));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg1, 1)));
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg4, 1)));
+  const uint32_t extra_out =
+      (uint32_t)_mm_cvtsi128_si32(_mm_srli_si128(avg4, 4));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(               avg1    ));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(               avg4    ));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg1, 1)));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg4, 1)));
 
   // these two are hard to get and irregular
   DST(3, 2) = (extra_out >> 0) & 0xff;
@@ -829,10 +962,10 @@ static WEBP_INLINE void RD4_SSE2(uint8_t* dst,
   const __m128i lsb = _mm_and_si128(_mm_xor_si128(JIXABCD__, LKJIXABCD), one);
   const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
   const __m128i abcdefg = _mm_avg_epu8(avg2, KJIXABCD_);
-  WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(               abcdefg    ));
-  WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
-  WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
-  WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
+  WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(               abcdefg    ));
+  WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
+  WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
+  WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
 }
 
 static WEBP_INLINE void HU4_SSE2(uint8_t* dst, const uint8_t* top) {
@@ -875,14 +1008,14 @@ static WEBP_INLINE void HD4_SSE2(uint8_t* dst, const uint8_t* top) {
 
 static WEBP_INLINE void TM4_SSE2(uint8_t* dst, const uint8_t* top) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i top_values = _mm_cvtsi32_si128(WebPMemToUint32(top));
+  const __m128i top_values = _mm_cvtsi32_si128(WebPMemToInt32(top));
   const __m128i top_base = _mm_unpacklo_epi8(top_values, zero);
   int y;
   for (y = 0; y < 4; ++y, dst += BPS) {
     const int val = top[-2 - y] - top[-1];
     const __m128i base = _mm_set1_epi16(val);
     const __m128i out = _mm_packus_epi16(_mm_add_epi16(base, top_base), zero);
-    WebPUint32ToMem(dst, _mm_cvtsi128_si32(out));
+    WebPInt32ToMem(dst, _mm_cvtsi128_si32(out));
   }
 }
 

3rdparty/libwebp/src/dsp/filters.c

@@ -189,6 +189,12 @@ static void GradientFilter_C(const uint8_t* data, int width, int height,
 
 //------------------------------------------------------------------------------
 
+static void NoneUnfilter_C(const uint8_t* prev, const uint8_t* in,
+                           uint8_t* out, int width) {
+  (void)prev;
+  if (out != in) memcpy(out, in, width * sizeof(*out));
+}
+
 static void HorizontalUnfilter_C(const uint8_t* prev, const uint8_t* in,
                                  uint8_t* out, int width) {
   uint8_t pred = (prev == NULL) ? 0 : prev[0];
@@ -233,13 +239,14 @@ static void GradientUnfilter_C(const uint8_t* prev, const uint8_t* in,
 WebPFilterFunc WebPFilters[WEBP_FILTER_LAST];
 WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST];
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8FiltersInitMIPSdspR2(void);
 extern void VP8FiltersInitMSA(void);
 extern void VP8FiltersInitNEON(void);
 extern void VP8FiltersInitSSE2(void);
 
 WEBP_DSP_INIT_FUNC(VP8FiltersInit) {
-  WebPUnfilters[WEBP_FILTER_NONE] = NULL;
+  WebPUnfilters[WEBP_FILTER_NONE] = NoneUnfilter_C;
 #if !WEBP_NEON_OMIT_C_CODE
   WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter_C;
   WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter_C;
@@ -254,7 +261,7 @@ WEBP_DSP_INIT_FUNC(VP8FiltersInit) {
 #endif
 
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8FiltersInitSSE2();
     }
@@ -271,13 +278,14 @@ WEBP_DSP_INIT_FUNC(VP8FiltersInit) {
 #endif
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     VP8FiltersInitNEON();
   }
 #endif
 
+  assert(WebPUnfilters[WEBP_FILTER_NONE] != NULL);
   assert(WebPUnfilters[WEBP_FILTER_HORIZONTAL] != NULL);
   assert(WebPUnfilters[WEBP_FILTER_VERTICAL] != NULL);
   assert(WebPUnfilters[WEBP_FILTER_GRADIENT] != NULL);

3rdparty/libwebp/src/dsp/filters_sse2.c

@@ -320,7 +320,12 @@ extern void VP8FiltersInitSSE2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8FiltersInitSSE2(void) {
   WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter_SSE2;
+#if defined(CHROMIUM)
+  // TODO(crbug.com/654974)
+  (void)VerticalUnfilter_SSE2;
+#else
   WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter_SSE2;
+#endif
   WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter_SSE2;
 
   WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter_SSE2;

3rdparty/libwebp/src/dsp/lossless.c

@@ -49,7 +49,7 @@ static WEBP_INLINE uint32_t Clip255(uint32_t a) {
 }
 
 static WEBP_INLINE int AddSubtractComponentFull(int a, int b, int c) {
-  return Clip255(a + b - c);
+  return Clip255((uint32_t)(a + b - c));
 }
 
 static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
@@ -66,7 +66,7 @@ static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
 }
 
 static WEBP_INLINE int AddSubtractComponentHalf(int a, int b) {
-  return Clip255(a + (a - b) / 2);
+  return Clip255((uint32_t)(a + (a - b) / 2));
 }
 
 static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
@@ -107,63 +107,77 @@ static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
 //------------------------------------------------------------------------------
 // Predictors
 
-uint32_t VP8LPredictor0_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor0_C(const uint32_t* const left,
+                          const uint32_t* const top) {
   (void)top;
   (void)left;
   return ARGB_BLACK;
 }
-uint32_t VP8LPredictor1_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor1_C(const uint32_t* const left,
+                          const uint32_t* const top) {
   (void)top;
-  return left;
+  return *left;
 }
-uint32_t VP8LPredictor2_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor2_C(const uint32_t* const left,
+                          const uint32_t* const top) {
   (void)left;
   return top[0];
 }
-uint32_t VP8LPredictor3_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor3_C(const uint32_t* const left,
+                          const uint32_t* const top) {
   (void)left;
   return top[1];
 }
-uint32_t VP8LPredictor4_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor4_C(const uint32_t* const left,
+                          const uint32_t* const top) {
   (void)left;
   return top[-1];
 }
-uint32_t VP8LPredictor5_C(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average3(left, top[0], top[1]);
+uint32_t VP8LPredictor5_C(const uint32_t* const left,
+                          const uint32_t* const top) {
+  const uint32_t pred = Average3(*left, top[0], top[1]);
   return pred;
 }
-uint32_t VP8LPredictor6_C(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(left, top[-1]);
+uint32_t VP8LPredictor6_C(const uint32_t* const left,
+                          const uint32_t* const top) {
+  const uint32_t pred = Average2(*left, top[-1]);
   return pred;
 }
-uint32_t VP8LPredictor7_C(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2(left, top[0]);
+uint32_t VP8LPredictor7_C(const uint32_t* const left,
+                          const uint32_t* const top) {
+  const uint32_t pred = Average2(*left, top[0]);
   return pred;
 }
-uint32_t VP8LPredictor8_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor8_C(const uint32_t* const left,
+                          const uint32_t* const top) {
   const uint32_t pred = Average2(top[-1], top[0]);
   (void)left;
   return pred;
 }
-uint32_t VP8LPredictor9_C(uint32_t left, const uint32_t* const top) {
+uint32_t VP8LPredictor9_C(const uint32_t* const left,
+                          const uint32_t* const top) {
   const uint32_t pred = Average2(top[0], top[1]);
   (void)left;
   return pred;
 }
-uint32_t VP8LPredictor10_C(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average4(left, top[-1], top[0], top[1]);
+uint32_t VP8LPredictor10_C(const uint32_t* const left,
+                           const uint32_t* const top) {
+  const uint32_t pred = Average4(*left, top[-1], top[0], top[1]);
   return pred;
 }
-uint32_t VP8LPredictor11_C(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Select(top[0], left, top[-1]);
+uint32_t VP8LPredictor11_C(const uint32_t* const left,
+                           const uint32_t* const top) {
+  const uint32_t pred = Select(top[0], *left, top[-1]);
   return pred;
 }
-uint32_t VP8LPredictor12_C(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]);
+uint32_t VP8LPredictor12_C(const uint32_t* const left,
+                           const uint32_t* const top) {
+  const uint32_t pred = ClampedAddSubtractFull(*left, top[0], top[-1]);
   return pred;
 }
-uint32_t VP8LPredictor13_C(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]);
+uint32_t VP8LPredictor13_C(const uint32_t* const left,
+                           const uint32_t* const top) {
+  const uint32_t pred = ClampedAddSubtractHalf(*left, top[0], top[-1]);
   return pred;
 }
 
@@ -279,10 +293,10 @@ void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
     const uint32_t red = argb >> 16;
     int new_red = red & 0xff;
     int new_blue = argb & 0xff;
-    new_red += ColorTransformDelta(m->green_to_red_, green);
+    new_red += ColorTransformDelta((int8_t)m->green_to_red_, green);
     new_red &= 0xff;
-    new_blue += ColorTransformDelta(m->green_to_blue_, green);
-    new_blue += ColorTransformDelta(m->red_to_blue_, (int8_t)new_red);
+    new_blue += ColorTransformDelta((int8_t)m->green_to_blue_, green);
+    new_blue += ColorTransformDelta((int8_t)m->red_to_blue_, (int8_t)new_red);
     new_blue &= 0xff;
     dst[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue);
   }
@@ -381,7 +395,7 @@ void VP8LInverseTransform(const VP8LTransform* const transform,
   assert(row_start < row_end);
   assert(row_end <= transform->ysize_);
   switch (transform->type_) {
-    case SUBTRACT_GREEN:
+    case SUBTRACT_GREEN_TRANSFORM:
       VP8LAddGreenToBlueAndRed(in, (row_end - row_start) * width, out);
       break;
     case PREDICTOR_TRANSFORM:
@@ -574,7 +588,9 @@ VP8LConvertFunc VP8LConvertBGRAToBGR;
 VP8LMapARGBFunc VP8LMapColor32b;
 VP8LMapAlphaFunc VP8LMapColor8b;
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8LDspInitSSE2(void);
+extern void VP8LDspInitSSE41(void);
 extern void VP8LDspInitNEON(void);
 extern void VP8LDspInitMIPSdspR2(void);
 extern void VP8LDspInitMSA(void);
@@ -621,9 +637,14 @@ WEBP_DSP_INIT_FUNC(VP8LDspInit) {
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8LDspInitSSE2();
+#if defined(WEBP_HAVE_SSE41)
+      if (VP8GetCPUInfo(kSSE4_1)) {
+        VP8LDspInitSSE41();
+      }
+#endif
     }
 #endif
 #if defined(WEBP_USE_MIPS_DSP_R2)
@@ -638,7 +659,7 @@ WEBP_DSP_INIT_FUNC(VP8LDspInit) {
 #endif
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     VP8LDspInitNEON();

3rdparty/libwebp/src/dsp/lossless.h

@@ -28,23 +28,38 @@ extern "C" {
 //------------------------------------------------------------------------------
 // Decoding
 
-typedef uint32_t (*VP8LPredictorFunc)(uint32_t left, const uint32_t* const top);
+typedef uint32_t (*VP8LPredictorFunc)(const uint32_t* const left,
+                                      const uint32_t* const top);
 extern VP8LPredictorFunc VP8LPredictors[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);
+uint32_t VP8LPredictor0_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor1_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor2_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor3_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor4_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor5_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor6_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor7_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor8_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor9_C(const uint32_t* const left,
+                          const uint32_t* const top);
+uint32_t VP8LPredictor10_C(const uint32_t* const left,
+                           const uint32_t* const top);
+uint32_t VP8LPredictor11_C(const uint32_t* const left,
+                           const uint32_t* const top);
+uint32_t VP8LPredictor12_C(const uint32_t* const left,
+                           const uint32_t* const top);
+uint32_t VP8LPredictor13_C(const uint32_t* const 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,
@@ -167,9 +182,9 @@ extern VP8LPredictorAddSubFunc VP8LPredictorsSub_C[16];
 // -----------------------------------------------------------------------------
 // Huffman-cost related functions.
 
-typedef double (*VP8LCostFunc)(const uint32_t* population, int length);
-typedef double (*VP8LCostCombinedFunc)(const uint32_t* X, const uint32_t* Y,
-                                       int length);
+typedef uint32_t (*VP8LCostFunc)(const uint32_t* population, int length);
+typedef uint32_t (*VP8LCostCombinedFunc)(const uint32_t* X, const uint32_t* Y,
+                                         int length);
 typedef float (*VP8LCombinedShannonEntropyFunc)(const int X[256],
                                                 const int Y[256]);
 
@@ -183,7 +198,7 @@ typedef struct {        // small struct to hold counters
 } VP8LStreaks;
 
 typedef struct {            // small struct to hold bit entropy results
-  double entropy;           // entropy
+  float entropy;            // entropy
   uint32_t sum;             // sum of the population
   int nonzeros;             // number of non-zero elements in the population
   uint32_t max_val;         // maximum value in the population

3rdparty/libwebp/src/dsp/lossless_common.h

@@ -16,9 +16,9 @@
 #ifndef WEBP_DSP_LOSSLESS_COMMON_H_
 #define WEBP_DSP_LOSSLESS_COMMON_H_
 
-#include "src/webp/types.h"
-
+#include "src/dsp/cpu.h"
 #include "src/utils/utils.h"
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -166,7 +166,7 @@ uint32_t VP8LSubPixels(uint32_t a, uint32_t b) {
 }
 
 //------------------------------------------------------------------------------
-// Transform-related functions use din both encoding and decoding.
+// Transform-related functions used in both encoding and decoding.
 
 // Macros used to create a batch predictor that iteratively uses a
 // one-pixel predictor.
@@ -179,7 +179,7 @@ static void PREDICTOR_ADD(const uint32_t* in, const uint32_t* upper, \
   int x;                                                             \
   assert(upper != NULL);                                             \
   for (x = 0; x < num_pixels; ++x) {                                 \
-    const uint32_t pred = (PREDICTOR)(out[x - 1], upper + x);        \
+    const uint32_t pred = (PREDICTOR)(&out[x - 1], upper + x);       \
     out[x] = VP8LAddPixels(in[x], pred);                             \
   }                                                                  \
 }

3rdparty/libwebp/src/dsp/lossless_enc.c

@@ -329,6 +329,15 @@ const uint8_t kPrefixEncodeExtraBitsValue[PREFIX_LOOKUP_IDX_MAX] = {
 static float FastSLog2Slow_C(uint32_t v) {
   assert(v >= LOG_LOOKUP_IDX_MAX);
   if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
+#if !defined(WEBP_HAVE_SLOW_CLZ_CTZ)
+    // use clz if available
+    const int log_cnt = BitsLog2Floor(v) - 7;
+    const uint32_t y = 1 << log_cnt;
+    int correction = 0;
+    const float v_f = (float)v;
+    const uint32_t orig_v = v;
+    v >>= log_cnt;
+#else
     int log_cnt = 0;
     uint32_t y = 1;
     int correction = 0;
@@ -339,6 +348,7 @@ static float FastSLog2Slow_C(uint32_t v) {
       v = v >> 1;
       y = y << 1;
     } while (v >= LOG_LOOKUP_IDX_MAX);
+#endif
     // vf = (2^log_cnt) * Xf; where y = 2^log_cnt and Xf < 256
     // Xf = floor(Xf) * (1 + (v % y) / v)
     // log2(Xf) = log2(floor(Xf)) + log2(1 + (v % y) / v)
@@ -355,6 +365,14 @@ static float FastSLog2Slow_C(uint32_t v) {
 static float FastLog2Slow_C(uint32_t v) {
   assert(v >= LOG_LOOKUP_IDX_MAX);
   if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
+#if !defined(WEBP_HAVE_SLOW_CLZ_CTZ)
+    // use clz if available
+    const int log_cnt = BitsLog2Floor(v) - 7;
+    const uint32_t y = 1 << log_cnt;
+    const uint32_t orig_v = v;
+    double log_2;
+    v >>= log_cnt;
+#else
     int log_cnt = 0;
     uint32_t y = 1;
     const uint32_t orig_v = v;
@@ -364,6 +382,7 @@ static float FastLog2Slow_C(uint32_t v) {
       v = v >> 1;
       y = y << 1;
     } while (v >= LOG_LOOKUP_IDX_MAX);
+#endif
     log_2 = kLog2Table[v] + log_cnt;
     if (orig_v >= APPROX_LOG_MAX) {
       // Since the division is still expensive, add this correction factor only
@@ -383,7 +402,7 @@ static float FastLog2Slow_C(uint32_t v) {
 // Compute the combined Shanon's entropy for distribution {X} and {X+Y}
 static float CombinedShannonEntropy_C(const int X[256], const int Y[256]) {
   int i;
-  double retval = 0.;
+  float retval = 0.f;
   int sumX = 0, sumXY = 0;
   for (i = 0; i < 256; ++i) {
     const int x = X[i];
@@ -399,7 +418,7 @@ static float CombinedShannonEntropy_C(const int X[256], const int Y[256]) {
     }
   }
   retval += VP8LFastSLog2(sumX) + VP8LFastSLog2(sumXY);
-  return (float)retval;
+  return retval;
 }
 
 void VP8LBitEntropyInit(VP8LBitEntropy* const entropy) {
@@ -503,11 +522,11 @@ static void GetCombinedEntropyUnrefined_C(const uint32_t X[],
 void VP8LSubtractGreenFromBlueAndRed_C(uint32_t* argb_data, int num_pixels) {
   int i;
   for (i = 0; i < num_pixels; ++i) {
-    const int argb = argb_data[i];
+    const int argb = (int)argb_data[i];
     const int green = (argb >> 8) & 0xff;
     const uint32_t new_r = (((argb >> 16) & 0xff) - green) & 0xff;
     const uint32_t new_b = (((argb >>  0) & 0xff) - green) & 0xff;
-    argb_data[i] = (argb & 0xff00ff00u) | (new_r << 16) | new_b;
+    argb_data[i] = ((uint32_t)argb & 0xff00ff00u) | (new_r << 16) | new_b;
   }
 }
 
@@ -528,10 +547,10 @@ void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
     const int8_t red   = U32ToS8(argb >> 16);
     int new_red = red & 0xff;
     int new_blue = argb & 0xff;
-    new_red -= ColorTransformDelta(m->green_to_red_, green);
+    new_red -= ColorTransformDelta((int8_t)m->green_to_red_, green);
     new_red &= 0xff;
-    new_blue -= ColorTransformDelta(m->green_to_blue_, green);
-    new_blue -= ColorTransformDelta(m->red_to_blue_, red);
+    new_blue -= ColorTransformDelta((int8_t)m->green_to_blue_, green);
+    new_blue -= ColorTransformDelta((int8_t)m->red_to_blue_, red);
     new_blue &= 0xff;
     data[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue);
   }
@@ -541,7 +560,7 @@ static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
                                              uint32_t argb) {
   const int8_t green = U32ToS8(argb >> 8);
   int new_red = argb >> 16;
-  new_red -= ColorTransformDelta(green_to_red, green);
+  new_red -= ColorTransformDelta((int8_t)green_to_red, green);
   return (new_red & 0xff);
 }
 
@@ -550,9 +569,9 @@ static WEBP_INLINE uint8_t TransformColorBlue(uint8_t green_to_blue,
                                               uint32_t argb) {
   const int8_t green = U32ToS8(argb >>  8);
   const int8_t red   = U32ToS8(argb >> 16);
-  uint8_t new_blue = argb & 0xff;
-  new_blue -= ColorTransformDelta(green_to_blue, green);
-  new_blue -= ColorTransformDelta(red_to_blue, red);
+  int new_blue = argb & 0xff;
+  new_blue -= ColorTransformDelta((int8_t)green_to_blue, green);
+  new_blue -= ColorTransformDelta((int8_t)red_to_blue, red);
   return (new_blue & 0xff);
 }
 
@@ -617,20 +636,25 @@ void VP8LBundleColorMap_C(const uint8_t* const row, int width, int xbits,
 
 //------------------------------------------------------------------------------
 
-static double ExtraCost_C(const uint32_t* population, int length) {
+static uint32_t ExtraCost_C(const uint32_t* population, int length) {
   int i;
-  double cost = 0.;
-  for (i = 2; i < length - 2; ++i) cost += (i >> 1) * population[i + 2];
+  uint32_t cost = population[4] + population[5];
+  assert(length % 2 == 0);
+  for (i = 2; i < length / 2 - 1; ++i) {
+    cost += i * (population[2 * i + 2] + population[2 * i + 3]);
+  }
   return cost;
 }
 
-static double ExtraCostCombined_C(const uint32_t* X, const uint32_t* Y,
-                                  int length) {
+static uint32_t ExtraCostCombined_C(const uint32_t* X, const uint32_t* Y,
+                                    int length) {
   int i;
-  double cost = 0.;
-  for (i = 2; i < length - 2; ++i) {
-    const int xy = X[i + 2] + Y[i + 2];
-    cost += (i >> 1) * xy;
+  uint32_t cost = X[4] + Y[4] + X[5] + Y[5];
+  assert(length % 2 == 0);
+  for (i = 2; i < length / 2 - 1; ++i) {
+    const int xy0 = X[2 * i + 2] + Y[2 * i + 2];
+    const int xy1 = X[2 * i + 3] + Y[2 * i + 3];
+    cost += i * (xy0 + xy1);
   }
   return cost;
 }
@@ -726,7 +750,7 @@ static void PredictorSub##PREDICTOR_I##_C(const uint32_t* in,              \
   assert(upper != NULL);                                                   \
   for (x = 0; x < num_pixels; ++x) {                                       \
     const uint32_t pred =                                                  \
-        VP8LPredictor##PREDICTOR_I##_C(in[x - 1], upper + x);              \
+        VP8LPredictor##PREDICTOR_I##_C(&in[x - 1], upper + x);             \
     out[x] = VP8LSubPixels(in[x], pred);                                   \
   }                                                                        \
 }
@@ -772,6 +796,7 @@ VP8LBundleColorMapFunc VP8LBundleColorMap;
 VP8LPredictorAddSubFunc VP8LPredictorsSub[16];
 VP8LPredictorAddSubFunc VP8LPredictorsSub_C[16];
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8LEncDspInitSSE2(void);
 extern void VP8LEncDspInitSSE41(void);
 extern void VP8LEncDspInitNEON(void);
@@ -843,10 +868,10 @@ WEBP_DSP_INIT_FUNC(VP8LEncDspInit) {
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8LEncDspInitSSE2();
-#if defined(WEBP_USE_SSE41)
+#if defined(WEBP_HAVE_SSE41)
       if (VP8GetCPUInfo(kSSE4_1)) {
         VP8LEncDspInitSSE41();
       }
@@ -870,7 +895,7 @@ WEBP_DSP_INIT_FUNC(VP8LEncDspInit) {
 #endif
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     VP8LEncDspInitNEON();

3rdparty/libwebp/src/dsp/lossless_enc_mips32.c

@@ -103,8 +103,8 @@ static float FastLog2Slow_MIPS32(uint32_t v) {
 //     cost += i * *(pop + 1);
 //     pop += 2;
 //   }
-//   return (double)cost;
-static double ExtraCost_MIPS32(const uint32_t* const population, int length) {
+//   return cost;
+static uint32_t ExtraCost_MIPS32(const uint32_t* const population, int length) {
   int i, temp0, temp1;
   const uint32_t* pop = &population[4];
   const uint32_t* const LoopEnd = &population[length];
@@ -130,7 +130,7 @@ static double ExtraCost_MIPS32(const uint32_t* const population, int length) {
     : "memory", "hi", "lo"
   );
 
-  return (double)((int64_t)temp0 << 32 | temp1);
+  return ((int64_t)temp0 << 32 | temp1);
 }
 
 // C version of this function:
@@ -148,9 +148,9 @@ static double ExtraCost_MIPS32(const uint32_t* const population, int length) {
 //     pX += 2;
 //     pY += 2;
 //   }
-//   return (double)cost;
-static double ExtraCostCombined_MIPS32(const uint32_t* const X,
-                                       const uint32_t* const Y, int length) {
+//   return cost;
+static uint32_t ExtraCostCombined_MIPS32(const uint32_t* const X,
+                                         const uint32_t* const Y, int length) {
   int i, temp0, temp1, temp2, temp3;
   const uint32_t* pX = &X[4];
   const uint32_t* pY = &Y[4];
@@ -183,7 +183,7 @@ static double ExtraCostCombined_MIPS32(const uint32_t* const X,
     : "memory", "hi", "lo"
   );
 
-  return (double)((int64_t)temp0 << 32 | temp1);
+  return ((int64_t)temp0 << 32 | temp1);
 }
 
 #define HUFFMAN_COST_PASS                                 \
@@ -347,24 +347,24 @@ static void GetCombinedEntropyUnrefined_MIPS32(const uint32_t X[],
 static void AddVector_MIPS32(const uint32_t* pa, const uint32_t* pb,
                              uint32_t* pout, int size) {
   uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
-  const uint32_t end = ((size) / 4) * 4;
+  const int end = ((size) / 4) * 4;
   const uint32_t* const LoopEnd = pa + end;
   int i;
   ASM_START
   ADD_TO_OUT(0, 4, 8, 12, 1, pa, pb, pout)
   ASM_END_0
-  for (i = end; i < size; ++i) pout[i] = pa[i] + pb[i];
+  for (i = 0; i < size - end; ++i) pout[i] = pa[i] + pb[i];
 }
 
 static void AddVectorEq_MIPS32(const uint32_t* pa, uint32_t* pout, int size) {
   uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
-  const uint32_t end = ((size) / 4) * 4;
+  const int end = ((size) / 4) * 4;
   const uint32_t* const LoopEnd = pa + end;
   int i;
   ASM_START
   ADD_TO_OUT(0, 4, 8, 12, 0, pa, pout, pout)
   ASM_END_1
-  for (i = end; i < size; ++i) pout[i] += pa[i];
+  for (i = 0; i < size - end; ++i) pout[i] += pa[i];
 }
 
 #undef ASM_END_1

3rdparty/libwebp/src/dsp/lossless_enc_neon.c

@@ -25,7 +25,7 @@
 
 // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
 // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
 #define USE_VTBLQ
 #endif

3rdparty/libwebp/src/dsp/lossless_enc_sse2.c

@@ -54,8 +54,8 @@ static void TransformColor_SSE2(const VP8LMultipliers* const m,
   const __m128i mults_rb = MK_CST_16(CST_5b(m->green_to_red_),
                                      CST_5b(m->green_to_blue_));
   const __m128i mults_b2 = MK_CST_16(CST_5b(m->red_to_blue_), 0);
-  const __m128i mask_ag = _mm_set1_epi32(0xff00ff00);  // alpha-green masks
-  const __m128i mask_rb = _mm_set1_epi32(0x00ff00ff);  // red-blue masks
+  const __m128i mask_ag = _mm_set1_epi32((int)0xff00ff00);  // alpha-green masks
+  const __m128i mask_rb = _mm_set1_epi32(0x00ff00ff);       // red-blue masks
   int i;
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]); // argb
@@ -232,79 +232,55 @@ static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
 //------------------------------------------------------------------------------
 // Entropy
 
-// Checks whether the X or Y contribution is worth computing and adding.
-// Used in loop unrolling.
-#define ANALYZE_X_OR_Y(x_or_y, j)                                           \
-  do {                                                                      \
-    if ((x_or_y)[i + (j)] != 0) retval -= VP8LFastSLog2((x_or_y)[i + (j)]); \
-  } while (0)
-
-// Checks whether the X + Y contribution is worth computing and adding.
-// Used in loop unrolling.
-#define ANALYZE_XY(j)                  \
-  do {                                 \
-    if (tmp[j] != 0) {                 \
-      retval -= VP8LFastSLog2(tmp[j]); \
-      ANALYZE_X_OR_Y(X, j);            \
-    }                                  \
-  } while (0)
-
-#if !(defined(__i386__) || defined(_M_IX86))
+// 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(WEBP_HAVE_SLOW_CLZ_CTZ) || defined(__i386__) || defined(_M_IX86))
+
 static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
   int i;
-  double retval = 0.;
-  int sumX, sumXY;
-  int32_t tmp[4];
-  __m128i zero = _mm_setzero_si128();
-  // Sums up X + Y, 4 ints at a time (and will merge it at the end for sumXY).
-  __m128i sumXY_128 = zero;
-  __m128i sumX_128 = zero;
-
-  for (i = 0; i < 256; i += 4) {
-    const __m128i x = _mm_loadu_si128((const __m128i*)(X + i));
-    const __m128i y = _mm_loadu_si128((const __m128i*)(Y + i));
-
-    // Check if any X is non-zero: this actually provides a speedup as X is
-    // usually sparse.
-    if (_mm_movemask_epi8(_mm_cmpeq_epi32(x, zero)) != 0xFFFF) {
-      const __m128i xy_128 = _mm_add_epi32(x, y);
-      sumXY_128 = _mm_add_epi32(sumXY_128, xy_128);
-
-      sumX_128 = _mm_add_epi32(sumX_128, x);
-
-      // Analyze the different X + Y.
-      _mm_storeu_si128((__m128i*)tmp, xy_128);
-
-      ANALYZE_XY(0);
-      ANALYZE_XY(1);
-      ANALYZE_XY(2);
-      ANALYZE_XY(3);
-    } else {
-      // X is fully 0, so only deal with Y.
-      sumXY_128 = _mm_add_epi32(sumXY_128, y);
-
-      ANALYZE_X_OR_Y(Y, 0);
-      ANALYZE_X_OR_Y(Y, 1);
-      ANALYZE_X_OR_Y(Y, 2);
-      ANALYZE_X_OR_Y(Y, 3);
+  float retval = 0.f;
+  int sumX = 0, sumXY = 0;
+  const __m128i zero = _mm_setzero_si128();
+
+  for (i = 0; i < 256; i += 16) {
+    const __m128i x0 = _mm_loadu_si128((const __m128i*)(X + i +  0));
+    const __m128i y0 = _mm_loadu_si128((const __m128i*)(Y + i +  0));
+    const __m128i x1 = _mm_loadu_si128((const __m128i*)(X + i +  4));
+    const __m128i y1 = _mm_loadu_si128((const __m128i*)(Y + i +  4));
+    const __m128i x2 = _mm_loadu_si128((const __m128i*)(X + i +  8));
+    const __m128i y2 = _mm_loadu_si128((const __m128i*)(Y + i +  8));
+    const __m128i x3 = _mm_loadu_si128((const __m128i*)(X + i + 12));
+    const __m128i y3 = _mm_loadu_si128((const __m128i*)(Y + i + 12));
+    const __m128i x4 = _mm_packs_epi16(_mm_packs_epi32(x0, x1),
+                                       _mm_packs_epi32(x2, x3));
+    const __m128i y4 = _mm_packs_epi16(_mm_packs_epi32(y0, y1),
+                                       _mm_packs_epi32(y2, y3));
+    const int32_t mx = _mm_movemask_epi8(_mm_cmpgt_epi8(x4, zero));
+    int32_t my = _mm_movemask_epi8(_mm_cmpgt_epi8(y4, zero)) | mx;
+    while (my) {
+      const int32_t j = BitsCtz(my);
+      int xy;
+      if ((mx >> j) & 1) {
+        const int x = X[i + j];
+        sumXY += x;
+        retval -= VP8LFastSLog2(x);
+      }
+      xy = X[i + j] + Y[i + j];
+      sumX += xy;
+      retval -= VP8LFastSLog2(xy);
+      my &= my - 1;
     }
   }
-
-  // Sum up sumX_128 to get sumX.
-  _mm_storeu_si128((__m128i*)tmp, sumX_128);
-  sumX = tmp[3] + tmp[2] + tmp[1] + tmp[0];
-
-  // Sum up sumXY_128 to get sumXY.
-  _mm_storeu_si128((__m128i*)tmp, sumXY_128);
-  sumXY = tmp[3] + tmp[2] + tmp[1] + tmp[0];
-
   retval += VP8LFastSLog2(sumX) + VP8LFastSLog2(sumXY);
-  return (float)retval;
+  return retval;
 }
-#endif  // !(defined(__i386__) || defined(_M_IX86))
 
-#undef ANALYZE_X_OR_Y
-#undef ANALYZE_XY
+#else
+
+#define DONT_USE_COMBINED_SHANNON_ENTROPY_SSE2_FUNC   // won't be faster
+
+#endif
 
 //------------------------------------------------------------------------------
 
@@ -400,7 +376,7 @@ static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
       break;
     }
     case 2: {
-      const __m128i mask_or = _mm_set1_epi32(0xff000000);
+      const __m128i mask_or = _mm_set1_epi32((int)0xff000000);
       const __m128i mul_cst = _mm_set1_epi16(0x0104);
       const __m128i mask_mul = _mm_set1_epi16(0x0f00);
       for (x = 0; x + 16 <= width; x += 16, dst += 4) {
@@ -451,7 +427,7 @@ static WEBP_INLINE void Average2_m128i(const __m128i* const a0,
 static void PredictorSub0_SSE2(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) {
   int i;
-  const __m128i black = _mm_set1_epi32(ARGB_BLACK);
+  const __m128i black = _mm_set1_epi32((int)ARGB_BLACK);
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
     const __m128i res = _mm_sub_epi8(src, black);
@@ -662,10 +638,7 @@ 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))
+#if !defined(DONT_USE_COMBINED_SHANNON_ENTROPY_SSE2_FUNC)
   VP8LCombinedShannonEntropy = CombinedShannonEntropy_SSE2;
 #endif
   VP8LVectorMismatch = VectorMismatch_SSE2;

3rdparty/libwebp/src/dsp/lossless_enc_sse41.c

@@ -18,8 +18,53 @@
 #include <smmintrin.h>
 #include "src/dsp/lossless.h"
 
-// For sign-extended multiplying constants, pre-shifted by 5:
-#define CST_5b(X)  (((int16_t)((uint16_t)(X) << 8)) >> 5)
+//------------------------------------------------------------------------------
+// Cost operations.
+
+static WEBP_INLINE uint32_t HorizontalSum_SSE41(__m128i cost) {
+  cost = _mm_add_epi32(cost, _mm_srli_si128(cost, 8));
+  cost = _mm_add_epi32(cost, _mm_srli_si128(cost, 4));
+  return _mm_cvtsi128_si32(cost);
+}
+
+static uint32_t ExtraCost_SSE41(const uint32_t* const a, int length) {
+  int i;
+  __m128i cost = _mm_set_epi32(2 * a[7], 2 * a[6], a[5], a[4]);
+  assert(length % 8 == 0);
+
+  for (i = 8; i + 8 <= length; i += 8) {
+    const int j = (i - 2) >> 1;
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i]);
+    const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i + 4]);
+    const __m128i w = _mm_set_epi32(j + 3, j + 2, j + 1, j);
+    const __m128i a2 = _mm_hadd_epi32(a0, a1);
+    const __m128i mul = _mm_mullo_epi32(a2, w);
+    cost = _mm_add_epi32(mul, cost);
+  }
+  return HorizontalSum_SSE41(cost);
+}
+
+static uint32_t ExtraCostCombined_SSE41(const uint32_t* const a,
+                                        const uint32_t* const b, int length) {
+  int i;
+  __m128i cost = _mm_add_epi32(_mm_set_epi32(2 * a[7], 2 * a[6], a[5], a[4]),
+                               _mm_set_epi32(2 * b[7], 2 * b[6], b[5], b[4]));
+  assert(length % 8 == 0);
+
+  for (i = 8; i + 8 <= length; i += 8) {
+    const int j = (i - 2) >> 1;
+    const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i]);
+    const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i + 4]);
+    const __m128i b0 = _mm_loadu_si128((const __m128i*)&b[i]);
+    const __m128i b1 = _mm_loadu_si128((const __m128i*)&b[i + 4]);
+    const __m128i w = _mm_set_epi32(j + 3, j + 2, j + 1, j);
+    const __m128i a2 = _mm_hadd_epi32(a0, a1);
+    const __m128i b2 = _mm_hadd_epi32(b0, b1);
+    const __m128i mul = _mm_mullo_epi32(_mm_add_epi32(a2, b2), w);
+    cost = _mm_add_epi32(mul, cost);
+  }
+  return HorizontalSum_SSE41(cost);
+}
 
 //------------------------------------------------------------------------------
 // Subtract-Green Transform
@@ -44,46 +89,50 @@ static void SubtractGreenFromBlueAndRed_SSE41(uint32_t* argb_data,
 //------------------------------------------------------------------------------
 // Color Transform
 
-#define SPAN 8
+// For sign-extended multiplying constants, pre-shifted by 5:
+#define CST_5b(X) (((int16_t)((uint16_t)(X) << 8)) >> 5)
+
+#define MK_CST_16(HI, LO) \
+  _mm_set1_epi32((int)(((uint32_t)(HI) << 16) | ((LO) & 0xffff)))
+
 static void CollectColorBlueTransforms_SSE41(const uint32_t* argb, int stride,
                                              int tile_width, int tile_height,
                                              int green_to_blue, int red_to_blue,
                                              int histo[]) {
-  const __m128i mults_r = _mm_set1_epi16(CST_5b(red_to_blue));
-  const __m128i mults_g = _mm_set1_epi16(CST_5b(green_to_blue));
-  const __m128i mask_g = _mm_set1_epi16((short)0xff00);   // green mask
-  const __m128i mask_gb = _mm_set1_epi32(0xffff);         // green/blue mask
-  const __m128i mask_b = _mm_set1_epi16(0x00ff);          // blue mask
-  const __m128i shuffler_lo = _mm_setr_epi8(-1, 2, -1, 6, -1, 10, -1, 14, -1,
-                                            -1, -1, -1, -1, -1, -1, -1);
-  const __m128i shuffler_hi = _mm_setr_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1,
-                                            2, -1, 6, -1, 10, -1, 14);
-  int y;
-  for (y = 0; y < tile_height; ++y) {
-    const uint32_t* const src = argb + y * stride;
-    int i, x;
-    for (x = 0; x + SPAN <= tile_width; x += SPAN) {
-      uint16_t values[SPAN];
-      const __m128i in0 = _mm_loadu_si128((__m128i*)&src[x + 0]);
-      const __m128i in1 = _mm_loadu_si128((__m128i*)&src[x + SPAN / 2]);
-      const __m128i r0 = _mm_shuffle_epi8(in0, shuffler_lo);
-      const __m128i r1 = _mm_shuffle_epi8(in1, shuffler_hi);
-      const __m128i r = _mm_or_si128(r0, r1);         // r 0
-      const __m128i gb0 = _mm_and_si128(in0, mask_gb);
-      const __m128i gb1 = _mm_and_si128(in1, mask_gb);
-      const __m128i gb = _mm_packus_epi32(gb0, gb1);  // g b
-      const __m128i g = _mm_and_si128(gb, mask_g);    // g 0
-      const __m128i A = _mm_mulhi_epi16(r, mults_r);  // x dbr
-      const __m128i B = _mm_mulhi_epi16(g, mults_g);  // x dbg
-      const __m128i C = _mm_sub_epi8(gb, B);          // x b'
-      const __m128i D = _mm_sub_epi8(C, A);           // x b''
-      const __m128i E = _mm_and_si128(D, mask_b);     // 0 b''
-      _mm_storeu_si128((__m128i*)values, E);
-      for (i = 0; i < SPAN; ++i) ++histo[values[i]];
+  const __m128i mult =
+      MK_CST_16(CST_5b(red_to_blue) + 256,CST_5b(green_to_blue));
+  const __m128i perm =
+      _mm_setr_epi8(-1, 1, -1, 2, -1, 5, -1, 6, -1, 9, -1, 10, -1, 13, -1, 14);
+  if (tile_width >= 4) {
+    int y;
+    for (y = 0; y < tile_height; ++y) {
+      const uint32_t* const src = argb + y * stride;
+      const __m128i A1 = _mm_loadu_si128((const __m128i*)src);
+      const __m128i B1 = _mm_shuffle_epi8(A1, perm);
+      const __m128i C1 = _mm_mulhi_epi16(B1, mult);
+      const __m128i D1 = _mm_sub_epi16(A1, C1);
+      __m128i E = _mm_add_epi16(_mm_srli_epi32(D1, 16), D1);
+      int x;
+      for (x = 4; x + 4 <= tile_width; x += 4) {
+        const __m128i A2 = _mm_loadu_si128((const __m128i*)(src + x));
+        __m128i B2, C2, D2;
+        ++histo[_mm_extract_epi8(E,  0)];
+        B2 = _mm_shuffle_epi8(A2, perm);
+        ++histo[_mm_extract_epi8(E,  4)];
+        C2 = _mm_mulhi_epi16(B2, mult);
+        ++histo[_mm_extract_epi8(E,  8)];
+        D2 = _mm_sub_epi16(A2, C2);
+        ++histo[_mm_extract_epi8(E, 12)];
+        E = _mm_add_epi16(_mm_srli_epi32(D2, 16), D2);
+      }
+      ++histo[_mm_extract_epi8(E,  0)];
+      ++histo[_mm_extract_epi8(E,  4)];
+      ++histo[_mm_extract_epi8(E,  8)];
+      ++histo[_mm_extract_epi8(E, 12)];
     }
   }
   {
-    const int left_over = tile_width & (SPAN - 1);
+    const int left_over = tile_width & 3;
     if (left_over > 0) {
       VP8LCollectColorBlueTransforms_C(argb + tile_width - left_over, stride,
                                        left_over, tile_height,
@@ -95,33 +144,37 @@ static void CollectColorBlueTransforms_SSE41(const uint32_t* argb, int stride,
 static void CollectColorRedTransforms_SSE41(const uint32_t* argb, int stride,
                                             int tile_width, int tile_height,
                                             int green_to_red, int histo[]) {
-  const __m128i mults_g = _mm_set1_epi16(CST_5b(green_to_red));
-  const __m128i mask_g = _mm_set1_epi32(0x00ff00);  // green mask
-  const __m128i mask = _mm_set1_epi16(0xff);
-
-  int y;
-  for (y = 0; y < tile_height; ++y) {
-    const uint32_t* const src = argb + y * stride;
-    int i, x;
-    for (x = 0; x + SPAN <= tile_width; x += SPAN) {
-      uint16_t values[SPAN];
-      const __m128i in0 = _mm_loadu_si128((__m128i*)&src[x + 0]);
-      const __m128i in1 = _mm_loadu_si128((__m128i*)&src[x + SPAN / 2]);
-      const __m128i g0 = _mm_and_si128(in0, mask_g);  // 0 0  | g 0
-      const __m128i g1 = _mm_and_si128(in1, mask_g);
-      const __m128i g = _mm_packus_epi32(g0, g1);     // g 0
-      const __m128i A0 = _mm_srli_epi32(in0, 16);     // 0 0  | x r
-      const __m128i A1 = _mm_srli_epi32(in1, 16);
-      const __m128i A = _mm_packus_epi32(A0, A1);     // x r
-      const __m128i B = _mm_mulhi_epi16(g, mults_g);  // x dr
-      const __m128i C = _mm_sub_epi8(A, B);           // x r'
-      const __m128i D = _mm_and_si128(C, mask);       // 0 r'
-      _mm_storeu_si128((__m128i*)values, D);
-      for (i = 0; i < SPAN; ++i) ++histo[values[i]];
+
+  const __m128i mult = MK_CST_16(0, CST_5b(green_to_red));
+  const __m128i mask_g = _mm_set1_epi32(0x0000ff00);
+  if (tile_width >= 4) {
+    int y;
+    for (y = 0; y < tile_height; ++y) {
+      const uint32_t* const src = argb + y * stride;
+      const __m128i A1 = _mm_loadu_si128((const __m128i*)src);
+      const __m128i B1 = _mm_and_si128(A1, mask_g);
+      const __m128i C1 = _mm_madd_epi16(B1, mult);
+      __m128i D = _mm_sub_epi16(A1, C1);
+      int x;
+      for (x = 4; x + 4 <= tile_width; x += 4) {
+        const __m128i A2 = _mm_loadu_si128((const __m128i*)(src + x));
+        __m128i B2, C2;
+        ++histo[_mm_extract_epi8(D,  2)];
+        B2 = _mm_and_si128(A2, mask_g);
+        ++histo[_mm_extract_epi8(D,  6)];
+        C2 = _mm_madd_epi16(B2, mult);
+        ++histo[_mm_extract_epi8(D, 10)];
+        ++histo[_mm_extract_epi8(D, 14)];
+        D = _mm_sub_epi16(A2, C2);
+      }
+      ++histo[_mm_extract_epi8(D,  2)];
+      ++histo[_mm_extract_epi8(D,  6)];
+      ++histo[_mm_extract_epi8(D, 10)];
+      ++histo[_mm_extract_epi8(D, 14)];
     }
   }
   {
-    const int left_over = tile_width & (SPAN - 1);
+    const int left_over = tile_width & 3;
     if (left_over > 0) {
       VP8LCollectColorRedTransforms_C(argb + tile_width - left_over, stride,
                                       left_over, tile_height, green_to_red,
@@ -130,12 +183,16 @@ static void CollectColorRedTransforms_SSE41(const uint32_t* argb, int stride,
   }
 }
 
+#undef MK_CST_16
+
 //------------------------------------------------------------------------------
 // Entry point
 
 extern void VP8LEncDspInitSSE41(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE41(void) {
+  VP8LExtraCost = ExtraCost_SSE41;
+  VP8LExtraCostCombined = ExtraCostCombined_SSE41;
   VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed_SSE41;
   VP8LCollectColorBlueTransforms = CollectColorBlueTransforms_SSE41;
   VP8LCollectColorRedTransforms = CollectColorRedTransforms_SSE41;

3rdparty/libwebp/src/dsp/lossless_mips_dsp_r2.c

@@ -188,46 +188,51 @@ static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
   return Average2(Average2(a0, a1), Average2(a2, a3));
 }
 
-static uint32_t Predictor5_MIPSdspR2(uint32_t left, const uint32_t* const top) {
-  return Average3(left, top[0], top[1]);
+static uint32_t Predictor5_MIPSdspR2(const uint32_t* const left,
+                                     const uint32_t* const top) {
+  return Average3(*left, top[0], top[1]);
 }
 
-static uint32_t Predictor6_MIPSdspR2(uint32_t left, const uint32_t* const top) {
-  return Average2(left, top[-1]);
+static uint32_t Predictor6_MIPSdspR2(const uint32_t* const left,
+                                     const uint32_t* const top) {
+  return Average2(*left, top[-1]);
 }
 
-static uint32_t Predictor7_MIPSdspR2(uint32_t left, const uint32_t* const top) {
-  return Average2(left, top[0]);
+static uint32_t Predictor7_MIPSdspR2(const uint32_t* const left,
+                                     const uint32_t* const top) {
+  return Average2(*left, top[0]);
 }
 
-static uint32_t Predictor8_MIPSdspR2(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor8_MIPSdspR2(const uint32_t* const left,
+                                     const uint32_t* const top) {
   (void)left;
   return Average2(top[-1], top[0]);
 }
 
-static uint32_t Predictor9_MIPSdspR2(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor9_MIPSdspR2(const uint32_t* const left,
+                                     const uint32_t* const top) {
   (void)left;
   return Average2(top[0], top[1]);
 }
 
-static uint32_t Predictor10_MIPSdspR2(uint32_t left,
+static uint32_t Predictor10_MIPSdspR2(const uint32_t* const left,
                                       const uint32_t* const top) {
-  return Average4(left, top[-1], top[0], top[1]);
+  return Average4(*left, top[-1], top[0], top[1]);
 }
 
-static uint32_t Predictor11_MIPSdspR2(uint32_t left,
+static uint32_t Predictor11_MIPSdspR2(const uint32_t* const left,
                                       const uint32_t* const top) {
-  return Select(top[0], left, top[-1]);
+  return Select(top[0], *left, top[-1]);
 }
 
-static uint32_t Predictor12_MIPSdspR2(uint32_t left,
+static uint32_t Predictor12_MIPSdspR2(const uint32_t* const left,
                                       const uint32_t* const top) {
-  return ClampedAddSubtractFull(left, top[0], top[-1]);
+  return ClampedAddSubtractFull(*left, top[0], top[-1]);
 }
 
-static uint32_t Predictor13_MIPSdspR2(uint32_t left,
+static uint32_t Predictor13_MIPSdspR2(const uint32_t* const left,
                                       const uint32_t* const top) {
-  return ClampedAddSubtractHalf(left, top[0], top[-1]);
+  return ClampedAddSubtractHalf(*left, top[0], top[-1]);
 }
 
 // Add green to blue and red channels (i.e. perform the inverse transform of

3rdparty/libwebp/src/dsp/lossless_neon.c

@@ -188,17 +188,21 @@ static WEBP_INLINE uint32_t Average3_NEON(uint32_t a0, uint32_t a1,
   return avg;
 }
 
-static uint32_t Predictor5_NEON(uint32_t left, const uint32_t* const top) {
-  return Average3_NEON(left, top[0], top[1]);
+static uint32_t Predictor5_NEON(const uint32_t* const left,
+                                const uint32_t* const top) {
+  return Average3_NEON(*left, top[0], top[1]);
 }
-static uint32_t Predictor6_NEON(uint32_t left, const uint32_t* const top) {
-  return Average2_NEON(left, top[-1]);
+static uint32_t Predictor6_NEON(const uint32_t* const left,
+                                const uint32_t* const top) {
+  return Average2_NEON(*left, top[-1]);
 }
-static uint32_t Predictor7_NEON(uint32_t left, const uint32_t* const top) {
-  return Average2_NEON(left, top[0]);
+static uint32_t Predictor7_NEON(const uint32_t* const left,
+                                const uint32_t* const top) {
+  return Average2_NEON(*left, top[0]);
 }
-static uint32_t Predictor13_NEON(uint32_t left, const uint32_t* const top) {
-  return ClampedAddSubtractHalf_NEON(left, top[0], top[-1]);
+static uint32_t Predictor13_NEON(const uint32_t* const left,
+                                 const uint32_t* const top) {
+  return ClampedAddSubtractHalf_NEON(*left, top[0], top[-1]);
 }
 
 // Batch versions of those functions.
@@ -494,7 +498,7 @@ static void PredictorAdd13_NEON(const uint32_t* in, const uint32_t* upper,
 
 // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
 // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
 #define USE_VTBLQ
 #endif

3rdparty/libwebp/src/dsp/lossless_sse2.c

@@ -18,7 +18,6 @@
 #include "src/dsp/common_sse2.h"
 #include "src/dsp/lossless.h"
 #include "src/dsp/lossless_common.h"
-#include <assert.h>
 #include <emmintrin.h>
 
 //------------------------------------------------------------------------------
@@ -28,23 +27,22 @@ static WEBP_INLINE uint32_t ClampedAddSubtractFull_SSE2(uint32_t c0,
                                                         uint32_t c1,
                                                         uint32_t c2) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
-  const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
-  const __m128i C2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero);
+  const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c0), zero);
+  const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c1), zero);
+  const __m128i C2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c2), zero);
   const __m128i V1 = _mm_add_epi16(C0, C1);
   const __m128i V2 = _mm_sub_epi16(V1, C2);
   const __m128i b = _mm_packus_epi16(V2, V2);
-  const uint32_t output = _mm_cvtsi128_si32(b);
-  return output;
+  return (uint32_t)_mm_cvtsi128_si32(b);
 }
 
 static WEBP_INLINE uint32_t ClampedAddSubtractHalf_SSE2(uint32_t c0,
                                                         uint32_t c1,
                                                         uint32_t c2) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
-  const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
-  const __m128i B0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero);
+  const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c0), zero);
+  const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c1), zero);
+  const __m128i B0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c2), zero);
   const __m128i avg = _mm_add_epi16(C1, C0);
   const __m128i A0 = _mm_srli_epi16(avg, 1);
   const __m128i A1 = _mm_sub_epi16(A0, B0);
@@ -53,16 +51,15 @@ static WEBP_INLINE uint32_t ClampedAddSubtractHalf_SSE2(uint32_t c0,
   const __m128i A3 = _mm_srai_epi16(A2, 1);
   const __m128i A4 = _mm_add_epi16(A0, A3);
   const __m128i A5 = _mm_packus_epi16(A4, A4);
-  const uint32_t output = _mm_cvtsi128_si32(A5);
-  return output;
+  return (uint32_t)_mm_cvtsi128_si32(A5);
 }
 
 static WEBP_INLINE uint32_t Select_SSE2(uint32_t a, uint32_t b, uint32_t c) {
   int pa_minus_pb;
   const __m128i zero = _mm_setzero_si128();
-  const __m128i A0 = _mm_cvtsi32_si128(a);
-  const __m128i B0 = _mm_cvtsi32_si128(b);
-  const __m128i C0 = _mm_cvtsi32_si128(c);
+  const __m128i A0 = _mm_cvtsi32_si128((int)a);
+  const __m128i B0 = _mm_cvtsi32_si128((int)b);
+  const __m128i C0 = _mm_cvtsi32_si128((int)c);
   const __m128i AC0 = _mm_subs_epu8(A0, C0);
   const __m128i CA0 = _mm_subs_epu8(C0, A0);
   const __m128i BC0 = _mm_subs_epu8(B0, C0);
@@ -95,8 +92,8 @@ static WEBP_INLINE void Average2_uint32_SSE2(const uint32_t a0,
                                              __m128i* const avg) {
   // (a + b) >> 1 = ((a + b + 1) >> 1) - ((a ^ b) & 1)
   const __m128i ones = _mm_set1_epi8(1);
-  const __m128i A0 = _mm_cvtsi32_si128(a0);
-  const __m128i A1 = _mm_cvtsi32_si128(a1);
+  const __m128i A0 = _mm_cvtsi32_si128((int)a0);
+  const __m128i A1 = _mm_cvtsi32_si128((int)a1);
   const __m128i avg1 = _mm_avg_epu8(A0, A1);
   const __m128i one = _mm_and_si128(_mm_xor_si128(A0, A1), ones);
   *avg = _mm_sub_epi8(avg1, one);
@@ -104,8 +101,8 @@ static WEBP_INLINE void Average2_uint32_SSE2(const uint32_t a0,
 
 static WEBP_INLINE __m128i Average2_uint32_16_SSE2(uint32_t a0, uint32_t a1) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a0), zero);
-  const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
+  const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a0), zero);
+  const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a1), zero);
   const __m128i sum = _mm_add_epi16(A1, A0);
   return _mm_srli_epi16(sum, 1);
 }
@@ -113,19 +110,18 @@ static WEBP_INLINE __m128i Average2_uint32_16_SSE2(uint32_t a0, uint32_t a1) {
 static WEBP_INLINE uint32_t Average2_SSE2(uint32_t a0, uint32_t a1) {
   __m128i output;
   Average2_uint32_SSE2(a0, a1, &output);
-  return _mm_cvtsi128_si32(output);
+  return (uint32_t)_mm_cvtsi128_si32(output);
 }
 
 static WEBP_INLINE uint32_t Average3_SSE2(uint32_t a0, uint32_t a1,
                                           uint32_t a2) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i avg1 = Average2_uint32_16_SSE2(a0, a2);
-  const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
+  const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a1), zero);
   const __m128i sum = _mm_add_epi16(avg1, A1);
   const __m128i avg2 = _mm_srli_epi16(sum, 1);
   const __m128i A2 = _mm_packus_epi16(avg2, avg2);
-  const uint32_t output = _mm_cvtsi128_si32(A2);
-  return output;
+  return (uint32_t)_mm_cvtsi128_si32(A2);
 }
 
 static WEBP_INLINE uint32_t Average4_SSE2(uint32_t a0, uint32_t a1,
@@ -135,46 +131,54 @@ static WEBP_INLINE uint32_t Average4_SSE2(uint32_t a0, uint32_t a1,
   const __m128i sum = _mm_add_epi16(avg2, avg1);
   const __m128i avg3 = _mm_srli_epi16(sum, 1);
   const __m128i A0 = _mm_packus_epi16(avg3, avg3);
-  const uint32_t output = _mm_cvtsi128_si32(A0);
-  return output;
+  return (uint32_t)_mm_cvtsi128_si32(A0);
 }
 
-static uint32_t Predictor5_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average3_SSE2(left, top[0], top[1]);
+static uint32_t Predictor5_SSE2(const uint32_t* const left,
+                                const uint32_t* const top) {
+  const uint32_t pred = Average3_SSE2(*left, top[0], top[1]);
   return pred;
 }
-static uint32_t Predictor6_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2_SSE2(left, top[-1]);
+static uint32_t Predictor6_SSE2(const uint32_t* const left,
+                                const uint32_t* const top) {
+  const uint32_t pred = Average2_SSE2(*left, top[-1]);
   return pred;
 }
-static uint32_t Predictor7_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average2_SSE2(left, top[0]);
+static uint32_t Predictor7_SSE2(const uint32_t* const left,
+                                const uint32_t* const top) {
+  const uint32_t pred = Average2_SSE2(*left, top[0]);
   return pred;
 }
-static uint32_t Predictor8_SSE2(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor8_SSE2(const uint32_t* const left,
+                                const uint32_t* const top) {
   const uint32_t pred = Average2_SSE2(top[-1], top[0]);
   (void)left;
   return pred;
 }
-static uint32_t Predictor9_SSE2(uint32_t left, const uint32_t* const top) {
+static uint32_t Predictor9_SSE2(const uint32_t* const left,
+                                const uint32_t* const top) {
   const uint32_t pred = Average2_SSE2(top[0], top[1]);
   (void)left;
   return pred;
 }
-static uint32_t Predictor10_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Average4_SSE2(left, top[-1], top[0], top[1]);
+static uint32_t Predictor10_SSE2(const uint32_t* const left,
+                                 const uint32_t* const top) {
+  const uint32_t pred = Average4_SSE2(*left, top[-1], top[0], top[1]);
   return pred;
 }
-static uint32_t Predictor11_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = Select_SSE2(top[0], left, top[-1]);
+static uint32_t Predictor11_SSE2(const uint32_t* const left,
+                                 const uint32_t* const top) {
+  const uint32_t pred = Select_SSE2(top[0], *left, top[-1]);
   return pred;
 }
-static uint32_t Predictor12_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = ClampedAddSubtractFull_SSE2(left, top[0], top[-1]);
+static uint32_t Predictor12_SSE2(const uint32_t* const left,
+                                 const uint32_t* const top) {
+  const uint32_t pred = ClampedAddSubtractFull_SSE2(*left, top[0], top[-1]);
   return pred;
 }
-static uint32_t Predictor13_SSE2(uint32_t left, const uint32_t* const top) {
-  const uint32_t pred = ClampedAddSubtractHalf_SSE2(left, top[0], top[-1]);
+static uint32_t Predictor13_SSE2(const uint32_t* const left,
+                                 const uint32_t* const top) {
+  const uint32_t pred = ClampedAddSubtractHalf_SSE2(*left, top[0], top[-1]);
   return pred;
 }
 
@@ -184,7 +188,7 @@ static uint32_t Predictor13_SSE2(uint32_t left, const uint32_t* const top) {
 static void PredictorAdd0_SSE2(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) {
   int i;
-  const __m128i black = _mm_set1_epi32(ARGB_BLACK);
+  const __m128i black = _mm_set1_epi32((int)ARGB_BLACK);
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
     const __m128i res = _mm_add_epi8(src, black);
@@ -200,7 +204,7 @@ static void PredictorAdd0_SSE2(const uint32_t* in, const uint32_t* upper,
 static void PredictorAdd1_SSE2(const uint32_t* in, const uint32_t* upper,
                                int num_pixels, uint32_t* out) {
   int i;
-  __m128i prev = _mm_set1_epi32(out[-1]);
+  __m128i prev = _mm_set1_epi32((int)out[-1]);
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     // a | b | c | d
     const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
@@ -277,12 +281,12 @@ GENERATE_PREDICTOR_2(9, upper[i + 1])
 #undef GENERATE_PREDICTOR_2
 
 // Predictor10: average of (average of (L,TL), average of (T, TR)).
-#define DO_PRED10(OUT) do {               \
-  __m128i avgLTL, avg;                    \
-  Average2_m128i(&L, &TL, &avgLTL);       \
-  Average2_m128i(&avgTTR, &avgLTL, &avg); \
-  L = _mm_add_epi8(avg, src);             \
-  out[i + (OUT)] = _mm_cvtsi128_si32(L);  \
+#define DO_PRED10(OUT) do {                         \
+  __m128i avgLTL, avg;                              \
+  Average2_m128i(&L, &TL, &avgLTL);                 \
+  Average2_m128i(&avgTTR, &avgLTL, &avg);           \
+  L = _mm_add_epi8(avg, src);                       \
+  out[i + (OUT)] = (uint32_t)_mm_cvtsi128_si32(L);  \
 } while (0)
 
 #define DO_PRED10_SHIFT do {                                  \
@@ -295,7 +299,7 @@ GENERATE_PREDICTOR_2(9, upper[i + 1])
 static void PredictorAdd10_SSE2(const uint32_t* in, const uint32_t* upper,
                                 int num_pixels, uint32_t* out) {
   int i;
-  __m128i L = _mm_cvtsi32_si128(out[-1]);
+  __m128i L = _mm_cvtsi32_si128((int)out[-1]);
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
     __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
@@ -328,7 +332,7 @@ static void PredictorAdd10_SSE2(const uint32_t* in, const uint32_t* upper,
   const __m128i B = _mm_andnot_si128(mask, T);                         \
   const __m128i pred = _mm_or_si128(A, B); /* pred = (pa > b)? L : T*/ \
   L = _mm_add_epi8(src, pred);                                         \
-  out[i + (OUT)] = _mm_cvtsi128_si32(L);                               \
+  out[i + (OUT)] = (uint32_t)_mm_cvtsi128_si32(L);                     \
 } while (0)
 
 #define DO_PRED11_SHIFT do {                                \
@@ -343,7 +347,7 @@ static void PredictorAdd11_SSE2(const uint32_t* in, const uint32_t* upper,
                                 int num_pixels, uint32_t* out) {
   int i;
   __m128i pa;
-  __m128i L = _mm_cvtsi32_si128(out[-1]);
+  __m128i L = _mm_cvtsi32_si128((int)out[-1]);
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     __m128i T = _mm_loadu_si128((const __m128i*)&upper[i]);
     __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
@@ -376,12 +380,12 @@ static void PredictorAdd11_SSE2(const uint32_t* in, const uint32_t* upper,
 #undef DO_PRED11_SHIFT
 
 // Predictor12: ClampedAddSubtractFull.
-#define DO_PRED12(DIFF, LANE, OUT) do {            \
-  const __m128i all = _mm_add_epi16(L, (DIFF));    \
-  const __m128i alls = _mm_packus_epi16(all, all); \
-  const __m128i res = _mm_add_epi8(src, alls);     \
-  out[i + (OUT)] = _mm_cvtsi128_si32(res);         \
-  L = _mm_unpacklo_epi8(res, zero);                \
+#define DO_PRED12(DIFF, LANE, OUT) do {              \
+  const __m128i all = _mm_add_epi16(L, (DIFF));      \
+  const __m128i alls = _mm_packus_epi16(all, all);   \
+  const __m128i res = _mm_add_epi8(src, alls);       \
+  out[i + (OUT)] = (uint32_t)_mm_cvtsi128_si32(res); \
+  L = _mm_unpacklo_epi8(res, zero);                  \
 } while (0)
 
 #define DO_PRED12_SHIFT(DIFF, LANE) do {                    \
@@ -394,7 +398,7 @@ static void PredictorAdd12_SSE2(const uint32_t* in, const uint32_t* upper,
                                 int num_pixels, uint32_t* out) {
   int i;
   const __m128i zero = _mm_setzero_si128();
-  const __m128i L8 = _mm_cvtsi32_si128(out[-1]);
+  const __m128i L8 = _mm_cvtsi32_si128((int)out[-1]);
   __m128i L = _mm_unpacklo_epi8(L8, zero);
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     // Load 4 pixels at a time.
@@ -460,7 +464,7 @@ static void TransformColorInverse_SSE2(const VP8LMultipliers* const m,
   const __m128i mults_b2 = MK_CST_16(CST(red_to_blue_), 0);
 #undef MK_CST_16
 #undef CST
-  const __m128i mask_ag = _mm_set1_epi32(0xff00ff00);  // alpha-green masks
+  const __m128i mask_ag = _mm_set1_epi32((int)0xff00ff00);  // alpha-green masks
   int i;
   for (i = 0; i + 4 <= num_pixels; i += 4) {
     const __m128i in = _mm_loadu_si128((const __m128i*)&src[i]); // argb
@@ -524,7 +528,7 @@ static void ConvertBGRAToRGB_SSE2(const uint32_t* src, int num_pixels,
 
 static void ConvertBGRAToRGBA_SSE2(const uint32_t* src,
                                    int num_pixels, uint8_t* dst) {
-  const __m128i red_blue_mask = _mm_set1_epi32(0x00ff00ffu);
+  const __m128i red_blue_mask = _mm_set1_epi32(0x00ff00ff);
   const __m128i* in = (const __m128i*)src;
   __m128i* out = (__m128i*)dst;
   while (num_pixels >= 8) {
@@ -553,7 +557,7 @@ static void ConvertBGRAToRGBA_SSE2(const uint32_t* src,
 static void ConvertBGRAToRGBA4444_SSE2(const uint32_t* src,
                                        int num_pixels, uint8_t* dst) {
   const __m128i mask_0x0f = _mm_set1_epi8(0x0f);
-  const __m128i mask_0xf0 = _mm_set1_epi8(0xf0);
+  const __m128i mask_0xf0 = _mm_set1_epi8((char)0xf0);
   const __m128i* in = (const __m128i*)src;
   __m128i* out = (__m128i*)dst;
   while (num_pixels >= 8) {
@@ -588,8 +592,8 @@ static void ConvertBGRAToRGBA4444_SSE2(const uint32_t* src,
 
 static void ConvertBGRAToRGB565_SSE2(const uint32_t* src,
                                      int num_pixels, uint8_t* dst) {
-  const __m128i mask_0xe0 = _mm_set1_epi8(0xe0);
-  const __m128i mask_0xf8 = _mm_set1_epi8(0xf8);
+  const __m128i mask_0xe0 = _mm_set1_epi8((char)0xe0);
+  const __m128i mask_0xf8 = _mm_set1_epi8((char)0xf8);
   const __m128i mask_0x07 = _mm_set1_epi8(0x07);
   const __m128i* in = (const __m128i*)src;
   __m128i* out = (__m128i*)dst;

3rdparty/libwebp/src/dsp/lossless_sse41.c

@@ -0,0 +1,133 @@
+// Copyright 2021 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// SSE41 variant of methods for lossless decoder
+
+#include "src/dsp/dsp.h"
+
+#if defined(WEBP_USE_SSE41)
+
+#include "src/dsp/common_sse41.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
+
+//------------------------------------------------------------------------------
+// Color-space conversion functions
+
+static void TransformColorInverse_SSE41(const VP8LMultipliers* const m,
+                                        const uint32_t* const src,
+                                        int num_pixels, uint32_t* dst) {
+// sign-extended multiplying constants, pre-shifted by 5.
+#define CST(X)  (((int16_t)(m->X << 8)) >> 5)   // sign-extend
+  const __m128i mults_rb =
+      _mm_set1_epi32((int)((uint32_t)CST(green_to_red_) << 16 |
+                           (CST(green_to_blue_) & 0xffff)));
+  const __m128i mults_b2 = _mm_set1_epi32(CST(red_to_blue_));
+#undef CST
+  const __m128i mask_ag = _mm_set1_epi32((int)0xff00ff00);
+  const __m128i perm1 = _mm_setr_epi8(-1, 1, -1, 1, -1, 5, -1, 5,
+                                      -1, 9, -1, 9, -1, 13, -1, 13);
+  const __m128i perm2 = _mm_setr_epi8(-1, 2, -1, -1, -1, 6, -1, -1,
+                                      -1, 10, -1, -1, -1, 14, -1, -1);
+  int i;
+  for (i = 0; i + 4 <= num_pixels; i += 4) {
+    const __m128i A = _mm_loadu_si128((const __m128i*)(src + i));
+    const __m128i B = _mm_shuffle_epi8(A, perm1); // argb -> g0g0
+    const __m128i C = _mm_mulhi_epi16(B, mults_rb);
+    const __m128i D = _mm_add_epi8(A, C);
+    const __m128i E = _mm_shuffle_epi8(D, perm2);
+    const __m128i F = _mm_mulhi_epi16(E, mults_b2);
+    const __m128i G = _mm_add_epi8(D, F);
+    const __m128i out = _mm_blendv_epi8(G, A, mask_ag);
+    _mm_storeu_si128((__m128i*)&dst[i], out);
+  }
+  // Fall-back to C-version for left-overs.
+  if (i != num_pixels) {
+    VP8LTransformColorInverse_C(m, src + i, num_pixels - i, dst + i);
+  }
+}
+
+//------------------------------------------------------------------------------
+
+#define ARGB_TO_RGB_SSE41 do {                        \
+  while (num_pixels >= 16) {                          \
+    const __m128i in0 = _mm_loadu_si128(in + 0);      \
+    const __m128i in1 = _mm_loadu_si128(in + 1);      \
+    const __m128i in2 = _mm_loadu_si128(in + 2);      \
+    const __m128i in3 = _mm_loadu_si128(in + 3);      \
+    const __m128i a0 = _mm_shuffle_epi8(in0, perm0);  \
+    const __m128i a1 = _mm_shuffle_epi8(in1, perm1);  \
+    const __m128i a2 = _mm_shuffle_epi8(in2, perm2);  \
+    const __m128i a3 = _mm_shuffle_epi8(in3, perm3);  \
+    const __m128i b0 = _mm_blend_epi16(a0, a1, 0xc0); \
+    const __m128i b1 = _mm_blend_epi16(a1, a2, 0xf0); \
+    const __m128i b2 = _mm_blend_epi16(a2, a3, 0xfc); \
+    _mm_storeu_si128(out + 0, b0);                    \
+    _mm_storeu_si128(out + 1, b1);                    \
+    _mm_storeu_si128(out + 2, b2);                    \
+    in += 4;                                          \
+    out += 3;                                         \
+    num_pixels -= 16;                                 \
+  }                                                   \
+} while (0)
+
+static void ConvertBGRAToRGB_SSE41(const uint32_t* src, int num_pixels,
+                                   uint8_t* dst) {
+  const __m128i* in = (const __m128i*)src;
+  __m128i* out = (__m128i*)dst;
+  const __m128i perm0 = _mm_setr_epi8(2, 1, 0, 6, 5, 4, 10, 9,
+                                      8, 14, 13, 12, -1, -1, -1, -1);
+  const __m128i perm1 = _mm_shuffle_epi32(perm0, 0x39);
+  const __m128i perm2 = _mm_shuffle_epi32(perm0, 0x4e);
+  const __m128i perm3 = _mm_shuffle_epi32(perm0, 0x93);
+
+  ARGB_TO_RGB_SSE41;
+
+  // left-overs
+  if (num_pixels > 0) {
+    VP8LConvertBGRAToRGB_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
+  }
+}
+
+static void ConvertBGRAToBGR_SSE41(const uint32_t* src,
+                                   int num_pixels, uint8_t* dst) {
+  const __m128i* in = (const __m128i*)src;
+  __m128i* out = (__m128i*)dst;
+  const __m128i perm0 = _mm_setr_epi8(0, 1, 2, 4, 5, 6, 8, 9, 10,
+                                      12, 13, 14, -1, -1, -1, -1);
+  const __m128i perm1 = _mm_shuffle_epi32(perm0, 0x39);
+  const __m128i perm2 = _mm_shuffle_epi32(perm0, 0x4e);
+  const __m128i perm3 = _mm_shuffle_epi32(perm0, 0x93);
+
+  ARGB_TO_RGB_SSE41;
+
+  // left-overs
+  if (num_pixels > 0) {
+    VP8LConvertBGRAToBGR_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
+  }
+}
+
+#undef ARGB_TO_RGB_SSE41
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void VP8LDspInitSSE41(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitSSE41(void) {
+  VP8LTransformColorInverse = TransformColorInverse_SSE41;
+  VP8LConvertBGRAToRGB = ConvertBGRAToRGB_SSE41;
+  VP8LConvertBGRAToBGR = ConvertBGRAToBGR_SSE41;
+}
+
+#else  // !WEBP_USE_SSE41
+
+WEBP_DSP_INIT_STUB(VP8LDspInitSSE41)
+
+#endif  // WEBP_USE_SSE41

3rdparty/libwebp/src/dsp/msa_macro.h

@@ -14,6 +14,10 @@
 #ifndef WEBP_DSP_MSA_MACRO_H_
 #define WEBP_DSP_MSA_MACRO_H_
 
+#include "src/dsp/dsp.h"
+
+#if defined(WEBP_USE_MSA)
+
 #include <stdint.h>
 #include <msa.h>
 
@@ -69,27 +73,25 @@
 #define ST_UW(...) ST_W(v4u32, __VA_ARGS__)
 #define ST_SW(...) ST_W(v4i32, __VA_ARGS__)
 
-#define MSA_LOAD_FUNC(TYPE, INSTR, FUNC_NAME)             \
-  static inline TYPE FUNC_NAME(const void* const psrc) {  \
-    const uint8_t* const psrc_m = (const uint8_t*)psrc;   \
-    TYPE val_m;                                           \
-    __asm__ volatile (                                        \
-      "" #INSTR " %[val_m], %[psrc_m]  \n\t"              \
-      : [val_m] "=r" (val_m)                              \
-      : [psrc_m] "m" (*psrc_m));                          \
-    return val_m;                                         \
+#define MSA_LOAD_FUNC(TYPE, INSTR, FUNC_NAME)               \
+  static inline TYPE FUNC_NAME(const void* const psrc) {    \
+    const uint8_t* const psrc_m = (const uint8_t*)psrc;     \
+    TYPE val_m;                                             \
+    __asm__ volatile("" #INSTR " %[val_m], %[psrc_m]  \n\t" \
+                     : [val_m] "=r"(val_m)                  \
+                     : [psrc_m] "m"(*psrc_m));              \
+    return val_m;                                           \
   }
 
 #define MSA_LOAD(psrc, FUNC_NAME)  FUNC_NAME(psrc)
 
-#define MSA_STORE_FUNC(TYPE, INSTR, FUNC_NAME)               \
-  static inline void FUNC_NAME(TYPE val, void* const pdst) { \
-    uint8_t* const pdst_m = (uint8_t*)pdst;                  \
-    TYPE val_m = val;                                        \
-    __asm__ volatile (                                           \
-      " " #INSTR "  %[val_m],  %[pdst_m]  \n\t"              \
-      : [pdst_m] "=m" (*pdst_m)                              \
-      : [val_m] "r" (val_m));                                \
+#define MSA_STORE_FUNC(TYPE, INSTR, FUNC_NAME)                 \
+  static inline void FUNC_NAME(TYPE val, void* const pdst) {   \
+    uint8_t* const pdst_m = (uint8_t*)pdst;                    \
+    TYPE val_m = val;                                          \
+    __asm__ volatile(" " #INSTR "  %[val_m],  %[pdst_m]  \n\t" \
+                     : [pdst_m] "=m"(*pdst_m)                  \
+                     : [val_m] "r"(val_m));                    \
   }
 
 #define MSA_STORE(val, pdst, FUNC_NAME)  FUNC_NAME(val, pdst)
@@ -1389,4 +1391,5 @@ static WEBP_INLINE uint32_t func_hadd_uh_u32(v8u16 in) {
 } while (0)
 #define AVER_UB2_UB(...) AVER_UB2(v16u8, __VA_ARGS__)
 
+#endif  // WEBP_USE_MSA
 #endif  // WEBP_DSP_MSA_MACRO_H_

3rdparty/libwebp/src/dsp/neon.h

@@ -12,14 +12,16 @@
 #ifndef WEBP_DSP_NEON_H_
 #define WEBP_DSP_NEON_H_
 
-#include <arm_neon.h>
-
 #include "src/dsp/dsp.h"
 
+#if defined(WEBP_USE_NEON)
+
+#include <arm_neon.h>
+
 // Right now, some intrinsics functions seem slower, so we disable them
 // everywhere except newer clang/gcc or aarch64 where the inline assembly is
 // incompatible.
-#if LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,9) || defined(__aarch64__)
+#if LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 9) || WEBP_AARCH64
 #define WEBP_USE_INTRINSICS   // use intrinsics when possible
 #endif
 
@@ -44,7 +46,7 @@
 // if using intrinsics, this flag avoids some functions that make gcc-4.6.3
 // crash ("internal compiler error: in immed_double_const, at emit-rtl.").
 // (probably similar to gcc.gnu.org/bugzilla/show_bug.cgi?id=48183)
-#if !(LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
+#if !(LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 8) || WEBP_AARCH64)
 #define WORK_AROUND_GCC
 #endif
 
@@ -98,4 +100,5 @@ static WEBP_INLINE int32x4x4_t Transpose4x4_NEON(const int32x4x4_t rows) {
 } while (0)
 #endif
 
+#endif  // WEBP_USE_NEON
 #endif  // WEBP_DSP_NEON_H_

3rdparty/libwebp/src/dsp/quant.h

@@ -21,18 +21,24 @@
 
 #define IsFlat IsFlat_NEON
 
-static uint32x2_t horizontal_add_uint32x4(const uint32x4_t a) {
+static uint32_t horizontal_add_uint32x4(const uint32x4_t a) {
+#if WEBP_AARCH64
+  return vaddvq_u32(a);
+#else
   const uint64x2_t b = vpaddlq_u32(a);
-  return vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
-                  vreinterpret_u32_u64(vget_high_u64(b)));
+  const uint32x2_t c = vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
+                                vreinterpret_u32_u64(vget_high_u64(b)));
+  return vget_lane_u32(c, 0);
+#endif
 }
 
 static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
                               int thresh) {
   const int16x8_t tst_ones = vdupq_n_s16(-1);
   uint32x4_t sum = vdupq_n_u32(0);
+  int i;
 
-  for (int i = 0; i < num_blocks; ++i) {
+  for (i = 0; i < num_blocks; ++i) {
     // Set DC to zero.
     const int16x8_t a_0 = vsetq_lane_s16(0, vld1q_s16(levels), 0);
     const int16x8_t a_1 = vld1q_s16(levels + 8);
@@ -45,7 +51,7 @@ static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
 
     levels += 16;
   }
-  return thresh >= (int32_t)vget_lane_u32(horizontal_add_uint32x4(sum), 0);
+  return thresh >= (int)horizontal_add_uint32x4(sum);
 }
 
 #else

3rdparty/libwebp/src/dsp/rescaler.c

@@ -38,8 +38,9 @@ void WebPRescalerImportRowExpand_C(WebPRescaler* const wrk,
     int x_out = channel;
     // simple bilinear interpolation
     int accum = wrk->x_add;
-    int left = src[x_in];
-    int right = (wrk->src_width > 1) ? src[x_in + x_stride] : left;
+    rescaler_t left = (rescaler_t)src[x_in];
+    rescaler_t right =
+        (wrk->src_width > 1) ? (rescaler_t)src[x_in + x_stride] : left;
     x_in += x_stride;
     while (1) {
       wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum;
@@ -50,7 +51,7 @@ void WebPRescalerImportRowExpand_C(WebPRescaler* const wrk,
         left = right;
         x_in += x_stride;
         assert(x_in < wrk->src_width * x_stride);
-        right = src[x_in];
+        right = (rescaler_t)src[x_in];
         accum += wrk->x_add;
       }
     }
@@ -196,6 +197,7 @@ WebPRescalerImportRowFunc WebPRescalerImportRowShrink;
 WebPRescalerExportRowFunc WebPRescalerExportRowExpand;
 WebPRescalerExportRowFunc WebPRescalerExportRowShrink;
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void WebPRescalerDspInitSSE2(void);
 extern void WebPRescalerDspInitMIPS32(void);
 extern void WebPRescalerDspInitMIPSdspR2(void);
@@ -213,7 +215,7 @@ WEBP_DSP_INIT_FUNC(WebPRescalerDspInit) {
   WebPRescalerImportRowShrink = WebPRescalerImportRowShrink_C;
 
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       WebPRescalerDspInitSSE2();
     }
@@ -235,7 +237,7 @@ WEBP_DSP_INIT_FUNC(WebPRescalerDspInit) {
 #endif
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     WebPRescalerDspInitNEON();

3rdparty/libwebp/src/dsp/rescaler_sse2.c

@@ -85,7 +85,7 @@ static void RescalerImportRowExpand_SSE2(WebPRescaler* const wrk,
       const __m128i mult = _mm_cvtsi32_si128(((x_add - accum) << 16) | accum);
       const __m128i out = _mm_madd_epi16(cur_pixels, mult);
       assert(sizeof(*frow) == sizeof(uint32_t));
-      WebPUint32ToMem((uint8_t*)frow, _mm_cvtsi128_si32(out));
+      WebPInt32ToMem((uint8_t*)frow, _mm_cvtsi128_si32(out));
       frow += 1;
       if (frow >= frow_end) break;
       accum -= wrk->x_sub;
@@ -132,7 +132,7 @@ static void RescalerImportRowShrink_SSE2(WebPRescaler* const wrk,
     __m128i base = zero;
     accum += wrk->x_add;
     while (accum > 0) {
-      const __m128i A = _mm_cvtsi32_si128(WebPMemToUint32(src));
+      const __m128i A = _mm_cvtsi32_si128(WebPMemToInt32(src));
       src += 4;
       base = _mm_unpacklo_epi8(A, zero);
       // To avoid overflow, we need: base * x_add / x_sub < 32768
@@ -198,7 +198,7 @@ static WEBP_INLINE void ProcessRow_SSE2(const __m128i* const A0,
                                         const __m128i* const mult,
                                         uint8_t* const dst) {
   const __m128i rounder = _mm_set_epi32(0, ROUNDER, 0, ROUNDER);
-  const __m128i mask = _mm_set_epi32(0xffffffffu, 0, 0xffffffffu, 0);
+  const __m128i mask = _mm_set_epi32(~0, 0, ~0, 0);
   const __m128i B0 = _mm_mul_epu32(*A0, *mult);
   const __m128i B1 = _mm_mul_epu32(*A1, *mult);
   const __m128i B2 = _mm_mul_epu32(*A2, *mult);

3rdparty/libwebp/src/dsp/ssim.c

@@ -137,6 +137,7 @@ VP8SSIMGetClippedFunc VP8SSIMGetClipped;
 VP8AccumulateSSEFunc VP8AccumulateSSE;
 #endif
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void VP8SSIMDspInitSSE2(void);
 
 WEBP_DSP_INIT_FUNC(VP8SSIMDspInit) {
@@ -150,7 +151,7 @@ WEBP_DSP_INIT_FUNC(VP8SSIMDspInit) {
 #endif
 
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8SSIMDspInitSSE2();
     }

3rdparty/libwebp/src/dsp/upsampling.c

@@ -215,6 +215,7 @@ static void EmptyYuv444Func(const uint8_t* y,
 
 WebPYUV444Converter WebPYUV444Converters[MODE_LAST];
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void WebPInitYUV444ConvertersMIPSdspR2(void);
 extern void WebPInitYUV444ConvertersSSE2(void);
 extern void WebPInitYUV444ConvertersSSE41(void);
@@ -233,12 +234,12 @@ WEBP_DSP_INIT_FUNC(WebPInitYUV444Converters) {
   WebPYUV444Converters[MODE_rgbA_4444] = WebPYuv444ToRgba4444_C;
 
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       WebPInitYUV444ConvertersSSE2();
     }
 #endif
-#if defined(WEBP_USE_SSE41)
+#if defined(WEBP_HAVE_SSE41)
     if (VP8GetCPUInfo(kSSE4_1)) {
       WebPInitYUV444ConvertersSSE41();
     }
@@ -278,12 +279,12 @@ WEBP_DSP_INIT_FUNC(WebPInitUpsamplers) {
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       WebPInitUpsamplersSSE2();
     }
 #endif
-#if defined(WEBP_USE_SSE41)
+#if defined(WEBP_HAVE_SSE41)
     if (VP8GetCPUInfo(kSSE4_1)) {
       WebPInitUpsamplersSSE41();
     }
@@ -300,7 +301,7 @@ WEBP_DSP_INIT_FUNC(WebPInitUpsamplers) {
 #endif
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     WebPInitUpsamplersNEON();

3rdparty/libwebp/src/dsp/upsampling_neon.c

@@ -111,7 +111,7 @@ static const int16_t kCoeffs1[4] = { 19077, 26149, 6419, 13320 };
   vst4_u8(out, v255_r_g_b);                                             \
 } while (0)
 
-#if !defined(WEBP_SWAP_16BIT_CSP)
+#if (WEBP_SWAP_16BIT_CSP == 0)
 #define ZIP_U8(lo, hi) vzip_u8((lo), (hi))
 #else
 #define ZIP_U8(lo, hi) vzip_u8((hi), (lo))

3rdparty/libwebp/src/dsp/upsampling_sse2.c

@@ -121,7 +121,7 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
   int uv_pos, pos;                                                             \
   /* 16byte-aligned array to cache reconstructed u and v */                    \
   uint8_t uv_buf[14 * 32 + 15] = { 0 };                                        \
-  uint8_t* const r_u = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~15);             \
+  uint8_t* const r_u = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~(uintptr_t)15);  \
   uint8_t* const r_v = r_u + 32;                                               \
                                                                                \
   assert(top_y != NULL);                                                       \

3rdparty/libwebp/src/dsp/yuv.c

@@ -70,6 +70,7 @@ void WebPSamplerProcessPlane(const uint8_t* y, int y_stride,
 
 WebPSamplerRowFunc WebPSamplers[MODE_LAST];
 
+extern VP8CPUInfo VP8GetCPUInfo;
 extern void WebPInitSamplersSSE2(void);
 extern void WebPInitSamplersSSE41(void);
 extern void WebPInitSamplersMIPS32(void);
@@ -90,16 +91,16 @@ WEBP_DSP_INIT_FUNC(WebPInitSamplers) {
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       WebPInitSamplersSSE2();
     }
-#endif  // WEBP_USE_SSE2
-#if defined(WEBP_USE_SSE41)
+#endif  // WEBP_HAVE_SSE2
+#if defined(WEBP_HAVE_SSE41)
     if (VP8GetCPUInfo(kSSE4_1)) {
       WebPInitSamplersSSE41();
     }
-#endif  // WEBP_USE_SSE41
+#endif  // WEBP_HAVE_SSE41
 #if defined(WEBP_USE_MIPS32)
     if (VP8GetCPUInfo(kMIPS32)) {
       WebPInitSamplersMIPS32();
@@ -194,50 +195,6 @@ void WebPConvertRGBA32ToUV_C(const uint16_t* rgb,
 
 //-----------------------------------------------------------------------------
 
-#if !WEBP_NEON_OMIT_C_CODE
-#define MAX_Y ((1 << 10) - 1)    // 10b precision over 16b-arithmetic
-static uint16_t clip_y(int v) {
-  return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
-}
-
-static uint64_t SharpYUVUpdateY_C(const uint16_t* ref, const uint16_t* src,
-                                  uint16_t* dst, int len) {
-  uint64_t diff = 0;
-  int i;
-  for (i = 0; i < len; ++i) {
-    const int diff_y = ref[i] - src[i];
-    const int new_y = (int)dst[i] + diff_y;
-    dst[i] = clip_y(new_y);
-    diff += (uint64_t)abs(diff_y);
-  }
-  return diff;
-}
-
-static void SharpYUVUpdateRGB_C(const int16_t* ref, const int16_t* src,
-                                int16_t* dst, int len) {
-  int i;
-  for (i = 0; i < len; ++i) {
-    const int diff_uv = ref[i] - src[i];
-    dst[i] += diff_uv;
-  }
-}
-
-static void SharpYUVFilterRow_C(const int16_t* A, const int16_t* B, int len,
-                                const uint16_t* best_y, uint16_t* out) {
-  int i;
-  for (i = 0; i < len; ++i, ++A, ++B) {
-    const int v0 = (A[0] * 9 + A[1] * 3 + B[0] * 3 + B[1] + 8) >> 4;
-    const int v1 = (A[1] * 9 + A[0] * 3 + B[1] * 3 + B[0] + 8) >> 4;
-    out[2 * i + 0] = clip_y(best_y[2 * i + 0] + v0);
-    out[2 * i + 1] = clip_y(best_y[2 * i + 1] + v1);
-  }
-}
-#endif  // !WEBP_NEON_OMIT_C_CODE
-
-#undef MAX_Y
-
-//-----------------------------------------------------------------------------
-
 void (*WebPConvertRGB24ToY)(const uint8_t* rgb, uint8_t* y, int width);
 void (*WebPConvertBGR24ToY)(const uint8_t* bgr, uint8_t* y, int width);
 void (*WebPConvertRGBA32ToUV)(const uint16_t* rgb,
@@ -247,18 +204,9 @@ void (*WebPConvertARGBToY)(const uint32_t* argb, uint8_t* y, int width);
 void (*WebPConvertARGBToUV)(const uint32_t* argb, uint8_t* u, uint8_t* v,
                             int src_width, int do_store);
 
-uint64_t (*WebPSharpYUVUpdateY)(const uint16_t* ref, const uint16_t* src,
-                                uint16_t* dst, int len);
-void (*WebPSharpYUVUpdateRGB)(const int16_t* ref, const int16_t* src,
-                              int16_t* dst, int len);
-void (*WebPSharpYUVFilterRow)(const int16_t* A, const int16_t* B, int len,
-                              const uint16_t* best_y, uint16_t* out);
-
 extern void WebPInitConvertARGBToYUVSSE2(void);
 extern void WebPInitConvertARGBToYUVSSE41(void);
 extern void WebPInitConvertARGBToYUVNEON(void);
-extern void WebPInitSharpYUVSSE2(void);
-extern void WebPInitSharpYUVNEON(void);
 
 WEBP_DSP_INIT_FUNC(WebPInitConvertARGBToYUV) {
   WebPConvertARGBToY = ConvertARGBToY_C;
@@ -269,40 +217,29 @@ WEBP_DSP_INIT_FUNC(WebPInitConvertARGBToYUV) {
 
   WebPConvertRGBA32ToUV = WebPConvertRGBA32ToUV_C;
 
-#if !WEBP_NEON_OMIT_C_CODE
-  WebPSharpYUVUpdateY = SharpYUVUpdateY_C;
-  WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_C;
-  WebPSharpYUVFilterRow = SharpYUVFilterRow_C;
-#endif
-
   if (VP8GetCPUInfo != NULL) {
-#if defined(WEBP_USE_SSE2)
+#if defined(WEBP_HAVE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       WebPInitConvertARGBToYUVSSE2();
-      WebPInitSharpYUVSSE2();
     }
-#endif  // WEBP_USE_SSE2
-#if defined(WEBP_USE_SSE41)
+#endif  // WEBP_HAVE_SSE2
+#if defined(WEBP_HAVE_SSE41)
     if (VP8GetCPUInfo(kSSE4_1)) {
       WebPInitConvertARGBToYUVSSE41();
     }
-#endif  // WEBP_USE_SSE41
+#endif  // WEBP_HAVE_SSE41
   }
 
-#if defined(WEBP_USE_NEON)
+#if defined(WEBP_HAVE_NEON)
   if (WEBP_NEON_OMIT_C_CODE ||
       (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
     WebPInitConvertARGBToYUVNEON();
-    WebPInitSharpYUVNEON();
   }
-#endif  // WEBP_USE_NEON
+#endif  // WEBP_HAVE_NEON
 
   assert(WebPConvertARGBToY != NULL);
   assert(WebPConvertARGBToUV != NULL);
   assert(WebPConvertRGB24ToY != NULL);
   assert(WebPConvertBGR24ToY != NULL);
   assert(WebPConvertRGBA32ToUV != NULL);
-  assert(WebPSharpYUVUpdateY != NULL);
-  assert(WebPSharpYUVUpdateRGB != NULL);
-  assert(WebPSharpYUVFilterRow != NULL);
 }

3rdparty/libwebp/src/dsp/yuv.h

@@ -10,7 +10,7 @@
 // inline YUV<->RGB conversion function
 //
 // The exact naming is Y'CbCr, following the ITU-R BT.601 standard.
-// More information at: http://en.wikipedia.org/wiki/YCbCr
+// 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

3rdparty/libwebp/src/dsp/yuv_neon.c

@@ -173,116 +173,8 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUVNEON(void) {
   WebPConvertRGBA32ToUV = ConvertRGBA32ToUV_NEON;
 }
 
-//------------------------------------------------------------------------------
-
-#define MAX_Y ((1 << 10) - 1)    // 10b precision over 16b-arithmetic
-static uint16_t clip_y_NEON(int v) {
-  return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
-}
-
-static uint64_t SharpYUVUpdateY_NEON(const uint16_t* ref, const uint16_t* src,
-                                     uint16_t* dst, int len) {
-  int i;
-  const int16x8_t zero = vdupq_n_s16(0);
-  const int16x8_t max = vdupq_n_s16(MAX_Y);
-  uint64x2_t sum = vdupq_n_u64(0);
-  uint64_t diff;
-
-  for (i = 0; i + 8 <= len; i += 8) {
-    const int16x8_t A = vreinterpretq_s16_u16(vld1q_u16(ref + i));
-    const int16x8_t B = vreinterpretq_s16_u16(vld1q_u16(src + i));
-    const int16x8_t C = vreinterpretq_s16_u16(vld1q_u16(dst + i));
-    const int16x8_t D = vsubq_s16(A, B);       // diff_y
-    const int16x8_t F = vaddq_s16(C, D);       // new_y
-    const uint16x8_t H =
-        vreinterpretq_u16_s16(vmaxq_s16(vminq_s16(F, max), zero));
-    const int16x8_t I = vabsq_s16(D);          // abs(diff_y)
-    vst1q_u16(dst + i, H);
-    sum = vpadalq_u32(sum, vpaddlq_u16(vreinterpretq_u16_s16(I)));
-  }
-  diff = vgetq_lane_u64(sum, 0) + vgetq_lane_u64(sum, 1);
-  for (; i < len; ++i) {
-    const int diff_y = ref[i] - src[i];
-    const int new_y = (int)(dst[i]) + diff_y;
-    dst[i] = clip_y_NEON(new_y);
-    diff += (uint64_t)(abs(diff_y));
-  }
-  return diff;
-}
-
-static void SharpYUVUpdateRGB_NEON(const int16_t* ref, const int16_t* src,
-                                   int16_t* dst, int len) {
-  int i;
-  for (i = 0; i + 8 <= len; i += 8) {
-    const int16x8_t A = vld1q_s16(ref + i);
-    const int16x8_t B = vld1q_s16(src + i);
-    const int16x8_t C = vld1q_s16(dst + i);
-    const int16x8_t D = vsubq_s16(A, B);   // diff_uv
-    const int16x8_t E = vaddq_s16(C, D);   // new_uv
-    vst1q_s16(dst + i, E);
-  }
-  for (; i < len; ++i) {
-    const int diff_uv = ref[i] - src[i];
-    dst[i] += diff_uv;
-  }
-}
-
-static void SharpYUVFilterRow_NEON(const int16_t* A, const int16_t* B, int len,
-                                   const uint16_t* best_y, uint16_t* out) {
-  int i;
-  const int16x8_t max = vdupq_n_s16(MAX_Y);
-  const int16x8_t zero = vdupq_n_s16(0);
-  for (i = 0; i + 8 <= len; i += 8) {
-    const int16x8_t a0 = vld1q_s16(A + i + 0);
-    const int16x8_t a1 = vld1q_s16(A + i + 1);
-    const int16x8_t b0 = vld1q_s16(B + i + 0);
-    const int16x8_t b1 = vld1q_s16(B + i + 1);
-    const int16x8_t a0b1 = vaddq_s16(a0, b1);
-    const int16x8_t a1b0 = vaddq_s16(a1, b0);
-    const int16x8_t a0a1b0b1 = vaddq_s16(a0b1, a1b0);  // A0+A1+B0+B1
-    const int16x8_t a0b1_2 = vaddq_s16(a0b1, a0b1);    // 2*(A0+B1)
-    const int16x8_t a1b0_2 = vaddq_s16(a1b0, a1b0);    // 2*(A1+B0)
-    const int16x8_t c0 = vshrq_n_s16(vaddq_s16(a0b1_2, a0a1b0b1), 3);
-    const int16x8_t c1 = vshrq_n_s16(vaddq_s16(a1b0_2, a0a1b0b1), 3);
-    const int16x8_t d0 = vaddq_s16(c1, a0);
-    const int16x8_t d1 = vaddq_s16(c0, a1);
-    const int16x8_t e0 = vrshrq_n_s16(d0, 1);
-    const int16x8_t e1 = vrshrq_n_s16(d1, 1);
-    const int16x8x2_t f = vzipq_s16(e0, e1);
-    const int16x8_t g0 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 0));
-    const int16x8_t g1 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 8));
-    const int16x8_t h0 = vaddq_s16(g0, f.val[0]);
-    const int16x8_t h1 = vaddq_s16(g1, f.val[1]);
-    const int16x8_t i0 = vmaxq_s16(vminq_s16(h0, max), zero);
-    const int16x8_t i1 = vmaxq_s16(vminq_s16(h1, max), zero);
-    vst1q_u16(out + 2 * i + 0, vreinterpretq_u16_s16(i0));
-    vst1q_u16(out + 2 * i + 8, vreinterpretq_u16_s16(i1));
-  }
-  for (; i < len; ++i) {
-    const int a0b1 = A[i + 0] + B[i + 1];
-    const int a1b0 = A[i + 1] + B[i + 0];
-    const int a0a1b0b1 = a0b1 + a1b0 + 8;
-    const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
-    const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
-    out[2 * i + 0] = clip_y_NEON(best_y[2 * i + 0] + v0);
-    out[2 * i + 1] = clip_y_NEON(best_y[2 * i + 1] + v1);
-  }
-}
-#undef MAX_Y
-
-//------------------------------------------------------------------------------
-
-extern void WebPInitSharpYUVNEON(void);
-
-WEBP_TSAN_IGNORE_FUNCTION void WebPInitSharpYUVNEON(void) {
-  WebPSharpYUVUpdateY = SharpYUVUpdateY_NEON;
-  WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_NEON;
-  WebPSharpYUVFilterRow = SharpYUVFilterRow_NEON;
-}
-
 #else  // !WEBP_USE_NEON
 
 WEBP_DSP_INIT_STUB(WebPInitConvertARGBToYUVNEON)
-WEBP_DSP_INIT_STUB(WebPInitSharpYUVNEON)
 
 #endif  // WEBP_USE_NEON

3rdparty/libwebp/src/dsp/yuv_sse2.c

@@ -15,10 +15,12 @@
 
 #if defined(WEBP_USE_SSE2)
 
-#include "src/dsp/common_sse2.h"
 #include <stdlib.h>
 #include <emmintrin.h>
 
+#include "src/dsp/common_sse2.h"
+#include "src/utils/utils.h"
+
 //-----------------------------------------------------------------------------
 // Convert spans of 32 pixels to various RGB formats for the fancy upsampler.
 
@@ -74,7 +76,7 @@ static WEBP_INLINE __m128i Load_HI_16_SSE2(const uint8_t* src) {
 // Load and replicate the U/V samples
 static WEBP_INLINE __m128i Load_UV_HI_8_SSE2(const uint8_t* src) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i tmp0 = _mm_cvtsi32_si128(*(const uint32_t*)src);
+  const __m128i tmp0 = _mm_cvtsi32_si128(WebPMemToInt32(src));
   const __m128i tmp1 = _mm_unpacklo_epi8(zero, tmp0);
   return _mm_unpacklo_epi16(tmp1, tmp1);   // replicate samples
 }
@@ -130,7 +132,7 @@ static WEBP_INLINE void PackAndStore4444_SSE2(const __m128i* const R,
   const __m128i rg0 = _mm_packus_epi16(*B, *A);
   const __m128i ba0 = _mm_packus_epi16(*R, *G);
 #endif
-  const __m128i mask_0xf0 = _mm_set1_epi8(0xf0);
+  const __m128i mask_0xf0 = _mm_set1_epi8((char)0xf0);
   const __m128i rb1 = _mm_unpacklo_epi8(rg0, ba0);  // rbrbrbrbrb...
   const __m128i ga1 = _mm_unpackhi_epi8(rg0, ba0);  // gagagagaga...
   const __m128i rb2 = _mm_and_si128(rb1, mask_0xf0);
@@ -147,9 +149,10 @@ static WEBP_INLINE void PackAndStore565_SSE2(const __m128i* const R,
   const __m128i r0 = _mm_packus_epi16(*R, *R);
   const __m128i g0 = _mm_packus_epi16(*G, *G);
   const __m128i b0 = _mm_packus_epi16(*B, *B);
-  const __m128i r1 = _mm_and_si128(r0, _mm_set1_epi8(0xf8));
+  const __m128i r1 = _mm_and_si128(r0, _mm_set1_epi8((char)0xf8));
   const __m128i b1 = _mm_and_si128(_mm_srli_epi16(b0, 3), _mm_set1_epi8(0x1f));
-  const __m128i g1 = _mm_srli_epi16(_mm_and_si128(g0, _mm_set1_epi8(0xe0)), 5);
+  const __m128i g1 =
+      _mm_srli_epi16(_mm_and_si128(g0, _mm_set1_epi8((char)0xe0)), 5);
   const __m128i g2 = _mm_slli_epi16(_mm_and_si128(g0, _mm_set1_epi8(0x1c)), 3);
   const __m128i rg = _mm_or_si128(r1, g1);
   const __m128i gb = _mm_or_si128(g2, b1);
@@ -747,128 +750,9 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUVSSE2(void) {
   WebPConvertRGBA32ToUV = ConvertRGBA32ToUV_SSE2;
 }
 
-//------------------------------------------------------------------------------
-
-#define MAX_Y ((1 << 10) - 1)    // 10b precision over 16b-arithmetic
-static uint16_t clip_y(int v) {
-  return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
-}
-
-static uint64_t SharpYUVUpdateY_SSE2(const uint16_t* ref, const uint16_t* src,
-                                     uint16_t* dst, int len) {
-  uint64_t diff = 0;
-  uint32_t tmp[4];
-  int i;
-  const __m128i zero = _mm_setzero_si128();
-  const __m128i max = _mm_set1_epi16(MAX_Y);
-  const __m128i one = _mm_set1_epi16(1);
-  __m128i sum = zero;
-
-  for (i = 0; i + 8 <= len; i += 8) {
-    const __m128i A = _mm_loadu_si128((const __m128i*)(ref + i));
-    const __m128i B = _mm_loadu_si128((const __m128i*)(src + i));
-    const __m128i C = _mm_loadu_si128((const __m128i*)(dst + i));
-    const __m128i D = _mm_sub_epi16(A, B);       // diff_y
-    const __m128i E = _mm_cmpgt_epi16(zero, D);  // sign (-1 or 0)
-    const __m128i F = _mm_add_epi16(C, D);       // new_y
-    const __m128i G = _mm_or_si128(E, one);      // -1 or 1
-    const __m128i H = _mm_max_epi16(_mm_min_epi16(F, max), zero);
-    const __m128i I = _mm_madd_epi16(D, G);      // sum(abs(...))
-    _mm_storeu_si128((__m128i*)(dst + i), H);
-    sum = _mm_add_epi32(sum, I);
-  }
-  _mm_storeu_si128((__m128i*)tmp, sum);
-  diff = tmp[3] + tmp[2] + tmp[1] + tmp[0];
-  for (; i < len; ++i) {
-    const int diff_y = ref[i] - src[i];
-    const int new_y = (int)dst[i] + diff_y;
-    dst[i] = clip_y(new_y);
-    diff += (uint64_t)abs(diff_y);
-  }
-  return diff;
-}
-
-static void SharpYUVUpdateRGB_SSE2(const int16_t* ref, const int16_t* src,
-                                   int16_t* dst, int len) {
-  int i = 0;
-  for (i = 0; i + 8 <= len; i += 8) {
-    const __m128i A = _mm_loadu_si128((const __m128i*)(ref + i));
-    const __m128i B = _mm_loadu_si128((const __m128i*)(src + i));
-    const __m128i C = _mm_loadu_si128((const __m128i*)(dst + i));
-    const __m128i D = _mm_sub_epi16(A, B);   // diff_uv
-    const __m128i E = _mm_add_epi16(C, D);   // new_uv
-    _mm_storeu_si128((__m128i*)(dst + i), E);
-  }
-  for (; i < len; ++i) {
-    const int diff_uv = ref[i] - src[i];
-    dst[i] += diff_uv;
-  }
-}
-
-static void SharpYUVFilterRow_SSE2(const int16_t* A, const int16_t* B, int len,
-                                   const uint16_t* best_y, uint16_t* out) {
-  int i;
-  const __m128i kCst8 = _mm_set1_epi16(8);
-  const __m128i max = _mm_set1_epi16(MAX_Y);
-  const __m128i zero = _mm_setzero_si128();
-  for (i = 0; i + 8 <= len; i += 8) {
-    const __m128i a0 = _mm_loadu_si128((const __m128i*)(A + i + 0));
-    const __m128i a1 = _mm_loadu_si128((const __m128i*)(A + i + 1));
-    const __m128i b0 = _mm_loadu_si128((const __m128i*)(B + i + 0));
-    const __m128i b1 = _mm_loadu_si128((const __m128i*)(B + i + 1));
-    const __m128i a0b1 = _mm_add_epi16(a0, b1);
-    const __m128i a1b0 = _mm_add_epi16(a1, b0);
-    const __m128i a0a1b0b1 = _mm_add_epi16(a0b1, a1b0);  // A0+A1+B0+B1
-    const __m128i a0a1b0b1_8 = _mm_add_epi16(a0a1b0b1, kCst8);
-    const __m128i a0b1_2 = _mm_add_epi16(a0b1, a0b1);    // 2*(A0+B1)
-    const __m128i a1b0_2 = _mm_add_epi16(a1b0, a1b0);    // 2*(A1+B0)
-    const __m128i c0 = _mm_srai_epi16(_mm_add_epi16(a0b1_2, a0a1b0b1_8), 3);
-    const __m128i c1 = _mm_srai_epi16(_mm_add_epi16(a1b0_2, a0a1b0b1_8), 3);
-    const __m128i d0 = _mm_add_epi16(c1, a0);
-    const __m128i d1 = _mm_add_epi16(c0, a1);
-    const __m128i e0 = _mm_srai_epi16(d0, 1);
-    const __m128i e1 = _mm_srai_epi16(d1, 1);
-    const __m128i f0 = _mm_unpacklo_epi16(e0, e1);
-    const __m128i f1 = _mm_unpackhi_epi16(e0, e1);
-    const __m128i g0 = _mm_loadu_si128((const __m128i*)(best_y + 2 * i + 0));
-    const __m128i g1 = _mm_loadu_si128((const __m128i*)(best_y + 2 * i + 8));
-    const __m128i h0 = _mm_add_epi16(g0, f0);
-    const __m128i h1 = _mm_add_epi16(g1, f1);
-    const __m128i i0 = _mm_max_epi16(_mm_min_epi16(h0, max), zero);
-    const __m128i i1 = _mm_max_epi16(_mm_min_epi16(h1, max), zero);
-    _mm_storeu_si128((__m128i*)(out + 2 * i + 0), i0);
-    _mm_storeu_si128((__m128i*)(out + 2 * i + 8), i1);
-  }
-  for (; i < len; ++i) {
-    //   (9 * A0 + 3 * A1 + 3 * B0 + B1 + 8) >> 4 =
-    // = (8 * A0 + 2 * (A1 + B0) + (A0 + A1 + B0 + B1 + 8)) >> 4
-    // We reuse the common sub-expressions.
-    const int a0b1 = A[i + 0] + B[i + 1];
-    const int a1b0 = A[i + 1] + B[i + 0];
-    const int a0a1b0b1 = a0b1 + a1b0 + 8;
-    const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
-    const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
-    out[2 * i + 0] = clip_y(best_y[2 * i + 0] + v0);
-    out[2 * i + 1] = clip_y(best_y[2 * i + 1] + v1);
-  }
-}
-
-#undef MAX_Y
-
-//------------------------------------------------------------------------------
-
-extern void WebPInitSharpYUVSSE2(void);
-
-WEBP_TSAN_IGNORE_FUNCTION void WebPInitSharpYUVSSE2(void) {
-  WebPSharpYUVUpdateY = SharpYUVUpdateY_SSE2;
-  WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_SSE2;
-  WebPSharpYUVFilterRow = SharpYUVFilterRow_SSE2;
-}
-
 #else  // !WEBP_USE_SSE2
 
 WEBP_DSP_INIT_STUB(WebPInitSamplersSSE2)
 WEBP_DSP_INIT_STUB(WebPInitConvertARGBToYUVSSE2)
-WEBP_DSP_INIT_STUB(WebPInitSharpYUVSSE2)
 
 #endif  // WEBP_USE_SSE2

3rdparty/libwebp/src/dsp/yuv_sse41.c

@@ -15,10 +15,12 @@
 
 #if defined(WEBP_USE_SSE41)
 
-#include "src/dsp/common_sse41.h"
 #include <stdlib.h>
 #include <smmintrin.h>
 
+#include "src/dsp/common_sse41.h"
+#include "src/utils/utils.h"
+
 //-----------------------------------------------------------------------------
 // Convert spans of 32 pixels to various RGB formats for the fancy upsampler.
 
@@ -74,7 +76,7 @@ static WEBP_INLINE __m128i Load_HI_16_SSE41(const uint8_t* src) {
 // Load and replicate the U/V samples
 static WEBP_INLINE __m128i Load_UV_HI_8_SSE41(const uint8_t* src) {
   const __m128i zero = _mm_setzero_si128();
-  const __m128i tmp0 = _mm_cvtsi32_si128(*(const uint32_t*)src);
+  const __m128i tmp0 = _mm_cvtsi32_si128(WebPMemToInt32(src));
   const __m128i tmp1 = _mm_unpacklo_epi8(zero, tmp0);
   return _mm_unpacklo_epi16(tmp1, tmp1);   // replicate samples
 }

3rdparty/libwebp/src/enc/alpha_enc.c

@@ -13,6 +13,7 @@
 
 #include <assert.h>
 #include <stdlib.h>
+#include <string.h>
 
 #include "src/enc/vp8i_enc.h"
 #include "src/dsp/dsp.h"
@@ -54,7 +55,7 @@ static int EncodeLossless(const uint8_t* const data, int width, int height,
   WebPConfig config;
   WebPPicture picture;
 
-  WebPPictureInit(&picture);
+  if (!WebPPictureInit(&picture)) return 0;
   picture.width = width;
   picture.height = height;
   picture.use_argb = 1;
@@ -86,7 +87,7 @@ static int EncodeLossless(const uint8_t* const data, int width, int height,
   // a decoder bug related to alpha with color cache.
   // See: https://code.google.com/p/webp/issues/detail?id=239
   // Need to re-enable this later.
-  ok = (VP8LEncodeStream(&config, &picture, bw, 0 /*use_cache*/) == VP8_ENC_OK);
+  ok = VP8LEncodeStream(&config, &picture, bw, /*use_cache=*/0);
   WebPPictureFree(&picture);
   ok = ok && !bw->error_;
   if (!ok) {
@@ -140,6 +141,11 @@ static int EncodeAlphaInternal(const uint8_t* const data, int width, int height,
                               !reduce_levels, &tmp_bw, &result->stats);
     if (ok) {
       output = VP8LBitWriterFinish(&tmp_bw);
+      if (tmp_bw.error_) {
+        VP8LBitWriterWipeOut(&tmp_bw);
+        memset(&result->bw, 0, sizeof(result->bw));
+        return 0;
+      }
       output_size = VP8LBitWriterNumBytes(&tmp_bw);
       if (output_size > data_size) {
         // compressed size is larger than source! Revert to uncompressed mode.
@@ -148,6 +154,7 @@ static int EncodeAlphaInternal(const uint8_t* const data, int width, int height,
       }
     } else {
       VP8LBitWriterWipeOut(&tmp_bw);
+      memset(&result->bw, 0, sizeof(result->bw));
       return 0;
     }
   }
@@ -162,7 +169,7 @@ static int EncodeAlphaInternal(const uint8_t* const data, int width, int height,
   header = method | (filter << 2);
   if (reduce_levels) header |= ALPHA_PREPROCESSED_LEVELS << 4;
 
-  VP8BitWriterInit(&result->bw, ALPHA_HEADER_LEN + output_size);
+  if (!VP8BitWriterInit(&result->bw, ALPHA_HEADER_LEN + output_size)) ok = 0;
   ok = ok && VP8BitWriterAppend(&result->bw, &header, ALPHA_HEADER_LEN);
   ok = ok && VP8BitWriterAppend(&result->bw, output, output_size);
 
@@ -303,7 +310,7 @@ static int EncodeAlpha(VP8Encoder* const enc,
   int ok = 1;
   const int reduce_levels = (quality < 100);
 
-  // quick sanity checks
+  // quick correctness checks
   assert((uint64_t)data_size == (uint64_t)width * height);  // as per spec
   assert(enc != NULL && pic != NULL && pic->a != NULL);
   assert(output != NULL && output_size != NULL);
@@ -312,11 +319,11 @@ static int EncodeAlpha(VP8Encoder* const enc,
   assert(filter >= WEBP_FILTER_NONE && filter <= WEBP_FILTER_FAST);
 
   if (quality < 0 || quality > 100) {
-    return 0;
+    return WebPEncodingSetError(pic, VP8_ENC_ERROR_INVALID_CONFIGURATION);
   }
 
   if (method < ALPHA_NO_COMPRESSION || method > ALPHA_LOSSLESS_COMPRESSION) {
-    return 0;
+    return WebPEncodingSetError(pic, VP8_ENC_ERROR_INVALID_CONFIGURATION);
   }
 
   if (method == ALPHA_NO_COMPRESSION) {
@@ -326,7 +333,7 @@ static int EncodeAlpha(VP8Encoder* const enc,
 
   quant_alpha = (uint8_t*)WebPSafeMalloc(1ULL, data_size);
   if (quant_alpha == NULL) {
-    return 0;
+    return WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
   }
 
   // Extract alpha data (width x height) from raw_data (stride x height).
@@ -346,6 +353,9 @@ static int EncodeAlpha(VP8Encoder* const enc,
     ok = ApplyFiltersAndEncode(quant_alpha, width, height, data_size, method,
                                filter, reduce_levels, effort_level, output,
                                output_size, pic->stats);
+    if (!ok) {
+      WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);  // imprecise
+    }
 #if !defined(WEBP_DISABLE_STATS)
     if (pic->stats != NULL) {  // need stats?
       pic->stats->coded_size += (int)(*output_size);
@@ -361,7 +371,7 @@ static int EncodeAlpha(VP8Encoder* const enc,
 //------------------------------------------------------------------------------
 // Main calls
 
-static int CompressAlphaJob(void* arg1, void* dummy) {
+static int CompressAlphaJob(void* arg1, void* unused) {
   VP8Encoder* const enc = (VP8Encoder*)arg1;
   const WebPConfig* config = enc->config_;
   uint8_t* alpha_data = NULL;
@@ -375,13 +385,13 @@ static int CompressAlphaJob(void* arg1, void* dummy) {
                    filter, effort_level, &alpha_data, &alpha_size)) {
     return 0;
   }
-  if (alpha_size != (uint32_t)alpha_size) {  // Sanity check.
+  if (alpha_size != (uint32_t)alpha_size) {  // Soundness check.
     WebPSafeFree(alpha_data);
     return 0;
   }
   enc->alpha_data_size_ = (uint32_t)alpha_size;
   enc->alpha_data_ = alpha_data;
-  (void)dummy;
+  (void)unused;
   return 1;
 }
 
@@ -405,7 +415,7 @@ int VP8EncStartAlpha(VP8Encoder* const enc) {
       WebPWorker* const worker = &enc->alpha_worker_;
       // Makes sure worker is good to go.
       if (!WebPGetWorkerInterface()->Reset(worker)) {
-        return 0;
+        return WebPEncodingSetError(enc->pic_, VP8_ENC_ERROR_OUT_OF_MEMORY);
       }
       WebPGetWorkerInterface()->Launch(worker);
       return 1;

3rdparty/libwebp/src/enc/analysis_enc.c

@@ -391,12 +391,14 @@ static int DoSegmentsJob(void* arg1, void* arg2) {
   return ok;
 }
 
+#ifdef WEBP_USE_THREAD
 static void MergeJobs(const SegmentJob* const src, SegmentJob* const dst) {
   int i;
   for (i = 0; i <= MAX_ALPHA; ++i) dst->alphas[i] += src->alphas[i];
   dst->alpha += src->alpha;
   dst->uv_alpha += src->uv_alpha;
 }
+#endif
 
 // initialize the job struct with some tasks to perform
 static void InitSegmentJob(VP8Encoder* const enc, SegmentJob* const job,
@@ -425,10 +427,10 @@ int VP8EncAnalyze(VP8Encoder* const enc) {
       (enc->method_ <= 1);  // for method 0 - 1, we need preds_[] to be filled.
   if (do_segments) {
     const int last_row = enc->mb_h_;
-    // We give a little more than a half work to the main thread.
-    const int split_row = (9 * last_row + 15) >> 4;
     const int total_mb = last_row * enc->mb_w_;
 #ifdef WEBP_USE_THREAD
+    // We give a little more than a half work to the main thread.
+    const int split_row = (9 * last_row + 15) >> 4;
     const int kMinSplitRow = 2;  // minimal rows needed for mt to be worth it
     const int do_mt = (enc->thread_level_ > 0) && (split_row >= kMinSplitRow);
 #else
@@ -438,6 +440,7 @@ int VP8EncAnalyze(VP8Encoder* const enc) {
         WebPGetWorkerInterface();
     SegmentJob main_job;
     if (do_mt) {
+#ifdef WEBP_USE_THREAD
       SegmentJob side_job;
       // Note the use of '&' instead of '&&' because we must call the functions
       // no matter what.
@@ -455,6 +458,7 @@ int VP8EncAnalyze(VP8Encoder* const enc) {
       }
       worker_interface->End(&side_job.worker);
       if (ok) MergeJobs(&side_job, &main_job);  // merge results together
+#endif  // WEBP_USE_THREAD
     } else {
       // Even for single-thread case, we use the generic Worker tools.
       InitSegmentJob(enc, &main_job, 0, last_row);
@@ -470,6 +474,10 @@ int VP8EncAnalyze(VP8Encoder* const enc) {
   } else {   // Use only one default segment.
     ResetAllMBInfo(enc);
   }
+  if (!ok) {
+    return WebPEncodingSetError(enc->pic_,
+                                VP8_ENC_ERROR_OUT_OF_MEMORY);  // imprecise
+  }
   return ok;
 }
 

3rdparty/libwebp/src/enc/backward_references_cost_enc.c

@@ -15,10 +15,11 @@
 //
 
 #include <assert.h>
+#include <float.h>
 
+#include "src/dsp/lossless_common.h"
 #include "src/enc/backward_references_enc.h"
 #include "src/enc/histogram_enc.h"
-#include "src/dsp/lossless_common.h"
 #include "src/utils/color_cache_utils.h"
 #include "src/utils/utils.h"
 
@@ -30,15 +31,15 @@ extern void VP8LBackwardRefsCursorAdd(VP8LBackwardRefs* const refs,
                                       const PixOrCopy v);
 
 typedef struct {
-  double alpha_[VALUES_IN_BYTE];
-  double red_[VALUES_IN_BYTE];
-  double blue_[VALUES_IN_BYTE];
-  double distance_[NUM_DISTANCE_CODES];
-  double* literal_;
+  float alpha_[VALUES_IN_BYTE];
+  float red_[VALUES_IN_BYTE];
+  float blue_[VALUES_IN_BYTE];
+  float distance_[NUM_DISTANCE_CODES];
+  float* literal_;
 } CostModel;
 
 static void ConvertPopulationCountTableToBitEstimates(
-    int num_symbols, const uint32_t population_counts[], double output[]) {
+    int num_symbols, const uint32_t population_counts[], float output[]) {
   uint32_t sum = 0;
   int nonzeros = 0;
   int i;
@@ -51,7 +52,7 @@ static void ConvertPopulationCountTableToBitEstimates(
   if (nonzeros <= 1) {
     memset(output, 0, num_symbols * sizeof(*output));
   } else {
-    const double logsum = VP8LFastLog2(sum);
+    const float logsum = VP8LFastLog2(sum);
     for (i = 0; i < num_symbols; ++i) {
       output[i] = logsum - VP8LFastLog2(population_counts[i]);
     }
@@ -75,8 +76,8 @@ static int CostModelBuild(CostModel* const m, int xsize, int cache_bits,
   }
 
   ConvertPopulationCountTableToBitEstimates(
-      VP8LHistogramNumCodes(histo->palette_code_bits_),
-      histo->literal_, m->literal_);
+      VP8LHistogramNumCodes(histo->palette_code_bits_), histo->literal_,
+      m->literal_);
   ConvertPopulationCountTableToBitEstimates(
       VALUES_IN_BYTE, histo->red_, m->red_);
   ConvertPopulationCountTableToBitEstimates(
@@ -92,27 +93,27 @@ static int CostModelBuild(CostModel* const m, int xsize, int cache_bits,
   return ok;
 }
 
-static WEBP_INLINE double GetLiteralCost(const CostModel* const m, uint32_t v) {
+static WEBP_INLINE float GetLiteralCost(const CostModel* const m, uint32_t v) {
   return m->alpha_[v >> 24] +
          m->red_[(v >> 16) & 0xff] +
          m->literal_[(v >> 8) & 0xff] +
          m->blue_[v & 0xff];
 }
 
-static WEBP_INLINE double GetCacheCost(const CostModel* const m, uint32_t idx) {
+static WEBP_INLINE float GetCacheCost(const CostModel* const m, uint32_t idx) {
   const int literal_idx = VALUES_IN_BYTE + NUM_LENGTH_CODES + idx;
   return m->literal_[literal_idx];
 }
 
-static WEBP_INLINE double GetLengthCost(const CostModel* const m,
-                                        uint32_t length) {
+static WEBP_INLINE float GetLengthCost(const CostModel* const m,
+                                       uint32_t length) {
   int code, extra_bits;
   VP8LPrefixEncodeBits(length, &code, &extra_bits);
   return m->literal_[VALUES_IN_BYTE + code] + extra_bits;
 }
 
-static WEBP_INLINE double GetDistanceCost(const CostModel* const m,
-                                          uint32_t distance) {
+static WEBP_INLINE float GetDistanceCost(const CostModel* const m,
+                                         uint32_t distance) {
   int code, extra_bits;
   VP8LPrefixEncodeBits(distance, &code, &extra_bits);
   return m->distance_[code] + extra_bits;
@@ -122,20 +123,20 @@ static WEBP_INLINE void AddSingleLiteralWithCostModel(
     const uint32_t* const argb, VP8LColorCache* const hashers,
     const CostModel* const cost_model, int idx, int use_color_cache,
     float prev_cost, float* const cost, uint16_t* const dist_array) {
-  double cost_val = prev_cost;
+  float cost_val = prev_cost;
   const uint32_t color = argb[idx];
   const int ix = use_color_cache ? VP8LColorCacheContains(hashers, color) : -1;
   if (ix >= 0) {
     // use_color_cache is true and hashers contains color
-    const double mul0 = 0.68;
+    const float mul0 = 0.68f;
     cost_val += GetCacheCost(cost_model, ix) * mul0;
   } else {
-    const double mul1 = 0.82;
+    const float mul1 = 0.82f;
     if (use_color_cache) VP8LColorCacheInsert(hashers, color);
     cost_val += GetLiteralCost(cost_model, color) * mul1;
   }
   if (cost[idx] > cost_val) {
-    cost[idx] = (float)cost_val;
+    cost[idx] = cost_val;
     dist_array[idx] = 1;  // only one is inserted.
   }
 }
@@ -172,7 +173,7 @@ struct CostInterval {
 
 // The GetLengthCost(cost_model, k) are cached in a CostCacheInterval.
 typedef struct {
-  double cost_;
+  float cost_;
   int start_;
   int end_;       // Exclusive.
 } CostCacheInterval;
@@ -187,7 +188,7 @@ typedef struct {
   int count_;  // The number of stored intervals.
   CostCacheInterval* cache_intervals_;
   size_t cache_intervals_size_;
-  double cost_cache_[MAX_LENGTH];  // Contains the GetLengthCost(cost_model, k).
+  float cost_cache_[MAX_LENGTH];  // Contains the GetLengthCost(cost_model, k).
   float* costs_;
   uint16_t* dist_array_;
   // Most of the time, we only need few intervals -> use a free-list, to avoid
@@ -262,10 +263,13 @@ static int CostManagerInit(CostManager* const manager,
   CostManagerInitFreeList(manager);
 
   // Fill in the cost_cache_.
+  // Has to be done in two passes due to a GCC bug on i686
+  // related to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=323
+  for (i = 0; i < cost_cache_size; ++i) {
+    manager->cost_cache_[i] = GetLengthCost(cost_model, i);
+  }
   manager->cache_intervals_size_ = 1;
-  manager->cost_cache_[0] = GetLengthCost(cost_model, 0);
   for (i = 1; i < cost_cache_size; ++i) {
-    manager->cost_cache_[i] = GetLengthCost(cost_model, i);
     // Get the number of bound intervals.
     if (manager->cost_cache_[i] != manager->cost_cache_[i - 1]) {
       ++manager->cache_intervals_size_;
@@ -294,7 +298,7 @@ static int CostManagerInit(CostManager* const manager,
     cur->end_ = 1;
     cur->cost_ = manager->cost_cache_[0];
     for (i = 1; i < cost_cache_size; ++i) {
-      const double cost_val = manager->cost_cache_[i];
+      const float cost_val = manager->cost_cache_[i];
       if (cost_val != cur->cost_) {
         ++cur;
         // Initialize an interval.
@@ -303,6 +307,8 @@ static int CostManagerInit(CostManager* const manager,
       }
       cur->end_ = i + 1;
     }
+    assert((size_t)(cur - manager->cache_intervals_) + 1 ==
+           manager->cache_intervals_size_);
   }
 
   manager->costs_ = (float*)WebPSafeMalloc(pix_count, sizeof(*manager->costs_));
@@ -311,7 +317,7 @@ static int CostManagerInit(CostManager* const manager,
     return 0;
   }
   // 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;
+  for (i = 0; i < pix_count; ++i) manager->costs_[i] = FLT_MAX;
 
   return 1;
 }
@@ -457,7 +463,7 @@ static WEBP_INLINE void InsertInterval(CostManager* const manager,
 // If handling the interval or one of its subintervals becomes to heavy, its
 // contribution is added to the costs right away.
 static WEBP_INLINE void PushInterval(CostManager* const manager,
-                                     double distance_cost, int position,
+                                     float distance_cost, int position,
                                      int len) {
   size_t i;
   CostInterval* interval = manager->head_;
@@ -474,7 +480,7 @@ static WEBP_INLINE void PushInterval(CostManager* const manager,
       const int k = j - position;
       float cost_tmp;
       assert(k >= 0 && k < MAX_LENGTH);
-      cost_tmp = (float)(distance_cost + manager->cost_cache_[k]);
+      cost_tmp = distance_cost + manager->cost_cache_[k];
 
       if (manager->costs_[j] > cost_tmp) {
         manager->costs_[j] = cost_tmp;
@@ -492,7 +498,7 @@ static WEBP_INLINE void PushInterval(CostManager* const manager,
     const int end = position + (cost_cache_intervals[i].end_ > len
                                  ? len
                                  : cost_cache_intervals[i].end_);
-    const float cost = (float)(distance_cost + cost_cache_intervals[i].cost_);
+    const float cost = distance_cost + cost_cache_intervals[i].cost_;
 
     for (; interval != NULL && interval->start_ < end;
          interval = interval_next) {
@@ -570,22 +576,21 @@ static int BackwardReferencesHashChainDistanceOnly(
   const int pix_count = xsize * ysize;
   const int use_color_cache = (cache_bits > 0);
   const size_t literal_array_size =
-      sizeof(double) * (NUM_LITERAL_CODES + NUM_LENGTH_CODES +
-                        ((cache_bits > 0) ? (1 << cache_bits) : 0));
+      sizeof(float) * (VP8LHistogramNumCodes(cache_bits));
   const size_t cost_model_size = sizeof(CostModel) + literal_array_size;
   CostModel* const cost_model =
       (CostModel*)WebPSafeCalloc(1ULL, cost_model_size);
   VP8LColorCache hashers;
   CostManager* cost_manager =
-      (CostManager*)WebPSafeMalloc(1ULL, sizeof(*cost_manager));
+      (CostManager*)WebPSafeCalloc(1ULL, sizeof(*cost_manager));
   int offset_prev = -1, len_prev = -1;
-  double offset_cost = -1;
+  float offset_cost = -1.f;
   int first_offset_is_constant = -1;  // initialized with 'impossible' value
   int reach = 0;
 
   if (cost_model == NULL || cost_manager == NULL) goto Error;
 
-  cost_model->literal_ = (double*)(cost_model + 1);
+  cost_model->literal_ = (float*)(cost_model + 1);
   if (use_color_cache) {
     cc_init = VP8LColorCacheInit(&hashers, cache_bits);
     if (!cc_init) goto Error;
@@ -675,7 +680,7 @@ static int BackwardReferencesHashChainDistanceOnly(
   }
 
   ok = !refs->error_;
-Error:
+ Error:
   if (cc_init) VP8LColorCacheClear(&hashers);
   CostManagerClear(cost_manager);
   WebPSafeFree(cost_model);

3rdparty/libwebp/src/enc/backward_references_enc.c

@@ -10,6 +10,8 @@
 // Author: Jyrki Alakuijala (jyrki@google.com)
 //
 
+#include "src/enc/backward_references_enc.h"
+
 #include <assert.h>
 #include <float.h>
 #include <math.h>
@@ -17,10 +19,11 @@
 #include "src/dsp/dsp.h"
 #include "src/dsp/lossless.h"
 #include "src/dsp/lossless_common.h"
-#include "src/enc/backward_references_enc.h"
 #include "src/enc/histogram_enc.h"
+#include "src/enc/vp8i_enc.h"
 #include "src/utils/color_cache_utils.h"
 #include "src/utils/utils.h"
+#include "src/webp/encode.h"
 
 #define MIN_BLOCK_SIZE 256  // minimum block size for backward references
 
@@ -255,10 +258,13 @@ static WEBP_INLINE int MaxFindCopyLength(int len) {
 
 int VP8LHashChainFill(VP8LHashChain* const p, int quality,
                       const uint32_t* const argb, int xsize, int ysize,
-                      int low_effort) {
+                      int low_effort, const WebPPicture* const pic,
+                      int percent_range, int* const percent) {
   const int size = xsize * ysize;
   const int iter_max = GetMaxItersForQuality(quality);
   const uint32_t window_size = GetWindowSizeForHashChain(quality, xsize);
+  int remaining_percent = percent_range;
+  int percent_start = *percent;
   int pos;
   int argb_comp;
   uint32_t base_position;
@@ -276,7 +282,12 @@ int VP8LHashChainFill(VP8LHashChain* const p, int quality,
 
   hash_to_first_index =
       (int32_t*)WebPSafeMalloc(HASH_SIZE, sizeof(*hash_to_first_index));
-  if (hash_to_first_index == NULL) return 0;
+  if (hash_to_first_index == NULL) {
+    return WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
+  }
+
+  percent_range = remaining_percent / 2;
+  remaining_percent -= percent_range;
 
   // Set the int32_t array to -1.
   memset(hash_to_first_index, 0xff, HASH_SIZE * sizeof(*hash_to_first_index));
@@ -323,12 +334,22 @@ int VP8LHashChainFill(VP8LHashChain* const p, int quality,
       hash_to_first_index[hash_code] = pos++;
       argb_comp = argb_comp_next;
     }
+
+    if (!WebPReportProgress(
+            pic, percent_start + percent_range * pos / (size - 2), percent)) {
+      WebPSafeFree(hash_to_first_index);
+      return 0;
+    }
   }
   // Process the penultimate pixel.
   chain[pos] = hash_to_first_index[GetPixPairHash64(argb + pos)];
 
   WebPSafeFree(hash_to_first_index);
 
+  percent_start += percent_range;
+  if (!WebPReportProgress(pic, percent_start, percent)) return 0;
+  percent_range = remaining_percent;
+
   // Find the best match interval at each pixel, defined by an offset to the
   // pixel and a length. The right-most pixel cannot match anything to the right
   // (hence a best length of 0) and the left-most pixel nothing to the left
@@ -417,8 +438,17 @@ int VP8LHashChainFill(VP8LHashChain* const p, int quality,
         max_base_position = base_position;
       }
     }
+
+    if (!WebPReportProgress(pic,
+                            percent_start + percent_range *
+                                                (size - 2 - base_position) /
+                                                (size - 2),
+                            percent)) {
+      return 0;
+    }
   }
-  return 1;
+
+  return WebPReportProgress(pic, percent_start + percent_range, percent);
 }
 
 static WEBP_INLINE void AddSingleLiteral(uint32_t pixel, int use_color_cache,
@@ -728,7 +758,7 @@ static int CalculateBestCacheSize(const uint32_t* argb, int quality,
                                   int* const best_cache_bits) {
   int i;
   const int cache_bits_max = (quality <= 25) ? 0 : *best_cache_bits;
-  double entropy_min = MAX_ENTROPY;
+  float entropy_min = MAX_ENTROPY;
   int cc_init[MAX_COLOR_CACHE_BITS + 1] = { 0 };
   VP8LColorCache hashers[MAX_COLOR_CACHE_BITS + 1];
   VP8LRefsCursor c = VP8LRefsCursorInit(refs);
@@ -813,14 +843,14 @@ static int CalculateBestCacheSize(const uint32_t* argb, int quality,
   }
 
   for (i = 0; i <= cache_bits_max; ++i) {
-    const double entropy = VP8LHistogramEstimateBits(histos[i]);
+    const float entropy = VP8LHistogramEstimateBits(histos[i]);
     if (i == 0 || entropy < entropy_min) {
       entropy_min = entropy;
       *best_cache_bits = i;
     }
   }
   ok = 1;
-Error:
+ Error:
   for (i = 0; i <= cache_bits_max; ++i) {
     if (cc_init[i]) VP8LColorCacheClear(&hashers[i]);
     VP8LFreeHistogram(histos[i]);
@@ -890,7 +920,7 @@ static int GetBackwardReferences(int width, int height,
   int i, lz77_type;
   // Index 0 is for a color cache, index 1 for no cache (if needed).
   int lz77_types_best[2] = {0, 0};
-  double bit_costs_best[2] = {DBL_MAX, DBL_MAX};
+  float bit_costs_best[2] = {FLT_MAX, FLT_MAX};
   VP8LHashChain hash_chain_box;
   VP8LBackwardRefs* const refs_tmp = &refs[do_no_cache ? 2 : 1];
   int status = 0;
@@ -902,7 +932,7 @@ static int GetBackwardReferences(int width, int height,
   for (lz77_type = 1; lz77_types_to_try;
        lz77_types_to_try &= ~lz77_type, lz77_type <<= 1) {
     int res = 0;
-    double bit_cost = 0.;
+    float bit_cost = 0.f;
     if ((lz77_types_to_try & lz77_type) == 0) continue;
     switch (lz77_type) {
       case kLZ77RLE:
@@ -976,15 +1006,16 @@ static int GetBackwardReferences(int width, int height,
       const VP8LHashChain* const hash_chain_tmp =
           (lz77_types_best[i] == kLZ77Standard) ? hash_chain : &hash_chain_box;
       const int cache_bits = (i == 1) ? 0 : *cache_bits_best;
-      if (VP8LBackwardReferencesTraceBackwards(width, height, argb, cache_bits,
-                                               hash_chain_tmp, &refs[i],
-                                               refs_tmp)) {
-        double bit_cost_trace;
-        VP8LHistogramCreate(histo, refs_tmp, cache_bits);
-        bit_cost_trace = VP8LHistogramEstimateBits(histo);
-        if (bit_cost_trace < bit_costs_best[i]) {
-          BackwardRefsSwap(refs_tmp, &refs[i]);
-        }
+      float bit_cost_trace;
+      if (!VP8LBackwardReferencesTraceBackwards(width, height, argb, cache_bits,
+                                                hash_chain_tmp, &refs[i],
+                                                refs_tmp)) {
+        goto Error;
+      }
+      VP8LHistogramCreate(histo, refs_tmp, cache_bits);
+      bit_cost_trace = VP8LHistogramEstimateBits(histo);
+      if (bit_cost_trace < bit_costs_best[i]) {
+        BackwardRefsSwap(refs_tmp, &refs[i]);
       }
     }
 
@@ -1000,31 +1031,35 @@ static int GetBackwardReferences(int width, int height,
   }
   status = 1;
 
-Error:
+ Error:
   VP8LHashChainClear(&hash_chain_box);
   VP8LFreeHistogram(histo);
   return status;
 }
 
-WebPEncodingError VP8LGetBackwardReferences(
+int VP8LGetBackwardReferences(
     int width, int height, const uint32_t* const argb, int quality,
     int low_effort, int lz77_types_to_try, int cache_bits_max, int do_no_cache,
     const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs,
-    int* const cache_bits_best) {
+    int* const cache_bits_best, const WebPPicture* const pic, int percent_range,
+    int* const percent) {
   if (low_effort) {
     VP8LBackwardRefs* refs_best;
     *cache_bits_best = cache_bits_max;
     refs_best = GetBackwardReferencesLowEffort(
         width, height, argb, cache_bits_best, hash_chain, refs);
-    if (refs_best == NULL) return VP8_ENC_ERROR_OUT_OF_MEMORY;
+    if (refs_best == NULL) {
+      return WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
+    }
     // Set it in first position.
     BackwardRefsSwap(refs_best, &refs[0]);
   } else {
     if (!GetBackwardReferences(width, height, argb, quality, lz77_types_to_try,
                                cache_bits_max, do_no_cache, hash_chain, refs,
                                cache_bits_best)) {
-      return VP8_ENC_ERROR_OUT_OF_MEMORY;
+      return WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
     }
   }
-  return VP8_ENC_OK;
+
+  return WebPReportProgress(pic, *percent + percent_range, percent);
 }

3rdparty/libwebp/src/enc/backward_references_enc.h

@@ -134,10 +134,11 @@ struct VP8LHashChain {
 
 // Must be called first, to set size.
 int VP8LHashChainInit(VP8LHashChain* const p, int size);
-// Pre-compute the best matches for argb.
+// Pre-compute the best matches for argb. pic and percent are for progress.
 int VP8LHashChainFill(VP8LHashChain* const p, int quality,
                       const uint32_t* const argb, int xsize, int ysize,
-                      int low_effort);
+                      int low_effort, const WebPPicture* const pic,
+                      int percent_range, int* const percent);
 void VP8LHashChainClear(VP8LHashChain* const p);  // release memory
 
 static WEBP_INLINE int VP8LHashChainFindOffset(const VP8LHashChain* const p,
@@ -227,11 +228,14 @@ enum VP8LLZ77Type {
 // VP8LBackwardRefs is put in the first element, the best value with no-cache in
 // the second element.
 // In both cases, the last element is used as temporary internally.
-WebPEncodingError VP8LGetBackwardReferences(
+// pic and percent are for progress.
+// Returns false in case of error (stored in pic->error_code).
+int VP8LGetBackwardReferences(
     int width, int height, const uint32_t* const argb, int quality,
     int low_effort, int lz77_types_to_try, int cache_bits_max, int do_no_cache,
     const VP8LHashChain* const hash_chain, VP8LBackwardRefs* const refs,
-    int* const cache_bits_best);
+    int* const cache_bits_best, const WebPPicture* const pic, int percent_range,
+    int* const percent);
 
 #ifdef __cplusplus
 }

3rdparty/libwebp/src/enc/frame_enc.c

@@ -578,7 +578,7 @@ static uint64_t OneStatPass(VP8Encoder* const enc, VP8RDLevel rd_opt,
   uint64_t size = 0;
   uint64_t size_p0 = 0;
   uint64_t distortion = 0;
-  const uint64_t pixel_count = nb_mbs * 384;
+  const uint64_t pixel_count = (uint64_t)nb_mbs * 384;
 
   VP8IteratorInit(enc, &it);
   SetLoopParams(enc, s->q);
@@ -689,7 +689,7 @@ static int PreLoopInitialize(VP8Encoder* const enc) {
   }
   if (!ok) {
     VP8EncFreeBitWriters(enc);  // malloc error occurred
-    WebPEncodingSetError(enc->pic_, VP8_ENC_ERROR_OUT_OF_MEMORY);
+    return WebPEncodingSetError(enc->pic_, VP8_ENC_ERROR_OUT_OF_MEMORY);
   }
   return ok;
 }
@@ -719,6 +719,7 @@ static int PostLoopFinalize(VP8EncIterator* const it, int ok) {
   } else {
     // Something bad happened -> need to do some memory cleanup.
     VP8EncFreeBitWriters(enc);
+    return WebPEncodingSetError(enc->pic_, VP8_ENC_ERROR_OUT_OF_MEMORY);
   }
   return ok;
 }
@@ -754,6 +755,11 @@ int VP8EncLoop(VP8Encoder* const enc) {
     // *then* decide how to code the skip decision if there's one.
     if (!VP8Decimate(&it, &info, rd_opt) || dont_use_skip) {
       CodeResiduals(it.bw_, &it, &info);
+      if (it.bw_->error_) {
+        // enc->pic_->error_code is set in PostLoopFinalize().
+        ok = 0;
+        break;
+      }
     } else {   // reset predictors after a skip
       ResetAfterSkip(&it);
     }
@@ -778,11 +784,12 @@ int VP8EncTokenLoop(VP8Encoder* const enc) {
   // Roughly refresh the proba eight times per pass
   int max_count = (enc->mb_w_ * enc->mb_h_) >> 3;
   int num_pass_left = enc->config_->pass;
+  int remaining_progress = 40;  // percents
   const int do_search = enc->do_search_;
   VP8EncIterator it;
   VP8EncProba* const proba = &enc->proba_;
   const VP8RDLevel rd_opt = enc->rd_opt_level_;
-  const uint64_t pixel_count = enc->mb_w_ * enc->mb_h_ * 384;
+  const uint64_t pixel_count = (uint64_t)enc->mb_w_ * enc->mb_h_ * 384;
   PassStats stats;
   int ok;
 
@@ -805,6 +812,9 @@ int VP8EncTokenLoop(VP8Encoder* const enc) {
     uint64_t size_p0 = 0;
     uint64_t distortion = 0;
     int cnt = max_count;
+    // The final number of passes is not trivial to know in advance.
+    const int pass_progress = remaining_progress / (2 + num_pass_left);
+    remaining_progress -= pass_progress;
     VP8IteratorInit(enc, &it);
     SetLoopParams(enc, stats.q);
     if (is_last_pass) {
@@ -832,7 +842,7 @@ int VP8EncTokenLoop(VP8Encoder* const enc) {
         StoreSideInfo(&it);
         VP8StoreFilterStats(&it);
         VP8IteratorExport(&it);
-        ok = VP8IteratorProgress(&it, 20);
+        ok = VP8IteratorProgress(&it, pass_progress);
       }
       VP8IteratorSaveBoundary(&it);
     } while (ok && VP8IteratorNext(&it));
@@ -878,7 +888,8 @@ int VP8EncTokenLoop(VP8Encoder* const enc) {
     ok = VP8EmitTokens(&enc->tokens_, enc->parts_ + 0,
                        (const uint8_t*)proba->coeffs_, 1);
   }
-  ok = ok && WebPReportProgress(enc->pic_, enc->percent_ + 20, &enc->percent_);
+  ok = ok && WebPReportProgress(enc->pic_, enc->percent_ + remaining_progress,
+                                &enc->percent_);
   return PostLoopFinalize(&it, ok);
 }
 

3rdparty/libwebp/src/enc/histogram_enc.c

@@ -13,15 +13,17 @@
 #include "src/webp/config.h"
 #endif
 
+#include <float.h>
 #include <math.h>
 
-#include "src/enc/backward_references_enc.h"
-#include "src/enc/histogram_enc.h"
 #include "src/dsp/lossless.h"
 #include "src/dsp/lossless_common.h"
+#include "src/enc/backward_references_enc.h"
+#include "src/enc/histogram_enc.h"
+#include "src/enc/vp8i_enc.h"
 #include "src/utils/utils.h"
 
-#define MAX_COST 1.e38
+#define MAX_BIT_COST FLT_MAX
 
 // Number of partitions for the three dominant (literal, red and blue) symbol
 // costs.
@@ -228,8 +230,8 @@ void VP8LHistogramAddSinglePixOrCopy(VP8LHistogram* const histo,
 // -----------------------------------------------------------------------------
 // Entropy-related functions.
 
-static WEBP_INLINE double BitsEntropyRefine(const VP8LBitEntropy* entropy) {
-  double mix;
+static WEBP_INLINE float BitsEntropyRefine(const VP8LBitEntropy* entropy) {
+  float mix;
   if (entropy->nonzeros < 5) {
     if (entropy->nonzeros <= 1) {
       return 0;
@@ -238,67 +240,67 @@ static WEBP_INLINE double BitsEntropyRefine(const VP8LBitEntropy* entropy) {
     // Let's mix in a bit of entropy to favor good clustering when
     // distributions of these are combined.
     if (entropy->nonzeros == 2) {
-      return 0.99 * entropy->sum + 0.01 * entropy->entropy;
+      return 0.99f * entropy->sum + 0.01f * entropy->entropy;
     }
     // No matter what the entropy says, we cannot be better than min_limit
     // with Huffman coding. I am mixing a bit of entropy into the
     // min_limit since it produces much better (~0.5 %) compression results
     // perhaps because of better entropy clustering.
     if (entropy->nonzeros == 3) {
-      mix = 0.95;
+      mix = 0.95f;
     } else {
-      mix = 0.7;  // nonzeros == 4.
+      mix = 0.7f;  // nonzeros == 4.
     }
   } else {
-    mix = 0.627;
+    mix = 0.627f;
   }
 
   {
-    double min_limit = 2 * entropy->sum - entropy->max_val;
-    min_limit = mix * min_limit + (1.0 - mix) * entropy->entropy;
+    float min_limit = 2.f * entropy->sum - entropy->max_val;
+    min_limit = mix * min_limit + (1.f - mix) * entropy->entropy;
     return (entropy->entropy < min_limit) ? min_limit : entropy->entropy;
   }
 }
 
-double VP8LBitsEntropy(const uint32_t* const array, int n) {
+float VP8LBitsEntropy(const uint32_t* const array, int n) {
   VP8LBitEntropy entropy;
   VP8LBitsEntropyUnrefined(array, n, &entropy);
 
   return BitsEntropyRefine(&entropy);
 }
 
-static double InitialHuffmanCost(void) {
+static float InitialHuffmanCost(void) {
   // Small bias because Huffman code length is typically not stored in
   // full length.
   static const int kHuffmanCodeOfHuffmanCodeSize = CODE_LENGTH_CODES * 3;
-  static const double kSmallBias = 9.1;
+  static const float kSmallBias = 9.1f;
   return kHuffmanCodeOfHuffmanCodeSize - kSmallBias;
 }
 
 // Finalize the Huffman cost based on streak numbers and length type (<3 or >=3)
-static double FinalHuffmanCost(const VP8LStreaks* const stats) {
+static float FinalHuffmanCost(const VP8LStreaks* const stats) {
   // The constants in this function are experimental and got rounded from
   // their original values in 1/8 when switched to 1/1024.
-  double retval = InitialHuffmanCost();
+  float retval = InitialHuffmanCost();
   // Second coefficient: Many zeros in the histogram are covered efficiently
   // by a run-length encode. Originally 2/8.
-  retval += stats->counts[0] * 1.5625 + 0.234375 * stats->streaks[0][1];
+  retval += stats->counts[0] * 1.5625f + 0.234375f * stats->streaks[0][1];
   // Second coefficient: Constant values are encoded less efficiently, but still
   // RLE'ed. Originally 6/8.
-  retval += stats->counts[1] * 2.578125 + 0.703125 * stats->streaks[1][1];
+  retval += stats->counts[1] * 2.578125f + 0.703125f * stats->streaks[1][1];
   // 0s are usually encoded more efficiently than non-0s.
   // Originally 15/8.
-  retval += 1.796875 * stats->streaks[0][0];
+  retval += 1.796875f * stats->streaks[0][0];
   // Originally 26/8.
-  retval += 3.28125 * stats->streaks[1][0];
+  retval += 3.28125f * stats->streaks[1][0];
   return retval;
 }
 
 // Get the symbol entropy for the distribution 'population'.
 // Set 'trivial_sym', if there's only one symbol present in the distribution.
-static double PopulationCost(const uint32_t* const population, int length,
-                             uint32_t* const trivial_sym,
-                             uint8_t* const is_used) {
+static float PopulationCost(const uint32_t* const population, int length,
+                            uint32_t* const trivial_sym,
+                            uint8_t* const is_used) {
   VP8LBitEntropy bit_entropy;
   VP8LStreaks stats;
   VP8LGetEntropyUnrefined(population, length, &bit_entropy, &stats);
@@ -314,11 +316,10 @@ static double PopulationCost(const uint32_t* const population, int length,
 
 // trivial_at_end is 1 if the two histograms only have one element that is
 // non-zero: both the zero-th one, or both the last one.
-static WEBP_INLINE double GetCombinedEntropy(const uint32_t* const X,
-                                             const uint32_t* const Y,
-                                             int length, int is_X_used,
-                                             int is_Y_used,
-                                             int trivial_at_end) {
+static WEBP_INLINE float GetCombinedEntropy(const uint32_t* const X,
+                                            const uint32_t* const Y, int length,
+                                            int is_X_used, int is_Y_used,
+                                            int trivial_at_end) {
   VP8LStreaks stats;
   if (trivial_at_end) {
     // This configuration is due to palettization that transforms an indexed
@@ -356,16 +357,18 @@ static WEBP_INLINE double GetCombinedEntropy(const uint32_t* const X,
 }
 
 // Estimates the Entropy + Huffman + other block overhead size cost.
-double VP8LHistogramEstimateBits(VP8LHistogram* const p) {
-  return
-      PopulationCost(p->literal_, VP8LHistogramNumCodes(p->palette_code_bits_),
-                     NULL, &p->is_used_[0])
-      + PopulationCost(p->red_, NUM_LITERAL_CODES, NULL, &p->is_used_[1])
-      + PopulationCost(p->blue_, NUM_LITERAL_CODES, NULL, &p->is_used_[2])
-      + PopulationCost(p->alpha_, NUM_LITERAL_CODES, NULL, &p->is_used_[3])
-      + PopulationCost(p->distance_, NUM_DISTANCE_CODES, NULL, &p->is_used_[4])
-      + VP8LExtraCost(p->literal_ + NUM_LITERAL_CODES, NUM_LENGTH_CODES)
-      + VP8LExtraCost(p->distance_, NUM_DISTANCE_CODES);
+float VP8LHistogramEstimateBits(VP8LHistogram* const p) {
+  return PopulationCost(p->literal_,
+                        VP8LHistogramNumCodes(p->palette_code_bits_), NULL,
+                        &p->is_used_[0]) +
+         PopulationCost(p->red_, NUM_LITERAL_CODES, NULL, &p->is_used_[1]) +
+         PopulationCost(p->blue_, NUM_LITERAL_CODES, NULL, &p->is_used_[2]) +
+         PopulationCost(p->alpha_, NUM_LITERAL_CODES, NULL, &p->is_used_[3]) +
+         PopulationCost(p->distance_, NUM_DISTANCE_CODES, NULL,
+                        &p->is_used_[4]) +
+         (float)VP8LExtraCost(p->literal_ + NUM_LITERAL_CODES,
+                              NUM_LENGTH_CODES) +
+         (float)VP8LExtraCost(p->distance_, NUM_DISTANCE_CODES);
 }
 
 // -----------------------------------------------------------------------------
@@ -373,17 +376,16 @@ double VP8LHistogramEstimateBits(VP8LHistogram* const p) {
 
 static int GetCombinedHistogramEntropy(const VP8LHistogram* const a,
                                        const VP8LHistogram* const b,
-                                       double cost_threshold,
-                                       double* cost) {
+                                       float cost_threshold, float* cost) {
   const int palette_code_bits = a->palette_code_bits_;
   int trivial_at_end = 0;
   assert(a->palette_code_bits_ == b->palette_code_bits_);
   *cost += GetCombinedEntropy(a->literal_, b->literal_,
                               VP8LHistogramNumCodes(palette_code_bits),
                               a->is_used_[0], b->is_used_[0], 0);
-  *cost += VP8LExtraCostCombined(a->literal_ + NUM_LITERAL_CODES,
-                                 b->literal_ + NUM_LITERAL_CODES,
-                                 NUM_LENGTH_CODES);
+  *cost += (float)VP8LExtraCostCombined(a->literal_ + NUM_LITERAL_CODES,
+                                        b->literal_ + NUM_LITERAL_CODES,
+                                        NUM_LENGTH_CODES);
   if (*cost > cost_threshold) return 0;
 
   if (a->trivial_symbol_ != VP8L_NON_TRIVIAL_SYM &&
@@ -417,8 +419,8 @@ static int GetCombinedHistogramEntropy(const VP8LHistogram* const a,
   *cost +=
       GetCombinedEntropy(a->distance_, b->distance_, NUM_DISTANCE_CODES,
                          a->is_used_[4], b->is_used_[4], 0);
-  *cost +=
-      VP8LExtraCostCombined(a->distance_, b->distance_, NUM_DISTANCE_CODES);
+  *cost += (float)VP8LExtraCostCombined(a->distance_, b->distance_,
+                                        NUM_DISTANCE_CODES);
   if (*cost > cost_threshold) return 0;
 
   return 1;
@@ -439,12 +441,11 @@ static WEBP_INLINE void HistogramAdd(const VP8LHistogram* const a,
 // Since the previous score passed is 'cost_threshold', we only need to compare
 // the partial cost against 'cost_threshold + C(a) + C(b)' to possibly bail-out
 // early.
-static double HistogramAddEval(const VP8LHistogram* const a,
-                               const VP8LHistogram* const b,
-                               VP8LHistogram* const out,
-                               double cost_threshold) {
-  double cost = 0;
-  const double sum_cost = a->bit_cost_ + b->bit_cost_;
+static float HistogramAddEval(const VP8LHistogram* const a,
+                              const VP8LHistogram* const b,
+                              VP8LHistogram* const out, float cost_threshold) {
+  float cost = 0;
+  const float sum_cost = a->bit_cost_ + b->bit_cost_;
   cost_threshold += sum_cost;
 
   if (GetCombinedHistogramEntropy(a, b, cost_threshold, &cost)) {
@@ -459,10 +460,10 @@ static double HistogramAddEval(const VP8LHistogram* const a,
 // Same as HistogramAddEval(), except that the resulting histogram
 // is not stored. Only the cost C(a+b) - C(a) is evaluated. We omit
 // the term C(b) which is constant over all the evaluations.
-static double HistogramAddThresh(const VP8LHistogram* const a,
-                                 const VP8LHistogram* const b,
-                                 double cost_threshold) {
-  double cost;
+static float HistogramAddThresh(const VP8LHistogram* const a,
+                                const VP8LHistogram* const b,
+                                float cost_threshold) {
+  float cost;
   assert(a != NULL && b != NULL);
   cost = -a->bit_cost_;
   GetCombinedHistogramEntropy(a, b, cost_threshold, &cost);
@@ -473,24 +474,22 @@ static double HistogramAddThresh(const VP8LHistogram* const a,
 
 // The structure to keep track of cost range for the three dominant entropy
 // symbols.
-// TODO(skal): Evaluate if float can be used here instead of double for
-// representing the entropy costs.
 typedef struct {
-  double literal_max_;
-  double literal_min_;
-  double red_max_;
-  double red_min_;
-  double blue_max_;
-  double blue_min_;
+  float literal_max_;
+  float literal_min_;
+  float red_max_;
+  float red_min_;
+  float blue_max_;
+  float blue_min_;
 } DominantCostRange;
 
 static void DominantCostRangeInit(DominantCostRange* const c) {
   c->literal_max_ = 0.;
-  c->literal_min_ = MAX_COST;
+  c->literal_min_ = MAX_BIT_COST;
   c->red_max_ = 0.;
-  c->red_min_ = MAX_COST;
+  c->red_min_ = MAX_BIT_COST;
   c->blue_max_ = 0.;
-  c->blue_min_ = MAX_COST;
+  c->blue_min_ = MAX_BIT_COST;
 }
 
 static void UpdateDominantCostRange(
@@ -505,16 +504,15 @@ static void UpdateDominantCostRange(
 
 static void UpdateHistogramCost(VP8LHistogram* const h) {
   uint32_t alpha_sym, red_sym, blue_sym;
-  const double alpha_cost =
-      PopulationCost(h->alpha_, NUM_LITERAL_CODES, &alpha_sym,
-                     &h->is_used_[3]);
-  const double distance_cost =
+  const float alpha_cost =
+      PopulationCost(h->alpha_, NUM_LITERAL_CODES, &alpha_sym, &h->is_used_[3]);
+  const float distance_cost =
       PopulationCost(h->distance_, NUM_DISTANCE_CODES, NULL, &h->is_used_[4]) +
-      VP8LExtraCost(h->distance_, NUM_DISTANCE_CODES);
+      (float)VP8LExtraCost(h->distance_, NUM_DISTANCE_CODES);
   const int num_codes = VP8LHistogramNumCodes(h->palette_code_bits_);
   h->literal_cost_ =
       PopulationCost(h->literal_, num_codes, NULL, &h->is_used_[0]) +
-          VP8LExtraCost(h->literal_ + NUM_LITERAL_CODES, NUM_LENGTH_CODES);
+      (float)VP8LExtraCost(h->literal_ + NUM_LITERAL_CODES, NUM_LENGTH_CODES);
   h->red_cost_ =
       PopulationCost(h->red_, NUM_LITERAL_CODES, &red_sym, &h->is_used_[1]);
   h->blue_cost_ =
@@ -529,10 +527,10 @@ static void UpdateHistogramCost(VP8LHistogram* const h) {
   }
 }
 
-static int GetBinIdForEntropy(double min, double max, double val) {
-  const double range = max - min;
+static int GetBinIdForEntropy(float min, float max, float val) {
+  const float range = max - min;
   if (range > 0.) {
-    const double delta = val - min;
+    const float delta = val - min;
     return (int)((NUM_PARTITIONS - 1e-6) * delta / range);
   } else {
     return 0;
@@ -641,15 +639,11 @@ static void HistogramAnalyzeEntropyBin(VP8LHistogramSet* const image_histo,
 
 // Merges some histograms with same bin_id together if it's advantageous.
 // Sets the remaining histograms to NULL.
-static void HistogramCombineEntropyBin(VP8LHistogramSet* const image_histo,
-                                       int* num_used,
-                                       const uint16_t* const clusters,
-                                       uint16_t* const cluster_mappings,
-                                       VP8LHistogram* cur_combo,
-                                       const uint16_t* const bin_map,
-                                       int num_bins,
-                                       double combine_cost_factor,
-                                       int low_effort) {
+static void HistogramCombineEntropyBin(
+    VP8LHistogramSet* const image_histo, int* num_used,
+    const uint16_t* const clusters, uint16_t* const cluster_mappings,
+    VP8LHistogram* cur_combo, const uint16_t* const bin_map, int num_bins,
+    float combine_cost_factor, int low_effort) {
   VP8LHistogram** const histograms = image_histo->histograms;
   int idx;
   struct {
@@ -679,11 +673,10 @@ static void HistogramCombineEntropyBin(VP8LHistogramSet* const image_histo,
       cluster_mappings[clusters[idx]] = clusters[first];
     } else {
       // try to merge #idx into #first (both share the same bin_id)
-      const double bit_cost = histograms[idx]->bit_cost_;
-      const double bit_cost_thresh = -bit_cost * combine_cost_factor;
-      const double curr_cost_diff =
-          HistogramAddEval(histograms[first], histograms[idx],
-                           cur_combo, bit_cost_thresh);
+      const float bit_cost = histograms[idx]->bit_cost_;
+      const float bit_cost_thresh = -bit_cost * combine_cost_factor;
+      const float curr_cost_diff = HistogramAddEval(
+          histograms[first], histograms[idx], cur_combo, bit_cost_thresh);
       if (curr_cost_diff < bit_cost_thresh) {
         // Try to merge two histograms only if the combo is a trivial one or
         // the two candidate histograms are already non-trivial.
@@ -731,8 +724,8 @@ static uint32_t MyRand(uint32_t* const seed) {
 typedef struct {
   int idx1;
   int idx2;
-  double cost_diff;
-  double cost_combo;
+  float cost_diff;
+  float cost_combo;
 } HistogramPair;
 
 typedef struct {
@@ -787,10 +780,9 @@ static void HistoQueueUpdateHead(HistoQueue* const histo_queue,
 // Update the cost diff and combo of a pair of histograms. This needs to be
 // called when the the histograms have been merged with a third one.
 static void HistoQueueUpdatePair(const VP8LHistogram* const h1,
-                                 const VP8LHistogram* const h2,
-                                 double threshold,
+                                 const VP8LHistogram* const h2, float threshold,
                                  HistogramPair* const pair) {
-  const double sum_cost = h1->bit_cost_ + h2->bit_cost_;
+  const float sum_cost = h1->bit_cost_ + h2->bit_cost_;
   pair->cost_combo = 0.;
   GetCombinedHistogramEntropy(h1, h2, sum_cost + threshold, &pair->cost_combo);
   pair->cost_diff = pair->cost_combo - sum_cost;
@@ -799,9 +791,9 @@ static void HistoQueueUpdatePair(const VP8LHistogram* const h1,
 // Create a pair from indices "idx1" and "idx2" provided its cost
 // is inferior to "threshold", a negative entropy.
 // It returns the cost of the pair, or 0. if it superior to threshold.
-static double HistoQueuePush(HistoQueue* const histo_queue,
-                             VP8LHistogram** const histograms, int idx1,
-                             int idx2, double threshold) {
+static float HistoQueuePush(HistoQueue* const histo_queue,
+                            VP8LHistogram** const histograms, int idx1,
+                            int idx2, float threshold) {
   const VP8LHistogram* h1;
   const VP8LHistogram* h2;
   HistogramPair pair;
@@ -945,8 +937,8 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
            ++tries_with_no_success < num_tries_no_success;
        ++iter) {
     int* mapping_index;
-    double best_cost =
-        (histo_queue.size == 0) ? 0. : histo_queue.queue[0].cost_diff;
+    float best_cost =
+        (histo_queue.size == 0) ? 0.f : histo_queue.queue[0].cost_diff;
     int best_idx1 = -1, best_idx2 = 1;
     const uint32_t rand_range = (*num_used - 1) * (*num_used);
     // (*num_used) / 2 was chosen empirically. Less means faster but worse
@@ -955,7 +947,7 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
 
     // Pick random samples.
     for (j = 0; *num_used >= 2 && j < num_tries; ++j) {
-      double curr_cost;
+      float curr_cost;
       // Choose two different histograms at random and try to combine them.
       const uint32_t tmp = MyRand(&seed) % rand_range;
       uint32_t idx1 = tmp / (*num_used - 1);
@@ -1034,7 +1026,7 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
   *do_greedy = (*num_used <= min_cluster_size);
   ok = 1;
 
-End:
+ End:
   HistoQueueClear(&histo_queue);
   WebPSafeFree(mappings);
   return ok;
@@ -1057,7 +1049,7 @@ static void HistogramRemap(const VP8LHistogramSet* const in,
   if (out_size > 1) {
     for (i = 0; i < in_size; ++i) {
       int best_out = 0;
-      double best_bits = MAX_COST;
+      float best_bits = MAX_BIT_COST;
       int k;
       if (in_histo[i] == NULL) {
         // Arbitrarily set to the previous value if unused to help future LZ77.
@@ -1065,7 +1057,7 @@ static void HistogramRemap(const VP8LHistogramSet* const in,
         continue;
       }
       for (k = 0; k < out_size; ++k) {
-        double cur_bits;
+        float cur_bits;
         cur_bits = HistogramAddThresh(out_histo[k], in_histo[i], best_bits);
         if (k == 0 || cur_bits < best_bits) {
           best_bits = cur_bits;
@@ -1093,13 +1085,13 @@ static void HistogramRemap(const VP8LHistogramSet* const in,
   }
 }
 
-static double GetCombineCostFactor(int histo_size, int quality) {
-  double combine_cost_factor = 0.16;
+static float GetCombineCostFactor(int histo_size, int quality) {
+  float combine_cost_factor = 0.16f;
   if (quality < 90) {
-    if (histo_size > 256) combine_cost_factor /= 2.;
-    if (histo_size > 512) combine_cost_factor /= 2.;
-    if (histo_size > 1024) combine_cost_factor /= 2.;
-    if (quality <= 50) combine_cost_factor /= 2.;
+    if (histo_size > 256) combine_cost_factor /= 2.f;
+    if (histo_size > 512) combine_cost_factor /= 2.f;
+    if (histo_size > 1024) combine_cost_factor /= 2.f;
+    if (quality <= 50) combine_cost_factor /= 2.f;
   }
   return combine_cost_factor;
 }
@@ -1169,15 +1161,17 @@ static void RemoveEmptyHistograms(VP8LHistogramSet* const image_histo) {
 }
 
 int VP8LGetHistoImageSymbols(int xsize, int ysize,
-                             const VP8LBackwardRefs* const refs,
-                             int quality, int low_effort,
-                             int histo_bits, int cache_bits,
+                             const VP8LBackwardRefs* const refs, int quality,
+                             int low_effort, int histogram_bits, int cache_bits,
                              VP8LHistogramSet* const image_histo,
                              VP8LHistogram* const tmp_histo,
-                             uint16_t* const histogram_symbols) {
-  int ok = 0;
-  const int histo_xsize = histo_bits ? VP8LSubSampleSize(xsize, histo_bits) : 1;
-  const int histo_ysize = histo_bits ? VP8LSubSampleSize(ysize, histo_bits) : 1;
+                             uint16_t* const histogram_symbols,
+                             const WebPPicture* const pic, int percent_range,
+                             int* const percent) {
+  const int histo_xsize =
+      histogram_bits ? VP8LSubSampleSize(xsize, histogram_bits) : 1;
+  const int histo_ysize =
+      histogram_bits ? VP8LSubSampleSize(ysize, histogram_bits) : 1;
   const int image_histo_raw_size = histo_xsize * histo_ysize;
   VP8LHistogramSet* const orig_histo =
       VP8LAllocateHistogramSet(image_histo_raw_size, cache_bits);
@@ -1187,13 +1181,16 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
   const int entropy_combine_num_bins = low_effort ? NUM_PARTITIONS : BIN_SIZE;
   int entropy_combine;
   uint16_t* const map_tmp =
-      WebPSafeMalloc(2 * image_histo_raw_size, sizeof(map_tmp));
+      WebPSafeMalloc(2 * image_histo_raw_size, sizeof(*map_tmp));
   uint16_t* const cluster_mappings = map_tmp + image_histo_raw_size;
   int num_used = image_histo_raw_size;
-  if (orig_histo == NULL || map_tmp == NULL) goto Error;
+  if (orig_histo == NULL || map_tmp == NULL) {
+    WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
+    goto Error;
+  }
 
   // Construct the histograms from backward references.
-  HistogramBuild(xsize, histo_bits, refs, orig_histo);
+  HistogramBuild(xsize, histogram_bits, refs, orig_histo);
   // Copies the histograms and computes its bit_cost.
   // histogram_symbols is optimized
   HistogramCopyAndAnalyze(orig_histo, image_histo, &num_used,
@@ -1204,16 +1201,15 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
 
   if (entropy_combine) {
     uint16_t* const bin_map = map_tmp;
-    const double combine_cost_factor =
+    const float combine_cost_factor =
         GetCombineCostFactor(image_histo_raw_size, quality);
     const uint32_t num_clusters = num_used;
 
     HistogramAnalyzeEntropyBin(image_histo, bin_map, low_effort);
     // Collapse histograms with similar entropy.
-    HistogramCombineEntropyBin(image_histo, &num_used, histogram_symbols,
-                               cluster_mappings, tmp_histo, bin_map,
-                               entropy_combine_num_bins, combine_cost_factor,
-                               low_effort);
+    HistogramCombineEntropyBin(
+        image_histo, &num_used, histogram_symbols, cluster_mappings, tmp_histo,
+        bin_map, entropy_combine_num_bins, combine_cost_factor, low_effort);
     OptimizeHistogramSymbols(image_histo, cluster_mappings, num_clusters,
                              map_tmp, histogram_symbols);
   }
@@ -1227,11 +1223,13 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
     int do_greedy;
     if (!HistogramCombineStochastic(image_histo, &num_used, threshold_size,
                                     &do_greedy)) {
+      WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
       goto Error;
     }
     if (do_greedy) {
       RemoveEmptyHistograms(image_histo);
       if (!HistogramCombineGreedy(image_histo, &num_used)) {
+        WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
         goto Error;
       }
     }
@@ -1241,10 +1239,12 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
   RemoveEmptyHistograms(image_histo);
   HistogramRemap(orig_histo, image_histo, histogram_symbols);
 
-  ok = 1;
+  if (!WebPReportProgress(pic, *percent + percent_range, percent)) {
+    goto Error;
+  }
 
  Error:
   VP8LFreeHistogramSet(orig_histo);
   WebPSafeFree(map_tmp);
-  return ok;
+  return (pic->error_code == VP8_ENC_OK);
 }

3rdparty/libwebp/src/enc/histogram_enc.h

@@ -40,10 +40,10 @@ typedef struct {
   int palette_code_bits_;
   uint32_t trivial_symbol_;  // True, if histograms for Red, Blue & Alpha
                              // literal symbols are single valued.
-  double bit_cost_;          // cached value of bit cost.
-  double literal_cost_;      // Cached values of dominant entropy costs:
-  double red_cost_;          // literal, red & blue.
-  double blue_cost_;
+  float bit_cost_;           // cached value of bit cost.
+  float literal_cost_;       // Cached values of dominant entropy costs:
+  float red_cost_;           // literal, red & blue.
+  float blue_cost_;
   uint8_t is_used_[5];       // 5 for literal, red, blue, alpha, distance
 } VP8LHistogram;
 
@@ -64,8 +64,8 @@ void VP8LHistogramCreate(VP8LHistogram* const p,
                          const VP8LBackwardRefs* const refs,
                          int palette_code_bits);
 
-// Return the size of the histogram for a given palette_code_bits.
-int VP8LGetHistogramSize(int palette_code_bits);
+// Return the size of the histogram for a given cache_bits.
+int VP8LGetHistogramSize(int cache_bits);
 
 // Set the palette_code_bits and reset the stats.
 // If init_arrays is true, the arrays are also filled with 0's.
@@ -105,21 +105,23 @@ static WEBP_INLINE int VP8LHistogramNumCodes(int palette_code_bits) {
       ((palette_code_bits > 0) ? (1 << palette_code_bits) : 0);
 }
 
-// Builds the histogram image.
+// Builds the histogram image. pic and percent are for progress.
+// Returns false in case of error (stored in pic->error_code).
 int VP8LGetHistoImageSymbols(int xsize, int ysize,
-                             const VP8LBackwardRefs* const refs,
-                             int quality, int low_effort,
-                             int histogram_bits, int cache_bits,
-                             VP8LHistogramSet* const image_in,
+                             const VP8LBackwardRefs* const refs, int quality,
+                             int low_effort, int histogram_bits, int cache_bits,
+                             VP8LHistogramSet* const image_histo,
                              VP8LHistogram* const tmp_histo,
-                             uint16_t* const histogram_symbols);
+                             uint16_t* const histogram_symbols,
+                             const WebPPicture* const pic, int percent_range,
+                             int* const percent);
 
 // Returns the entropy for the symbols in the input array.
-double VP8LBitsEntropy(const uint32_t* const array, int n);
+float VP8LBitsEntropy(const uint32_t* const array, int n);
 
 // Estimate how many bits the combined entropy of literals and distance
 // approximately maps to.
-double VP8LHistogramEstimateBits(VP8LHistogram* const p);
+float VP8LHistogramEstimateBits(VP8LHistogram* const p);
 
 #ifdef __cplusplus
 }

3rdparty/libwebp/src/enc/picture_csp_enc.c

@@ -15,12 +15,19 @@
 #include <stdlib.h>
 #include <math.h>
 
+#include "sharpyuv/sharpyuv.h"
+#include "sharpyuv/sharpyuv_csp.h"
 #include "src/enc/vp8i_enc.h"
 #include "src/utils/random_utils.h"
 #include "src/utils/utils.h"
 #include "src/dsp/dsp.h"
 #include "src/dsp/lossless.h"
 #include "src/dsp/yuv.h"
+#include "src/dsp/cpu.h"
+
+#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
+#include <pthread.h>
+#endif
 
 // Uncomment to disable gamma-compression during RGB->U/V averaging
 #define USE_GAMMA_COMPRESSION
@@ -62,10 +69,12 @@ static int CheckNonOpaque(const uint8_t* alpha, int width, int height,
 int WebPPictureHasTransparency(const WebPPicture* picture) {
   if (picture == NULL) return 0;
   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));
+    if (picture->argb != NULL) {
+      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);
@@ -76,30 +85,31 @@ int WebPPictureHasTransparency(const WebPPicture* picture) {
 
 #if defined(USE_GAMMA_COMPRESSION)
 
-// gamma-compensates loss of resolution during chroma subsampling
-#define kGamma 0.80      // for now we use a different gamma value than kGammaF
-#define kGammaFix 12     // fixed-point precision for linear values
-#define kGammaScale ((1 << kGammaFix) - 1)
-#define kGammaTabFix 7   // fixed-point fractional bits precision
-#define kGammaTabScale (1 << kGammaTabFix)
-#define kGammaTabRounder (kGammaTabScale >> 1)
-#define kGammaTabSize (1 << (kGammaFix - kGammaTabFix))
+// Gamma correction compensates loss of resolution during chroma subsampling.
+#define GAMMA_FIX 12      // fixed-point precision for linear values
+#define GAMMA_TAB_FIX 7   // fixed-point fractional bits precision
+#define GAMMA_TAB_SIZE (1 << (GAMMA_FIX - GAMMA_TAB_FIX))
+static const double kGamma = 0.80;
+static const int kGammaScale = ((1 << GAMMA_FIX) - 1);
+static const int kGammaTabScale = (1 << GAMMA_TAB_FIX);
+static const int kGammaTabRounder = (1 << GAMMA_TAB_FIX >> 1);
 
-static int kLinearToGammaTab[kGammaTabSize + 1];
+static int kLinearToGammaTab[GAMMA_TAB_SIZE + 1];
 static uint16_t kGammaToLinearTab[256];
 static volatile int kGammaTablesOk = 0;
 static void InitGammaTables(void);
+extern VP8CPUInfo VP8GetCPUInfo;
 
 WEBP_DSP_INIT_FUNC(InitGammaTables) {
   if (!kGammaTablesOk) {
     int v;
-    const double scale = (double)(1 << kGammaTabFix) / kGammaScale;
+    const double scale = (double)(1 << GAMMA_TAB_FIX) / kGammaScale;
     const double norm = 1. / 255.;
     for (v = 0; v <= 255; ++v) {
       kGammaToLinearTab[v] =
           (uint16_t)(pow(norm * v, kGamma) * kGammaScale + .5);
     }
-    for (v = 0; v <= kGammaTabSize; ++v) {
+    for (v = 0; v <= GAMMA_TAB_SIZE; ++v) {
       kLinearToGammaTab[v] = (int)(255. * pow(scale * v, 1. / kGamma) + .5);
     }
     kGammaTablesOk = 1;
@@ -111,12 +121,12 @@ static WEBP_INLINE uint32_t GammaToLinear(uint8_t v) {
 }
 
 static WEBP_INLINE int Interpolate(int v) {
-  const int tab_pos = v >> (kGammaTabFix + 2);    // integer part
+  const int tab_pos = v >> (GAMMA_TAB_FIX + 2);    // integer part
   const int x = v & ((kGammaTabScale << 2) - 1);  // fractional part
   const int v0 = kLinearToGammaTab[tab_pos];
   const int v1 = kLinearToGammaTab[tab_pos + 1];
   const int y = v1 * x + v0 * ((kGammaTabScale << 2) - x);   // interpolate
-  assert(tab_pos + 1 < kGammaTabSize + 1);
+  assert(tab_pos + 1 < GAMMA_TAB_SIZE + 1);
   return y;
 }
 
@@ -124,7 +134,7 @@ static WEBP_INLINE int Interpolate(int v) {
 // U/V value, suitable for RGBToU/V calls.
 static WEBP_INLINE int LinearToGamma(uint32_t base_value, int shift) {
   const int y = Interpolate(base_value << shift);   // final uplifted value
-  return (y + kGammaTabRounder) >> kGammaTabFix;    // descale
+  return (y + kGammaTabRounder) >> GAMMA_TAB_FIX;    // descale
 }
 
 #else
@@ -158,415 +168,26 @@ static int RGBToV(int r, int g, int b, VP8Random* const rg) {
 //------------------------------------------------------------------------------
 // Sharp RGB->YUV conversion
 
-static const int kNumIterations = 4;
 static const int kMinDimensionIterativeConversion = 4;
 
-// We could use SFIX=0 and only uint8_t for fixed_y_t, but it produces some
-// banding sometimes. Better use extra precision.
-#define SFIX 2                // fixed-point precision of RGB and Y/W
-typedef int16_t fixed_t;      // signed type with extra SFIX precision for UV
-typedef uint16_t fixed_y_t;   // unsigned type with extra SFIX precision for W
-
-#define SHALF (1 << SFIX >> 1)
-#define MAX_Y_T ((256 << SFIX) - 1)
-#define SROUNDER (1 << (YUV_FIX + SFIX - 1))
-
-#if defined(USE_GAMMA_COMPRESSION)
-
-// We use tables of different size and precision for the Rec709 / BT2020
-// transfer function.
-#define kGammaF (1./0.45)
-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);
-
-WEBP_DSP_INIT_FUNC(InitGammaTablesS) {
-  assert(2 * GAMMA_TO_LINEAR_BITS < 32);  // we use uint32_t intermediate values
-  if (!kGammaTablesSOk) {
-    int v;
-    const double norm = 1. / MAX_Y_T;
-    const double scale = 1. / kGammaTabSize;
-    const double a = 0.09929682680944;
-    const double thresh = 0.018053968510807;
-    const double final_scale = 1 << GAMMA_TO_LINEAR_BITS;
-    for (v = 0; v <= MAX_Y_T; ++v) {
-      const double g = norm * v;
-      double value;
-      if (g <= thresh * 4.5) {
-        value = g / 4.5;
-      } else {
-        const double a_rec = 1. / (1. + a);
-        value = pow(a_rec * (g + a), kGammaF);
-      }
-      kGammaToLinearTabS[v] = (uint32_t)(value * final_scale + .5);
-    }
-    for (v = 0; v <= kGammaTabSize; ++v) {
-      const double g = scale * v;
-      double value;
-      if (g <= thresh) {
-        value = 4.5 * g;
-      } else {
-        value = (1. + a) * pow(g, 1. / kGammaF) - a;
-      }
-      // we already incorporate the 1/2 rounding constant here
-      kLinearToGammaTabS[v] =
-          (uint32_t)(MAX_Y_T * value) + (1 << GAMMA_TO_LINEAR_BITS >> 1);
-    }
-    // to prevent small rounding errors to cause read-overflow:
-    kLinearToGammaTabS[kGammaTabSize + 1] = kLinearToGammaTabS[kGammaTabSize];
-    kGammaTablesSOk = 1;
-  }
-}
-
-// return value has a fixed-point precision of GAMMA_TO_LINEAR_BITS
-static WEBP_INLINE uint32_t GammaToLinearS(int v) {
-  return kGammaToLinearTabS[v];
-}
-
-static WEBP_INLINE uint32_t LinearToGammaS(uint32_t value) {
-  // 'value' is in GAMMA_TO_LINEAR_BITS fractional precision
-  const uint32_t v = value * kGammaTabSize;
-  const uint32_t tab_pos = v >> GAMMA_TO_LINEAR_BITS;
-  // fractional part, in GAMMA_TO_LINEAR_BITS fixed-point precision
-  const uint32_t x = v - (tab_pos << GAMMA_TO_LINEAR_BITS);  // fractional part
-  // v0 / v1 are in GAMMA_TO_LINEAR_BITS fixed-point precision (range [0..1])
-  const uint32_t v0 = kLinearToGammaTabS[tab_pos + 0];
-  const uint32_t v1 = kLinearToGammaTabS[tab_pos + 1];
-  // Final interpolation. Note that rounding is already included.
-  const uint32_t v2 = (v1 - v0) * x;    // note: v1 >= v0.
-  const uint32_t result = v0 + (v2 >> GAMMA_TO_LINEAR_BITS);
-  return result;
-}
-
-#else
-
-static void InitGammaTablesS(void) {}
-static WEBP_INLINE uint32_t GammaToLinearS(int v) {
-  return (v << GAMMA_TO_LINEAR_BITS) / MAX_Y_T;
-}
-static WEBP_INLINE uint32_t LinearToGammaS(uint32_t value) {
-  return (MAX_Y_T * value) >> GAMMA_TO_LINEAR_BITS;
-}
-
-#endif    // USE_GAMMA_COMPRESSION
-
-//------------------------------------------------------------------------------
-
-static uint8_t clip_8b(fixed_t v) {
-  return (!(v & ~0xff)) ? (uint8_t)v : (v < 0) ? 0u : 255u;
-}
-
-static fixed_y_t clip_y(int y) {
-  return (!(y & ~MAX_Y_T)) ? (fixed_y_t)y : (y < 0) ? 0 : MAX_Y_T;
-}
-
-//------------------------------------------------------------------------------
-
-static int RGBToGray(int r, int g, int b) {
-  const int luma = 13933 * r + 46871 * g + 4732 * b + YUV_HALF;
-  return (luma >> YUV_FIX);
-}
-
-static uint32_t ScaleDown(int a, int b, int c, int d) {
-  const uint32_t A = GammaToLinearS(a);
-  const uint32_t B = GammaToLinearS(b);
-  const uint32_t C = GammaToLinearS(c);
-  const uint32_t D = GammaToLinearS(d);
-  return LinearToGammaS((A + B + C + D + 2) >> 2);
-}
-
-static WEBP_INLINE void UpdateW(const fixed_y_t* src, fixed_y_t* dst, int w) {
-  int i;
-  for (i = 0; i < w; ++i) {
-    const uint32_t R = GammaToLinearS(src[0 * w + i]);
-    const uint32_t G = GammaToLinearS(src[1 * w + i]);
-    const uint32_t B = GammaToLinearS(src[2 * w + i]);
-    const uint32_t Y = RGBToGray(R, G, B);
-    dst[i] = (fixed_y_t)LinearToGammaS(Y);
-  }
-}
-
-static void UpdateChroma(const fixed_y_t* src1, const fixed_y_t* src2,
-                         fixed_t* dst, int uv_w) {
-  int i;
-  for (i = 0; i < uv_w; ++i) {
-    const int r = ScaleDown(src1[0 * uv_w + 0], src1[0 * uv_w + 1],
-                            src2[0 * uv_w + 0], src2[0 * uv_w + 1]);
-    const int g = ScaleDown(src1[2 * uv_w + 0], src1[2 * uv_w + 1],
-                            src2[2 * uv_w + 0], src2[2 * uv_w + 1]);
-    const int b = ScaleDown(src1[4 * uv_w + 0], src1[4 * uv_w + 1],
-                            src2[4 * uv_w + 0], src2[4 * uv_w + 1]);
-    const int W = RGBToGray(r, g, b);
-    dst[0 * uv_w] = (fixed_t)(r - W);
-    dst[1 * uv_w] = (fixed_t)(g - W);
-    dst[2 * uv_w] = (fixed_t)(b - W);
-    dst  += 1;
-    src1 += 2;
-    src2 += 2;
-  }
-}
-
-static void StoreGray(const fixed_y_t* rgb, fixed_y_t* y, int w) {
-  int i;
-  for (i = 0; i < w; ++i) {
-    y[i] = RGBToGray(rgb[0 * w + i], rgb[1 * w + i], rgb[2 * w + i]);
-  }
-}
-
-//------------------------------------------------------------------------------
-
-static WEBP_INLINE fixed_y_t Filter2(int A, int B, int W0) {
-  const int v0 = (A * 3 + B + 2) >> 2;
-  return clip_y(v0 + W0);
-}
-
-//------------------------------------------------------------------------------
-
-static WEBP_INLINE fixed_y_t UpLift(uint8_t a) {  // 8bit -> SFIX
-  return ((fixed_y_t)a << SFIX) | SHALF;
-}
-
-static void ImportOneRow(const uint8_t* const r_ptr,
-                         const uint8_t* const g_ptr,
-                         const uint8_t* const b_ptr,
-                         int step,
-                         int pic_width,
-                         fixed_y_t* const dst) {
-  int i;
-  const int w = (pic_width + 1) & ~1;
-  for (i = 0; i < pic_width; ++i) {
-    const int off = i * step;
-    dst[i + 0 * w] = UpLift(r_ptr[off]);
-    dst[i + 1 * w] = UpLift(g_ptr[off]);
-    dst[i + 2 * w] = UpLift(b_ptr[off]);
-  }
-  if (pic_width & 1) {  // replicate rightmost pixel
-    dst[pic_width + 0 * w] = dst[pic_width + 0 * w - 1];
-    dst[pic_width + 1 * w] = dst[pic_width + 1 * w - 1];
-    dst[pic_width + 2 * w] = dst[pic_width + 2 * w - 1];
-  }
-}
-
-static void InterpolateTwoRows(const fixed_y_t* const best_y,
-                               const fixed_t* prev_uv,
-                               const fixed_t* cur_uv,
-                               const fixed_t* next_uv,
-                               int w,
-                               fixed_y_t* out1,
-                               fixed_y_t* out2) {
-  const int uv_w = w >> 1;
-  const int len = (w - 1) >> 1;   // length to filter
-  int k = 3;
-  while (k-- > 0) {   // process each R/G/B segments in turn
-    // special boundary case for i==0
-    out1[0] = Filter2(cur_uv[0], prev_uv[0], best_y[0]);
-    out2[0] = Filter2(cur_uv[0], next_uv[0], best_y[w]);
-
-    WebPSharpYUVFilterRow(cur_uv, prev_uv, len, best_y + 0 + 1, out1 + 1);
-    WebPSharpYUVFilterRow(cur_uv, next_uv, len, best_y + w + 1, out2 + 1);
-
-    // special boundary case for i == w - 1 when w is even
-    if (!(w & 1)) {
-      out1[w - 1] = Filter2(cur_uv[uv_w - 1], prev_uv[uv_w - 1],
-                            best_y[w - 1 + 0]);
-      out2[w - 1] = Filter2(cur_uv[uv_w - 1], next_uv[uv_w - 1],
-                            best_y[w - 1 + w]);
-    }
-    out1 += w;
-    out2 += w;
-    prev_uv += uv_w;
-    cur_uv  += uv_w;
-    next_uv += uv_w;
-  }
-}
-
-static WEBP_INLINE uint8_t ConvertRGBToY(int r, int g, int b) {
-  const int luma = 16839 * r + 33059 * g + 6420 * b + SROUNDER;
-  return clip_8b(16 + (luma >> (YUV_FIX + SFIX)));
-}
-
-static WEBP_INLINE uint8_t ConvertRGBToU(int r, int g, int b) {
-  const int u =  -9719 * r - 19081 * g + 28800 * b + SROUNDER;
-  return clip_8b(128 + (u >> (YUV_FIX + SFIX)));
-}
-
-static WEBP_INLINE uint8_t ConvertRGBToV(int r, int g, int b) {
-  const int v = +28800 * r - 24116 * g -  4684 * b + SROUNDER;
-  return clip_8b(128 + (v >> (YUV_FIX + SFIX)));
-}
-
-static int ConvertWRGBToYUV(const fixed_y_t* best_y, const fixed_t* best_uv,
-                            WebPPicture* const picture) {
-  int i, j;
-  uint8_t* dst_y = picture->y;
-  uint8_t* dst_u = picture->u;
-  uint8_t* dst_v = picture->v;
-  const fixed_t* const best_uv_base = best_uv;
-  const int w = (picture->width + 1) & ~1;
-  const int h = (picture->height + 1) & ~1;
-  const int uv_w = w >> 1;
-  const int uv_h = h >> 1;
-  for (best_uv = best_uv_base, j = 0; j < picture->height; ++j) {
-    for (i = 0; i < picture->width; ++i) {
-      const int off = (i >> 1);
-      const int W = best_y[i];
-      const int r = best_uv[off + 0 * uv_w] + W;
-      const int g = best_uv[off + 1 * uv_w] + W;
-      const int b = best_uv[off + 2 * uv_w] + W;
-      dst_y[i] = ConvertRGBToY(r, g, b);
-    }
-    best_y += w;
-    best_uv += (j & 1) * 3 * uv_w;
-    dst_y += picture->y_stride;
-  }
-  for (best_uv = best_uv_base, j = 0; j < uv_h; ++j) {
-    for (i = 0; i < uv_w; ++i) {
-      const int off = i;
-      const int r = best_uv[off + 0 * uv_w];
-      const int g = best_uv[off + 1 * uv_w];
-      const int b = best_uv[off + 2 * uv_w];
-      dst_u[i] = ConvertRGBToU(r, g, b);
-      dst_v[i] = ConvertRGBToV(r, g, b);
-    }
-    best_uv += 3 * uv_w;
-    dst_u += picture->uv_stride;
-    dst_v += picture->uv_stride;
-  }
-  return 1;
-}
-
 //------------------------------------------------------------------------------
 // Main function
 
-#define SAFE_ALLOC(W, H, T) ((T*)WebPSafeMalloc((W) * (H), sizeof(T)))
-
 static int PreprocessARGB(const uint8_t* r_ptr,
                           const uint8_t* g_ptr,
                           const uint8_t* b_ptr,
                           int step, int rgb_stride,
                           WebPPicture* const picture) {
-  // we expand the right/bottom border if needed
-  const int w = (picture->width + 1) & ~1;
-  const int h = (picture->height + 1) & ~1;
-  const int uv_w = w >> 1;
-  const int uv_h = h >> 1;
-  uint64_t prev_diff_y_sum = ~0;
-  int j, iter;
-
-  // TODO(skal): allocate one big memory chunk. But for now, it's easier
-  // for valgrind debugging to have several chunks.
-  fixed_y_t* const tmp_buffer = SAFE_ALLOC(w * 3, 2, fixed_y_t);   // scratch
-  fixed_y_t* const best_y_base = SAFE_ALLOC(w, h, fixed_y_t);
-  fixed_y_t* const target_y_base = SAFE_ALLOC(w, h, fixed_y_t);
-  fixed_y_t* const best_rgb_y = SAFE_ALLOC(w, 2, fixed_y_t);
-  fixed_t* const best_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t);
-  fixed_t* const target_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t);
-  fixed_t* const best_rgb_uv = SAFE_ALLOC(uv_w * 3, 1, fixed_t);
-  fixed_y_t* best_y = best_y_base;
-  fixed_y_t* target_y = target_y_base;
-  fixed_t* best_uv = best_uv_base;
-  fixed_t* target_uv = target_uv_base;
-  const uint64_t diff_y_threshold = (uint64_t)(3.0 * w * h);
-  int ok;
-
-  if (best_y_base == NULL || best_uv_base == NULL ||
-      target_y_base == NULL || target_uv_base == NULL ||
-      best_rgb_y == NULL || best_rgb_uv == NULL ||
-      tmp_buffer == NULL) {
-    ok = WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY);
-    goto End;
-  }
-  assert(picture->width >= kMinDimensionIterativeConversion);
-  assert(picture->height >= kMinDimensionIterativeConversion);
-
-  WebPInitConvertARGBToYUV();
-
-  // Import RGB samples to W/RGB representation.
-  for (j = 0; j < picture->height; j += 2) {
-    const int is_last_row = (j == picture->height - 1);
-    fixed_y_t* const src1 = tmp_buffer + 0 * w;
-    fixed_y_t* const src2 = tmp_buffer + 3 * w;
-
-    // prepare two rows of input
-    ImportOneRow(r_ptr, g_ptr, b_ptr, step, picture->width, src1);
-    if (!is_last_row) {
-      ImportOneRow(r_ptr + rgb_stride, g_ptr + rgb_stride, b_ptr + rgb_stride,
-                   step, picture->width, src2);
-    } else {
-      memcpy(src2, src1, 3 * w * sizeof(*src2));
-    }
-    StoreGray(src1, best_y + 0, w);
-    StoreGray(src2, best_y + w, w);
-
-    UpdateW(src1, target_y, w);
-    UpdateW(src2, target_y + w, w);
-    UpdateChroma(src1, src2, target_uv, uv_w);
-    memcpy(best_uv, target_uv, 3 * uv_w * sizeof(*best_uv));
-    best_y += 2 * w;
-    best_uv += 3 * uv_w;
-    target_y += 2 * w;
-    target_uv += 3 * uv_w;
-    r_ptr += 2 * rgb_stride;
-    g_ptr += 2 * rgb_stride;
-    b_ptr += 2 * rgb_stride;
-  }
-
-  // Iterate and resolve clipping conflicts.
-  for (iter = 0; iter < kNumIterations; ++iter) {
-    const fixed_t* cur_uv = best_uv_base;
-    const fixed_t* prev_uv = best_uv_base;
-    uint64_t diff_y_sum = 0;
-
-    best_y = best_y_base;
-    best_uv = best_uv_base;
-    target_y = target_y_base;
-    target_uv = target_uv_base;
-    for (j = 0; j < h; j += 2) {
-      fixed_y_t* const src1 = tmp_buffer + 0 * w;
-      fixed_y_t* const src2 = tmp_buffer + 3 * w;
-      {
-        const fixed_t* const next_uv = cur_uv + ((j < h - 2) ? 3 * uv_w : 0);
-        InterpolateTwoRows(best_y, prev_uv, cur_uv, next_uv, w, src1, src2);
-        prev_uv = cur_uv;
-        cur_uv = next_uv;
-      }
-
-      UpdateW(src1, best_rgb_y + 0 * w, w);
-      UpdateW(src2, best_rgb_y + 1 * w, w);
-      UpdateChroma(src1, src2, best_rgb_uv, uv_w);
-
-      // update two rows of Y and one row of RGB
-      diff_y_sum += WebPSharpYUVUpdateY(target_y, best_rgb_y, best_y, 2 * w);
-      WebPSharpYUVUpdateRGB(target_uv, best_rgb_uv, best_uv, 3 * uv_w);
-
-      best_y += 2 * w;
-      best_uv += 3 * uv_w;
-      target_y += 2 * w;
-      target_uv += 3 * uv_w;
-    }
-    // test exit condition
-    if (iter > 0) {
-      if (diff_y_sum < diff_y_threshold) break;
-      if (diff_y_sum > prev_diff_y_sum) break;
-    }
-    prev_diff_y_sum = diff_y_sum;
+  const int ok = SharpYuvConvert(
+      r_ptr, g_ptr, b_ptr, step, rgb_stride, /*rgb_bit_depth=*/8,
+      picture->y, picture->y_stride, picture->u, picture->uv_stride, picture->v,
+      picture->uv_stride, /*yuv_bit_depth=*/8, picture->width,
+      picture->height, SharpYuvGetConversionMatrix(kSharpYuvMatrixWebp));
+  if (!ok) {
+    return WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY);
   }
-  // final reconstruction
-  ok = ConvertWRGBToYUV(best_y_base, best_uv_base, picture);
-
- End:
-  WebPSafeFree(best_y_base);
-  WebPSafeFree(best_uv_base);
-  WebPSafeFree(target_y_base);
-  WebPSafeFree(target_uv_base);
-  WebPSafeFree(best_rgb_y);
-  WebPSafeFree(best_rgb_uv);
-  WebPSafeFree(tmp_buffer);
   return ok;
 }
-#undef SAFE_ALLOC
 
 //------------------------------------------------------------------------------
 // "Fast" regular RGB->YUV
@@ -591,8 +212,8 @@ static const int kAlphaFix = 19;
 // and constant are adjusted very tightly to fit 32b arithmetic.
 // In particular, they use the fact that the operands for 'v / a' are actually
 // derived as v = (a0.p0 + a1.p1 + a2.p2 + a3.p3) and a = a0 + a1 + a2 + a3
-// with ai in [0..255] and pi in [0..1<<kGammaFix). The constraint to avoid
-// overflow is: kGammaFix + kAlphaFix <= 31.
+// with ai in [0..255] and pi in [0..1<<GAMMA_FIX). The constraint to avoid
+// overflow is: GAMMA_FIX + kAlphaFix <= 31.
 static const uint32_t kInvAlpha[4 * 0xff + 1] = {
   0,  /* alpha = 0 */
   524288, 262144, 174762, 131072, 104857, 87381, 74898, 65536,
@@ -818,11 +439,20 @@ static WEBP_INLINE void AccumulateRGB(const uint8_t* const r_ptr,
     dst[0] = SUM4(r_ptr + j, step);
     dst[1] = SUM4(g_ptr + j, step);
     dst[2] = SUM4(b_ptr + j, step);
+    // MemorySanitizer may raise false positives with data that passes through
+    // RGBA32PackedToPlanar_16b_SSE41() due to incorrect modeling of shuffles.
+    // See https://crbug.com/webp/573.
+#ifdef WEBP_MSAN
+    dst[3] = 0;
+#endif
   }
   if (width & 1) {
     dst[0] = SUM2(r_ptr + j);
     dst[1] = SUM2(g_ptr + j);
     dst[2] = SUM2(b_ptr + j);
+#ifdef WEBP_MSAN
+    dst[3] = 0;
+#endif
   }
 }
 
@@ -839,6 +469,8 @@ static WEBP_INLINE void ConvertRowsToUV(const uint16_t* rgb,
   }
 }
 
+extern void SharpYuvInit(VP8CPUInfo cpu_info_func);
+
 static int ImportYUVAFromRGBA(const uint8_t* r_ptr,
                               const uint8_t* g_ptr,
                               const uint8_t* b_ptr,
@@ -863,18 +495,18 @@ static int ImportYUVAFromRGBA(const uint8_t* r_ptr,
     use_iterative_conversion = 0;
   }
 
-  if (!WebPPictureAllocYUVA(picture, width, height)) {
+  if (!WebPPictureAllocYUVA(picture)) {
     return 0;
   }
   if (has_alpha) {
     assert(step == 4);
 #if defined(USE_GAMMA_COMPRESSION) && defined(USE_INVERSE_ALPHA_TABLE)
-    assert(kAlphaFix + kGammaFix <= 31);
+    assert(kAlphaFix + GAMMA_FIX <= 31);
 #endif
   }
 
   if (use_iterative_conversion) {
-    InitGammaTablesS();
+    SharpYuvInit(VP8GetCPUInfo);
     if (!PreprocessARGB(r_ptr, g_ptr, b_ptr, step, rgb_stride, picture)) {
       return 0;
     }
@@ -903,7 +535,9 @@ static int ImportYUVAFromRGBA(const uint8_t* r_ptr,
     WebPInitConvertARGBToYUV();
     InitGammaTables();
 
-    if (tmp_rgb == NULL) return 0;  // malloc error
+    if (tmp_rgb == NULL) {
+      return WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY);
+    }
 
     // Downsample Y/U/V planes, two rows at a time
     for (y = 0; y < (height >> 1); ++y) {
@@ -1044,7 +678,7 @@ int WebPPictureYUVAToARGB(WebPPicture* picture) {
     return WebPEncodingSetError(picture, VP8_ENC_ERROR_INVALID_CONFIGURATION);
   }
   // Allocate a new argb buffer (discarding the previous one).
-  if (!WebPPictureAllocARGB(picture, picture->width, picture->height)) return 0;
+  if (!WebPPictureAllocARGB(picture)) return 0;
   picture->use_argb = 1;
 
   // Convert
@@ -1106,6 +740,8 @@ static int Import(WebPPicture* const picture,
   const int width = picture->width;
   const int height = picture->height;
 
+  if (abs(rgb_stride) < (import_alpha ? 4 : 3) * width) return 0;
+
   if (!picture->use_argb) {
     const uint8_t* a_ptr = import_alpha ? rgb + 3 : NULL;
     return ImportYUVAFromRGBA(r_ptr, g_ptr, b_ptr, a_ptr, step, rgb_stride,
@@ -1163,24 +799,24 @@ static int Import(WebPPicture* const picture,
 #if !defined(WEBP_REDUCE_CSP)
 
 int WebPPictureImportBGR(WebPPicture* picture,
-                         const uint8_t* rgb, int rgb_stride) {
-  return (picture != NULL && rgb != NULL)
-             ? Import(picture, rgb, rgb_stride, 3, 1, 0)
+                         const uint8_t* bgr, int bgr_stride) {
+  return (picture != NULL && bgr != NULL)
+             ? Import(picture, bgr, bgr_stride, 3, 1, 0)
              : 0;
 }
 
 int WebPPictureImportBGRA(WebPPicture* picture,
-                          const uint8_t* rgba, int rgba_stride) {
-  return (picture != NULL && rgba != NULL)
-             ? Import(picture, rgba, rgba_stride, 4, 1, 1)
+                          const uint8_t* bgra, int bgra_stride) {
+  return (picture != NULL && bgra != NULL)
+             ? Import(picture, bgra, bgra_stride, 4, 1, 1)
              : 0;
 }
 
 
 int WebPPictureImportBGRX(WebPPicture* picture,
-                          const uint8_t* rgba, int rgba_stride) {
-  return (picture != NULL && rgba != NULL)
-             ? Import(picture, rgba, rgba_stride, 4, 1, 0)
+                          const uint8_t* bgrx, int bgrx_stride) {
+  return (picture != NULL && bgrx != NULL)
+             ? Import(picture, bgrx, bgrx_stride, 4, 1, 0)
              : 0;
 }
 
@@ -1201,9 +837,9 @@ int WebPPictureImportRGBA(WebPPicture* picture,
 }
 
 int WebPPictureImportRGBX(WebPPicture* picture,
-                          const uint8_t* rgba, int rgba_stride) {
-  return (picture != NULL && rgba != NULL)
-             ? Import(picture, rgba, rgba_stride, 4, 0, 0)
+                          const uint8_t* rgbx, int rgbx_stride) {
+  return (picture != NULL && rgbx != NULL)
+             ? Import(picture, rgbx, rgbx_stride, 4, 0, 0)
              : 0;
 }
 

3rdparty/libwebp/src/enc/picture_enc.c

@@ -12,10 +12,10 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include <assert.h>
+#include <limits.h>
 #include <stdlib.h>
 
 #include "src/enc/vp8i_enc.h"
-#include "src/dsp/dsp.h"
 #include "src/utils/utils.h"
 
 //------------------------------------------------------------------------------
@@ -45,6 +45,22 @@ int WebPPictureInitInternal(WebPPicture* picture, int version) {
 
 //------------------------------------------------------------------------------
 
+int WebPValidatePicture(const WebPPicture* const picture) {
+  if (picture == NULL) return 0;
+  if (picture->width <= 0 || picture->height <= 0) {
+    return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
+  }
+  if (picture->width <= 0 || picture->width / 4 > INT_MAX / 4 ||
+      picture->height <= 0 || picture->height / 4 > INT_MAX / 4) {
+    return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
+  }
+  if (picture->colorspace != WEBP_YUV420 &&
+      picture->colorspace != WEBP_YUV420A) {
+    return WebPEncodingSetError(picture, VP8_ENC_ERROR_INVALID_CONFIGURATION);
+  }
+  return 1;
+}
+
 static void WebPPictureResetBufferARGB(WebPPicture* const picture) {
   picture->memory_argb_ = NULL;
   picture->argb = NULL;
@@ -63,18 +79,17 @@ void WebPPictureResetBuffers(WebPPicture* const picture) {
   WebPPictureResetBufferYUVA(picture);
 }
 
-int WebPPictureAllocARGB(WebPPicture* const picture, int width, int height) {
+int WebPPictureAllocARGB(WebPPicture* const picture) {
   void* memory;
+  const int width = picture->width;
+  const int height = picture->height;
   const uint64_t argb_size = (uint64_t)width * height;
 
-  assert(picture != NULL);
+  if (!WebPValidatePicture(picture)) return 0;
 
   WebPSafeFree(picture->memory_argb_);
   WebPPictureResetBufferARGB(picture);
 
-  if (width <= 0 || height <= 0) {
-    return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
-  }
   // allocate a new buffer.
   memory = WebPSafeMalloc(argb_size + WEBP_ALIGN_CST, sizeof(*picture->argb));
   if (memory == NULL) {
@@ -86,10 +101,10 @@ int WebPPictureAllocARGB(WebPPicture* const picture, int width, int height) {
   return 1;
 }
 
-int WebPPictureAllocYUVA(WebPPicture* const picture, int width, int height) {
-  const WebPEncCSP uv_csp =
-      (WebPEncCSP)((int)picture->colorspace & WEBP_CSP_UV_MASK);
+int WebPPictureAllocYUVA(WebPPicture* const picture) {
   const int has_alpha = (int)picture->colorspace & WEBP_CSP_ALPHA_BIT;
+  const int width = picture->width;
+  const int height = picture->height;
   const int y_stride = width;
   const int uv_width = (int)(((int64_t)width + 1) >> 1);
   const int uv_height = (int)(((int64_t)height + 1) >> 1);
@@ -98,15 +113,11 @@ int WebPPictureAllocYUVA(WebPPicture* const picture, int width, int height) {
   uint64_t y_size, uv_size, a_size, total_size;
   uint8_t* mem;
 
-  assert(picture != NULL);
+  if (!WebPValidatePicture(picture)) return 0;
 
   WebPSafeFree(picture->memory_);
   WebPPictureResetBufferYUVA(picture);
 
-  if (uv_csp != WEBP_YUV420) {
-    return WebPEncodingSetError(picture, VP8_ENC_ERROR_INVALID_CONFIGURATION);
-  }
-
   // alpha
   a_width = has_alpha ? width : 0;
   a_stride = a_width;
@@ -152,15 +163,12 @@ int WebPPictureAllocYUVA(WebPPicture* const picture, int width, int height) {
 
 int WebPPictureAlloc(WebPPicture* picture) {
   if (picture != NULL) {
-    const int width = picture->width;
-    const int height = picture->height;
-
     WebPPictureFree(picture);   // erase previous buffer
 
     if (!picture->use_argb) {
-      return WebPPictureAllocYUVA(picture, width, height);
+      return WebPPictureAllocYUVA(picture);
     } else {
-      return WebPPictureAllocARGB(picture, width, height);
+      return WebPPictureAllocARGB(picture);
     }
   }
   return 1;

3rdparty/libwebp/src/enc/picture_rescale_enc.c

@@ -13,14 +13,15 @@
 
 #include "src/webp/encode.h"
 
-#if !defined(WEBP_REDUCE_SIZE)
-
 #include <assert.h>
 #include <stdlib.h>
 
 #include "src/enc/vp8i_enc.h"
+
+#if !defined(WEBP_REDUCE_SIZE)
 #include "src/utils/rescaler_utils.h"
 #include "src/utils/utils.h"
+#endif  // !defined(WEBP_REDUCE_SIZE)
 
 #define HALVE(x) (((x) + 1) >> 1)
 
@@ -56,6 +57,7 @@ static int AdjustAndCheckRectangle(const WebPPicture* const pic,
   return 1;
 }
 
+#if !defined(WEBP_REDUCE_SIZE)
 int WebPPictureCopy(const WebPPicture* src, WebPPicture* dst) {
   if (src == NULL || dst == NULL) return 0;
   if (src == dst) return 1;
@@ -81,6 +83,7 @@ int WebPPictureCopy(const WebPPicture* src, WebPPicture* dst) {
   }
   return 1;
 }
+#endif  // !defined(WEBP_REDUCE_SIZE)
 
 int WebPPictureIsView(const WebPPicture* picture) {
   if (picture == NULL) return 0;
@@ -120,6 +123,7 @@ int WebPPictureView(const WebPPicture* src,
   return 1;
 }
 
+#if !defined(WEBP_REDUCE_SIZE)
 //------------------------------------------------------------------------------
 // Picture cropping
 
@@ -133,7 +137,9 @@ int WebPPictureCrop(WebPPicture* pic,
   PictureGrabSpecs(pic, &tmp);
   tmp.width = width;
   tmp.height = height;
-  if (!WebPPictureAlloc(&tmp)) return 0;
+  if (!WebPPictureAlloc(&tmp)) {
+    return WebPEncodingSetError(pic, tmp.error_code);
+  }
 
   if (!pic->use_argb) {
     const int y_offset = top * pic->y_stride + left;
@@ -164,22 +170,25 @@ int WebPPictureCrop(WebPPicture* pic,
 //------------------------------------------------------------------------------
 // Simple picture rescaler
 
-static void RescalePlane(const uint8_t* src,
-                         int src_width, int src_height, int src_stride,
-                         uint8_t* dst,
-                         int dst_width, int dst_height, int dst_stride,
-                         rescaler_t* const work,
-                         int num_channels) {
+static int RescalePlane(const uint8_t* src,
+                        int src_width, int src_height, int src_stride,
+                        uint8_t* dst,
+                        int dst_width, int dst_height, int dst_stride,
+                        rescaler_t* const work,
+                        int num_channels) {
   WebPRescaler rescaler;
   int y = 0;
-  WebPRescalerInit(&rescaler, src_width, src_height,
-                   dst, dst_width, dst_height, dst_stride,
-                   num_channels, work);
+  if (!WebPRescalerInit(&rescaler, src_width, src_height,
+                        dst, dst_width, dst_height, dst_stride,
+                        num_channels, work)) {
+    return 0;
+  }
   while (y < src_height) {
     y += WebPRescalerImport(&rescaler, src_height - y,
                             src + y * src_stride, src_stride);
     WebPRescalerExport(&rescaler);
   }
+  return 1;
 }
 
 static void AlphaMultiplyARGB(WebPPicture* const pic, int inverse) {
@@ -195,73 +204,76 @@ static void AlphaMultiplyY(WebPPicture* const pic, int inverse) {
   }
 }
 
-int WebPPictureRescale(WebPPicture* pic, int width, int height) {
+int WebPPictureRescale(WebPPicture* picture, int width, int height) {
   WebPPicture tmp;
   int prev_width, prev_height;
   rescaler_t* work;
 
-  if (pic == NULL) return 0;
-  prev_width = pic->width;
-  prev_height = pic->height;
+  if (picture == NULL) return 0;
+  prev_width = picture->width;
+  prev_height = picture->height;
   if (!WebPRescalerGetScaledDimensions(
           prev_width, prev_height, &width, &height)) {
-    return 0;
+    return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
   }
 
-  PictureGrabSpecs(pic, &tmp);
+  PictureGrabSpecs(picture, &tmp);
   tmp.width = width;
   tmp.height = height;
-  if (!WebPPictureAlloc(&tmp)) return 0;
+  if (!WebPPictureAlloc(&tmp)) {
+    return WebPEncodingSetError(picture, tmp.error_code);
+  }
 
-  if (!pic->use_argb) {
+  if (!picture->use_argb) {
     work = (rescaler_t*)WebPSafeMalloc(2ULL * width, sizeof(*work));
     if (work == NULL) {
       WebPPictureFree(&tmp);
-      return 0;
+      return WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY);
     }
     // If present, we need to rescale alpha first (for AlphaMultiplyY).
-    if (pic->a != NULL) {
+    if (picture->a != NULL) {
       WebPInitAlphaProcessing();
-      RescalePlane(pic->a, prev_width, prev_height, pic->a_stride,
-                   tmp.a, width, height, tmp.a_stride, work, 1);
+      if (!RescalePlane(picture->a, prev_width, prev_height, picture->a_stride,
+                        tmp.a, width, height, tmp.a_stride, work, 1)) {
+        return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
+      }
     }
 
     // We take transparency into account on the luma plane only. That's not
     // totally exact blending, but still is a good approximation.
-    AlphaMultiplyY(pic, 0);
-    RescalePlane(pic->y, prev_width, prev_height, pic->y_stride,
-                 tmp.y, width, height, tmp.y_stride, work, 1);
+    AlphaMultiplyY(picture, 0);
+    if (!RescalePlane(picture->y, prev_width, prev_height, picture->y_stride,
+                      tmp.y, width, height, tmp.y_stride, work, 1) ||
+        !RescalePlane(picture->u, HALVE(prev_width), HALVE(prev_height),
+                      picture->uv_stride, tmp.u, HALVE(width), HALVE(height),
+                      tmp.uv_stride, work, 1) ||
+        !RescalePlane(picture->v, HALVE(prev_width), HALVE(prev_height),
+                      picture->uv_stride, tmp.v, HALVE(width), HALVE(height),
+                      tmp.uv_stride, work, 1)) {
+      return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
+    }
     AlphaMultiplyY(&tmp, 1);
-
-    RescalePlane(pic->u,
-                 HALVE(prev_width), HALVE(prev_height), pic->uv_stride,
-                 tmp.u,
-                 HALVE(width), HALVE(height), tmp.uv_stride, work, 1);
-    RescalePlane(pic->v,
-                 HALVE(prev_width), HALVE(prev_height), pic->uv_stride,
-                 tmp.v,
-                 HALVE(width), HALVE(height), tmp.uv_stride, work, 1);
   } else {
     work = (rescaler_t*)WebPSafeMalloc(2ULL * width * 4, sizeof(*work));
     if (work == NULL) {
       WebPPictureFree(&tmp);
-      return 0;
+      return WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY);
     }
     // In order to correctly interpolate colors, we need to apply the alpha
     // weighting first (black-matting), scale the RGB values, and remove
     // the premultiplication afterward (while preserving the alpha channel).
     WebPInitAlphaProcessing();
-    AlphaMultiplyARGB(pic, 0);
-    RescalePlane((const uint8_t*)pic->argb, prev_width, prev_height,
-                 pic->argb_stride * 4,
-                 (uint8_t*)tmp.argb, width, height,
-                 tmp.argb_stride * 4,
-                 work, 4);
+    AlphaMultiplyARGB(picture, 0);
+    if (!RescalePlane((const uint8_t*)picture->argb, prev_width, prev_height,
+                      picture->argb_stride * 4, (uint8_t*)tmp.argb, width,
+                      height, tmp.argb_stride * 4, work, 4)) {
+      return WebPEncodingSetError(picture, VP8_ENC_ERROR_BAD_DIMENSION);
+    }
     AlphaMultiplyARGB(&tmp, 1);
   }
-  WebPPictureFree(pic);
+  WebPPictureFree(picture);
   WebPSafeFree(work);
-  *pic = tmp;
+  *picture = tmp;
   return 1;
 }
 
@@ -273,23 +285,6 @@ int WebPPictureCopy(const WebPPicture* src, WebPPicture* dst) {
   return 0;
 }
 
-int WebPPictureIsView(const WebPPicture* picture) {
-  (void)picture;
-  return 0;
-}
-
-int WebPPictureView(const WebPPicture* src,
-                    int left, int top, int width, int height,
-                    WebPPicture* dst) {
-  (void)src;
-  (void)left;
-  (void)top;
-  (void)width;
-  (void)height;
-  (void)dst;
-  return 0;
-}
-
 int WebPPictureCrop(WebPPicture* pic,
                     int left, int top, int width, int height) {
   (void)pic;

3rdparty/libwebp/src/enc/picture_tools_enc.c

@@ -190,27 +190,28 @@ static WEBP_INLINE uint32_t MakeARGB32(int r, int g, int b) {
   return (0xff000000u | (r << 16) | (g << 8) | b);
 }
 
-void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb) {
+void WebPBlendAlpha(WebPPicture* picture, uint32_t background_rgb) {
   const int red = (background_rgb >> 16) & 0xff;
   const int green = (background_rgb >> 8) & 0xff;
   const int blue = (background_rgb >> 0) & 0xff;
   int x, y;
-  if (pic == NULL) return;
-  if (!pic->use_argb) {
-    const int uv_width = (pic->width >> 1);  // omit last pixel during u/v loop
+  if (picture == NULL) return;
+  if (!picture->use_argb) {
+    // omit last pixel during u/v loop
+    const int uv_width = (picture->width >> 1);
     const int Y0 = VP8RGBToY(red, green, blue, YUV_HALF);
     // VP8RGBToU/V expects the u/v values summed over four pixels
     const int U0 = VP8RGBToU(4 * red, 4 * green, 4 * blue, 4 * YUV_HALF);
     const int V0 = VP8RGBToV(4 * red, 4 * green, 4 * blue, 4 * YUV_HALF);
-    const int has_alpha = pic->colorspace & WEBP_CSP_ALPHA_BIT;
-    uint8_t* y_ptr = pic->y;
-    uint8_t* u_ptr = pic->u;
-    uint8_t* v_ptr = pic->v;
-    uint8_t* a_ptr = pic->a;
+    const int has_alpha = picture->colorspace & WEBP_CSP_ALPHA_BIT;
+    uint8_t* y_ptr = picture->y;
+    uint8_t* u_ptr = picture->u;
+    uint8_t* v_ptr = picture->v;
+    uint8_t* a_ptr = picture->a;
     if (!has_alpha || a_ptr == NULL) return;    // nothing to do
-    for (y = 0; y < pic->height; ++y) {
+    for (y = 0; y < picture->height; ++y) {
       // Luma blending
-      for (x = 0; x < pic->width; ++x) {
+      for (x = 0; x < picture->width; ++x) {
         const uint8_t alpha = a_ptr[x];
         if (alpha < 0xff) {
           y_ptr[x] = BLEND(Y0, y_ptr[x], alpha);
@@ -219,7 +220,7 @@ void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb) {
       // Chroma blending every even line
       if ((y & 1) == 0) {
         uint8_t* const a_ptr2 =
-            (y + 1 == pic->height) ? a_ptr : a_ptr + pic->a_stride;
+            (y + 1 == picture->height) ? a_ptr : a_ptr + picture->a_stride;
         for (x = 0; x < uv_width; ++x) {
           // Average four alpha values into a single blending weight.
           // TODO(skal): might lead to visible contouring. Can we do better?
@@ -229,24 +230,24 @@ void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb) {
           u_ptr[x] = BLEND_10BIT(U0, u_ptr[x], alpha);
           v_ptr[x] = BLEND_10BIT(V0, v_ptr[x], alpha);
         }
-        if (pic->width & 1) {   // rightmost pixel
+        if (picture->width & 1) {  // rightmost pixel
           const uint32_t alpha = 2 * (a_ptr[2 * x + 0] + a_ptr2[2 * x + 0]);
           u_ptr[x] = BLEND_10BIT(U0, u_ptr[x], alpha);
           v_ptr[x] = BLEND_10BIT(V0, v_ptr[x], alpha);
         }
       } else {
-        u_ptr += pic->uv_stride;
-        v_ptr += pic->uv_stride;
+        u_ptr += picture->uv_stride;
+        v_ptr += picture->uv_stride;
       }
-      memset(a_ptr, 0xff, pic->width);  // reset alpha value to opaque
-      a_ptr += pic->a_stride;
-      y_ptr += pic->y_stride;
+      memset(a_ptr, 0xff, picture->width);  // reset alpha value to opaque
+      a_ptr += picture->a_stride;
+      y_ptr += picture->y_stride;
     }
   } else {
-    uint32_t* argb = pic->argb;
+    uint32_t* argb = picture->argb;
     const uint32_t background = MakeARGB32(red, green, blue);
-    for (y = 0; y < pic->height; ++y) {
-      for (x = 0; x < pic->width; ++x) {
+    for (y = 0; y < picture->height; ++y) {
+      for (x = 0; x < picture->width; ++x) {
         const int alpha = (argb[x] >> 24) & 0xff;
         if (alpha != 0xff) {
           if (alpha > 0) {
@@ -262,7 +263,7 @@ void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb) {
           }
         }
       }
-      argb += pic->argb_stride;
+      argb += picture->argb_stride;
     }
   }
 }

3rdparty/libwebp/src/enc/predictor_enc.c

@@ -16,6 +16,7 @@
 
 #include "src/dsp/lossless.h"
 #include "src/dsp/lossless_common.h"
+#include "src/enc/vp8i_enc.h"
 #include "src/enc/vp8li_enc.h"
 
 #define MAX_DIFF_COST (1e30f)
@@ -31,10 +32,10 @@ static WEBP_INLINE int GetMin(int a, int b) { return (a > b) ? b : a; }
 // Methods to calculate Entropy (Shannon).
 
 static float PredictionCostSpatial(const int counts[256], int weight_0,
-                                   double exp_val) {
+                                   float exp_val) {
   const int significant_symbols = 256 >> 4;
-  const double exp_decay_factor = 0.6;
-  double bits = weight_0 * counts[0];
+  const float exp_decay_factor = 0.6f;
+  float bits = (float)weight_0 * counts[0];
   int i;
   for (i = 1; i < significant_symbols; ++i) {
     bits += exp_val * (counts[i] + counts[256 - i]);
@@ -46,9 +47,9 @@ static float PredictionCostSpatial(const int counts[256], int weight_0,
 static float PredictionCostSpatialHistogram(const int accumulated[4][256],
                                             const int tile[4][256]) {
   int i;
-  double retval = 0;
+  float retval = 0.f;
   for (i = 0; i < 4; ++i) {
-    const double kExpValue = 0.94;
+    const float kExpValue = 0.94f;
     retval += PredictionCostSpatial(tile[i], 1, kExpValue);
     retval += VP8LCombinedShannonEntropy(tile[i], accumulated[i]);
   }
@@ -249,7 +250,7 @@ static WEBP_INLINE void GetResidual(
       } else if (x == 0) {
         predict = upper_row[x];  // Top.
       } else {
-        predict = pred_func(current_row[x - 1], upper_row + x);
+        predict = pred_func(&current_row[x - 1], upper_row + x);
       }
 #if (WEBP_NEAR_LOSSLESS == 1)
       if (max_quantization == 1 || mode == 0 || y == 0 || y == height - 1 ||
@@ -472,12 +473,15 @@ static void CopyImageWithPrediction(int width, int height,
 // with respect to predictions. If near_lossless_quality < 100, applies
 // near lossless processing, shaving off more bits of residuals for lower
 // qualities.
-void VP8LResidualImage(int width, int height, int bits, int low_effort,
-                       uint32_t* const argb, uint32_t* const argb_scratch,
-                       uint32_t* const image, int near_lossless_quality,
-                       int exact, int used_subtract_green) {
+int VP8LResidualImage(int width, int height, int bits, int low_effort,
+                      uint32_t* const argb, uint32_t* const argb_scratch,
+                      uint32_t* const image, int near_lossless_quality,
+                      int exact, int used_subtract_green,
+                      const WebPPicture* const pic, int percent_range,
+                      int* const percent) {
   const int tiles_per_row = VP8LSubSampleSize(width, bits);
   const int tiles_per_col = VP8LSubSampleSize(height, bits);
+  int percent_start = *percent;
   int tile_y;
   int histo[4][256];
   const int max_quantization = 1 << VP8LNearLosslessBits(near_lossless_quality);
@@ -491,17 +495,24 @@ void VP8LResidualImage(int width, int height, int bits, int low_effort,
     for (tile_y = 0; tile_y < tiles_per_col; ++tile_y) {
       int tile_x;
       for (tile_x = 0; tile_x < tiles_per_row; ++tile_x) {
-        const int pred = GetBestPredictorForTile(width, height, tile_x, tile_y,
-            bits, histo, argb_scratch, argb, max_quantization, exact,
-            used_subtract_green, image);
+        const int pred = GetBestPredictorForTile(
+            width, height, tile_x, tile_y, bits, histo, argb_scratch, argb,
+            max_quantization, exact, used_subtract_green, image);
         image[tile_y * tiles_per_row + tile_x] = ARGB_BLACK | (pred << 8);
       }
+
+      if (!WebPReportProgress(
+              pic, percent_start + percent_range * tile_y / tiles_per_col,
+              percent)) {
+        return 0;
+      }
     }
   }
 
   CopyImageWithPrediction(width, height, bits, image, argb_scratch, argb,
                           low_effort, max_quantization, exact,
                           used_subtract_green);
+  return WebPReportProgress(pic, percent_start + percent_range, percent);
 }
 
 //------------------------------------------------------------------------------
@@ -532,7 +543,7 @@ static float PredictionCostCrossColor(const int accumulated[256],
                                       const int counts[256]) {
   // Favor low entropy, locally and globally.
   // Favor small absolute values for PredictionCostSpatial
-  static const double kExpValue = 2.4;
+  static const float kExpValue = 2.4f;
   return VP8LCombinedShannonEntropy(counts, accumulated) +
          PredictionCostSpatial(counts, 3, kExpValue);
 }
@@ -714,11 +725,14 @@ static void CopyTileWithColorTransform(int xsize, int ysize,
   }
 }
 
-void VP8LColorSpaceTransform(int width, int height, int bits, int quality,
-                             uint32_t* const argb, uint32_t* image) {
+int VP8LColorSpaceTransform(int width, int height, int bits, int quality,
+                            uint32_t* const argb, uint32_t* image,
+                            const WebPPicture* const pic, int percent_range,
+                            int* const percent) {
   const int max_tile_size = 1 << bits;
   const int tile_xsize = VP8LSubSampleSize(width, bits);
   const int tile_ysize = VP8LSubSampleSize(height, bits);
+  int percent_start = *percent;
   int accumulated_red_histo[256] = { 0 };
   int accumulated_blue_histo[256] = { 0 };
   int tile_x, tile_y;
@@ -768,5 +782,11 @@ void VP8LColorSpaceTransform(int width, int height, int bits, int quality,
         }
       }
     }
+    if (!WebPReportProgress(
+            pic, percent_start + percent_range * tile_y / tile_ysize,
+            percent)) {
+      return 0;
+    }
   }
+  return 1;
 }

3rdparty/libwebp/src/enc/quant_enc.c

@@ -533,7 +533,8 @@ static void InitScore(VP8ModeScore* const rd) {
   rd->score = MAX_COST;
 }
 
-static void CopyScore(VP8ModeScore* const dst, const VP8ModeScore* const src) {
+static void CopyScore(VP8ModeScore* WEBP_RESTRICT const dst,
+                      const VP8ModeScore* WEBP_RESTRICT const src) {
   dst->D  = src->D;
   dst->SD = src->SD;
   dst->R  = src->R;
@@ -542,7 +543,8 @@ static void CopyScore(VP8ModeScore* const dst, const VP8ModeScore* const src) {
   dst->score = src->score;
 }
 
-static void AddScore(VP8ModeScore* const dst, const VP8ModeScore* const src) {
+static void AddScore(VP8ModeScore* WEBP_RESTRICT const dst,
+                     const VP8ModeScore* WEBP_RESTRICT const src) {
   dst->D  += src->D;
   dst->SD += src->SD;
   dst->R  += src->R;
@@ -585,15 +587,18 @@ static WEBP_INLINE score_t RDScoreTrellis(int lambda, score_t rate,
   return rate * lambda + RD_DISTO_MULT * distortion;
 }
 
-static int TrellisQuantizeBlock(const VP8Encoder* const enc,
+// Coefficient type.
+enum { TYPE_I16_AC = 0, TYPE_I16_DC = 1, TYPE_CHROMA_A = 2, TYPE_I4_AC = 3 };
+
+static int TrellisQuantizeBlock(const VP8Encoder* WEBP_RESTRICT const enc,
                                 int16_t in[16], int16_t out[16],
                                 int ctx0, int coeff_type,
-                                const VP8Matrix* const mtx,
+                                const VP8Matrix* WEBP_RESTRICT const mtx,
                                 int lambda) {
   const ProbaArray* const probas = enc->proba_.coeffs_[coeff_type];
   CostArrayPtr const costs =
       (CostArrayPtr)enc->proba_.remapped_costs_[coeff_type];
-  const int first = (coeff_type == 0) ? 1 : 0;
+  const int first = (coeff_type == TYPE_I16_AC) ? 1 : 0;
   Node nodes[16][NUM_NODES];
   ScoreState score_states[2][NUM_NODES];
   ScoreState* ss_cur = &SCORE_STATE(0, MIN_DELTA);
@@ -657,16 +662,17 @@ static int TrellisQuantizeBlock(const VP8Encoder* const enc,
     // test all alternate level values around level0.
     for (m = -MIN_DELTA; m <= MAX_DELTA; ++m) {
       Node* const cur = &NODE(n, m);
-      int level = level0 + m;
+      const int level = level0 + m;
       const int ctx = (level > 2) ? 2 : level;
       const int band = VP8EncBands[n + 1];
       score_t base_score;
-      score_t best_cur_score = MAX_COST;
-      int best_prev = 0;   // default, in case
+      score_t best_cur_score;
+      int best_prev;
+      score_t cost, score;
 
-      ss_cur[m].score = MAX_COST;
       ss_cur[m].costs = costs[n + 1][ctx];
       if (level < 0 || level > thresh_level) {
+        ss_cur[m].score = MAX_COST;
         // Node is dead.
         continue;
       }
@@ -682,18 +688,24 @@ static int TrellisQuantizeBlock(const VP8Encoder* const enc,
       }
 
       // Inspect all possible non-dead predecessors. Retain only the best one.
-      for (p = -MIN_DELTA; p <= MAX_DELTA; ++p) {
+      // The base_score is added to all scores so it is only added for the final
+      // value after the loop.
+      cost = VP8LevelCost(ss_prev[-MIN_DELTA].costs, level);
+      best_cur_score =
+          ss_prev[-MIN_DELTA].score + RDScoreTrellis(lambda, cost, 0);
+      best_prev = -MIN_DELTA;
+      for (p = -MIN_DELTA + 1; p <= MAX_DELTA; ++p) {
         // Dead nodes (with ss_prev[p].score >= MAX_COST) are automatically
         // eliminated since their score can't be better than the current best.
-        const score_t cost = VP8LevelCost(ss_prev[p].costs, level);
+        cost = VP8LevelCost(ss_prev[p].costs, level);
         // Examine node assuming it's a non-terminal one.
-        const score_t score =
-            base_score + ss_prev[p].score + RDScoreTrellis(lambda, cost, 0);
+        score = ss_prev[p].score + RDScoreTrellis(lambda, cost, 0);
         if (score < best_cur_score) {
           best_cur_score = score;
           best_prev = p;
         }
       }
+      best_cur_score += base_score;
       // Store best finding in current node.
       cur->sign = sign;
       cur->level = level;
@@ -701,11 +713,11 @@ static int TrellisQuantizeBlock(const VP8Encoder* const enc,
       ss_cur[m].score = best_cur_score;
 
       // Now, record best terminal node (and thus best entry in the graph).
-      if (level != 0) {
+      if (level != 0 && best_cur_score < best_score) {
         const score_t last_pos_cost =
             (n < 15) ? VP8BitCost(0, probas[band][ctx][0]) : 0;
         const score_t last_pos_score = RDScoreTrellis(lambda, last_pos_cost, 0);
-        const score_t score = best_cur_score + last_pos_score;
+        score = best_cur_score + last_pos_score;
         if (score < best_score) {
           best_score = score;
           best_path[0] = n;                     // best eob position
@@ -717,10 +729,16 @@ static int TrellisQuantizeBlock(const VP8Encoder* const enc,
   }
 
   // Fresh start
-  memset(in + first, 0, (16 - first) * sizeof(*in));
-  memset(out + first, 0, (16 - first) * sizeof(*out));
+  // Beware! We must preserve in[0]/out[0] value for TYPE_I16_AC case.
+  if (coeff_type == TYPE_I16_AC) {
+    memset(in + 1, 0, 15 * sizeof(*in));
+    memset(out + 1, 0, 15 * sizeof(*out));
+  } else {
+    memset(in, 0, 16 * sizeof(*in));
+    memset(out, 0, 16 * sizeof(*out));
+  }
   if (best_path[0] == -1) {
-    return 0;   // skip!
+    return 0;  // skip!
   }
 
   {
@@ -751,9 +769,9 @@ static int TrellisQuantizeBlock(const VP8Encoder* const enc,
 // all at once. Output is the reconstructed block in *yuv_out, and the
 // quantized levels in *levels.
 
-static int ReconstructIntra16(VP8EncIterator* const it,
-                              VP8ModeScore* const rd,
-                              uint8_t* const yuv_out,
+static int ReconstructIntra16(VP8EncIterator* WEBP_RESTRICT const it,
+                              VP8ModeScore* WEBP_RESTRICT const rd,
+                              uint8_t* WEBP_RESTRICT const yuv_out,
                               int mode) {
   const VP8Encoder* const enc = it->enc_;
   const uint8_t* const ref = it->yuv_p_ + VP8I16ModeOffsets[mode];
@@ -775,9 +793,9 @@ static int ReconstructIntra16(VP8EncIterator* const it,
     for (y = 0, n = 0; y < 4; ++y) {
       for (x = 0; x < 4; ++x, ++n) {
         const int ctx = it->top_nz_[x] + it->left_nz_[y];
-        const int non_zero =
-            TrellisQuantizeBlock(enc, tmp[n], rd->y_ac_levels[n], ctx, 0,
-                                 &dqm->y1_, dqm->lambda_trellis_i16_);
+        const int non_zero = TrellisQuantizeBlock(
+            enc, tmp[n], rd->y_ac_levels[n], ctx, TYPE_I16_AC, &dqm->y1_,
+            dqm->lambda_trellis_i16_);
         it->top_nz_[x] = it->left_nz_[y] = non_zero;
         rd->y_ac_levels[n][0] = 0;
         nz |= non_zero << n;
@@ -803,10 +821,10 @@ static int ReconstructIntra16(VP8EncIterator* const it,
   return nz;
 }
 
-static int ReconstructIntra4(VP8EncIterator* const it,
+static int ReconstructIntra4(VP8EncIterator* WEBP_RESTRICT const it,
                              int16_t levels[16],
-                             const uint8_t* const src,
-                             uint8_t* const yuv_out,
+                             const uint8_t* WEBP_RESTRICT const src,
+                             uint8_t* WEBP_RESTRICT const yuv_out,
                              int mode) {
   const VP8Encoder* const enc = it->enc_;
   const uint8_t* const ref = it->yuv_p_ + VP8I4ModeOffsets[mode];
@@ -818,7 +836,7 @@ static int ReconstructIntra4(VP8EncIterator* const it,
   if (DO_TRELLIS_I4 && it->do_trellis_) {
     const int x = it->i4_ & 3, y = it->i4_ >> 2;
     const int ctx = it->top_nz_[x] + it->left_nz_[y];
-    nz = TrellisQuantizeBlock(enc, tmp, levels, ctx, 3, &dqm->y1_,
+    nz = TrellisQuantizeBlock(enc, tmp, levels, ctx, TYPE_I4_AC, &dqm->y1_,
                               dqm->lambda_trellis_i4_);
   } else {
     nz = VP8EncQuantizeBlock(tmp, levels, &dqm->y1_);
@@ -839,7 +857,8 @@ static int ReconstructIntra4(VP8EncIterator* const it,
 
 // Quantize as usual, but also compute and return the quantization error.
 // Error is already divided by DSHIFT.
-static int QuantizeSingle(int16_t* const v, const VP8Matrix* const mtx) {
+static int QuantizeSingle(int16_t* WEBP_RESTRICT const v,
+                          const VP8Matrix* WEBP_RESTRICT const mtx) {
   int V = *v;
   const int sign = (V < 0);
   if (sign) V = -V;
@@ -853,9 +872,10 @@ static int QuantizeSingle(int16_t* const v, const VP8Matrix* const mtx) {
   return (sign ? -V : V) >> DSCALE;
 }
 
-static void CorrectDCValues(const VP8EncIterator* const it,
-                            const VP8Matrix* const mtx,
-                            int16_t tmp[][16], VP8ModeScore* const rd) {
+static void CorrectDCValues(const VP8EncIterator* WEBP_RESTRICT const it,
+                            const VP8Matrix* WEBP_RESTRICT const mtx,
+                            int16_t tmp[][16],
+                            VP8ModeScore* WEBP_RESTRICT const rd) {
   //         | top[0] | top[1]
   // --------+--------+---------
   // left[0] | tmp[0]   tmp[1]  <->   err0 err1
@@ -886,8 +906,8 @@ static void CorrectDCValues(const VP8EncIterator* const it,
   }
 }
 
-static void StoreDiffusionErrors(VP8EncIterator* const it,
-                                 const VP8ModeScore* const rd) {
+static void StoreDiffusionErrors(VP8EncIterator* WEBP_RESTRICT const it,
+                                 const VP8ModeScore* WEBP_RESTRICT const rd) {
   int ch;
   for (ch = 0; ch <= 1; ++ch) {
     int8_t* const top = it->top_derr_[it->x_][ch];
@@ -906,8 +926,9 @@ static void StoreDiffusionErrors(VP8EncIterator* const it,
 
 //------------------------------------------------------------------------------
 
-static int ReconstructUV(VP8EncIterator* const it, VP8ModeScore* const rd,
-                         uint8_t* const yuv_out, int mode) {
+static int ReconstructUV(VP8EncIterator* WEBP_RESTRICT const it,
+                         VP8ModeScore* WEBP_RESTRICT const rd,
+                         uint8_t* WEBP_RESTRICT const yuv_out, int mode) {
   const VP8Encoder* const enc = it->enc_;
   const uint8_t* const ref = it->yuv_p_ + VP8UVModeOffsets[mode];
   const uint8_t* const src = it->yuv_in_ + U_OFF_ENC;
@@ -927,9 +948,9 @@ static int ReconstructUV(VP8EncIterator* const it, VP8ModeScore* const rd,
       for (y = 0; y < 2; ++y) {
         for (x = 0; x < 2; ++x, ++n) {
           const int ctx = it->top_nz_[4 + ch + x] + it->left_nz_[4 + ch + y];
-          const int non_zero =
-              TrellisQuantizeBlock(enc, tmp[n], rd->uv_levels[n], ctx, 2,
-                                   &dqm->uv_, dqm->lambda_trellis_uv_);
+          const int non_zero = TrellisQuantizeBlock(
+              enc, tmp[n], rd->uv_levels[n], ctx, TYPE_CHROMA_A, &dqm->uv_,
+              dqm->lambda_trellis_uv_);
           it->top_nz_[4 + ch + x] = it->left_nz_[4 + ch + y] = non_zero;
           nz |= non_zero << n;
         }
@@ -978,7 +999,8 @@ static void SwapOut(VP8EncIterator* const it) {
   SwapPtr(&it->yuv_out_, &it->yuv_out2_);
 }
 
-static void PickBestIntra16(VP8EncIterator* const it, VP8ModeScore* rd) {
+static void PickBestIntra16(VP8EncIterator* WEBP_RESTRICT const it,
+                            VP8ModeScore* WEBP_RESTRICT rd) {
   const int kNumBlocks = 16;
   VP8SegmentInfo* const dqm = &it->enc_->dqm_[it->mb_->segment_];
   const int lambda = dqm->lambda_i16_;
@@ -1038,7 +1060,7 @@ static void PickBestIntra16(VP8EncIterator* const it, VP8ModeScore* rd) {
 //------------------------------------------------------------------------------
 
 // return the cost array corresponding to the surrounding prediction modes.
-static const uint16_t* GetCostModeI4(VP8EncIterator* const it,
+static const uint16_t* GetCostModeI4(VP8EncIterator* WEBP_RESTRICT const it,
                                      const uint8_t modes[16]) {
   const int preds_w = it->enc_->preds_w_;
   const int x = (it->i4_ & 3), y = it->i4_ >> 2;
@@ -1047,7 +1069,8 @@ static const uint16_t* GetCostModeI4(VP8EncIterator* const it,
   return VP8FixedCostsI4[top][left];
 }
 
-static int PickBestIntra4(VP8EncIterator* const it, VP8ModeScore* const rd) {
+static int PickBestIntra4(VP8EncIterator* WEBP_RESTRICT const it,
+                          VP8ModeScore* WEBP_RESTRICT const rd) {
   const VP8Encoder* const enc = it->enc_;
   const VP8SegmentInfo* const dqm = &enc->dqm_[it->mb_->segment_];
   const int lambda = dqm->lambda_i4_;
@@ -1143,7 +1166,8 @@ static int PickBestIntra4(VP8EncIterator* const it, VP8ModeScore* const rd) {
 
 //------------------------------------------------------------------------------
 
-static void PickBestUV(VP8EncIterator* const it, VP8ModeScore* const rd) {
+static void PickBestUV(VP8EncIterator* WEBP_RESTRICT const it,
+                       VP8ModeScore* WEBP_RESTRICT const rd) {
   const int kNumBlocks = 8;
   const VP8SegmentInfo* const dqm = &it->enc_->dqm_[it->mb_->segment_];
   const int lambda = dqm->lambda_uv_;
@@ -1195,7 +1219,8 @@ static void PickBestUV(VP8EncIterator* const it, VP8ModeScore* const rd) {
 //------------------------------------------------------------------------------
 // Final reconstruction and quantization.
 
-static void SimpleQuantize(VP8EncIterator* const it, VP8ModeScore* const rd) {
+static void SimpleQuantize(VP8EncIterator* WEBP_RESTRICT const it,
+                           VP8ModeScore* WEBP_RESTRICT const rd) {
   const VP8Encoder* const enc = it->enc_;
   const int is_i16 = (it->mb_->type_ == 1);
   int nz = 0;
@@ -1220,9 +1245,9 @@ static void SimpleQuantize(VP8EncIterator* const it, VP8ModeScore* const rd) {
 }
 
 // Refine intra16/intra4 sub-modes based on distortion only (not rate).
-static void RefineUsingDistortion(VP8EncIterator* const it,
+static void RefineUsingDistortion(VP8EncIterator* WEBP_RESTRICT const it,
                                   int try_both_modes, int refine_uv_mode,
-                                  VP8ModeScore* const rd) {
+                                  VP8ModeScore* WEBP_RESTRICT const rd) {
   score_t best_score = MAX_COST;
   int nz = 0;
   int mode;
@@ -1336,7 +1361,8 @@ static void RefineUsingDistortion(VP8EncIterator* const it,
 //------------------------------------------------------------------------------
 // Entry point
 
-int VP8Decimate(VP8EncIterator* const it, VP8ModeScore* const rd,
+int VP8Decimate(VP8EncIterator* WEBP_RESTRICT const it,
+                VP8ModeScore* WEBP_RESTRICT const rd,
                 VP8RDLevel rd_opt) {
   int is_skipped;
   const int method = it->enc_->method_;

3rdparty/libwebp/src/enc/syntax_enc.c

@@ -258,7 +258,10 @@ static int EmitPartitionsSize(const VP8Encoder* const enc,
     buf[3 * p + 1] = (part_size >>  8) & 0xff;
     buf[3 * p + 2] = (part_size >> 16) & 0xff;
   }
-  return p ? pic->writer(buf, 3 * p, pic) : 1;
+  if (p && !pic->writer(buf, 3 * p, pic)) {
+    return WebPEncodingSetError(pic, VP8_ENC_ERROR_BAD_WRITE);
+  }
+  return 1;
 }
 
 //------------------------------------------------------------------------------
@@ -349,7 +352,7 @@ int VP8EncWrite(VP8Encoder* const enc) {
                                        (enc->alpha_data_size_ & 1);
     riff_size += CHUNK_HEADER_SIZE + padded_alpha_size;
   }
-  // Sanity check.
+  // RIFF size should fit in 32-bits.
   if (riff_size > 0xfffffffeU) {
     return WebPEncodingSetError(pic, VP8_ENC_ERROR_FILE_TOO_BIG);
   }
@@ -381,6 +384,7 @@ int VP8EncWrite(VP8Encoder* const enc) {
 
   enc->coded_size_ = (int)(CHUNK_HEADER_SIZE + riff_size);
   ok = ok && WebPReportProgress(pic, final_percent, &enc->percent_);
+  if (!ok) WebPEncodingSetError(pic, VP8_ENC_ERROR_BAD_WRITE);
   return ok;
 }
 

3rdparty/libwebp/src/enc/vp8i_enc.h

@@ -31,8 +31,8 @@ extern "C" {
 
 // version numbers
 #define ENC_MAJ_VERSION 1
-#define ENC_MIN_VERSION 2
-#define ENC_REV_VERSION 0
+#define ENC_MIN_VERSION 3
+#define ENC_REV_VERSION 1
 
 enum { MAX_LF_LEVELS = 64,       // Maximum loop filter level
        MAX_VARIABLE_LEVEL = 67,  // last (inclusive) level with variable cost
@@ -286,8 +286,7 @@ int VP8IteratorNext(VP8EncIterator* const it);
 // save the yuv_out_ boundary values to top_/left_ arrays for next iterations.
 void VP8IteratorSaveBoundary(VP8EncIterator* const it);
 // Report progression based on macroblock rows. Return 0 for user-abort request.
-int VP8IteratorProgress(const VP8EncIterator* const it,
-                        int final_delta_percent);
+int VP8IteratorProgress(const VP8EncIterator* const it, int delta);
 // Intra4x4 iterations
 void VP8IteratorStartI4(VP8EncIterator* const it);
 // returns true if not done.
@@ -471,7 +470,8 @@ int VP8EncAnalyze(VP8Encoder* const enc);
 // Sets up segment's quantization values, base_quant_ and filter strengths.
 void VP8SetSegmentParams(VP8Encoder* const enc, float quality);
 // Pick best modes and fills the levels. Returns true if skipped.
-int VP8Decimate(VP8EncIterator* const it, VP8ModeScore* const rd,
+int VP8Decimate(VP8EncIterator* WEBP_RESTRICT const it,
+                VP8ModeScore* WEBP_RESTRICT const rd,
                 VP8RDLevel rd_opt);
 
   // in alpha.c
@@ -491,19 +491,24 @@ int VP8FilterStrengthFromDelta(int sharpness, int delta);
 
   // misc utils for picture_*.c:
 
+// Returns true if 'picture' is non-NULL and dimensions/colorspace are within
+// their valid ranges. If returning false, the 'error_code' in 'picture' is
+// updated.
+int WebPValidatePicture(const WebPPicture* const picture);
+
 // Remove reference to the ARGB/YUVA buffer (doesn't free anything).
 void WebPPictureResetBuffers(WebPPicture* const picture);
 
-// Allocates ARGB buffer of given dimension (previous one is always free'd).
-// Preserves the YUV(A) buffer. Returns false in case of error (invalid param,
-// out-of-memory).
-int WebPPictureAllocARGB(WebPPicture* const picture, int width, int height);
+// Allocates ARGB buffer according to set width/height (previous one is
+// always free'd). Preserves the YUV(A) buffer. Returns false in case of error
+// (invalid param, out-of-memory).
+int WebPPictureAllocARGB(WebPPicture* const picture);
 
-// Allocates YUVA buffer of given dimension (previous one is always free'd).
-// Uses picture->csp to determine whether an alpha buffer is needed.
+// Allocates YUVA buffer according to set width/height (previous one is always
+// free'd). Uses picture->csp to determine whether an alpha buffer is needed.
 // Preserves the ARGB buffer.
 // Returns false in case of error (invalid param, out-of-memory).
-int WebPPictureAllocYUVA(WebPPicture* const picture, int width, int height);
+int WebPPictureAllocYUVA(WebPPicture* const picture);
 
 // Replace samples that are fully transparent by 'color' to help compressibility
 // (no guarantee, though). Assumes pic->use_argb is true.

3rdparty/libwebp/src/enc/vp8li_enc.h

@@ -69,6 +69,8 @@ typedef struct {
   int use_palette_;
   int palette_size_;
   uint32_t palette_[MAX_PALETTE_SIZE];
+  // Sorted version of palette_ for cache purposes.
+  uint32_t palette_sorted_[MAX_PALETTE_SIZE];
 
   // Some 'scratch' (potentially large) objects.
   struct VP8LBackwardRefs refs_[4];  // Backward Refs array for temporaries.
@@ -87,9 +89,10 @@ int VP8LEncodeImage(const WebPConfig* const config,
 
 // Encodes the main image stream using the supplied bit writer.
 // If 'use_cache' is false, disables the use of color cache.
-WebPEncodingError VP8LEncodeStream(const WebPConfig* const config,
-                                   const WebPPicture* const picture,
-                                   VP8LBitWriter* const bw, int use_cache);
+// Returns false in case of error (stored in picture->error_code).
+int VP8LEncodeStream(const WebPConfig* const config,
+                     const WebPPicture* const picture, VP8LBitWriter* const bw,
+                     int use_cache);
 
 #if (WEBP_NEAR_LOSSLESS == 1)
 // in near_lossless.c
@@ -101,13 +104,18 @@ int VP8ApplyNearLossless(const WebPPicture* const picture, int quality,
 //------------------------------------------------------------------------------
 // Image transforms in predictor.c.
 
-void VP8LResidualImage(int width, int height, int bits, int low_effort,
-                       uint32_t* const argb, uint32_t* const argb_scratch,
-                       uint32_t* const image, int near_lossless, int exact,
-                       int used_subtract_green);
-
-void VP8LColorSpaceTransform(int width, int height, int bits, int quality,
-                             uint32_t* const argb, uint32_t* image);
+// pic and percent are for progress.
+// Returns false in case of error (stored in pic->error_code).
+int VP8LResidualImage(int width, int height, int bits, int low_effort,
+                      uint32_t* const argb, uint32_t* const argb_scratch,
+                      uint32_t* const image, int near_lossless, int exact,
+                      int used_subtract_green, const WebPPicture* const pic,
+                      int percent_range, int* const percent);
+
+int VP8LColorSpaceTransform(int width, int height, int bits, int quality,
+                            uint32_t* const argb, uint32_t* image,
+                            const WebPPicture* const pic, int percent_range,
+                            int* const percent);
 
 //------------------------------------------------------------------------------
 

3rdparty/libwebp/src/enc/webp_enc.c

@@ -307,7 +307,10 @@ int WebPEncodingSetError(const WebPPicture* const pic,
                          WebPEncodingError error) {
   assert((int)error < VP8_ENC_ERROR_LAST);
   assert((int)error >= VP8_ENC_OK);
-  ((WebPPicture*)pic)->error_code = error;
+  // The oldest error reported takes precedence over the new one.
+  if (pic->error_code == VP8_ENC_OK) {
+    ((WebPPicture*)pic)->error_code = error;
+  }
   return 0;
 }
 
@@ -317,8 +320,7 @@ int WebPReportProgress(const WebPPicture* const pic,
     *percent_store = percent;
     if (pic->progress_hook && !pic->progress_hook(percent, pic)) {
       // user abort requested
-      WebPEncodingSetError(pic, VP8_ENC_ERROR_USER_ABORT);
-      return 0;
+      return WebPEncodingSetError(pic, VP8_ENC_ERROR_USER_ABORT);
     }
   }
   return 1;  // ok
@@ -329,16 +331,14 @@ int WebPEncode(const WebPConfig* config, WebPPicture* pic) {
   int ok = 0;
   if (pic == NULL) return 0;
 
-  WebPEncodingSetError(pic, VP8_ENC_OK);  // all ok so far
+  pic->error_code = VP8_ENC_OK;  // all ok so far
   if (config == NULL) {  // bad params
     return WebPEncodingSetError(pic, VP8_ENC_ERROR_NULL_PARAMETER);
   }
   if (!WebPValidateConfig(config)) {
     return WebPEncodingSetError(pic, VP8_ENC_ERROR_INVALID_CONFIGURATION);
   }
-  if (pic->width <= 0 || pic->height <= 0) {
-    return WebPEncodingSetError(pic, VP8_ENC_ERROR_BAD_DIMENSION);
-  }
+  if (!WebPValidatePicture(pic)) return 0;
   if (pic->width > WEBP_MAX_DIMENSION || pic->height > WEBP_MAX_DIMENSION) {
     return WebPEncodingSetError(pic, VP8_ENC_ERROR_BAD_DIMENSION);
   }

3rdparty/libwebp/src/mux/anim_encode.c

@@ -248,9 +248,6 @@ WebPAnimEncoder* WebPAnimEncoderNewInternal(
 
   enc = (WebPAnimEncoder*)WebPSafeCalloc(1, sizeof(*enc));
   if (enc == NULL) return NULL;
-  // sanity inits, so we can call WebPAnimEncoderDelete():
-  enc->encoded_frames_ = NULL;
-  enc->mux_ = NULL;
   MarkNoError(enc);
 
   // Dimensions and options.
@@ -421,7 +418,7 @@ static void MinimizeChangeRectangle(const WebPPicture* const src,
   const int max_allowed_diff_lossy = QualityToMaxDiff(quality);
   const int max_allowed_diff = is_lossless ? 0 : max_allowed_diff_lossy;
 
-  // Sanity checks.
+  // Assumption/correctness checks.
   assert(src->width == dst->width && src->height == dst->height);
   assert(rect->x_offset_ + rect->width_ <= dst->width);
   assert(rect->y_offset_ + rect->height_ <= dst->height);
@@ -596,16 +593,17 @@ int WebPAnimEncoderRefineRect(
     int is_lossless, float quality, int* const x_offset, int* const y_offset,
     int* const width, int* const height) {
   FrameRectangle rect;
-  const int right = clip(*x_offset + *width, 0, curr_canvas->width);
-  const int left = clip(*x_offset, 0, curr_canvas->width - 1);
-  const int bottom = clip(*y_offset + *height, 0, curr_canvas->height);
-  const int top = clip(*y_offset, 0, curr_canvas->height - 1);
+  int right, left, bottom, top;
   if (prev_canvas == NULL || curr_canvas == NULL ||
       prev_canvas->width != curr_canvas->width ||
       prev_canvas->height != curr_canvas->height ||
       !prev_canvas->use_argb || !curr_canvas->use_argb) {
     return 0;
   }
+  right = clip(*x_offset + *width, 0, curr_canvas->width);
+  left = clip(*x_offset, 0, curr_canvas->width - 1);
+  bottom = clip(*y_offset + *height, 0, curr_canvas->height);
+  top = clip(*y_offset, 0, curr_canvas->height - 1);
   rect.x_offset_ = left;
   rect.y_offset_ = top;
   rect.width_ = clip(right - left, 0, curr_canvas->width - rect.x_offset_);
@@ -949,7 +947,8 @@ static int IncreasePreviousDuration(WebPAnimEncoder* const enc, int duration) {
   int new_duration;
 
   assert(enc->count_ >= 1);
-  assert(prev_enc_frame->sub_frame_.duration ==
+  assert(!prev_enc_frame->is_key_frame_ ||
+         prev_enc_frame->sub_frame_.duration ==
          prev_enc_frame->key_frame_.duration);
   assert(prev_enc_frame->sub_frame_.duration ==
          (prev_enc_frame->sub_frame_.duration & (MAX_DURATION - 1)));
@@ -966,7 +965,7 @@ static int IncreasePreviousDuration(WebPAnimEncoder* const enc, int duration) {
       0x10, 0x88, 0x88, 0x08
     };
     const WebPData lossless_1x1 = {
-        lossless_1x1_bytes, sizeof(lossless_1x1_bytes)
+      lossless_1x1_bytes, sizeof(lossless_1x1_bytes)
     };
     const uint8_t lossy_1x1_bytes[] = {
       0x52, 0x49, 0x46, 0x46, 0x40, 0x00, 0x00, 0x00, 0x57, 0x45, 0x42, 0x50,
@@ -1358,6 +1357,12 @@ int WebPAnimEncoderAdd(WebPAnimEncoder* enc, WebPPicture* frame, int timestamp,
     if (!IncreasePreviousDuration(enc, (int)prev_frame_duration)) {
       return 0;
     }
+    // IncreasePreviousDuration() may add a frame to avoid exceeding
+    // MAX_DURATION which could cause CacheFrame() to over read encoded_frames_
+    // before the next flush.
+    if (enc->count_ == enc->size_ && !FlushFrames(enc)) {
+      return 0;
+    }
   } else {
     enc->first_timestamp_ = timestamp;
   }

3rdparty/libwebp/src/mux/muxedit.c

@@ -70,6 +70,7 @@ void WebPMuxDelete(WebPMux* mux) {
     err = ChunkAssignData(&chunk, data, copy_data, tag);                       \
     if (err == WEBP_MUX_OK) {                                                  \
       err = ChunkSetHead(&chunk, (LIST));                                      \
+      if (err != WEBP_MUX_OK) ChunkRelease(&chunk);                            \
     }                                                                          \
     return err;                                                                \
   }
@@ -235,7 +236,6 @@ WebPMuxError WebPMuxSetImage(WebPMux* mux, const WebPData* bitstream,
   WebPMuxImage wpi;
   WebPMuxError err;
 
-  // Sanity checks.
   if (mux == NULL || bitstream == NULL || bitstream->bytes == NULL ||
       bitstream->size > MAX_CHUNK_PAYLOAD) {
     return WEBP_MUX_INVALID_ARGUMENT;
@@ -267,7 +267,6 @@ WebPMuxError WebPMuxPushFrame(WebPMux* mux, const WebPMuxFrameInfo* info,
   WebPMuxImage wpi;
   WebPMuxError err;
 
-  // Sanity checks.
   if (mux == NULL || info == NULL) return WEBP_MUX_INVALID_ARGUMENT;
 
   if (info->id != WEBP_CHUNK_ANMF) return WEBP_MUX_INVALID_ARGUMENT;
@@ -556,7 +555,8 @@ static WebPMuxError MuxCleanup(WebPMux* const mux) {
   if (num_frames == 1) {
     WebPMuxImage* frame = NULL;
     err = MuxImageGetNth((const WebPMuxImage**)&mux->images_, 1, &frame);
-    assert(err == WEBP_MUX_OK);  // We know that one frame does exist.
+    if (err != WEBP_MUX_OK) return err;
+    // We know that one frame does exist.
     assert(frame != NULL);
     if (frame->header_ != NULL &&
         ((mux->canvas_width_ == 0 && mux->canvas_height_ == 0) ||

3rdparty/libwebp/src/mux/muxi.h

@@ -28,8 +28,8 @@ extern "C" {
 // Defines and constants.
 
 #define MUX_MAJ_VERSION 1
-#define MUX_MIN_VERSION 2
-#define MUX_REV_VERSION 0
+#define MUX_MIN_VERSION 3
+#define MUX_REV_VERSION 1
 
 // Chunk object.
 typedef struct WebPChunk WebPChunk;

3rdparty/libwebp/src/mux/muxinternal.c

@@ -155,17 +155,18 @@ WebPMuxError ChunkSetHead(WebPChunk* const chunk,
 
 WebPMuxError ChunkAppend(WebPChunk* const chunk,
                          WebPChunk*** const chunk_list) {
+  WebPMuxError err;
   assert(chunk_list != NULL && *chunk_list != NULL);
 
   if (**chunk_list == NULL) {
-    ChunkSetHead(chunk, *chunk_list);
+    err = ChunkSetHead(chunk, *chunk_list);
   } else {
     WebPChunk* last_chunk = **chunk_list;
     while (last_chunk->next_ != NULL) last_chunk = last_chunk->next_;
-    ChunkSetHead(chunk, &last_chunk->next_);
-    *chunk_list = &last_chunk->next_;
+    err = ChunkSetHead(chunk, &last_chunk->next_);
+    if (err == WEBP_MUX_OK) *chunk_list = &last_chunk->next_;
   }
-  return WEBP_MUX_OK;
+  return err;
 }
 
 //------------------------------------------------------------------------------

3rdparty/libwebp/src/mux/muxread.c

@@ -56,7 +56,7 @@ static WebPMuxError ChunkVerifyAndAssign(WebPChunk* chunk,
   uint32_t chunk_size;
   WebPData chunk_data;
 
-  // Sanity checks.
+  // Correctness checks.
   if (data_size < CHUNK_HEADER_SIZE) return WEBP_MUX_NOT_ENOUGH_DATA;
   chunk_size = GetLE32(data + TAG_SIZE);
   if (chunk_size > MAX_CHUNK_PAYLOAD) return WEBP_MUX_BAD_DATA;
@@ -116,9 +116,12 @@ static int MuxImageParse(const WebPChunk* const chunk, int copy_data,
     // Each of ANMF chunk contain a header at the beginning. So, its size should
     // be at least 'hdr_size'.
     if (size < hdr_size) goto Fail;
-    ChunkAssignData(&subchunk, &temp, copy_data, chunk->tag_);
+    if (ChunkAssignData(&subchunk, &temp, copy_data,
+                        chunk->tag_) != WEBP_MUX_OK) {
+      goto Fail;
+    }
   }
-  ChunkSetHead(&subchunk, &wpi->header_);
+  if (ChunkSetHead(&subchunk, &wpi->header_) != WEBP_MUX_OK) goto Fail;
   wpi->is_partial_ = 1;  // Waiting for ALPH and/or VP8/VP8L chunks.
 
   // Rest of the chunks.
@@ -186,7 +189,6 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
   WebPChunk** chunk_list_ends[WEBP_CHUNK_NIL + 1] = { NULL };
   ChunkInit(&chunk);
 
-  // Sanity checks.
   if (WEBP_ABI_IS_INCOMPATIBLE(version, WEBP_MUX_ABI_VERSION)) {
     return NULL;  // version mismatch
   }
@@ -481,7 +483,6 @@ WebPMuxError WebPMuxGetFrame(
   WebPMuxError err;
   WebPMuxImage* wpi;
 
-  // Sanity checks.
   if (mux == NULL || frame == NULL) {
     return WEBP_MUX_INVALID_ARGUMENT;
   }

3rdparty/libwebp/src/utils/bit_reader_inl_utils.h

@@ -55,7 +55,7 @@ void VP8LoadFinalBytes(VP8BitReader* const br);
 
 // makes sure br->value_ has at least BITS bits worth of data
 static WEBP_UBSAN_IGNORE_UNDEF WEBP_INLINE
-void VP8LoadNewBytes(VP8BitReader* const br) {
+void VP8LoadNewBytes(VP8BitReader* WEBP_RESTRICT const br) {
   assert(br != NULL && br->buf_ != NULL);
   // Read 'BITS' bits at a time if possible.
   if (br->buf_ < br->buf_max_) {
@@ -104,7 +104,7 @@ void VP8LoadNewBytes(VP8BitReader* const br) {
 }
 
 // Read a bit with proba 'prob'. Speed-critical function!
-static WEBP_INLINE int VP8GetBit(VP8BitReader* const br,
+static WEBP_INLINE int VP8GetBit(VP8BitReader* WEBP_RESTRICT const br,
                                  int prob, const char label[]) {
   // Don't move this declaration! It makes a big speed difference to store
   // 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
@@ -137,7 +137,8 @@ static WEBP_INLINE int VP8GetBit(VP8BitReader* const br,
 
 // simplified version of VP8GetBit() for prob=0x80 (note shift is always 1 here)
 static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
-int VP8GetSigned(VP8BitReader* const br, int v, const char label[]) {
+int VP8GetSigned(VP8BitReader* WEBP_RESTRICT const br, int v,
+                 const char label[]) {
   if (br->bits_ < 0) {
     VP8LoadNewBytes(br);
   }
@@ -147,15 +148,15 @@ int VP8GetSigned(VP8BitReader* const br, int v, const char label[]) {
     const range_t value = (range_t)(br->value_ >> pos);
     const int32_t mask = (int32_t)(split - value) >> 31;  // -1 or 0
     br->bits_ -= 1;
-    br->range_ += mask;
+    br->range_ += (range_t)mask;
     br->range_ |= 1;
-    br->value_ -= (bit_t)((split + 1) & mask) << pos;
+    br->value_ -= (bit_t)((split + 1) & (uint32_t)mask) << pos;
     BT_TRACK(br);
     return (v ^ mask) - mask;
   }
 }
 
-static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br,
+static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* WEBP_RESTRICT const br,
                                     int prob, const char label[]) {
   // Don't move this declaration! It makes a big speed difference to store
   // 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't

3rdparty/libwebp/src/utils/bit_reader_utils.c

@@ -15,6 +15,7 @@
 #include "src/webp/config.h"
 #endif
 
+#include "src/dsp/cpu.h"
 #include "src/utils/bit_reader_inl_utils.h"
 #include "src/utils/utils.h"
 
@@ -121,7 +122,7 @@ int32_t VP8GetSignedValue(VP8BitReader* const br, int bits,
 
 #define VP8L_LOG8_WBITS 4  // Number of bytes needed to store VP8L_WBITS bits.
 
-#if defined(__arm__) || defined(_M_ARM) || defined(__aarch64__) || \
+#if defined(__arm__) || defined(_M_ARM) || WEBP_AARCH64 || \
     defined(__i386__) || defined(_M_IX86) || \
     defined(__x86_64__) || defined(_M_X64)
 #define VP8L_USE_FAST_LOAD

3rdparty/libwebp/src/utils/bit_reader_utils.h

@@ -19,6 +19,7 @@
 #ifdef _MSC_VER
 #include <stdlib.h>  // _byteswap_ulong
 #endif
+#include "src/dsp/cpu.h"
 #include "src/webp/types.h"
 
 // Warning! This macro triggers quite some MACRO wizardry around func signature!
@@ -64,7 +65,7 @@ extern "C" {
 #define BITS 56
 #elif defined(__arm__) || defined(_M_ARM)      // ARM
 #define BITS 24
-#elif defined(__aarch64__)                     // ARM 64bit
+#elif WEBP_AARCH64                             // ARM 64bit
 #define BITS 56
 #elif defined(__mips__)                        // MIPS
 #define BITS 24

3rdparty/libwebp/src/utils/bit_writer_utils.c

@@ -278,7 +278,7 @@ void VP8LPutBitsFlushBits(VP8LBitWriter* const bw) {
   // If needed, make some room by flushing some bits out.
   if (bw->cur_ + VP8L_WRITER_BYTES > bw->end_) {
     const uint64_t extra_size = (bw->end_ - bw->buf_) + MIN_EXTRA_SIZE;
-    if (extra_size != (size_t)extra_size ||
+    if (!CheckSizeOverflow(extra_size) ||
         !VP8LBitWriterResize(bw, (size_t)extra_size)) {
       bw->cur_ = bw->buf_;
       bw->error_ = 1;
@@ -314,7 +314,7 @@ void VP8LPutBitsInternal(VP8LBitWriter* const bw, uint32_t bits, int n_bits) {
     while (used >= VP8L_WRITER_BITS) {
       if (bw->cur_ + VP8L_WRITER_BYTES > bw->end_) {
         const uint64_t extra_size = (bw->end_ - bw->buf_) + MIN_EXTRA_SIZE;
-        if (extra_size != (size_t)extra_size ||
+        if (!CheckSizeOverflow(extra_size) ||
             !VP8LBitWriterResize(bw, (size_t)extra_size)) {
           bw->cur_ = bw->buf_;
           bw->error_ = 1;

3rdparty/libwebp/src/utils/color_cache_utils.c

@@ -20,22 +20,22 @@
 //------------------------------------------------------------------------------
 // VP8LColorCache.
 
-int VP8LColorCacheInit(VP8LColorCache* const cc, int hash_bits) {
+int VP8LColorCacheInit(VP8LColorCache* const color_cache, int hash_bits) {
   const int hash_size = 1 << hash_bits;
-  assert(cc != NULL);
+  assert(color_cache != NULL);
   assert(hash_bits > 0);
-  cc->colors_ = (uint32_t*)WebPSafeCalloc((uint64_t)hash_size,
-                                          sizeof(*cc->colors_));
-  if (cc->colors_ == NULL) return 0;
-  cc->hash_shift_ = 32 - hash_bits;
-  cc->hash_bits_ = hash_bits;
+  color_cache->colors_ = (uint32_t*)WebPSafeCalloc(
+      (uint64_t)hash_size, sizeof(*color_cache->colors_));
+  if (color_cache->colors_ == NULL) return 0;
+  color_cache->hash_shift_ = 32 - hash_bits;
+  color_cache->hash_bits_ = hash_bits;
   return 1;
 }
 
-void VP8LColorCacheClear(VP8LColorCache* const cc) {
-  if (cc != NULL) {
-    WebPSafeFree(cc->colors_);
-    cc->colors_ = NULL;
+void VP8LColorCacheClear(VP8LColorCache* const color_cache) {
+  if (color_cache != NULL) {
+    WebPSafeFree(color_cache->colors_);
+    color_cache->colors_ = NULL;
   }
 }