libwebp: Sync with upstream 1.2.4

Changes: https://chromium.googlesource.com/webm/libwebp/+/1.2.4/NEWS (cherry picked from commit e5e34f21fc)
2022-08-23 16:47:48 +02:00 · 2022-08-23 16:47:48 +02:00 · b129ae2127
parent a8c01896cc
commit b129ae2127
47 changed files with 2611 additions and 1669 deletions
--- a/modules/webp/SCsub
+++ b/modules/webp/SCsub
@ -12,123 +12,129 @@ thirdparty_obj = []
 if env["builtin_libwebp"]:
    thirdparty_dir = "#thirdparty/libwebp/"
    thirdparty_sources = [
-        "dec/alpha_dec.c",
-        "dec/buffer_dec.c",
-        "dec/frame_dec.c",
-        "dec/idec_dec.c",
-        "dec/io_dec.c",
-        "dec/quant_dec.c",
-        "dec/tree_dec.c",
-        "dec/vp8_dec.c",
-        "dec/vp8l_dec.c",
-        "dec/webp_dec.c",
-        "demux/anim_decode.c",
-        "demux/demux.c",
-        "dsp/alpha_processing.c",
-        "dsp/alpha_processing_mips_dsp_r2.c",
-        "dsp/alpha_processing_neon.c",
-        "dsp/alpha_processing_sse2.c",
-        "dsp/alpha_processing_sse41.c",
-        "dsp/cost.c",
-        "dsp/cost_mips32.c",
-        "dsp/cost_mips_dsp_r2.c",
-        "dsp/cost_neon.c",
-        "dsp/cost_sse2.c",
-        "dsp/cpu.c",
-        "dsp/dec.c",
-        "dsp/dec_clip_tables.c",
-        "dsp/dec_mips32.c",
-        "dsp/dec_mips_dsp_r2.c",
-        "dsp/dec_msa.c",
-        "dsp/dec_neon.c",
-        "dsp/dec_sse2.c",
-        "dsp/dec_sse41.c",
-        "dsp/enc.c",
-        "dsp/enc_mips32.c",
-        "dsp/enc_mips_dsp_r2.c",
-        "dsp/enc_msa.c",
-        "dsp/enc_neon.c",
-        "dsp/enc_sse2.c",
-        "dsp/enc_sse41.c",
-        "dsp/filters.c",
-        "dsp/filters_mips_dsp_r2.c",
-        "dsp/filters_msa.c",
-        "dsp/filters_neon.c",
-        "dsp/filters_sse2.c",
-        "dsp/lossless.c",
-        "dsp/lossless_enc.c",
-        "dsp/lossless_enc_mips32.c",
-        "dsp/lossless_enc_mips_dsp_r2.c",
-        "dsp/lossless_enc_msa.c",
-        "dsp/lossless_enc_neon.c",
-        "dsp/lossless_enc_sse2.c",
-        "dsp/lossless_enc_sse41.c",
-        "dsp/lossless_mips_dsp_r2.c",
-        "dsp/lossless_msa.c",
-        "dsp/lossless_neon.c",
-        "dsp/lossless_sse2.c",
-        "dsp/lossless_sse41.c",
-        "dsp/rescaler.c",
-        "dsp/rescaler_mips32.c",
-        "dsp/rescaler_mips_dsp_r2.c",
-        "dsp/rescaler_msa.c",
-        "dsp/rescaler_neon.c",
-        "dsp/rescaler_sse2.c",
-        "dsp/ssim.c",
-        "dsp/ssim_sse2.c",
-        "dsp/upsampling.c",
-        "dsp/upsampling_mips_dsp_r2.c",
-        "dsp/upsampling_msa.c",
-        "dsp/upsampling_neon.c",
-        "dsp/upsampling_sse2.c",
-        "dsp/upsampling_sse41.c",
-        "dsp/yuv.c",
-        "dsp/yuv_mips32.c",
-        "dsp/yuv_mips_dsp_r2.c",
-        "dsp/yuv_neon.c",
-        "dsp/yuv_sse2.c",
-        "dsp/yuv_sse41.c",
-        "enc/alpha_enc.c",
-        "enc/analysis_enc.c",
-        "enc/backward_references_cost_enc.c",
-        "enc/backward_references_enc.c",
-        "enc/config_enc.c",
-        "enc/cost_enc.c",
-        "enc/filter_enc.c",
-        "enc/frame_enc.c",
-        "enc/histogram_enc.c",
-        "enc/iterator_enc.c",
-        "enc/near_lossless_enc.c",
-        "enc/picture_csp_enc.c",
-        "enc/picture_enc.c",
-        "enc/picture_psnr_enc.c",
-        "enc/picture_rescale_enc.c",
-        "enc/picture_tools_enc.c",
-        "enc/predictor_enc.c",
-        "enc/quant_enc.c",
-        "enc/syntax_enc.c",
-        "enc/token_enc.c",
-        "enc/tree_enc.c",
-        "enc/vp8l_enc.c",
-        "enc/webp_enc.c",
-        "mux/anim_encode.c",
-        "mux/muxedit.c",
-        "mux/muxinternal.c",
-        "mux/muxread.c",
-        "utils/bit_reader_utils.c",
-        "utils/bit_writer_utils.c",
-        "utils/color_cache_utils.c",
-        "utils/filters_utils.c",
-        "utils/huffman_encode_utils.c",
-        "utils/huffman_utils.c",
-        "utils/quant_levels_dec_utils.c",
-        "utils/quant_levels_utils.c",
-        "utils/random_utils.c",
-        "utils/rescaler_utils.c",
-        "utils/thread_utils.c",
-        "utils/utils.c",
+        "sharpyuv/sharpyuv.c",
+        "sharpyuv/sharpyuv_csp.c",
+        "sharpyuv/sharpyuv_dsp.c",
+        "sharpyuv/sharpyuv_gamma.c",
+        "sharpyuv/sharpyuv_neon.c",
+        "sharpyuv/sharpyuv_sse2.c",
+        "src/dec/alpha_dec.c",
+        "src/dec/buffer_dec.c",
+        "src/dec/frame_dec.c",
+        "src/dec/idec_dec.c",
+        "src/dec/io_dec.c",
+        "src/dec/quant_dec.c",
+        "src/dec/tree_dec.c",
+        "src/dec/vp8_dec.c",
+        "src/dec/vp8l_dec.c",
+        "src/dec/webp_dec.c",
+        "src/demux/anim_decode.c",
+        "src/demux/demux.c",
+        "src/dsp/alpha_processing.c",
+        "src/dsp/alpha_processing_mips_dsp_r2.c",
+        "src/dsp/alpha_processing_neon.c",
+        "src/dsp/alpha_processing_sse2.c",
+        "src/dsp/alpha_processing_sse41.c",
+        "src/dsp/cost.c",
+        "src/dsp/cost_mips32.c",
+        "src/dsp/cost_mips_dsp_r2.c",
+        "src/dsp/cost_neon.c",
+        "src/dsp/cost_sse2.c",
+        "src/dsp/cpu.c",
+        "src/dsp/dec.c",
+        "src/dsp/dec_clip_tables.c",
+        "src/dsp/dec_mips32.c",
+        "src/dsp/dec_mips_dsp_r2.c",
+        "src/dsp/dec_msa.c",
+        "src/dsp/dec_neon.c",
+        "src/dsp/dec_sse2.c",
+        "src/dsp/dec_sse41.c",
+        "src/dsp/enc.c",
+        "src/dsp/enc_mips32.c",
+        "src/dsp/enc_mips_dsp_r2.c",
+        "src/dsp/enc_msa.c",
+        "src/dsp/enc_neon.c",
+        "src/dsp/enc_sse2.c",
+        "src/dsp/enc_sse41.c",
+        "src/dsp/filters.c",
+        "src/dsp/filters_mips_dsp_r2.c",
+        "src/dsp/filters_msa.c",
+        "src/dsp/filters_neon.c",
+        "src/dsp/filters_sse2.c",
+        "src/dsp/lossless.c",
+        "src/dsp/lossless_enc.c",
+        "src/dsp/lossless_enc_mips32.c",
+        "src/dsp/lossless_enc_mips_dsp_r2.c",
+        "src/dsp/lossless_enc_msa.c",
+        "src/dsp/lossless_enc_neon.c",
+        "src/dsp/lossless_enc_sse2.c",
+        "src/dsp/lossless_enc_sse41.c",
+        "src/dsp/lossless_mips_dsp_r2.c",
+        "src/dsp/lossless_msa.c",
+        "src/dsp/lossless_neon.c",
+        "src/dsp/lossless_sse2.c",
+        "src/dsp/lossless_sse41.c",
+        "src/dsp/rescaler.c",
+        "src/dsp/rescaler_mips32.c",
+        "src/dsp/rescaler_mips_dsp_r2.c",
+        "src/dsp/rescaler_msa.c",
+        "src/dsp/rescaler_neon.c",
+        "src/dsp/rescaler_sse2.c",
+        "src/dsp/ssim.c",
+        "src/dsp/ssim_sse2.c",
+        "src/dsp/upsampling.c",
+        "src/dsp/upsampling_mips_dsp_r2.c",
+        "src/dsp/upsampling_msa.c",
+        "src/dsp/upsampling_neon.c",
+        "src/dsp/upsampling_sse2.c",
+        "src/dsp/upsampling_sse41.c",
+        "src/dsp/yuv.c",
+        "src/dsp/yuv_mips32.c",
+        "src/dsp/yuv_mips_dsp_r2.c",
+        "src/dsp/yuv_neon.c",
+        "src/dsp/yuv_sse2.c",
+        "src/dsp/yuv_sse41.c",
+        "src/enc/alpha_enc.c",
+        "src/enc/analysis_enc.c",
+        "src/enc/backward_references_cost_enc.c",
+        "src/enc/backward_references_enc.c",
+        "src/enc/config_enc.c",
+        "src/enc/cost_enc.c",
+        "src/enc/filter_enc.c",
+        "src/enc/frame_enc.c",
+        "src/enc/histogram_enc.c",
+        "src/enc/iterator_enc.c",
+        "src/enc/near_lossless_enc.c",
+        "src/enc/picture_csp_enc.c",
+        "src/enc/picture_enc.c",
+        "src/enc/picture_psnr_enc.c",
+        "src/enc/picture_rescale_enc.c",
+        "src/enc/picture_tools_enc.c",
+        "src/enc/predictor_enc.c",
+        "src/enc/quant_enc.c",
+        "src/enc/syntax_enc.c",
+        "src/enc/token_enc.c",
+        "src/enc/tree_enc.c",
+        "src/enc/vp8l_enc.c",
+        "src/enc/webp_enc.c",
+        "src/mux/anim_encode.c",
+        "src/mux/muxedit.c",
+        "src/mux/muxinternal.c",
+        "src/mux/muxread.c",
+        "src/utils/bit_reader_utils.c",
+        "src/utils/bit_writer_utils.c",
+        "src/utils/color_cache_utils.c",
+        "src/utils/filters_utils.c",
+        "src/utils/huffman_encode_utils.c",
+        "src/utils/huffman_utils.c",
+        "src/utils/quant_levels_dec_utils.c",
+        "src/utils/quant_levels_utils.c",
+        "src/utils/random_utils.c",
+        "src/utils/rescaler_utils.c",
+        "src/utils/thread_utils.c",
+        "src/utils/utils.c",
    ]
-    thirdparty_sources = [thirdparty_dir + "src/" + file for file in thirdparty_sources]
+    thirdparty_sources = [thirdparty_dir + file for file in thirdparty_sources]

    env_webp.Prepend(CPPPATH=[thirdparty_dir, thirdparty_dir + "src/"])

--- a/thirdparty/README.md
+++ b/thirdparty/README.md
@ -266,12 +266,12 @@ from the Android NDK r18.
 ## libwebp

 - Upstream: https://chromium.googlesource.com/webm/libwebp/
- Version: 1.2.2 (b0a860891dcd4c0c2d7c6149e5cccb6eb881cc21, 2022)
+- Version: 1.2.4 (0d1f12546bd803099a60c070517a552483f3790e, 2022)
 - License: BSD-3-Clause

 Files extracted from upstream source:

- `src/*` except from: `.am`, `.rc` and `.in` files
+- `src/` and `sharpyuv/` except from: `.am`, `.rc` and `.in` files
 - `AUTHORS`, `COPYING`, `PATENTS`


--- a/thirdparty/libwebp/AUTHORS
+++ b/thirdparty/libwebp/AUTHORS
@ -1,12 +1,15 @@
 Contributors:
 - Aidan O'Loan (aidanol at gmail dot com)
 - Alan Browning (browning at google dot com)
+- Alexandru Ardelean (ardeleanalex at gmail dot com)
+- Brian Ledger (brianpl at google dot com)
 - Charles Munger (clm at google dot com)
 - Cheng Yi (cyi at google dot com)
 - Christian Duvivier (cduvivier at google dot com)
 - Christopher Degawa (ccom at randomderp dot com)
 - Clement Courbet (courbet at google dot com)
 - Djordje Pesut (djordje dot pesut at imgtec dot com)
+- Frank Barchard (fbarchard at google dot com)
 - Hui Su (huisu at google dot com)
 - Ilya Kurdyukov (jpegqs at gmail dot com)
 - Ingvar Stepanyan (rreverser at google dot com)
@ -22,6 +25,7 @@ Contributors:
 - Mans Rullgard (mans at mansr dot com)
 - Marcin Kowalczyk (qrczak at google dot com)
 - Martin Olsson (mnemo at minimum dot se)
+- Maryla Ustarroz-Calonge (maryla at google dot com)
 - Mikołaj Zalewski (mikolajz at google dot com)
 - Mislav Bradac (mislavm at google dot com)
 - Nico Weber (thakis at chromium dot org)
--- a/thirdparty/libwebp/sharpyuv/sharpyuv.c
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv.c
@ -0,0 +1,498 @@
+// 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 <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "src/webp/types.h"
+#include "src/dsp/cpu.h"
+#include "sharpyuv/sharpyuv_dsp.h"
+#include "sharpyuv/sharpyuv_gamma.h"
+
+//------------------------------------------------------------------------------
+// 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) {
+  const int bit_depth = rgb_bit_depth + GetPrecisionShift(rgb_bit_depth);
+  const uint32_t A = SharpYuvGammaToLinear(a, bit_depth);
+  const uint32_t B = SharpYuvGammaToLinear(b, bit_depth);
+  const uint32_t C = SharpYuvGammaToLinear(c, bit_depth);
+  const uint32_t D = SharpYuvGammaToLinear(d, bit_depth);
+  return SharpYuvLinearToGamma((A + B + C + D + 2) >> 2, bit_depth);
+}
+
+static WEBP_INLINE void UpdateW(const fixed_y_t* src, fixed_y_t* dst, int w,
+                                int rgb_bit_depth) {
+  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);
+    const uint32_t G = SharpYuvGammaToLinear(src[1 * w + i], bit_depth);
+    const uint32_t B = SharpYuvGammaToLinear(src[2 * w + i], bit_depth);
+    const uint32_t Y = RGBToGray(R, G, B);
+    dst[i] = (fixed_y_t)SharpYuvLinearToGamma(Y, bit_depth);
+  }
+}
+
+static void UpdateChroma(const fixed_y_t* src1, const fixed_y_t* src2,
+                         fixed_t* dst, int uv_w, int rgb_bit_depth) {
+  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);
+    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);
+    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);
+    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) {
+  // 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);
+    UpdateW(src2, target_y + w, w, rgb_bit_depth);
+    UpdateChroma(src1, src2, target_uv, uv_w, rgb_bit_depth);
+    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);
+      UpdateW(src2, best_rgb_y + 1 * w, w, rgb_bit_depth);
+      UpdateChroma(src1, src2, best_rgb_uv, uv_w, rgb_bit_depth);
+
+      // 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
+
+// Hidden exported init function.
+// By default SharpYuvConvert calls it with NULL. If needed, users can declare
+// it as extern and call it with a VP8CPUInfo function.
+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;
+  const int initialized =
+      (sharpyuv_last_cpuinfo_used != (VP8CPUInfo)&sharpyuv_last_cpuinfo_used);
+  if (cpu_info_func == NULL && initialized) return;
+  if (sharpyuv_last_cpuinfo_used == cpu_info_func) return;
+
+  SharpYuvInitDsp(cpu_info_func);
+  if (!initialized) {
+    SharpYuvInitGammaTables();
+  }
+
+  sharpyuv_last_cpuinfo_used = cpu_info_func;
+}
+
+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) {
+  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;
+  }
+  SharpYuvInit(NULL);
+
+  // 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);
+}
+
+//------------------------------------------------------------------------------
--- a/thirdparty/libwebp/sharpyuv/sharpyuv.h
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv.h
@ -0,0 +1,81 @@
+// 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_
+
+#include <inttypes.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// SharpYUV API version following the convention from semver.org
+#define SHARPYUV_VERSION_MAJOR 0
+#define SHARPYUV_VERSION_MINOR 1
+#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)
+
+// 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;
+
+// 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
+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);
+
+// 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_
--- a/thirdparty/libwebp/sharpyuv/sharpyuv_csp.c
+++ b/thirdparty/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 <string.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;
+}
--- a/thirdparty/libwebp/sharpyuv/sharpyuv_csp.h
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv_csp.h
@ -0,0 +1,59 @@
+// 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.
+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.
+const SharpYuvConversionMatrix* SharpYuvGetConversionMatrix(
+    SharpYuvMatrixType matrix_type);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // WEBP_SHARPYUV_SHARPYUV_CSP_H_
--- a/thirdparty/libwebp/sharpyuv/sharpyuv_dsp.c
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv_dsp.c
@ -0,0 +1,102 @@
+// 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 "src/dsp/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 void InitSharpYuvSSE2(void);
+extern void InitSharpYuvNEON(void);
+
+void SharpYuvInitDsp(VP8CPUInfo cpu_info_func) {
+  (void)cpu_info_func;
+
+#if !WEBP_NEON_OMIT_C_CODE
+  SharpYuvUpdateY = SharpYuvUpdateY_C;
+  SharpYuvUpdateRGB = SharpYuvUpdateRGB_C;
+  SharpYuvFilterRow = SharpYuvFilterRow_C;
+#endif
+
+#if defined(WEBP_HAVE_SSE2)
+  if (cpu_info_func == NULL || cpu_info_func(kSSE2)) {
+    InitSharpYuvSSE2();
+  }
+#endif  // WEBP_HAVE_SSE2
+
+#if defined(WEBP_HAVE_NEON)
+  if (WEBP_NEON_OMIT_C_CODE || cpu_info_func == NULL || cpu_info_func(kNEON)) {
+    InitSharpYuvNEON();
+  }
+#endif  // WEBP_HAVE_NEON
+
+  assert(SharpYuvUpdateY != NULL);
+  assert(SharpYuvUpdateRGB != NULL);
+  assert(SharpYuvFilterRow != NULL);
+}
--- a/thirdparty/libwebp/sharpyuv/sharpyuv_dsp.h
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv_dsp.h
@ -0,0 +1,29 @@
+// 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 <stdint.h>
+
+#include "src/dsp/cpu.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(VP8CPUInfo cpu_info_func);
+
+#endif  // WEBP_SHARPYUV_SHARPYUV_DSP_H_
--- a/thirdparty/libwebp/sharpyuv/sharpyuv_gamma.c
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv_gamma.c
@ -0,0 +1,114 @@
+// 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 <math.h>
+#include <stdint.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;
+}
+
+uint32_t SharpYuvGammaToLinear(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);
+}
+
+uint16_t SharpYuvLinearToGamma(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);
+}
--- a/thirdparty/libwebp/sharpyuv/sharpyuv_gamma.h
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv_gamma.h
@ -0,0 +1,35 @@
+// 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 <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Initializes precomputed tables. Must be called once before calling
+// SharpYuvGammaToLinear or SharpYuvLinearToGamma.
+void SharpYuvInitGammaTables(void);
+
+// Converts a gamma color value on 'bit_depth' bits to a 16 bit linear value.
+uint32_t SharpYuvGammaToLinear(uint16_t v, int bit_depth);
+
+// Converts a 16 bit linear color value to a gamma value on 'bit_depth' bits.
+uint16_t SharpYuvLinearToGamma(uint32_t value, int bit_depth);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // WEBP_SHARPYUV_SHARPYUV_GAMMA_H_
--- a/thirdparty/libwebp/sharpyuv/sharpyuv_neon.c
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv_neon.c
@ -0,0 +1,182 @@
+// 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>
+#endif
+
+extern void InitSharpYuvNEON(void);
+
+#if defined(WEBP_USE_NEON)
+
+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);
+  }
+}
+
+//------------------------------------------------------------------------------
+
+WEBP_TSAN_IGNORE_FUNCTION void InitSharpYuvNEON(void) {
+  SharpYuvUpdateY = SharpYuvUpdateY_NEON;
+  SharpYuvUpdateRGB = SharpYuvUpdateRGB_NEON;
+  SharpYuvFilterRow = SharpYuvFilterRow_NEON;
+}
+
+#else  // !WEBP_USE_NEON
+
+void InitSharpYuvNEON(void) {}
+
+#endif  // WEBP_USE_NEON
--- a/thirdparty/libwebp/sharpyuv/sharpyuv_sse2.c
+++ b/thirdparty/libwebp/sharpyuv/sharpyuv_sse2.c
@ -0,0 +1,204 @@
+// 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>
+#endif
+
+extern void InitSharpYuvSSE2(void);
+
+#if defined(WEBP_USE_SSE2)
+
+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
+
+void InitSharpYuvSSE2(void) {}
+
+#endif  // WEBP_USE_SSE2
--- a/thirdparty/libwebp/src/dec/vp8i_dec.h
+++ b/thirdparty/libwebp/src/dec/vp8i_dec.h
@ -32,7 +32,7 @@ extern "C" {
 // version numbers
 #define DEC_MAJ_VERSION 1
 #define DEC_MIN_VERSION 2
-#define DEC_REV_VERSION 2
+#define DEC_REV_VERSION 4

 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
 // Constraints are: We need to store one 16x16 block of luma samples (y),
--- a/thirdparty/libwebp/src/dec/vp8l_dec.c
+++ b/thirdparty/libwebp/src/dec/vp8l_dec.c
@ -178,7 +178,7 @@ 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) {
@ -321,7 +321,7 @@ 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;
@ -1281,7 +1281,7 @@ 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) {
--- a/thirdparty/libwebp/src/demux/demux.c
+++ b/thirdparty/libwebp/src/demux/demux.c
@ -25,7 +25,7 @@

 #define DMUX_MAJ_VERSION 1
 #define DMUX_MIN_VERSION 2
-#define DMUX_REV_VERSION 2
+#define DMUX_REV_VERSION 4

 typedef struct {
  size_t start_;        // start location of the data
@ -614,7 +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_) {
@ -649,8 +648,6 @@ static int IsValidExtendedFormat(const WebPDemuxer* const dmux) {
                            dmux->canvas_width_, dmux->canvas_height_)) {
        return 0;
      }
-
-      ++frame_count;
    }
  }
  return 1;
--- a/thirdparty/libwebp/src/dsp/alpha_processing_neon.c
+++ b/thirdparty/libwebp/src/dsp/alpha_processing_neon.c
@ -83,7 +83,7 @@ static void ApplyAlphaMultiply_NEON(uint8_t* rgba, int alpha_first,
 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 = 0xffffffffu;
+  uint32_t alpha_mask = 0xffu;
  uint8x8_t mask8 = vdup_n_u8(0xff);
  uint32_t tmp[2];
  int i, j;
@ -107,6 +107,7 @@ static int DispatchAlpha_NEON(const uint8_t* WEBP_RESTRICT alpha,
    dst += dst_stride;
  }
  vst1_u8((uint8_t*)tmp, mask8);
+  alpha_mask *= 0x01010101;
  alpha_mask &= tmp[0];
  alpha_mask &= tmp[1];
  return (alpha_mask != 0xffffffffu);
@ -135,7 +136,7 @@ static void DispatchAlphaToGreen_NEON(const uint8_t* WEBP_RESTRICT alpha,
 static int ExtractAlpha_NEON(const uint8_t* WEBP_RESTRICT argb, int argb_stride,
                             int width, int height,
                             uint8_t* WEBP_RESTRICT alpha, int alpha_stride) {
-  uint32_t alpha_mask = 0xffffffffu;
+  uint32_t alpha_mask = 0xffu;
  uint8x8_t mask8 = vdup_n_u8(0xff);
  uint32_t tmp[2];
  int i, j;
@ -157,6 +158,7 @@ static int ExtractAlpha_NEON(const uint8_t* WEBP_RESTRICT 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);
--- a/thirdparty/libwebp/src/dsp/cpu.c
+++ b/thirdparty/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>
--- a/thirdparty/libwebp/src/dsp/cpu.h
+++ b/thirdparty/libwebp/src/dsp/cpu.h
@ -0,0 +1,254 @@
+// 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_
+
+#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
+
+#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
+
+// 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)))
+#define WEBP_USE_NEON
+#define WEBP_USE_INTRINSICS
+#endif
+
+#if defined(WEBP_USE_NEON) && !defined(WEBP_HAVE_NEON)
+#define WEBP_HAVE_NEON
+#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(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) || \
+      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(__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;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// returns true if the CPU supports the feature.
+typedef int (*VP8CPUInfo)(CPUFeature feature);
+WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif
+
+#endif  // WEBP_DSP_CPU_H_
--- a/thirdparty/libwebp/src/dsp/dsp.h
+++ b/thirdparty/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
@ -43,225 +44,6 @@ extern "C" {
 #define WEBP_RESTRICT
 #endif

-//------------------------------------------------------------------------------
-// CPU detection
-
-#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
-
-#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
-
-// 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.
-#if defined(_MSC_VER) && \
-  ((_MSC_VER >= 1700 && defined(_M_ARM)) || \
-   (_MSC_VER >= 1920 && defined(_M_ARM64)))
-#define WEBP_USE_NEON
-#define WEBP_USE_INTRINSICS
-#endif
-
-#if defined(WEBP_USE_NEON) && !defined(WEBP_HAVE_NEON)
-#define WEBP_HAVE_NEON
-#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(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) || 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
@ -550,15 +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);

--- a/thirdparty/libwebp/src/dsp/lossless.h
+++ b/thirdparty/libwebp/src/dsp/lossless.h
@ -182,9 +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 float (*VP8LCostFunc)(const uint32_t* population, int length);
+typedef float (*VP8LCostCombinedFunc)(const uint32_t* X, const uint32_t* Y,
+                                      int length);
 typedef float (*VP8LCombinedShannonEntropyFunc)(const int X[256],
                                                const int Y[256]);

@ -198,7 +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
--- a/thirdparty/libwebp/src/dsp/lossless_enc.c
+++ b/thirdparty/libwebp/src/dsp/lossless_enc.c
@ -402,7 +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];
@ -418,7 +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) {
@ -636,17 +636,17 @@ void VP8LBundleColorMap_C(const uint8_t* const row, int width, int xbits,

 //------------------------------------------------------------------------------

-static double ExtraCost_C(const uint32_t* population, int length) {
+static float ExtraCost_C(const uint32_t* population, int length) {
  int i;
-  double cost = 0.;
+  float cost = 0.f;
  for (i = 2; i < length - 2; ++i) cost += (i >> 1) * population[i + 2];
  return cost;
 }

-static double ExtraCostCombined_C(const uint32_t* X, const uint32_t* Y,
+static float ExtraCostCombined_C(const uint32_t* X, const uint32_t* Y,
                                  int length) {
  int i;
-  double cost = 0.;
+  float cost = 0.f;
  for (i = 2; i < length - 2; ++i) {
    const int xy = X[i + 2] + Y[i + 2];
    cost += (i >> 1) * xy;
--- a/thirdparty/libwebp/src/dsp/lossless_enc_mips32.c
+++ b/thirdparty/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 (float)cost;
+static float 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 (float)((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 (float)cost;
+static float 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 (float)((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
--- a/thirdparty/libwebp/src/dsp/lossless_enc_sse2.c
+++ b/thirdparty/libwebp/src/dsp/lossless_enc_sse2.c
@ -239,7 +239,7 @@ static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {

 static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
  int i;
-  double retval = 0.;
+  float retval = 0.f;
  int sumX = 0, sumXY = 0;
  const __m128i zero = _mm_setzero_si128();

@ -273,7 +273,7 @@ static float CombinedShannonEntropy_SSE2(const int X[256], const int Y[256]) {
    }
  }
  retval += VP8LFastSLog2(sumX) + VP8LFastSLog2(sumXY);
-  return (float)retval;
+  return retval;
 }

 #else
--- a/thirdparty/libwebp/src/dsp/yuv.c
+++ b/thirdparty/libwebp/src/dsp/yuv.c
@ -194,50 +194,6 @@ 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 +203,9 @@ 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;
@ -269,17 +216,10 @@ 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_HAVE_SSE2)
    if (VP8GetCPUInfo(kSSE2)) {
      WebPInitConvertARGBToYUVSSE2();
-      WebPInitSharpYUVSSE2();
    }
 #endif  // WEBP_HAVE_SSE2
 #if defined(WEBP_HAVE_SSE41)
@ -293,7 +233,6 @@ WEBP_DSP_INIT_FUNC(WebPInitConvertARGBToYUV) {
  if (WEBP_NEON_OMIT_C_CODE ||
      (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
    WebPInitConvertARGBToYUVNEON();
-    WebPInitSharpYUVNEON();
  }
 #endif  // WEBP_HAVE_NEON

@ -302,7 +241,4 @@ WEBP_DSP_INIT_FUNC(WebPInitConvertARGBToYUV) {
  assert(WebPConvertRGB24ToY != NULL);
  assert(WebPConvertBGR24ToY != NULL);
  assert(WebPConvertRGBA32ToUV != NULL);
-  assert(WebPSharpYUVUpdateY != NULL);
-  assert(WebPSharpYUVUpdateRGB != NULL);
-  assert(WebPSharpYUVFilterRow != NULL);
 }
--- a/thirdparty/libwebp/src/dsp/yuv_neon.c
+++ b/thirdparty/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
--- a/thirdparty/libwebp/src/dsp/yuv_sse2.c
+++ b/thirdparty/libwebp/src/dsp/yuv_sse2.c
@ -747,128 +747,9 @@ 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
--- a/thirdparty/libwebp/src/enc/alpha_enc.c
+++ b/thirdparty/libwebp/src/enc/alpha_enc.c
@ -86,7 +86,7 @@ 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) {
--- a/thirdparty/libwebp/src/enc/backward_references_cost_enc.c
+++ b/thirdparty/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_.
-  manager->cache_intervals_size_ = 1;
-  manager->cost_cache_[0] = GetLengthCost(cost_model, 0);
-  for (i = 1; i < cost_cache_size; ++i) {
+  // 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;
+  for (i = 1; i < cost_cache_size; ++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);
--- a/thirdparty/libwebp/src/enc/backward_references_enc.c
+++ b/thirdparty/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 +282,13 @@ 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) {
+    WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
+    return 0;
+  }
+
+  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 +335,22 @@ 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 +439,17 @@ 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 +759,7 @@ 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 +844,14 @@ 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 +921,7 @@ 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 +933,7 @@ 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 +1007,16 @@ 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 +1032,37 @@ 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) {
+      WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
+      return 0;
+    }
    // 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;
+      WebPEncodingSetError(pic, VP8_ENC_ERROR_OUT_OF_MEMORY);
+      return 0;
    }
  }
-  return VP8_ENC_OK;
+
+  return WebPReportProgress(pic, *percent + percent_range, percent);
 }
--- a/thirdparty/libwebp/src/enc/backward_references_enc.h
+++ b/thirdparty/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
 }
--- a/thirdparty/libwebp/src/enc/histogram_enc.c
+++ b/thirdparty/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
@ -356,7 +357,7 @@ static WEBP_INLINE double GetCombinedEntropy(const uint32_t* const X,
 }

 // Estimates the Entropy + Huffman + other block overhead size cost.
-double VP8LHistogramEstimateBits(VP8LHistogram* const p) {
+float VP8LHistogramEstimateBits(VP8LHistogram* const p) {
  return
      PopulationCost(p->literal_, VP8LHistogramNumCodes(p->palette_code_bits_),
                     NULL, &p->is_used_[0])
@ -373,8 +374,7 @@ 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_);
@ -439,12 +439,11 @@ 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 +458,10 @@ 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 +472,22 @@ 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(
@ -505,10 +502,9 @@ 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);
  const int num_codes = VP8LHistogramNumCodes(h->palette_code_bits_);
@ -529,10 +525,10 @@ 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 +637,11 @@ 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 +671,10 @@ 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 +722,8 @@ 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 +778,9 @@ 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 +789,9 @@ 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 +935,8 @@ 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 +945,7 @@ 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 +1024,7 @@ 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 +1047,7 @@ 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 +1055,7 @@ 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 +1083,13 @@ 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;
 }
@ -1169,13 +1159,13 @@ static void RemoveEmptyHistograms(VP8LHistogramSet* const image_histo) {
 }

 int VP8LGetHistoImageSymbols(int xsize, int ysize,
-                             const VP8LBackwardRefs* const refs,
-                             int quality, int low_effort,
-                             int histogram_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;
+                             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 =
@ -1192,7 +1182,10 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
      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, histogram_bits, refs, orig_histo);
@ -1206,16 +1199,15 @@ 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);
  }
@ -1229,11 +1221,13 @@ 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;
      }
    }
@ -1243,10 +1237,12 @@ 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);
 }
--- a/thirdparty/libwebp/src/enc/histogram_enc.h
+++ b/thirdparty/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;

@ -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,
+                             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
 }
--- a/thirdparty/libwebp/src/enc/picture_csp_enc.c
+++ b/thirdparty/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
@ -76,16 +83,16 @@ 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);
@ -93,13 +100,13 @@ static void InitGammaTables(void);
 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 +118,12 @@ 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 +131,7 @@ 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 +165,41 @@ 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)))
+extern void SharpYuvInit(VP8CPUInfo cpu_info_func);
+
+static void SafeInitSharpYuv(void) {
+#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
+  static pthread_mutex_t initsharpyuv_lock = PTHREAD_MUTEX_INITIALIZER;
+  if (pthread_mutex_lock(&initsharpyuv_lock)) return;
+#endif
+
+  SharpYuvInit(VP8GetCPUInfo);
+
+#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
+  (void)pthread_mutex_unlock(&initsharpyuv_lock);
+#endif
+}

 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;
+  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);
  }
-  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;
-  }
-  // 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 +224,8 @@ 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 +451,20 @@ 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
  }
 }

@ -863,18 +505,18 @@ 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();
+    SafeInitSharpYuv();
    if (!PreprocessARGB(r_ptr, g_ptr, b_ptr, step, rgb_stride, picture)) {
      return 0;
    }
@ -1044,7 +686,7 @@ 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 +748,8 @@ 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 +807,24 @@ 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 +845,9 @@ 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;
 }

--- a/thirdparty/libwebp/src/enc/picture_enc.c
+++ b/thirdparty/libwebp/src/enc/picture_enc.c
@ -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;
--- a/thirdparty/libwebp/src/enc/picture_rescale_enc.c
+++ b/thirdparty/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

@ -198,34 +202,34 @@ 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;
  }

-  PictureGrabSpecs(pic, &tmp);
+  PictureGrabSpecs(picture, &tmp);
  tmp.width = width;
  tmp.height = height;
  if (!WebPPictureAlloc(&tmp)) return 0;

-  if (!pic->use_argb) {
+  if (!picture->use_argb) {
    work = (rescaler_t*)WebPSafeMalloc(2ULL * width, sizeof(*work));
    if (work == NULL) {
      WebPPictureFree(&tmp);
      return 0;
    }
    // If present, we need to rescale alpha first (for AlphaMultiplyY).
-    if (pic->a != NULL) {
+    if (picture->a != NULL) {
      WebPInitAlphaProcessing();
-      if (!RescalePlane(pic->a, prev_width, prev_height, pic->a_stride,
+      if (!RescalePlane(picture->a, prev_width, prev_height, picture->a_stride,
                        tmp.a, width, height, tmp.a_stride, work, 1)) {
        return 0;
      }
@ -233,17 +237,15 @@ int WebPPictureRescale(WebPPicture* pic, int width, int height) {

    // 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);
-    if (!RescalePlane(pic->y, prev_width, prev_height, pic->y_stride,
+    AlphaMultiplyY(picture, 0);
+    if (!RescalePlane(picture->y, prev_width, prev_height, picture->y_stride,
                      tmp.y, width, height, tmp.y_stride, work, 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)) {
+        !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 0;
    }
    AlphaMultiplyY(&tmp, 1);
@ -257,18 +259,17 @@ int WebPPictureRescale(WebPPicture* pic, int width, int height) {
    // weighting first (black-matting), scale the RGB values, and remove
    // the premultiplication afterward (while preserving the alpha channel).
    WebPInitAlphaProcessing();
-    AlphaMultiplyARGB(pic, 0);
-    if (!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 0;
    }
    AlphaMultiplyARGB(&tmp, 1);
  }
-  WebPPictureFree(pic);
+  WebPPictureFree(picture);
  WebPSafeFree(work);
-  *pic = tmp;
+  *picture = tmp;
  return 1;
 }

@ -280,23 +281,6 @@ 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;
--- a/thirdparty/libwebp/src/enc/picture_tools_enc.c
+++ b/thirdparty/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;
    }
  }
 }
--- a/thirdparty/libwebp/src/enc/predictor_enc.c
+++ b/thirdparty/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]);
  }
@ -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;
 }
--- a/thirdparty/libwebp/src/enc/quant_enc.c
+++ b/thirdparty/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;
@ -588,10 +590,10 @@ static WEBP_INLINE score_t RDScoreTrellis(int lambda, score_t rate,
 // Coefficient type.
 enum { TYPE_I16_AC = 0, TYPE_I16_DC = 1, TYPE_CHROMA_A = 2, TYPE_I4_AC = 3 };

-static int TrellisQuantizeBlock(const VP8Encoder* const enc,
+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 =
@ -767,9 +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];
@ -819,10 +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];
@ -855,7 +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;
@ -869,9 +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
@ -902,8 +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];
@ -922,8 +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;
@ -994,7 +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_;
@ -1054,7 +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;
@ -1063,7 +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_;
@ -1159,7 +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_;
@ -1211,7 +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;
@ -1236,9 +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;
@ -1352,7 +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_;
--- a/thirdparty/libwebp/src/enc/vp8i_enc.h
+++ b/thirdparty/libwebp/src/enc/vp8i_enc.h
@ -32,7 +32,7 @@ extern "C" {
 // version numbers
 #define ENC_MAJ_VERSION 1
 #define ENC_MIN_VERSION 2
-#define ENC_REV_VERSION 2
+#define ENC_REV_VERSION 4

 enum { MAX_LF_LEVELS = 64,       // Maximum loop filter level
       MAX_VARIABLE_LEVEL = 67,  // last (inclusive) level with variable cost
@ -470,7 +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
@ -490,19 +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.
--- a/thirdparty/libwebp/src/enc/vp8l_enc.c
+++ b/thirdparty/libwebp/src/enc/vp8l_enc.c
--- a/thirdparty/libwebp/src/enc/vp8li_enc.h
+++ b/thirdparty/libwebp/src/enc/vp8li_enc.h
@ -89,9 +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
@ -103,13 +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);
+// 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);

-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);

 //------------------------------------------------------------------------------

--- a/thirdparty/libwebp/src/enc/webp_enc.c
+++ b/thirdparty/libwebp/src/enc/webp_enc.c
@ -336,9 +336,7 @@ int WebPEncode(const WebPConfig* config, WebPPicture* pic) {
  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);
  }
--- a/thirdparty/libwebp/src/mux/muxedit.c
+++ b/thirdparty/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;                                                                \
  }
--- a/thirdparty/libwebp/src/mux/muxi.h
+++ b/thirdparty/libwebp/src/mux/muxi.h
@ -29,7 +29,7 @@ extern "C" {

 #define MUX_MAJ_VERSION 1
 #define MUX_MIN_VERSION 2
-#define MUX_REV_VERSION 2
+#define MUX_REV_VERSION 4

 // Chunk object.
 typedef struct WebPChunk WebPChunk;
--- a/thirdparty/libwebp/src/mux/muxinternal.c
+++ b/thirdparty/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;
 }

 //------------------------------------------------------------------------------
--- a/thirdparty/libwebp/src/webp/encode.h
+++ b/thirdparty/libwebp/src/webp/encode.h
@ -441,7 +441,7 @@ WEBP_EXTERN int WebPPictureCrop(WebPPicture* picture,
 // the original dimension will be lost). Picture 'dst' need not be initialized
 // with WebPPictureInit() if it is different from 'src', since its content will
 // be overwritten.
-// Returns false in case of memory allocation error or invalid parameters.
+// Returns false in case of invalid parameters.
 WEBP_EXTERN int WebPPictureView(const WebPPicture* src,
                                int left, int top, int width, int height,
                                WebPPicture* dst);
@ -455,7 +455,7 @@ WEBP_EXTERN int WebPPictureIsView(const WebPPicture* picture);
 // dimension will be calculated preserving the aspect ratio.
 // No gamma correction is applied.
 // Returns false in case of error (invalid parameter or insufficient memory).
-WEBP_EXTERN int WebPPictureRescale(WebPPicture* pic, int width, int height);
+WEBP_EXTERN int WebPPictureRescale(WebPPicture* picture, int width, int height);

 // Colorspace conversion function to import RGB samples.
 // Previous buffer will be free'd, if any.
@ -526,7 +526,7 @@ WEBP_EXTERN int WebPPictureHasTransparency(const WebPPicture* picture);
 // Remove the transparency information (if present) by blending the color with
 // the background color 'background_rgb' (specified as 24bit RGB triplet).
 // After this call, all alpha values are reset to 0xff.
-WEBP_EXTERN void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb);
+WEBP_EXTERN void WebPBlendAlpha(WebPPicture* picture, uint32_t background_rgb);

 //------------------------------------------------------------------------------
 // Main call