Update libwebp to 0.6.0

16 files changed, 195 insertions, 118 deletions

diff --git a/thirdparty/libwebp/dec/alpha.c b/thirdparty/libwebp/dec/alpha_dec.c
index d88f01d8de..83ffd4b609 100644
--- a/thirdparty/libwebp/dec/alpha.c
+++ b/thirdparty/libwebp/dec/alpha_dec.c
@@ -12,11 +12,11 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include <stdlib.h>
-#include "./alphai.h"
-#include "./vp8i.h"
-#include "./vp8li.h"
+#include "./alphai_dec.h"
+#include "./vp8i_dec.h"
+#include "./vp8li_dec.h"
 #include "../dsp/dsp.h"
-#include "../utils/quant_levels_dec.h"
+#include "../utils/quant_levels_dec_utils.h"
 #include "../utils/utils.h"
 #include "../webp/format_constants.h"
 
diff --git a/thirdparty/libwebp/dec/alphai.h b/thirdparty/libwebp/dec/alphai_dec.h
index 69dd7c0f5d..561e8151ee 100644
--- a/thirdparty/libwebp/dec/alphai.h
+++ b/thirdparty/libwebp/dec/alphai_dec.h
@@ -14,8 +14,8 @@
 #ifndef WEBP_DEC_ALPHAI_H_
 #define WEBP_DEC_ALPHAI_H_
 
-#include "./webpi.h"
-#include "../utils/filters.h"
+#include "./webpi_dec.h"
+#include "../utils/filters_utils.h"
 
 #ifdef __cplusplus
 extern "C" {
diff --git a/thirdparty/libwebp/dec/buffer.c b/thirdparty/libwebp/dec/buffer_dec.c
index 547e69b434..c685fd5646 100644
--- a/thirdparty/libwebp/dec/buffer.c
+++ b/thirdparty/libwebp/dec/buffer_dec.c
@@ -13,8 +13,8 @@
 
 #include <stdlib.h>
 
-#include "./vp8i.h"
-#include "./webpi.h"
+#include "./vp8i_dec.h"
+#include "./webpi_dec.h"
 #include "../utils/utils.h"
 
 //------------------------------------------------------------------------------
diff --git a/thirdparty/libwebp/dec/common.h b/thirdparty/libwebp/dec/common_dec.h
index 6961e22470..6961e22470 100644
--- a/thirdparty/libwebp/dec/common.h
+++ b/thirdparty/libwebp/dec/common_dec.h
diff --git a/thirdparty/libwebp/dec/frame.c b/thirdparty/libwebp/dec/frame_dec.c
index 22d291d2cd..f91e27f7c8 100644
--- a/thirdparty/libwebp/dec/frame.c
+++ b/thirdparty/libwebp/dec/frame_dec.c
@@ -12,7 +12,7 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include <stdlib.h>
-#include "./vp8i.h"
+#include "./vp8i_dec.h"
 #include "../utils/utils.h"
 
 //------------------------------------------------------------------------------
@@ -723,7 +723,7 @@ static int AllocateMemory(VP8Decoder* const dec) {
       return VP8SetError(dec, VP8_STATUS_OUT_OF_MEMORY,
                          "no memory during frame initialization.");
     }
-    // down-cast is ok, thanks to WebPSafeAlloc() above.
+    // down-cast is ok, thanks to WebPSafeMalloc() above.
     dec->mem_size_ = (size_t)needed;
   }
 
diff --git a/thirdparty/libwebp/dec/idec.c b/thirdparty/libwebp/dec/idec_dec.c
index 8de131916e..78fb2e7186 100644
--- a/thirdparty/libwebp/dec/idec.c
+++ b/thirdparty/libwebp/dec/idec_dec.c
@@ -15,9 +15,9 @@
 #include <string.h>
 #include <stdlib.h>
 
-#include "./alphai.h"
-#include "./webpi.h"
-#include "./vp8i.h"
+#include "./alphai_dec.h"
+#include "./webpi_dec.h"
+#include "./vp8i_dec.h"
 #include "../utils/utils.h"
 
 // In append mode, buffer allocations increase as multiples of this value.
diff --git a/thirdparty/libwebp/dec/io.c b/thirdparty/libwebp/dec/io_dec.c
index 8d5c43f325..8bfab86959 100644
--- a/thirdparty/libwebp/dec/io.c
+++ b/thirdparty/libwebp/dec/io_dec.c
@@ -13,8 +13,8 @@
 
 #include <assert.h>
 #include <stdlib.h>
-#include "../dec/vp8i.h"
-#include "./webpi.h"
+#include "../dec/vp8i_dec.h"
+#include "./webpi_dec.h"
 #include "../dsp/dsp.h"
 #include "../dsp/yuv.h"
 #include "../utils/utils.h"
@@ -256,7 +256,7 @@ static int Rescale(const uint8_t* src, int src_stride,
 static int EmitRescaledYUV(const VP8Io* const io, WebPDecParams* const p) {
   const int mb_h = io->mb_h;
   const int uv_mb_h = (mb_h + 1) >> 1;
-  WebPRescaler* const scaler = &p->scaler_y;
+  WebPRescaler* const scaler = p->scaler_y;
   int num_lines_out = 0;
   if (WebPIsAlphaMode(p->output->colorspace) && io->a != NULL) {
     // Before rescaling, we premultiply the luma directly into the io->y
@@ -267,29 +267,28 @@ static int EmitRescaledYUV(const VP8Io* const io, WebPDecParams* const p) {
                  io->a, io->width, io->mb_w, mb_h, 0);
   }
   num_lines_out = Rescale(io->y, io->y_stride, mb_h, scaler);
-  Rescale(io->u, io->uv_stride, uv_mb_h, &p->scaler_u);
-  Rescale(io->v, io->uv_stride, uv_mb_h, &p->scaler_v);
+  Rescale(io->u, io->uv_stride, uv_mb_h, p->scaler_u);
+  Rescale(io->v, io->uv_stride, uv_mb_h, p->scaler_v);
   return num_lines_out;
 }
 
 static int EmitRescaledAlphaYUV(const VP8Io* const io, WebPDecParams* const p,
                                 int expected_num_lines_out) {
   const WebPYUVABuffer* const buf = &p->output->u.YUVA;
+  uint8_t* const dst_a = buf->a + p->last_y * buf->a_stride;
   if (io->a != NULL) {
-    uint8_t* dst_y = buf->y + p->last_y * buf->y_stride;
-    const uint8_t* src_a = buf->a + p->last_y * buf->a_stride;
-    const int num_lines_out = Rescale(io->a, io->width, io->mb_h, &p->scaler_a);
-    (void)expected_num_lines_out;
+    uint8_t* const dst_y = buf->y + p->last_y * buf->y_stride;
+    const int num_lines_out = Rescale(io->a, io->width, io->mb_h, p->scaler_a);
     assert(expected_num_lines_out == num_lines_out);
     if (num_lines_out > 0) {   // unmultiply the Y
-      WebPMultRows(dst_y, buf->y_stride, src_a, buf->a_stride,
-                   p->scaler_a.dst_width, num_lines_out, 1);
+      WebPMultRows(dst_y, buf->y_stride, dst_a, buf->a_stride,
+                   p->scaler_a->dst_width, num_lines_out, 1);
     }
   } else if (buf->a != NULL) {
     // the user requested alpha, but there is none, set it to opaque.
     assert(p->last_y + expected_num_lines_out <= io->scaled_height);
-    FillAlphaPlane(buf->a + p->last_y * buf->a_stride,
-                   io->scaled_width, expected_num_lines_out, buf->a_stride);
+    FillAlphaPlane(dst_a, io->scaled_width, expected_num_lines_out,
+                   buf->a_stride);
   }
   return 0;
 }
@@ -305,31 +304,42 @@ static int InitYUVRescaler(const VP8Io* const io, WebPDecParams* const p) {
   const int uv_in_height = (io->mb_h + 1) >> 1;
   const size_t work_size = 2 * out_width;   // scratch memory for luma rescaler
   const size_t uv_work_size = 2 * uv_out_width;  // and for each u/v ones
-  size_t tmp_size;
+  size_t tmp_size, rescaler_size;
   rescaler_t* work;
+  WebPRescaler* scalers;
+  const int num_rescalers = has_alpha ? 4 : 3;
 
   tmp_size = (work_size + 2 * uv_work_size) * sizeof(*work);
   if (has_alpha) {
     tmp_size += work_size * sizeof(*work);
   }
-  p->memory = WebPSafeMalloc(1ULL, tmp_size);
+  rescaler_size = num_rescalers * sizeof(*p->scaler_y) + WEBP_ALIGN_CST;
+
+  p->memory = WebPSafeMalloc(1ULL, tmp_size + rescaler_size);
   if (p->memory == NULL) {
     return 0;   // memory error
   }
   work = (rescaler_t*)p->memory;
-  WebPRescalerInit(&p->scaler_y, io->mb_w, io->mb_h,
+
+  scalers = (WebPRescaler*)WEBP_ALIGN((const uint8_t*)work + tmp_size);
+  p->scaler_y = &scalers[0];
+  p->scaler_u = &scalers[1];
+  p->scaler_v = &scalers[2];
+  p->scaler_a = has_alpha ? &scalers[3] : NULL;
+
+  WebPRescalerInit(p->scaler_y, io->mb_w, io->mb_h,
                    buf->y, out_width, out_height, buf->y_stride, 1,
                    work);
-  WebPRescalerInit(&p->scaler_u, uv_in_width, uv_in_height,
+  WebPRescalerInit(p->scaler_u, uv_in_width, uv_in_height,
                    buf->u, uv_out_width, uv_out_height, buf->u_stride, 1,
                    work + work_size);
-  WebPRescalerInit(&p->scaler_v, uv_in_width, uv_in_height,
+  WebPRescalerInit(p->scaler_v, uv_in_width, uv_in_height,
                    buf->v, uv_out_width, uv_out_height, buf->v_stride, 1,
                    work + work_size + uv_work_size);
   p->emit = EmitRescaledYUV;
 
   if (has_alpha) {
-    WebPRescalerInit(&p->scaler_a, io->mb_w, io->mb_h,
+    WebPRescalerInit(p->scaler_a, io->mb_w, io->mb_h,
                      buf->a, out_width, out_height, buf->a_stride, 1,
                      work + work_size + 2 * uv_work_size);
     p->emit_alpha = EmitRescaledAlphaYUV;
@@ -349,15 +359,15 @@ static int ExportRGB(WebPDecParams* const p, int y_pos) {
   int num_lines_out = 0;
   // For RGB rescaling, because of the YUV420, current scan position
   // U/V can be +1/-1 line from the Y one.  Hence the double test.
-  while (WebPRescalerHasPendingOutput(&p->scaler_y) &&
-         WebPRescalerHasPendingOutput(&p->scaler_u)) {
+  while (WebPRescalerHasPendingOutput(p->scaler_y) &&
+         WebPRescalerHasPendingOutput(p->scaler_u)) {
     assert(y_pos + num_lines_out < p->output->height);
-    assert(p->scaler_u.y_accum == p->scaler_v.y_accum);
-    WebPRescalerExportRow(&p->scaler_y);
-    WebPRescalerExportRow(&p->scaler_u);
-    WebPRescalerExportRow(&p->scaler_v);
-    convert(p->scaler_y.dst, p->scaler_u.dst, p->scaler_v.dst,
-            dst, p->scaler_y.dst_width);
+    assert(p->scaler_u->y_accum == p->scaler_v->y_accum);
+    WebPRescalerExportRow(p->scaler_y);
+    WebPRescalerExportRow(p->scaler_u);
+    WebPRescalerExportRow(p->scaler_v);
+    convert(p->scaler_y->dst, p->scaler_u->dst, p->scaler_v->dst,
+            dst, p->scaler_y->dst_width);
     dst += buf->stride;
     ++num_lines_out;
   }
@@ -371,15 +381,15 @@ static int EmitRescaledRGB(const VP8Io* const io, WebPDecParams* const p) {
   int num_lines_out = 0;
   while (j < mb_h) {
     const int y_lines_in =
-        WebPRescalerImport(&p->scaler_y, mb_h - j,
+        WebPRescalerImport(p->scaler_y, mb_h - j,
                            io->y + j * io->y_stride, io->y_stride);
     j += y_lines_in;
-    if (WebPRescaleNeededLines(&p->scaler_u, uv_mb_h - uv_j)) {
+    if (WebPRescaleNeededLines(p->scaler_u, uv_mb_h - uv_j)) {
       const int u_lines_in =
-          WebPRescalerImport(&p->scaler_u, uv_mb_h - uv_j,
+          WebPRescalerImport(p->scaler_u, uv_mb_h - uv_j,
                              io->u + uv_j * io->uv_stride, io->uv_stride);
       const int v_lines_in =
-          WebPRescalerImport(&p->scaler_v, uv_mb_h - uv_j,
+          WebPRescalerImport(p->scaler_v, uv_mb_h - uv_j,
                              io->v + uv_j * io->uv_stride, io->uv_stride);
       (void)v_lines_in;   // remove a gcc warning
       assert(u_lines_in == v_lines_in);
@@ -400,13 +410,13 @@ static int ExportAlpha(WebPDecParams* const p, int y_pos, int max_lines_out) {
   int num_lines_out = 0;
   const int is_premult_alpha = WebPIsPremultipliedMode(colorspace);
   uint32_t non_opaque = 0;
-  const int width = p->scaler_a.dst_width;
+  const int width = p->scaler_a->dst_width;
 
-  while (WebPRescalerHasPendingOutput(&p->scaler_a) &&
+  while (WebPRescalerHasPendingOutput(p->scaler_a) &&
          num_lines_out < max_lines_out) {
     assert(y_pos + num_lines_out < p->output->height);
-    WebPRescalerExportRow(&p->scaler_a);
-    non_opaque |= WebPDispatchAlpha(p->scaler_a.dst, 0, width, 1, dst, 0);
+    WebPRescalerExportRow(p->scaler_a);
+    non_opaque |= WebPDispatchAlpha(p->scaler_a->dst, 0, width, 1, dst, 0);
     dst += buf->stride;
     ++num_lines_out;
   }
@@ -428,18 +438,18 @@ static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos,
 #endif
   int num_lines_out = 0;
   const WEBP_CSP_MODE colorspace = p->output->colorspace;
-  const int width = p->scaler_a.dst_width;
+  const int width = p->scaler_a->dst_width;
   const int is_premult_alpha = WebPIsPremultipliedMode(colorspace);
   uint32_t alpha_mask = 0x0f;
 
-  while (WebPRescalerHasPendingOutput(&p->scaler_a) &&
+  while (WebPRescalerHasPendingOutput(p->scaler_a) &&
          num_lines_out < max_lines_out) {
     int i;
     assert(y_pos + num_lines_out < p->output->height);
-    WebPRescalerExportRow(&p->scaler_a);
+    WebPRescalerExportRow(p->scaler_a);
     for (i = 0; i < width; ++i) {
       // Fill in the alpha value (converted to 4 bits).
-      const uint32_t alpha_value = p->scaler_a.dst[i] >> 4;
+      const uint32_t alpha_value = p->scaler_a->dst[i] >> 4;
       alpha_dst[2 * i] = (alpha_dst[2 * i] & 0xf0) | alpha_value;
       alpha_mask &= alpha_value;
     }
@@ -455,7 +465,7 @@ static int ExportAlphaRGBA4444(WebPDecParams* const p, int y_pos,
 static int EmitRescaledAlphaRGB(const VP8Io* const io, WebPDecParams* const p,
                                 int expected_num_out_lines) {
   if (io->a != NULL) {
-    WebPRescaler* const scaler = &p->scaler_a;
+    WebPRescaler* const scaler = p->scaler_a;
     int lines_left = expected_num_out_lines;
     const int y_end = p->last_y + lines_left;
     while (lines_left > 0) {
@@ -477,7 +487,9 @@ static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) {
   const size_t work_size = 2 * out_width;   // scratch memory for one rescaler
   rescaler_t* work;  // rescalers work area
   uint8_t* tmp;   // tmp storage for scaled YUV444 samples before RGB conversion
-  size_t tmp_size1, tmp_size2, total_size;
+  size_t tmp_size1, tmp_size2, total_size, rescaler_size;
+  WebPRescaler* scalers;
+  const int num_rescalers = has_alpha ? 4 : 3;
 
   tmp_size1 = 3 * work_size;
   tmp_size2 = 3 * out_width;
@@ -486,26 +498,35 @@ static int InitRGBRescaler(const VP8Io* const io, WebPDecParams* const p) {
     tmp_size2 += out_width;
   }
   total_size = tmp_size1 * sizeof(*work) + tmp_size2 * sizeof(*tmp);
-  p->memory = WebPSafeMalloc(1ULL, total_size);
+  rescaler_size = num_rescalers * sizeof(*p->scaler_y) + WEBP_ALIGN_CST;
+
+  p->memory = WebPSafeMalloc(1ULL, total_size + rescaler_size);
   if (p->memory == NULL) {
     return 0;   // memory error
   }
   work = (rescaler_t*)p->memory;
   tmp = (uint8_t*)(work + tmp_size1);
-  WebPRescalerInit(&p->scaler_y, io->mb_w, io->mb_h,
+
+  scalers = (WebPRescaler*)WEBP_ALIGN((const uint8_t*)work + total_size);
+  p->scaler_y = &scalers[0];
+  p->scaler_u = &scalers[1];
+  p->scaler_v = &scalers[2];
+  p->scaler_a = has_alpha ? &scalers[3] : NULL;
+
+  WebPRescalerInit(p->scaler_y, io->mb_w, io->mb_h,
                    tmp + 0 * out_width, out_width, out_height, 0, 1,
                    work + 0 * work_size);
-  WebPRescalerInit(&p->scaler_u, uv_in_width, uv_in_height,
+  WebPRescalerInit(p->scaler_u, uv_in_width, uv_in_height,
                    tmp + 1 * out_width, out_width, out_height, 0, 1,
                    work + 1 * work_size);
-  WebPRescalerInit(&p->scaler_v, uv_in_width, uv_in_height,
+  WebPRescalerInit(p->scaler_v, uv_in_width, uv_in_height,
                    tmp + 2 * out_width, out_width, out_height, 0, 1,
                    work + 2 * work_size);
   p->emit = EmitRescaledRGB;
   WebPInitYUV444Converters();
 
   if (has_alpha) {
-    WebPRescalerInit(&p->scaler_a, io->mb_w, io->mb_h,
+    WebPRescalerInit(p->scaler_a, io->mb_w, io->mb_h,
                      tmp + 3 * out_width, out_width, out_height, 0, 1,
                      work + 3 * work_size);
     p->emit_alpha = EmitRescaledAlphaRGB;
diff --git a/thirdparty/libwebp/dec/quant.c b/thirdparty/libwebp/dec/quant_dec.c
index 5b648f942c..14e3198946 100644
--- a/thirdparty/libwebp/dec/quant.c
+++ b/thirdparty/libwebp/dec/quant_dec.c
@@ -11,7 +11,7 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./vp8i.h"
+#include "./vp8i_dec.h"
 
 static WEBP_INLINE int clip(int v, int M) {
   return v < 0 ? 0 : v > M ? M : v;
diff --git a/thirdparty/libwebp/dec/tree.c b/thirdparty/libwebp/dec/tree_dec.c
index c2007ea733..9e805f60f3 100644
--- a/thirdparty/libwebp/dec/tree.c
+++ b/thirdparty/libwebp/dec/tree_dec.c
@@ -11,10 +11,13 @@
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./vp8i.h"
-#include "../utils/bit_reader_inl.h"
+#include "./vp8i_dec.h"
+#include "../utils/bit_reader_inl_utils.h"
 
+#if !defined(__arm__) && !defined(_M_ARM) && !defined(__aarch64__)
+// using a table is ~1-2% slower on ARM. Prefer the coded-tree approach then.
 #define USE_GENERIC_TREE
+#endif
 
 #ifdef USE_GENERIC_TREE
 static const int8_t kYModesIntra4[18] = {
diff --git a/thirdparty/libwebp/dec/vp8.c b/thirdparty/libwebp/dec/vp8_dec.c
index 336680c38c..fad8d9cf35 100644
--- a/thirdparty/libwebp/dec/vp8.c
+++ b/thirdparty/libwebp/dec/vp8_dec.c
@@ -13,11 +13,11 @@
 
 #include <stdlib.h>
 
-#include "./alphai.h"
-#include "./vp8i.h"
-#include "./vp8li.h"
-#include "./webpi.h"
-#include "../utils/bit_reader_inl.h"
+#include "./alphai_dec.h"
+#include "./vp8i_dec.h"
+#include "./vp8li_dec.h"
+#include "./webpi_dec.h"
+#include "../utils/bit_reader_inl_utils.h"
 #include "../utils/utils.h"
 
 //------------------------------------------------------------------------------
@@ -27,6 +27,16 @@ int WebPGetDecoderVersion(void) {
 }
 
 //------------------------------------------------------------------------------
+// Signature and pointer-to-function for GetCoeffs() variants below.
+
+typedef int (*GetCoeffsFunc)(VP8BitReader* const br,
+                             const VP8BandProbas* const prob[],
+                             int ctx, const quant_t dq, int n, int16_t* out);
+static volatile GetCoeffsFunc GetCoeffs = NULL;
+
+static void InitGetCoeffs(void);
+
+//------------------------------------------------------------------------------
 // VP8Decoder
 
 static void SetOk(VP8Decoder* const dec) {
@@ -51,6 +61,7 @@ VP8Decoder* VP8New(void) {
     WebPGetWorkerInterface()->Init(&dec->worker_);
     dec->ready_ = 0;
     dec->num_parts_minus_one_ = 0;
+    InitGetCoeffs();
   }
   return dec;
 }
@@ -273,12 +284,14 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
     frm_hdr->profile_ = (bits >> 1) & 7;
     frm_hdr->show_ = (bits >> 4) & 1;
     frm_hdr->partition_length_ = (bits >> 5);
-    if (frm_hdr->profile_ > 3)
+    if (frm_hdr->profile_ > 3) {
       return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
                          "Incorrect keyframe parameters.");
-    if (!frm_hdr->show_)
+    }
+    if (!frm_hdr->show_) {
       return VP8SetError(dec, VP8_STATUS_UNSUPPORTED_FEATURE,
                          "Frame not displayable.");
+    }
     buf += 3;
     buf_size -= 3;
   }
@@ -420,8 +433,9 @@ static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) {
 }
 
 // Returns the position of the last non-zero coeff plus one
-static int GetCoeffs(VP8BitReader* const br, const VP8BandProbas* const prob[],
-                     int ctx, const quant_t dq, int n, int16_t* out) {
+static int GetCoeffsFast(VP8BitReader* const br,
+                         const VP8BandProbas* const prob[],
+                         int ctx, const quant_t dq, int n, int16_t* out) {
   const uint8_t* p = prob[n]->probas_[ctx];
   for (; n < 16; ++n) {
     if (!VP8GetBit(br, p[0])) {
@@ -447,6 +461,46 @@ static int GetCoeffs(VP8BitReader* const br, const VP8BandProbas* const prob[],
   return 16;
 }
 
+// This version of GetCoeffs() uses VP8GetBitAlt() which is an alternate version
+// of VP8GetBitAlt() targeting specific platforms.
+static int GetCoeffsAlt(VP8BitReader* const br,
+                        const VP8BandProbas* const prob[],
+                        int ctx, const quant_t dq, int n, int16_t* out) {
+  const uint8_t* p = prob[n]->probas_[ctx];
+  for (; n < 16; ++n) {
+    if (!VP8GetBitAlt(br, p[0])) {
+      return n;  // previous coeff was last non-zero coeff
+    }
+    while (!VP8GetBitAlt(br, p[1])) {       // sequence of zero coeffs
+      p = prob[++n]->probas_[0];
+      if (n == 16) return 16;
+    }
+    {        // non zero coeff
+      const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
+      int v;
+      if (!VP8GetBitAlt(br, p[2])) {
+        v = 1;
+        p = p_ctx[1];
+      } else {
+        v = GetLargeValue(br, p);
+        p = p_ctx[2];
+      }
+      out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
+    }
+  }
+  return 16;
+}
+
+WEBP_TSAN_IGNORE_FUNCTION static void InitGetCoeffs(void) {
+  if (GetCoeffs == NULL) {
+    if (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kSlowSSSE3)) {
+      GetCoeffs = GetCoeffsAlt;
+    } else {
+      GetCoeffs = GetCoeffsFast;
+    }
+  }
+}
+
 static WEBP_INLINE uint32_t NzCodeBits(uint32_t nz_coeffs, int nz, int dc_nz) {
   nz_coeffs <<= 2;
   nz_coeffs |= (nz > 3) ? 3 : (nz > 1) ? 2 : dc_nz;
diff --git a/thirdparty/libwebp/dec/decode_vp8.h b/thirdparty/libwebp/dec/vp8_dec.h
index b9337bbec0..b9337bbec0 100644
--- a/thirdparty/libwebp/dec/decode_vp8.h
+++ b/thirdparty/libwebp/dec/vp8_dec.h
diff --git a/thirdparty/libwebp/dec/vp8i.h b/thirdparty/libwebp/dec/vp8i_dec.h
index 313d8a7b94..555853e8f8 100644
--- a/thirdparty/libwebp/dec/vp8i.h
+++ b/thirdparty/libwebp/dec/vp8i_dec.h
@@ -15,11 +15,11 @@
 #define WEBP_DEC_VP8I_H_
 
 #include <string.h>     // for memcpy()
-#include "./common.h"
-#include "./vp8li.h"
-#include "../utils/bit_reader.h"
-#include "../utils/random.h"
-#include "../utils/thread.h"
+#include "./common_dec.h"
+#include "./vp8li_dec.h"
+#include "../utils/bit_reader_utils.h"
+#include "../utils/random_utils.h"
+#include "../utils/thread_utils.h"
 #include "../dsp/dsp.h"
 
 #ifdef __cplusplus
@@ -31,8 +31,8 @@ extern "C" {
 
 // version numbers
 #define DEC_MAJ_VERSION 0
-#define DEC_MIN_VERSION 5
-#define DEC_REV_VERSION 2
+#define DEC_MIN_VERSION 6
+#define DEC_REV_VERSION 0
 
 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
 // Constraints are: We need to store one 16x16 block of luma samples (y),
diff --git a/thirdparty/libwebp/dec/vp8l.c b/thirdparty/libwebp/dec/vp8l_dec.c
index cb2e3176b6..ef359a91f0 100644
--- a/thirdparty/libwebp/dec/vp8l.c
+++ b/thirdparty/libwebp/dec/vp8l_dec.c
@@ -14,13 +14,14 @@
 
 #include <stdlib.h>
 
-#include "./alphai.h"
-#include "./vp8li.h"
+#include "./alphai_dec.h"
+#include "./vp8li_dec.h"
 #include "../dsp/dsp.h"
 #include "../dsp/lossless.h"
+#include "../dsp/lossless_common.h"
 #include "../dsp/yuv.h"
-#include "../utils/endian_inl.h"
-#include "../utils/huffman.h"
+#include "../utils/endian_inl_utils.h"
+#include "../utils/huffman_utils.h"
 #include "../utils/utils.h"
 
 #define NUM_ARGB_CACHE_ROWS          16
@@ -547,11 +548,14 @@ static int EmitRescaledRowsRGBA(const VP8LDecoder* const dec,
     uint8_t* const row_out = out + num_lines_out * out_stride;
     const int lines_left = mb_h - num_lines_in;
     const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left);
+    int lines_imported;
     assert(needed_lines > 0 && needed_lines <= lines_left);
     WebPMultARGBRows(row_in, in_stride,
                      dec->rescaler->src_width, needed_lines, 0);
-    WebPRescalerImport(dec->rescaler, lines_left, row_in, in_stride);
-    num_lines_in += needed_lines;
+    lines_imported =
+        WebPRescalerImport(dec->rescaler, lines_left, row_in, in_stride);
+    assert(lines_imported == needed_lines);
+    num_lines_in += lines_imported;
     num_lines_out += Export(dec->rescaler, colorspace, out_stride, row_out);
   }
   return num_lines_out;
@@ -623,9 +627,12 @@ static int EmitRescaledRowsYUVA(const VP8LDecoder* const dec,
   while (num_lines_in < mb_h) {
     const int lines_left = mb_h - num_lines_in;
     const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left);
+    int lines_imported;
     WebPMultARGBRows(in, in_stride, dec->rescaler->src_width, needed_lines, 0);
-    WebPRescalerImport(dec->rescaler, lines_left, in, in_stride);
-    num_lines_in += needed_lines;
+    lines_imported =
+        WebPRescalerImport(dec->rescaler, lines_left, in, in_stride);
+    assert(lines_imported == needed_lines);
+    num_lines_in += lines_imported;
     in += needed_lines * in_stride;
     y_pos += ExportYUVA(dec, y_pos);
   }
@@ -705,13 +712,15 @@ static void ApplyInverseTransforms(VP8LDecoder* const dec, int num_rows,
   uint32_t* const rows_out = dec->argb_cache_;
 
   // Inverse transforms.
-  // TODO: most transforms only need to operate on the cropped region only.
-  memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out));
   while (n-- > 0) {
     VP8LTransform* const transform = &dec->transforms_[n];
     VP8LInverseTransform(transform, start_row, end_row, rows_in, rows_out);
     rows_in = rows_out;
   }
+  if (rows_in != rows_out) {
+    // No transform called, hence just copy.
+    memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out));
+  }
 }
 
 // Processes (transforms, scales & color-converts) the rows decoded after the
@@ -1210,8 +1219,9 @@ static int ExpandColorMap(int num_colors, VP8LTransform* const transform) {
       // Equivalent to AddPixelEq(), on a byte-basis.
       new_data[i] = (data[i] + new_data[i - 4]) & 0xff;
     }
-    for (; i < 4 * final_num_colors; ++i)
+    for (; i < 4 * final_num_colors; ++i) {
       new_data[i] = 0;  // black tail.
+    }
     WebPSafeFree(transform->data_);
     transform->data_ = new_color_map;
   }
@@ -1482,9 +1492,8 @@ static void ExtractAlphaRows(VP8LDecoder* const dec, int last_row) {
     const int cache_pixs = width * num_rows_to_process;
     uint8_t* const dst = output + width * cur_row;
     const uint32_t* const src = dec->argb_cache_;
-    int i;
     ApplyInverseTransforms(dec, num_rows_to_process, in);
-    for (i = 0; i < cache_pixs; ++i) dst[i] = (src[i] >> 8) & 0xff;
+    WebPExtractGreen(src, dst, cache_pixs);
     AlphaApplyFilter(alph_dec,
                      cur_row, cur_row + num_rows_to_process, dst, width);
     num_rows -= num_rows_to_process;
@@ -1552,6 +1561,8 @@ int VP8LDecodeAlphaImageStream(ALPHDecoder* const alph_dec, int last_row) {
     return 1;  // done
   }
 
+  if (!alph_dec->use_8b_decode_) WebPInitAlphaProcessing();
+
   // Decode (with special row processing).
   return alph_dec->use_8b_decode_ ?
       DecodeAlphaData(dec, (uint8_t*)dec->pixels_, dec->width_, dec->height_,
diff --git a/thirdparty/libwebp/dec/vp8li.h b/thirdparty/libwebp/dec/vp8li_dec.h
index 9313bdc0af..097a9d0589 100644
--- a/thirdparty/libwebp/dec/vp8li.h
+++ b/thirdparty/libwebp/dec/vp8li_dec.h
@@ -16,10 +16,10 @@
 #define WEBP_DEC_VP8LI_H_
 
 #include <string.h>     // for memcpy()
-#include "./webpi.h"
-#include "../utils/bit_reader.h"
-#include "../utils/color_cache.h"
-#include "../utils/huffman.h"
+#include "./webpi_dec.h"
+#include "../utils/bit_reader_utils.h"
+#include "../utils/color_cache_utils.h"
+#include "../utils/huffman_utils.h"
 
 #ifdef __cplusplus
 extern "C" {
diff --git a/thirdparty/libwebp/dec/webp.c b/thirdparty/libwebp/dec/webp_dec.c
index d0b912f02f..a8e9c2c510 100644
--- a/thirdparty/libwebp/dec/webp.c
+++ b/thirdparty/libwebp/dec/webp_dec.c
@@ -13,9 +13,9 @@
 
 #include <stdlib.h>
 
-#include "./vp8i.h"
-#include "./vp8li.h"
-#include "./webpi.h"
+#include "./vp8i_dec.h"
+#include "./vp8li_dec.h"
+#include "./webpi_dec.h"
 #include "../utils/utils.h"
 #include "../webp/mux_types.h"  // ALPHA_FLAG
 
@@ -39,8 +39,8 @@
 //   20..23  VP8X flags bit-map corresponding to the chunk-types present.
 //   24..26  Width of the Canvas Image.
 //   27..29  Height of the Canvas Image.
-// There can be extra chunks after the "VP8X" chunk (ICCP, FRGM, ANMF, VP8,
-// VP8L, XMP, EXIF  ...)
+// There can be extra chunks after the "VP8X" chunk (ICCP, ANMF, VP8, VP8L,
+// XMP, EXIF  ...)
 // All sizes are in little-endian order.
 // Note: chunk data size must be padded to multiple of 2 when written.
 
@@ -289,7 +289,6 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
   int found_riff = 0;
   int found_vp8x = 0;
   int animation_present = 0;
-  int fragments_present = 0;
   const int have_all_data = (headers != NULL) ? headers->have_all_data : 0;
 
   VP8StatusCode status;
@@ -318,7 +317,6 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
       return status;  // Wrong VP8X / insufficient data.
     }
     animation_present = !!(flags & ANIMATION_FLAG);
-    fragments_present = !!(flags & FRAGMENTS_FLAG);
     if (!found_riff && found_vp8x) {
       // Note: This restriction may be removed in the future, if it becomes
       // necessary to send VP8X chunk to the decoder.
@@ -330,8 +328,7 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
 
     image_width = canvas_width;
     image_height = canvas_height;
-    if (found_vp8x && (animation_present || fragments_present) &&
-        headers == NULL) {
+    if (found_vp8x && animation_present && headers == NULL) {
       status = VP8_STATUS_OK;
       goto ReturnWidthHeight;  // Just return features from VP8X header.
     }
@@ -362,7 +359,7 @@ static VP8StatusCode ParseHeadersInternal(const uint8_t* data,
     return VP8_STATUS_BITSTREAM_ERROR;
   }
 
-  if (format != NULL && !(animation_present || fragments_present)) {
+  if (format != NULL && !animation_present) {
     *format = hdrs.is_lossless ? 2 : 1;
   }
 
diff --git a/thirdparty/libwebp/dec/webpi.h b/thirdparty/libwebp/dec/webpi_dec.h
index 991b194c22..696abc1958 100644
--- a/thirdparty/libwebp/dec/webpi.h
+++ b/thirdparty/libwebp/dec/webpi_dec.h
@@ -18,8 +18,8 @@
 extern "C" {
 #endif
 
-#include "../utils/rescaler.h"
-#include "./decode_vp8.h"
+#include "../utils/rescaler_utils.h"
+#include "./vp8_dec.h"
 
 //------------------------------------------------------------------------------
 // WebPDecParams: Decoding output parameters. Transient internal object.
@@ -38,27 +38,18 @@ struct WebPDecParams {
 
   int last_y;                 // coordinate of the line that was last output
   const WebPDecoderOptions* options;  // if not NULL, use alt decoding features
-  // rescalers
-  WebPRescaler scaler_y, scaler_u, scaler_v, scaler_a;
+
+  WebPRescaler* scaler_y, *scaler_u, *scaler_v, *scaler_a;  // rescalers
   void* memory;                  // overall scratch memory for the output work.
 
   OutputFunc emit;               // output RGB or YUV samples
   OutputAlphaFunc emit_alpha;    // output alpha channel
   OutputRowFunc emit_alpha_row;  // output one line of rescaled alpha values
-
-  WebPDecBuffer* final_output;   // In case the user supplied a slow-memory
-                                 // output, we decode image in temporary buffer
-                                 // (this::output) and copy it here.
-  WebPDecBuffer tmp_buffer;      // this::output will point to this one in case
-                                 // of slow memory.
 };
 
 // Should be called first, before any use of the WebPDecParams object.
 void WebPResetDecParams(WebPDecParams* const params);
 
-// Delete all memory (after an error occurred, for instance)
-void WebPFreeDecParams(WebPDecParams* const params);
-
 //------------------------------------------------------------------------------
 // Header parsing helpers