Update libwebp to 1.0.1

modules/webp/SCsub

@@ -42,7 +42,6 @@ if env['builtin_libwebp']:
         "dsp/dec_neon.c",
         "dsp/dec_sse2.c",
         "dsp/dec_sse41.c",
-        "dsp/enc_avx2.c",
         "dsp/enc.c",
         "dsp/enc_mips32.c",
         "dsp/enc_mips_dsp_r2.c",
@@ -90,7 +89,6 @@ if env['builtin_libwebp']:
         "enc/backward_references_enc.c",
         "enc/config_enc.c",
         "enc/cost_enc.c",
-        "enc/delta_palettization_enc.c",
         "enc/filter_enc.c",
         "enc/frame_enc.c",
         "enc/histogram_enc.c",

thirdparty/README.md

@@ -233,7 +233,7 @@ Godot-made change marked with `// -- GODOT --` comments.
 ## libwebp
 
 - Upstream: https://chromium.googlesource.com/webm/libwebp/
-- Version: 1.0.0
+- Version: 1.0.1
 - License: BSD-3-Clause
 
 Files extracted from upstream source:

thirdparty/libwebp/src/dec/alphai_dec.h

@@ -51,4 +51,4 @@ void WebPDeallocateAlphaMemory(VP8Decoder* const dec);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_ALPHAI_DEC_H_ */
+#endif  // WEBP_DEC_ALPHAI_DEC_H_

thirdparty/libwebp/src/dec/buffer_dec.c

@@ -74,7 +74,8 @@ static VP8StatusCode CheckDecBuffer(const WebPDecBuffer* const buffer) {
   } else {    // RGB checks
     const WebPRGBABuffer* const buf = &buffer->u.RGBA;
     const int stride = abs(buf->stride);
-    const uint64_t size = MIN_BUFFER_SIZE(width, height, stride);
+    const uint64_t size =
+        MIN_BUFFER_SIZE(width * kModeBpp[mode], height, stride);
     ok &= (size <= buf->size);
     ok &= (stride >= width * kModeBpp[mode]);
     ok &= (buf->rgba != NULL);

thirdparty/libwebp/src/dec/common_dec.h

@@ -51,4 +51,4 @@ enum { MB_FEATURE_TREE_PROBS = 3,
        NUM_PROBAS = 11
      };
 
-#endif    // WEBP_DEC_COMMON_DEC_H_
+#endif  // WEBP_DEC_COMMON_DEC_H_

thirdparty/libwebp/src/dec/frame_dec.c

@@ -338,7 +338,6 @@ void VP8InitDithering(const WebPDecoderOptions* const options,
       for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
         VP8QuantMatrix* const dqm = &dec->dqm_[s];
         if (dqm->uv_quant_ < DITHER_AMP_TAB_SIZE) {
-          // TODO(skal): should we specially dither more for uv_quant_ < 0?
           const int idx = (dqm->uv_quant_ < 0) ? 0 : dqm->uv_quant_;
           dqm->dither_ = (f * kQuantToDitherAmp[idx]) >> 3;
         }
@@ -669,15 +668,9 @@ int VP8GetThreadMethod(const WebPDecoderOptions* const options,
   (void)height;
   assert(headers == NULL || !headers->is_lossless);
 #if defined(WEBP_USE_THREAD)
-  if (width < MIN_WIDTH_FOR_THREADS) return 0;
-  // TODO(skal): tune the heuristic further
-#if 0
-  if (height < 2 * width) return 2;
+  if (width >= MIN_WIDTH_FOR_THREADS) return 2;
 #endif
-  return 2;
-#else   // !WEBP_USE_THREAD
   return 0;
-#endif
 }
 
 #undef MT_CACHE_LINES

thirdparty/libwebp/src/dec/idec_dec.c

@@ -140,10 +140,9 @@ static void DoRemap(WebPIDecoder* const idec, ptrdiff_t offset) {
       if (NeedCompressedAlpha(idec)) {
         ALPHDecoder* const alph_dec = dec->alph_dec_;
         dec->alpha_data_ += offset;
-        if (alph_dec != NULL) {
+        if (alph_dec != NULL && alph_dec->vp8l_dec_ != NULL) {
           if (alph_dec->method_ == ALPHA_LOSSLESS_COMPRESSION) {
             VP8LDecoder* const alph_vp8l_dec = alph_dec->vp8l_dec_;
-            assert(alph_vp8l_dec != NULL);
             assert(dec->alpha_data_size_ >= ALPHA_HEADER_LEN);
             VP8LBitReaderSetBuffer(&alph_vp8l_dec->br_,
                                    dec->alpha_data_ + ALPHA_HEADER_LEN,
@@ -283,10 +282,8 @@ static void RestoreContext(const MBContext* context, VP8Decoder* const dec,
 
 static VP8StatusCode IDecError(WebPIDecoder* const idec, VP8StatusCode error) {
   if (idec->state_ == STATE_VP8_DATA) {
-    VP8Io* const io = &idec->io_;
-    if (io->teardown != NULL) {
-      io->teardown(io);
-    }
+    // Synchronize the thread, clean-up and check for errors.
+    VP8ExitCritical((VP8Decoder*)idec->dec_, &idec->io_);
   }
   idec->state_ = STATE_ERROR;
   return error;
@@ -451,7 +448,10 @@ static VP8StatusCode DecodeRemaining(WebPIDecoder* const idec) {
   VP8Decoder* const dec = (VP8Decoder*)idec->dec_;
   VP8Io* const io = &idec->io_;
 
-  assert(dec->ready_);
+  // Make sure partition #0 has been read before, to set dec to ready_.
+  if (!dec->ready_) {
+    return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
+  }
   for (; dec->mb_y_ < dec->mb_h_; ++dec->mb_y_) {
     if (idec->last_mb_y_ != dec->mb_y_) {
       if (!VP8ParseIntraModeRow(&dec->br_, dec)) {
@@ -473,6 +473,12 @@ static VP8StatusCode DecodeRemaining(WebPIDecoder* const idec) {
             MemDataSize(&idec->mem_) > MAX_MB_SIZE) {
           return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
         }
+        // Synchronize the threads.
+        if (dec->mt_method_ > 0) {
+          if (!WebPGetWorkerInterface()->Sync(&dec->worker_)) {
+            return IDecError(idec, VP8_STATUS_BITSTREAM_ERROR);
+          }
+        }
         RestoreContext(&context, dec, token_br);
         return VP8_STATUS_SUSPENDED;
       }
@@ -491,6 +497,7 @@ static VP8StatusCode DecodeRemaining(WebPIDecoder* const idec) {
   }
   // Synchronize the thread and check for errors.
   if (!VP8ExitCritical(dec, io)) {
+    idec->state_ = STATE_ERROR;  // prevent re-entry in IDecError
     return IDecError(idec, VP8_STATUS_USER_ABORT);
   }
   dec->ready_ = 0;
@@ -571,6 +578,10 @@ static VP8StatusCode IDecode(WebPIDecoder* idec) {
     status = DecodePartition0(idec);
   }
   if (idec->state_ == STATE_VP8_DATA) {
+    const VP8Decoder* const dec = (VP8Decoder*)idec->dec_;
+    if (dec == NULL) {
+      return VP8_STATUS_SUSPENDED;  // can't continue if we have no decoder.
+    }
     status = DecodeRemaining(idec);
   }
   if (idec->state_ == STATE_VP8L_HEADER) {

thirdparty/libwebp/src/dec/vp8_dec.h

@@ -182,4 +182,4 @@ WEBP_EXTERN int VP8LGetInfo(
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_VP8_DEC_H_ */
+#endif  // WEBP_DEC_VP8_DEC_H_

thirdparty/libwebp/src/dec/vp8i_dec.h

@@ -32,7 +32,7 @@ extern "C" {
 // version numbers
 #define DEC_MAJ_VERSION 1
 #define DEC_MIN_VERSION 0
-#define DEC_REV_VERSION 0
+#define DEC_REV_VERSION 1
 
 // YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
 // Constraints are: We need to store one 16x16 block of luma samples (y),
@@ -316,4 +316,4 @@ const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_VP8I_DEC_H_ */
+#endif  // WEBP_DEC_VP8I_DEC_H_

thirdparty/libwebp/src/dec/vp8l_dec.c

@@ -362,12 +362,19 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
   VP8LMetadata* const hdr = &dec->hdr_;
   uint32_t* huffman_image = NULL;
   HTreeGroup* htree_groups = NULL;
+  // When reading htrees, some might be unused, as the format allows it.
+  // We will still read them but put them in this htree_group_bogus.
+  HTreeGroup htree_group_bogus;
   HuffmanCode* huffman_tables = NULL;
+  HuffmanCode* huffman_tables_bogus = NULL;
   HuffmanCode* next = NULL;
   int num_htree_groups = 1;
+  int num_htree_groups_max = 1;
   int max_alphabet_size = 0;
   int* code_lengths = NULL;
   const int table_size = kTableSize[color_cache_bits];
+  int* mapping = NULL;
+  int ok = 0;
 
   if (allow_recursion && VP8LReadBits(br, 1)) {
     // use meta Huffman codes.
@@ -384,9 +391,41 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
       // The huffman data is stored in red and green bytes.
       const int group = (huffman_image[i] >> 8) & 0xffff;
       huffman_image[i] = group;
-      if (group >= num_htree_groups) {
-        num_htree_groups = group + 1;
+      if (group >= num_htree_groups_max) {
+        num_htree_groups_max = group + 1;
+      }
+    }
+    // Check the validity of num_htree_groups_max. If it seems too big, use a
+    // smaller value for later. This will prevent big memory allocations to end
+    // up with a bad bitstream anyway.
+    // The value of 1000 is totally arbitrary. We know that num_htree_groups_max
+    // is smaller than (1 << 16) and should be smaller than the number of pixels
+    // (though the format allows it to be bigger).
+    if (num_htree_groups_max > 1000 || num_htree_groups_max > xsize * ysize) {
+      // Create a mapping from the used indices to the minimal set of used
+      // values [0, num_htree_groups)
+      mapping = (int*)WebPSafeMalloc(num_htree_groups_max, sizeof(*mapping));
+      if (mapping == NULL) {
+        dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
+        goto Error;
+      }
+      // -1 means a value is unmapped, and therefore unused in the Huffman
+      // image.
+      memset(mapping, 0xff, num_htree_groups_max * sizeof(*mapping));
+      for (num_htree_groups = 0, i = 0; i < huffman_pixs; ++i) {
+        // Get the current mapping for the group and remap the Huffman image.
+        int* const mapped_group = &mapping[huffman_image[i]];
+        if (*mapped_group == -1) *mapped_group = num_htree_groups++;
+        huffman_image[i] = *mapped_group;
+      }
+      huffman_tables_bogus = (HuffmanCode*)WebPSafeMalloc(
+          table_size, sizeof(*huffman_tables_bogus));
+      if (huffman_tables_bogus == NULL) {
+        dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
+        goto Error;
       }
+    } else {
+      num_htree_groups = num_htree_groups_max;
     }
   }
 
@@ -403,11 +442,11 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
     }
   }
 
+  code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size,
+                                      sizeof(*code_lengths));
   huffman_tables = (HuffmanCode*)WebPSafeMalloc(num_htree_groups * table_size,
                                                 sizeof(*huffman_tables));
   htree_groups = VP8LHtreeGroupsNew(num_htree_groups);
-  code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size,
-                                      sizeof(*code_lengths));
 
   if (htree_groups == NULL || code_lengths == NULL || huffman_tables == NULL) {
     dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
@@ -415,28 +454,35 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
   }
 
   next = huffman_tables;
-  for (i = 0; i < num_htree_groups; ++i) {
-    HTreeGroup* const htree_group = &htree_groups[i];
+  for (i = 0; i < num_htree_groups_max; ++i) {
+    // If the index "i" is unused in the Huffman image, read the coefficients
+    // but store them to a bogus htree_group.
+    const int is_bogus = (mapping != NULL && mapping[i] == -1);
+    HTreeGroup* const htree_group =
+        is_bogus ? &htree_group_bogus :
+        &htree_groups[(mapping == NULL) ? i : mapping[i]];
     HuffmanCode** const htrees = htree_group->htrees;
+    HuffmanCode* huffman_tables_i = is_bogus ? huffman_tables_bogus : next;
     int size;
     int total_size = 0;
     int is_trivial_literal = 1;
     int max_bits = 0;
     for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
       int alphabet_size = kAlphabetSize[j];
-      htrees[j] = next;
+      htrees[j] = huffman_tables_i;
       if (j == 0 && color_cache_bits > 0) {
         alphabet_size += 1 << color_cache_bits;
       }
-      size = ReadHuffmanCode(alphabet_size, dec, code_lengths, next);
+      size =
+          ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_tables_i);
       if (size == 0) {
         goto Error;
       }
       if (is_trivial_literal && kLiteralMap[j] == 1) {
-        is_trivial_literal = (next->bits == 0);
+        is_trivial_literal = (huffman_tables_i->bits == 0);
       }
-      total_size += next->bits;
-      next += size;
+      total_size += huffman_tables_i->bits;
+      huffman_tables_i += size;
       if (j <= ALPHA) {
         int local_max_bits = code_lengths[0];
         int k;
@@ -448,38 +494,41 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
         max_bits += local_max_bits;
       }
     }
+    if (!is_bogus) next = huffman_tables_i;
     htree_group->is_trivial_literal = is_trivial_literal;
     htree_group->is_trivial_code = 0;
     if (is_trivial_literal) {
       const int red = htrees[RED][0].value;
       const int blue = htrees[BLUE][0].value;
       const int alpha = htrees[ALPHA][0].value;
-      htree_group->literal_arb =
-          ((uint32_t)alpha << 24) | (red << 16) | blue;
+      htree_group->literal_arb = ((uint32_t)alpha << 24) | (red << 16) | blue;
       if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) {
         htree_group->is_trivial_code = 1;
         htree_group->literal_arb |= htrees[GREEN][0].value << 8;
       }
     }
-    htree_group->use_packed_table = !htree_group->is_trivial_code &&
-                                    (max_bits < HUFFMAN_PACKED_BITS);
+    htree_group->use_packed_table =
+        !htree_group->is_trivial_code && (max_bits < HUFFMAN_PACKED_BITS);
     if (htree_group->use_packed_table) BuildPackedTable(htree_group);
   }
-  WebPSafeFree(code_lengths);
+  ok = 1;
 
-  // All OK. Finalize pointers and return.
+  // All OK. Finalize pointers.
   hdr->huffman_image_ = huffman_image;
   hdr->num_htree_groups_ = num_htree_groups;
   hdr->htree_groups_ = htree_groups;
   hdr->huffman_tables_ = huffman_tables;
-  return 1;
 
  Error:
   WebPSafeFree(code_lengths);
-  WebPSafeFree(huffman_image);
-  WebPSafeFree(huffman_tables);
-  VP8LHtreeGroupsFree(htree_groups);
-  return 0;
+  WebPSafeFree(huffman_tables_bogus);
+  WebPSafeFree(mapping);
+  if (!ok) {
+    WebPSafeFree(huffman_image);
+    WebPSafeFree(huffman_tables);
+    VP8LHtreeGroupsFree(htree_groups);
+  }
+  return ok;
 }
 
 //------------------------------------------------------------------------------
@@ -884,7 +933,11 @@ static WEBP_INLINE void CopyBlock8b(uint8_t* const dst, int dist, int length) {
 #endif
         break;
       case 2:
+#if !defined(WORDS_BIGENDIAN)
         memcpy(&pattern, src, sizeof(uint16_t));
+#else
+        pattern = ((uint32_t)src[0] << 8) | src[1];
+#endif
 #if defined(__arm__) || defined(_M_ARM)
         pattern |= pattern << 16;
 #elif defined(WEBP_USE_MIPS_DSP_R2)
@@ -1523,7 +1576,6 @@ int VP8LDecodeAlphaHeader(ALPHDecoder* const alph_dec,
   if (dec == NULL) return 0;
 
   assert(alph_dec != NULL);
-  alph_dec->vp8l_dec_ = dec;
 
   dec->width_ = alph_dec->width_;
   dec->height_ = alph_dec->height_;
@@ -1555,11 +1607,12 @@ int VP8LDecodeAlphaHeader(ALPHDecoder* const alph_dec,
 
   if (!ok) goto Err;
 
+  // Only set here, once we are sure it is valid (to avoid thread races).
+  alph_dec->vp8l_dec_ = dec;
   return 1;
 
  Err:
-  VP8LDelete(alph_dec->vp8l_dec_);
-  alph_dec->vp8l_dec_ = NULL;
+  VP8LDelete(dec);
   return 0;
 }
 

thirdparty/libwebp/src/dec/vp8li_dec.h

@@ -132,4 +132,4 @@ void VP8LDelete(VP8LDecoder* const dec);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_VP8LI_DEC_H_ */
+#endif  // WEBP_DEC_VP8LI_DEC_H_

thirdparty/libwebp/src/dec/webpi_dec.h

@@ -130,4 +130,4 @@ int WebPAvoidSlowMemory(const WebPDecBuffer* const output,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DEC_WEBPI_DEC_H_ */
+#endif  // WEBP_DEC_WEBPI_DEC_H_

thirdparty/libwebp/src/demux/demux.c

@@ -25,7 +25,7 @@
 
 #define DMUX_MAJ_VERSION 1
 #define DMUX_MIN_VERSION 0
-#define DMUX_REV_VERSION 0
+#define DMUX_REV_VERSION 1
 
 typedef struct {
   size_t start_;        // start location of the data

thirdparty/libwebp/src/dsp/dsp.h

@@ -76,10 +76,6 @@ extern "C" {
 #define WEBP_USE_SSE41
 #endif
 
-#if defined(__AVX2__) || defined(WEBP_HAVE_AVX2)
-#define WEBP_USE_AVX2
-#endif
-
 // 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__) || \
@@ -679,4 +675,4 @@ void VP8FiltersInit(void);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DSP_DSP_H_ */
+#endif  // WEBP_DSP_DSP_H_

thirdparty/libwebp/src/dsp/enc.c

@@ -734,7 +734,6 @@ VP8BlockCopy VP8Copy16x8;
 
 extern void VP8EncDspInitSSE2(void);
 extern void VP8EncDspInitSSE41(void);
-extern void VP8EncDspInitAVX2(void);
 extern void VP8EncDspInitNEON(void);
 extern void VP8EncDspInitMIPS32(void);
 extern void VP8EncDspInitMIPSdspR2(void);
@@ -784,11 +783,6 @@ WEBP_DSP_INIT_FUNC(VP8EncDspInit) {
 #endif
     }
 #endif
-#if defined(WEBP_USE_AVX2)
-    if (VP8GetCPUInfo(kAVX2)) {
-      VP8EncDspInitAVX2();
-    }
-#endif
 #if defined(WEBP_USE_MIPS32)
     if (VP8GetCPUInfo(kMIPS32)) {
       VP8EncDspInitMIPS32();

thirdparty/libwebp/src/dsp/enc_avx2.c

@@ -1,21 +0,0 @@
-// Copyright 2014 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.
-// -----------------------------------------------------------------------------
-//
-// AVX2 version of speed-critical encoding functions.
-
-#include "src/dsp/dsp.h"
-
-#if defined(WEBP_USE_AVX2)
-
-#endif  // WEBP_USE_AVX2
-
-//------------------------------------------------------------------------------
-// Entry point
-
-WEBP_DSP_INIT_STUB(VP8EncDspInitAVX2)

thirdparty/libwebp/src/dsp/lossless.c

@@ -23,8 +23,6 @@
 #include "src/dsp/lossless.h"
 #include "src/dsp/lossless_common.h"
 
-#define MAX_DIFF_COST (1e30f)
-
 //------------------------------------------------------------------------------
 // Image transforms.
 

thirdparty/libwebp/src/dsp/lossless.h

@@ -163,7 +163,7 @@ extern VP8LCostCombinedFunc VP8LExtraCostCombined;
 extern VP8LCombinedShannonEntropyFunc VP8LCombinedShannonEntropy;
 
 typedef struct {        // small struct to hold counters
-  int counts[2];        // index: 0=zero steak, 1=non-zero streak
+  int counts[2];        // index: 0=zero streak, 1=non-zero streak
   int streaks[2][2];    // [zero/non-zero][streak<3 / streak>=3]
 } VP8LStreaks;
 
@@ -194,10 +194,14 @@ extern VP8LGetEntropyUnrefinedFunc VP8LGetEntropyUnrefined;
 void VP8LBitsEntropyUnrefined(const uint32_t* const array, int n,
                               VP8LBitEntropy* const entropy);
 
-typedef void (*VP8LHistogramAddFunc)(const VP8LHistogram* const a,
-                                     const VP8LHistogram* const b,
-                                     VP8LHistogram* const out);
-extern VP8LHistogramAddFunc VP8LHistogramAdd;
+typedef void (*VP8LAddVectorFunc)(const uint32_t* a, const uint32_t* b,
+                                  uint32_t* out, int size);
+extern VP8LAddVectorFunc VP8LAddVector;
+typedef void (*VP8LAddVectorEqFunc)(const uint32_t* a, uint32_t* out, int size);
+extern VP8LAddVectorEqFunc VP8LAddVectorEq;
+void VP8LHistogramAdd(const VP8LHistogram* const a,
+                      const VP8LHistogram* const b,
+                      VP8LHistogram* const out);
 
 // -----------------------------------------------------------------------------
 // PrefixEncode()

thirdparty/libwebp/src/dsp/lossless_enc.c

@@ -632,38 +632,67 @@ static double ExtraCostCombined_C(const uint32_t* X, const uint32_t* Y,
 
 //------------------------------------------------------------------------------
 
-static void HistogramAdd_C(const VP8LHistogram* const a,
-                           const VP8LHistogram* const b,
-                           VP8LHistogram* const out) {
+static void AddVector_C(const uint32_t* a, const uint32_t* b, uint32_t* out,
+                        int size) {
+  int i;
+  for (i = 0; i < size; ++i) out[i] = a[i] + b[i];
+}
+
+static void AddVectorEq_C(const uint32_t* a, uint32_t* out, int size) {
+  int i;
+  for (i = 0; i < size; ++i) out[i] += a[i];
+}
+
+#define ADD(X, ARG, LEN) do {                                                  \
+  if (a->is_used_[X]) {                                                        \
+    if (b->is_used_[X]) {                                                      \
+      VP8LAddVector(a->ARG, b->ARG, out->ARG, (LEN));                          \
+    } else {                                                                   \
+      memcpy(&out->ARG[0], &a->ARG[0], (LEN) * sizeof(out->ARG[0]));           \
+    }                                                                          \
+  } else if (b->is_used_[X]) {                                                 \
+    memcpy(&out->ARG[0], &b->ARG[0], (LEN) * sizeof(out->ARG[0]));             \
+  } else {                                                                     \
+    memset(&out->ARG[0], 0, (LEN) * sizeof(out->ARG[0]));                      \
+  }                                                                            \
+} while (0)
+
+#define ADD_EQ(X, ARG, LEN) do {                                               \
+  if (a->is_used_[X]) {                                                        \
+    if (out->is_used_[X]) {                                                    \
+      VP8LAddVectorEq(a->ARG, out->ARG, (LEN));                                \
+    } else {                                                                   \
+      memcpy(&out->ARG[0], &a->ARG[0], (LEN) * sizeof(out->ARG[0]));           \
+    }                                                                          \
+  }                                                                            \
+} while (0)
+
+void VP8LHistogramAdd(const VP8LHistogram* const a,
+                      const VP8LHistogram* const b, VP8LHistogram* const out) {
   int i;
   const int literal_size = VP8LHistogramNumCodes(a->palette_code_bits_);
   assert(a->palette_code_bits_ == b->palette_code_bits_);
+
   if (b != out) {
-    for (i = 0; i < literal_size; ++i) {
-      out->literal_[i] = a->literal_[i] + b->literal_[i];
-    }
-    for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
-      out->distance_[i] = a->distance_[i] + b->distance_[i];
-    }
-    for (i = 0; i < NUM_LITERAL_CODES; ++i) {
-      out->red_[i] = a->red_[i] + b->red_[i];
-      out->blue_[i] = a->blue_[i] + b->blue_[i];
-      out->alpha_[i] = a->alpha_[i] + b->alpha_[i];
+    ADD(0, literal_, literal_size);
+    ADD(1, red_, NUM_LITERAL_CODES);
+    ADD(2, blue_, NUM_LITERAL_CODES);
+    ADD(3, alpha_, NUM_LITERAL_CODES);
+    ADD(4, distance_, NUM_DISTANCE_CODES);
+    for (i = 0; i < 5; ++i) {
+      out->is_used_[i] = (a->is_used_[i] | b->is_used_[i]);
     }
   } else {
-    for (i = 0; i < literal_size; ++i) {
-      out->literal_[i] += a->literal_[i];
-    }
-    for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
-      out->distance_[i] += a->distance_[i];
-    }
-    for (i = 0; i < NUM_LITERAL_CODES; ++i) {
-      out->red_[i] += a->red_[i];
-      out->blue_[i] += a->blue_[i];
-      out->alpha_[i] += a->alpha_[i];
-    }
+    ADD_EQ(0, literal_, literal_size);
+    ADD_EQ(1, red_, NUM_LITERAL_CODES);
+    ADD_EQ(2, blue_, NUM_LITERAL_CODES);
+    ADD_EQ(3, alpha_, NUM_LITERAL_CODES);
+    ADD_EQ(4, distance_, NUM_DISTANCE_CODES);
+    for (i = 0; i < 5; ++i) out->is_used_[i] |= a->is_used_[i];
   }
 }
+#undef ADD
+#undef ADD_EQ
 
 //------------------------------------------------------------------------------
 // Image transforms.
@@ -848,7 +877,8 @@ VP8LCombinedShannonEntropyFunc VP8LCombinedShannonEntropy;
 VP8LGetEntropyUnrefinedFunc VP8LGetEntropyUnrefined;
 VP8LGetCombinedEntropyUnrefinedFunc VP8LGetCombinedEntropyUnrefined;
 
-VP8LHistogramAddFunc VP8LHistogramAdd;
+VP8LAddVectorFunc VP8LAddVector;
+VP8LAddVectorEqFunc VP8LAddVectorEq;
 
 VP8LVectorMismatchFunc VP8LVectorMismatch;
 VP8LBundleColorMapFunc VP8LBundleColorMap;
@@ -885,7 +915,8 @@ WEBP_DSP_INIT_FUNC(VP8LEncDspInit) {
   VP8LGetEntropyUnrefined = GetEntropyUnrefined_C;
   VP8LGetCombinedEntropyUnrefined = GetCombinedEntropyUnrefined_C;
 
-  VP8LHistogramAdd = HistogramAdd_C;
+  VP8LAddVector = AddVector_C;
+  VP8LAddVectorEq = AddVectorEq_C;
 
   VP8LVectorMismatch = VectorMismatch_C;
   VP8LBundleColorMap = VP8LBundleColorMap_C;
@@ -971,7 +1002,8 @@ WEBP_DSP_INIT_FUNC(VP8LEncDspInit) {
   assert(VP8LCombinedShannonEntropy != NULL);
   assert(VP8LGetEntropyUnrefined != NULL);
   assert(VP8LGetCombinedEntropyUnrefined != NULL);
-  assert(VP8LHistogramAdd != NULL);
+  assert(VP8LAddVector != NULL);
+  assert(VP8LAddVectorEq != NULL);
   assert(VP8LVectorMismatch != NULL);
   assert(VP8LBundleColorMap != NULL);
   assert(VP8LPredictorsSub[0] != NULL);

thirdparty/libwebp/src/dsp/lossless_enc_mips32.c

@@ -344,65 +344,29 @@ static void GetCombinedEntropyUnrefined_MIPS32(const uint32_t X[],
     ASM_END_COMMON_0                                    \
     ASM_END_COMMON_1
 
-#define ADD_VECTOR(A, B, OUT, SIZE, EXTRA_SIZE)  do {   \
-  const uint32_t* pa = (const uint32_t*)(A);            \
-  const uint32_t* pb = (const uint32_t*)(B);            \
-  uint32_t* pout = (uint32_t*)(OUT);                    \
-  const uint32_t* const LoopEnd = pa + (SIZE);          \
-  assert((SIZE) % 4 == 0);                              \
-  ASM_START                                             \
-  ADD_TO_OUT(0, 4, 8, 12, 1, pa, pb, pout)              \
-  ASM_END_0                                             \
-  if ((EXTRA_SIZE) > 0) {                               \
-    const int last = (EXTRA_SIZE);                      \
-    int i;                                              \
-    for (i = 0; i < last; ++i) pout[i] = pa[i] + pb[i]; \
-  }                                                     \
-} while (0)
-
-#define ADD_VECTOR_EQ(A, OUT, SIZE, EXTRA_SIZE)  do {   \
-  const uint32_t* pa = (const uint32_t*)(A);            \
-  uint32_t* pout = (uint32_t*)(OUT);                    \
-  const uint32_t* const LoopEnd = pa + (SIZE);          \
-  assert((SIZE) % 4 == 0);                              \
-  ASM_START                                             \
-  ADD_TO_OUT(0, 4, 8, 12, 0, pa, pout, pout)            \
-  ASM_END_1                                             \
-  if ((EXTRA_SIZE) > 0) {                               \
-    const int last = (EXTRA_SIZE);                      \
-    int i;                                              \
-    for (i = 0; i < last; ++i) pout[i] += pa[i];        \
-  }                                                     \
-} while (0)
-
-static void HistogramAdd_MIPS32(const VP8LHistogram* const a,
-                                const VP8LHistogram* const b,
-                                VP8LHistogram* const out) {
+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 int extra_cache_size = VP8LHistogramNumCodes(a->palette_code_bits_)
-                             - (NUM_LITERAL_CODES + NUM_LENGTH_CODES);
-  assert(a->palette_code_bits_ == b->palette_code_bits_);
-
-  if (b != out) {
-    ADD_VECTOR(a->literal_, b->literal_, out->literal_,
-               NUM_LITERAL_CODES + NUM_LENGTH_CODES, extra_cache_size);
-    ADD_VECTOR(a->distance_, b->distance_, out->distance_,
-               NUM_DISTANCE_CODES, 0);
-    ADD_VECTOR(a->red_, b->red_, out->red_, NUM_LITERAL_CODES, 0);
-    ADD_VECTOR(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES, 0);
-    ADD_VECTOR(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES, 0);
-  } else {
-    ADD_VECTOR_EQ(a->literal_, out->literal_,
-                  NUM_LITERAL_CODES + NUM_LENGTH_CODES, extra_cache_size);
-    ADD_VECTOR_EQ(a->distance_, out->distance_, NUM_DISTANCE_CODES, 0);
-    ADD_VECTOR_EQ(a->red_, out->red_, NUM_LITERAL_CODES, 0);
-    ADD_VECTOR_EQ(a->blue_, out->blue_, NUM_LITERAL_CODES, 0);
-    ADD_VECTOR_EQ(a->alpha_, out->alpha_, NUM_LITERAL_CODES, 0);
-  }
+  const uint32_t 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];
+}
+
+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 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];
 }
 
-#undef ADD_VECTOR_EQ
-#undef ADD_VECTOR
 #undef ASM_END_1
 #undef ASM_END_0
 #undef ASM_END_COMMON_1
@@ -422,7 +386,8 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitMIPS32(void) {
   VP8LExtraCostCombined = ExtraCostCombined_MIPS32;
   VP8LGetEntropyUnrefined = GetEntropyUnrefined_MIPS32;
   VP8LGetCombinedEntropyUnrefined = GetCombinedEntropyUnrefined_MIPS32;
-  VP8LHistogramAdd = HistogramAdd_MIPS32;
+  VP8LAddVector = AddVector_MIPS32;
+  VP8LAddVectorEq = AddVectorEq_MIPS32;
 }
 
 #else  // !WEBP_USE_MIPS32

thirdparty/libwebp/src/dsp/lossless_enc_sse2.c

@@ -170,12 +170,13 @@ static void CollectColorRedTransforms_SSE2(const uint32_t* argb, int stride,
 
 //------------------------------------------------------------------------------
 
+// Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
+// that's ok since the histogram values are less than 1<<28 (max picture size).
 #define LINE_SIZE 16    // 8 or 16
 static void AddVector_SSE2(const uint32_t* a, const uint32_t* b, uint32_t* out,
                            int size) {
   int i;
-  assert(size % LINE_SIZE == 0);
-  for (i = 0; i < size; i += LINE_SIZE) {
+  for (i = 0; i + LINE_SIZE <= size; i += LINE_SIZE) {
     const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i +  0]);
     const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i +  4]);
 #if (LINE_SIZE == 16)
@@ -195,12 +196,14 @@ static void AddVector_SSE2(const uint32_t* a, const uint32_t* b, uint32_t* out,
     _mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
 #endif
   }
+  for (; i < size; ++i) {
+    out[i] = a[i] + b[i];
+  }
 }
 
 static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
   int i;
-  assert(size % LINE_SIZE == 0);
-  for (i = 0; i < size; i += LINE_SIZE) {
+  for (i = 0; i + LINE_SIZE <= size; i += LINE_SIZE) {
     const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i +  0]);
     const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i +  4]);
 #if (LINE_SIZE == 16)
@@ -220,35 +223,11 @@ static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
     _mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
 #endif
   }
-}
-#undef LINE_SIZE
-
-// Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
-// that's ok since the histogram values are less than 1<<28 (max picture size).
-static void HistogramAdd_SSE2(const VP8LHistogram* const a,
-                              const VP8LHistogram* const b,
-                              VP8LHistogram* const out) {
-  int i;
-  const int literal_size = VP8LHistogramNumCodes(a->palette_code_bits_);
-  assert(a->palette_code_bits_ == b->palette_code_bits_);
-  if (b != out) {
-    AddVector_SSE2(a->literal_, b->literal_, out->literal_, NUM_LITERAL_CODES);
-    AddVector_SSE2(a->red_, b->red_, out->red_, NUM_LITERAL_CODES);
-    AddVector_SSE2(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES);
-    AddVector_SSE2(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES);
-  } else {
-    AddVectorEq_SSE2(a->literal_, out->literal_, NUM_LITERAL_CODES);
-    AddVectorEq_SSE2(a->red_, out->red_, NUM_LITERAL_CODES);
-    AddVectorEq_SSE2(a->blue_, out->blue_, NUM_LITERAL_CODES);
-    AddVectorEq_SSE2(a->alpha_, out->alpha_, NUM_LITERAL_CODES);
-  }
-  for (i = NUM_LITERAL_CODES; i < literal_size; ++i) {
-    out->literal_[i] = a->literal_[i] + b->literal_[i];
-  }
-  for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
-    out->distance_[i] = a->distance_[i] + b->distance_[i];
+  for (; i < size; ++i) {
+    out[i] += a[i];
   }
 }
+#undef LINE_SIZE
 
 //------------------------------------------------------------------------------
 // Entropy
@@ -675,7 +654,8 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE2(void) {
   VP8LTransformColor = TransformColor_SSE2;
   VP8LCollectColorBlueTransforms = CollectColorBlueTransforms_SSE2;
   VP8LCollectColorRedTransforms = CollectColorRedTransforms_SSE2;
-  VP8LHistogramAdd = HistogramAdd_SSE2;
+  VP8LAddVector = AddVector_SSE2;
+  VP8LAddVectorEq = AddVectorEq_SSE2;
   VP8LCombinedShannonEntropy = CombinedShannonEntropy_SSE2;
   VP8LVectorMismatch = VectorMismatch_SSE2;
   VP8LBundleColorMap = BundleColorMap_SSE2;

thirdparty/libwebp/src/dsp/msa_macro.h

@@ -1389,4 +1389,4 @@ static WEBP_INLINE uint32_t func_hadd_uh_u32(v8u16 in) {
 } while (0)
 #define AVER_UB2_UB(...) AVER_UB2(v16u8, __VA_ARGS__)
 
-#endif  /* WEBP_DSP_MSA_MACRO_H_ */
+#endif  // WEBP_DSP_MSA_MACRO_H_

thirdparty/libwebp/src/dsp/rescaler.c

@@ -21,6 +21,7 @@
 
 #define ROUNDER (WEBP_RESCALER_ONE >> 1)
 #define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
 
 //------------------------------------------------------------------------------
 // Row import
@@ -138,7 +139,7 @@ void WebPRescalerExportRowShrink_C(WebPRescaler* const wrk) {
   if (yscale) {
     for (x_out = 0; x_out < x_out_max; ++x_out) {
       const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
-      const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+      const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
       assert(v >= 0 && v <= 255);
       dst[x_out] = v;
       irow[x_out] = frac;   // new fractional start
@@ -153,6 +154,7 @@ void WebPRescalerExportRowShrink_C(WebPRescaler* const wrk) {
   }
 }
 
+#undef MULT_FIX_FLOOR
 #undef MULT_FIX
 #undef ROUNDER
 

thirdparty/libwebp/src/dsp/rescaler_mips32.c

@@ -209,6 +209,7 @@ static void ExportRowExpand_MIPS32(WebPRescaler* const wrk) {
   }
 }
 
+#if 0  // disabled for now. TODO(skal): make match the C-code
 static void ExportRowShrink_MIPS32(WebPRescaler* const wrk) {
   const int x_out_max = wrk->dst_width * wrk->num_channels;
   uint8_t* dst = wrk->dst;
@@ -273,6 +274,7 @@ static void ExportRowShrink_MIPS32(WebPRescaler* const wrk) {
     );
   }
 }
+#endif  // 0
 
 //------------------------------------------------------------------------------
 // Entry point
@@ -283,7 +285,7 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPS32(void) {
   WebPRescalerImportRowExpand = ImportRowExpand_MIPS32;
   WebPRescalerImportRowShrink = ImportRowShrink_MIPS32;
   WebPRescalerExportRowExpand = ExportRowExpand_MIPS32;
-  WebPRescalerExportRowShrink = ExportRowShrink_MIPS32;
+//  WebPRescalerExportRowShrink = ExportRowShrink_MIPS32;
 }
 
 #else  // !WEBP_USE_MIPS32

thirdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c

@@ -20,10 +20,12 @@
 
 #define ROUNDER (WEBP_RESCALER_ONE >> 1)
 #define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
 
 //------------------------------------------------------------------------------
 // Row export
 
+#if 0  // disabled for now. TODO(skal): make match the C-code
 static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
   int i;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
@@ -106,7 +108,7 @@ static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
     }
     for (i = 0; i < (x_out_max & 0x3); ++i) {
       const uint32_t frac = (uint32_t)MULT_FIX(*frow++, yscale);
-      const int v = (int)MULT_FIX(*irow - frac, wrk->fxy_scale);
+      const int v = (int)MULT_FIX_FLOOR(*irow - frac, wrk->fxy_scale);
       assert(v >= 0 && v <= 255);
       *dst++ = v;
       *irow++ = frac;   // new fractional start
@@ -154,13 +156,14 @@ static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
       );
     }
     for (i = 0; i < (x_out_max & 0x3); ++i) {
-      const int v = (int)MULT_FIX(*irow, wrk->fxy_scale);
+      const int v = (int)MULT_FIX_FLOOR(*irow, wrk->fxy_scale);
       assert(v >= 0 && v <= 255);
       *dst++ = v;
       *irow++ = 0;
     }
   }
 }
+#endif  // 0
 
 static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
   int i;
@@ -294,6 +297,7 @@ static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
   }
 }
 
+#undef MULT_FIX_FLOOR
 #undef MULT_FIX
 #undef ROUNDER
 
@@ -304,7 +308,7 @@ extern void WebPRescalerDspInitMIPSdspR2(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPSdspR2(void) {
   WebPRescalerExportRowExpand = ExportRowExpand_MIPSdspR2;
-  WebPRescalerExportRowShrink = ExportRowShrink_MIPSdspR2;
+//  WebPRescalerExportRowShrink = ExportRowShrink_MIPSdspR2;
 }
 
 #else  // !WEBP_USE_MIPS_DSP_R2

thirdparty/libwebp/src/dsp/rescaler_msa.c

@@ -22,6 +22,7 @@
 
 #define ROUNDER (WEBP_RESCALER_ONE >> 1)
 #define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
 
 #define CALC_MULT_FIX_16(in0, in1, in2, in3, scale, shift, dst) do {  \
   v4u32 tmp0, tmp1, tmp2, tmp3;                                       \
@@ -262,6 +263,7 @@ static void RescalerExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
   }
 }
 
+#if 0  // disabled for now. TODO(skal): make match the C-code
 static WEBP_INLINE void ExportRowShrink_0(const uint32_t* frow, uint32_t* irow,
                                           uint8_t* dst, int length,
                                           const uint32_t yscale,
@@ -341,7 +343,7 @@ static WEBP_INLINE void ExportRowShrink_0(const uint32_t* frow, uint32_t* irow,
     }
     for (x_out = 0; x_out < length; ++x_out) {
       const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
-      const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+      const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
       assert(v >= 0 && v <= 255);
       dst[x_out] = v;
       irow[x_out] = frac;
@@ -426,6 +428,7 @@ static void RescalerExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
     ExportRowShrink_1(irow, dst, x_out_max, wrk);
   }
 }
+#endif  // 0
 
 //------------------------------------------------------------------------------
 // Entry point
@@ -434,7 +437,7 @@ extern void WebPRescalerDspInitMSA(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMSA(void) {
   WebPRescalerExportRowExpand = RescalerExportRowExpand_MIPSdspR2;
-  WebPRescalerExportRowShrink = RescalerExportRowShrink_MIPSdspR2;
+//  WebPRescalerExportRowShrink = RescalerExportRowShrink_MIPSdspR2;
 }
 
 #else     // !WEBP_USE_MSA

thirdparty/libwebp/src/dsp/rescaler_neon.c

@@ -22,6 +22,7 @@
 
 #define ROUNDER (WEBP_RESCALER_ONE >> 1)
 #define MULT_FIX_C(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR_C(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
 
 #define LOAD_32x4(SRC, DST) const uint32x4_t DST = vld1q_u32((SRC))
 #define LOAD_32x8(SRC, DST0, DST1)                                    \
@@ -35,8 +36,11 @@
 
 #if (WEBP_RESCALER_RFIX == 32)
 #define MAKE_HALF_CST(C) vdupq_n_s32((int32_t)((C) >> 1))
-#define MULT_FIX(A, B) /* note: B is actualy scale>>1. See MAKE_HALF_CST */ \
+// note: B is actualy scale>>1. See MAKE_HALF_CST
+#define MULT_FIX(A, B) \
     vreinterpretq_u32_s32(vqrdmulhq_s32(vreinterpretq_s32_u32((A)), (B)))
+#define MULT_FIX_FLOOR(A, B) \
+    vreinterpretq_u32_s32(vqdmulhq_s32(vreinterpretq_s32_u32((A)), (B)))
 #else
 #error "MULT_FIX/WEBP_RESCALER_RFIX need some more work"
 #endif
@@ -135,8 +139,8 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
       const uint32x4_t A1 = MULT_FIX(in1, yscale_half);
       const uint32x4_t B0 = vqsubq_u32(in2, A0);
       const uint32x4_t B1 = vqsubq_u32(in3, A1);
-      const uint32x4_t C0 = MULT_FIX(B0, fxy_scale_half);
-      const uint32x4_t C1 = MULT_FIX(B1, fxy_scale_half);
+      const uint32x4_t C0 = MULT_FIX_FLOOR(B0, fxy_scale_half);
+      const uint32x4_t C1 = MULT_FIX_FLOOR(B1, fxy_scale_half);
       const uint16x4_t D0 = vmovn_u32(C0);
       const uint16x4_t D1 = vmovn_u32(C1);
       const uint8x8_t E = vmovn_u16(vcombine_u16(D0, D1));
@@ -145,7 +149,7 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
     }
     for (; x_out < x_out_max; ++x_out) {
       const uint32_t frac = (uint32_t)MULT_FIX_C(frow[x_out], yscale);
-      const int v = (int)MULT_FIX_C(irow[x_out] - frac, wrk->fxy_scale);
+      const int v = (int)MULT_FIX_FLOOR_C(irow[x_out] - frac, fxy_scale);
       assert(v >= 0 && v <= 255);
       dst[x_out] = v;
       irow[x_out] = frac;   // new fractional start
@@ -170,6 +174,12 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
   }
 }
 
+#undef MULT_FIX_FLOOR_C
+#undef MULT_FIX_C
+#undef MULT_FIX_FLOOR
+#undef MULT_FIX
+#undef ROUNDER
+
 //------------------------------------------------------------------------------
 
 extern void WebPRescalerDspInitNEON(void);

thirdparty/libwebp/src/dsp/rescaler_sse2.c

@@ -25,6 +25,7 @@
 
 #define ROUNDER (WEBP_RESCALER_ONE >> 1)
 #define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
 
 // input: 8 bytes ABCDEFGH -> output: A0E0B0F0C0G0D0H0
 static void LoadTwoPixels_SSE2(const uint8_t* const src, __m128i* out) {
@@ -224,6 +225,35 @@ static WEBP_INLINE void ProcessRow_SSE2(const __m128i* const A0,
   _mm_storel_epi64((__m128i*)dst, G);
 }
 
+static WEBP_INLINE void ProcessRow_Floor_SSE2(const __m128i* const A0,
+                                              const __m128i* const A1,
+                                              const __m128i* const A2,
+                                              const __m128i* const A3,
+                                              const __m128i* const mult,
+                                              uint8_t* const dst) {
+  const __m128i mask = _mm_set_epi32(0xffffffffu, 0, 0xffffffffu, 0);
+  const __m128i B0 = _mm_mul_epu32(*A0, *mult);
+  const __m128i B1 = _mm_mul_epu32(*A1, *mult);
+  const __m128i B2 = _mm_mul_epu32(*A2, *mult);
+  const __m128i B3 = _mm_mul_epu32(*A3, *mult);
+  const __m128i D0 = _mm_srli_epi64(B0, WEBP_RESCALER_RFIX);
+  const __m128i D1 = _mm_srli_epi64(B1, WEBP_RESCALER_RFIX);
+#if (WEBP_RESCALER_RFIX < 32)
+  const __m128i D2 =
+      _mm_and_si128(_mm_slli_epi64(B2, 32 - WEBP_RESCALER_RFIX), mask);
+  const __m128i D3 =
+      _mm_and_si128(_mm_slli_epi64(B3, 32 - WEBP_RESCALER_RFIX), mask);
+#else
+  const __m128i D2 = _mm_and_si128(B2, mask);
+  const __m128i D3 = _mm_and_si128(B3, mask);
+#endif
+  const __m128i E0 = _mm_or_si128(D0, D2);
+  const __m128i E1 = _mm_or_si128(D1, D3);
+  const __m128i F = _mm_packs_epi32(E0, E1);
+  const __m128i G = _mm_packus_epi16(F, F);
+  _mm_storel_epi64((__m128i*)dst, G);
+}
+
 static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
   int x_out;
   uint8_t* const dst = wrk->dst;
@@ -322,12 +352,12 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
         const __m128i G1 = _mm_or_si128(D1, F3);
         _mm_storeu_si128((__m128i*)(irow + x_out + 0), G0);
         _mm_storeu_si128((__m128i*)(irow + x_out + 4), G1);
-        ProcessRow_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
+        ProcessRow_Floor_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
       }
     }
     for (; x_out < x_out_max; ++x_out) {
       const uint32_t frac = (int)MULT_FIX(frow[x_out], yscale);
-      const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+      const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
       assert(v >= 0 && v <= 255);
       dst[x_out] = v;
       irow[x_out] = frac;   // new fractional start
@@ -352,6 +382,7 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
   }
 }
 
+#undef MULT_FIX_FLOOR
 #undef MULT_FIX
 #undef ROUNDER
 

thirdparty/libwebp/src/dsp/yuv.h

@@ -207,4 +207,4 @@ static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_DSP_YUV_H_ */
+#endif  // WEBP_DSP_YUV_H_

thirdparty/libwebp/src/enc/analysis_enc.c

@@ -458,7 +458,7 @@ static void MergeJobs(const SegmentJob* const src, SegmentJob* const dst) {
   dst->uv_alpha += src->uv_alpha;
 }
 
-// initialize the job struct with some TODOs
+// initialize the job struct with some tasks to perform
 static void InitSegmentJob(VP8Encoder* const enc, SegmentJob* const job,
                            int start_row, int end_row) {
   WebPGetWorkerInterface()->Init(&job->worker);

thirdparty/libwebp/src/enc/backward_references_cost_enc.c

@@ -67,7 +67,7 @@ static int CostModelBuild(CostModel* const m, int xsize, int cache_bits,
 
   // The following code is similar to VP8LHistogramCreate but converts the
   // distance to plane code.
-  VP8LHistogramInit(histo, cache_bits);
+  VP8LHistogramInit(histo, cache_bits, /*init_arrays=*/ 1);
   while (VP8LRefsCursorOk(&c)) {
     VP8LHistogramAddSinglePixOrCopy(histo, c.cur_pos, VP8LDistanceToPlaneCode,
                                     xsize);

thirdparty/libwebp/src/enc/backward_references_enc.c

@@ -715,6 +715,7 @@ static int CalculateBestCacheSize(const uint32_t* argb, int quality,
   for (i = 0; i <= cache_bits_max; ++i) {
     histos[i] = VP8LAllocateHistogram(i);
     if (histos[i] == NULL) goto Error;
+    VP8LHistogramInit(histos[i], i, /*init_arrays=*/ 1);
     if (i == 0) continue;
     cc_init[i] = VP8LColorCacheInit(&hashers[i], i);
     if (!cc_init[i]) goto Error;

thirdparty/libwebp/src/enc/cost_enc.h

@@ -79,4 +79,4 @@ extern const uint16_t VP8FixedCostsI4[NUM_BMODES][NUM_BMODES][NUM_BMODES];
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_ENC_COST_ENC_H_ */
+#endif  // WEBP_ENC_COST_ENC_H_

thirdparty/libwebp/src/enc/delta_palettization_enc.c

@@ -1,455 +0,0 @@
-// Copyright 2015 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.
-// -----------------------------------------------------------------------------
-//
-// Author: Mislav Bradac ([email protected])
-//
-
-#include "src/enc/delta_palettization_enc.h"
-
-#ifdef WEBP_EXPERIMENTAL_FEATURES
-#include "src/webp/types.h"
-#include "src/dsp/lossless.h"
-
-#define MK_COL(r, g, b) (((r) << 16) + ((g) << 8) + (b))
-
-// Format allows palette up to 256 entries, but more palette entries produce
-// bigger entropy. In the future it will probably be useful to add more entries
-// that are far from the origin of the palette or choose remaining entries
-// dynamically.
-#define DELTA_PALETTE_SIZE 226
-
-// Palette used for delta_palettization. Entries are roughly sorted by distance
-// of their signed equivalents from the origin.
-static const uint32_t kDeltaPalette[DELTA_PALETTE_SIZE] = {
-  MK_COL(0u, 0u, 0u),
-  MK_COL(255u, 255u, 255u),
-  MK_COL(1u, 1u, 1u),
-  MK_COL(254u, 254u, 254u),
-  MK_COL(2u, 2u, 2u),
-  MK_COL(4u, 4u, 4u),
-  MK_COL(252u, 252u, 252u),
-  MK_COL(250u, 0u, 0u),
-  MK_COL(0u, 250u, 0u),
-  MK_COL(0u, 0u, 250u),
-  MK_COL(6u, 0u, 0u),
-  MK_COL(0u, 6u, 0u),
-  MK_COL(0u, 0u, 6u),
-  MK_COL(0u, 0u, 248u),
-  MK_COL(0u, 0u, 8u),
-  MK_COL(0u, 248u, 0u),
-  MK_COL(0u, 248u, 248u),
-  MK_COL(0u, 248u, 8u),
-  MK_COL(0u, 8u, 0u),
-  MK_COL(0u, 8u, 248u),
-  MK_COL(0u, 8u, 8u),
-  MK_COL(8u, 8u, 8u),
-  MK_COL(248u, 0u, 0u),
-  MK_COL(248u, 0u, 248u),
-  MK_COL(248u, 0u, 8u),
-  MK_COL(248u, 248u, 0u),
-  MK_COL(248u, 8u, 0u),
-  MK_COL(8u, 0u, 0u),
-  MK_COL(8u, 0u, 248u),
-  MK_COL(8u, 0u, 8u),
-  MK_COL(8u, 248u, 0u),
-  MK_COL(8u, 8u, 0u),
-  MK_COL(23u, 23u, 23u),
-  MK_COL(13u, 13u, 13u),
-  MK_COL(232u, 232u, 232u),
-  MK_COL(244u, 244u, 244u),
-  MK_COL(245u, 245u, 250u),
-  MK_COL(50u, 50u, 50u),
-  MK_COL(204u, 204u, 204u),
-  MK_COL(236u, 236u, 236u),
-  MK_COL(16u, 16u, 16u),
-  MK_COL(240u, 16u, 16u),
-  MK_COL(16u, 240u, 16u),
-  MK_COL(240u, 240u, 16u),
-  MK_COL(16u, 16u, 240u),
-  MK_COL(240u, 16u, 240u),
-  MK_COL(16u, 240u, 240u),
-  MK_COL(240u, 240u, 240u),
-  MK_COL(0u, 0u, 232u),
-  MK_COL(0u, 232u, 0u),
-  MK_COL(232u, 0u, 0u),
-  MK_COL(0u, 0u, 24u),
-  MK_COL(0u, 24u, 0u),
-  MK_COL(24u, 0u, 0u),
-  MK_COL(32u, 32u, 32u),
-  MK_COL(224u, 32u, 32u),
-  MK_COL(32u, 224u, 32u),
-  MK_COL(224u, 224u, 32u),
-  MK_COL(32u, 32u, 224u),
-  MK_COL(224u, 32u, 224u),
-  MK_COL(32u, 224u, 224u),
-  MK_COL(224u, 224u, 224u),
-  MK_COL(0u, 0u, 176u),
-  MK_COL(0u, 0u, 80u),
-  MK_COL(0u, 176u, 0u),
-  MK_COL(0u, 176u, 176u),
-  MK_COL(0u, 176u, 80u),
-  MK_COL(0u, 80u, 0u),
-  MK_COL(0u, 80u, 176u),
-  MK_COL(0u, 80u, 80u),
-  MK_COL(176u, 0u, 0u),
-  MK_COL(176u, 0u, 176u),
-  MK_COL(176u, 0u, 80u),
-  MK_COL(176u, 176u, 0u),
-  MK_COL(176u, 80u, 0u),
-  MK_COL(80u, 0u, 0u),
-  MK_COL(80u, 0u, 176u),
-  MK_COL(80u, 0u, 80u),
-  MK_COL(80u, 176u, 0u),
-  MK_COL(80u, 80u, 0u),
-  MK_COL(0u, 0u, 152u),
-  MK_COL(0u, 0u, 104u),
-  MK_COL(0u, 152u, 0u),
-  MK_COL(0u, 152u, 152u),
-  MK_COL(0u, 152u, 104u),
-  MK_COL(0u, 104u, 0u),
-  MK_COL(0u, 104u, 152u),
-  MK_COL(0u, 104u, 104u),
-  MK_COL(152u, 0u, 0u),
-  MK_COL(152u, 0u, 152u),
-  MK_COL(152u, 0u, 104u),
-  MK_COL(152u, 152u, 0u),
-  MK_COL(152u, 104u, 0u),
-  MK_COL(104u, 0u, 0u),
-  MK_COL(104u, 0u, 152u),
-  MK_COL(104u, 0u, 104u),
-  MK_COL(104u, 152u, 0u),
-  MK_COL(104u, 104u, 0u),
-  MK_COL(216u, 216u, 216u),
-  MK_COL(216u, 216u, 40u),
-  MK_COL(216u, 216u, 176u),
-  MK_COL(216u, 216u, 80u),
-  MK_COL(216u, 40u, 216u),
-  MK_COL(216u, 40u, 40u),
-  MK_COL(216u, 40u, 176u),
-  MK_COL(216u, 40u, 80u),
-  MK_COL(216u, 176u, 216u),
-  MK_COL(216u, 176u, 40u),
-  MK_COL(216u, 176u, 176u),
-  MK_COL(216u, 176u, 80u),
-  MK_COL(216u, 80u, 216u),
-  MK_COL(216u, 80u, 40u),
-  MK_COL(216u, 80u, 176u),
-  MK_COL(216u, 80u, 80u),
-  MK_COL(40u, 216u, 216u),
-  MK_COL(40u, 216u, 40u),
-  MK_COL(40u, 216u, 176u),
-  MK_COL(40u, 216u, 80u),
-  MK_COL(40u, 40u, 216u),
-  MK_COL(40u, 40u, 40u),
-  MK_COL(40u, 40u, 176u),
-  MK_COL(40u, 40u, 80u),
-  MK_COL(40u, 176u, 216u),
-  MK_COL(40u, 176u, 40u),
-  MK_COL(40u, 176u, 176u),
-  MK_COL(40u, 176u, 80u),
-  MK_COL(40u, 80u, 216u),
-  MK_COL(40u, 80u, 40u),
-  MK_COL(40u, 80u, 176u),
-  MK_COL(40u, 80u, 80u),
-  MK_COL(80u, 216u, 216u),
-  MK_COL(80u, 216u, 40u),
-  MK_COL(80u, 216u, 176u),
-  MK_COL(80u, 216u, 80u),
-  MK_COL(80u, 40u, 216u),
-  MK_COL(80u, 40u, 40u),
-  MK_COL(80u, 40u, 176u),
-  MK_COL(80u, 40u, 80u),
-  MK_COL(80u, 176u, 216u),
-  MK_COL(80u, 176u, 40u),
-  MK_COL(80u, 176u, 176u),
-  MK_COL(80u, 176u, 80u),
-  MK_COL(80u, 80u, 216u),
-  MK_COL(80u, 80u, 40u),
-  MK_COL(80u, 80u, 176u),
-  MK_COL(80u, 80u, 80u),
-  MK_COL(0u, 0u, 192u),
-  MK_COL(0u, 0u, 64u),
-  MK_COL(0u, 0u, 128u),
-  MK_COL(0u, 192u, 0u),
-  MK_COL(0u, 192u, 192u),
-  MK_COL(0u, 192u, 64u),
-  MK_COL(0u, 192u, 128u),
-  MK_COL(0u, 64u, 0u),
-  MK_COL(0u, 64u, 192u),
-  MK_COL(0u, 64u, 64u),
-  MK_COL(0u, 64u, 128u),
-  MK_COL(0u, 128u, 0u),
-  MK_COL(0u, 128u, 192u),
-  MK_COL(0u, 128u, 64u),
-  MK_COL(0u, 128u, 128u),
-  MK_COL(176u, 216u, 216u),
-  MK_COL(176u, 216u, 40u),
-  MK_COL(176u, 216u, 176u),
-  MK_COL(176u, 216u, 80u),
-  MK_COL(176u, 40u, 216u),
-  MK_COL(176u, 40u, 40u),
-  MK_COL(176u, 40u, 176u),
-  MK_COL(176u, 40u, 80u),
-  MK_COL(176u, 176u, 216u),
-  MK_COL(176u, 176u, 40u),
-  MK_COL(176u, 176u, 176u),
-  MK_COL(176u, 176u, 80u),
-  MK_COL(176u, 80u, 216u),
-  MK_COL(176u, 80u, 40u),
-  MK_COL(176u, 80u, 176u),
-  MK_COL(176u, 80u, 80u),
-  MK_COL(192u, 0u, 0u),
-  MK_COL(192u, 0u, 192u),
-  MK_COL(192u, 0u, 64u),
-  MK_COL(192u, 0u, 128u),
-  MK_COL(192u, 192u, 0u),
-  MK_COL(192u, 192u, 192u),
-  MK_COL(192u, 192u, 64u),
-  MK_COL(192u, 192u, 128u),
-  MK_COL(192u, 64u, 0u),
-  MK_COL(192u, 64u, 192u),
-  MK_COL(192u, 64u, 64u),
-  MK_COL(192u, 64u, 128u),
-  MK_COL(192u, 128u, 0u),
-  MK_COL(192u, 128u, 192u),
-  MK_COL(192u, 128u, 64u),
-  MK_COL(192u, 128u, 128u),
-  MK_COL(64u, 0u, 0u),
-  MK_COL(64u, 0u, 192u),
-  MK_COL(64u, 0u, 64u),
-  MK_COL(64u, 0u, 128u),
-  MK_COL(64u, 192u, 0u),
-  MK_COL(64u, 192u, 192u),
-  MK_COL(64u, 192u, 64u),
-  MK_COL(64u, 192u, 128u),
-  MK_COL(64u, 64u, 0u),
-  MK_COL(64u, 64u, 192u),
-  MK_COL(64u, 64u, 64u),
-  MK_COL(64u, 64u, 128u),
-  MK_COL(64u, 128u, 0u),
-  MK_COL(64u, 128u, 192u),
-  MK_COL(64u, 128u, 64u),
-  MK_COL(64u, 128u, 128u),
-  MK_COL(128u, 0u, 0u),
-  MK_COL(128u, 0u, 192u),
-  MK_COL(128u, 0u, 64u),
-  MK_COL(128u, 0u, 128u),
-  MK_COL(128u, 192u, 0u),
-  MK_COL(128u, 192u, 192u),
-  MK_COL(128u, 192u, 64u),
-  MK_COL(128u, 192u, 128u),
-  MK_COL(128u, 64u, 0u),
-  MK_COL(128u, 64u, 192u),
-  MK_COL(128u, 64u, 64u),
-  MK_COL(128u, 64u, 128u),
-  MK_COL(128u, 128u, 0u),
-  MK_COL(128u, 128u, 192u),
-  MK_COL(128u, 128u, 64u),
-  MK_COL(128u, 128u, 128u),
-};
-
-#undef MK_COL
-
-//------------------------------------------------------------------------------
-// TODO(skal): move the functions to dsp/lossless.c when the correct
-// granularity is found. For now, we'll just copy-paste some useful bits
-// here instead.
-
-// In-place sum of each component with mod 256.
-static WEBP_INLINE void AddPixelsEq(uint32_t* a, uint32_t b) {
-  const uint32_t alpha_and_green = (*a & 0xff00ff00u) + (b & 0xff00ff00u);
-  const uint32_t red_and_blue = (*a & 0x00ff00ffu) + (b & 0x00ff00ffu);
-  *a = (alpha_and_green & 0xff00ff00u) | (red_and_blue & 0x00ff00ffu);
-}
-
-static WEBP_INLINE uint32_t Clip255(uint32_t a) {
-  if (a < 256) {
-    return a;
-  }
-  // return 0, when a is a negative integer.
-  // return 255, when a is positive.
-  return ~a >> 24;
-}
-
-// Delta palettization functions.
-static WEBP_INLINE int Square(int x) {
-  return x * x;
-}
-
-static WEBP_INLINE uint32_t Intensity(uint32_t a) {
-  return
-      30 * ((a >> 16) & 0xff) +
-      59 * ((a >>  8) & 0xff) +
-      11 * ((a >>  0) & 0xff);
-}
-
-static uint32_t CalcDist(uint32_t predicted_value, uint32_t actual_value,
-                         uint32_t palette_entry) {
-  int i;
-  uint32_t distance = 0;
-  AddPixelsEq(&predicted_value, palette_entry);
-  for (i = 0; i < 32; i += 8) {
-    const int32_t av = (actual_value >> i) & 0xff;
-    const int32_t pv = (predicted_value >> i) & 0xff;
-    distance += Square(pv - av);
-  }
-  // We sum square of intensity difference with factor 10, but because Intensity
-  // returns 100 times real intensity we need to multiply differences of colors
-  // by 1000.
-  distance *= 1000u;
-  distance += Square(Intensity(predicted_value)
-                     - Intensity(actual_value));
-  return distance;
-}
-
-static uint32_t Predict(int x, int y, uint32_t* image) {
-  const uint32_t t = (y == 0) ? ARGB_BLACK : image[x];
-  const uint32_t l = (x == 0) ? ARGB_BLACK : image[x - 1];
-  const uint32_t p =
-      (((((t >> 24) & 0xff) + ((l >> 24) & 0xff)) / 2) << 24) +
-      (((((t >> 16) & 0xff) + ((l >> 16) & 0xff)) / 2) << 16) +
-      (((((t >>  8) & 0xff) + ((l >>  8) & 0xff)) / 2) <<  8) +
-      (((((t >>  0) & 0xff) + ((l >>  0) & 0xff)) / 2) <<  0);
-  if (x == 0 && y == 0) return ARGB_BLACK;
-  if (x == 0) return t;
-  if (y == 0) return l;
-  return p;
-}
-
-static WEBP_INLINE int AddSubtractComponentFullWithCoefficient(
-    int a, int b, int c) {
-  return Clip255(a + ((b - c) >> 2));
-}
-
-static WEBP_INLINE uint32_t ClampedAddSubtractFullWithCoefficient(
-    uint32_t c0, uint32_t c1, uint32_t c2) {
-  const int a = AddSubtractComponentFullWithCoefficient(
-      c0 >> 24, c1 >> 24, c2 >> 24);
-  const int r = AddSubtractComponentFullWithCoefficient((c0 >> 16) & 0xff,
-                                                       (c1 >> 16) & 0xff,
-                                                       (c2 >> 16) & 0xff);
-  const int g = AddSubtractComponentFullWithCoefficient((c0 >> 8) & 0xff,
-                                                       (c1 >> 8) & 0xff,
-                                                       (c2 >> 8) & 0xff);
-  const int b = AddSubtractComponentFullWithCoefficient(
-      c0 & 0xff, c1 & 0xff, c2 & 0xff);
-  return ((uint32_t)a << 24) | (r << 16) | (g << 8) | b;
-}
-
-//------------------------------------------------------------------------------
-
-// Find palette entry with minimum error from difference of actual pixel value
-// and predicted pixel value. Propagate error of pixel to its top and left pixel
-// in src array. Write predicted_value + palette_entry to new_image. Return
-// index of best palette entry.
-static int FindBestPaletteEntry(uint32_t src, uint32_t predicted_value,
-                                const uint32_t palette[], int palette_size) {
-  int i;
-  int idx = 0;
-  uint32_t best_distance = CalcDist(predicted_value, src, palette[0]);
-  for (i = 1; i < palette_size; ++i) {
-    const uint32_t distance = CalcDist(predicted_value, src, palette[i]);
-    if (distance < best_distance) {
-      best_distance = distance;
-      idx = i;
-    }
-  }
-  return idx;
-}
-
-static void ApplyBestPaletteEntry(int x, int y,
-                                  uint32_t new_value, uint32_t palette_value,
-                                  uint32_t* src, int src_stride,
-                                  uint32_t* new_image) {
-  AddPixelsEq(&new_value, palette_value);
-  if (x > 0) {
-    src[x - 1] = ClampedAddSubtractFullWithCoefficient(src[x - 1],
-                                                       new_value, src[x]);
-  }
-  if (y > 0) {
-    src[x - src_stride] =
-        ClampedAddSubtractFullWithCoefficient(src[x - src_stride],
-                                              new_value, src[x]);
-  }
-  new_image[x] = new_value;
-}
-
-//------------------------------------------------------------------------------
-// Main entry point
-
-static WebPEncodingError ApplyDeltaPalette(uint32_t* src, uint32_t* dst,
-                                           uint32_t src_stride,
-                                           uint32_t dst_stride,
-                                           const uint32_t* palette,
-                                           int palette_size,
-                                           int width, int height,
-                                           int num_passes) {
-  int x, y;
-  WebPEncodingError err = VP8_ENC_OK;
-  uint32_t* new_image = (uint32_t*)WebPSafeMalloc(width, sizeof(*new_image));
-  uint8_t* const tmp_row = (uint8_t*)WebPSafeMalloc(width, sizeof(*tmp_row));
-  if (new_image == NULL || tmp_row == NULL) {
-    err = VP8_ENC_ERROR_OUT_OF_MEMORY;
-    goto Error;
-  }
-
-  while (num_passes--) {
-    uint32_t* cur_src = src;
-    uint32_t* cur_dst = dst;
-    for (y = 0; y < height; ++y) {
-      for (x = 0; x < width; ++x) {
-        const uint32_t predicted_value = Predict(x, y, new_image);
-        tmp_row[x] = FindBestPaletteEntry(cur_src[x], predicted_value,
-                                          palette, palette_size);
-        ApplyBestPaletteEntry(x, y, predicted_value, palette[tmp_row[x]],
-                              cur_src, src_stride, new_image);
-      }
-      for (x = 0; x < width; ++x) {
-        cur_dst[x] = palette[tmp_row[x]];
-      }
-      cur_src += src_stride;
-      cur_dst += dst_stride;
-    }
-  }
- Error:
-  WebPSafeFree(new_image);
-  WebPSafeFree(tmp_row);
-  return err;
-}
-
-// replaces enc->argb_ by a palettizable approximation of it,
-// and generates optimal enc->palette_[]
-WebPEncodingError WebPSearchOptimalDeltaPalette(VP8LEncoder* const enc) {
-  const WebPPicture* const pic = enc->pic_;
-  uint32_t* src = pic->argb;
-  uint32_t* dst = enc->argb_;
-  const int width = pic->width;
-  const int height = pic->height;
-
-  WebPEncodingError err = VP8_ENC_OK;
-  memcpy(enc->palette_, kDeltaPalette, sizeof(kDeltaPalette));
-  enc->palette_[DELTA_PALETTE_SIZE - 1] = src[0] - 0xff000000u;
-  enc->palette_size_ = DELTA_PALETTE_SIZE;
-  err = ApplyDeltaPalette(src, dst, pic->argb_stride, enc->current_width_,
-                          enc->palette_, enc->palette_size_,
-                          width, height, 2);
-  if (err != VP8_ENC_OK) goto Error;
-
- Error:
-  return err;
-}
-
-#else  // !WEBP_EXPERIMENTAL_FEATURES
-
-WebPEncodingError WebPSearchOptimalDeltaPalette(VP8LEncoder* const enc) {
-  (void)enc;
-  return VP8_ENC_ERROR_INVALID_CONFIGURATION;
-}
-
-#endif  // WEBP_EXPERIMENTAL_FEATURES

thirdparty/libwebp/src/enc/delta_palettization_enc.h

@@ -1,25 +0,0 @@
-// Copyright 2015 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.
-// -----------------------------------------------------------------------------
-//
-// Author: Mislav Bradac ([email protected])
-//
-
-#ifndef WEBP_ENC_DELTA_PALETTIZATION_ENC_H_
-#define WEBP_ENC_DELTA_PALETTIZATION_ENC_H_
-
-#include "src/webp/encode.h"
-#include "src/enc/vp8li_enc.h"
-
-// Replaces enc->argb_[] input by a palettizable approximation of it,
-// and generates optimal enc->palette_[].
-// This function can revert enc->use_palette_ / enc->use_predict_ flag
-// if delta-palettization is not producing expected saving.
-WebPEncodingError WebPSearchOptimalDeltaPalette(VP8LEncoder* const enc);
-
-#endif  // WEBP_ENC_DELTA_PALETTIZATION_ENC_H_

thirdparty/libwebp/src/enc/histogram_enc.c

@@ -51,10 +51,12 @@ static void HistogramCopy(const VP8LHistogram* const src,
                           VP8LHistogram* const dst) {
   uint32_t* const dst_literal = dst->literal_;
   const int dst_cache_bits = dst->palette_code_bits_;
+  const int literal_size = VP8LHistogramNumCodes(dst_cache_bits);
   const int histo_size = VP8LGetHistogramSize(dst_cache_bits);
   assert(src->palette_code_bits_ == dst_cache_bits);
   memcpy(dst, src, histo_size);
   dst->literal_ = dst_literal;
+  memcpy(dst->literal_, src->literal_, literal_size * sizeof(*dst->literal_));
 }
 
 int VP8LGetHistogramSize(int cache_bits) {
@@ -91,9 +93,19 @@ void VP8LHistogramCreate(VP8LHistogram* const p,
   VP8LHistogramStoreRefs(refs, p);
 }
 
-void VP8LHistogramInit(VP8LHistogram* const p, int palette_code_bits) {
+void VP8LHistogramInit(VP8LHistogram* const p, int palette_code_bits,
+                       int init_arrays) {
   p->palette_code_bits_ = palette_code_bits;
-  HistogramClear(p);
+  if (init_arrays) {
+    HistogramClear(p);
+  } else {
+    p->trivial_symbol_ = 0;
+    p->bit_cost_ = 0.;
+    p->literal_cost_ = 0.;
+    p->red_cost_ = 0.;
+    p->blue_cost_ = 0.;
+    memset(p->is_used_, 0, sizeof(p->is_used_));
+  }
 }
 
 VP8LHistogram* VP8LAllocateHistogram(int cache_bits) {
@@ -104,37 +116,70 @@ VP8LHistogram* VP8LAllocateHistogram(int cache_bits) {
   histo = (VP8LHistogram*)memory;
   // literal_ won't necessary be aligned.
   histo->literal_ = (uint32_t*)(memory + sizeof(VP8LHistogram));
-  VP8LHistogramInit(histo, cache_bits);
+  VP8LHistogramInit(histo, cache_bits, /*init_arrays=*/ 0);
   return histo;
 }
 
+// Resets the pointers of the histograms to point to the bit buffer in the set.
+static void HistogramSetResetPointers(VP8LHistogramSet* const set,
+                                      int cache_bits) {
+  int i;
+  const int histo_size = VP8LGetHistogramSize(cache_bits);
+  uint8_t* memory = (uint8_t*) (set->histograms);
+  memory += set->max_size * sizeof(*set->histograms);
+  for (i = 0; i < set->max_size; ++i) {
+    memory = (uint8_t*) WEBP_ALIGN(memory);
+    set->histograms[i] = (VP8LHistogram*) memory;
+    // literal_ won't necessary be aligned.
+    set->histograms[i]->literal_ = (uint32_t*)(memory + sizeof(VP8LHistogram));
+    memory += histo_size;
+  }
+}
+
+// Returns the total size of the VP8LHistogramSet.
+static size_t HistogramSetTotalSize(int size, int cache_bits) {
+  const int histo_size = VP8LGetHistogramSize(cache_bits);
+  return (sizeof(VP8LHistogramSet) + size * (sizeof(VP8LHistogram*) +
+          histo_size + WEBP_ALIGN_CST));
+}
+
 VP8LHistogramSet* VP8LAllocateHistogramSet(int size, int cache_bits) {
   int i;
   VP8LHistogramSet* set;
-  const int histo_size = VP8LGetHistogramSize(cache_bits);
-  const size_t total_size =
-      sizeof(*set) + size * (sizeof(*set->histograms) +
-      histo_size + WEBP_ALIGN_CST);
+  const size_t total_size = HistogramSetTotalSize(size, cache_bits);
   uint8_t* memory = (uint8_t*)WebPSafeMalloc(total_size, sizeof(*memory));
   if (memory == NULL) return NULL;
 
   set = (VP8LHistogramSet*)memory;
   memory += sizeof(*set);
   set->histograms = (VP8LHistogram**)memory;
-  memory += size * sizeof(*set->histograms);
   set->max_size = size;
   set->size = size;
+  HistogramSetResetPointers(set, cache_bits);
   for (i = 0; i < size; ++i) {
-    memory = (uint8_t*)WEBP_ALIGN(memory);
-    set->histograms[i] = (VP8LHistogram*)memory;
-    // literal_ won't necessary be aligned.
-    set->histograms[i]->literal_ = (uint32_t*)(memory + sizeof(VP8LHistogram));
-    VP8LHistogramInit(set->histograms[i], cache_bits);
-    memory += histo_size;
+    VP8LHistogramInit(set->histograms[i], cache_bits, /*init_arrays=*/ 0);
   }
   return set;
 }
 
+void VP8LHistogramSetClear(VP8LHistogramSet* const set) {
+  int i;
+  const int cache_bits = set->histograms[0]->palette_code_bits_;
+  const int size = set->size;
+  const size_t total_size = HistogramSetTotalSize(size, cache_bits);
+  uint8_t* memory = (uint8_t*)set;
+
+  memset(memory, 0, total_size);
+  memory += sizeof(*set);
+  set->histograms = (VP8LHistogram**)memory;
+  set->max_size = size;
+  set->size = size;
+  HistogramSetResetPointers(set, cache_bits);
+  for (i = 0; i < size; ++i) {
+    set->histograms[i]->palette_code_bits_ = cache_bits;
+  }
+}
+
 // -----------------------------------------------------------------------------
 
 void VP8LHistogramAddSinglePixOrCopy(VP8LHistogram* const histo,
@@ -237,7 +282,8 @@ static double FinalHuffmanCost(const VP8LStreaks* const stats) {
 // 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) {
+                             uint32_t* const trivial_sym,
+                             uint8_t* const is_used) {
   VP8LBitEntropy bit_entropy;
   VP8LStreaks stats;
   VP8LGetEntropyUnrefined(population, length, &bit_entropy, &stats);
@@ -245,6 +291,8 @@ static double PopulationCost(const uint32_t* const population, int length,
     *trivial_sym = (bit_entropy.nonzeros == 1) ? bit_entropy.nonzero_code
                                                : VP8L_NON_TRIVIAL_SYM;
   }
+  // The histogram is used if there is at least one non-zero streak.
+  *is_used = (stats.streaks[1][0] != 0 || stats.streaks[1][1] != 0);
 
   return BitsEntropyRefine(&bit_entropy) + FinalHuffmanCost(&stats);
 }
@@ -253,7 +301,9 @@ static double PopulationCost(const uint32_t* const population, int length,
 // 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 trivial_at_end) {
+                                             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
@@ -262,28 +312,43 @@ static WEBP_INLINE double GetCombinedEntropy(const uint32_t* const X,
     // Only FinalHuffmanCost needs to be evaluated.
     memset(&stats, 0, sizeof(stats));
     // Deal with the non-zero value at index 0 or length-1.
-    stats.streaks[1][0] += 1;
+    stats.streaks[1][0] = 1;
     // Deal with the following/previous zero streak.
-    stats.counts[0] += 1;
-    stats.streaks[0][1] += length - 1;
+    stats.counts[0] = 1;
+    stats.streaks[0][1] = length - 1;
     return FinalHuffmanCost(&stats);
   } else {
     VP8LBitEntropy bit_entropy;
-    VP8LGetCombinedEntropyUnrefined(X, Y, length, &bit_entropy, &stats);
+    if (is_X_used) {
+      if (is_Y_used) {
+        VP8LGetCombinedEntropyUnrefined(X, Y, length, &bit_entropy, &stats);
+      } else {
+        VP8LGetEntropyUnrefined(X, length, &bit_entropy, &stats);
+      }
+    } else {
+      if (is_Y_used) {
+        VP8LGetEntropyUnrefined(Y, length, &bit_entropy, &stats);
+      } else {
+        memset(&stats, 0, sizeof(stats));
+        stats.counts[0] = 1;
+        stats.streaks[0][length > 3] = length;
+        VP8LBitEntropyInit(&bit_entropy);
+      }
+    }
 
     return BitsEntropyRefine(&bit_entropy) + FinalHuffmanCost(&stats);
   }
 }
 
 // Estimates the Entropy + Huffman + other block overhead size cost.
-double VP8LHistogramEstimateBits(const VP8LHistogram* const p) {
+double VP8LHistogramEstimateBits(VP8LHistogram* const p) {
   return
-      PopulationCost(
-          p->literal_, VP8LHistogramNumCodes(p->palette_code_bits_), NULL)
-      + PopulationCost(p->red_, NUM_LITERAL_CODES, NULL)
-      + PopulationCost(p->blue_, NUM_LITERAL_CODES, NULL)
-      + PopulationCost(p->alpha_, NUM_LITERAL_CODES, NULL)
-      + PopulationCost(p->distance_, NUM_DISTANCE_CODES, NULL)
+      PopulationCost(p->literal_, VP8LHistogramNumCodes(p->palette_code_bits_),
+                     NULL, &p->is_used_[0])
+      + PopulationCost(p->red_, NUM_LITERAL_CODES, NULL, &p->is_used_[1])
+      + PopulationCost(p->blue_, NUM_LITERAL_CODES, NULL, &p->is_used_[2])
+      + PopulationCost(p->alpha_, NUM_LITERAL_CODES, NULL, &p->is_used_[3])
+      + PopulationCost(p->distance_, NUM_DISTANCE_CODES, NULL, &p->is_used_[4])
       + VP8LExtraCost(p->literal_ + NUM_LITERAL_CODES, NUM_LENGTH_CODES)
       + VP8LExtraCost(p->distance_, NUM_DISTANCE_CODES);
 }
@@ -299,7 +364,8 @@ static int GetCombinedHistogramEntropy(const VP8LHistogram* const a,
   int trivial_at_end = 0;
   assert(a->palette_code_bits_ == b->palette_code_bits_);
   *cost += GetCombinedEntropy(a->literal_, b->literal_,
-                              VP8LHistogramNumCodes(palette_code_bits), 0);
+                              VP8LHistogramNumCodes(palette_code_bits),
+                              a->is_used_[0], b->is_used_[0], 0);
   *cost += VP8LExtraCostCombined(a->literal_ + NUM_LITERAL_CODES,
                                  b->literal_ + NUM_LITERAL_CODES,
                                  NUM_LENGTH_CODES);
@@ -319,19 +385,23 @@ static int GetCombinedHistogramEntropy(const VP8LHistogram* const a,
   }
 
   *cost +=
-      GetCombinedEntropy(a->red_, b->red_, NUM_LITERAL_CODES, trivial_at_end);
+      GetCombinedEntropy(a->red_, b->red_, NUM_LITERAL_CODES, a->is_used_[1],
+                         b->is_used_[1], trivial_at_end);
   if (*cost > cost_threshold) return 0;
 
   *cost +=
-      GetCombinedEntropy(a->blue_, b->blue_, NUM_LITERAL_CODES, trivial_at_end);
+      GetCombinedEntropy(a->blue_, b->blue_, NUM_LITERAL_CODES, a->is_used_[2],
+                         b->is_used_[2], trivial_at_end);
   if (*cost > cost_threshold) return 0;
 
-  *cost += GetCombinedEntropy(a->alpha_, b->alpha_, NUM_LITERAL_CODES,
-                              trivial_at_end);
+  *cost +=
+      GetCombinedEntropy(a->alpha_, b->alpha_, NUM_LITERAL_CODES,
+                         a->is_used_[3], b->is_used_[3], trivial_at_end);
   if (*cost > cost_threshold) return 0;
 
   *cost +=
-      GetCombinedEntropy(a->distance_, b->distance_, NUM_DISTANCE_CODES, 0);
+      GetCombinedEntropy(a->distance_, b->distance_, NUM_DISTANCE_CODES,
+                         a->is_used_[4], b->is_used_[4], 0);
   *cost +=
       VP8LExtraCostCombined(a->distance_, b->distance_, NUM_DISTANCE_CODES);
   if (*cost > cost_threshold) return 0;
@@ -419,16 +489,19 @@ 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);
+      PopulationCost(h->alpha_, NUM_LITERAL_CODES, &alpha_sym,
+                     &h->is_used_[3]);
   const double distance_cost =
-      PopulationCost(h->distance_, NUM_DISTANCE_CODES, NULL) +
+      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_);
-  h->literal_cost_ = PopulationCost(h->literal_, num_codes, NULL) +
-                     VP8LExtraCost(h->literal_ + NUM_LITERAL_CODES,
-                                   NUM_LENGTH_CODES);
-  h->red_cost_ = PopulationCost(h->red_, NUM_LITERAL_CODES, &red_sym);
-  h->blue_cost_ = PopulationCost(h->blue_, NUM_LITERAL_CODES, &blue_sym);
+  h->literal_cost_ =
+      PopulationCost(h->literal_, num_codes, NULL, &h->is_used_[0]) +
+          VP8LExtraCost(h->literal_ + NUM_LITERAL_CODES, NUM_LENGTH_CODES);
+  h->red_cost_ =
+      PopulationCost(h->red_, NUM_LITERAL_CODES, &red_sym, &h->is_used_[1]);
+  h->blue_cost_ =
+      PopulationCost(h->blue_, NUM_LITERAL_CODES, &blue_sym, &h->is_used_[2]);
   h->bit_cost_ = h->literal_cost_ + h->red_cost_ + h->blue_cost_ +
                  alpha_cost + distance_cost;
   if ((alpha_sym | red_sym | blue_sym) == VP8L_NON_TRIVIAL_SYM) {
@@ -473,6 +546,7 @@ static void HistogramBuild(
   VP8LHistogram** const histograms = image_histo->histograms;
   VP8LRefsCursor c = VP8LRefsCursorInit(backward_refs);
   assert(histo_bits > 0);
+  VP8LHistogramSetClear(image_histo);
   while (VP8LRefsCursorOk(&c)) {
     const PixOrCopy* const v = c.cur_pos;
     const int ix = (y >> histo_bits) * histo_xsize + (x >> histo_bits);
@@ -493,11 +567,19 @@ static void HistogramCopyAndAnalyze(
   const int histo_size = orig_histo->size;
   VP8LHistogram** const orig_histograms = orig_histo->histograms;
   VP8LHistogram** const histograms = image_histo->histograms;
+  image_histo->size = 0;
   for (i = 0; i < histo_size; ++i) {
     VP8LHistogram* const histo = orig_histograms[i];
     UpdateHistogramCost(histo);
+
+    // Skip the histogram if it is completely empty, which can happen for tiles
+    // with no information (when they are skipped because of LZ77).
+    if (!histo->is_used_[0] && !histo->is_used_[1] && !histo->is_used_[2]
+        && !histo->is_used_[3] && !histo->is_used_[4]) {
+      continue;
+    }
     // Copy histograms from orig_histo[] to image_histo[].
-    HistogramCopy(histo, histograms[i]);
+    HistogramCopy(histo, histograms[image_histo->size++]);
   }
 }
 
@@ -674,6 +756,18 @@ 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,
+                                 HistogramPair* const pair) {
+  const double 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;
+}
+
 // 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.
@@ -683,7 +777,6 @@ static double HistoQueuePush(HistoQueue* const histo_queue,
   const VP8LHistogram* h1;
   const VP8LHistogram* h2;
   HistogramPair pair;
-  double sum_cost;
 
   assert(threshold <= 0.);
   if (idx1 > idx2) {
@@ -695,10 +788,8 @@ static double HistoQueuePush(HistoQueue* const histo_queue,
   pair.idx2 = idx2;
   h1 = histograms[idx1];
   h2 = histograms[idx2];
-  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;
+
+  HistoQueueUpdatePair(h1, h2, threshold, &pair);
 
   // Do not even consider the pair if it does not improve the entropy.
   if (pair.cost_diff >= threshold) return 0.;
@@ -891,8 +982,7 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
       }
       if (do_eval) {
         // Re-evaluate the cost of an updated pair.
-        GetCombinedHistogramEntropy(histograms[p->idx1], histograms[p->idx2], 0,
-                                    &p->cost_diff);
+        HistoQueueUpdatePair(histograms[p->idx1], histograms[p->idx2], 0., p);
         if (p->cost_diff >= 0.) {
           HistoQueuePopPair(&histo_queue, p);
           continue;
@@ -987,8 +1077,7 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
   // histograms of small sizes (as bin_map will be very sparse) and
   // maximum quality q==100 (to preserve the compression gains at that level).
   const int entropy_combine_num_bins = low_effort ? NUM_PARTITIONS : BIN_SIZE;
-  const int entropy_combine =
-      (orig_histo->size > entropy_combine_num_bins * 2) && (quality < 100);
+  int entropy_combine;
 
   if (orig_histo == NULL) goto Error;
 
@@ -996,15 +1085,16 @@ int VP8LGetHistoImageSymbols(int xsize, int ysize,
   HistogramBuild(xsize, histo_bits, refs, orig_histo);
   // Copies the histograms and computes its bit_cost.
   HistogramCopyAndAnalyze(orig_histo, image_histo);
-
+  entropy_combine =
+      (image_histo->size > entropy_combine_num_bins * 2) && (quality < 100);
   if (entropy_combine) {
-    const int bin_map_size = orig_histo->size;
+    const int bin_map_size = image_histo->size;
     // Reuse histogram_symbols storage. By definition, it's guaranteed to be ok.
     uint16_t* const bin_map = histogram_symbols;
     const double combine_cost_factor =
         GetCombineCostFactor(image_histo_raw_size, quality);
 
-    HistogramAnalyzeEntropyBin(orig_histo, bin_map, low_effort);
+    HistogramAnalyzeEntropyBin(image_histo, bin_map, low_effort);
     // Collapse histograms with similar entropy.
     HistogramCombineEntropyBin(image_histo, tmp_histo, bin_map, bin_map_size,
                                entropy_combine_num_bins, combine_cost_factor,

thirdparty/libwebp/src/enc/histogram_enc.h

@@ -44,6 +44,7 @@ typedef struct {
   double literal_cost_;      // Cached values of dominant entropy costs:
   double red_cost_;          // literal, red & blue.
   double blue_cost_;
+  uint8_t is_used_[5];       // 5 for literal, red, blue, alpha, distance
 } VP8LHistogram;
 
 // Collection of histograms with fixed capacity, allocated as one
@@ -67,7 +68,9 @@ void VP8LHistogramCreate(VP8LHistogram* const p,
 int VP8LGetHistogramSize(int palette_code_bits);
 
 // Set the palette_code_bits and reset the stats.
-void VP8LHistogramInit(VP8LHistogram* const p, int palette_code_bits);
+// If init_arrays is true, the arrays are also filled with 0's.
+void VP8LHistogramInit(VP8LHistogram* const p, int palette_code_bits,
+                       int init_arrays);
 
 // Collect all the references into a histogram (without reset)
 void VP8LHistogramStoreRefs(const VP8LBackwardRefs* const refs,
@@ -83,6 +86,9 @@ void VP8LFreeHistogramSet(VP8LHistogramSet* const histo);
 // using 'cache_bits'. Return NULL in case of memory error.
 VP8LHistogramSet* VP8LAllocateHistogramSet(int size, int cache_bits);
 
+// Set the histograms in set to 0.
+void VP8LHistogramSetClear(VP8LHistogramSet* const set);
+
 // Allocate and initialize histogram object with specified 'cache_bits'.
 // Returns NULL in case of memory error.
 // Special case of VP8LAllocateHistogramSet, with size equals 1.
@@ -113,7 +119,7 @@ double VP8LBitsEntropy(const uint32_t* const array, int n);
 
 // Estimate how many bits the combined entropy of literals and distance
 // approximately maps to.
-double VP8LHistogramEstimateBits(const VP8LHistogram* const p);
+double VP8LHistogramEstimateBits(VP8LHistogram* const p);
 
 #ifdef __cplusplus
 }

thirdparty/libwebp/src/enc/iterator_enc.c

@@ -128,7 +128,7 @@ static void ImportLine(const uint8_t* src, int src_stride,
   for (; i < total_len; ++i) dst[i] = dst[len - 1];
 }
 
-void VP8IteratorImport(VP8EncIterator* const it, uint8_t* tmp_32) {
+void VP8IteratorImport(VP8EncIterator* const it, uint8_t* const tmp_32) {
   const VP8Encoder* const enc = it->enc_;
   const int x = it->x_, y = it->y_;
   const WebPPicture* const pic = enc->pic_;

thirdparty/libwebp/src/enc/picture_tools_enc.c

@@ -16,10 +16,6 @@
 #include "src/enc/vp8i_enc.h"
 #include "src/dsp/yuv.h"
 
-static WEBP_INLINE uint32_t MakeARGB32(int r, int g, int b) {
-  return (0xff000000u | (r << 16) | (g << 8) | b);
-}
-
 //------------------------------------------------------------------------------
 // Helper: clean up fully transparent area to help compressibility.
 
@@ -195,6 +191,10 @@ void WebPCleanupTransparentAreaLossless(WebPPicture* const pic) {
 #define BLEND_10BIT(V0, V1, ALPHA) \
     ((((V0) * (1020 - (ALPHA)) + (V1) * (ALPHA)) * 0x101 + 1024) >> 18)
 
+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) {
   const int red = (background_rgb >> 16) & 0xff;
   const int green = (background_rgb >> 8) & 0xff;
@@ -208,39 +208,44 @@ void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb) {
     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;
-    if (!has_alpha || pic->a == NULL) return;    // nothing to do
+    uint8_t* y_ptr = pic->y;
+    uint8_t* u_ptr = pic->u;
+    uint8_t* v_ptr = pic->v;
+    uint8_t* a_ptr = pic->a;
+    if (!has_alpha || a_ptr == NULL) return;    // nothing to do
     for (y = 0; y < pic->height; ++y) {
       // Luma blending
-      uint8_t* const y_ptr = pic->y + y * pic->y_stride;
-      uint8_t* const a_ptr = pic->a + y * pic->a_stride;
       for (x = 0; x < pic->width; ++x) {
-        const int alpha = a_ptr[x];
+        const uint8_t alpha = a_ptr[x];
         if (alpha < 0xff) {
-          y_ptr[x] = BLEND(Y0, y_ptr[x], a_ptr[x]);
+          y_ptr[x] = BLEND(Y0, y_ptr[x], alpha);
         }
       }
       // Chroma blending every even line
       if ((y & 1) == 0) {
-        uint8_t* const u = pic->u + (y >> 1) * pic->uv_stride;
-        uint8_t* const v = pic->v + (y >> 1) * pic->uv_stride;
         uint8_t* const a_ptr2 =
             (y + 1 == pic->height) ? a_ptr : a_ptr + pic->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?
-          const int alpha =
+          const uint32_t alpha =
               a_ptr[2 * x + 0] + a_ptr[2 * x + 1] +
               a_ptr2[2 * x + 0] + a_ptr2[2 * x + 1];
-          u[x] = BLEND_10BIT(U0, u[x], alpha);
-          v[x] = BLEND_10BIT(V0, v[x], alpha);
+          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
-          const int alpha = 2 * (a_ptr[2 * x + 0] + a_ptr2[2 * x + 0]);
-          u[x] = BLEND_10BIT(U0, u[x], alpha);
-          v[x] = BLEND_10BIT(V0, v[x], alpha);
+          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;
       }
-      memset(a_ptr, 0xff, pic->width);
+      memset(a_ptr, 0xff, pic->width);  // reset alpha value to opaque
+      a_ptr += pic->a_stride;
+      y_ptr += pic->y_stride;
     }
   } else {
     uint32_t* argb = pic->argb;

thirdparty/libwebp/src/enc/vp8i_enc.h

@@ -32,7 +32,7 @@ extern "C" {
 // version numbers
 #define ENC_MAJ_VERSION 1
 #define ENC_MIN_VERSION 0
-#define ENC_REV_VERSION 0
+#define ENC_REV_VERSION 1
 
 enum { MAX_LF_LEVELS = 64,       // Maximum loop filter level
        MAX_VARIABLE_LEVEL = 67,  // last (inclusive) level with variable cost
@@ -278,7 +278,7 @@ int VP8IteratorIsDone(const VP8EncIterator* const it);
 // Import uncompressed samples from source.
 // If tmp_32 is not NULL, import boundary samples too.
 // tmp_32 is a 32-bytes scratch buffer that must be aligned in memory.
-void VP8IteratorImport(VP8EncIterator* const it, uint8_t* tmp_32);
+void VP8IteratorImport(VP8EncIterator* const it, uint8_t* const tmp_32);
 // export decimated samples
 void VP8IteratorExport(const VP8EncIterator* const it);
 // go to next macroblock. Returns false if not finished.
@@ -515,4 +515,4 @@ void WebPCleanupTransparentAreaLossless(WebPPicture* const pic);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_ENC_VP8I_ENC_H_ */
+#endif  // WEBP_ENC_VP8I_ENC_H_

thirdparty/libwebp/src/enc/vp8l_enc.c

@@ -809,6 +809,7 @@ static WebPEncodingError EncodeImageNoHuffman(VP8LBitWriter* const bw,
     err = VP8_ENC_ERROR_OUT_OF_MEMORY;
     goto Error;
   }
+  VP8LHistogramSetClear(histogram_image);
 
   // Build histogram image and symbols from backward references.
   VP8LHistogramStoreRefs(refs, histogram_image->histograms[0]);
@@ -1248,14 +1249,20 @@ static WebPEncodingError MakeInputImageCopy(VP8LEncoder* const enc) {
   const WebPPicture* const picture = enc->pic_;
   const int width = picture->width;
   const int height = picture->height;
-  int y;
+
   err = AllocateTransformBuffer(enc, width, height);
   if (err != VP8_ENC_OK) return err;
   if (enc->argb_content_ == kEncoderARGB) return VP8_ENC_OK;
-  for (y = 0; y < height; ++y) {
-    memcpy(enc->argb_ + y * width,
-           picture->argb + y * picture->argb_stride,
-           width * sizeof(*enc->argb_));
+
+  {
+    uint32_t* dst = enc->argb_;
+    const uint32_t* src = picture->argb;
+    int y;
+    for (y = 0; y < height; ++y) {
+      memcpy(dst, src, width * sizeof(*dst));
+      dst += width;
+      src += picture->argb_stride;
+    }
   }
   enc->argb_content_ = kEncoderARGB;
   assert(enc->current_width_ == width);

thirdparty/libwebp/src/enc/vp8li_enc.h

@@ -115,4 +115,4 @@ void VP8LColorSpaceTransform(int width, int height, int bits, int quality,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_ENC_VP8LI_ENC_H_ */
+#endif  // WEBP_ENC_VP8LI_ENC_H_

thirdparty/libwebp/src/mux/animi.h

@@ -40,4 +40,4 @@ int WebPAnimEncoderRefineRect(
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_MUX_ANIMI_H_ */
+#endif  // WEBP_MUX_ANIMI_H_

thirdparty/libwebp/src/mux/muxedit.c

@@ -69,12 +69,12 @@ void WebPMuxDelete(WebPMux* mux) {
   if (idx == (INDEX)) {                                                        \
     err = ChunkAssignData(&chunk, data, copy_data, tag);                       \
     if (err == WEBP_MUX_OK) {                                                  \
-      err = ChunkSetNth(&chunk, (LIST), nth);                                  \
+      err = ChunkSetHead(&chunk, (LIST));                                      \
     }                                                                          \
     return err;                                                                \
   }
 
-static WebPMuxError MuxSet(WebPMux* const mux, uint32_t tag, uint32_t nth,
+static WebPMuxError MuxSet(WebPMux* const mux, uint32_t tag,
                            const WebPData* const data, int copy_data) {
   WebPChunk chunk;
   WebPMuxError err = WEBP_MUX_NOT_FOUND;
@@ -190,7 +190,7 @@ WebPMuxError WebPMuxSetChunk(WebPMux* mux, const char fourcc[4],
   if (err != WEBP_MUX_OK && err != WEBP_MUX_NOT_FOUND) return err;
 
   // Add the given chunk.
-  return MuxSet(mux, tag, 1, chunk_data, copy_data);
+  return MuxSet(mux, tag, chunk_data, copy_data);
 }
 
 // Creates a chunk from given 'data' and sets it as 1st chunk in 'chunk_list'.
@@ -202,7 +202,7 @@ static WebPMuxError AddDataToChunkList(
   ChunkInit(&chunk);
   err = ChunkAssignData(&chunk, data, copy_data, tag);
   if (err != WEBP_MUX_OK) goto Err;
-  err = ChunkSetNth(&chunk, chunk_list, 1);
+  err = ChunkSetHead(&chunk, chunk_list);
   if (err != WEBP_MUX_OK) goto Err;
   return WEBP_MUX_OK;
  Err:
@@ -266,14 +266,14 @@ WebPMuxError WebPMuxPushFrame(WebPMux* mux, const WebPMuxFrameInfo* info,
                               int copy_data) {
   WebPMuxImage wpi;
   WebPMuxError err;
-  const WebPData* const bitstream = &info->bitstream;
 
   // Sanity checks.
   if (mux == NULL || info == NULL) return WEBP_MUX_INVALID_ARGUMENT;
 
   if (info->id != WEBP_CHUNK_ANMF) return WEBP_MUX_INVALID_ARGUMENT;
 
-  if (bitstream->bytes == NULL || bitstream->size > MAX_CHUNK_PAYLOAD) {
+  if (info->bitstream.bytes == NULL ||
+      info->bitstream.size > MAX_CHUNK_PAYLOAD) {
     return WEBP_MUX_INVALID_ARGUMENT;
   }
 
@@ -287,7 +287,7 @@ WebPMuxError WebPMuxPushFrame(WebPMux* mux, const WebPMuxFrameInfo* info,
   }
 
   MuxImageInit(&wpi);
-  err = SetAlphaAndImageChunks(bitstream, copy_data, &wpi);
+  err = SetAlphaAndImageChunks(&info->bitstream, copy_data, &wpi);
   if (err != WEBP_MUX_OK) goto Err;
   assert(wpi.img_ != NULL);  // As SetAlphaAndImageChunks() was successful.
 
@@ -342,7 +342,7 @@ WebPMuxError WebPMuxSetAnimationParams(WebPMux* mux,
   // Set the animation parameters.
   PutLE32(data, params->bgcolor);
   PutLE16(data + 4, params->loop_count);
-  return MuxSet(mux, kChunks[IDX_ANIM].tag, 1, &anim, 1);
+  return MuxSet(mux, kChunks[IDX_ANIM].tag, &anim, 1);
 }
 
 WebPMuxError WebPMuxSetCanvasSize(WebPMux* mux,
@@ -540,7 +540,7 @@ static WebPMuxError CreateVP8XChunk(WebPMux* const mux) {
   PutLE24(data + 4, width - 1);   // canvas width.
   PutLE24(data + 7, height - 1);  // canvas height.
 
-  return MuxSet(mux, kChunks[IDX_VP8X].tag, 1, &vp8x, 1);
+  return MuxSet(mux, kChunks[IDX_VP8X].tag, &vp8x, 1);
 }
 
 // Cleans up 'mux' by removing any unnecessary chunks.

thirdparty/libwebp/src/mux/muxi.h

@@ -14,6 +14,7 @@
 #ifndef WEBP_MUX_MUXI_H_
 #define WEBP_MUX_MUXI_H_
 
+#include <assert.h>
 #include <stdlib.h>
 #include "src/dec/vp8i_dec.h"
 #include "src/dec/vp8li_dec.h"
@@ -28,7 +29,7 @@ extern "C" {
 
 #define MUX_MAJ_VERSION 1
 #define MUX_MIN_VERSION 0
-#define MUX_REV_VERSION 0
+#define MUX_REV_VERSION 1
 
 // Chunk object.
 typedef struct WebPChunk WebPChunk;
@@ -126,11 +127,14 @@ WebPChunk* ChunkSearchList(WebPChunk* first, uint32_t nth, uint32_t tag);
 WebPMuxError ChunkAssignData(WebPChunk* chunk, const WebPData* const data,
                              int copy_data, uint32_t tag);
 
-// Sets 'chunk' at nth position in the 'chunk_list'.
-// nth = 0 has the special meaning "last of the list".
+// Sets 'chunk' as the only element in 'chunk_list' if it is empty.
 // On success ownership is transferred from 'chunk' to the 'chunk_list'.
-WebPMuxError ChunkSetNth(WebPChunk* chunk, WebPChunk** chunk_list,
-                         uint32_t nth);
+WebPMuxError ChunkSetHead(WebPChunk* const chunk, WebPChunk** const chunk_list);
+// Sets 'chunk' at last position in the 'chunk_list'.
+// On success ownership is transferred from 'chunk' to the 'chunk_list'.
+// *chunk_list also points towards the last valid element of the initial
+// *chunk_list.
+WebPMuxError ChunkAppend(WebPChunk* const chunk, WebPChunk*** const chunk_list);
 
 // Releases chunk and returns chunk->next_.
 WebPChunk* ChunkRelease(WebPChunk* const chunk);
@@ -143,13 +147,13 @@ void ChunkListDelete(WebPChunk** const chunk_list);
 
 // Returns size of the chunk including chunk header and padding byte (if any).
 static WEBP_INLINE size_t SizeWithPadding(size_t chunk_size) {
+  assert(chunk_size <= MAX_CHUNK_PAYLOAD);
   return CHUNK_HEADER_SIZE + ((chunk_size + 1) & ~1U);
 }
 
 // Size of a chunk including header and padding.
 static WEBP_INLINE size_t ChunkDiskSize(const WebPChunk* chunk) {
   const size_t data_size = chunk->data_.size;
-  assert(data_size < MAX_CHUNK_PAYLOAD);
   return SizeWithPadding(data_size);
 }
 
@@ -227,4 +231,4 @@ WebPMuxError MuxValidate(const WebPMux* const mux);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_MUX_MUXI_H_ */
+#endif  // WEBP_MUX_MUXI_H_

thirdparty/libwebp/src/mux/muxinternal.c

@@ -111,27 +111,6 @@ WebPChunk* ChunkSearchList(WebPChunk* first, uint32_t nth, uint32_t tag) {
   return ((nth > 0) && (iter > 0)) ? NULL : first;
 }
 
-// Outputs a pointer to 'prev_chunk->next_',
-//   where 'prev_chunk' is the pointer to the chunk at position (nth - 1).
-// Returns true if nth chunk was found.
-static int ChunkSearchListToSet(WebPChunk** chunk_list, uint32_t nth,
-                                WebPChunk*** const location) {
-  uint32_t count = 0;
-  assert(chunk_list != NULL);
-  *location = chunk_list;
-
-  while (*chunk_list != NULL) {
-    WebPChunk* const cur_chunk = *chunk_list;
-    ++count;
-    if (count == nth) return 1;  // Found.
-    chunk_list = &cur_chunk->next_;
-    *location = chunk_list;
-  }
-
-  // *chunk_list is ok to be NULL if adding at last location.
-  return (nth == 0 || (count == nth - 1)) ? 1 : 0;
-}
-
 //------------------------------------------------------------------------------
 // Chunk writer methods.
 
@@ -156,11 +135,12 @@ WebPMuxError ChunkAssignData(WebPChunk* chunk, const WebPData* const data,
   return WEBP_MUX_OK;
 }
 
-WebPMuxError ChunkSetNth(WebPChunk* chunk, WebPChunk** chunk_list,
-                         uint32_t nth) {
+WebPMuxError ChunkSetHead(WebPChunk* const chunk,
+                          WebPChunk** const chunk_list) {
   WebPChunk* new_chunk;
 
-  if (!ChunkSearchListToSet(chunk_list, nth, &chunk_list)) {
+  assert(chunk_list != NULL);
+  if (*chunk_list != NULL) {
     return WEBP_MUX_NOT_FOUND;
   }
 
@@ -168,11 +148,26 @@ WebPMuxError ChunkSetNth(WebPChunk* chunk, WebPChunk** chunk_list,
   if (new_chunk == NULL) return WEBP_MUX_MEMORY_ERROR;
   *new_chunk = *chunk;
   chunk->owner_ = 0;
-  new_chunk->next_ = *chunk_list;
+  new_chunk->next_ = NULL;
   *chunk_list = new_chunk;
   return WEBP_MUX_OK;
 }
 
+WebPMuxError ChunkAppend(WebPChunk* const chunk,
+                         WebPChunk*** const chunk_list) {
+  assert(chunk_list != NULL && *chunk_list != NULL);
+
+  if (**chunk_list == NULL) {
+    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_;
+  }
+  return WEBP_MUX_OK;
+}
+
 //------------------------------------------------------------------------------
 // Chunk deletion method(s).
 
@@ -232,9 +227,11 @@ void MuxImageInit(WebPMuxImage* const wpi) {
 WebPMuxImage* MuxImageRelease(WebPMuxImage* const wpi) {
   WebPMuxImage* next;
   if (wpi == NULL) return NULL;
-  ChunkDelete(wpi->header_);
-  ChunkDelete(wpi->alpha_);
-  ChunkDelete(wpi->img_);
+  // There should be at most one chunk of header_, alpha_, img_ but we call
+  // ChunkListDelete to be safe
+  ChunkListDelete(&wpi->header_);
+  ChunkListDelete(&wpi->alpha_);
+  ChunkListDelete(&wpi->img_);
   ChunkListDelete(&wpi->unknown_);
 
   next = wpi->next_;

thirdparty/libwebp/src/mux/muxread.c

@@ -59,6 +59,7 @@ static WebPMuxError ChunkVerifyAndAssign(WebPChunk* chunk,
   // Sanity checks.
   if (data_size < CHUNK_HEADER_SIZE) return WEBP_MUX_NOT_ENOUGH_DATA;
   chunk_size = GetLE32(data + TAG_SIZE);
+  if (chunk_size > MAX_CHUNK_PAYLOAD) return WEBP_MUX_BAD_DATA;
 
   {
     const size_t chunk_disk_size = SizeWithPadding(chunk_size);
@@ -102,6 +103,7 @@ static int MuxImageParse(const WebPChunk* const chunk, int copy_data,
   const uint8_t* const last = bytes + size;
   WebPChunk subchunk;
   size_t subchunk_size;
+  WebPChunk** unknown_chunk_list = &wpi->unknown_;
   ChunkInit(&subchunk);
 
   assert(chunk->tag_ == kChunks[IDX_ANMF].tag);
@@ -116,7 +118,7 @@ static int MuxImageParse(const WebPChunk* const chunk, int copy_data,
     if (size < hdr_size) goto Fail;
     ChunkAssignData(&subchunk, &temp, copy_data, chunk->tag_);
   }
-  ChunkSetNth(&subchunk, &wpi->header_, 1);
+  ChunkSetHead(&subchunk, &wpi->header_);
   wpi->is_partial_ = 1;  // Waiting for ALPH and/or VP8/VP8L chunks.
 
   // Rest of the chunks.
@@ -133,18 +135,23 @@ static int MuxImageParse(const WebPChunk* const chunk, int copy_data,
     switch (ChunkGetIdFromTag(subchunk.tag_)) {
       case WEBP_CHUNK_ALPHA:
         if (wpi->alpha_ != NULL) goto Fail;  // Consecutive ALPH chunks.
-        if (ChunkSetNth(&subchunk, &wpi->alpha_, 1) != WEBP_MUX_OK) goto Fail;
+        if (ChunkSetHead(&subchunk, &wpi->alpha_) != WEBP_MUX_OK) goto Fail;
         wpi->is_partial_ = 1;  // Waiting for a VP8 chunk.
         break;
       case WEBP_CHUNK_IMAGE:
-        if (ChunkSetNth(&subchunk, &wpi->img_, 1) != WEBP_MUX_OK) goto Fail;
+        if (wpi->img_ != NULL) goto Fail;  // Only 1 image chunk allowed.
+        if (ChunkSetHead(&subchunk, &wpi->img_) != WEBP_MUX_OK) goto Fail;
         if (!MuxImageFinalize(wpi)) goto Fail;
         wpi->is_partial_ = 0;  // wpi is completely filled.
         break;
       case WEBP_CHUNK_UNKNOWN:
-        if (wpi->is_partial_) goto Fail;  // Encountered an unknown chunk
-                                          // before some image chunks.
-        if (ChunkSetNth(&subchunk, &wpi->unknown_, 0) != WEBP_MUX_OK) goto Fail;
+        if (wpi->is_partial_) {
+          goto Fail;  // Encountered an unknown chunk
+                      // before some image chunks.
+        }
+        if (ChunkAppend(&subchunk, &unknown_chunk_list) != WEBP_MUX_OK) {
+          goto Fail;
+        }
         break;
       default:
         goto Fail;
@@ -175,6 +182,9 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
   const uint8_t* data;
   size_t size;
   WebPChunk chunk;
+  // Stores the end of the chunk lists so that it is faster to append data to
+  // their ends.
+  WebPChunk** chunk_list_ends[WEBP_CHUNK_NIL + 1] = { NULL };
   ChunkInit(&chunk);
 
   // Sanity checks.
@@ -187,7 +197,7 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
   size = bitstream->size;
 
   if (data == NULL) return NULL;
-  if (size < RIFF_HEADER_SIZE) return NULL;
+  if (size < RIFF_HEADER_SIZE + CHUNK_HEADER_SIZE) return NULL;
   if (GetLE32(data + 0) != MKFOURCC('R', 'I', 'F', 'F') ||
       GetLE32(data + CHUNK_HEADER_SIZE) != MKFOURCC('W', 'E', 'B', 'P')) {
     return NULL;
@@ -196,8 +206,6 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
   mux = WebPMuxNew();
   if (mux == NULL) return NULL;
 
-  if (size < RIFF_HEADER_SIZE + TAG_SIZE) goto Err;
-
   tag = GetLE32(data + RIFF_HEADER_SIZE);
   if (tag != kChunks[IDX_VP8].tag &&
       tag != kChunks[IDX_VP8L].tag &&
@@ -205,13 +213,17 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
     goto Err;  // First chunk should be VP8, VP8L or VP8X.
   }
 
-  riff_size = SizeWithPadding(GetLE32(data + TAG_SIZE));
-  if (riff_size > MAX_CHUNK_PAYLOAD || riff_size > size) {
-    goto Err;
-  } else {
-    if (riff_size < size) {  // Redundant data after last chunk.
-      size = riff_size;  // To make sure we don't read any data beyond mux_size.
-    }
+  riff_size = GetLE32(data + TAG_SIZE);
+  if (riff_size > MAX_CHUNK_PAYLOAD) goto Err;
+
+  // Note this padding is historical and differs from demux.c which does not
+  // pad the file size.
+  riff_size = SizeWithPadding(riff_size);
+  if (riff_size < CHUNK_HEADER_SIZE) goto Err;
+  if (riff_size > size) goto Err;
+  // There's no point in reading past the end of the RIFF chunk.
+  if (size > riff_size + CHUNK_HEADER_SIZE) {
+    size = riff_size + CHUNK_HEADER_SIZE;
   }
 
   end = data + size;
@@ -226,7 +238,6 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
   while (data != end) {
     size_t data_size;
     WebPChunkId id;
-    WebPChunk** chunk_list;
     if (ChunkVerifyAndAssign(&chunk, data, size, riff_size,
                              copy_data) != WEBP_MUX_OK) {
       goto Err;
@@ -236,11 +247,11 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
     switch (id) {
       case WEBP_CHUNK_ALPHA:
         if (wpi->alpha_ != NULL) goto Err;  // Consecutive ALPH chunks.
-        if (ChunkSetNth(&chunk, &wpi->alpha_, 1) != WEBP_MUX_OK) goto Err;
+        if (ChunkSetHead(&chunk, &wpi->alpha_) != WEBP_MUX_OK) goto Err;
         wpi->is_partial_ = 1;  // Waiting for a VP8 chunk.
         break;
       case WEBP_CHUNK_IMAGE:
-        if (ChunkSetNth(&chunk, &wpi->img_, 1) != WEBP_MUX_OK) goto Err;
+        if (ChunkSetHead(&chunk, &wpi->img_) != WEBP_MUX_OK) goto Err;
         if (!MuxImageFinalize(wpi)) goto Err;
         wpi->is_partial_ = 0;  // wpi is completely filled.
  PushImage:
@@ -257,9 +268,13 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
       default:  // A non-image chunk.
         if (wpi->is_partial_) goto Err;  // Encountered a non-image chunk before
                                          // getting all chunks of an image.
-        chunk_list = MuxGetChunkListFromId(mux, id);  // List to add this chunk.
-        if (ChunkSetNth(&chunk, chunk_list, 0) != WEBP_MUX_OK) goto Err;
+        if (chunk_list_ends[id] == NULL) {
+          chunk_list_ends[id] =
+              MuxGetChunkListFromId(mux, id);  // List to add this chunk.
+        }
+        if (ChunkAppend(&chunk, &chunk_list_ends[id]) != WEBP_MUX_OK) goto Err;
         if (id == WEBP_CHUNK_VP8X) {  // grab global specs
+          if (data_size < CHUNK_HEADER_SIZE + VP8X_CHUNK_SIZE) goto Err;
           mux->canvas_width_ = GetLE24(data + 12) + 1;
           mux->canvas_height_ = GetLE24(data + 15) + 1;
         }
@@ -385,6 +400,10 @@ static WebPMuxError SynthesizeBitstream(const WebPMuxImage* const wpi,
   uint8_t* const data = (uint8_t*)WebPSafeMalloc(1ULL, size);
   if (data == NULL) return WEBP_MUX_MEMORY_ERROR;
 
+  // There should be at most one alpha_ chunk and exactly one img_ chunk.
+  assert(wpi->alpha_ == NULL || wpi->alpha_->next_ == NULL);
+  assert(wpi->img_ != NULL && wpi->img_->next_ == NULL);
+
   // Main RIFF header.
   dst = MuxEmitRiffHeader(data, size);
 

thirdparty/libwebp/src/utils/bit_reader_inl_utils.h

@@ -187,4 +187,4 @@ static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br, int prob) {
 }    // extern "C"
 #endif
 
-#endif   // WEBP_UTILS_BIT_READER_INL_UTILS_H_
+#endif  // WEBP_UTILS_BIT_READER_INL_UTILS_H_

thirdparty/libwebp/src/utils/bit_reader_utils.h

@@ -182,4 +182,4 @@ static WEBP_INLINE void VP8LFillBitWindow(VP8LBitReader* const br) {
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_BIT_READER_UTILS_H_ */
+#endif  // WEBP_UTILS_BIT_READER_UTILS_H_

thirdparty/libwebp/src/utils/bit_writer_utils.h

@@ -151,4 +151,4 @@ static WEBP_INLINE void VP8LPutBits(VP8LBitWriter* const bw,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_BIT_WRITER_UTILS_H_ */
+#endif  // WEBP_UTILS_BIT_WRITER_UTILS_H_

thirdparty/libwebp/src/utils/filters_utils.h

@@ -29,4 +29,4 @@ WEBP_FILTER_TYPE WebPEstimateBestFilter(const uint8_t* data,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_FILTERS_UTILS_H_ */
+#endif  // WEBP_UTILS_FILTERS_UTILS_H_

thirdparty/libwebp/src/utils/quant_levels_dec_utils.c

@@ -261,9 +261,15 @@ static void CleanupParams(SmoothParams* const p) {
 
 int WebPDequantizeLevels(uint8_t* const data, int width, int height, int stride,
                          int strength) {
-  const int radius = 4 * strength / 100;
+  int radius = 4 * strength / 100;
+
   if (strength < 0 || strength > 100) return 0;
   if (data == NULL || width <= 0 || height <= 0) return 0;  // bad params
+
+  // limit the filter size to not exceed the image dimensions
+  if (2 * radius + 1 > width) radius = (width - 1) >> 1;
+  if (2 * radius + 1 > height) radius = (height - 1) >> 1;
+
   if (radius > 0) {
     SmoothParams p;
     memset(&p, 0, sizeof(p));

thirdparty/libwebp/src/utils/quant_levels_dec_utils.h

@@ -32,4 +32,4 @@ int WebPDequantizeLevels(uint8_t* const data, int width, int height, int stride,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_QUANT_LEVELS_DEC_UTILS_H_ */
+#endif  // WEBP_UTILS_QUANT_LEVELS_DEC_UTILS_H_

thirdparty/libwebp/src/utils/quant_levels_utils.h

@@ -33,4 +33,4 @@ int QuantizeLevels(uint8_t* const data, int width, int height, int num_levels,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_QUANT_LEVELS_UTILS_H_ */
+#endif  // WEBP_UTILS_QUANT_LEVELS_UTILS_H_

thirdparty/libwebp/src/utils/random_utils.h

@@ -60,4 +60,4 @@ static WEBP_INLINE int VP8RandomBits(VP8Random* const rg, int num_bits) {
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_RANDOM_UTILS_H_ */
+#endif  // WEBP_UTILS_RANDOM_UTILS_H_

thirdparty/libwebp/src/utils/rescaler_utils.h

@@ -98,4 +98,4 @@ int WebPRescalerHasPendingOutput(const WebPRescaler* const rescaler) {
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_RESCALER_UTILS_H_ */
+#endif  // WEBP_UTILS_RESCALER_UTILS_H_

thirdparty/libwebp/src/utils/thread_utils.h

@@ -87,4 +87,4 @@ WEBP_EXTERN const WebPWorkerInterface* WebPGetWorkerInterface(void);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_THREAD_UTILS_H_ */
+#endif  // WEBP_UTILS_THREAD_UTILS_H_

thirdparty/libwebp/src/utils/utils.h

@@ -175,4 +175,4 @@ WEBP_EXTERN int WebPGetColorPalette(const struct WebPPicture* const pic,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_UTILS_UTILS_H_ */
+#endif  // WEBP_UTILS_UTILS_H_

thirdparty/libwebp/src/webp/decode.h

@@ -491,4 +491,4 @@ WEBP_EXTERN VP8StatusCode WebPDecode(const uint8_t* data, size_t data_size,
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_WEBP_DECODE_H_ */
+#endif  // WEBP_WEBP_DECODE_H_

thirdparty/libwebp/src/webp/demux.h

@@ -360,4 +360,4 @@ WEBP_EXTERN void WebPAnimDecoderDelete(WebPAnimDecoder* dec);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_WEBP_DEMUX_H_ */
+#endif  // WEBP_WEBP_DEMUX_H_

thirdparty/libwebp/src/webp/encode.h

@@ -542,4 +542,4 @@ WEBP_EXTERN int WebPEncode(const WebPConfig* config, WebPPicture* picture);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_WEBP_ENCODE_H_ */
+#endif  // WEBP_WEBP_ENCODE_H_

thirdparty/libwebp/src/webp/format_constants.h

@@ -84,4 +84,4 @@ typedef enum {
 // overflow a uint32_t.
 #define MAX_CHUNK_PAYLOAD (~0U - CHUNK_HEADER_SIZE - 1)
 
-#endif  /* WEBP_WEBP_FORMAT_CONSTANTS_H_ */
+#endif  // WEBP_WEBP_FORMAT_CONSTANTS_H_

thirdparty/libwebp/src/webp/mux.h

@@ -527,4 +527,4 @@ WEBP_EXTERN void WebPAnimEncoderDelete(WebPAnimEncoder* enc);
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_WEBP_MUX_H_ */
+#endif  // WEBP_WEBP_MUX_H_

thirdparty/libwebp/src/webp/mux_types.h

@@ -95,4 +95,4 @@ static WEBP_INLINE int WebPDataCopy(const WebPData* src, WebPData* dst) {
 }    // extern "C"
 #endif
 
-#endif  /* WEBP_WEBP_MUX_TYPES_H_ */
+#endif  // WEBP_WEBP_MUX_TYPES_H_

thirdparty/libwebp/src/webp/types.h

@@ -49,4 +49,4 @@ typedef long long int int64_t;
 // Macro to check ABI compatibility (same major revision number)
 #define WEBP_ABI_IS_INCOMPATIBLE(a, b) (((a) >> 8) != ((b) >> 8))
 
-#endif  /* WEBP_WEBP_TYPES_H_ */
+#endif  // WEBP_WEBP_TYPES_H_