GetBackwardReferences: fail on alloc error

previously failures in the call to
VP8LBackwardReferencesTraceBackwards() would be ignored which, though it
wouldn't result in a crash, would produce non-deterministic output

Change-Id: Id9890a60883c3270ec75e968506d46eea32b76d4
(cherry picked from commit e3cfafaf71)
(cherry picked from commit 20ef03ee35)
(cherry picked from commit 89e226a3c7)

examples/anim_diff.c

@@ -143,8 +143,18 @@ static int CompareAnimatedImagePair(const AnimatedImage* const img1,
   if (!ok) return 0;  // These are fatal failures, can't proceed.
 
   if (is_multi_frame_image) {  // Checks relevant for multi-frame images only.
-    ok = CompareValues(img1->loop_count, img2->loop_count,
+    int max_loop_count_workaround = 0;
-                       "Loop count mismatch") && ok;
+    // Transcodes to webp increase the gif loop count by 1 for compatibility.
+    // When the gif has the maximum value the webp value will be off by one.
+    if ((img1->format == ANIM_GIF && img1->loop_count == 65536 &&
+         img2->format == ANIM_WEBP && img2->loop_count == 65535) ||
+        (img1->format == ANIM_WEBP && img1->loop_count == 65535 &&
+         img2->format == ANIM_GIF && img2->loop_count == 65536)) {
+      max_loop_count_workaround = 1;
+    }
+    ok = (max_loop_count_workaround ||
+          CompareValues(img1->loop_count, img2->loop_count,
+                        "Loop count mismatch")) && ok;
     ok = CompareBackgroundColor(img1->bgcolor, img2->bgcolor,
                                 premultiply) && ok;
   }

examples/anim_util.c

@@ -275,6 +275,7 @@ static int ReadAnimatedWebP(const char filename[],
     prev_frame_timestamp = timestamp;
   }
   ok = dump_ok;
+  if (ok) image->format = ANIM_WEBP;
 
  End:
   WebPAnimDecoderDelete(dec);
@@ -687,6 +688,7 @@ static int ReadAnimatedGIF(const char filename[], AnimatedImage* const image,
       }
     }
   }
+  image->format = ANIM_GIF;
   DGifCloseFile(gif, NULL);
   return 1;
 }

examples/anim_util.h

@@ -22,6 +22,11 @@
 extern "C" {
 #endif
 
+typedef enum {
+  ANIM_GIF,
+  ANIM_WEBP
+} AnimatedFileFormat;
+
 typedef struct {
   uint8_t* rgba;         // Decoded and reconstructed full frame.
   int duration;          // Frame duration in milliseconds.
@@ -29,6 +34,7 @@ typedef struct {
 } DecodedFrame;
 
 typedef struct {
+  AnimatedFileFormat format;
   uint32_t canvas_width;
   uint32_t canvas_height;
   uint32_t bgcolor;

examples/gif2webp.c

@@ -478,7 +478,7 @@ int main(int argc, const char *argv[]) {
         stored_loop_count = 1;
         loop_count = 1;
       }
-    } else if (loop_count > 0) {
+    } else if (loop_count > 0 && loop_count < 65535) {
       // adapt GIF's semantic to WebP's (except in the infinite-loop case)
       loop_count += 1;
     }

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);

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_;
+    // Synchronize the thread, clean-up and check for errors.
-    if (io->teardown != NULL) {
+    VP8ExitCritical((VP8Decoder*)idec->dec_, &idec->io_);
-      io->teardown(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)) {
@@ -491,6 +491,7 @@ 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 +572,10 @@ 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) {

src/dec/vp8l_dec.c

@@ -884,7 +884,11 @@ 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 +1527,6 @@ 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 +1558,12 @@ 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_);
+  VP8LDelete(dec);
-  alph_dec->vp8l_dec_ = NULL;
   return 0;
 }
 

src/dsp/alpha_processing_neon.c

@@ -83,7 +83,7 @@ static void ApplyAlphaMultiply_NEON(uint8_t* rgba, int alpha_first,
 static int DispatchAlpha_NEON(const uint8_t* alpha, int alpha_stride,
                               int width, int height,
                               uint8_t* dst, int dst_stride) {
-  uint32_t alpha_mask = 0xffffffffu;
+  uint32_t alpha_mask = 0xffu;
   uint8x8_t mask8 = vdup_n_u8(0xff);
   uint32_t tmp[2];
   int i, j;
@@ -107,6 +107,7 @@ static int DispatchAlpha_NEON(const uint8_t* alpha, int alpha_stride,
     dst += dst_stride;
   }
   vst1_u8((uint8_t*)tmp, mask8);
+  alpha_mask *= 0x01010101;
   alpha_mask &= tmp[0];
   alpha_mask &= tmp[1];
   return (alpha_mask != 0xffffffffu);
@@ -134,7 +135,7 @@ static void DispatchAlphaToGreen_NEON(const uint8_t* alpha, int alpha_stride,
 static int ExtractAlpha_NEON(const uint8_t* argb, int argb_stride,
                              int width, int height,
                              uint8_t* alpha, int alpha_stride) {
-  uint32_t alpha_mask = 0xffffffffu;
+  uint32_t alpha_mask = 0xffu;
   uint8x8_t mask8 = vdup_n_u8(0xff);
   uint32_t tmp[2];
   int i, j;
@@ -156,6 +157,7 @@ static int ExtractAlpha_NEON(const uint8_t* argb, int argb_stride,
     alpha += alpha_stride;
   }
   vst1_u8((uint8_t*)tmp, mask8);
+  alpha_mask *= 0x01010101;
   alpha_mask &= tmp[0];
   alpha_mask &= tmp[1];
   return (alpha_mask == 0xffffffffu);

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
@@ -108,8 +109,7 @@ void WebPRescalerExportRowExpand_C(WebPRescaler* const wrk) {
     for (x_out = 0; x_out < x_out_max; ++x_out) {
       const uint32_t J = frow[x_out];
       const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
     }
   } else {
     const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
@@ -119,8 +119,7 @@ void WebPRescalerExportRowExpand_C(WebPRescaler* const wrk) {
                        + (uint64_t)B * irow[x_out];
       const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
       const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
     }
   }
 }
@@ -137,22 +136,21 @@ void WebPRescalerExportRowShrink_C(WebPRescaler* const wrk) {
   assert(!wrk->y_expand);
   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 uint32_t frac = (uint32_t)MULT_FIX_FLOOR(frow[x_out], yscale);
       const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
       irow[x_out] = frac;   // new fractional start
     }
   } else {
     for (x_out = 0; x_out < x_out_max; ++x_out) {
       const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
       irow[x_out] = 0;
     }
   }
 }
 
+#undef MULT_FIX_FLOOR
 #undef MULT_FIX
 #undef ROUNDER
 

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

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;
@@ -105,10 +107,9 @@ 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 uint32_t frac = (uint32_t)MULT_FIX_FLOOR(*frow++, yscale);
       const int v = (int)MULT_FIX(*irow - frac, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
+      *dst++ = (v > 255) ? 255u : (uint8_t)v;
-      *dst++ = v;
       *irow++ = frac;   // new fractional start
     }
   } else {
@@ -154,13 +155,13 @@ 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 > 255) ? 255u : (uint8_t)v;
-      *dst++ = v;
       *irow++ = 0;
     }
   }
 }
+#endif  // 0
 
 static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
   int i;
@@ -216,8 +217,7 @@ static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
     for (i = 0; i < (x_out_max & 0x3); ++i) {
       const uint32_t J = *frow++;
       const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
+      *dst++ = (v > 255) ? 255u : (uint8_t)v;
-      *dst++ = v;
     }
   } else {
     const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
@@ -288,12 +288,12 @@ static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
                        + (uint64_t)B * *irow++;
       const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
       const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
+      *dst++ = (v > 255) ? 255u : (uint8_t)v;
-      *dst++ = v;
     }
   }
 }
 
+#undef MULT_FIX_FLOOR
 #undef MULT_FIX
 #undef ROUNDER
 
@@ -304,7 +304,7 @@ 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

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;                                       \
@@ -165,8 +166,7 @@ static WEBP_INLINE void ExportRowExpand_0(const uint32_t* frow, uint8_t* dst,
     for (x_out = 0; x_out < length; ++x_out) {
       const uint32_t J = frow[x_out];
       const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
     }
   }
 }
@@ -240,8 +240,7 @@ static WEBP_INLINE void ExportRowExpand_1(const uint32_t* frow, uint32_t* irow,
                        + (uint64_t)B * irow[x_out];
       const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
       const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
     }
   }
 }
@@ -262,6 +261,7 @@ 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,
@@ -340,10 +340,9 @@ static WEBP_INLINE void ExportRowShrink_0(const uint32_t* frow, uint32_t* irow,
       length -= 4;
     }
     for (x_out = 0; x_out < length; ++x_out) {
-      const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
+      const uint32_t frac = (uint32_t)MULT_FIX_FLOOR(frow[x_out], yscale);
       const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
       irow[x_out] = frac;
     }
   }
@@ -404,8 +403,7 @@ static WEBP_INLINE void ExportRowShrink_1(uint32_t* irow, uint8_t* dst,
     }
     for (x_out = 0; x_out < length; ++x_out) {
       const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
       irow[x_out] = 0;
     }
   }
@@ -426,6 +424,7 @@ static void RescalerExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
     ExportRowShrink_1(irow, dst, x_out_max, wrk);
   }
 }
+#endif  // 0
 
 //------------------------------------------------------------------------------
 // Entry point
@@ -434,7 +433,7 @@ extern void WebPRescalerDspInitMSA(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMSA(void) {
   WebPRescalerExportRowExpand = RescalerExportRowExpand_MIPSdspR2;
-  WebPRescalerExportRowShrink = RescalerExportRowShrink_MIPSdspR2;
+//  WebPRescalerExportRowShrink = RescalerExportRowShrink_MIPSdspR2;
 }
 
 #else     // !WEBP_USE_MSA

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
@@ -77,14 +81,13 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
       const uint32x4_t B1 = MULT_FIX(A1, fy_scale_half);
       const uint16x4_t C0 = vmovn_u32(B0);
       const uint16x4_t C1 = vmovn_u32(B1);
-      const uint8x8_t D = vmovn_u16(vcombine_u16(C0, C1));
+      const uint8x8_t D = vqmovn_u16(vcombine_u16(C0, C1));
       vst1_u8(dst + x_out, D);
     }
     for (; x_out < x_out_max; ++x_out) {
       const uint32_t J = frow[x_out];
       const int v = (int)MULT_FIX_C(J, fy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
     }
   } else {
     const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
@@ -98,7 +101,7 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
       const uint32x4_t D1 = MULT_FIX(C1, fy_scale_half);
       const uint16x4_t E0 = vmovn_u32(D0);
       const uint16x4_t E1 = vmovn_u32(D1);
-      const uint8x8_t F = vmovn_u16(vcombine_u16(E0, E1));
+      const uint8x8_t F = vqmovn_u16(vcombine_u16(E0, E1));
       vst1_u8(dst + x_out, F);
     }
     for (; x_out < x_out_max; ++x_out) {
@@ -106,8 +109,7 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
                        + (uint64_t)B * irow[x_out];
       const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
       const int v = (int)MULT_FIX_C(J, fy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
     }
   }
 }
@@ -131,23 +133,22 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
     for (x_out = 0; x_out < max_span; x_out += 8) {
       LOAD_32x8(frow + x_out, in0, in1);
       LOAD_32x8(irow + x_out, in2, in3);
-      const uint32x4_t A0 = MULT_FIX(in0, yscale_half);
+      const uint32x4_t A0 = MULT_FIX_FLOOR(in0, yscale_half);
-      const uint32x4_t A1 = MULT_FIX(in1, yscale_half);
+      const uint32x4_t A1 = MULT_FIX_FLOOR(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 uint16x4_t D0 = vmovn_u32(C0);
       const uint16x4_t D1 = vmovn_u32(C1);
-      const uint8x8_t E = vmovn_u16(vcombine_u16(D0, D1));
+      const uint8x8_t E = vqmovn_u16(vcombine_u16(D0, D1));
       vst1_u8(dst + x_out, E);
       STORE_32x8(A0, A1, irow + x_out);
     }
     for (; x_out < x_out_max; ++x_out) {
-      const uint32_t frac = (uint32_t)MULT_FIX_C(frow[x_out], yscale);
+      const uint32_t frac = (uint32_t)MULT_FIX_FLOOR_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_C(irow[x_out] - frac, fxy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
       irow[x_out] = frac;   // new fractional start
     }
   } else {
@@ -157,19 +158,24 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
       const uint32x4_t A1 = MULT_FIX(in1, fxy_scale_half);
       const uint16x4_t B0 = vmovn_u32(A0);
       const uint16x4_t B1 = vmovn_u32(A1);
-      const uint8x8_t C = vmovn_u16(vcombine_u16(B0, B1));
+      const uint8x8_t C = vqmovn_u16(vcombine_u16(B0, B1));
       vst1_u8(dst + x_out, C);
       STORE_32x8(zero, zero, irow + x_out);
     }
     for (; x_out < x_out_max; ++x_out) {
       const int v = (int)MULT_FIX_C(irow[x_out], fxy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
       irow[x_out] = 0;
     }
   }
 }
 
+#undef MULT_FIX_FLOOR_C
+#undef MULT_FIX_C
+#undef MULT_FIX_FLOOR
+#undef MULT_FIX
+#undef ROUNDER
+
 //------------------------------------------------------------------------------
 
 extern void WebPRescalerDspInitNEON(void);

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) {
@@ -244,8 +245,7 @@ static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
     for (; x_out < x_out_max; ++x_out) {
       const uint32_t J = frow[x_out];
       const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
     }
   } else {
     const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
@@ -278,8 +278,7 @@ static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
                        + (uint64_t)B * irow[x_out];
       const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
       const int v = (int)MULT_FIX(J, wrk->fy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
     }
   }
 }
@@ -298,20 +297,15 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
     const int scale_xy = wrk->fxy_scale;
     const __m128i mult_xy = _mm_set_epi32(0, scale_xy, 0, scale_xy);
     const __m128i mult_y = _mm_set_epi32(0, yscale, 0, yscale);
-    const __m128i rounder = _mm_set_epi32(0, ROUNDER, 0, ROUNDER);
     for (x_out = 0; x_out + 8 <= x_out_max; x_out += 8) {
       __m128i A0, A1, A2, A3, B0, B1, B2, B3;
       LoadDispatchAndMult_SSE2(irow + x_out, NULL, &A0, &A1, &A2, &A3);
       LoadDispatchAndMult_SSE2(frow + x_out, &mult_y, &B0, &B1, &B2, &B3);
       {
-        const __m128i C0 = _mm_add_epi64(B0, rounder);
+        const __m128i D0 = _mm_srli_epi64(B0, WEBP_RESCALER_RFIX);   // = frac
-        const __m128i C1 = _mm_add_epi64(B1, rounder);
+        const __m128i D1 = _mm_srli_epi64(B1, WEBP_RESCALER_RFIX);
-        const __m128i C2 = _mm_add_epi64(B2, rounder);
+        const __m128i D2 = _mm_srli_epi64(B2, WEBP_RESCALER_RFIX);
-        const __m128i C3 = _mm_add_epi64(B3, rounder);
+        const __m128i D3 = _mm_srli_epi64(B3, WEBP_RESCALER_RFIX);
-        const __m128i D0 = _mm_srli_epi64(C0, WEBP_RESCALER_RFIX);   // = frac
-        const __m128i D1 = _mm_srli_epi64(C1, WEBP_RESCALER_RFIX);
-        const __m128i D2 = _mm_srli_epi64(C2, WEBP_RESCALER_RFIX);
-        const __m128i D3 = _mm_srli_epi64(C3, WEBP_RESCALER_RFIX);
         const __m128i E0 = _mm_sub_epi64(A0, D0);   // irow[x] - frac
         const __m128i E1 = _mm_sub_epi64(A1, D1);
         const __m128i E2 = _mm_sub_epi64(A2, D2);
@@ -326,10 +320,9 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
       }
     }
     for (; x_out < x_out_max; ++x_out) {
-      const uint32_t frac = (int)MULT_FIX(frow[x_out], yscale);
+      const uint32_t frac = (int)MULT_FIX_FLOOR(frow[x_out], yscale);
       const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
       irow[x_out] = frac;   // new fractional start
     }
   } else {
@@ -345,13 +338,13 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
     }
     for (; x_out < x_out_max; ++x_out) {
       const int v = (int)MULT_FIX(irow[x_out], scale);
-      assert(v >= 0 && v <= 255);
+      dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
-      dst[x_out] = v;
       irow[x_out] = 0;
     }
   }
 }
 
+#undef MULT_FIX_FLOOR
 #undef MULT_FIX
 #undef ROUNDER
 

src/enc/backward_references_cost_enc.c

@@ -577,7 +577,7 @@ static int BackwardReferencesHashChainDistanceOnly(
       (CostModel*)WebPSafeCalloc(1ULL, cost_model_size);
   VP8LColorCache hashers;
   CostManager* cost_manager =
-      (CostManager*)WebPSafeMalloc(1ULL, sizeof(*cost_manager));
+      (CostManager*)WebPSafeCalloc(1ULL, sizeof(*cost_manager));
   int offset_prev = -1, len_prev = -1;
   double offset_cost = -1;
   int first_offset_is_constant = -1;  // initialized with 'impossible' value

src/enc/backward_references_enc.c

@@ -910,13 +910,14 @@ static VP8LBackwardRefs* GetBackwardReferences(
       quality >= 25) {
     const VP8LHashChain* const hash_chain_tmp =
         (lz77_type_best == kLZ77Standard) ? hash_chain : &hash_chain_box;
-    if (VP8LBackwardReferencesTraceBackwards(width, height, argb, *cache_bits,
+    double bit_cost_trace;
-                                             hash_chain_tmp, best, worst)) {
+    if (!VP8LBackwardReferencesTraceBackwards(width, height, argb, *cache_bits,
-      double bit_cost_trace;
+                                              hash_chain_tmp, best, worst)) {
-      VP8LHistogramCreate(histo, worst, *cache_bits);
+      goto Error;
-      bit_cost_trace = VP8LHistogramEstimateBits(histo);
-      if (bit_cost_trace < bit_cost_best) best = worst;
     }
+    VP8LHistogramCreate(histo, worst, *cache_bits);
+    bit_cost_trace = VP8LHistogramEstimateBits(histo);
+    if (bit_cost_trace < bit_cost_best) best = worst;
   }
 
   BackwardReferences2DLocality(width, best);

src/enc/vp8l_enc.c

@@ -1685,11 +1685,16 @@ WebPEncodingError VP8LEncodeStream(const WebPConfig* const config,
   const WebPWorkerInterface* const worker_interface = WebPGetWorkerInterface();
   int ok_main;
 
+  if (enc_main == NULL || !VP8LBitWriterInit(&bw_side, 0)) {
+    WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY);
+    VP8LEncoderDelete(enc_main);
+    return 0;
+  }
+
   // Analyze image (entropy, num_palettes etc)
-  if (enc_main == NULL ||
+  if (!EncoderAnalyze(enc_main, crunch_configs, &num_crunch_configs_main,
-      !EncoderAnalyze(enc_main, crunch_configs, &num_crunch_configs_main,
                       &red_and_blue_always_zero) ||
-      !EncoderInit(enc_main) || !VP8LBitWriterInit(&bw_side, 0)) {
+      !EncoderInit(enc_main)) {
     err = VP8_ENC_ERROR_OUT_OF_MEMORY;
     goto Error;
   }

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"
@@ -143,13 +144,13 @@ 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);
 }
 

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);
@@ -137,6 +138,7 @@ static int MuxImageParse(const WebPChunk* const chunk, int copy_data,
         wpi->is_partial_ = 1;  // Waiting for a VP8 chunk.
         break;
       case WEBP_CHUNK_IMAGE:
+        if (wpi->img_ != NULL) goto Fail;  // Only 1 image chunk allowed.
         if (ChunkSetNth(&subchunk, &wpi->img_, 1) != WEBP_MUX_OK) goto Fail;
         if (!MuxImageFinalize(wpi)) goto Fail;
         wpi->is_partial_ = 0;  // wpi is completely filled.
@@ -187,7 +189,7 @@ 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 +198,6 @@ 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 +205,17 @@ 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));
+  riff_size = GetLE32(data + TAG_SIZE);
-  if (riff_size > MAX_CHUNK_PAYLOAD || riff_size > size) {
+  if (riff_size > MAX_CHUNK_PAYLOAD) goto Err;
-    goto Err;
+
-  } else {
+  // Note this padding is historical and differs from demux.c which does not
-    if (riff_size < size) {  // Redundant data after last chunk.
+  // pad the file size.
-      size = riff_size;  // To make sure we don't read any data beyond mux_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;
@@ -260,6 +264,7 @@ WebPMux* WebPMuxCreateInternal(const WebPData* bitstream, int copy_data,
         chunk_list = MuxGetChunkListFromId(mux, id);  // List to add this chunk.
         if (ChunkSetNth(&chunk, chunk_list, 0) != 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;
         }

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));