libwebp: update to 0.5.0

third_party/libwebp/dsp/enc.c

 // Copyright 2011 Google Inc. All Rights Reserved.
 //
 // Use of this source code is governed by a BSD-style license
 // that can be found in the COPYING file in the root of the source
 // tree. An additional intellectual property rights grant can be found
 // in the file PATENTS. All contributing project authors may
 // be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // Speed-critical encoding functions.
@@ skipping 22 matching lines @@
   // Luma
   0 +  0 * BPS,  4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS,
   0 +  4 * BPS,  4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS,
   0 +  8 * BPS,  4 +  8 * BPS, 8 +  8 * BPS, 12 +  8 * BPS,
   0 + 12 * BPS,  4 + 12 * BPS, 8 + 12 * BPS, 12 + 12 * BPS,
 
   0 + 0 * BPS,   4 + 0 * BPS, 0 + 4 * BPS,  4 + 4 * BPS,    // U
   8 + 0 * BPS,  12 + 0 * BPS, 8 + 4 * BPS, 12 + 4 * BPS     // V
 };
 
+// general-purpose util function
+void VP8SetHistogramData(const int distribution[MAX_COEFF_THRESH + 1],
+                         VP8Histogram* const histo) {
+  int max_value = 0, last_non_zero = 1;
+  int k;
+  for (k = 0; k <= MAX_COEFF_THRESH; ++k) {
+    const int value = distribution[k];
+    if (value > 0) {
+      if (value > max_value) max_value = value;
+      last_non_zero = k;
+    }
+  }
+  histo->max_value = max_value;
+  histo->last_non_zero = last_non_zero;
+}
+
 static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
                              int start_block, int end_block,
                              VP8Histogram* const histo) {
   int j;
+  int distribution[MAX_COEFF_THRESH + 1] = { 0 };
   for (j = start_block; j < end_block; ++j) {
     int k;
     int16_t out[16];
 
     VP8FTransform(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
 
     // Convert coefficients to bin.
     for (k = 0; k < 16; ++k) {
       const int v = abs(out[k]) >> 3;  // TODO(skal): add rounding?
       const int clipped_value = clip_max(v, MAX_COEFF_THRESH);
-      histo->distribution[clipped_value]++;
+      ++distribution[clipped_value];
     }
   }
+  VP8SetHistogramData(distribution, histo);
 }
 
 //------------------------------------------------------------------------------
 // run-time tables (~4k)
 
 static uint8_t clip1[255 + 510 + 1];    // clips [-255,510] to [0,255]
 
 // We declare this variable 'volatile' to prevent instruction reordering
 // and make sure it's set to true _last_ (so as to be thread-safe)
 static volatile int tables_ok = 0;
 
-static void InitTables(void) {
+static WEBP_TSAN_IGNORE_FUNCTION void InitTables(void) {
   if (!tables_ok) {
     int i;
     for (i = -255; i <= 255 + 255; ++i) {
       clip1[255 + i] = clip_8b(i);
     }
     tables_ok = 1;
   }
 }
 
 
@@ skipping 70 matching lines @@
     const int a1 = (tmp[4 + i] + tmp[ 8 + i]);
     const int a2 = (tmp[4 + i] - tmp[ 8 + i]);
     const int a3 = (tmp[0 + i] - tmp[12 + i]);
     out[0 + i] = (a0 + a1 + 7) >> 4;            // 12b
     out[4 + i] = ((a2 * 2217 + a3 * 5352 + 12000) >> 16) + (a3 != 0);
     out[8 + i] = (a0 - a1 + 7) >> 4;
     out[12+ i] = ((a3 * 2217 - a2 * 5352 + 51000) >> 16);
   }
 }
 
+static void FTransform2(const uint8_t* src, const uint8_t* ref, int16_t* out) {
+  VP8FTransform(src, ref, out);
+  VP8FTransform(src + 4, ref + 4, out + 16);
+}
+
 static void FTransformWHT(const int16_t* in, int16_t* out) {
   // input is 12b signed
   int32_t tmp[16];
   int i;
   for (i = 0; i < 4; ++i, in += 64) {
     const int a0 = (in[0 * 16] + in[2 * 16]);  // 13b
     const int a1 = (in[1 * 16] + in[3 * 16]);
     const int a2 = (in[1 * 16] - in[3 * 16]);
     const int a3 = (in[0 * 16] - in[2 * 16]);
     tmp[0 + i * 4] = a0 + a1;   // 14b
@@ skipping 16 matching lines @@
     out[12 + i] = b3 >> 1;
   }
 }
 
 #undef MUL
 #undef STORE
 
 //------------------------------------------------------------------------------
 // Intra predictions
 
-#define DST(x, y) dst[(x) + (y) * BPS]
-
 static WEBP_INLINE void Fill(uint8_t* dst, int value, int size) {
   int j;
   for (j = 0; j < size; ++j) {
     memset(dst + j * BPS, value, size);
   }
 }
 
 static WEBP_INLINE void VerticalPred(uint8_t* dst,
                                      const uint8_t* top, int size) {
   int j;
-  if (top) {
+  if (top != NULL) {
     for (j = 0; j < size; ++j) memcpy(dst + j * BPS, top, size);
   } else {
     Fill(dst, 127, size);
   }
 }
 
 static WEBP_INLINE void HorizontalPred(uint8_t* dst,
                                        const uint8_t* left, int size) {
-  if (left) {
+  if (left != NULL) {
     int j;
     for (j = 0; j < size; ++j) {
       memset(dst + j * BPS, left[j], size);
     }
   } else {
     Fill(dst, 129, size);
   }
 }
 
 static WEBP_INLINE void TrueMotion(uint8_t* dst, const uint8_t* left,
                                    const uint8_t* top, int size) {
   int y;
-  if (left) {
-    if (top) {
+  if (left != NULL) {
+    if (top != NULL) {
       const uint8_t* const clip = clip1 + 255 - left[-1];
       for (y = 0; y < size; ++y) {
         const uint8_t* const clip_table = clip + left[y];
         int x;
         for (x = 0; x < size; ++x) {
           dst[x] = clip_table[top[x]];
         }
         dst += BPS;
       }
     } else {
       HorizontalPred(dst, left, size);
     }
   } else {
     // true motion without left samples (hence: with default 129 value)
     // is equivalent to VE prediction where you just copy the top samples.
     // Note that if top samples are not available, the default value is
     // then 129, and not 127 as in the VerticalPred case.
-    if (top) {
+    if (top != NULL) {
       VerticalPred(dst, top, size);
     } else {
       Fill(dst, 129, size);
     }
   }
 }
 
 static WEBP_INLINE void DCMode(uint8_t* dst, const uint8_t* left,
                                const uint8_t* top,
                                int size, int round, int shift) {
   int DC = 0;
   int j;
-  if (top) {
+  if (top != NULL) {
     for (j = 0; j < size; ++j) DC += top[j];
-    if (left) {   // top and left present
+    if (left != NULL) {   // top and left present
       for (j = 0; j < size; ++j) DC += left[j];
     } else {      // top, but no left
       DC += DC;
     }
     DC = (DC + round) >> shift;
-  } else if (left) {   // left but no top
+  } else if (left != NULL) {   // left but no top
     for (j = 0; j < size; ++j) DC += left[j];
     DC += DC;
     DC = (DC + round) >> shift;
   } else {   // no top, no left, nothing.
     DC = 0x80;
   }
   Fill(dst, DC, size);
 }
 
 //------------------------------------------------------------------------------
 // Chroma 8x8 prediction (paragraph 12.2)
 
 static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
                              const uint8_t* top) {
   // U block
   DCMode(C8DC8 + dst, left, top, 8, 8, 4);
   VerticalPred(C8VE8 + dst, top, 8);
   HorizontalPred(C8HE8 + dst, left, 8);
   TrueMotion(C8TM8 + dst, left, top, 8);
   // V block
   dst += 8;
-  if (top) top += 8;
-  if (left) left += 16;
+  if (top != NULL) top += 8;
+  if (left != NULL) left += 16;
   DCMode(C8DC8 + dst, left, top, 8, 8, 4);
   VerticalPred(C8VE8 + dst, top, 8);
   HorizontalPred(C8HE8 + dst, left, 8);
   TrueMotion(C8TM8 + dst, left, top, 8);
 }
 
 //------------------------------------------------------------------------------
 // luma 16x16 prediction (paragraph 12.3)
 
 static void Intra16Preds(uint8_t* dst,
                          const uint8_t* left, const uint8_t* top) {
   DCMode(I16DC16 + dst, left, top, 16, 16, 5);
   VerticalPred(I16VE16 + dst, top, 16);
   HorizontalPred(I16HE16 + dst, left, 16);
   TrueMotion(I16TM16 + dst, left, top, 16);
 }
 
 //------------------------------------------------------------------------------
 // luma 4x4 prediction
 
+#define DST(x, y) dst[(x) + (y) * BPS]
 #define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
 
 static void VE4(uint8_t* dst, const uint8_t* top) {    // vertical
   const uint8_t vals[4] = {
     AVG3(top[-1], top[0], top[1]),
     AVG3(top[ 0], top[1], top[2]),
     AVG3(top[ 1], top[2], top[3]),
     AVG3(top[ 2], top[3], top[4])
   };
   int i;
   for (i = 0; i < 4; ++i) {
     memcpy(dst + i * BPS, vals, 4);
   }
 }
 
 static void HE4(uint8_t* dst, const uint8_t* top) {    // horizontal
   const int X = top[-1];
   const int I = top[-2];
   const int J = top[-3];
   const int K = top[-4];
   const int L = top[-5];
-  *(uint32_t*)(dst + 0 * BPS) = 0x01010101U * AVG3(X, I, J);
-  *(uint32_t*)(dst + 1 * BPS) = 0x01010101U * AVG3(I, J, K);
-  *(uint32_t*)(dst + 2 * BPS) = 0x01010101U * AVG3(J, K, L);
-  *(uint32_t*)(dst + 3 * BPS) = 0x01010101U * AVG3(K, L, L);
+  WebPUint32ToMem(dst + 0 * BPS, 0x01010101U * AVG3(X, I, J));
+  WebPUint32ToMem(dst + 1 * BPS, 0x01010101U * AVG3(I, J, K));
+  WebPUint32ToMem(dst + 2 * BPS, 0x01010101U * AVG3(J, K, L));
+  WebPUint32ToMem(dst + 3 * BPS, 0x01010101U * AVG3(K, L, L));
 }
 
 static void DC4(uint8_t* dst, const uint8_t* top) {
   uint32_t dc = 4;
   int i;
   for (i = 0; i < 4; ++i) dc += top[i] + top[-5 + i];
   Fill(dst, dc >> 3, 4);
 }
 
 static void RD4(uint8_t* dst, const uint8_t* top) {
@@ skipping 266 matching lines @@
       out[n] = level;
       if (level) last = n;
     } else {
       out[n] = 0;
       in[j] = 0;
     }
   }
   return (last >= 0);
 }
 
+static int Quantize2Blocks(int16_t in[32], int16_t out[32],
+                           const VP8Matrix* const mtx) {
+  int nz;
+  nz  = VP8EncQuantizeBlock(in + 0 * 16, out + 0 * 16, mtx) << 0;
+  nz |= VP8EncQuantizeBlock(in + 1 * 16, out + 1 * 16, mtx) << 1;
+  return nz;
+}
+
 static int QuantizeBlockWHT(int16_t in[16], int16_t out[16],
                             const VP8Matrix* const mtx) {
   int n, last = -1;
   for (n = 0; n < 16; ++n) {
     const int j = kZigzag[n];
     const int sign = (in[j] < 0);
     const uint32_t coeff = sign ? -in[j] : in[j];
     assert(mtx->sharpen_[j] == 0);
     if (coeff > mtx->zthresh_[j]) {
       const uint32_t Q = mtx->q_[j];
       const uint32_t iQ = mtx->iq_[j];
       const uint32_t B = mtx->bias_[j];
       int level = QUANTDIV(coeff, iQ, B);
       if (level > MAX_LEVEL) level = MAX_LEVEL;
       if (sign) level = -level;
       in[j] = level * Q;
       out[n] = level;
       if (level) last = n;
     } else {
       out[n] = 0;
       in[j] = 0;
     }
   }
   return (last >= 0);
 }
 
 //------------------------------------------------------------------------------
 // Block copy
 
-static WEBP_INLINE void Copy(const uint8_t* src, uint8_t* dst, int size) {
+static WEBP_INLINE void Copy(const uint8_t* src, uint8_t* dst, int w, int h) {
   int y;
-  for (y = 0; y < size; ++y) {
-    memcpy(dst, src, size);
+  for (y = 0; y < h; ++y) {
+    memcpy(dst, src, w);
     src += BPS;
     dst += BPS;
   }
 }
 
-static void Copy4x4(const uint8_t* src, uint8_t* dst) { Copy(src, dst, 4); }
+static void Copy4x4(const uint8_t* src, uint8_t* dst) {
+  Copy(src, dst, 4, 4);
+}
+
+static void Copy16x8(const uint8_t* src, uint8_t* dst) {
+  Copy(src, dst, 16, 8);
+}
 
 //------------------------------------------------------------------------------
 // Initialization
 
 // Speed-critical function pointers. We have to initialize them to the default
 // implementations within VP8EncDspInit().
 VP8CHisto VP8CollectHistogram;
 VP8Idct VP8ITransform;
 VP8Fdct VP8FTransform;
+VP8Fdct VP8FTransform2;
 VP8WHT VP8FTransformWHT;
 VP8Intra4Preds VP8EncPredLuma4;
 VP8IntraPreds VP8EncPredLuma16;
 VP8IntraPreds VP8EncPredChroma8;
 VP8Metric VP8SSE16x16;
 VP8Metric VP8SSE8x8;
 VP8Metric VP8SSE16x8;
 VP8Metric VP8SSE4x4;
 VP8WMetric VP8TDisto4x4;
 VP8WMetric VP8TDisto16x16;
 VP8QuantizeBlock VP8EncQuantizeBlock;
+VP8Quantize2Blocks VP8EncQuantize2Blocks;
 VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT;
 VP8BlockCopy VP8Copy4x4;
+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);
 
 static volatile VP8CPUInfo enc_last_cpuinfo_used =
     (VP8CPUInfo)&enc_last_cpuinfo_used;
 
-void VP8EncDspInit(void) {
+WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInit(void) {
   if (enc_last_cpuinfo_used == VP8GetCPUInfo) return;
 
   VP8DspInit();  // common inverse transforms
   InitTables();
 
   // default C implementations
   VP8CollectHistogram = CollectHistogram;
   VP8ITransform = ITransform;
   VP8FTransform = FTransform;
+  VP8FTransform2 = FTransform2;
   VP8FTransformWHT = FTransformWHT;
   VP8EncPredLuma4 = Intra4Preds;
   VP8EncPredLuma16 = Intra16Preds;
   VP8EncPredChroma8 = IntraChromaPreds;
   VP8SSE16x16 = SSE16x16;
   VP8SSE8x8 = SSE8x8;
   VP8SSE16x8 = SSE16x8;
   VP8SSE4x4 = SSE4x4;
   VP8TDisto4x4 = Disto4x4;
   VP8TDisto16x16 = Disto16x16;
   VP8EncQuantizeBlock = QuantizeBlock;
+  VP8EncQuantize2Blocks = Quantize2Blocks;
   VP8EncQuantizeBlockWHT = QuantizeBlockWHT;
   VP8Copy4x4 = Copy4x4;
+  VP8Copy16x8 = Copy16x8;
 
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8EncDspInitSSE2();
+#if defined(WEBP_USE_SSE41)
+      if (VP8GetCPUInfo(kSSE4_1)) {
+        VP8EncDspInitSSE41();
+      }
+#endif
     }
 #endif
 #if defined(WEBP_USE_AVX2)
     if (VP8GetCPUInfo(kAVX2)) {
       VP8EncDspInitAVX2();
     }
 #endif
 #if defined(WEBP_USE_NEON)
     if (VP8GetCPUInfo(kNEON)) {
       VP8EncDspInitNEON();
     }
 #endif
 #if defined(WEBP_USE_MIPS32)
     if (VP8GetCPUInfo(kMIPS32)) {
       VP8EncDspInitMIPS32();
     }
 #endif
+#if defined(WEBP_USE_MIPS_DSP_R2)
+    if (VP8GetCPUInfo(kMIPSdspR2)) {
+      VP8EncDspInitMIPSdspR2();
+    }
+#endif
   }
   enc_last_cpuinfo_used = VP8GetCPUInfo;
 }
-