libvpx: Pull from upstream

source/libvpx/third_party/libyuv/source/convert_from_argb.cc

 /*
  *  Copyright 2012 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE 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.
  */
 
@@ skipping 54 matching lines @@
     }
   }
 #endif
 #if defined(HAS_ARGBTOYROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToYRow = ARGBToYRow_Any_SSSE3;
     if (IS_ALIGNED(width, 16)) {
       ARGBToYRow = ARGBToYRow_SSSE3;
     }
   }
-
+#endif
+#if defined(HAS_ARGBTOYROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToYRow = ARGBToYRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      ARGBToYRow = ARGBToYRow_AVX2;
+    }
+  }
 #endif
 #if defined(HAS_ARGBTOYROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToYRow = ARGBToYRow_Any_NEON;
     if (IS_ALIGNED(width, 8)) {
       ARGBToYRow = ARGBToYRow_NEON;
     }
   }
 #endif
 
@@ skipping 46 matching lines @@
   }
 #endif
 #if defined(HAS_ARGBTOUV422ROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToUV422Row = ARGBToUV422Row_Any_NEON;
     if (IS_ALIGNED(width, 16)) {
       ARGBToUV422Row = ARGBToUV422Row_NEON;
     }
   }
 #endif
-
 #if defined(HAS_ARGBTOYROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToYRow = ARGBToYRow_Any_SSSE3;
     if (IS_ALIGNED(width, 16)) {
       ARGBToYRow = ARGBToYRow_SSSE3;
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToYRow = ARGBToYRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      ARGBToYRow = ARGBToYRow_AVX2;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToYRow = ARGBToYRow_Any_NEON;
     if (IS_ALIGNED(width, 8)) {
       ARGBToYRow = ARGBToYRow_NEON;
     }
   }
 #endif
 
   for (y = 0; y < height; ++y) {
@@ skipping 107 matching lines @@
 #if defined(HAS_ARGBTOYROW_SSSE3) && defined(HAS_ARGBTOUVROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToUVRow = ARGBToUVRow_Any_SSSE3;
     ARGBToYRow = ARGBToYRow_Any_SSSE3;
     if (IS_ALIGNED(width, 16)) {
       ARGBToUVRow = ARGBToUVRow_SSSE3;
       ARGBToYRow = ARGBToYRow_SSSE3;
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_AVX2) && defined(HAS_ARGBTOUVROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToUVRow = ARGBToUVRow_Any_AVX2;
+    ARGBToYRow = ARGBToYRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      ARGBToUVRow = ARGBToUVRow_AVX2;
+      ARGBToYRow = ARGBToYRow_AVX2;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToYRow = ARGBToYRow_Any_NEON;
     if (IS_ALIGNED(width, 8)) {
       ARGBToYRow = ARGBToYRow_NEON;
     }
   }
 #endif
 #if defined(HAS_ARGBTOUVROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
@@ skipping 22 matching lines @@
 #if defined(HAS_MERGEUVROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     MergeUVRow_ = MergeUVRow_Any_NEON;
     if (IS_ALIGNED(halfwidth, 16)) {
       MergeUVRow_ = MergeUVRow_NEON;
     }
   }
 #endif
   {
     // Allocate a rows of uv.
-    align_buffer_64(row_u, ((halfwidth + 15) & ~15) * 2);
-    uint8* row_v = row_u + ((halfwidth + 15) & ~15);
+    align_buffer_64(row_u, ((halfwidth + 31) & ~31) * 2);
+    uint8* row_v = row_u + ((halfwidth + 31) & ~31);
 
     for (y = 0; y < height - 1; y += 2) {
       ARGBToUVRow(src_argb, src_stride_argb, row_u, row_v, width);
       MergeUVRow_(row_u, row_v, dst_uv, halfwidth);
       ARGBToYRow(src_argb, dst_y, width);
       ARGBToYRow(src_argb + src_stride_argb, dst_y + dst_stride_y, width);
       src_argb += src_stride_argb * 2;
       dst_y += dst_stride_y * 2;
       dst_uv += dst_stride_uv;
     }
@@ skipping 35 matching lines @@
 #if defined(HAS_ARGBTOYROW_SSSE3) && defined(HAS_ARGBTOUVROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToUVRow = ARGBToUVRow_Any_SSSE3;
     ARGBToYRow = ARGBToYRow_Any_SSSE3;
     if (IS_ALIGNED(width, 16)) {
       ARGBToUVRow = ARGBToUVRow_SSSE3;
       ARGBToYRow = ARGBToYRow_SSSE3;
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_AVX2) && defined(HAS_ARGBTOUVROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToUVRow = ARGBToUVRow_Any_AVX2;
+    ARGBToYRow = ARGBToYRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      ARGBToUVRow = ARGBToUVRow_AVX2;
+      ARGBToYRow = ARGBToYRow_AVX2;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToYRow = ARGBToYRow_Any_NEON;
     if (IS_ALIGNED(width, 8)) {
       ARGBToYRow = ARGBToYRow_NEON;
     }
   }
 #endif
 #if defined(HAS_ARGBTOUVROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
@@ skipping 22 matching lines @@
 #if defined(HAS_MERGEUVROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     MergeUVRow_ = MergeUVRow_Any_NEON;
     if (IS_ALIGNED(halfwidth, 16)) {
       MergeUVRow_ = MergeUVRow_NEON;
     }
   }
 #endif
   {
     // Allocate a rows of uv.
-    align_buffer_64(row_u, ((halfwidth + 15) & ~15) * 2);
-    uint8* row_v = row_u + ((halfwidth + 15) & ~15);
+    align_buffer_64(row_u, ((halfwidth + 31) & ~31) * 2);
+    uint8* row_v = row_u + ((halfwidth + 31) & ~31);
 
     for (y = 0; y < height - 1; y += 2) {
       ARGBToUVRow(src_argb, src_stride_argb, row_u, row_v, width);
       MergeUVRow_(row_v, row_u, dst_uv, halfwidth);
       ARGBToYRow(src_argb, dst_y, width);
       ARGBToYRow(src_argb + src_stride_argb, dst_y + dst_stride_y, width);
       src_argb += src_stride_argb * 2;
       dst_y += dst_stride_y * 2;
       dst_uv += dst_stride_uv;
     }
@@ skipping 54 matching lines @@
   }
 #endif
 #if defined(HAS_ARGBTOYROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToYRow = ARGBToYRow_Any_SSSE3;
     if (IS_ALIGNED(width, 16)) {
       ARGBToYRow = ARGBToYRow_SSSE3;
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToYRow = ARGBToYRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      ARGBToYRow = ARGBToYRow_AVX2;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToYRow = ARGBToYRow_Any_NEON;
     if (IS_ALIGNED(width, 8)) {
       ARGBToYRow = ARGBToYRow_NEON;
     }
   }
 #endif
 
 #if defined(HAS_I422TOYUY2ROW_SSE2)
@@ skipping 79 matching lines @@
   }
 #endif
 #if defined(HAS_ARGBTOYROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToYRow = ARGBToYRow_Any_SSSE3;
     if (IS_ALIGNED(width, 16)) {
       ARGBToYRow = ARGBToYRow_SSSE3;
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToYRow = ARGBToYRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      ARGBToYRow = ARGBToYRow_AVX2;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToYRow = ARGBToYRow_Any_NEON;
     if (IS_ALIGNED(width, 8)) {
       ARGBToYRow = ARGBToYRow_NEON;
     }
   }
 #endif
 
 #if defined(HAS_I422TOUYVYROW_SSE2)
@@ skipping 193 matching lines @@
 #endif
 
   for (y = 0; y < height; ++y) {
     ARGBToRAWRow(src_argb, dst_raw, width);
     src_argb += src_stride_argb;
     dst_raw += dst_stride_raw;
   }
   return 0;
 }
 
-static const uint8 kDither8x8[64] = {
-  0, 128, 32, 160,  8, 136, 40, 168,
-  192, 64, 224, 96, 200, 72, 232, 104,
-  48, 176, 16, 144, 56, 184, 24, 152,
-  240, 112, 208, 80, 248, 120, 216, 88,
-  12, 140, 44, 172,  4, 132, 36, 164,
-  204, 76, 236, 108, 196, 68, 228, 100,
-  60, 188, 28, 156, 52, 180, 20, 148,
-  252, 124, 220, 92, 244, 116, 212, 84,
+// Ordered 8x8 dither for 888 to 565.  Values from 0 to 7.
+static const uint8 kDither565_4x4[16] = {
+  0, 4, 1, 5,
+  6, 2, 7, 3,
+  1, 5, 0, 4,
+  7, 3, 6, 2,
 };
 
-// Convert ARGB To RGB565 with 8x8 dither matrix (64 bytes).
+// Convert ARGB To RGB565 with 4x4 dither matrix (16 bytes).
 LIBYUV_API
 int ARGBToRGB565Dither(const uint8* src_argb, int src_stride_argb,
                        uint8* dst_rgb565, int dst_stride_rgb565,
-                       const uint8* dither8x8, int width, int height) {
+                       const uint8* dither4x4, int width, int height) {
   int y;
   void (*ARGBToRGB565DitherRow)(const uint8* src_argb, uint8* dst_rgb,
-      const uint8* dither8x8, int pix) = ARGBToRGB565DitherRow_C;
+      const uint32 dither4, int pix) = ARGBToRGB565DitherRow_C;
   if (!src_argb || !dst_rgb565 || width <= 0 || height == 0) {
     return -1;
   }
   if (height < 0) {
     height = -height;
     src_argb = src_argb + (height - 1) * src_stride_argb;
     src_stride_argb = -src_stride_argb;
   }
-  if (!dither8x8) {
-    dither8x8 = kDither8x8;
-
+  if (!dither4x4) {
+    dither4x4 = kDither565_4x4;
   }
+#if defined(HAS_ARGBTORGB565DITHERROW_SSE2)
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_Any_SSE2;
+    if (IS_ALIGNED(width, 4)) {
+      ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_SSE2;
+    }
+  }
+#endif
+#if defined(HAS_ARGBTORGB565DITHERROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_Any_AVX2;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_AVX2;
+    }
+  }
+#endif
+#if defined(HAS_ARGBTORGB565DITHERROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_Any_NEON;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_NEON;
+    }
+  }
+#endif
   for (y = 0; y < height; ++y) {
     ARGBToRGB565DitherRow(src_argb, dst_rgb565,
-                          dither8x8 + ((y & 7) << 3), width);
+                          *(uint32*)(dither4x4 + ((y & 3) << 2)), width);
     src_argb += src_stride_argb;
     dst_rgb565 += dst_stride_rgb565;
   }
   return 0;
 }
 
 // Convert ARGB To RGB565.
+// TODO(fbarchard): Consider using dither function low level with zeros.
 LIBYUV_API
 int ARGBToRGB565(const uint8* src_argb, int src_stride_argb,
                  uint8* dst_rgb565, int dst_stride_rgb565,
                  int width, int height) {
   int y;
   void (*ARGBToRGB565Row)(const uint8* src_argb, uint8* dst_rgb, int pix) =
       ARGBToRGB565Row_C;
   if (!src_argb || !dst_rgb565 || width <= 0 || height == 0) {
     return -1;
   }
@@ skipping 156 matching lines @@
 
 // Convert ARGB to J420. (JPeg full range I420).
 LIBYUV_API
 int ARGBToJ420(const uint8* src_argb, int src_stride_argb,
                uint8* dst_yj, int dst_stride_yj,
                uint8* dst_u, int dst_stride_u,
                uint8* dst_v, int dst_stride_v,
                int width, int height) {
   int y;
   void (*ARGBToUVJRow)(const uint8* src_argb0, int src_stride_argb,
-                      uint8* dst_u, uint8* dst_v, int width) = ARGBToUVJRow_C;
+                       uint8* dst_u, uint8* dst_v, int width) = ARGBToUVJRow_C;
   void (*ARGBToYJRow)(const uint8* src_argb, uint8* dst_yj, int pix) =
       ARGBToYJRow_C;
   if (!src_argb ||
       !dst_yj || !dst_u || !dst_v ||
       width <= 0 || height == 0) {
     return -1;
   }
   // Negative height means invert the image.
   if (height < 0) {
     height = -height;
     src_argb = src_argb + (height - 1) * src_stride_argb;
     src_stride_argb = -src_stride_argb;
   }
 #if defined(HAS_ARGBTOYJROW_SSSE3) && defined(HAS_ARGBTOUVJROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToUVJRow = ARGBToUVJRow_Any_SSSE3;
     ARGBToYJRow = ARGBToYJRow_Any_SSSE3;
     if (IS_ALIGNED(width, 16)) {
       ARGBToUVJRow = ARGBToUVJRow_SSSE3;
       ARGBToYJRow = ARGBToYJRow_SSSE3;
     }
   }
 #endif
-#if defined(HAS_ARGBTOYJROW_AVX2) && defined(HAS_ARGBTOUVJROW_AVX2)
+#if defined(HAS_ARGBTOYJROW_AVX2)
   if (TestCpuFlag(kCpuHasAVX2)) {
     ARGBToYJRow = ARGBToYJRow_Any_AVX2;
     if (IS_ALIGNED(width, 32)) {
       ARGBToYJRow = ARGBToYJRow_AVX2;
     }
   }
 #endif
 #if defined(HAS_ARGBTOYJROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToYJRow = ARGBToYJRow_Any_NEON;
@@ skipping 74 matching lines @@
 #endif
 
 #if defined(HAS_ARGBTOYJROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToYJRow = ARGBToYJRow_Any_SSSE3;
     if (IS_ALIGNED(width, 16)) {
       ARGBToYJRow = ARGBToYJRow_SSSE3;
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToYJRow = ARGBToYJRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      ARGBToYJRow = ARGBToYJRow_AVX2;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYJROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToYJRow = ARGBToYJRow_Any_NEON;
     if (IS_ALIGNED(width, 8)) {
       ARGBToYJRow = ARGBToYJRow_NEON;
     }
   }
 #endif
 
   for (y = 0; y < height; ++y) {
@@ skipping 60 matching lines @@
     src_argb += src_stride_argb;
     dst_yj += dst_stride_yj;
   }
   return 0;
 }
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif