Optimize BlendPlane SIMD support for odd width.

source/planar_functions.cc

 /*
  *  Copyright 2011 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 592 matching lines @@
       src_stride_y1 == width &&
       alpha_stride == width &&
       dst_stride_y == width) {
     width *= height;
     height = 1;
     src_stride_y0 = src_stride_y1 = alpha_stride = dst_stride_y = 0;
   }
 
 #if defined(HAS_BLENDPLANEROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
-// TODO(fbarchard): Implement any versions for odd width.
-//  BlendPlaneRow = BlendPlaneRow_Any_SSSE3;
+  BlendPlaneRow = BlendPlaneRow_Any_SSSE3;
     if (IS_ALIGNED(width, 8)) {
       BlendPlaneRow = BlendPlaneRow_SSSE3;
     }
   }
 #endif
 #if defined(HAS_BLENDPLANEROW_AVX2)
   if (TestCpuFlag(kCpuHasAVX2)) {
-//  BlendPlaneRow = BlendPlaneRow_Any_AVX2;
-    if (IS_ALIGNED(width, 16)) {
+  BlendPlaneRow = BlendPlaneRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
       BlendPlaneRow = BlendPlaneRow_AVX2;
     }
   }
 #endif
 
   for (y = 0; y < height; ++y) {
     BlendPlaneRow(src_y0, src_y1, alpha, dst_y, width);
     src_y0 += src_stride_y0;
     src_y1 += src_stride_y1;
     alpha += alpha_stride;
@@ skipping 37 matching lines @@
 
   // Blend Y plane.
   BlendPlane(src_y0, src_stride_y0,
              src_y1, src_stride_y1,
              alpha, alpha_stride,
              dst_y, dst_stride_y,
              width, height);
 
 #if defined(HAS_BLENDPLANEROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
-// TODO(fbarchard): Implement any versions for odd width.
-//  BlendPlaneRow = BlendPlaneRow_Any_SSSE3;
-    if (IS_ALIGNED(halfwidth, 8)) {
+  BlendPlaneRow = BlendPlaneRow_Any_SSSE3;
+    if (IS_ALIGNED(width, 8)) {
       BlendPlaneRow = BlendPlaneRow_SSSE3;
     }
   }
 #endif
 #if defined(HAS_BLENDPLANEROW_AVX2)
   if (TestCpuFlag(kCpuHasAVX2)) {
-//  BlendPlaneRow = BlendPlaneRow_Any_AVX2;
-    if (IS_ALIGNED(halfwidth, 16)) {
+  BlendPlaneRow = BlendPlaneRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {

[Diony Rosa, 2015/12/08 20:06:53]

You're also using BlendPlaneRow on the chroma channels, so don't you need to
verify that halfwidth is a multiple of 32?

[fbarchard, 2015/12/08 20:09:47]

Done.

       BlendPlaneRow = BlendPlaneRow_AVX2;
     }
   }
 #endif
 #if defined(HAS_SCALEROWDOWN2_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ScaleRowDown2 = ScaleRowDown2Box_Any_NEON;
     if (IS_ALIGNED(halfwidth, 16)) {
       ScaleRowDown2 = ScaleRowDown2Box_NEON;
     }
@@ skipping 1925 matching lines @@
     }
     free_aligned_buffer_64(rows);
   }
   return 0;
 }
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif