Reenable AVX2 scaling with bug fix for any width

BUG=376
TESTED=unittest on scale functions
R=brucedawson@google.com, harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/30759004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1135 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1134
+Version: 1135
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -201,8 +201,7 @@ extern "C" {
 #define HAS_ARGBTOYJROW_AVX2
 #define HAS_ARGBTOYROW_AVX2
 #define HAS_I422TOARGBROW_AVX2
-// TODO(fbarchard): fix bug #376.
-// #define HAS_INTERPOLATEROW_AVX2
+#define HAS_INTERPOLATEROW_AVX2
 #define HAS_MERGEUVROW_AVX2
 #define HAS_MIRRORROW_AVX2
 #define HAS_SPLITUVROW_AVX2

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1134
+#define LIBYUV_VERSION 1135
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_any.cc

@@ -581,24 +581,20 @@ YANY(ARGBShuffleRow_Any_NEON, ARGBShuffleRow_NEON,
     }
 
 #ifdef HAS_INTERPOLATEROW_AVX2
-NANY(InterpolateRow_Any_AVX2, InterpolateRow_AVX2,
-     InterpolateRow_C, 1, 1, 32)
+NANY(InterpolateRow_Any_AVX2, InterpolateRow_AVX2, InterpolateRow_C, 1, 1, 31)
 #endif
 #ifdef HAS_INTERPOLATEROW_SSSE3
-NANY(InterpolateRow_Any_SSSE3, InterpolateRow_SSSE3,
-     InterpolateRow_C, 1, 1, 15)
+NANY(InterpolateRow_Any_SSSE3, InterpolateRow_SSSE3, InterpolateRow_C, 1, 1, 15)
 #endif
 #ifdef HAS_INTERPOLATEROW_SSE2
-NANY(InterpolateRow_Any_SSE2, InterpolateRow_SSE2,
-     InterpolateRow_C, 1, 1, 15)
+NANY(InterpolateRow_Any_SSE2, InterpolateRow_SSE2, InterpolateRow_C, 1, 1, 15)
 #endif
 #ifdef HAS_INTERPOLATEROW_NEON
-NANY(InterpolateRow_Any_NEON, InterpolateRow_NEON,
-     InterpolateRow_C, 1, 1, 15)
+NANY(InterpolateRow_Any_NEON, InterpolateRow_NEON, InterpolateRow_C, 1, 1, 15)
 #endif
 #ifdef HAS_INTERPOLATEROW_MIPS_DSPR2
-NANY(InterpolateRow_Any_MIPS_DSPR2, InterpolateRow_MIPS_DSPR2,
-     InterpolateRow_C, 1, 1, 3)
+NANY(InterpolateRow_Any_MIPS_DSPR2, InterpolateRow_MIPS_DSPR2, InterpolateRow_C,
+     1, 1, 3)
 #endif
 #undef NANY
 

source/row_win.cc

@@ -4972,11 +4972,11 @@ void ARGBAffineRow_SSE2(const uint8* src_argb, int src_argb_stride,
 #endif  // HAS_ARGBAFFINEROW_SSE2
 
 #ifdef HAS_INTERPOLATEROW_AVX2
-// Bilinear filter 16x2 -> 16x1
+// Bilinear filter 32x2 -> 32x1
 __declspec(naked) __declspec(align(16))
 void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr,
-                          ptrdiff_t src_stride, int dst_width,
-                          int source_y_fraction) {
+                         ptrdiff_t src_stride, int dst_width,
+                         int source_y_fraction) {
   __asm {
     push       esi
     push       edi
@@ -5023,45 +5023,48 @@ void InterpolateRow_AVX2(uint8* dst_ptr, const uint8* src_ptr,
     jg         xloop
     jmp        xloop99
 
-    // Blend 25 / 75.
-    align      4
-  xloop25:
-    vmovdqu    ymm0, [esi]
-    vpavgb     ymm0, ymm0, [esi + edx]
-    vpavgb     ymm0, ymm0, [esi + edx]
-    sub        ecx, 32
-    vmovdqu    [esi + edi], ymm0
-    lea        esi, [esi + 32]
-    jg         xloop25
-    jmp        xloop99
-
-    // Blend 50 / 50.
-    align      4
-  xloop50:
-    vmovdqu    ymm0, [esi]
-    vpavgb     ymm0, ymm0, [esi + edx]
-    sub        ecx, 32
-    vmovdqu    [esi + edi], ymm0
-    lea        esi, [esi + 32]
-    jg         xloop50
-    jmp        xloop99
-
-    // Blend 75 / 25.
-    align      4
-  xloop75:
-    vmovdqu    ymm0, [esi + edx]
-    vpavgb     ymm0, ymm0, [esi]
-    vpavgb     ymm0, ymm0, [esi]
-    sub        ecx, 32
-    vmovdqu     [esi + edi], ymm0
-    lea        esi, [esi + 32]
-    jg         xloop75
-    jmp        xloop99
-
-    // Blend 100 / 0 - Copy row unchanged.
-    align      4
-  xloop100:
-    rep movsb
+   // Blend 25 / 75.
+   align      4
+ xloop25:
+   vmovdqu    ymm0, [esi]
+   vmovdqu    ymm1, [esi + edx]
+   vpavgb     ymm0, ymm0, ymm1
+   vpavgb     ymm0, ymm0, ymm1
+   sub        ecx, 32
+   vmovdqu    [esi + edi], ymm0
+   lea        esi, [esi + 32]
+   jg         xloop25
+   jmp        xloop99
+
+   // Blend 50 / 50.
+   align      4
+ xloop50:
+   vmovdqu    ymm0, [esi]
+   vmovdqu    ymm1, [esi + edx]
+   vpavgb     ymm0, ymm0, ymm1
+   sub        ecx, 32
+   vmovdqu    [esi + edi], ymm0
+   lea        esi, [esi + 32]
+   jg         xloop50
+   jmp        xloop99
+
+   // Blend 75 / 25.
+   align      4
+ xloop75:
+   vmovdqu    ymm1, [esi]
+   vmovdqu    ymm0, [esi + edx]
+   vpavgb     ymm0, ymm0, ymm1
+   vpavgb     ymm0, ymm0, ymm1
+   sub        ecx, 32
+   vmovdqu    [esi + edi], ymm0
+   lea        esi, [esi + 32]
+   jg         xloop75
+   jmp        xloop99
+
+   // Blend 100 / 0 - Copy row unchanged.
+   align      4
+ xloop100:
+   rep movsb
 
   xloop99:
     pop        edi