From 278d88f872154b32bc93363d90ea605955dbad3d Mon Sep 17 00:00:00 2001
From: Frank Barchard <fbarchard@google.com>
Date: Thu, 13 Aug 2015 15:05:14 -0700
Subject: [PATCH] Copy Alpha odd width support

R=harryjin@google.com
BUG=none

Review URL: https://webrtc-codereview.appspot.com/59369004.
---
 include/libyuv/row.h       |  8 ++++++++
 source/convert_argb.cc     | 14 ++++++++++----
 source/planar_functions.cc | 28 ++++++++++++++++++++--------
 source/row_any.cc          | 12 ++++++++++++
 4 files changed, 50 insertions(+), 12 deletions(-)

diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index ebae3e719..2285c3416 100644
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -816,10 +816,18 @@ void CopyRow_16_C(const uint16* src, uint16* dst, int count);
 void ARGBCopyAlphaRow_C(const uint8* src_argb, uint8* dst_argb, int width);
 void ARGBCopyAlphaRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width);
 void ARGBCopyAlphaRow_AVX2(const uint8* src_argb, uint8* dst_argb, int width);
+void ARGBCopyAlphaRow_Any_SSE2(const uint8* src_argb, uint8* dst_argb,
+                               int width);
+void ARGBCopyAlphaRow_Any_AVX2(const uint8* src_argb, uint8* dst_argb,
+                               int width);
 
 void ARGBCopyYToAlphaRow_C(const uint8* src_y, uint8* dst_argb, int width);
 void ARGBCopyYToAlphaRow_SSE2(const uint8* src_y, uint8* dst_argb, int width);
 void ARGBCopyYToAlphaRow_AVX2(const uint8* src_y, uint8* dst_argb, int width);
+void ARGBCopyYToAlphaRow_Any_SSE2(const uint8* src_y, uint8* dst_argb,
+                                  int width);
+void ARGBCopyYToAlphaRow_Any_AVX2(const uint8* src_y, uint8* dst_argb,
+                                  int width);
 
 void SetRow_C(uint8* dst, uint8 v8, int count);
 void SetRow_X86(uint8* dst, uint8 v8, int count);
diff --git a/source/convert_argb.cc b/source/convert_argb.cc
index 87434b172..1f8ee632c 100644
--- a/source/convert_argb.cc
+++ b/source/convert_argb.cc
@@ -319,13 +319,19 @@ int I420AlphaToARGB(const uint8* src_y, int src_stride_y,
   }
 #endif
 #if defined(HAS_ARGBCOPYYTOALPHAROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 8)) {
-    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_SSE2;
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_Any_SSE2;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_SSE2;
+    }
   }
 #endif
 #if defined(HAS_ARGBCOPYYTOALPHAROW_AVX2)
-  if (TestCpuFlag(kCpuHasAVX2) && IS_ALIGNED(width, 16)) {
-    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_AVX2;
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_Any_AVX2;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_AVX2;
+    }
   }
 #endif
 #if defined(HAS_ARGBATTENUATEROW_SSE2)
diff --git a/source/planar_functions.cc b/source/planar_functions.cc
index b96bd5020..3cca5f48b 100644
--- a/source/planar_functions.cc
+++ b/source/planar_functions.cc
@@ -2280,13 +2280,19 @@ int ARGBCopyAlpha(const uint8* src_argb, int src_stride_argb,
     src_stride_argb = dst_stride_argb = 0;
   }
 #if defined(HAS_ARGBCOPYALPHAROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 8)) {
-    ARGBCopyAlphaRow = ARGBCopyAlphaRow_SSE2;
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    ARGBCopyAlphaRow = ARGBCopyAlphaRow_Any_SSE2;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBCopyAlphaRow = ARGBCopyAlphaRow_SSE2;
+    }
   }
 #endif
 #if defined(HAS_ARGBCOPYALPHAROW_AVX2)
-  if (TestCpuFlag(kCpuHasAVX2) && IS_ALIGNED(width, 16)) {
-    ARGBCopyAlphaRow = ARGBCopyAlphaRow_AVX2;
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBCopyAlphaRow = ARGBCopyAlphaRow_Any_AVX2;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBCopyAlphaRow = ARGBCopyAlphaRow_AVX2;
+    }
   }
 #endif
 
@@ -2323,13 +2329,19 @@ int ARGBCopyYToAlpha(const uint8* src_y, int src_stride_y,
     src_stride_y = dst_stride_argb = 0;
   }
 #if defined(HAS_ARGBCOPYYTOALPHAROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 8)) {
-    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_SSE2;
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_Any_SSE2;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_SSE2;
+    }
   }
 #endif
 #if defined(HAS_ARGBCOPYYTOALPHAROW_AVX2)
-  if (TestCpuFlag(kCpuHasAVX2) && IS_ALIGNED(width, 16)) {
-    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_AVX2;
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_Any_AVX2;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBCopyYToAlphaRow = ARGBCopyYToAlphaRow_AVX2;
+    }
   }
 #endif
 
diff --git a/source/row_any.cc b/source/row_any.cc
index 1cb1f6b93..2ef5359bd 100644
--- a/source/row_any.cc
+++ b/source/row_any.cc
@@ -245,6 +245,18 @@ ANY11(CopyRow_Any_SSE2, CopyRow_SSE2, 0, 1, 1, 31)
 #ifdef HAS_COPYROW_NEON
 ANY11(CopyRow_Any_NEON, CopyRow_NEON, 0, 1, 1, 31)
 #endif
+#ifdef HAS_ARGBCOPYALPHAROW_AVX2
+ANY11(ARGBCopyAlphaRow_Any_AVX2, ARGBCopyAlphaRow_AVX2, 0, 1, 4, 15)
+#endif
+#ifdef HAS_ARGBCOPYYTOALPHAROW_SSE2
+ANY11(ARGBCopyAlphaRow_Any_SSE2, ARGBCopyAlphaRow_SSE2, 0, 1, 4, 7)
+#endif
+#ifdef HAS_ARGBCOPYYTOALPHAROW_AVX2
+ANY11(ARGBCopyYToAlphaRow_Any_AVX2, ARGBCopyYToAlphaRow_AVX2, 0, 1, 4, 15)
+#endif
+#ifdef HAS_ARGBCOPYYTOALPHAROW_SSE2
+ANY11(ARGBCopyYToAlphaRow_Any_SSE2, ARGBCopyYToAlphaRow_SSE2, 0, 1, 4, 7)
+#endif
 #if defined(HAS_ARGBTORGB24ROW_SSSE3)
 ANY11(ARGBToRGB24Row_Any_SSSE3, ARGBToRGB24Row_SSSE3, 0, 4, 3, 15)
 ANY11(ARGBToRAWRow_Any_SSSE3, ARGBToRAWRow_SSSE3, 0, 4, 3, 15)