InterpolateRow used for scale handle unaligned memory. Remove HalfRow which is not used.

BUG=367
TESTED=unittest on I422ToI420
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/28639004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1107 16f28f9a-4ce2-e073-06de-1de4eb20be90

8 files changed, 53 insertions, 196 deletions

diff --git a/README.chromium b/README.chromium
index cff56a3..baa50e9 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1105
+Version: 1106
 License: BSD
 License File: LICENSE
 
diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index dd83afa..e9f62b8 100644
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -124,7 +124,6 @@ extern "C" {
 #define HAS_COPYROW_ERMS
 #define HAS_COPYROW_SSE2
 #define HAS_COPYROW_X86
-#define HAS_HALFROW_SSE2
 #define HAS_I400TOARGBROW_SSE2
 #define HAS_I411TOARGBROW_SSSE3
 #define HAS_I422TOARGB1555ROW_SSSE3
@@ -212,7 +211,6 @@ extern "C" {
 #define HAS_ARGBTOUVROW_AVX2
 #define HAS_ARGBTOYJROW_AVX2
 #define HAS_ARGBTOYROW_AVX2
-#define HAS_HALFROW_AVX2
 #define HAS_I422TOARGBROW_AVX2
 #define HAS_INTERPOLATEROW_AVX2
 #define HAS_MERGEUVROW_AVX2
@@ -300,7 +298,6 @@ extern "C" {
 #define HAS_UYVYTOUV422ROW_NEON
 #define HAS_YUY2TOUVROW_NEON
 #define HAS_UYVYTOUVROW_NEON
-#define HAS_HALFROW_NEON
 #define HAS_ARGBTOBAYERROW_NEON
 #define HAS_ARGBTOBAYERGGROW_NEON
 #define HAS_ARGBSHUFFLEROW_NEON
@@ -378,7 +375,6 @@ extern "C" {
 #define HAS_BGRATOUVROW_NEON
 #define HAS_BGRATOYROW_NEON
 #define HAS_COPYROW_NEON
-#define HAS_HALFROW_NEON
 #define HAS_I400TOARGBROW_NEON
 #define HAS_I411TOARGBROW_NEON
 #define HAS_I422TOABGRROW_NEON
@@ -1577,18 +1573,6 @@ void UYVYToUVRow_Any_NEON(const uint8* src_uyvy, int stride_uyvy,
 void UYVYToUV422Row_Any_NEON(const uint8* src_uyvy,
                              uint8* dst_u, uint8* dst_v, int pix);
 
-void HalfRow_C(const uint8* src_uv, int src_uv_stride,
-               uint8* dst_uv, int pix);
-void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride,
-                  uint8* dst_uv, int pix);
-void HalfRow_AVX2(const uint8* src_uv, int src_uv_stride,
-                  uint8* dst_uv, int pix);
-void HalfRow_NEON(const uint8* src_uv, int src_uv_stride,
-                  uint8* dst_uv, int pix);
-
-void HalfRow_16_C(const uint16* src_uv, int src_uv_stride,
-                  uint16* dst_uv, int pix);
-
 void ARGBToBayerRow_C(const uint8* src_argb, uint8* dst_bayer,
                       uint32 selector, int pix);
 void ARGBToBayerRow_SSSE3(const uint8* src_argb, uint8* dst_bayer,
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index 8a89025..24b888f 100644
--- a/include/libyuv/version.h
+++ b/include/libyuv/version.h
@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1105
+#define LIBYUV_VERSION 1106
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT
diff --git a/source/row_common.cc b/source/row_common.cc
index b998819..40a8261 100644
--- a/source/row_common.cc
+++ b/source/row_common.cc
@@ -1885,17 +1885,17 @@ void ARGBAffineRow_C(const uint8* src_argb, int src_argb_stride,
   }
 }
 
-// Blend 2 rows into 1 for conversions such as I422ToI420.
-void HalfRow_C(const uint8* src_uv, int src_uv_stride,
-               uint8* dst_uv, int pix) {
+// Blend 2 rows into 1.
+static void HalfRow_C(const uint8* src_uv, int src_uv_stride,
+                      uint8* dst_uv, int pix) {
   int x;
   for (x = 0; x < pix; ++x) {
     dst_uv[x] = (src_uv[x] + src_uv[src_uv_stride + x] + 1) >> 1;
   }
 }
 
-void HalfRow_16_C(const uint16* src_uv, int src_uv_stride,
-                  uint16* dst_uv, int pix) {
+static void HalfRow_16_C(const uint16* src_uv, int src_uv_stride,
+                         uint16* dst_uv, int pix) {
   int x;
   for (x = 0; x < pix; ++x) {
     dst_uv[x] = (src_uv[x] + src_uv[src_uv_stride + x] + 1) >> 1;
diff --git a/source/row_neon.cc b/source/row_neon.cc
index 12c294e..ac1c5e5 100644
--- a/source/row_neon.cc
+++ b/source/row_neon.cc
@@ -1274,30 +1274,6 @@ void UYVYToUVRow_NEON(const uint8* src_uyvy, int stride_uyvy,
   );
 }
 
-void HalfRow_NEON(const uint8* src_uv, int src_uv_stride,
-                  uint8* dst_uv, int pix) {
-  asm volatile (
-    // change the stride to row 2 pointer
-    "add        %1, %0                         \n"
-  "1:                                          \n"
-    MEMACCESS(0)
-    "vld1.8     {q0}, [%0]!                    \n"  // load row 1 16 pixels.
-    "subs       %3, %3, #16                    \n"  // 16 processed per loop
-    MEMACCESS(1)
-    "vld1.8     {q1}, [%1]!                    \n"  // load row 2 16 pixels.
-    "vrhadd.u8  q0, q1                         \n"  // average row 1 and 2
-    MEMACCESS(2)
-    "vst1.8     {q0}, [%2]!                    \n"
-    "bgt        1b                             \n"
-  : "+r"(src_uv),         // %0
-    "+r"(src_uv_stride),  // %1
-    "+r"(dst_uv),         // %2
-    "+r"(pix)             // %3
-  :
-  : "cc", "memory", "q0", "q1"  // Clobber List
-  );
-}
-
 // Select 2 channels from ARGB on alternating pixels.  e.g.  BGBGBGBG
 void ARGBToBayerRow_NEON(const uint8* src_argb, uint8* dst_bayer,
                          uint32 selector, int pix) {
diff --git a/source/row_neon64.cc b/source/row_neon64.cc
index 164786d..1829af8 100644
--- a/source/row_neon64.cc
+++ b/source/row_neon64.cc
@@ -1260,32 +1260,6 @@ void UYVYToUVRow_NEON(const uint8* src_uyvy, int stride_uyvy,
 }
 #endif  // HAS_UYVYTOUVROW_NEON
 
-#ifdef HAS_HALFROW_NEON
-void HalfRow_NEON(const uint8* src_uv, int src_uv_stride,
-                  uint8* dst_uv, int pix) {
-  const uint8* src_uvb = src_uv + src_uv_stride;
-  asm volatile (
-    // change the stride to row 2 pointer
-  "1:                                          \n"
-    MEMACCESS(0)
-    "ld1        {v0.16b}, [%0], #16            \n"  // load row 1 16 pixels.
-    "subs       %3, %3, #16                    \n"  // 16 processed per loop
-    MEMACCESS(1)
-    "ld1        {v1.16b}, [%1], #16            \n"  // load row 2 16 pixels.
-    "urhadd     v0.16b, v0.16b, v1.16b         \n"  // average row 1 and 2
-    MEMACCESS(2)
-    "st1        {v0.16b}, [%2], #16            \n"
-    "b.gt       1b                             \n"
-  : "+r"(src_uv),         // %0
-    "+r"(src_uvb),        // %1
-    "+r"(dst_uv),         // %2
-    "+r"(pix)             // %3
-  :
-  : "cc", "memory", "v0", "v1"  // Clobber List
-  );
-}
-#endif  // HAS_HALFROW_NEON
-
 // Select 2 channels from ARGB on alternating pixels.  e.g.  BGBGBGBG
 #ifdef HAS_ARGBTOBAYERROW_NEON
 void ARGBToBayerRow_NEON(const uint8* src_argb, uint8* dst_bayer,
diff --git a/source/row_posix.cc b/source/row_posix.cc
index 5eefec4..36111c1 100644
--- a/source/row_posix.cc
+++ b/source/row_posix.cc
@@ -3182,14 +3182,14 @@ void MergeUVRow_SSE2(const uint8* src_u, const uint8* src_v, uint8* dst_uv,
     "sub       %0,%1                             \n"
     LABELALIGN
   "1:                                            \n"
-    "movdqa    " MEMACCESS(0) ",%%xmm0           \n"
-    MEMOPREG(movdqa,0x00,0,1,1,xmm1)             //  movdqa    (%0,%1,1),%%xmm1
+    "movdqu    " MEMACCESS(0) ",%%xmm0           \n"
+    MEMOPREG(movdqu,0x00,0,1,1,xmm1)             //  movdqa    (%0,%1,1),%%xmm1
     "lea       " MEMLEA(0x10,0) ",%0             \n"
     "movdqa    %%xmm0,%%xmm2                     \n"
     "punpcklbw %%xmm1,%%xmm0                     \n"
     "punpckhbw %%xmm1,%%xmm2                     \n"
-    "movdqa    %%xmm0," MEMACCESS(2) "           \n"
-    "movdqa    %%xmm2," MEMACCESS2(0x10,2) "     \n"
+    "movdqu    %%xmm0," MEMACCESS(2) "           \n"
+    "movdqu    %%xmm2," MEMACCESS2(0x10,2) "     \n"
     "lea       " MEMLEA(0x20,2) ",%2             \n"
     "sub       $0x10,%3                          \n"
     "jg        1b                                \n"
@@ -3246,11 +3246,11 @@ void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
   asm volatile (
     LABELALIGN
   "1:                                          \n"
-    "movdqa    " MEMACCESS(0) ",%%xmm0         \n"
-    "movdqa    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
+    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
+    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
     "lea       " MEMLEA(0x20,0) ",%0           \n"
-    "movdqa    %%xmm0," MEMACCESS(1) "         \n"
-    "movdqa    %%xmm1," MEMACCESS2(0x10,1) "   \n"
+    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
+    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
     "lea       " MEMLEA(0x20,1) ",%1           \n"
     "sub       $0x20,%2                        \n"
     "jg        1b                              \n"
@@ -3331,19 +3331,19 @@ void ARGBCopyAlphaRow_SSE2(const uint8* src, uint8* dst, int width) {
     "psrld     $0x8,%%xmm1                     \n"
     LABELALIGN
   "1:                                          \n"
-    "movdqa    " MEMACCESS(0) ",%%xmm2         \n"
-    "movdqa    " MEMACCESS2(0x10,0) ",%%xmm3   \n"
+    "movdqu    " MEMACCESS(0) ",%%xmm2         \n"
+    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm3   \n"
     "lea       " MEMLEA(0x20,0) ",%0           \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm4         \n"
-    "movdqa    " MEMACCESS2(0x10,1) ",%%xmm5   \n"
+    "movdqu    " MEMACCESS(1) ",%%xmm4         \n"
+    "movdqu    " MEMACCESS2(0x10,1) ",%%xmm5   \n"
     "pand      %%xmm0,%%xmm2                   \n"
     "pand      %%xmm0,%%xmm3                   \n"
     "pand      %%xmm1,%%xmm4                   \n"
     "pand      %%xmm1,%%xmm5                   \n"
     "por       %%xmm4,%%xmm2                   \n"
     "por       %%xmm5,%%xmm3                   \n"
-    "movdqa    %%xmm2," MEMACCESS(1) "         \n"
-    "movdqa    %%xmm3," MEMACCESS2(0x10,1) "   \n"
+    "movdqu    %%xmm2," MEMACCESS(1) "         \n"
+    "movdqu    %%xmm3," MEMACCESS2(0x10,1) "   \n"
     "lea       " MEMLEA(0x20,1) ",%1           \n"
     "sub       $0x8,%2                         \n"
     "jg        1b                              \n"
@@ -5377,8 +5377,8 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
     // General purpose row blend.
     LABELALIGN
   "1:                                          \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm0         \n"
-    MEMOPREG(movdqa,0x00,1,4,1,xmm2)
+    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
+    MEMOPREG(movdqu,0x00,1,4,1,xmm2)
     "movdqa    %%xmm0,%%xmm1                   \n"
     "punpcklbw %%xmm2,%%xmm0                   \n"
     "punpckhbw %%xmm2,%%xmm1                   \n"
@@ -5389,7 +5389,7 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
     "packuswb  %%xmm1,%%xmm0                   \n"
     "sub       $0x10,%2                        \n"
     BUNDLEALIGN
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        1b                              \n"
     "jmp       99f                             \n"
@@ -5397,13 +5397,13 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
     // Blend 25 / 75.
     LABELALIGN
   "25:                                         \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm0         \n"
-    MEMOPREG(movdqa,0x00,1,4,1,xmm1)
+    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
+    MEMOPREG(movdqu,0x00,1,4,1,xmm1)
     "pavgb     %%xmm1,%%xmm0                   \n"
     "pavgb     %%xmm1,%%xmm0                   \n"
     "sub       $0x10,%2                        \n"
     BUNDLEALIGN
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        25b                             \n"
     "jmp       99f                             \n"
@@ -5411,12 +5411,12 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
     // Blend 50 / 50.
     LABELALIGN
   "50:                                         \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm0         \n"
-    MEMOPREG(movdqa,0x00,1,4,1,xmm1)
+    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
+    MEMOPREG(movdqu,0x00,1,4,1,xmm1)
     "pavgb     %%xmm1,%%xmm0                   \n"
     "sub       $0x10,%2                        \n"
     BUNDLEALIGN
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        50b                             \n"
     "jmp       99f                             \n"
@@ -5424,13 +5424,13 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
     // Blend 75 / 25.
     LABELALIGN
   "75:                                         \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm1         \n"
-    MEMOPREG(movdqa,0x00,1,4,1,xmm0)
+    "movdqu    " MEMACCESS(1) ",%%xmm1         \n"
+    MEMOPREG(movdqu,0x00,1,4,1,xmm0)
     "pavgb     %%xmm1,%%xmm0                   \n"
     "pavgb     %%xmm1,%%xmm0                   \n"
     "sub       $0x10,%2                        \n"
     BUNDLEALIGN
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        75b                             \n"
     "jmp       99f                             \n"
@@ -5438,9 +5438,9 @@ void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
     // Blend 100 / 0 - Copy row unchanged.
     LABELALIGN
   "100:                                        \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm0         \n"
+    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
     "sub       $0x10,%2                        \n"
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        100b                            \n"
 
@@ -5490,8 +5490,8 @@ void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr,
     // General purpose row blend.
     LABELALIGN
   "1:                                          \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm0         \n"
-    MEMOPREG(movdqa,0x00,1,4,1,xmm2)           //  movdqa    (%1,%4,1),%%xmm2
+    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
+    MEMOPREG(movdqu,0x00,1,4,1,xmm2)           //  movdqu    (%1,%4,1),%%xmm2
     "movdqa    %%xmm0,%%xmm1                   \n"
     "movdqa    %%xmm2,%%xmm3                   \n"
     "punpcklbw %%xmm4,%%xmm2                   \n"
@@ -5509,7 +5509,7 @@ void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr,
     "packuswb  %%xmm1,%%xmm0                   \n"
     "sub       $0x10,%2                        \n"
     BUNDLEALIGN
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)           //  movdqa    %%xmm0,(%1,%0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)           //  movdqu    %%xmm0,(%1,%0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        1b                              \n"
     "jmp       99f                             \n"
@@ -5517,13 +5517,13 @@ void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr,
     // Blend 25 / 75.
     LABELALIGN
   "25:                                         \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm0         \n"
-    MEMOPREG(movdqa,0x00,1,4,1,xmm1)           //  movdqa    (%1,%4,1),%%xmm1
+    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
+    MEMOPREG(movdqu,0x00,1,4,1,xmm1)           //  movdqu    (%1,%4,1),%%xmm1
     "pavgb     %%xmm1,%%xmm0                   \n"
     "pavgb     %%xmm1,%%xmm0                   \n"
     "sub       $0x10,%2                        \n"
     BUNDLEALIGN
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)           //  movdqa    %%xmm0,(%1,%0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)           //  movdqu    %%xmm0,(%1,%0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        25b                             \n"
     "jmp       99f                             \n"
@@ -5531,12 +5531,12 @@ void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr,
     // Blend 50 / 50.
     LABELALIGN
   "50:                                         \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm0         \n"
-    MEMOPREG(movdqa,0x00,1,4,1,xmm1)           //  movdqa    (%1,%4,1),%%xmm1
+    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
+    MEMOPREG(movdqu,0x00,1,4,1,xmm1)           //  movdqu    (%1,%4,1),%%xmm1
     "pavgb     %%xmm1,%%xmm0                   \n"
     "sub       $0x10,%2                        \n"
     BUNDLEALIGN
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)           //  movdqa    %%xmm0,(%1,%0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)           //  movdqu    %%xmm0,(%1,%0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        50b                             \n"
     "jmp       99f                             \n"
@@ -5544,13 +5544,13 @@ void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr,
     // Blend 75 / 25.
     LABELALIGN
   "75:                                         \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm1         \n"
-    MEMOPREG(movdqa,0x00,1,4,1,xmm0)           //  movdqa    (%1,%4,1),%%xmm0
+    "movdqu    " MEMACCESS(1) ",%%xmm1         \n"
+    MEMOPREG(movdqu,0x00,1,4,1,xmm0)           //  movdqu    (%1,%4,1),%%xmm0
     "pavgb     %%xmm1,%%xmm0                   \n"
     "pavgb     %%xmm1,%%xmm0                   \n"
     "sub       $0x10,%2                        \n"
     BUNDLEALIGN
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)           //  movdqa    %%xmm0,(%1,%0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)           //  movdqu    %%xmm0,(%1,%0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        75b                             \n"
     "jmp       99f                             \n"
@@ -5558,9 +5558,9 @@ void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr,
     // Blend 100 / 0 - Copy row unchanged.
     LABELALIGN
   "100:                                        \n"
-    "movdqa    " MEMACCESS(1) ",%%xmm0         \n"
+    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
     "sub       $0x10,%2                        \n"
-    MEMOPMEM(movdqa,xmm0,0x00,1,0,1)           //  movdqa    %%xmm0,(%1,%0,1)
+    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)           //  movdqu    %%xmm0,(%1,%0,1)
     "lea       " MEMLEA(0x10,1) ",%1           \n"
     "jg        100b                            \n"
 
@@ -5813,31 +5813,6 @@ void InterpolateRow_Unaligned_SSE2(uint8* dst_ptr, const uint8* src_ptr,
 }
 #endif  // HAS_INTERPOLATEROW_SSE2
 
-#ifdef HAS_HALFROW_SSE2
-void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride,
-                  uint8* dst_uv, int pix) {
-  asm volatile (
-    "sub       %0,%1                           \n"
-    LABELALIGN
-  "1:                                          \n"
-    "movdqa    " MEMACCESS(0) ",%%xmm0         \n"
-    MEMOPREG(pavgb,0x00,0,3,1,xmm0)            //  pavgb     (%0,%3),%%xmm0
-    "sub       $0x10,%2                        \n"
-    MEMOPMEM(movdqa,xmm0,0x00,0,1,1)           //  movdqa    %%xmm0,(%0,%1)
-    "lea       " MEMLEA(0x10,0) ",%0           \n"
-    "jg        1b                              \n"
-  : "+r"(src_uv),  // %0
-    "+r"(dst_uv),  // %1
-    "+r"(pix)      // %2
-  : "r"((intptr_t)(src_uv_stride))  // %3
-  : "memory", "cc"
-#if defined(__SSE2__)
-      , "xmm0"
-#endif
-  );
-}
-#endif  // HAS_HALFROW_SSE2
-
 #ifdef HAS_ARGBTOBAYERROW_SSSE3
 void ARGBToBayerRow_SSSE3(const uint8* src_argb, uint8* dst_bayer,
                           uint32 selector, int pix) {
diff --git a/source/row_win.cc b/source/row_win.cc
index 61602d8..1cf0b9a 100644
--- a/source/row_win.cc
+++ b/source/row_win.cc
@@ -3674,11 +3674,11 @@ void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
 
     align      4
   convertloop:
-    movdqa     xmm0, [eax]
-    movdqa     xmm1, [eax + 16]
+    movdqu     xmm0, [eax]
+    movdqu     xmm1, [eax + 16]
     lea        eax, [eax + 32]
-    movdqa     [edx], xmm0
-    movdqa     [edx + 16], xmm1
+    movdqu     [edx], xmm0
+    movdqu     [edx + 16], xmm1
     lea        edx, [edx + 32]
     sub        ecx, 32
     jg         convertloop
@@ -6541,58 +6541,6 @@ void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr,
 #endif  // HAS_INTERPOLATEROW_SSE2
 
 __declspec(naked) __declspec(align(16))
-void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride,
-                  uint8* dst_uv, int pix) {
-  __asm {
-    push       edi
-    mov        eax, [esp + 4 + 4]    // src_uv
-    mov        edx, [esp + 4 + 8]    // src_uv_stride
-    mov        edi, [esp + 4 + 12]   // dst_v
-    mov        ecx, [esp + 4 + 16]   // pix
-    sub        edi, eax
-
-    align      4
-  convertloop:
-    movdqa     xmm0, [eax]
-    pavgb      xmm0, [eax + edx]
-    sub        ecx, 16
-    movdqa     [eax + edi], xmm0
-    lea        eax,  [eax + 16]
-    jg         convertloop
-    pop        edi
-    ret
-  }
-}
-
-#ifdef HAS_HALFROW_AVX2
-__declspec(naked) __declspec(align(16))
-void HalfRow_AVX2(const uint8* src_uv, int src_uv_stride,
-                  uint8* dst_uv, int pix) {
-  __asm {
-    push       edi
-    mov        eax, [esp + 4 + 4]    // src_uv
-    mov        edx, [esp + 4 + 8]    // src_uv_stride
-    mov        edi, [esp + 4 + 12]   // dst_v
-    mov        ecx, [esp + 4 + 16]   // pix
-    sub        edi, eax
-
-    align      4
-  convertloop:
-    vmovdqu    ymm0, [eax]
-    vpavgb     ymm0, ymm0, [eax + edx]
-    sub        ecx, 32
-    vmovdqu    [eax + edi], ymm0
-    lea        eax,  [eax + 32]
-    jg         convertloop
-
-    pop        edi
-    vzeroupper
-    ret
-  }
-}
-#endif  // HAS_HALFROW_AVX2
-
-__declspec(naked) __declspec(align(16))
 void ARGBToBayerRow_SSSE3(const uint8* src_argb, uint8* dst_bayer,
                           uint32 selector, int pix) {
   __asm {