Neon YToARGB and fix SSE2 to match C version

BUG=none
TEST=YToARGB_Opt
Review URL: https://webrtc-codereview.appspot.com/966007

git-svn-id: http://libyuv.googlecode.com/svn/trunk@466 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 465
+Version: 466
 License: BSD
 License File: LICENSE
 

include/libyuv/convert_argb.h

@@ -75,6 +75,9 @@ int I400ToARGB(const uint8* src_y, int src_stride_y,
                uint8* dst_argb, int dst_stride_argb,
                int width, int height);
 
+// Alias.
+#define YToARGB I400ToARGB_Reference
+
 // Convert I400 to ARGB. Reverse of ARGBToI400.
 LIBYUV_API
 int I400ToARGB_Reference(const uint8* src_y, int src_stride_y,

include/libyuv/row.h

@@ -183,6 +183,7 @@ extern "C" {
 #define HAS_ARGBTOARGB4444ROW_NEON
 #define HAS_ARGBTOYROW_NEON
 #define HAS_MERGEUV_NEON
+#define HAS_YTOARGBROW_NEON
 #endif
 
 // The following are available on Mips platforms
@@ -718,6 +719,9 @@ void I422ToRAWRow_Any_SSSE3(const uint8* y_buf,
 void YToARGBRow_SSE2(const uint8* y_buf,
                      uint8* argb_buf,
                      int width);
+void YToARGBRow_NEON(const uint8* y_buf,
+                     uint8* argb_buf,
+                     int width);
 
 // ARGB preattenuated alpha blend.
 void ARGBBlendRow_SSSE3(const uint8* src_argb0, const uint8* src_argb1,

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 465
+#define LIBYUV_VERSION 466
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert_argb.cc

@@ -220,6 +220,10 @@ int I400ToARGB_Reference(const uint8* src_y, int src_stride_y,
       IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
     YToARGBRow = YToARGBRow_SSE2;
   }
+#elif defined(HAS_YTOARGBROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
+    YToARGBRow = YToARGBRow_NEON;
+  }
 #endif
 
   for (int y = 0; y < height; ++y) {

source/row_posix.cc

@@ -2334,12 +2334,13 @@ void YToARGBRow_SSE2(const uint8* y_buf,
                      uint8* rgb_buf,
                      int width) {
   asm volatile (
+    "pxor      %%xmm5,%%xmm5                   \n"
     "pcmpeqb   %%xmm4,%%xmm4                   \n"
     "pslld     $0x18,%%xmm4                    \n"
-    "mov       $0x10001000,%%eax               \n"
+    "mov       $0x00100010,%%eax               \n"
     "movd      %%eax,%%xmm3                    \n"
     "pshufd    $0x0,%%xmm3,%%xmm3              \n"
-    "mov       $0x012a012a,%%eax               \n"
+    "mov       $0x004a004a,%%eax               \n"
     "movd      %%eax,%%xmm2                    \n"
     "pshufd    $0x0,%%xmm2,%%xmm2              \n"
     ".p2align  4                               \n"
@@ -2347,9 +2348,10 @@ void YToARGBRow_SSE2(const uint8* y_buf,
     // Step 1: Scale Y contribution to 8 G values. G = (y - 16) * 1.164
     "movq      (%0),%%xmm0                     \n"
     "lea       0x8(%0),%0                      \n"
-    "punpcklbw %%xmm0,%%xmm0                   \n"
+    "punpcklbw %%xmm5,%%xmm0                   \n"
     "psubusw   %%xmm3,%%xmm0                   \n"
-    "pmulhuw   %%xmm2,%%xmm0                   \n"
+    "pmullw    %%xmm2,%%xmm0                   \n"
+    "psrlw     $6, %%xmm0                      \n"
     "packuswb  %%xmm0,%%xmm0                   \n"
 
     // Step 2: Weave into ARGB

source/row_win.cc

@@ -2483,12 +2483,13 @@ void YToARGBRow_SSE2(const uint8* y_buf,
                      uint8* rgb_buf,
                      int width) {
   __asm {
+    pxor       xmm5, xmm5
     pcmpeqb    xmm4, xmm4           // generate mask 0xff000000
     pslld      xmm4, 24
-    mov        eax,0x10001000
+    mov        eax,0x00100010
     movd       xmm3,eax
     pshufd     xmm3,xmm3,0
-    mov        eax,0x012a012a
+    mov        eax,0x004a004a       // 74
     movd       xmm2,eax
     pshufd     xmm2,xmm2,0
     mov        eax, [esp + 4]       // Y
@@ -2500,9 +2501,10 @@ void YToARGBRow_SSE2(const uint8* y_buf,
     // Step 1: Scale Y contribution to 8 G values. G = (y - 16) * 1.164
     movq       xmm0, qword ptr [eax]
     lea        eax, [eax + 8]
-    punpcklbw  xmm0, xmm0           // Y.Y
+    punpcklbw  xmm0, xmm5           // 0.Y
     psubusw    xmm0, xmm3
-    pmulhuw    xmm0, xmm2
+    pmullw     xmm0, xmm2
+    psrlw      xmm0, 6
     packuswb   xmm0, xmm0           // G
 
     // Step 2: Weave into ARGB

unit_test/convert_test.cc

@@ -707,6 +707,7 @@ TESTATOB(BayerGRBG, 1, 1, ARGB, 4, 0)
 TESTATOB(I400, 1, 1, ARGB, 4, 0)
 TESTATOB(I400, 1, 1, I400, 1, 0)
 TESTATOB(I400, 1, 1, I400Mirror, 1, 0)
+TESTATOB(Y, 1, 1, ARGB, 4, 0)
 TESTATOB(ARGB, 4, 4, ARGBMirror, 4, 0)
 
 #define TESTATOBRANDOM(FMT_A, BPP_A, STRIDE_A, FMT_B, BPP_B, STRIDE_B, DIFF)   \