Port I444ToARGB to AVX2.

BUG=403
TESTED=I444ToARGB unittests
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/45589004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1314 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1313
+Version: 1314
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -198,7 +198,7 @@ extern "C" {
 #define HAS_I422TORGB565ROW_AVX2
 #define HAS_I422TOARGB1555ROW_AVX2
 #define HAS_I422TOARGB4444ROW_AVX2
-
+#define HAS_I444TOARGBROW_AVX2
 // TODO(fbarchard): Port to Neon
 #define HAS_ARGBTORGB565DITHERROW_SSE2
 #define HAS_ARGBTORGB565DITHERROW_AVX2
@@ -1046,6 +1046,11 @@ void I444ToARGBRow_SSSE3(const uint8* src_y,
                          const uint8* src_v,
                          uint8* dst_argb,
                          int width);
+void I444ToARGBRow_AVX2(const uint8* src_y,
+                        const uint8* src_u,
+                        const uint8* src_v,
+                        uint8* dst_argb,
+                        int width);
 void I422ToARGBRow_SSSE3(const uint8* src_y,
                          const uint8* src_u,
                          const uint8* src_v,
@@ -1185,6 +1190,11 @@ void I444ToARGBRow_Any_SSSE3(const uint8* src_y,
                              const uint8* src_v,
                              uint8* dst_argb,
                              int width);
+void I444ToARGBRow_Any_AVX2(const uint8* src_y,
+                            const uint8* src_u,
+                            const uint8* src_v,
+                            uint8* dst_argb,
+                            int width);
 void I422ToARGBRow_Any_SSSE3(const uint8* src_y,
                              const uint8* src_u,
                              const uint8* src_v,

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1313
+#define LIBYUV_VERSION 1314
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert_argb.cc

@@ -85,6 +85,14 @@ int I444ToARGB(const uint8* src_y, int src_stride_y,
     }
   }
 #endif
+#if defined(HAS_I444TOARGBROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    I444ToARGBRow = I444ToARGBRow_Any_AVX2;
+    if (IS_ALIGNED(width, 16)) {
+      I444ToARGBRow = I444ToARGBRow_AVX2;
+    }
+  }
+#endif
 #if defined(HAS_I444TOARGBROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     I444ToARGBRow = I444ToARGBRow_Any_NEON;

source/row_any.cc

@@ -73,6 +73,9 @@ YANY(I422ToRGBARow_Any_AVX2, I422ToRGBARow_AVX2, I422ToRGBARow_C, 1, 4, 15)
 #ifdef HAS_I422TOABGRROW_AVX2
 YANY(I422ToABGRRow_Any_AVX2, I422ToABGRRow_AVX2, I422ToABGRRow_C, 1, 4, 15)
 #endif
+#ifdef HAS_I444TOARGBROW_AVX2
+YANY(I444ToARGBRow_Any_AVX2, I444ToARGBRow_AVX2, I444ToARGBRow_C, 0, 4, 15)
+#endif
 #ifdef HAS_I422TOARGB4444ROW_AVX2
 YANY(I422ToARGB4444Row_Any_AVX2, I422ToARGB4444Row_AVX2, I422ToARGB4444Row_C,
      1, 2, 7)

source/row_win.cc

@@ -1678,6 +1678,16 @@ void RGBAToUVRow_SSSE3(const uint8* src_argb0, int src_stride_argb,
 }
 #endif  // HAS_ARGBTOYROW_SSSE3
 
+// Read 16 UV from 444
+#define READYUV444_AVX2 __asm {                                                \
+    __asm vmovdqu    xmm0, [esi]                  /* U */         /* NOLINT */ \
+    __asm vmovdqu    xmm1, [esi + edi]            /* V */         /* NOLINT */ \
+    __asm lea        esi,  [esi + 16]                                          \
+    __asm vpermq     ymm0, ymm0, 0xd8                                          \
+    __asm vpermq     ymm1, ymm1, 0xd8                                          \
+    __asm vpunpcklbw ymm0, ymm0, ymm1             /* UV */                     \
+  }
+
 // Read 8 UV from 422, upsample to 16 UV.
 #define READYUV422_AVX2 __asm {                                                \
     __asm vmovq      xmm0, qword ptr [esi]        /* U */         /* NOLINT */ \
@@ -1775,6 +1785,44 @@ void I422ToARGBRow_AVX2(const uint8* y_buf,
 }
 #endif  // HAS_I422TOARGBROW_AVX2
 
+#ifdef HAS_I444TOARGBROW_AVX2
+// 16 pixels
+// 16 UV values with 16 Y producing 16 ARGB (64 bytes).
+__declspec(naked) __declspec(align(16))
+void I444ToARGBRow_AVX2(const uint8* y_buf,
+                        const uint8* u_buf,
+                        const uint8* v_buf,
+                        uint8* dst_argb,
+                        int width) {
+  __asm {
+    push       esi
+    push       edi
+    mov        eax, [esp + 8 + 4]   // Y
+    mov        esi, [esp + 8 + 8]   // U
+    mov        edi, [esp + 8 + 12]  // V
+    mov        edx, [esp + 8 + 16]  // argb
+    mov        ecx, [esp + 8 + 20]  // width
+    sub        edi, esi
+    vpcmpeqb   ymm5, ymm5, ymm5     // generate 0xffffffffffffffff for alpha
+
+ convertloop:
+    READYUV444_AVX2
+    YUVTORGB_AVX2(kYuvConstants)
+    STOREARGB_AVX2
+
+    sub        ecx, 16
+    jg         convertloop
+
+    pop        edi
+    pop        esi
+    vzeroupper
+    ret
+  }
+}
+#endif  // HAS_I444TOARGBROW_AVX2
+
+
+
 #ifdef HAS_NV12TOARGBROW_AVX2
 // 16 pixels.
 // 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes).

unit_test/color_test.cc

@@ -185,13 +185,7 @@ static void YToRGB(int y, int* r, int* g, int* b) {
 
 static int RoundToByte(double f) {
   int i = static_cast<int>(f + 0.5);
-  if (i < 0) {
-    i = 0;
-  }
-  if (i > 255) {
-    i = 255;
-  }
-  return i;
+  return (i & ~0xff) ?  (int)(((int32)(-i) >> 31) & 0xff) : i;
 }
 
 static void YUVToRGBReference(int y, int u, int v, int* r, int* g, int* b) {