yuy2 avx2 initial change

source/row_win.cc

 /*
  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS. All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 225 matching lines @@
 // Shuffle table for converting ARGBToRGB24 for I422ToRGB24.  First 8 + next 4
 static const uvec8 kShuffleMaskARGBToRGB24_0 = {
   0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 128u, 128u, 128u, 128u, 10u, 12u, 13u, 14u
 };
 
 // Shuffle table for converting ARGB to RAW.
 static const uvec8 kShuffleMaskARGBToRAW_0 = {
   2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 128u, 128u, 128u, 128u, 8u, 14u, 13u, 12u
 };
 
+// YUY2 shuf 16 Y to 32 Y.
+static const lvec8 kShuffleYUY2Y = {
+  0, 0, 2, 2, 4, 4, 6, 6, 8, 8, 10, 10, 12, 12, 14, 14,
+  0, 0, 2, 2, 4, 4, 6, 6, 8, 8, 10, 10, 12, 12, 14, 14
+};
+
+// YUY2 shuf 8 UV to 16 UV.
+static const lvec8 kShuffleYUY2UV = {
+  1, 3, 1, 3, 5, 7, 5, 7, 9, 11, 9, 11, 13, 15, 13, 15,
+  1, 3, 1, 3, 5, 7, 5, 7, 9, 11, 9, 11, 13, 15, 13, 15
+};
+
+// UYVY shuf 16 Y to 32 Y.
+static const lvec8 kShuffleUYVYY = {
+  1, 1, 3, 3, 5, 5, 7, 7, 9, 9, 11, 11, 13, 13, 15, 15,
+  1, 1, 3, 3, 5, 5, 7, 7, 9, 9, 11, 11, 13, 13, 15, 15
+};
+
+// UYVY shuf 8 UV to 16 UV.
+static const lvec8 kShuffleUYVYUV = {
+  0, 2, 0, 2, 4, 6, 4, 6, 8, 10, 8, 10, 12, 14, 12, 14,
+  0, 2, 0, 2, 4, 6, 4, 6, 8, 10, 8, 10, 12, 14, 12, 14
+};
+
 // Duplicates gray value 3 times and fills in alpha opaque.
 __declspec(naked)
 void J400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int pix) {
   __asm {
     mov        eax, [esp + 4]        // src_y
     mov        edx, [esp + 8]        // dst_argb
     mov        ecx, [esp + 12]       // pix
     pcmpeqb    xmm5, xmm5            // generate mask 0xff000000
     pslld      xmm5, 24
 
@@ skipping 1636 matching lines @@
     __asm vmovdqu    xmm0, [esi]                  /* UV */                     \
     __asm lea        esi,  [esi + 16]                                          \
     __asm vpermq     ymm0, ymm0, 0xd8                                          \
     __asm vpunpcklwd ymm0, ymm0, ymm0             /* UVUV (upsample) */        \
     __asm vmovdqu    xmm4, [eax]                  /* Y */                      \
     __asm vpermq     ymm4, ymm4, 0xd8                                          \
     __asm vpunpcklbw ymm4, ymm4, ymm4                                          \
     __asm lea        eax, [eax + 16]                                           \
   }
 
+// Read 8 YUY2 with 16 Y and upsample 8 UV to 16 UV.
+#define READYUY2_AVX2 __asm {                                                  \
+    __asm vmovdqu    ymm4, [eax]          /* YUY2 */                           \
+    __asm vpshufb    ymm4, ymm4, ymmword ptr kShuffleYUY2Y                     \
+    __asm vmovdqu    ymm0, [eax]          /* UV */                             \
+    __asm vpshufb    ymm0, ymm0, ymmword ptr kShuffleYUY2UV                    \
+    __asm lea        eax, [eax + 32]                                           \
+  }
+
+// Read 8 UYVY with 16 Y and upsample 8 UV to 16 UV.
+#define READUYVY_AVX2 __asm {                                                  \
+    __asm vmovdqu    ymm4, [eax]          /* UYVY */                           \
+    __asm vpshufb    ymm4, ymm4, ymmword ptr kShuffleUYVYY                     \
+    __asm vmovdqu    ymm0, [eax]          /* UV */                             \
+    __asm vpshufb    ymm0, ymm0, ymmword ptr kShuffleUYVYUV                    \
+    __asm lea        eax, [eax + 32]                                           \
+  }
+
 // Convert 16 pixels: 16 UV and 16 Y.
 #define YUVTORGB_AVX2(YuvConstants) __asm {                                    \
     __asm vpmaddubsw ymm2, ymm0, ymmword ptr [YuvConstants + KUVTOR] /* R UV */\
     __asm vpmaddubsw ymm1, ymm0, ymmword ptr [YuvConstants + KUVTOG] /* G UV */\
     __asm vpmaddubsw ymm0, ymm0, ymmword ptr [YuvConstants + KUVTOB] /* B UV */\
     __asm vmovdqu    ymm3, ymmword ptr [YuvConstants + KUVBIASR]               \
     __asm vpsubw     ymm2, ymm3, ymm2                                          \
     __asm vmovdqu    ymm3, ymmword ptr [YuvConstants + KUVBIASG]               \
     __asm vpsubw     ymm1, ymm3, ymm1                                          \
     __asm vmovdqu    ymm3, ymmword ptr [YuvConstants + KUVBIASB]               \
@@ skipping 249 matching lines @@
     jg         convertloop
 
     pop        ebp
     pop        esi
     vzeroupper
     ret
   }
 }
 #endif  // HAS_NV12TOARGBROW_AVX2
 
+// 16 pixels.
+// 8 YUY2 values with 16 Y and 8 UV producing 16 ARGB (64 bytes).
+__declspec(naked)
+void YUY2ToARGBRow_AVX2(const uint8* src_yuy2,
+                        uint8* dst_argb,
+                        struct YuvConstants* yuvconstants,
+                        int width) {
+  __asm {
+    push       ebp
+    mov        eax, [esp + 4 + 4]   // yuy2
+    mov        edx, [esp + 4 + 8]   // argb
+    mov        ebp, [esp + 4 + 12]  // yuvconstants
+    mov        ecx, [esp + 4 + 16]  // width
+    vpcmpeqb   ymm5, ymm5, ymm5     // generate 0xffffffffffffffff for alpha
+
+ convertloop:
+    READYUY2_AVX2
+    YUVTORGB_AVX2(ebp)
+    STOREARGB_AVX2
+
+    sub        ecx, 16
+    jg         convertloop
+
+    pop        ebp
+    vzeroupper
+    ret
+  }
+}
+
+// 16 pixels.
+// 8 UYVY values with 16 Y and 8 UV producing 16 ARGB (64 bytes).
+__declspec(naked)
+void UYVYToARGBRow_AVX2(const uint8* src_uyvy,
+                        uint8* dst_argb,
+                        struct YuvConstants* yuvconstants,
+                        int width) {
+  __asm {
+    push       ebp
+    mov        eax, [esp + 4 + 4]   // uyvy
+    mov        edx, [esp + 4 + 8]   // argb
+    mov        ebp, [esp + 4 + 12]  // yuvconstants
+    mov        ecx, [esp + 4 + 16]  // width
+    vpcmpeqb   ymm5, ymm5, ymm5     // generate 0xffffffffffffffff for alpha
+
+ convertloop:
+    READUYVY_AVX2
+    YUVTORGB_AVX2(ebp)
+    STOREARGB_AVX2
+
+    sub        ecx, 16
+    jg         convertloop
+
+    pop        ebp
+    vzeroupper
+    ret
+  }
+}
+
+
 #ifdef HAS_I422TOBGRAROW_AVX2
 // 16 pixels
 // 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 BGRA (64 bytes).
 // TODO(fbarchard): Use macros to reduce duplicate code.  See SSSE3.
 __declspec(naked)
 void I422ToBGRARow_AVX2(const uint8* y_buf,
                         const uint8* u_buf,
                         const uint8* v_buf,
                         uint8* dst_argb,
                         struct YuvConstants* yuvconstants,
@@ skipping 150 matching lines @@
 // Read 4 UV from NV12, upsample to 8 UV.
 #define READNV12 __asm {                                                       \
     __asm movq       xmm0, qword ptr [esi] /* UV */                            \
     __asm lea        esi,  [esi + 8]                                           \
     __asm punpcklwd  xmm0, xmm0           /* UVUV (upsample) */                \
     __asm movq       xmm4, qword ptr [eax]                                     \
     __asm punpcklbw  xmm4, xmm4                                                \
     __asm lea        eax, [eax + 8]                                            \
   }
 
-// YUY2 shuf 8 Y to 16 Y.
-static const vec8 kShuffleYUY2Y = {
-  0, 0, 2, 2, 4, 4, 6, 6, 8, 8, 10, 10, 12, 12, 14, 14
-};
-
-// YUY2 shuf 4 UV to 8 UV.
-static const vec8 kShuffleYUY2UV = {
-  1, 3, 1, 3, 5, 7, 5, 7, 9, 11, 9, 11, 13, 15, 13, 15
-};
-
-// Read 4 YUY2 with 8 Y and update 4 UV to 8 UV.
+// Read 4 YUY2 with 8 Y and upsample 4 UV to 8 UV.
 #define READYUY2 __asm {                                                       \
     __asm movdqu     xmm4, [eax]          /* YUY2 */                           \
     __asm pshufb     xmm4, xmmword ptr kShuffleYUY2Y                           \
     __asm movdqu     xmm0, [eax]          /* UV */                             \
     __asm pshufb     xmm0, xmmword ptr kShuffleYUY2UV                          \
     __asm lea        eax, [eax + 16]                                           \
   }
 
-// UYVY shuf 8 Y to 16 Y.
-static const vec8 kShuffleUYVYY = {
-  1, 1, 3, 3, 5, 5, 7, 7, 9, 9, 11, 11, 13, 13, 15, 15
-};
-
-// UYVY shuf 4 UV to 8 UV.
-static const vec8 kShuffleUYVYUV = {
-  0, 2, 0, 2, 4, 6, 4, 6, 8, 10, 8, 10, 12, 14, 12, 14
-};
-
-// Read 4 UYVY with 8 Y and update 4 UV to 8 UV.
+// Read 4 UYVY with 8 Y and upsample 4 UV to 8 UV.
 #define READUYVY __asm {                                                       \
     __asm movdqu     xmm4, [eax]          /* UYVY */                           \
     __asm pshufb     xmm4, xmmword ptr kShuffleUYVYY                           \
     __asm movdqu     xmm0, [eax]          /* UV */                             \
     __asm pshufb     xmm0, xmmword ptr kShuffleUYVYUV                          \
     __asm lea        eax, [eax + 16]                                           \
-    __asm lea        eax, [eax + 8]                                            \
   }
 
 // Convert 8 pixels: 8 UV and 8 Y.
 #define YUVTORGB(YuvConstants) __asm {                                         \
     __asm movdqa     xmm1, xmm0                                                \
     __asm movdqa     xmm2, xmm0                                                \
     __asm movdqa     xmm3, xmm0                                                \
     __asm movdqa     xmm0, xmmword ptr [YuvConstants + KUVBIASB]               \
     __asm pmaddubsw  xmm1, xmmword ptr [YuvConstants + KUVTOB]                 \
     __asm psubw      xmm0, xmm1                                                \
@@ skipping 3987 matching lines @@
 }
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(_M_X64)
 #endif  // !defined(LIBYUV_DISABLE_X86) && (defined(_M_IX86) || defined(_M_X64))
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif