Remove I411 support, update doc and switch to side by side test

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 1951 matching lines @@
     __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]                                           \
     __asm vmovdqu    xmm5, [ebp]                  /* A */                      \
     __asm vpermq     ymm5, ymm5, 0xd8                                          \
     __asm lea        ebp, [ebp + 16]                                           \
   }
 
-// Read 4 UV from 411, upsample to 16 UV.
-#define READYUV411_AVX2 __asm {                                                \
-    __asm vmovd      xmm0, dword ptr [esi]        /* U */                      \
-    __asm vmovd      xmm1, dword ptr [esi + edi]  /* V */                      \
-    __asm lea        esi,  [esi + 4]                                           \
-    __asm vpunpcklbw ymm0, ymm0, ymm1             /* UV */                     \
-    __asm vpunpcklwd ymm0, ymm0, ymm0             /* UVUV (upsample) */        \
-    __asm vpermq     ymm0, ymm0, 0xd8                                          \
-    __asm vpunpckldq ymm0, ymm0, ymm0             /* UVUVUVUV (upsample) */    \
-    __asm vmovdqu    xmm4, [eax]                  /* Y */                      \
-    __asm vpermq     ymm4, ymm4, 0xd8                                          \
-    __asm vpunpcklbw ymm4, ymm4, ymm4                                          \
-    __asm lea        eax, [eax + 16]                                           \
-  }
-
 // Read 8 UV from NV12, upsample to 16 UV.
 #define READNV12_AVX2 __asm {                                                  \
     __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]                                           \
@@ skipping 194 matching lines @@
 
     pop        ebx
     pop        edi
     pop        esi
     vzeroupper
     ret
   }
 }
 #endif  // HAS_I444TOARGBROW_AVX2
 
-#ifdef HAS_I411TOARGBROW_AVX2
-// 16 pixels
-// 4 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes).
-__declspec(naked)
-void I411ToARGBRow_AVX2(const uint8* y_buf,
-                        const uint8* u_buf,
-                        const uint8* v_buf,
-                        uint8* dst_argb,
-                        const struct YuvConstants* yuvconstants,
-                        int width) {
-  __asm {
-    push       esi
-    push       edi
-    push       ebx
-    mov        eax, [esp + 12 + 4]   // Y
-    mov        esi, [esp + 12 + 8]   // U
-    mov        edi, [esp + 12 + 12]  // V
-    mov        edx, [esp + 12 + 16]  // abgr
-    mov        ebx, [esp + 12 + 20]  // yuvconstants
-    mov        ecx, [esp + 12 + 24]  // width
-    sub        edi, esi
-    vpcmpeqb   ymm5, ymm5, ymm5     // generate 0xffffffffffffffff for alpha
-
- convertloop:
-    READYUV411_AVX2
-    YUVTORGB_AVX2(ebx)
-    STOREARGB_AVX2
-
-    sub        ecx, 16
-    jg         convertloop
-
-    pop        ebx
-    pop        edi
-    pop        esi
-    vzeroupper
-    ret
-  }
-}
-#endif  // HAS_I411TOARGBROW_AVX2
-
 #ifdef HAS_NV12TOARGBROW_AVX2
 // 16 pixels.
 // 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes).
 __declspec(naked)
 void NV12ToARGBRow_AVX2(const uint8* y_buf,
                         const uint8* uv_buf,
                         uint8* dst_argb,
                         const struct YuvConstants* yuvconstants,
                         int width) {
   __asm {
@@ skipping 193 matching lines @@
     __asm lea        esi,  [esi + 4]                                           \
     __asm punpcklbw  xmm0, xmm1           /* UV */                             \
     __asm punpcklwd  xmm0, xmm0           /* UVUV (upsample) */                \
     __asm movq       xmm4, qword ptr [eax]   /* Y */                           \
     __asm punpcklbw  xmm4, xmm4                                                \
     __asm lea        eax, [eax + 8]                                            \
     __asm movq       xmm5, qword ptr [ebp]   /* A */                           \
     __asm lea        ebp, [ebp + 8]                                            \
   }
 
-// Read 2 UV from 411, upsample to 8 UV.
-// drmemory fails with memory fault if pinsrw used. libyuv bug: 525
-//  __asm pinsrw     xmm0, [esi], 0        /* U */
-//  __asm pinsrw     xmm1, [esi + edi], 0  /* V */
-#define READYUV411_EBX __asm {                                                 \
-    __asm movzx      ebx, word ptr [esi]        /* U */                        \
-    __asm movd       xmm0, ebx                                                 \
-    __asm movzx      ebx, word ptr [esi + edi]  /* V */                        \
-    __asm movd       xmm1, ebx                                                 \
-    __asm lea        esi,  [esi + 2]                                           \
-    __asm punpcklbw  xmm0, xmm1            /* UV */                            \
-    __asm punpcklwd  xmm0, xmm0            /* UVUV (upsample) */               \
-    __asm punpckldq  xmm0, xmm0            /* UVUVUVUV (upsample) */           \
-    __asm movq       xmm4, qword ptr [eax]                                     \
-    __asm punpcklbw  xmm4, xmm4                                                \
-    __asm lea        eax, [eax + 8]                                            \
-  }
-
 // 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]                                            \
   }
 
@@ skipping 331 matching lines @@
 
     pop        ebp
     pop        ebx
     pop        edi
     pop        esi
     ret
   }
 }
 
 // 8 pixels.
-// 2 UV values upsampled to 8 UV, mixed with 8 Y producing 8 ARGB (32 bytes).
-// Similar to I420 but duplicate UV once more.
-__declspec(naked)
-void I411ToARGBRow_SSSE3(const uint8* y_buf,
-                         const uint8* u_buf,
-                         const uint8* v_buf,
-                         uint8* dst_argb,
-                         const struct YuvConstants* yuvconstants,
-                         int width) {
-  __asm {
-    push       esi
-    push       edi
-    push       ebx
-    push       ebp
-    mov        eax, [esp + 16 + 4]   // Y
-    mov        esi, [esp + 16 + 8]   // U
-    mov        edi, [esp + 16 + 12]  // V
-    mov        edx, [esp + 16 + 16]  // abgr
-    mov        ebp, [esp + 16 + 20]  // yuvconstants
-    mov        ecx, [esp + 16 + 24]  // width
-    sub        edi, esi
-    pcmpeqb    xmm5, xmm5            // generate 0xffffffff for alpha
-
- convertloop:
-    READYUV411_EBX
-    YUVTORGB(ebp)
-    STOREARGB
-
-    sub        ecx, 8
-    jg         convertloop
-
-    pop        ebp
-    pop        ebx
-    pop        edi
-    pop        esi
-    ret
-  }
-}
-
-// 8 pixels.
 // 4 UV values upsampled to 8 UV, mixed with 8 Y producing 8 ARGB (32 bytes).
 __declspec(naked)
 void NV12ToARGBRow_SSSE3(const uint8* y_buf,
                          const uint8* uv_buf,
                          uint8* dst_argb,
                          const struct YuvConstants* yuvconstants,
                          int width) {
   __asm {
     push       esi
     push       ebx
@@ skipping 3454 matching lines @@
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(_M_X64)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif
 
 #endif  // !defined(LIBYUV_DISABLE_X86) && (defined(_M_IX86) || defined(_M_X64))