Raw 24 bit RGB to RGB24 (bgr)

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 208 matching lines @@
 // Shuffle table for converting RGB24 to ARGB.
 static const uvec8 kShuffleMaskRGB24ToARGB = {
   0u, 1u, 2u, 12u, 3u, 4u, 5u, 13u, 6u, 7u, 8u, 14u, 9u, 10u, 11u, 15u
 };
 
 // Shuffle table for converting RAW to ARGB.
 static const uvec8 kShuffleMaskRAWToARGB = {
   2u, 1u, 0u, 12u, 5u, 4u, 3u, 13u, 8u, 7u, 6u, 14u, 11u, 10u, 9u, 15u
 };
 
+// Shuffle table for converting RAW to RGB24.  First 8.
+static const uvec8 kShuffleMaskRAWToRGB24_0 = {
+  2u, 1u, 0u, 5u, 4u, 3u, 8u, 7u,
+  128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u
+};
+
+// Shuffle table for converting RAW to RGB24.  Middle 8.
+static const uvec8 kShuffleMaskRAWToRGB24_1 = {
+  2u, 7u, 6u, 5u, 10u, 9u, 8u, 13u,
+  128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u
+};
+
+// Shuffle table for converting RAW to RGB24.  Last 8.
+static const uvec8 kShuffleMaskRAWToRGB24_2 = {
+  8u, 7u, 12u, 11u, 10u, 15u, 14u, 13u,
+  128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u
+};
+
 // Shuffle table for converting ARGB to RGB24.
 static const uvec8 kShuffleMaskARGBToRGB24 = {
   0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 10u, 12u, 13u, 14u, 128u, 128u, 128u, 128u
 };
 
 // Shuffle table for converting ARGB to RAW.
 static const uvec8 kShuffleMaskARGBToRAW = {
   2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 8u, 14u, 13u, 12u, 128u, 128u, 128u, 128u
 };
 
 // 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
@@ skipping 147 matching lines @@
     movdqu    [edx + 16], xmm1
     por       xmm3, xmm5
     movdqu    [edx + 48], xmm3
     lea       edx, [edx + 64]
     sub       ecx, 16
     jg        convertloop
     ret
   }
 }
 
+__declspec(naked)
+void RAWToRGB24Row_SSSE3(const uint8* src_raw, uint8* dst_rgb24, int width) {
+  __asm {
+    mov       eax, [esp + 4]   // src_raw
+    mov       edx, [esp + 8]   // dst_rgb24
+    mov       ecx, [esp + 12]  // width
+    movdqa    xmm3, xmmword ptr kShuffleMaskRAWToRGB24_0
+    movdqa    xmm4, xmmword ptr kShuffleMaskRAWToRGB24_1
+    movdqa    xmm5, xmmword ptr kShuffleMaskRAWToRGB24_2
+
+ convertloop:
+    movdqu    xmm0, [eax]
+    movdqu    xmm1, [eax + 4]
+    movdqu    xmm2, [eax + 8]
+    lea       eax, [eax + 24]
+    pshufb    xmm0, xmm3
+    pshufb    xmm1, xmm4
+    pshufb    xmm2, xmm5
+    movq      qword ptr [edx], xmm0
+    movq      qword ptr [edx + 8], xmm1
+    movq      qword ptr [edx + 16], xmm2
+    lea       edx, [edx + 24]
+    sub       ecx, 8
+    jg        convertloop
+    ret
+  }
+}
+
 // pmul method to replicate bits.
 // Math to replicate bits:
 // (v << 8) | (v << 3)
 // v * 256 + v * 8
 // v * (256 + 8)
 // G shift of 5 is incorporated, so shift is 5 + 8 and 5 + 3
 // 20 instructions.
 __declspec(naked)
 void RGB565ToARGBRow_SSE2(const uint8* src_rgb565, uint8* dst_argb,
                           int width) {
@@ skipping 5761 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