port I444 to ARGB to matrix

source/row_gcc.cc

 // VERSION 2
 /*
  *  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 1566 matching lines @@
     "pcmpeqb   %%xmm5,%%xmm5                                     \n"           \
     "punpcklbw %%xmm2,%%xmm1                                     \n"           \
     "punpcklbw %%xmm0,%%xmm5                                     \n"           \
     "movdqa    %%xmm5,%%xmm0                                     \n"           \
     "punpcklwd %%xmm1,%%xmm5                                     \n"           \
     "punpckhwd %%xmm1,%%xmm0                                     \n"           \
     "movdqu    %%xmm5," MEMACCESS([dst_rgba]) "                  \n"           \
     "movdqu    %%xmm0," MEMACCESS2(0x10, [dst_rgba]) "           \n"           \
     "lea       " MEMLEA(0x20, [dst_rgba]) ",%[dst_rgba]          \n"
 
-void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf,
-                                const uint8* u_buf,
-                                const uint8* v_buf,
-                                uint8* dst_argb,
-                                int width) {
+void OMITFP I444ToARGBMatrixRow_SSSE3(const uint8* y_buf,
+                                      const uint8* u_buf,
+                                      const uint8* v_buf,
+                                      uint8* dst_argb,
+                                      struct YuvConstants* YuvConstants,
+                                      int width) {
   asm volatile (
     "sub       %[u_buf],%[v_buf]               \n"
     "pcmpeqb   %%xmm5,%%xmm5                   \n"
     LABELALIGN
   "1:                                          \n"
     READYUV444
-    YUVTORGB(kYuvConstants)
+    YUVTORGB(YuvConstants)
     STOREARGB
     "sub       $0x8,%[width]                   \n"
     "jg        1b                              \n"
   : [y_buf]"+r"(y_buf),    // %[y_buf]
     [u_buf]"+r"(u_buf),    // %[u_buf]
     [v_buf]"+r"(v_buf),    // %[v_buf]
     [dst_argb]"+r"(dst_argb),  // %[dst_argb]
     [width]"+rm"(width)    // %[width]
-  : [kYuvConstants]"r"(&kYuvConstants.kUVToB) // %[kYuvConstants]
+  : [YuvConstants]"r"(YuvConstants)  // %[kYuvConstants]
   : "memory", "cc", NACL_R14
     "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
   );
+}
+
+void OMITFP I444ToABGRMatrixRow_SSSE3(const uint8* y_buf,
+                                      const uint8* u_buf,
+                                      const uint8* v_buf,
+                                      uint8* dst_abgr,
+                                      struct YuvConstants* YuvConstants,
+                                      int width) {
+  asm volatile (
+    "sub       %[u_buf],%[v_buf]               \n"
+    "pcmpeqb   %%xmm5,%%xmm5                   \n"
+    LABELALIGN
+  "1:                                          \n"
+    READYUV444
+    YUVTORGB(YuvConstants)
+    STOREABGR
+    "sub       $0x8,%[width]                   \n"
+    "jg        1b                              \n"
+  : [y_buf]"+r"(y_buf),    // %[y_buf]
+    [u_buf]"+r"(u_buf),    // %[u_buf]
+    [v_buf]"+r"(v_buf),    // %[v_buf]
+    [dst_abgr]"+r"(dst_abgr),  // %[dst_abgr]
+    [width]"+rm"(width)    // %[width]
+  : [YuvConstants]"r"(YuvConstants)  // %[kYuvConstants]
+  : "memory", "cc", NACL_R14
+    "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
+  );
 }
 
 // TODO(fbarchard): Consider putting masks into constants.
 void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf,
                                  const uint8* u_buf,
                                  const uint8* v_buf,
                                  uint8* dst_rgb24,
                                  int width) {
   asm volatile (
     "movdqa    %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n"
@@ skipping 83 matching lines @@
                                       const uint8* v_buf,
                                       uint8* dst_argb,
                                       struct YuvConstants* YuvConstants,
                                       int width) {
   asm volatile (
     "sub       %[u_buf],%[v_buf]               \n"
     "pcmpeqb   %%xmm5,%%xmm5                   \n"
     LABELALIGN
   "1:                                          \n"
     READYUV422
-    YUVTORGB(kYuvConstants)
+    YUVTORGB(YuvConstants)
     STOREARGB
     "sub       $0x8,%[width]                   \n"
     "jg        1b                              \n"
   : [y_buf]"+r"(y_buf),    // %[y_buf]
     [u_buf]"+r"(u_buf),    // %[u_buf]
     [v_buf]"+r"(v_buf),    // %[v_buf]
     [dst_argb]"+r"(dst_argb),  // %[dst_argb]
     [width]"+rm"(width)    // %[width]
-  : [kYuvConstants]"r"(YuvConstants)  // %[kYuvConstants]
+  : [YuvConstants]"r"(YuvConstants)  // %[kYuvConstants]
   : "memory", "cc", NACL_R14
     "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
   );
 }
 
 void OMITFP I411ToARGBRow_SSSE3(const uint8* y_buf,
                                 const uint8* u_buf,
                                 const uint8* v_buf,
                                 uint8* dst_argb,
                                 int width) {
@@ skipping 3615 matching lines @@
   );
 }
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(__x86_64__) || defined(__i386__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif