I422ToRGBARow_AVX2 ported to GCC.

BUG=269
TESTED=nacl build
R=brucedawson@google.com

Review URL: https://webrtc-codereview.appspot.com/32259004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1183 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1182
+Version: 1183
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -192,6 +192,7 @@ extern "C" {
 #define HAS_I422TOARGBROW_AVX2
 #define HAS_I422TOBGRAROW_AVX2
 #define HAS_I422TOABGRROW_AVX2
+#define HAS_I422TORGBAROW_AVX2
 #define HAS_YUY2TOYROW_AVX2
 #define HAS_YUY2TOUV422ROW_AVX2
 #define HAS_YUY2TOUVROW_AVX2
@@ -214,7 +215,6 @@ extern "C" {
 // The following are require VS2012.
 // TODO(fbarchard): Port to gcc.
 #if !defined(LIBYUV_DISABLE_X86) && defined(VISUALC_HAS_AVX2)
-#define HAS_I422TORGBAROW_AVX2
 #define HAS_ARGBTOUVROW_AVX2
 #define HAS_ARGBTOYJROW_AVX2
 #define HAS_ARGBTOYROW_AVX2

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1182
+#define LIBYUV_VERSION 1183
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_posix.cc

@@ -2187,6 +2187,7 @@ void OMITFP I422ToARGBRow_AVX2(const uint8* y_buf,
     "lea       " MEMLEA(0x40,[dst_argb]) ",%[dst_argb] \n"
     "sub       $0x10,%[width]                  \n"
     "jg        1b                              \n"
+    "vzeroupper                                \n"
   : [y_buf]"+r"(y_buf),    // %[y_buf]
     [u_buf]"+r"(u_buf),    // %[u_buf]
     [v_buf]"+r"(v_buf),    // %[v_buf]
@@ -2206,7 +2207,7 @@ void OMITFP I422ToARGBRow_AVX2(const uint8* y_buf,
 
 #if defined(HAS_I422TOABGRROW_AVX2)
 // 16 pixels
-// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes).
+// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ABGR (64 bytes).
 void OMITFP I422ToABGRRow_AVX2(const uint8* y_buf,
                                const uint8* u_buf,
                                const uint8* v_buf,
@@ -2233,6 +2234,7 @@ void OMITFP I422ToABGRRow_AVX2(const uint8* y_buf,
     "lea       " MEMLEA(0x40,[dst_argb]) ",%[dst_argb] \n"
     "sub       $0x10,%[width]                  \n"
     "jg        1b                              \n"
+    "vzeroupper                                \n"
   : [y_buf]"+r"(y_buf),    // %[y_buf]
     [u_buf]"+r"(u_buf),    // %[u_buf]
     [v_buf]"+r"(v_buf),    // %[v_buf]
@@ -2248,7 +2250,54 @@ void OMITFP I422ToABGRRow_AVX2(const uint8* y_buf,
 #endif
   );
 }
-#endif  // HAS_I422TOARGBROW_AVX2
+#endif  // HAS_I422TOABGRROW_AVX2
+
+#if defined(HAS_I422TORGBAROW_AVX2)
+// 16 pixels
+// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 RGBA (64 bytes).
+void OMITFP I422ToRGBARow_AVX2(const uint8* y_buf,
+                               const uint8* u_buf,
+                               const uint8* v_buf,
+                               uint8* dst_argb,
+                               int width) {
+  asm volatile (
+    "sub       %[u_buf],%[v_buf]               \n"
+    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
+    "vpxor      %%ymm4,%%ymm4,%%ymm4           \n"
+    LABELALIGN
+  "1:                                          \n"
+    READYUV422_AVX2
+    YUVTORGB_AVX2
+
+    // Step 3: Weave into RGBA
+    "vpunpcklbw %%ymm2,%%ymm1,%%ymm1           \n"
+    "vpermq     $0xd8,%%ymm1,%%ymm1            \n"
+    "vpunpcklbw %%ymm0,%%ymm5,%%ymm2           \n"
+    "vpermq     $0xd8,%%ymm2,%%ymm2            \n"
+    "vpunpcklwd %%ymm1,%%ymm2,%%ymm0           \n"
+    "vpunpckhwd %%ymm1,%%ymm2,%%ymm1           \n"
+    "vmovdqu    %%ymm0," MEMACCESS([dst_argb]) "\n"
+    "vmovdqu    %%ymm1," MEMACCESS2(0x20,[dst_argb]) "\n"
+    "lea       " MEMLEA(0x40,[dst_argb]) ",%[dst_argb] \n"
+    "sub       $0x10,%[width]                  \n"
+    "jg        1b                              \n"
+    "vzeroupper                                \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_AVX.kUVToB_AVX)  // %[kYuvConstants]
+  : "memory", "cc"
+#if defined(__native_client__) && defined(__x86_64__)
+    , "r14"
+#endif
+#if defined(__SSE2__)
+    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
+#endif
+  );
+}
+#endif  // HAS_I422TORGBAROW_AVX2
 
 #ifdef HAS_YTOARGBROW_SSE2
 void YToARGBRow_SSE2(const uint8* y_buf,