neon yuv matrix function

R=harryjin@google.com
BUG=libyuv:488

Review URL: https://codereview.chromium.org/1337973002 .

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1481
+Version: 1482
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -304,7 +304,9 @@ extern "C" {
 #define HAS_I422TOARGB4444ROW_NEON
 #define HAS_I422TOARGBROW_NEON
 // TODO(fbarchard): Implement NEON version
-// #define HAS_I422TOARGBMATRIXROW_NEON
+#ifndef __aarch64__
+#define HAS_I422TOARGBMATRIXROW_NEON
+#endif
 // #define HAS_I422TOABGRMATRIXROW_NEON
 #define HAS_I422TOBGRAROW_NEON
 #define HAS_I422TORAWROW_NEON
@@ -442,6 +444,13 @@ struct YuvConstants {
   lvec16 kYToRgb;
 };
 
+struct YuvConstantsNEON {
+    uvec8 kUVToRB;
+    uvec8 kUVToG;
+    vec16 kUVBiasBGR;
+    vec32 kYToRgb;
+};
+
 #if defined(__APPLE__) || defined(__x86_64__) || defined(__llvm__)
 #define OMITFP
 #else
@@ -541,7 +550,7 @@ void I422ToARGBMatrixRow_NEON(const uint8* src_y,
                               const uint8* src_u,
                               const uint8* src_v,
                               uint8* dst_argb,
-                              struct YuvConstants* YuvConstants,
+                              struct YuvConstantsNEON* YuvConstants,
                               int width);
 void I422ToABGRMatrixRow_NEON(const uint8* src_y,
                               const uint8* src_u,

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1481
+#define LIBYUV_VERSION 1482
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_common.cc

@@ -2276,6 +2276,7 @@ void I422ToUYVYRow_C(const uint8* src_y,
 extern struct YuvConstants kYuvConstants;
 extern struct YuvConstants kYuvJConstants;
 extern struct YuvConstants kYuvHConstants;
+extern struct YuvConstantsNEON kYuvConstantsNEON;
 
 #define ANYYUV(NAMEANY, ANY_SIMD, YUVCONSTANTS)                                \
     void NAMEANY(const uint8* y_buf,                                           \
@@ -2286,6 +2287,12 @@ extern struct YuvConstants kYuvHConstants;
       ANY_SIMD(y_buf, u_buf, v_buf, dst_argb, &YUVCONSTANTS, width);           \
     }
 
+#ifdef HAS_I422TOARGBMATRIXROW_NEON
+ANYYUV(I422ToARGBRow_NEON, I422ToARGBMatrixRow_NEON, kYuvConstantsNEON)
+//ANYYUV(J422ToARGBRow_NEON, I422ToARGBMatrixRow_NEON, kYuvJConstantsNEON)
+//ANYYUV(H422ToARGBRow_NEON, I422ToARGBMatrixRow_NEON, kYuvHConstantsNEON)
+#endif
+
 #ifdef HAS_I422TOARGBMATRIXROW_SSSE3
 ANYYUV(I422ToARGBRow_SSSE3, I422ToARGBMatrixRow_SSSE3, kYuvConstants)
 ANYYUV(J422ToARGBRow_SSSE3, I422ToARGBMatrixRow_SSSE3, kYuvJConstants)

source/row_neon.cc

@@ -150,6 +150,13 @@ extern "C" {
 #define BG (UG * 128 + VG * 128 - YGB)
 #define BR            (VR * 128 - YGB)
 
+YuvConstantsNEON SIMD_ALIGNED(kYuvConstantsNEON) = {
+  { 128, 128, 128, 128, 102, 102, 102, 102, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 25, 25, 25, 25, 52, 52, 52, 52, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { BB, BG, BR, 0, 0, 0, 0, 0 },
+  { 0x0101 * YG, 0, 0, 0 }
+};
+
 static uvec8 kUVToRB  = { 128, 128, 128, 128, 102, 102, 102, 102,
                           0, 0, 0, 0, 0, 0, 0, 0 };
 static uvec8 kUVToG = { 25, 25, 25, 25, 52, 52, 52, 52,
@@ -196,11 +203,12 @@ void I444ToARGBRow_NEON(const uint8* src_y,
   );
 }
 
-void I422ToARGBRow_NEON(const uint8* src_y,
-                        const uint8* src_u,
-                        const uint8* src_v,
-                        uint8* dst_argb,
-                        int width) {
+void I422ToARGBMatrixRow_NEON(const uint8* src_y,
+                              const uint8* src_u,
+                              const uint8* src_v,
+                              uint8* dst_argb,
+                              struct YuvConstantsNEON* YuvConstants,
+                              int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
@@ -216,10 +224,10 @@ void I422ToARGBRow_NEON(const uint8* src_y,
       "+r"(src_v),     // %2
       "+r"(dst_argb),  // %3
       "+r"(width)      // %4
-    : [kUVToRB]"r"(&kUVToRB),   // %5
-      [kUVToG]"r"(&kUVToG),     // %6
-      [kUVBiasBGR]"r"(&kUVBiasBGR),
-      [kYToRgb]"r"(&kYToRgb)
+    : [kUVToRB]"r"(&YuvConstants->kUVToRB),   // %5
+      [kUVToG]"r"(&YuvConstants->kUVToG),     // %6
+      [kUVBiasBGR]"r"(&YuvConstants->kUVBiasBGR),
+      [kYToRgb]"r"(&YuvConstants->kYToRgb)
     : "cc", "memory", "q0", "q1", "q2", "q3", "q4",
       "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
   );