Add MSA optimized ARGBExtractAlpha, ARGBBlend, ARGBQuantize and ARGBColorMatrix row functions

TBR=kjellander@chromium.org
R=fbarchard@google.com

Bug:libyuv:634
Change-Id: I17bd3f87336f613ad363af7d7b9d7af49d725e56
Reviewed-on: https://chromium-review.googlesource.com/613100Reviewed-by: Frank Barchard <fbarchard@google.com>

include/libyuv/row.h

@@ -421,6 +421,7 @@ extern "C" {
 #define HAS_YUY2TOUV422ROW_MSA
 #define HAS_YUY2TOUVROW_MSA
 #define HAS_YUY2TOYROW_MSA
+#define HAS_ARGBEXTRACTALPHAROW_MSA
 
 #ifndef DISABLE_CLANG_MSA
 #define HAS_ABGRTOUVROW_MSA
@@ -463,6 +464,9 @@ extern "C" {
 #define HAS_SOBELXYROW_MSA
 #define HAS_UYVYTOARGBROW_MSA
 #define HAS_YUY2TOARGBROW_MSA
+#define HAS_ARGBBLENDROW_MSA
+#define HAS_ARGBQUANTIZEROW_MSA
+#define HAS_ARGBCOLORMATRIXROW_MSA
 #endif
 #endif
 
@@ -1467,6 +1471,7 @@ void ARGBExtractAlphaRow_C(const uint8* src_argb, uint8* dst_a, int width);
 void ARGBExtractAlphaRow_SSE2(const uint8* src_argb, uint8* dst_a, int width);
 void ARGBExtractAlphaRow_AVX2(const uint8* src_argb, uint8* dst_a, int width);
 void ARGBExtractAlphaRow_NEON(const uint8* src_argb, uint8* dst_a, int width);
+void ARGBExtractAlphaRow_MSA(const uint8* src_argb, uint8* dst_a, int width);
 void ARGBExtractAlphaRow_Any_SSE2(const uint8* src_argb,
                                   uint8* dst_a,
                                   int width);
@@ -1476,6 +1481,9 @@ void ARGBExtractAlphaRow_Any_AVX2(const uint8* src_argb,
 void ARGBExtractAlphaRow_Any_NEON(const uint8* src_argb,
                                   uint8* dst_a,
                                   int width);
+void ARGBExtractAlphaRow_Any_MSA(const uint8* src_argb,
+                                 uint8* dst_a,
+                                 int width);
 
 void ARGBCopyYToAlphaRow_C(const uint8* src_y, uint8* dst_argb, int width);
 void ARGBCopyYToAlphaRow_SSE2(const uint8* src_y, uint8* dst_argb, int width);
@@ -2135,6 +2143,10 @@ void ARGBBlendRow_NEON(const uint8* src_argb,
                        const uint8* src_argb1,
                        uint8* dst_argb,
                        int width);
+void ARGBBlendRow_MSA(const uint8* src_argb,
+                      const uint8* src_argb1,
+                      uint8* dst_argb,
+                      int width);
 void ARGBBlendRow_C(const uint8* src_argb,
                     const uint8* src_argb1,
                     uint8* dst_argb,
@@ -2848,6 +2860,10 @@ void ARGBColorMatrixRow_NEON(const uint8* src_argb,
                              uint8* dst_argb,
                              const int8* matrix_argb,
                              int width);
+void ARGBColorMatrixRow_MSA(const uint8* src_argb,
+                            uint8* dst_argb,
+                            const int8* matrix_argb,
+                            int width);
 
 void ARGBColorTableRow_C(uint8* dst_argb, const uint8* table_argb, int width);
 void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, int width);
@@ -2870,6 +2886,11 @@ void ARGBQuantizeRow_NEON(uint8* dst_argb,
                           int interval_size,
                           int interval_offset,
                           int width);
+void ARGBQuantizeRow_MSA(uint8* dst_argb,
+                         int scale,
+                         int interval_size,
+                         int interval_offset,
+                         int width);
 
 void ARGBShadeRow_C(const uint8* src_argb,
                     uint8* dst_argb,

source/planar_functions.cc

@@ -845,6 +845,11 @@ ARGBBlendRow GetARGBBlend() {
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBBlendRow = ARGBBlendRow_NEON;
   }
+#endif
+#if defined(HAS_ARGBBLENDROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    ARGBBlendRow = ARGBBlendRow_MSA;
+  }
 #endif
   return ARGBBlendRow;
 }
@@ -1973,6 +1978,11 @@ int ARGBColorMatrix(const uint8* src_argb,
   if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
     ARGBColorMatrixRow = ARGBColorMatrixRow_NEON;
   }
+#endif
+#if defined(HAS_ARGBCOLORMATRIXROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA) && IS_ALIGNED(width, 8)) {
+    ARGBColorMatrixRow = ARGBColorMatrixRow_MSA;
+  }
 #endif
   for (y = 0; y < height; ++y) {
     ARGBColorMatrixRow(src_argb, dst_argb, matrix_argb, width);
@@ -2133,6 +2143,11 @@ int ARGBQuantize(uint8* dst_argb,
   if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
     ARGBQuantizeRow = ARGBQuantizeRow_NEON;
   }
+#endif
+#if defined(HAS_ARGBQUANTIZEROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA) && IS_ALIGNED(width, 8)) {
+    ARGBQuantizeRow = ARGBQuantizeRow_MSA;
+  }
 #endif
   for (y = 0; y < height; ++y) {
     ARGBQuantizeRow(dst, scale, interval_size, interval_offset, width);
@@ -3048,6 +3063,12 @@ int ARGBExtractAlpha(const uint8* src_argb,
                                                 : ARGBExtractAlphaRow_Any_NEON;
   }
 #endif
+#if defined(HAS_ARGBEXTRACTALPHAROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    ARGBExtractAlphaRow = IS_ALIGNED(width, 16) ? ARGBExtractAlphaRow_MSA
+                                                : ARGBExtractAlphaRow_Any_MSA;
+  }
+#endif
 
   for (int y = 0; y < height; ++y) {
     ARGBExtractAlphaRow(src_argb, dst_a, width);

source/row_any.cc

@@ -621,6 +621,9 @@ ANY11(ARGBExtractAlphaRow_Any_AVX2, ARGBExtractAlphaRow_AVX2, 0, 4, 1, 32)
 #ifdef HAS_ARGBEXTRACTALPHAROW_NEON
 ANY11(ARGBExtractAlphaRow_Any_NEON, ARGBExtractAlphaRow_NEON, 0, 4, 1, 15)
 #endif
+#ifdef HAS_ARGBEXTRACTALPHAROW_MSA
+ANY11(ARGBExtractAlphaRow_Any_MSA, ARGBExtractAlphaRow_MSA, 0, 4, 1, 15)
+#endif
 #undef ANY11
 
 // Any 1 to 1 blended.  Destination is read, modify, write.