Add MSA optimized I444/I400/J400/YUY2/UYVY to ARGB row functions

BUG=libyuv:634

Change-Id: Ida80027c36a938a3bcf6f4480626f8eb9495e1be

Performance Gain (vs C auto-vectorized)
I444ToARGBRow_MSA       - ~1.6x
I444ToARGBRow_Any_MSA   - ~1.6x
I400ToARGBRow_MSA       - ~5.5x
I400ToARGBRow_Any_MSA   - ~5.3x
J400ToARGBRow_MSA       - ~1.0x
J400ToARGBRow_Any_MSA   - ~1.0x
YUY2ToARGBRow_MSA       - ~1.6x
YUY2ToARGBRow_Any_MSA   - ~1.6x
UYVYToARGBRow_MSA       - ~1.6x
UYVYToARGBRow_Any_MSA   - ~1.6x

Performance Gain (vs C non-vectorized)
I444ToARGBRow_MSA       - ~7.3x
I444ToARGBRow_Any_MSA   - ~7.1x
I400ToARGBRow_MSA       - ~5.5x
I400ToARGBRow_Any_MSA   - ~5.2x
J400ToARGBRow_MSA       - ~6.8x
J400ToARGBRow_Any_MSA   - ~5.7x
YUY2ToARGBRow_MSA       - ~7.2x
YUY2ToARGBRow_Any_MSA   - ~7.0x
UYVYToARGBRow_MSA       - ~7.1x
UYVYToARGBRow_Any_MSA   - ~6.9x

Change-Id: Ida80027c36a938a3bcf6f4480626f8eb9495e1be
Reviewed-on: https://chromium-review.googlesource.com/439246Reviewed-by: Frank Barchard <fbarchard@google.com>
Commit-Queue: Frank Barchard <fbarchard@google.com>

include/libyuv/row.h

@@ -442,6 +442,11 @@ extern "C" {
 #define HAS_BGRATOUVROW_MSA
 #define HAS_ABGRTOUVROW_MSA
 #define HAS_RGBATOUVROW_MSA
+#define HAS_I444TOARGBROW_MSA
+#define HAS_I400TOARGBROW_MSA
+#define HAS_J400TOARGBROW_MSA
+#define HAS_YUY2TOARGBROW_MSA
+#define HAS_UYVYTOARGBROW_MSA
 #endif
 
 #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
@@ -754,6 +759,12 @@ void UYVYToARGBRow_NEON(const uint8* src_uyvy,
                         uint8* dst_argb,
                         const struct YuvConstants* yuvconstants,
                         int width);
+void I444ToARGBRow_MSA(const uint8* src_y,
+                       const uint8* src_u,
+                       const uint8* src_v,
+                       uint8* dst_argb,
+                       const struct YuvConstants* yuvconstants,
+                       int width);
 void I444ToARGBRow_DSPR2(const uint8* src_y,
                          const uint8* src_u,
                          const uint8* src_v,
@@ -836,6 +847,14 @@ void NV21ToARGBRow_MSA(const uint8* src_y,
                        uint8* dst_argb,
                        const struct YuvConstants* yuvconstants,
                        int width);
+void YUY2ToARGBRow_MSA(const uint8* src_yuy2,
+                       uint8* dst_argb,
+                       const struct YuvConstants* yuvconstants,
+                       int width);
+void UYVYToARGBRow_MSA(const uint8* src_uyvy,
+                       uint8* dst_argb,
+                       const struct YuvConstants* yuvconstants,
+                       int width);
 
 void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYRow_Any_AVX2(const uint8* src_argb, uint8* dst_y, int width);
@@ -1679,10 +1698,12 @@ void ARGBToARGB4444Row_C(const uint8* src_argb, uint8* dst_rgb, int width);
 void J400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int width);
 void J400ToARGBRow_AVX2(const uint8* src_y, uint8* dst_argb, int width);
 void J400ToARGBRow_NEON(const uint8* src_y, uint8* dst_argb, int width);
+void J400ToARGBRow_MSA(const uint8* src_y, uint8* dst_argb, int width);
 void J400ToARGBRow_C(const uint8* src_y, uint8* dst_argb, int width);
 void J400ToARGBRow_Any_SSE2(const uint8* src_y, uint8* dst_argb, int width);
 void J400ToARGBRow_Any_AVX2(const uint8* src_y, uint8* dst_argb, int width);
 void J400ToARGBRow_Any_NEON(const uint8* src_y, uint8* dst_argb, int width);
+void J400ToARGBRow_Any_MSA(const uint8* src_y, uint8* dst_argb, int width);
 
 void I444ToARGBRow_C(const uint8* src_y,
                      const uint8* src_u,
@@ -2079,9 +2100,11 @@ void I400ToARGBRow_C(const uint8* src_y, uint8* dst_argb, int width);
 void I400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int width);
 void I400ToARGBRow_AVX2(const uint8* src_y, uint8* dst_argb, int width);
 void I400ToARGBRow_NEON(const uint8* src_y, uint8* dst_argb, int width);
+void I400ToARGBRow_MSA(const uint8* src_y, uint8* dst_argb, int width);
 void I400ToARGBRow_Any_SSE2(const uint8* src_y, uint8* dst_argb, int width);
 void I400ToARGBRow_Any_AVX2(const uint8* src_y, uint8* dst_argb, int width);
 void I400ToARGBRow_Any_NEON(const uint8* src_y, uint8* dst_argb, int width);
+void I400ToARGBRow_Any_MSA(const uint8* src_y, uint8* dst_argb, int width);
 
 // ARGB preattenuated alpha blend.
 void ARGBBlendRow_SSSE3(const uint8* src_argb,
@@ -2413,6 +2436,12 @@ void I422ToARGBRow_DSPR2(const uint8* src_y,
                          uint8* dst_argb,
                          const struct YuvConstants* yuvconstants,
                          int width);
+void I444ToARGBRow_Any_MSA(const uint8* src_y,
+                           const uint8* src_u,
+                           const uint8* src_v,
+                           uint8* dst_argb,
+                           const struct YuvConstants* yuvconstants,
+                           int width);
 void I422ToARGBRow_Any_MSA(const uint8* src_y,
                            const uint8* src_u,
                            const uint8* src_v,
@@ -2471,6 +2500,14 @@ void NV21ToARGBRow_Any_MSA(const uint8* src_y,
                            uint8* dst_argb,
                            const struct YuvConstants* yuvconstants,
                            int width);
+void YUY2ToARGBRow_Any_MSA(const uint8* src_yuy2,
+                           uint8* dst_argb,
+                           const struct YuvConstants* yuvconstants,
+                           int width);
+void UYVYToARGBRow_Any_MSA(const uint8* src_uyvy,
+                           uint8* dst_argb,
+                           const struct YuvConstants* yuvconstants,
+                           int width);
 
 void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width);
 void YUY2ToUVRow_AVX2(const uint8* src_yuy2,

source/convert_argb.cc

@@ -493,6 +493,14 @@ static int I444ToARGBMatrix(const uint8* src_y,
     }
   }
 #endif
+#if defined(HAS_I444TOARGBROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    I444ToARGBRow = I444ToARGBRow_Any_MSA;
+    if (IS_ALIGNED(width, 8)) {
+      I444ToARGBRow = I444ToARGBRow_MSA;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     I444ToARGBRow(src_y, src_u, src_v, dst_argb, yuvconstants, width);
@@ -773,6 +781,14 @@ int I400ToARGB(const uint8* src_y,
     }
   }
 #endif
+#if defined(HAS_I400TOARGBROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    I400ToARGBRow = I400ToARGBRow_Any_MSA;
+    if (IS_ALIGNED(width, 16)) {
+      I400ToARGBRow = I400ToARGBRow_MSA;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     I400ToARGBRow(src_y, dst_argb, width);
@@ -831,6 +847,14 @@ int J400ToARGB(const uint8* src_y,
       J400ToARGBRow = J400ToARGBRow_NEON;
     }
   }
+#endif
+#if defined(HAS_J400TOARGBROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    J400ToARGBRow = J400ToARGBRow_Any_MSA;
+    if (IS_ALIGNED(width, 16)) {
+      J400ToARGBRow = J400ToARGBRow_MSA;
+    }
+  }
 #endif
   for (y = 0; y < height; ++y) {
     J400ToARGBRow(src_y, dst_argb, width);
@@ -1540,6 +1564,14 @@ int YUY2ToARGB(const uint8* src_yuy2,
       YUY2ToARGBRow = YUY2ToARGBRow_NEON;
     }
   }
+#endif
+#if defined(HAS_YUY2TOARGBROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    YUY2ToARGBRow = YUY2ToARGBRow_Any_MSA;
+    if (IS_ALIGNED(width, 8)) {
+      YUY2ToARGBRow = YUY2ToARGBRow_MSA;
+    }
+  }
 #endif
   for (y = 0; y < height; ++y) {
     YUY2ToARGBRow(src_yuy2, dst_argb, &kYuvI601Constants, width);
@@ -1599,6 +1631,14 @@ int UYVYToARGB(const uint8* src_uyvy,
       UYVYToARGBRow = UYVYToARGBRow_NEON;
     }
   }
+#endif
+#if defined(HAS_UYVYTOARGBROW_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    UYVYToARGBRow = UYVYToARGBRow_Any_MSA;
+    if (IS_ALIGNED(width, 8)) {
+      UYVYToARGBRow = UYVYToARGBRow_MSA;
+    }
+  }
 #endif
   for (y = 0; y < height; ++y) {
     UYVYToARGBRow(src_uyvy, dst_argb, &kYuvI601Constants, width);

source/row_any.cc

@@ -174,6 +174,7 @@ ANY31C(I422ToARGB4444Row_Any_DSPR2, I422ToARGB4444Row_DSPR2, 1, 0, 2, 7)
 ANY31C(I422ToARGB1555Row_Any_DSPR2, I422ToARGB1555Row_DSPR2, 1, 0, 2, 7)
 #endif
 #ifdef HAS_I422TOARGBROW_MSA
+ANY31C(I444ToARGBRow_Any_MSA, I444ToARGBRow_MSA, 0, 0, 4, 7)
 ANY31C(I422ToARGBRow_Any_MSA, I422ToARGBRow_MSA, 1, 0, 4, 7)
 ANY31C(I422ToRGBARow_Any_MSA, I422ToRGBARow_MSA, 1, 0, 4, 7)
 ANY31C(I422ToRGB24Row_Any_MSA, I422ToRGB24Row_MSA, 1, 0, 3, 15)
@@ -422,6 +423,8 @@ ANY11(ARGBToRAWRow_Any_MSA, ARGBToRAWRow_MSA, 0, 4, 3, 15)
 ANY11(ARGBToRGB565Row_Any_MSA, ARGBToRGB565Row_MSA, 0, 4, 2, 7)
 ANY11(ARGBToARGB1555Row_Any_MSA, ARGBToARGB1555Row_MSA, 0, 4, 2, 7)
 ANY11(ARGBToARGB4444Row_Any_MSA, ARGBToARGB4444Row_MSA, 0, 4, 2, 7)
+ANY11(J400ToARGBRow_Any_MSA, J400ToARGBRow_MSA, 0, 1, 4, 15)
+ANY11(I400ToARGBRow_Any_MSA, I400ToARGBRow_MSA, 0, 1, 4, 15)
 #endif
 #if defined(HAS_RAWTORGB24ROW_NEON)
 ANY11(RAWToRGB24Row_Any_NEON, RAWToRGB24Row_NEON, 0, 3, 3, 7)
@@ -759,6 +762,10 @@ ANY11C(UYVYToARGBRow_Any_AVX2, UYVYToARGBRow_AVX2, 1, 4, 4, 31)
 ANY11C(YUY2ToARGBRow_Any_NEON, YUY2ToARGBRow_NEON, 1, 4, 4, 7)
 ANY11C(UYVYToARGBRow_Any_NEON, UYVYToARGBRow_NEON, 1, 4, 4, 7)
 #endif
+#if defined(HAS_YUY2TOARGBROW_MSA)
+ANY11C(YUY2ToARGBRow_Any_MSA, YUY2ToARGBRow_MSA, 1, 4, 4, 7)
+ANY11C(UYVYToARGBRow_Any_MSA, UYVYToARGBRow_MSA, 1, 4, 4, 7)
+#endif
 #undef ANY11C
 
 // Any 1 to 1 interpolate.  Takes 2 rows of source via stride.