Enable I{422,444}AlphaToARGBRow_RVV & ARGBAttentuateRow_RVV

Run on SiFive internal FPGA:

I444AlphaToARGB_Opt (~16x vs scalar)
I422AlphaToARGB_Opt (~10x vs scalar)
ARGBAttenuate_Opt (~3x vs scalar)

LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10

Change-Id: I0046eb7af8104bc8e13cee1cb91a19f90940d5b0
Signed-off-by: Bruce Lai <bruce.lai@sifive.com>
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4535657
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>

1 files changed, 76 insertions, 1 deletions

diff --git a/source/convert_argb.cc b/source/convert_argb.cc
index 62884e5d..4d953fae 100644
--- a/source/convert_argb.cc
+++ b/source/convert_argb.cc
@@ -11,7 +11,6 @@
 #include "libyuv/convert_argb.h"
 
 #include <assert.h>
-
 #include "libyuv/convert_from_argb.h"
 #include "libyuv/cpu_id.h"
 #ifdef HAVE_JPEG
@@ -2059,6 +2058,11 @@ int I420AlphaToARGBMatrix(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_I422ALPHATOARGBROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    I422AlphaToARGBRow = I422AlphaToARGBRow_RVV;
+  }
+#endif
 #if defined(HAS_ARGBATTENUATEROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBAttenuateRow = ARGBAttenuateRow_Any_SSSE3;
@@ -2091,6 +2095,11 @@ int I420AlphaToARGBMatrix(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     I422AlphaToARGBRow(src_y, src_u, src_v, src_a, dst_argb, yuvconstants,
@@ -2192,6 +2201,11 @@ int I422AlphaToARGBMatrix(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_I422ALPHATOARGBROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    I422AlphaToARGBRow = I422AlphaToARGBRow_RVV;
+  }
+#endif
 #if defined(HAS_ARGBATTENUATEROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBAttenuateRow = ARGBAttenuateRow_Any_SSSE3;
@@ -2224,6 +2238,11 @@ int I422AlphaToARGBMatrix(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     I422AlphaToARGBRow(src_y, src_u, src_v, src_a, dst_argb, yuvconstants,
@@ -2307,6 +2326,11 @@ int I444AlphaToARGBMatrix(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_I444ALPHATOARGBROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    I444AlphaToARGBRow = I444AlphaToARGBRow_RVV;
+  }
+#endif
 #if defined(HAS_ARGBATTENUATEROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBAttenuateRow = ARGBAttenuateRow_Any_SSSE3;
@@ -2339,6 +2363,11 @@ int I444AlphaToARGBMatrix(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     I444AlphaToARGBRow(src_y, src_u, src_v, src_a, dst_argb, yuvconstants,
@@ -2567,6 +2596,11 @@ int I010AlphaToARGBMatrix(const uint16_t* src_y,
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     I210AlphaToARGBRow(src_y, src_u, src_v, src_a, dst_argb, yuvconstants,
@@ -2668,6 +2702,11 @@ int I210AlphaToARGBMatrix(const uint16_t* src_y,
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     I210AlphaToARGBRow(src_y, src_u, src_v, src_a, dst_argb, yuvconstants,
@@ -2767,6 +2806,11 @@ int I410AlphaToARGBMatrix(const uint16_t* src_y,
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     I410AlphaToARGBRow(src_y, src_u, src_v, src_a, dst_argb, yuvconstants,
@@ -6703,6 +6747,11 @@ static int I420AlphaToARGBMatrixBilinear(
     }
   }
 #endif
+#if defined(HAS_I444ALPHATOARGBROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    I444AlphaToARGBRow = I444AlphaToARGBRow_RVV;
+  }
+#endif
 #if defined(HAS_ARGBATTENUATEROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBAttenuateRow = ARGBAttenuateRow_Any_SSSE3;
@@ -6735,6 +6784,11 @@ static int I420AlphaToARGBMatrixBilinear(
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
 #if defined(HAS_SCALEROWUP2_BILINEAR_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
@@ -6895,6 +6949,11 @@ static int I422AlphaToARGBMatrixLinear(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_I444ALPHATOARGBROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    I444AlphaToARGBRow = I444AlphaToARGBRow_RVV;
+  }
+#endif
 #if defined(HAS_ARGBATTENUATEROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBAttenuateRow = ARGBAttenuateRow_Any_SSSE3;
@@ -6927,6 +6986,12 @@ static int I422AlphaToARGBMatrixLinear(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
+
 #if defined(HAS_SCALEROWUP2_LINEAR_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     ScaleRowUp2_Linear = ScaleRowUp2_Linear_Any_SSE2;
@@ -7060,6 +7125,11 @@ static int I010AlphaToARGBMatrixBilinear(
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
 #if defined(HAS_SCALEROWUP2_BILINEAR_12_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
@@ -7221,6 +7291,11 @@ static int I210AlphaToARGBMatrixLinear(const uint16_t* src_y,
     }
   }
 #endif
+#if defined(HAS_ARGBATTENUATEROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBAttenuateRow = ARGBAttenuateRow_RVV;
+  }
+#endif
 
 #if defined(HAS_SCALEROWUP2_LINEAR_12_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {