Enable ARGBToYMatrixRow_RVV/RGBAToYMatrixRow_RVV/RGBToYMatrixRow_RVV

Run on SiFive internal FPGA:

ARGBToJ400_Opt (~6x vs scalar)
RGBAToJ400_Opt (~6x vs scalar)
RGB24ToJ400_Opt (~5.5x vs scalar)

LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10

Change-Id: Ia3ce8cea7962fbd8618cc23e850a7913c9cabf4f
Signed-off-by: Bruce Lai <bruce.lai@sifive.com>
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4521783
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>

5 files changed, 334 insertions, 8 deletions

diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index a8375606..680982e7 100644
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -782,6 +782,11 @@ extern "C" {
 #define HAS_ARGBTOAR64ROW_RVV
 #define HAS_ARGBTORAWROW_RVV
 #define HAS_ARGBTORGB24ROW_RVV
+#define HAS_ARGBTOYROW_RVV
+#define HAS_ARGBTOYJROW_RVV
+#define HAS_ABGRTOYROW_RVV
+#define HAS_ABGRTOYJROW_RVV
+#define HAS_BGRATOYROW_RVV
 #define HAS_I422TOARGBROW_RVV
 #define HAS_I422TORGB24ROW_RVV
 #define HAS_I422TORGBAROW_RVV
@@ -790,13 +795,19 @@ extern "C" {
 #define HAS_MERGEARGBROW_RVV
 #define HAS_MERGERGBROW_RVV
 #define HAS_MERGEXRGBROW_RVV
-#define HAS_SPLITARGBROW_RVV
-#define HAS_SPLITRGBROW_RVV
-#define HAS_SPLITXRGBROW_RVV
 #define HAS_RAWTOARGBROW_RVV
 #define HAS_RAWTORGB24ROW_RVV
 #define HAS_RAWTORGBAROW_RVV
+#define HAS_RAWTOYJROW_RVV
+#define HAS_RAWTOYROW_RVV
 #define HAS_RGB24TOARGBROW_RVV
+#define HAS_RGB24TOYJROW_RVV
+#define HAS_RGB24TOYROW_RVV
+#define HAS_RGBATOYROW_RVV
+#define HAS_RGBATOYJROW_RVV
+#define HAS_SPLITARGBROW_RVV
+#define HAS_SPLITRGBROW_RVV
+#define HAS_SPLITXRGBROW_RVV
 #endif
 
 #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
@@ -1327,6 +1338,10 @@ void ARGBToYRow_NEON(const uint8_t* src_argb, uint8_t* dst_y, int width);
 void ARGBToYJRow_NEON(const uint8_t* src_argb, uint8_t* dst_yj, int width);
 void ABGRToYJRow_NEON(const uint8_t* src_abgr, uint8_t* dst_yj, int width);
 void RGBAToYJRow_NEON(const uint8_t* src_rgba, uint8_t* dst_yj, int width);
+void ARGBToYRow_RVV(const uint8_t* src_argb, uint8_t* dst_y, int width);
+void ARGBToYJRow_RVV(const uint8_t* src_argb, uint8_t* dst_yj, int width);
+void ABGRToYJRow_RVV(const uint8_t* src_rgba, uint8_t* dst_yj, int width);
+void RGBAToYJRow_RVV(const uint8_t* src_rgba, uint8_t* dst_yj, int width);
 void ARGBToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width);
 void ARGBToYJRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width);
 void ARGBToYRow_LSX(const uint8_t* src_argb0, uint8_t* dst_y, int width);
@@ -1548,6 +1563,13 @@ void ARGB1555ToYRow_NEON(const uint8_t* src_argb1555,
 void ARGB4444ToYRow_NEON(const uint8_t* src_argb4444,
                          uint8_t* dst_y,
                          int width);
+void BGRAToYRow_RVV(const uint8_t* src_bgra, uint8_t* dst_y, int width);
+void ABGRToYRow_RVV(const uint8_t* src_abgr, uint8_t* dst_y, int width);
+void RGBAToYRow_RVV(const uint8_t* src_rgba, uint8_t* dst_y, int width);
+void RGB24ToYRow_RVV(const uint8_t* src_rgb24, uint8_t* dst_y, int width);
+void RGB24ToYJRow_RVV(const uint8_t* src_rgb24, uint8_t* dst_yj, int width);
+void RAWToYRow_RVV(const uint8_t* src_raw, uint8_t* dst_y, int width);
+void RAWToYJRow_RVV(const uint8_t* src_raw, uint8_t* dst_yj, int width);
 void BGRAToYRow_MSA(const uint8_t* src_argb, uint8_t* dst_y, int width);
 void ABGRToYRow_MSA(const uint8_t* src_argb, uint8_t* dst_y, int width);
 void RGBAToYRow_MSA(const uint8_t* src_argb, uint8_t* dst_y, int width);
diff --git a/source/convert.cc b/source/convert.cc
index ad679c59..140be1ab 100644
--- a/source/convert.cc
+++ b/source/convert.cc
@@ -1950,6 +1950,11 @@ int BGRAToI420(const uint8_t* src_bgra,
     }
   }
 #endif
+#if defined(HAS_BGRATOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    BGRAToYRow = BGRAToYRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     BGRAToUVRow(src_bgra, src_stride_bgra, dst_u, dst_v, width);
@@ -2070,6 +2075,11 @@ int ABGRToI420(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ABGRToYRow = ABGRToYRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     ABGRToUVRow(src_abgr, src_stride_abgr, dst_u, dst_v, width);
@@ -2174,6 +2184,11 @@ int RGBAToI420(const uint8_t* src_rgba,
     }
   }
 #endif
+#if defined(HAS_RGBATOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RGBAToYRow = RGBAToYRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     RGBAToUVRow(src_rgba, src_stride_rgba, dst_u, dst_v, width);
@@ -2193,7 +2208,7 @@ int RGBAToI420(const uint8_t* src_rgba,
 
 // Enabled if 1 pass is available
 #if (defined(HAS_RGB24TOYROW_NEON) || defined(HAS_RGB24TOYROW_MSA) || \
-     defined(HAS_RGB24TOYROW_LSX))
+     defined(HAS_RGB24TOYROW_LSX) || defined(HAS_RGB24TOYROW_RVV))
 #define HAS_RGB24TOYROW
 #endif
 
@@ -2278,6 +2293,11 @@ int RGB24ToI420(const uint8_t* src_rgb24,
     }
   }
 #endif
+#if defined(HAS_RGB24TOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RGB24ToYRow = RGB24ToYRow_RVV;
+  }
+#endif
 
 // Other platforms do intermediate conversion from RGB24 to ARGB.
 #else  // HAS_RGB24TOYROW
@@ -2367,7 +2387,8 @@ int RGB24ToI420(const uint8_t* src_rgb24,
 #undef HAS_RGB24TOYROW
 
 // Enabled if 1 pass is available
-#if defined(HAS_RGB24TOYJROW_NEON) || defined(HAS_RGB24TOYJROW_MSA)
+#if defined(HAS_RGB24TOYJROW_NEON) || defined(HAS_RGB24TOYJROW_MSA) || \
+    defined(HAS_RGB24TOYJROW_RVV)
 #define HAS_RGB24TOYJROW
 #endif
 
@@ -2448,6 +2469,11 @@ int RGB24ToJ420(const uint8_t* src_rgb24,
     }
   }
 #endif
+#if defined(HAS_RGB24TOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RGB24ToYJRow = RGB24ToYJRow_RVV;
+  }
+#endif
 
 // Other platforms do intermediate conversion from RGB24 to ARGB.
 #else  // HAS_RGB24TOYJROW
@@ -2538,7 +2564,7 @@ int RGB24ToJ420(const uint8_t* src_rgb24,
 
 // Enabled if 1 pass is available
 #if (defined(HAS_RAWTOYROW_NEON) || defined(HAS_RAWTOYROW_MSA) || \
-     defined(HAS_RAWTOYROW_LSX))
+     defined(HAS_RAWTOYROW_LSX) || defined(HAS_RAWTOYROW_RVV))
 #define HAS_RAWTOYROW
 #endif
 
@@ -2622,6 +2648,11 @@ int RAWToI420(const uint8_t* src_raw,
     }
   }
 #endif
+#if defined(HAS_RAWTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RAWToYRow = RAWToYRow_RVV;
+  }
+#endif
 
 // Other platforms do intermediate conversion from RAW to ARGB.
 #else  // HAS_RAWTOYROW
@@ -2711,7 +2742,8 @@ int RAWToI420(const uint8_t* src_raw,
 #undef HAS_RAWTOYROW
 
 // Enabled if 1 pass is available
-#if defined(HAS_RAWTOYJROW_NEON) || defined(HAS_RAWTOYJROW_MSA)
+#if defined(HAS_RAWTOYJROW_NEON) || defined(HAS_RAWTOYJROW_MSA) || \
+    defined(HAS_RAWTOYJROW_RVV)
 #define HAS_RAWTOYJROW
 #endif
 
@@ -2792,6 +2824,11 @@ int RAWToJ420(const uint8_t* src_raw,
     }
   }
 #endif
+#if defined(HAS_RAWTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RAWToYJRow = RAWToYJRow_RVV;
+  }
+#endif
 
 // Other platforms do intermediate conversion from RAW to ARGB.
 #else  // HAS_RAWTOYJROW
@@ -3515,6 +3552,11 @@ int RGB24ToJ400(const uint8_t* src_rgb24,
     }
   }
 #endif
+#if defined(HAS_RGB24TOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RGB24ToYJRow = RGB24ToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     RGB24ToYJRow(src_rgb24, dst_yj, width);
@@ -3599,6 +3641,11 @@ int RAWToJ400(const uint8_t* src_raw,
     }
   }
 #endif
+#if defined(HAS_RAWTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RAWToYJRow = RAWToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     RAWToYJRow(src_raw, dst_yj, width);
diff --git a/source/convert_from_argb.cc b/source/convert_from_argb.cc
index 7e6d8647..897b2c11 100644
--- a/source/convert_from_argb.cc
+++ b/source/convert_from_argb.cc
@@ -132,6 +132,11 @@ int ARGBToI444(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBToYRow = ARGBToYRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToUV444Row(src_argb, dst_u, dst_v, width);
@@ -256,6 +261,11 @@ int ARGBToI422(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBToYRow = ARGBToYRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToUVRow(src_argb, 0, dst_u, dst_v, width);
@@ -373,6 +383,11 @@ int ARGBToNV12(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBToYRow = ARGBToYRow_RVV;
+  }
+#endif
 #if defined(HAS_MERGEUVROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     MergeUVRow_ = MergeUVRow_Any_SSE2;
@@ -551,6 +566,11 @@ int ARGBToNV21(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBToYRow = ARGBToYRow_RVV;
+  }
+#endif
 #if defined(HAS_MERGEUVROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     MergeUVRow_ = MergeUVRow_Any_SSE2;
@@ -726,6 +746,11 @@ int ABGRToNV12(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ABGRToYRow = ABGRToYRow_RVV;
+  }
+#endif
 #if defined(HAS_MERGEUVROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     MergeUVRow_ = MergeUVRow_Any_SSE2;
@@ -902,6 +927,11 @@ int ABGRToNV21(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ABGRToYRow = ABGRToYRow_RVV;
+  }
+#endif
 #if defined(HAS_MERGEUVROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     MergeUVRow_ = MergeUVRow_Any_SSE2;
@@ -1085,6 +1115,11 @@ int ARGBToYUY2(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBToYRow = ARGBToYRow_RVV;
+  }
+#endif
 #if defined(HAS_I422TOYUY2ROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     I422ToYUY2Row = I422ToYUY2Row_Any_SSE2;
@@ -1264,6 +1299,11 @@ int ARGBToUYVY(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBToYRow = ARGBToYRow_RVV;
+  }
+#endif
 #if defined(HAS_I422TOUYVYROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     I422ToUYVYRow = I422ToUYVYRow_Any_SSE2;
@@ -1405,6 +1445,11 @@ int ARGBToI400(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ARGBToYRow = ARGBToYRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToYRow(src_argb, dst_y, width);
@@ -2104,6 +2149,11 @@ int ARGBToJ420(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+     ARGBToYJRow = ARGBToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     ARGBToUVJRow(src_argb, src_stride_argb, dst_uj, dst_vj, width);
@@ -2235,6 +2285,11 @@ int ARGBToJ422(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+     ARGBToYJRow = ARGBToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToUVJRow(src_argb, 0, dst_uj, dst_vj, width);
@@ -2304,6 +2359,11 @@ int ARGBToJ400(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+     ARGBToYJRow = ARGBToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToYJRow(src_argb, dst_yj, width);
@@ -2386,6 +2446,11 @@ int RGBAToJ400(const uint8_t* src_rgba,
     }
   }
 #endif
+#if defined(HAS_RGBATOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RGBAToYJRow = RGBAToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     RGBAToYJRow(src_rgba, dst_yj, width);
@@ -2496,6 +2561,11 @@ int ABGRToJ420(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ABGRToYJRow = ABGRToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     ABGRToUVJRow(src_abgr, src_stride_abgr, dst_uj, dst_vj, width);
@@ -2623,6 +2693,11 @@ int ABGRToJ422(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ABGRToYJRow = ABGRToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ABGRToUVJRow(src_abgr, 0, dst_uj, dst_vj, width);
@@ -2708,6 +2783,11 @@ int ABGRToJ400(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    ABGRToYJRow = ABGRToYJRow_RVV;
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ABGRToYJRow(src_abgr, dst_yj, width);
@@ -2846,7 +2926,7 @@ int ARGBToAB64(const uint8_t* src_argb,
 }
 
 // Enabled if 1 pass is available
-#if defined(HAS_RAWTOYJROW_NEON) || defined(HAS_RAWTOYJROW_MSA)
+#if defined(HAS_RAWTOYJROW_NEON) || defined(HAS_RAWTOYJROW_MSA) || defined(HAS_RAWTOYJROW_RVV)
 #define HAS_RAWTOYJROW
 #endif
 
@@ -2928,6 +3008,11 @@ int RAWToJNV21(const uint8_t* src_raw,
     }
   }
 #endif
+#if defined(HAS_RAWTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+    RAWToYJRow = RAWToYJRow_RVV;
+  }
+#endif
 
 // Other platforms do intermediate conversion from RAW to ARGB.
 #else  // HAS_RAWTOYJROW
diff --git a/source/planar_functions.cc b/source/planar_functions.cc
index b0dc2f43..72ede82b 100644
--- a/source/planar_functions.cc
+++ b/source/planar_functions.cc
@@ -4709,6 +4709,11 @@ static int ARGBSobelize(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_RVV)
+  if (TestCpuFlag(kCpuHasRVV)) {
+     ARGBToYJRow = ARGBToYJRow_RVV;
+  }
+#endif
 
 #if defined(HAS_SOBELYROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
diff --git a/source/row_rvv.cc b/source/row_rvv.cc
index 39ce71e4..99f23165 100644
--- a/source/row_rvv.cc
+++ b/source/row_rvv.cc
@@ -543,6 +543,173 @@ void MergeXRGBRow_RVV(const uint8_t* src_r,
   } while (w > 0);
 }
 
+struct RgbConstants {
+  uint8_t kRGBToY[4];
+  uint16_t kAddY;
+  uint16_t pad;
+};
+
+// RGB to JPeg coefficients
+// B * 0.1140 coefficient = 29
+// G * 0.5870 coefficient = 150
+// R * 0.2990 coefficient = 77
+// Add 0.5 = 0x80
+static const struct RgbConstants kRgb24JPEGConstants = {{29, 150, 77, 0},
+                                                        128,
+                                                        0};
+
+static const struct RgbConstants kRawJPEGConstants = {{77, 150, 29, 0}, 128, 0};
+
+// RGB to BT.601 coefficients
+// B * 0.1016 coefficient = 25
+// G * 0.5078 coefficient = 129
+// R * 0.2578 coefficient = 66
+// Add 16.5 = 0x1080
+
+static const struct RgbConstants kRgb24I601Constants = {{25, 129, 66, 0},
+                                                        0x1080,
+                                                        0};
+
+static const struct RgbConstants kRawI601Constants = {{66, 129, 25, 0},
+                                                      0x1080,
+                                                      0};
+
+// ARGB expects first 3 values to contain RGB and 4th value is ignored.
+void ARGBToYMatrixRow_RVV(const uint8_t* src_argb,
+                          uint8_t* dst_y,
+                          int width,
+                          const struct RgbConstants* rgbconstants) {
+  assert(width != 0);
+  size_t w = (size_t)width;
+  vuint8m2_t v_by, v_gy, v_ry;  // vectors are to store RGBToY constant
+  vuint16m4_t v_addy;           // vector is to store kAddY
+  size_t vl = __riscv_vsetvl_e8m2(w);
+  v_by = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[0], vl);
+  v_gy = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[1], vl);
+  v_ry = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[2], vl);
+  v_addy = __riscv_vmv_v_x_u16m4(rgbconstants->kAddY, vl);
+  do {
+    vuint8m2_t v_b, v_g, v_r, v_a, v_y;
+    vuint16m4_t v_y_u16;
+    size_t vl = __riscv_vsetvl_e8m2(w);
+    __riscv_vlseg4e8_v_u8m2(&v_b, &v_g, &v_r, &v_a, src_argb, vl);
+    v_y_u16 = __riscv_vwmulu_vv_u16m4(v_r, v_ry, vl);
+    v_y_u16 = __riscv_vwmaccu_vv_u16m4(v_y_u16, v_gy, v_g, vl);
+    v_y_u16 = __riscv_vwmaccu_vv_u16m4(v_y_u16, v_by, v_b, vl);
+    v_y_u16 = __riscv_vadd_vv_u16m4(v_y_u16, v_addy, vl);
+    v_y = __riscv_vnsrl_wx_u8m2(v_y_u16, 8, vl);
+    __riscv_vse8_v_u8m2(dst_y, v_y, vl);
+    w -= vl;
+    src_argb += 4 * vl;
+    dst_y += vl;
+  } while (w > 0);
+}
+
+void ARGBToYRow_RVV(const uint8_t* src_argb, uint8_t* dst_y, int width) {
+  ARGBToYMatrixRow_RVV(src_argb, dst_y, width, &kRgb24I601Constants);
+}
+
+void ARGBToYJRow_RVV(const uint8_t* src_argb, uint8_t* dst_yj, int width) {
+  ARGBToYMatrixRow_RVV(src_argb, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void ABGRToYRow_RVV(const uint8_t* src_abgr, uint8_t* dst_y, int width) {
+  ARGBToYMatrixRow_RVV(src_abgr, dst_y, width, &kRawI601Constants);
+}
+
+void ABGRToYJRow_RVV(const uint8_t* src_abgr, uint8_t* dst_yj, int width) {
+  ARGBToYMatrixRow_RVV(src_abgr, dst_yj, width, &kRawJPEGConstants);
+}
+
+// RGBA expects first value to be A and ignored, then 3 values to contain RGB.
+void RGBAToYMatrixRow_RVV(const uint8_t* src_rgba,
+                          uint8_t* dst_y,
+                          int width,
+                          const struct RgbConstants* rgbconstants) {
+  assert(width != 0);
+  size_t w = (size_t)width;
+  vuint8m2_t v_by, v_gy, v_ry;  // vectors are to store RGBToY constant
+  vuint16m4_t v_addy;           // vector is to store kAddY
+  size_t vl = __riscv_vsetvl_e8m2(w);
+  v_by = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[0], vl);
+  v_gy = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[1], vl);
+  v_ry = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[2], vl);
+  v_addy = __riscv_vmv_v_x_u16m4(rgbconstants->kAddY, vl);
+  do {
+    vuint8m2_t v_b, v_g, v_r, v_a, v_y;
+    vuint16m4_t v_y_u16;
+    size_t vl = __riscv_vsetvl_e8m2(w);
+    __riscv_vlseg4e8_v_u8m2(&v_a, &v_b, &v_g, &v_r, src_rgba, vl);
+    v_y_u16 = __riscv_vwmulu_vv_u16m4(v_r, v_ry, vl);
+    v_y_u16 = __riscv_vwmaccu_vv_u16m4(v_y_u16, v_gy, v_g, vl);
+    v_y_u16 = __riscv_vwmaccu_vv_u16m4(v_y_u16, v_by, v_b, vl);
+    v_y_u16 = __riscv_vadd_vv_u16m4(v_y_u16, v_addy, vl);
+    v_y = __riscv_vnsrl_wx_u8m2(v_y_u16, 8, vl);
+    __riscv_vse8_v_u8m2(dst_y, v_y, vl);
+    w -= vl;
+    src_rgba += 4 * vl;
+    dst_y += vl;
+  } while (w > 0);
+}
+
+void RGBAToYRow_RVV(const uint8_t* src_rgba, uint8_t* dst_y, int width) {
+  RGBAToYMatrixRow_RVV(src_rgba, dst_y, width, &kRgb24I601Constants);
+}
+
+void RGBAToYJRow_RVV(const uint8_t* src_rgba, uint8_t* dst_yj, int width) {
+  RGBAToYMatrixRow_RVV(src_rgba, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void BGRAToYRow_RVV(const uint8_t* src_bgra, uint8_t* dst_y, int width) {
+  RGBAToYMatrixRow_RVV(src_bgra, dst_y, width, &kRawI601Constants);
+}
+
+void RGBToYMatrixRow_RVV(const uint8_t* src_rgb,
+                         uint8_t* dst_y,
+                         int width,
+                         const struct RgbConstants* rgbconstants) {
+  assert(width != 0);
+  size_t w = (size_t)width;
+  vuint8m2_t v_by, v_gy, v_ry;  // vectors are to store RGBToY constant
+  vuint16m4_t v_addy;           // vector is to store kAddY
+  size_t vl = __riscv_vsetvl_e8m2(w);
+  v_by = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[0], vl);
+  v_gy = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[1], vl);
+  v_ry = __riscv_vmv_v_x_u8m2(rgbconstants->kRGBToY[2], vl);
+  v_addy = __riscv_vmv_v_x_u16m4(rgbconstants->kAddY, vl);
+  do {
+    vuint8m2_t v_b, v_g, v_r, v_y;
+    vuint16m4_t v_y_u16;
+    size_t vl = __riscv_vsetvl_e8m2(w);
+    __riscv_vlseg3e8_v_u8m2(&v_b, &v_g, &v_r, src_rgb, vl);
+    v_y_u16 = __riscv_vwmulu_vv_u16m4(v_r, v_ry, vl);
+    v_y_u16 = __riscv_vwmaccu_vv_u16m4(v_y_u16, v_gy, v_g, vl);
+    v_y_u16 = __riscv_vwmaccu_vv_u16m4(v_y_u16, v_by, v_b, vl);
+    v_y_u16 = __riscv_vadd_vv_u16m4(v_y_u16, v_addy, vl);
+    v_y = __riscv_vnsrl_wx_u8m2(v_y_u16, 8, vl);
+    __riscv_vse8_v_u8m2(dst_y, v_y, vl);
+    w -= vl;
+    src_rgb += 3 * vl;
+    dst_y += vl;
+  } while (w > 0);
+}
+
+void RGB24ToYJRow_RVV(const uint8_t* src_rgb24, uint8_t* dst_yj, int width) {
+  RGBToYMatrixRow_RVV(src_rgb24, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void RAWToYJRow_RVV(const uint8_t* src_raw, uint8_t* dst_yj, int width) {
+  RGBToYMatrixRow_RVV(src_raw, dst_yj, width, &kRawJPEGConstants);
+}
+
+void RGB24ToYRow_RVV(const uint8_t* src_rgb24, uint8_t* dst_y, int width) {
+  RGBToYMatrixRow_RVV(src_rgb24, dst_y, width, &kRgb24I601Constants);
+}
+
+void RAWToYRow_RVV(const uint8_t* src_raw, uint8_t* dst_y, int width) {
+  RGBToYMatrixRow_RVV(src_raw, dst_y, width, &kRawI601Constants);
+}
+
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv