Refine some functions on the Longarch platform.

Add ARGBToYMatrixRow_LSX/LASX, RGBAToYMatrixRow_LSX/LASX and
RGBToYMatrixRow_LSX/LASX functions with RgbConstants argument.

Bug: libyuv:912
Change-Id: I956e639d1f0da4a47a55b79c9d41dcd29e29bdc5
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4167860
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: Mirko Bonadei <mbonadei@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>

6 files changed, 761 insertions, 338 deletions

diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index c9278126..8d998727 100644
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -686,6 +686,11 @@ extern "C" {
 #define HAS_SPLITUVROW_LSX
 #define HAS_UYVYTOARGBROW_LSX
 #define HAS_YUY2TOARGBROW_LSX
+#define HAS_ARGBTOYROW_LSX
+#define HAS_ABGRTOYJROW_LSX
+#define HAS_RGBATOYJROW_LSX
+#define HAS_RGB24TOYJROW_LSX
+#define HAS_RAWTOYJROW_LSX
 #endif
 
 #if !defined(LIBYUV_DISABLE_LASX) && defined(__loongarch_asx)
@@ -713,6 +718,8 @@ extern "C" {
 #define HAS_ARGBTOUVROW_LASX
 #define HAS_ARGBTOYJROW_LASX
 #define HAS_ARGBTOYROW_LASX
+#define HAS_ABGRTOYJROW_LASX
+#define HAS_ABGRTOYROW_LASX
 #define HAS_I422ALPHATOARGBROW_LASX
 #define HAS_I422TOARGB1555ROW_LASX
 #define HAS_I422TOARGB4444ROW_LASX
@@ -742,6 +749,11 @@ extern "C" {
 #define HAS_YUY2TOUV422ROW_LASX
 #define HAS_YUY2TOUVROW_LASX
 #define HAS_YUY2TOYROW_LASX
+#define HAS_RGBATOYROW_LASX
+#define HAS_RGBATOYJROW_LASX
+#define HAS_BGRATOYROW_LASX
+#define HAS_RGB24TOYJROW_LASX
+#define HAS_RAWTOYJROW_LASX
 #endif
 
 #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
@@ -1201,9 +1213,14 @@ 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_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);
 void ARGBToYRow_LASX(const uint8_t* src_argb0, uint8_t* dst_y, int width);
 void ARGBToYJRow_LSX(const uint8_t* src_argb0, uint8_t* dst_y, int width);
+void ABGRToYJRow_LSX(const uint8_t* src_abgr, uint8_t* dst_yj, int width);
+void RGBAToYJRow_LSX(const uint8_t* src_rgba, uint8_t* dst_yj, int width);
 void ARGBToYJRow_LASX(const uint8_t* src_argb0, uint8_t* dst_y, int width);
+void ABGRToYJRow_LASX(const uint8_t* src_abgr, uint8_t* dst_yj, int width);
+void RGBAToYJRow_LASX(const uint8_t* src_rgba, uint8_t* dst_yj, int width);
 void ARGBToUV444Row_NEON(const uint8_t* src_argb,
                          uint8_t* dst_u,
                          uint8_t* dst_v,
@@ -1427,6 +1444,8 @@ void BGRAToYRow_LSX(const uint8_t* src_bgra, uint8_t* dst_y, int width);
 void ABGRToYRow_LSX(const uint8_t* src_abgr, uint8_t* dst_y, int width);
 void RGBAToYRow_LSX(const uint8_t* src_rgba, uint8_t* dst_y, int width);
 void ARGB1555ToYRow_LSX(const uint8_t* src_argb1555, uint8_t* dst_y, int width);
+void RGB24ToYJRow_LSX(const uint8_t* src_rgb24, uint8_t* dst_yj, int width);
+void ABGRToYRow_LASX(const uint8_t* src_abgr, uint8_t* dst_y, int width);
 void ARGB1555ToYRow_LASX(const uint8_t* src_argb1555,
                          uint8_t* dst_y,
                          int width);
@@ -1436,6 +1455,11 @@ void RGB24ToYRow_LSX(const uint8_t* src_rgb24, uint8_t* dst_y, int width);
 void RGB24ToYRow_LASX(const uint8_t* src_rgb24, uint8_t* dst_y, int width);
 void RAWToYRow_LSX(const uint8_t* src_raw, uint8_t* dst_y, int width);
 void RAWToYRow_LASX(const uint8_t* src_raw, uint8_t* dst_y, int width);
+void RGBAToYRow_LASX(const uint8_t* src_rgba, uint8_t* dst_y, int width);
+void BGRAToYRow_LASX(const uint8_t* src_bgra, uint8_t* dst_y, int width);
+void RGB24ToYJRow_LASX(const uint8_t* src_rgb24, uint8_t* dst_yj, int width);
+void RAWToYJRow_LSX(const uint8_t* src_raw, uint8_t* dst_yj, int width);
+void RAWToYJRow_LASX(const uint8_t* src_raw, uint8_t* dst_yj, int width);
 
 void ARGBToYRow_C(const uint8_t* src_rgb, uint8_t* dst_y, int width);
 void ARGBToYJRow_C(const uint8_t* src_rgb, uint8_t* dst_y, int width);
@@ -1499,10 +1523,15 @@ void ARGB1555ToYRow_Any_MSA(const uint8_t* src_ptr,
 void BGRAToYRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void ABGRToYRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RGBAToYRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ARGBToYRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void ARGBToYJRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RGB24ToYRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RGB565ToYRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ABGRToYJRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RAWToYRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RGBAToYJRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RGB24ToYJRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RAWToYJRow_Any_LSX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void ARGB1555ToYRow_Any_LSX(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             int width);
@@ -1511,7 +1540,14 @@ void RGB565ToYRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RGB24ToYRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void ARGBToYJRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void ARGBToYRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ABGRToYRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ABGRToYJRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RAWToYRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RGBAToYRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RGBAToYJRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void BGRAToYRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RGB24ToYJRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RAWToYJRow_Any_LASX(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void ARGB1555ToYRow_Any_LASX(const uint8_t* src_ptr,
                              uint8_t* dst_ptr,
                              int width);
diff --git a/source/convert.cc b/source/convert.cc
index 9a263536..15c70a65 100644
--- a/source/convert.cc
+++ b/source/convert.cc
@@ -1793,6 +1793,14 @@ int ARGBToI420(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX) && defined(HAS_ARGBTOUVROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -1907,13 +1915,21 @@ int BGRAToI420(const uint8_t* src_bgra,
     }
   }
 #endif
-#if defined(HAS_BGRATOYROW_LASX) && defined(HAS_BGRATOUVROW_LASX)
+#if defined(HAS_BGRATOYROW_LSX) && defined(HAS_BGRATOUVROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    BGRAToYRow = BGRAToYRow_Any_LSX;
+    BGRAToUVRow = BGRAToUVRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      BGRAToYRow = BGRAToYRow_LSX;
+      BGRAToUVRow = BGRAToUVRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_BGRATOYROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     BGRAToYRow = BGRAToYRow_Any_LASX;
-    BGRAToUVRow = BGRAToUVRow_Any_LASX;
     if (IS_ALIGNED(width, 32)) {
       BGRAToYRow = BGRAToYRow_LASX;
-      BGRAToUVRow = BGRAToUVRow_LASX;
     }
   }
 #endif
@@ -2029,6 +2045,14 @@ int ABGRToI420(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    ABGRToYRow = ABGRToYRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      ABGRToYRow = ABGRToYRow_LASX;
+    }
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     ABGRToUVRow(src_abgr, src_stride_abgr, dst_u, dst_v, width);
@@ -2125,6 +2149,14 @@ int RGBAToI420(const uint8_t* src_rgba,
     }
   }
 #endif
+#if defined(HAS_RGBATOYROW_LASX)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    RGBAToYRow = RGBAToYRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      RGBAToYRow = RGBAToYRow_LASX;
+    }
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     RGBAToUVRow(src_rgba, src_stride_rgba, dst_u, dst_v, width);
@@ -2383,6 +2415,22 @@ int RGB24ToJ420(const uint8_t* src_rgb24,
     }
   }
 #endif
+#if defined(HAS_RGB24TOYJROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    RGB24ToYJRow = RGB24ToYJRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      RGB24ToYJRow = RGB24ToYJRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_RGB24TOYJROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    RGB24ToYJRow = RGB24ToYJRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      RGB24ToYJRow = RGB24ToYJRow_LASX;
+    }
+  }
+#endif
 
 // Other platforms do intermediate conversion from RGB24 to ARGB.
 #else  // HAS_RGB24TOYJROW
@@ -2711,6 +2759,22 @@ int RAWToJ420(const uint8_t* src_raw,
     }
   }
 #endif
+#if defined(HAS_RAWTOYJROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    RAWToYJRow = RAWToYJRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      RAWToYJRow = RAWToYJRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_RAWTOYJROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    RAWToYJRow = RAWToYJRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      RAWToYJRow = RAWToYJRow_LASX;
+    }
+  }
+#endif
 
 // Other platforms do intermediate conversion from RAW to ARGB.
 #else  // HAS_RAWTOYJROW
@@ -3298,6 +3362,14 @@ int ARGB4444ToI420(const uint8_t* src_argb4444,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX) && defined(HAS_ARGBTOUVROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -3410,6 +3482,22 @@ int RGB24ToJ400(const uint8_t* src_rgb24,
     }
   }
 #endif
+#if defined(HAS_RGB24TOYJROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    RGB24ToYJRow = RGB24ToYJRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      RGB24ToYJRow = RGB24ToYJRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_RGB24TOYJROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    RGB24ToYJRow = RGB24ToYJRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      RGB24ToYJRow = RGB24ToYJRow_LASX;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     RGB24ToYJRow(src_rgb24, dst_yj, width);
@@ -3478,6 +3566,22 @@ int RAWToJ400(const uint8_t* src_raw,
     }
   }
 #endif
+#if defined(HAS_RAWTOYJROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    RAWToYJRow = RAWToYJRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      RAWToYJRow = RAWToYJRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_RAWTOYJROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    RAWToYJRow = RAWToYJRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      RAWToYJRow = RAWToYJRow_LASX;
+    }
+  }
+#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 f7eab0c6..d548aec2 100644
--- a/source/convert_from_argb.cc
+++ b/source/convert_from_argb.cc
@@ -116,6 +116,14 @@ int ARGBToI444(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -230,7 +238,14 @@ int ARGBToI422(const uint8_t* src_argb,
     }
   }
 #endif
-
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX) && defined(HAS_ARGBTOUVROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -340,6 +355,14 @@ int ARGBToNV12(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX) && defined(HAS_ARGBTOUVROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -502,6 +525,14 @@ int ARGBToNV21(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX) && defined(HAS_ARGBTOUVROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -663,6 +694,22 @@ int ABGRToNV12(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ABGRToYRow = ABGRToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ABGRToYRow = ABGRToYRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_ABGRTOYROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    ABGRToYRow = ABGRToYRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      ABGRToYRow = ABGRToYRow_LASX;
+    }
+  }
+#endif
 #if defined(HAS_MERGEUVROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     MergeUVRow_ = MergeUVRow_Any_SSE2;
@@ -815,6 +862,22 @@ int ABGRToNV21(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ABGRToYRow = ABGRToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ABGRToYRow = ABGRToYRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_ABGRTOYROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    ABGRToYRow = ABGRToYRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      ABGRToYRow = ABGRToYRow_LASX;
+    }
+  }
+#endif
 #if defined(HAS_MERGEUVROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     MergeUVRow_ = MergeUVRow_Any_SSE2;
@@ -972,6 +1035,14 @@ int ARGBToYUY2(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX) && defined(HAS_ARGBTOUVROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -1135,6 +1206,14 @@ int ARGBToUYVY(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX) && defined(HAS_ARGBTOUVROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -1262,6 +1341,14 @@ int ARGBToI400(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYRow = ARGBToYRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYRow = ARGBToYRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYRow = ARGBToYRow_Any_LASX;
@@ -1939,6 +2026,16 @@ int ARGBToJ420(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_LSX) && defined(HAS_ARGBTOUVJROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ARGBToYJRow = ARGBToYJRow_Any_LSX;
+    ARGBToUVJRow = ARGBToUVJRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToYJRow = ARGBToYJRow_LSX;
+      ARGBToUVJRow = ARGBToUVJRow_LSX;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYJROW_LASX) && defined(HAS_ARGBTOUVJROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ARGBToYJRow = ARGBToYJRow_Any_LASX;
@@ -2215,6 +2312,22 @@ int RGBAToJ400(const uint8_t* src_rgba,
     }
   }
 #endif
+#if defined(HAS_RGBATOYJROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    RGBAToYJRow = RGBAToYJRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      RGBAToYJRow = RGBAToYJRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_RGBATOYJROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    RGBAToYJRow = RGBAToYJRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      RGBAToYJRow = RGBAToYJRow_LASX;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     RGBAToYJRow(src_rgba, dst_yj, width);
@@ -2309,13 +2422,19 @@ int ABGRToJ420(const uint8_t* src_abgr,
     }
   }
 #endif
-#if defined(HAS_ABGRTOYJROW_LSX) && defined(HAS_ABGRTOUVJROW_LSX)
+#if defined(HAS_ABGRTOYJROW_LSX)
   if (TestCpuFlag(kCpuHasLSX)) {
     ABGRToYJRow = ABGRToYJRow_Any_LSX;
-    ABGRToUVJRow = ABGRToUVJRow_Any_LSX;
     if (IS_ALIGNED(width, 16)) {
       ABGRToYJRow = ABGRToYJRow_LSX;
-      ABGRToUVJRow = ABGRToUVJRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_ABGRTOYJROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    ABGRToYJRow = ABGRToYJRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      ABGRToYJRow = ABGRToYJRow_LASX;
     }
   }
 #endif
@@ -2430,23 +2549,19 @@ int ABGRToJ422(const uint8_t* src_abgr,
     }
   }
 #endif
-#if defined(HAS_ABGRTOYJROW_LSX) && defined(HAS_ABGRTOUVJROW_LSX)
+#if defined(HAS_ABGRTOYJROW_LSX)
   if (TestCpuFlag(kCpuHasLSX)) {
     ABGRToYJRow = ABGRToYJRow_Any_LSX;
-    ABGRToUVJRow = ABGRToUVJRow_Any_LSX;
     if (IS_ALIGNED(width, 16)) {
       ABGRToYJRow = ABGRToYJRow_LSX;
-      ABGRToUVJRow = ABGRToUVJRow_LSX;
     }
   }
 #endif
-#if defined(HAS_ABGRTOYJROW_LASX) && defined(HAS_ABGRTOUVJROW_LASX)
+#if defined(HAS_ABGRTOYJROW_LASX)
   if (TestCpuFlag(kCpuHasLASX)) {
     ABGRToYJRow = ABGRToYJRow_Any_LASX;
-    ABGRToUVJRow = ABGRToUVJRow_Any_LASX;
     if (IS_ALIGNED(width, 32)) {
       ABGRToYJRow = ABGRToYJRow_LASX;
-      ABGRToUVJRow = ABGRToUVJRow_LASX;
     }
   }
 #endif
@@ -2519,6 +2634,22 @@ int ABGRToJ400(const uint8_t* src_abgr,
     }
   }
 #endif
+#if defined(HAS_ABGRTOYJROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    ABGRToYJRow = ABGRToYJRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      ABGRToYJRow = ABGRToYJRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_ABGRTOYJROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    ABGRToYJRow = ABGRToYJRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      ABGRToYJRow = ABGRToYJRow_LASX;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ABGRToYJRow(src_abgr, dst_yj, width);
@@ -2713,6 +2844,22 @@ int RAWToJNV21(const uint8_t* src_raw,
     }
   }
 #endif
+#if defined(HAS_RAWTOYJROW_LSX)
+  if (TestCpuFlag(kCpuHasLSX)) {
+    RAWToYJRow = RAWToYJRow_Any_LSX;
+    if (IS_ALIGNED(width, 16)) {
+      RAWToYJRow = RAWToYJRow_LSX;
+    }
+  }
+#endif
+#if defined(HAS_RAWTOYJROW_LASX)
+  if (TestCpuFlag(kCpuHasLASX)) {
+    RAWToYJRow = RAWToYJRow_Any_LASX;
+    if (IS_ALIGNED(width, 32)) {
+      RAWToYJRow = RAWToYJRow_LASX;
+    }
+  }
+#endif
 
 // Other platforms do intermediate conversion from RAW to ARGB.
 #else  // HAS_RAWTOYJROW
diff --git a/source/row_any.cc b/source/row_any.cc
index 3c7dc893..012f0fb2 100644
--- a/source/row_any.cc
+++ b/source/row_any.cc
@@ -1036,6 +1036,9 @@ ANY11(ARGBToYRow_Any_NEON, ARGBToYRow_NEON, 0, 4, 1, 15)
 #ifdef HAS_ARGBTOYROW_MSA
 ANY11(ARGBToYRow_Any_MSA, ARGBToYRow_MSA, 0, 4, 1, 15)
 #endif
+#ifdef HAS_ARGBTOYROW_LSX
+ANY11(ARGBToYRow_Any_LSX, ARGBToYRow_LSX, 0, 4, 1, 15)
+#endif
 #ifdef HAS_ARGBTOYROW_LASX
 ANY11(ARGBToYRow_Any_LASX, ARGBToYRow_LASX, 0, 4, 1, 31)
 #endif
@@ -1054,9 +1057,21 @@ ANY11(ARGBToYJRow_Any_MSA, ARGBToYJRow_MSA, 0, 4, 1, 15)
 #ifdef HAS_ARGBTOYJROW_LSX
 ANY11(ARGBToYJRow_Any_LSX, ARGBToYJRow_LSX, 0, 4, 1, 15)
 #endif
+#ifdef HAS_RGBATOYJROW_LSX
+ANY11(RGBAToYJRow_Any_LSX, RGBAToYJRow_LSX, 0, 4, 1, 15)
+#endif
+#ifdef HAS_ABGRTOYJROW_LSX
+ANY11(ABGRToYJRow_Any_LSX, ABGRToYJRow_LSX, 0, 4, 1, 15)
+#endif
+#ifdef HAS_RGBATOYJROW_LASX
+ANY11(RGBAToYJRow_Any_LASX, RGBAToYJRow_LASX, 0, 4, 1, 31)
+#endif
 #ifdef HAS_ARGBTOYJROW_LASX
 ANY11(ARGBToYJRow_Any_LASX, ARGBToYJRow_LASX, 0, 4, 1, 31)
 #endif
+#ifdef HAS_ABGRTOYJROW_LASX
+ANY11(ABGRToYJRow_Any_LASX, ABGRToYJRow_LASX, 0, 4, 1, 31)
+#endif
 #ifdef HAS_BGRATOYROW_NEON
 ANY11(BGRAToYRow_Any_NEON, BGRAToYRow_NEON, 0, 4, 1, 15)
 #endif
@@ -1066,6 +1081,9 @@ ANY11(BGRAToYRow_Any_MSA, BGRAToYRow_MSA, 0, 4, 1, 15)
 #ifdef HAS_BGRATOYROW_LSX
 ANY11(BGRAToYRow_Any_LSX, BGRAToYRow_LSX, 0, 4, 1, 15)
 #endif
+#ifdef HAS_BGRATOYROW_LASX
+ANY11(BGRAToYRow_Any_LASX, BGRAToYRow_LASX, 0, 4, 1, 31)
+#endif
 #ifdef HAS_ABGRTOYROW_NEON
 ANY11(ABGRToYRow_Any_NEON, ABGRToYRow_NEON, 0, 4, 1, 15)
 #endif
@@ -1075,6 +1093,9 @@ ANY11(ABGRToYRow_Any_MSA, ABGRToYRow_MSA, 0, 4, 1, 7)
 #ifdef HAS_ABGRTOYROW_LSX
 ANY11(ABGRToYRow_Any_LSX, ABGRToYRow_LSX, 0, 4, 1, 15)
 #endif
+#ifdef HAS_ABGRTOYROW_LASX
+ANY11(ABGRToYRow_Any_LASX, ABGRToYRow_LASX, 0, 4, 1, 31)
+#endif
 #ifdef HAS_RGBATOYROW_NEON
 ANY11(RGBAToYRow_Any_NEON, RGBAToYRow_NEON, 0, 4, 1, 15)
 #endif
@@ -1084,6 +1105,9 @@ ANY11(RGBAToYRow_Any_MSA, RGBAToYRow_MSA, 0, 4, 1, 15)
 #ifdef HAS_RGBATOYROW_LSX
 ANY11(RGBAToYRow_Any_LSX, RGBAToYRow_LSX, 0, 4, 1, 15)
 #endif
+#ifdef HAS_RGBATOYROW_LASX
+ANY11(RGBAToYRow_Any_LASX, RGBAToYRow_LASX, 0, 4, 1, 31)
+#endif
 #ifdef HAS_RGB24TOYROW_NEON
 ANY11(RGB24ToYRow_Any_NEON, RGB24ToYRow_NEON, 0, 3, 1, 15)
 #endif
@@ -1102,6 +1126,12 @@ ANY11(RGB24ToYRow_Any_MSA, RGB24ToYRow_MSA, 0, 3, 1, 15)
 #ifdef HAS_RGB24TOYROW_LSX
 ANY11(RGB24ToYRow_Any_LSX, RGB24ToYRow_LSX, 0, 3, 1, 15)
 #endif
+#ifdef HAS_RGB24TOYJROW_LSX
+ANY11(RGB24ToYJRow_Any_LSX, RGB24ToYJRow_LSX, 0, 3, 1, 15)
+#endif
+#ifdef HAS_RGB24TOYJROW_LASX
+ANY11(RGB24ToYJRow_Any_LASX, RGB24ToYJRow_LASX, 0, 3, 1, 31)
+#endif
 #ifdef HAS_RGB24TOYROW_LASX
 ANY11(RGB24ToYRow_Any_LASX, RGB24ToYRow_LASX, 0, 3, 1, 31)
 #endif
@@ -1126,6 +1156,12 @@ ANY11(RAWToYRow_Any_LSX, RAWToYRow_LSX, 0, 3, 1, 15)
 #ifdef HAS_RAWTOYROW_LASX
 ANY11(RAWToYRow_Any_LASX, RAWToYRow_LASX, 0, 3, 1, 31)
 #endif
+#ifdef HAS_RAWTOYJROW_LSX
+ANY11(RAWToYJRow_Any_LSX, RAWToYJRow_LSX, 0, 3, 1, 15)
+#endif
+#ifdef HAS_RAWTOYJROW_LASX
+ANY11(RAWToYJRow_Any_LASX, RAWToYJRow_LASX, 0, 3, 1, 31)
+#endif
 #ifdef HAS_RGB565TOYROW_NEON
 ANY11(RGB565ToYRow_Any_NEON, RGB565ToYRow_NEON, 0, 2, 1, 7)
 #endif
diff --git a/source/row_lasx.cc b/source/row_lasx.cc
index 7dd18f40..f824906d 100644
--- a/source/row_lasx.cc
+++ b/source/row_lasx.cc
@@ -775,40 +775,6 @@ void UYVYToUV422Row_LASX(const uint8_t* src_uyvy,
   }
 }
 
-void ARGBToYRow_LASX(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 32;
-  __m256i src0, src1, src2, src3, vec0, vec1, vec2, vec3;
-  __m256i tmp0, tmp1, dst0;
-  __m256i const_19 = __lasx_xvldi(0x19);
-  __m256i const_42 = __lasx_xvldi(0x42);
-  __m256i const_81 = __lasx_xvldi(0x81);
-  __m256i const_1080 = {0x1080108010801080, 0x1080108010801080,
-                        0x1080108010801080, 0x1080108010801080};
-  __m256i control = {0x0000000400000000, 0x0000000500000001, 0x0000000600000002,
-                     0x0000000700000003};
-
-  for (x = 0; x < len; x++) {
-    DUP4_ARG2(__lasx_xvld, src_argb0, 0, src_argb0, 32, src_argb0, 64,
-              src_argb0, 96, src0, src1, src2, src3);
-    vec0 = __lasx_xvpickev_b(src1, src0);
-    vec1 = __lasx_xvpickev_b(src3, src2);
-    vec2 = __lasx_xvpickod_b(src1, src0);
-    vec3 = __lasx_xvpickod_b(src3, src2);
-    tmp0 = __lasx_xvmaddwev_h_bu(const_1080, vec0, const_19);
-    tmp1 = __lasx_xvmaddwev_h_bu(const_1080, vec1, const_19);
-    tmp0 = __lasx_xvmaddwev_h_bu(tmp0, vec2, const_81);
-    tmp1 = __lasx_xvmaddwev_h_bu(tmp1, vec3, const_81);
-    tmp0 = __lasx_xvmaddwod_h_bu(tmp0, vec0, const_42);
-    tmp1 = __lasx_xvmaddwod_h_bu(tmp1, vec1, const_42);
-    dst0 = __lasx_xvssrani_b_h(tmp1, tmp0, 8);
-    dst0 = __lasx_xvperm_w(dst0, control);
-    __lasx_xvst(dst0, dst_y, 0);
-    src_argb0 += 128;
-    dst_y += 32;
-  }
-}
-
 void ARGBToUVRow_LASX(const uint8_t* src_argb0,
                       int src_stride_argb,
                       uint8_t* dst_u,
@@ -1811,48 +1777,6 @@ void RGB565ToUVRow_LASX(const uint8_t* src_rgb565,
   }
 }
 
-void RGB24ToYRow_LASX(const uint8_t* src_rgb24, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 32;
-  __m256i src0, src1, src2;
-  __m256i tmp0, tmp1, tmp2, tmp3;
-  __m256i reg0, reg1, reg2, dst0;
-  __m256i const_129 = __lasx_xvldi(129);
-  __m256i const_br = {0x4219421942194219, 0x4219421942194219,
-                      0x4219421942194219, 0x4219421942194219};
-  __m256i const_1080 = {0x1080108010801080, 0x1080108010801080,
-                        0x1080108010801080, 0x1080108010801080};
-  __m256i shuff0 = {0x0B09080605030200, 0x17151412110F0E0C, 0x0B09080605030200,
-                    0x17151412110F0E0C};
-  __m256i shuff1 = {0x0301001E1D1B1A18, 0x0F0D0C0A09070604, 0x0301001E1D1B1A18,
-                    0x0F0D0C0A09070604};
-  __m256i shuff2 = {0x000A000700040001, 0x001600130010000D, 0x000A000700040001,
-                    0x001600130010000D};
-  __m256i shuff3 = {0x0002001F001C0019, 0x000E000B00080005, 0x0002001F001C0019,
-                    0x000E000B00080005};
-
-  for (x = 0; x < len; x++) {
-    reg0 = __lasx_xvld(src_rgb24, 0);
-    reg1 = __lasx_xvld(src_rgb24, 32);
-    reg2 = __lasx_xvld(src_rgb24, 64);
-    src0 = __lasx_xvpermi_q(reg1, reg0, 0x30);
-    src1 = __lasx_xvpermi_q(reg2, reg0, 0x21);
-    src2 = __lasx_xvpermi_q(reg2, reg1, 0x30);
-    tmp0 = __lasx_xvshuf_b(src1, src0, shuff0);
-    tmp1 = __lasx_xvshuf_b(src1, src2, shuff1);
-    tmp2 = __lasx_xvshuf_b(src1, src0, shuff2);
-    tmp3 = __lasx_xvshuf_b(src1, src2, shuff3);
-    reg0 = __lasx_xvmaddwev_h_bu(const_1080, tmp2, const_129);
-    reg1 = __lasx_xvmaddwev_h_bu(const_1080, tmp3, const_129);
-    reg0 = __lasx_xvdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lasx_xvdp2add_h_bu(reg1, const_br, tmp1);
-    dst0 = __lasx_xvpickod_b(reg1, reg0);
-    __lasx_xvst(dst0, dst_y, 0);
-    dst_y += 32;
-    src_rgb24 += 96;
-  }
-}
-
 void RGB24ToUVRow_LASX(const uint8_t* src_rgb24,
                        int src_stride_rgb24,
                        uint8_t* dst_u,
@@ -1916,48 +1840,6 @@ void RGB24ToUVRow_LASX(const uint8_t* src_rgb24,
   }
 }
 
-void RAWToYRow_LASX(const uint8_t* src_raw, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 32;
-  __m256i src0, src1, src2;
-  __m256i tmp0, tmp1, tmp2, tmp3;
-  __m256i reg0, reg1, reg2, dst0;
-  __m256i const_129 = __lasx_xvldi(129);
-  __m256i const_br = {0x1942194219421942, 0x1942194219421942,
-                      0x1942194219421942, 0x1942194219421942};
-  __m256i const_1080 = {0x1080108010801080, 0x1080108010801080,
-                        0x1080108010801080, 0x1080108010801080};
-  __m256i shuff0 = {0x0B09080605030200, 0x17151412110F0E0C, 0x0B09080605030200,
-                    0x17151412110F0E0C};
-  __m256i shuff1 = {0x0301001E1D1B1A18, 0x0F0D0C0A09070604, 0x0301001E1D1B1A18,
-                    0x0F0D0C0A09070604};
-  __m256i shuff2 = {0x000A000700040001, 0x001600130010000D, 0x000A000700040001,
-                    0x001600130010000D};
-  __m256i shuff3 = {0x0002001F001C0019, 0x000E000B00080005, 0x0002001F001C0019,
-                    0x000E000B00080005};
-
-  for (x = 0; x < len; x++) {
-    reg0 = __lasx_xvld(src_raw, 0);
-    reg1 = __lasx_xvld(src_raw, 32);
-    reg2 = __lasx_xvld(src_raw, 64);
-    src0 = __lasx_xvpermi_q(reg1, reg0, 0x30);
-    src1 = __lasx_xvpermi_q(reg2, reg0, 0x21);
-    src2 = __lasx_xvpermi_q(reg2, reg1, 0x30);
-    tmp0 = __lasx_xvshuf_b(src1, src0, shuff0);
-    tmp1 = __lasx_xvshuf_b(src1, src2, shuff1);
-    tmp2 = __lasx_xvshuf_b(src1, src0, shuff2);
-    tmp3 = __lasx_xvshuf_b(src1, src2, shuff3);
-    reg0 = __lasx_xvmaddwev_h_bu(const_1080, tmp2, const_129);
-    reg1 = __lasx_xvmaddwev_h_bu(const_1080, tmp3, const_129);
-    reg0 = __lasx_xvdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lasx_xvdp2add_h_bu(reg1, const_br, tmp1);
-    dst0 = __lasx_xvpickod_b(reg1, reg0);
-    __lasx_xvst(dst0, dst_y, 0);
-    dst_y += 32;
-    src_raw += 96;
-  }
-}
-
 void RAWToUVRow_LASX(const uint8_t* src_raw,
                      int src_stride_raw,
                      uint8_t* dst_u,
@@ -2118,36 +2000,226 @@ void NV21ToARGBRow_LASX(const uint8_t* src_y,
   }
 }
 
-void ARGBToYJRow_LASX(const uint8_t* src_argb, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 32;
-  __m256i src0, src1, src2, src3, dst0;
-  __m256i tmp0, tmp1, tmp2, tmp3;
-  __m256i reg0, reg1;
-  __m256i const_128 = __lasx_xvldi(0x480);
-  __m256i const_150 = __lasx_xvldi(0x96);
-  __m256i const_br = {0x4D1D4D1D4D1D4D1D, 0x4D1D4D1D4D1D4D1D,
-                      0x4D1D4D1D4D1D4D1D, 0x4D1D4D1D4D1D4D1D};
-  __m256i shuff = {0x0000000400000000, 0x0000000500000001, 0x0000000600000002,
-                   0x0000000700000003};
+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.
+static void ARGBToYMatrixRow_LASX(const uint8_t* src_argb,
+                                  uint8_t* dst_y,
+                                  int width,
+                                  const struct RgbConstants* rgbconstants) {
+    int32_t shuff[8] = {0, 4, 1, 5, 2, 6, 3, 7};
+    asm volatile(
+      "xvldrepl.b      $xr0,  %3,    0             \n\t"  // load rgbconstants
+      "xvldrepl.b      $xr1,  %3,    1             \n\t"  // load rgbconstants
+      "xvldrepl.b      $xr2,  %3,    2             \n\t"  // load rgbconstants
+      "xvldrepl.h      $xr3,  %3,    4             \n\t"  // load rgbconstants
+      "xvld            $xr20, %4,    0             \n\t"  // load shuff
+      "1:                                          \n\t"
+      "xvld            $xr4,  %0,    0             \n\t"
+      "xvld            $xr5,  %0,    32            \n\t"
+      "xvld            $xr6,  %0,    64            \n\t"
+      "xvld            $xr7,  %0,    96            \n\t"  // load 32 pixels of ARGB
+      "xvor.v          $xr12, $xr3,  $xr3          \n\t"
+      "xvor.v          $xr13, $xr3,  $xr3          \n\t"
+      "addi.d          %2,    %2,    -32           \n\t"  // 32 processed per loop.
+      "xvpickev.b      $xr8,  $xr5,  $xr4          \n\t"  //BR
+      "xvpickev.b      $xr10, $xr7,  $xr6          \n\t"
+      "xvpickod.b      $xr9,  $xr5,  $xr4          \n\t"  //GA
+      "xvpickod.b      $xr11, $xr7,  $xr6          \n\t"
+      "xvmaddwev.h.bu  $xr12, $xr8,  $xr0          \n\t"  //B
+      "xvmaddwev.h.bu  $xr13, $xr10, $xr0          \n\t"
+      "xvmaddwev.h.bu  $xr12, $xr9,  $xr1          \n\t"  //G
+      "xvmaddwev.h.bu  $xr13, $xr11, $xr1          \n\t"
+      "xvmaddwod.h.bu  $xr12, $xr8,  $xr2          \n\t"  //R
+      "xvmaddwod.h.bu  $xr13, $xr10, $xr2          \n\t"
+      "addi.d          %0,    %0,    128           \n\t"
+      "xvpickod.b      $xr10, $xr13, $xr12         \n\t"
+      "xvperm.w        $xr11, $xr10, $xr20         \n\t"
+      "xvst            $xr11, %1,    0             \n\t"
+      "addi.d          %1,    %1,    32            \n\t"
+      "bnez            %2,    1b                   \n\t"
+      : "+&r"(src_argb),    // %0
+        "+&r"(dst_y),       // %1
+        "+&r"(width)        // %2
+      : "r"(rgbconstants),
+        "r"(shuff)
+      : "memory"
+    );
+}
 
-  for (x = 0; x < len; x++) {
-    DUP4_ARG2(__lasx_xvld, src_argb, 0, src_argb, 32, src_argb, 64, src_argb,
-              96, src0, src1, src2, src3);
-    tmp0 = __lasx_xvpickev_b(src1, src0);
-    tmp1 = __lasx_xvpickod_b(src1, src0);
-    tmp2 = __lasx_xvpickev_b(src3, src2);
-    tmp3 = __lasx_xvpickod_b(src3, src2);
-    reg0 = __lasx_xvmaddwev_h_bu(const_128, tmp1, const_150);
-    reg1 = __lasx_xvmaddwev_h_bu(const_128, tmp3, const_150);
-    reg0 = __lasx_xvdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lasx_xvdp2add_h_bu(reg1, const_br, tmp2);
-    dst0 = __lasx_xvpickod_b(reg1, reg0);
-    dst0 = __lasx_xvperm_w(dst0, shuff);
-    __lasx_xvst(dst0, dst_y, 0);
-    dst_y += 32;
-    src_argb += 128;
-  }
+void ARGBToYRow_LASX(const uint8_t* src_argb, uint8_t* dst_y, int width) {
+  ARGBToYMatrixRow_LASX(src_argb, dst_y, width, &kRgb24I601Constants);
+}
+
+void ARGBToYJRow_LASX(const uint8_t* src_argb, uint8_t* dst_yj, int width) {
+  ARGBToYMatrixRow_LASX(src_argb, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void ABGRToYRow_LASX(const uint8_t* src_abgr, uint8_t* dst_y, int width) {
+  ARGBToYMatrixRow_LASX(src_abgr, dst_y, width, &kRawI601Constants);
+}
+
+void ABGRToYJRow_LASX(const uint8_t* src_abgr, uint8_t* dst_yj, int width) {
+  ARGBToYMatrixRow_LASX(src_abgr, dst_yj, width, &kRawJPEGConstants);
+}
+
+// RGBA expects first value to be A and ignored, then 3 values to contain RGB.
+// Same code as ARGB, except the LD4
+static void RGBAToYMatrixRow_LASX(const uint8_t* src_rgba,
+                                  uint8_t* dst_y,
+                                  int width,
+                                  const struct RgbConstants* rgbconstants) {
+    int32_t shuff[8] = {0, 4, 1, 5, 2, 6, 3, 7};
+    asm volatile(
+      "xvldrepl.b      $xr0,  %3,    0             \n\t"  // load rgbconstants
+      "xvldrepl.b      $xr1,  %3,    1             \n\t"  // load rgbconstants
+      "xvldrepl.b      $xr2,  %3,    2             \n\t"  // load rgbconstants
+      "xvldrepl.h      $xr3,  %3,    4             \n\t"  // load rgbconstants
+      "xvld            $xr20, %4,    0             \n\t"  // load shuff
+      "1:                                          \n\t"
+      "xvld            $xr4,  %0,    0             \n\t"
+      "xvld            $xr5,  %0,    32            \n\t"
+      "xvld            $xr6,  %0,    64            \n\t"
+      "xvld            $xr7,  %0,    96            \n\t"  // load 32 pixels of RGBA
+      "xvor.v          $xr12, $xr3,  $xr3          \n\t"
+      "xvor.v          $xr13, $xr3,  $xr3          \n\t"
+      "addi.d          %2,    %2,    -32           \n\t"  // 32 processed per loop.
+      "xvpickev.b      $xr8,  $xr5,  $xr4          \n\t"  //AG
+      "xvpickev.b      $xr10, $xr7,  $xr6          \n\t"
+      "xvpickod.b      $xr9,  $xr5,  $xr4          \n\t"  //BR
+      "xvpickod.b      $xr11, $xr7,  $xr6          \n\t"
+      "xvmaddwev.h.bu  $xr12, $xr9,  $xr0          \n\t"  //B
+      "xvmaddwev.h.bu  $xr13, $xr11, $xr0          \n\t"
+      "xvmaddwod.h.bu  $xr12, $xr8,  $xr1          \n\t"  //G
+      "xvmaddwod.h.bu  $xr13, $xr10, $xr1          \n\t"
+      "xvmaddwod.h.bu  $xr12, $xr9,  $xr2          \n\t"  //R
+      "xvmaddwod.h.bu  $xr13, $xr11, $xr2          \n\t"
+      "addi.d          %0,    %0,    128           \n\t"
+      "xvpickod.b      $xr10, $xr13, $xr12         \n\t"
+      "xvperm.w        $xr11, $xr10, $xr20         \n\t"
+      "xvst            $xr11, %1,    0             \n\t"
+      "addi.d          %1,    %1,    32            \n\t"
+      "bnez            %2,    1b                   \n\t"
+      : "+&r"(src_rgba),    // %0
+        "+&r"(dst_y),       // %1
+        "+&r"(width)        // %2
+      : "r"(rgbconstants),
+        "r"(shuff)
+      : "memory"
+    );
+}
+
+void RGBAToYRow_LASX(const uint8_t* src_rgba, uint8_t* dst_y, int width) {
+  RGBAToYMatrixRow_LASX(src_rgba, dst_y, width, &kRgb24I601Constants);
+}
+
+void RGBAToYJRow_LASX(const uint8_t* src_rgba, uint8_t* dst_yj, int width) {
+  RGBAToYMatrixRow_LASX(src_rgba, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void BGRAToYRow_LASX(const uint8_t* src_bgra, uint8_t* dst_y, int width) {
+  RGBAToYMatrixRow_LASX(src_bgra, dst_y, width, &kRawI601Constants);
+}
+
+static void RGBToYMatrixRow_LASX(const uint8_t* src_rgba,
+                                  uint8_t* dst_y,
+                                  int width,
+                                  const struct RgbConstants* rgbconstants) {
+    int8_t shuff[128] = {0, 2, 3, 5, 6, 8, 9, 11, 12, 14, 15, 17, 18, 20, 21, 23,
+                         0, 2, 3, 5, 6, 8, 9, 11, 12, 14, 15, 17, 18, 20, 21, 23,
+                         24, 26, 27, 29, 30, 0, 1, 3, 4, 6, 7, 9, 10, 12, 13, 15,
+                         24, 26, 27, 29, 30, 0, 1, 3, 4, 6, 7, 9, 10, 12, 13, 15,
+                         1, 0, 4, 0, 7, 0, 10, 0, 13, 0, 16, 0, 19, 0, 22, 0,
+                         1, 0, 4, 0, 7, 0, 10, 0, 13, 0, 16, 0, 19, 0, 22, 0,
+                         25, 0, 28, 0, 31, 0, 2, 0, 5, 0, 8, 0, 11, 0, 14, 0,
+                         25, 0, 28, 0, 31, 0, 2, 0, 5, 0, 8, 0, 11, 0, 14, 0};
+    asm volatile(
+      "xvldrepl.b      $xr0,  %3,    0             \n\t"  // load rgbconstants
+      "xvldrepl.b      $xr1,  %3,    1             \n\t"  // load rgbconstants
+      "xvldrepl.b      $xr2,  %3,    2             \n\t"  // load rgbconstants
+      "xvldrepl.h      $xr3,  %3,    4             \n\t"  // load rgbconstants
+      "xvld            $xr4,  %4,    0             \n\t"  // load shuff
+      "xvld            $xr5,  %4,    32            \n\t"
+      "xvld            $xr6,  %4,    64            \n\t"
+      "xvld            $xr7,  %4,    96            \n\t"
+      "1:                                          \n\t"
+      "xvld            $xr8,  %0,    0             \n\t"
+      "xvld            $xr9,  %0,    32            \n\t"
+      "xvld            $xr10, %0,    64            \n\t"  // load 32 pixels of RGB
+      "xvor.v          $xr12, $xr3,  $xr3          \n\t"
+      "xvor.v          $xr13, $xr3,  $xr3          \n\t"
+      "xvor.v          $xr11, $xr9,  $xr9          \n\t"
+      "addi.d          %2,    %2,    -32           \n\t"  // 32 processed per loop.
+      "xvpermi.q       $xr9,  $xr8,  0x30          \n\t"  //src0
+      "xvpermi.q       $xr8,  $xr10, 0x03          \n\t"  //src1
+      "xvpermi.q       $xr10, $xr11, 0x30          \n\t"  //src2
+      "xvshuf.b        $xr14, $xr8,  $xr9,  $xr4   \n\t"
+      "xvshuf.b        $xr15, $xr8,  $xr10, $xr5   \n\t"
+      "xvshuf.b        $xr16, $xr8,  $xr9,  $xr6   \n\t"
+      "xvshuf.b        $xr17, $xr8,  $xr10, $xr7   \n\t"
+      "xvmaddwev.h.bu  $xr12, $xr16, $xr1          \n\t"  //G
+      "xvmaddwev.h.bu  $xr13, $xr17, $xr1          \n\t"
+      "xvmaddwev.h.bu  $xr12, $xr14, $xr0          \n\t"  //B
+      "xvmaddwev.h.bu  $xr13, $xr15, $xr0          \n\t"
+      "xvmaddwod.h.bu  $xr12, $xr14, $xr2          \n\t"  //R
+      "xvmaddwod.h.bu  $xr13, $xr15, $xr2          \n\t"
+      "addi.d          %0,    %0,    96            \n\t"
+      "xvpickod.b      $xr10, $xr13, $xr12         \n\t"
+      "xvst            $xr10, %1,    0             \n\t"
+      "addi.d          %1,    %1,    32            \n\t"
+      "bnez            %2,    1b                   \n\t"
+      : "+&r"(src_rgba),    // %0
+        "+&r"(dst_y),       // %1
+        "+&r"(width)        // %2
+      : "r"(rgbconstants),  // %3
+        "r"(shuff)          // %4
+      : "memory"
+    );
+}
+
+void RGB24ToYJRow_LASX(const uint8_t* src_rgb24, uint8_t* dst_yj, int width) {
+  RGBToYMatrixRow_LASX(src_rgb24, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void RAWToYJRow_LASX(const uint8_t* src_raw, uint8_t* dst_yj, int width) {
+  RGBToYMatrixRow_LASX(src_raw, dst_yj, width, &kRawJPEGConstants);
+}
+
+void RGB24ToYRow_LASX(const uint8_t* src_rgb24, uint8_t* dst_y, int width) {
+  RGBToYMatrixRow_LASX(src_rgb24, dst_y, width, &kRgb24I601Constants);
+}
+
+void RAWToYRow_LASX(const uint8_t* src_raw, uint8_t* dst_y, int width) {
+  RGBToYMatrixRow_LASX(src_raw, dst_y, width, &kRawI601Constants);
 }
 
 void ARGBToUVJRow_LASX(const uint8_t* src_argb,
diff --git a/source/row_lsx.cc b/source/row_lsx.cc
index 3e8b901a..0825b633 100644
--- a/source/row_lsx.cc
+++ b/source/row_lsx.cc
@@ -561,39 +561,6 @@ void RGB565ToUVRow_LSX(const uint8_t* src_rgb565,
   }
 }
 
-void RGB24ToYRow_LSX(const uint8_t* src_rgb24, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 16;
-  __m128i src0, src1, src2;
-  __m128i tmp0, tmp1, tmp2, tmp3;
-  __m128i reg0, reg1, dst0;
-  __m128i const_129 = __lsx_vldi(129);
-  __m128i const_br = {0x4219421942194219, 0x4219421942194219};
-  __m128i const_1080 = {0x1080108010801080, 0x1080108010801080};
-  __m128i shuff0 = {0x0B09080605030200, 0x17151412110F0E0C};
-  __m128i shuff1 = {0x0301001E1D1B1A18, 0x0F0D0C0A09070604};
-  __m128i shuff2 = {0x000A000700040001, 0x001600130010000D};
-  __m128i shuff3 = {0x0002001F001C0019, 0x000E000B00080005};
-
-  for (x = 0; x < len; x++) {
-    src0 = __lsx_vld(src_rgb24, 0);
-    src1 = __lsx_vld(src_rgb24, 16);
-    src2 = __lsx_vld(src_rgb24, 32);
-    tmp0 = __lsx_vshuf_b(src1, src0, shuff0);
-    tmp1 = __lsx_vshuf_b(src1, src2, shuff1);
-    tmp2 = __lsx_vshuf_b(src1, src0, shuff2);
-    tmp3 = __lsx_vshuf_b(src1, src2, shuff3);
-    reg0 = __lsx_vmaddwev_h_bu(const_1080, tmp2, const_129);
-    reg1 = __lsx_vmaddwev_h_bu(const_1080, tmp3, const_129);
-    reg0 = __lsx_vdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lsx_vdp2add_h_bu(reg1, const_br, tmp1);
-    dst0 = __lsx_vpickod_b(reg1, reg0);
-    __lsx_vst(dst0, dst_y, 0);
-    dst_y += 16;
-    src_rgb24 += 48;
-  }
-}
-
 void RGB24ToUVRow_LSX(const uint8_t* src_rgb24,
                       int src_stride_rgb24,
                       uint8_t* dst_u,
@@ -647,39 +614,6 @@ void RGB24ToUVRow_LSX(const uint8_t* src_rgb24,
   }
 }
 
-void RAWToYRow_LSX(const uint8_t* src_raw, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 16;
-  __m128i src0, src1, src2;
-  __m128i tmp0, tmp1, tmp2, tmp3;
-  __m128i reg0, reg1, dst0;
-  __m128i const_129 = __lsx_vldi(129);
-  __m128i const_br = {0x1942194219421942, 0x1942194219421942};
-  __m128i const_1080 = {0x1080108010801080, 0x1080108010801080};
-  __m128i shuff0 = {0x0B09080605030200, 0x17151412110F0E0C};
-  __m128i shuff1 = {0x0301001E1D1B1A18, 0x0F0D0C0A09070604};
-  __m128i shuff2 = {0x000A000700040001, 0x001600130010000D};
-  __m128i shuff3 = {0x0002001F001C0019, 0x000E000B00080005};
-
-  for (x = 0; x < len; x++) {
-    src0 = __lsx_vld(src_raw, 0);
-    src1 = __lsx_vld(src_raw, 16);
-    src2 = __lsx_vld(src_raw, 32);
-    tmp0 = __lsx_vshuf_b(src1, src0, shuff0);
-    tmp1 = __lsx_vshuf_b(src1, src2, shuff1);
-    tmp2 = __lsx_vshuf_b(src1, src0, shuff2);
-    tmp3 = __lsx_vshuf_b(src1, src2, shuff3);
-    reg0 = __lsx_vmaddwev_h_bu(const_1080, tmp2, const_129);
-    reg1 = __lsx_vmaddwev_h_bu(const_1080, tmp3, const_129);
-    reg0 = __lsx_vdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lsx_vdp2add_h_bu(reg1, const_br, tmp1);
-    dst0 = __lsx_vsrlni_b_h(reg1, reg0, 8);
-    __lsx_vst(dst0, dst_y, 0);
-    dst_y += 16;
-    src_raw += 48;
-  }
-}
-
 void RAWToUVRow_LSX(const uint8_t* src_raw,
                     int src_stride_raw,
                     uint8_t* dst_u,
@@ -914,62 +848,6 @@ void SobelXYRow_LSX(const uint8_t* src_sobelx,
   }
 }
 
-void ARGBToYJRow_LSX(const uint8_t* src_argb, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 16;
-  __m128i src0, src1, src2, src3, dst0;
-  __m128i tmp0, tmp1, tmp2, tmp3;
-  __m128i reg0, reg1;
-  __m128i const_128 = __lsx_vldi(0x480);
-  __m128i const_150 = __lsx_vldi(0x96);
-  __m128i const_br = {0x4D1D4D1D4D1D4D1D, 0x4D1D4D1D4D1D4D1D};
-
-  for (x = 0; x < len; x++) {
-    DUP4_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src_argb, 32, src_argb, 48,
-              src0, src1, src2, src3);
-    tmp0 = __lsx_vpickev_b(src1, src0);
-    tmp1 = __lsx_vpickod_b(src1, src0);
-    tmp2 = __lsx_vpickev_b(src3, src2);
-    tmp3 = __lsx_vpickod_b(src3, src2);
-    reg0 = __lsx_vmaddwev_h_bu(const_128, tmp1, const_150);
-    reg1 = __lsx_vmaddwev_h_bu(const_128, tmp3, const_150);
-    reg0 = __lsx_vdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lsx_vdp2add_h_bu(reg1, const_br, tmp2);
-    dst0 = __lsx_vpickod_b(reg1, reg0);
-    __lsx_vst(dst0, dst_y, 0);
-    dst_y += 16;
-    src_argb += 64;
-  }
-}
-
-void BGRAToYRow_LSX(const uint8_t* src_bgra, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 16;
-  __m128i src0, src1, src2, src3, dst0;
-  __m128i tmp0, tmp1, tmp2, tmp3;
-  __m128i reg0, reg1;
-  __m128i const_129 = __lsx_vldi(0x81);
-  __m128i const_br = {0x1942194219421942, 0x1942194219421942};
-  __m128i const_1080 = {0x1080108010801080, 0x1080108010801080};
-
-  for (x = 0; x < len; x++) {
-    DUP4_ARG2(__lsx_vld, src_bgra, 0, src_bgra, 16, src_bgra, 32, src_bgra, 48,
-              src0, src1, src2, src3);
-    tmp0 = __lsx_vpickod_b(src1, src0);
-    tmp1 = __lsx_vpickev_b(src1, src0);
-    tmp2 = __lsx_vpickod_b(src3, src2);
-    tmp3 = __lsx_vpickev_b(src3, src2);
-    reg0 = __lsx_vmaddwod_h_bu(const_1080, tmp1, const_129);
-    reg1 = __lsx_vmaddwod_h_bu(const_1080, tmp3, const_129);
-    reg0 = __lsx_vdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lsx_vdp2add_h_bu(reg1, const_br, tmp2);
-    dst0 = __lsx_vsrlni_b_h(reg1, reg0, 8);
-    __lsx_vst(dst0, dst_y, 0);
-    dst_y += 16;
-    src_bgra += 64;
-  }
-}
-
 void BGRAToUVRow_LSX(const uint8_t* src_bgra,
                      int src_stride_bgra,
                      uint8_t* dst_u,
@@ -1018,34 +896,6 @@ void BGRAToUVRow_LSX(const uint8_t* src_bgra,
   }
 }
 
-void ABGRToYRow_LSX(const uint8_t* src_abgr, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 16;
-  __m128i src0, src1, src2, src3, dst0;
-  __m128i tmp0, tmp1, tmp2, tmp3;
-  __m128i reg0, reg1;
-  __m128i const_129 = __lsx_vldi(0x81);
-  __m128i const_br = {0x1942194219421942, 0x1942194219421942};
-  __m128i const_1080 = {0x1080108010801080, 0x1080108010801080};
-
-  for (x = 0; x < len; x++) {
-    DUP4_ARG2(__lsx_vld, src_abgr, 0, src_abgr, 16, src_abgr, 32, src_abgr, 48,
-              src0, src1, src2, src3);
-    tmp0 = __lsx_vpickev_b(src1, src0);
-    tmp1 = __lsx_vpickod_b(src1, src0);
-    tmp2 = __lsx_vpickev_b(src3, src2);
-    tmp3 = __lsx_vpickod_b(src3, src2);
-    reg0 = __lsx_vmaddwev_h_bu(const_1080, tmp1, const_129);
-    reg1 = __lsx_vmaddwev_h_bu(const_1080, tmp3, const_129);
-    reg0 = __lsx_vdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lsx_vdp2add_h_bu(reg1, const_br, tmp2);
-    dst0 = __lsx_vsrlni_b_h(reg1, reg0, 8);
-    __lsx_vst(dst0, dst_y, 0);
-    dst_y += 16;
-    src_abgr += 64;
-  }
-}
-
 void ABGRToUVRow_LSX(const uint8_t* src_abgr,
                      int src_stride_abgr,
                      uint8_t* dst_u,
@@ -1094,34 +944,6 @@ void ABGRToUVRow_LSX(const uint8_t* src_abgr,
   }
 }
 
-void RGBAToYRow_LSX(const uint8_t* src_rgba, uint8_t* dst_y, int width) {
-  int x;
-  int len = width / 16;
-  __m128i src0, src1, src2, src3, dst0;
-  __m128i tmp0, tmp1, tmp2, tmp3;
-  __m128i reg0, reg1;
-  __m128i const_129 = __lsx_vldi(0x81);
-  __m128i const_br = {0x4219421942194219, 0x4219421942194219};
-  __m128i const_1080 = {0x1080108010801080, 0x1080108010801080};
-
-  for (x = 0; x < len; x++) {
-    DUP4_ARG2(__lsx_vld, src_rgba, 0, src_rgba, 16, src_rgba, 32, src_rgba, 48,
-              src0, src1, src2, src3);
-    tmp0 = __lsx_vpickod_b(src1, src0);
-    tmp1 = __lsx_vpickev_b(src1, src0);
-    tmp2 = __lsx_vpickod_b(src3, src2);
-    tmp3 = __lsx_vpickev_b(src3, src2);
-    reg0 = __lsx_vmaddwod_h_bu(const_1080, tmp1, const_129);
-    reg1 = __lsx_vmaddwod_h_bu(const_1080, tmp3, const_129);
-    reg0 = __lsx_vdp2add_h_bu(reg0, const_br, tmp0);
-    reg1 = __lsx_vdp2add_h_bu(reg1, const_br, tmp2);
-    dst0 = __lsx_vsrlni_b_h(reg1, reg0, 8);
-    __lsx_vst(dst0, dst_y, 0);
-    dst_y += 16;
-    src_rgba += 64;
-  }
-}
-
 void RGBAToUVRow_LSX(const uint8_t* src_rgba,
                      int src_stride_rgba,
                      uint8_t* dst_u,
@@ -1821,6 +1643,212 @@ void HalfFloatRow_LSX(const uint16_t* src,
   }
 }
 
+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.
+static void ARGBToYMatrixRow_LSX(const uint8_t* src_argb,
+                                 uint8_t* dst_y,
+                                 int width,
+                                 const struct RgbConstants* rgbconstants) {
+    asm volatile(
+      "vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
+      "vldrepl.b      $vr1,  %3,    1             \n\t"  // load rgbconstants
+      "vldrepl.b      $vr2,  %3,    2             \n\t"  // load rgbconstants
+      "vldrepl.h      $vr3,  %3,    4             \n\t"  // load rgbconstants
+      "1:                                         \n\t"
+      "vld            $vr4,  %0,    0             \n\t"
+      "vld            $vr5,  %0,    16            \n\t"
+      "vld            $vr6,  %0,    32            \n\t"
+      "vld            $vr7,  %0,    48            \n\t"  // load 16 pixels of ARGB
+      "vor.v          $vr12, $vr3,  $vr3          \n\t"
+      "vor.v          $vr13, $vr3,  $vr3          \n\t"
+      "addi.d         %2,    %2,    -16           \n\t"  // 16 processed per loop.
+      "vpickev.b      $vr8,  $vr5,  $vr4          \n\t"  //BR
+      "vpickev.b      $vr10, $vr7,  $vr6          \n\t"
+      "vpickod.b      $vr9,  $vr5,  $vr4          \n\t"  //GA
+      "vpickod.b      $vr11, $vr7,  $vr6          \n\t"
+      "vmaddwev.h.bu  $vr12, $vr8,  $vr0          \n\t"  //B
+      "vmaddwev.h.bu  $vr13, $vr10, $vr0          \n\t"
+      "vmaddwev.h.bu  $vr12, $vr9,  $vr1          \n\t"  //G
+      "vmaddwev.h.bu  $vr13, $vr11, $vr1          \n\t"
+      "vmaddwod.h.bu  $vr12, $vr8,  $vr2          \n\t"  //R
+      "vmaddwod.h.bu  $vr13, $vr10, $vr2          \n\t"
+      "addi.d         %0,    %0,    64            \n\t"
+      "vpickod.b      $vr10, $vr13, $vr12         \n\t"
+      "vst            $vr10, %1,    0             \n\t"
+      "addi.d         %1,    %1,    16            \n\t"
+      "bnez           %2,    1b                   \n\t"
+      : "+&r"(src_argb),    // %0
+        "+&r"(dst_y),       // %1
+        "+&r"(width)        // %2
+      : "r"(rgbconstants)
+      : "memory"
+    );
+}
+
+void ARGBToYRow_LSX(const uint8_t* src_argb, uint8_t* dst_y, int width) {
+  ARGBToYMatrixRow_LSX(src_argb, dst_y, width, &kRgb24I601Constants);
+}
+
+void ARGBToYJRow_LSX(const uint8_t* src_argb, uint8_t* dst_yj, int width) {
+  ARGBToYMatrixRow_LSX(src_argb, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void ABGRToYRow_LSX(const uint8_t* src_abgr, uint8_t* dst_y, int width) {
+  ARGBToYMatrixRow_LSX(src_abgr, dst_y, width, &kRawI601Constants);
+}
+
+void ABGRToYJRow_LSX(const uint8_t* src_abgr, uint8_t* dst_yj, int width) {
+  ARGBToYMatrixRow_LSX(src_abgr, dst_yj, width, &kRawJPEGConstants);
+}
+
+// RGBA expects first value to be A and ignored, then 3 values to contain RGB.
+// Same code as ARGB, except the LD4
+static void RGBAToYMatrixRow_LSX(const uint8_t* src_rgba,
+                                 uint8_t* dst_y,
+                                 int width,
+                                 const struct RgbConstants* rgbconstants) {
+    asm volatile(
+      "vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
+      "vldrepl.b      $vr1,  %3,    1             \n\t"  // load rgbconstants
+      "vldrepl.b      $vr2,  %3,    2             \n\t"  // load rgbconstants
+      "vldrepl.h      $vr3,  %3,    4             \n\t"  // load rgbconstants
+      "1:                                         \n\t"
+      "vld            $vr4,  %0,    0             \n\t"
+      "vld            $vr5,  %0,    16            \n\t"
+      "vld            $vr6,  %0,    32            \n\t"
+      "vld            $vr7,  %0,    48            \n\t"  // load 16 pixels of RGBA
+      "vor.v          $vr12, $vr3,  $vr3          \n\t"
+      "vor.v          $vr13, $vr3,  $vr3          \n\t"
+      "addi.d         %2,    %2,    -16           \n\t"  // 16 processed per loop.
+      "vpickev.b      $vr8,  $vr5,  $vr4          \n\t"  //AG
+      "vpickev.b      $vr10, $vr7,  $vr6          \n\t"
+      "vpickod.b      $vr9,  $vr5,  $vr4          \n\t"  //BR
+      "vpickod.b      $vr11, $vr7,  $vr6          \n\t"
+      "vmaddwev.h.bu  $vr12, $vr9,  $vr0          \n\t"  //B
+      "vmaddwev.h.bu  $vr13, $vr11, $vr0          \n\t"
+      "vmaddwod.h.bu  $vr12, $vr8,  $vr1          \n\t"  //G
+      "vmaddwod.h.bu  $vr13, $vr10, $vr1          \n\t"
+      "vmaddwod.h.bu  $vr12, $vr9,  $vr2          \n\t"  //R
+      "vmaddwod.h.bu  $vr13, $vr11, $vr2          \n\t"
+      "addi.d         %0,    %0,    64            \n\t"
+      "vpickod.b      $vr10, $vr13, $vr12         \n\t"
+      "vst            $vr10, %1,    0             \n\t"
+      "addi.d         %1,    %1,    16            \n\t"
+      "bnez           %2,    1b                   \n\t"
+      : "+&r"(src_rgba),    // %0
+        "+&r"(dst_y),       // %1
+        "+&r"(width)        // %2
+      : "r"(rgbconstants)
+      : "memory"
+    );
+}
+
+void RGBAToYRow_LSX(const uint8_t* src_rgba, uint8_t* dst_y, int width) {
+  RGBAToYMatrixRow_LSX(src_rgba, dst_y, width, &kRgb24I601Constants);
+}
+
+void RGBAToYJRow_LSX(const uint8_t* src_rgba, uint8_t* dst_yj, int width) {
+  RGBAToYMatrixRow_LSX(src_rgba, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void BGRAToYRow_LSX(const uint8_t* src_bgra, uint8_t* dst_y, int width) {
+  RGBAToYMatrixRow_LSX(src_bgra, dst_y, width, &kRawI601Constants);
+}
+
+static void RGBToYMatrixRow_LSX(const uint8_t* src_rgba,
+                                uint8_t* dst_y,
+                                int width,
+                                const struct RgbConstants* rgbconstants) {
+    int8_t shuff[64] = {0, 2, 3, 5, 6, 8, 9, 11, 12, 14, 15, 17, 18, 20, 21, 23,
+                        24, 26, 27, 29, 30, 0, 1, 3, 4, 6, 7, 9, 10, 12, 13, 15,
+                        1, 0, 4, 0, 7, 0, 10, 0, 13, 0, 16, 0, 19, 0, 22, 0,
+                        25, 0, 28, 0, 31, 0, 2, 0, 5, 0, 8, 0, 11, 0, 14, 0};
+    asm volatile(
+      "vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
+      "vldrepl.b      $vr1,  %3,    1             \n\t"  // load rgbconstants
+      "vldrepl.b      $vr2,  %3,    2             \n\t"  // load rgbconstants
+      "vldrepl.h      $vr3,  %3,    4             \n\t"  // load rgbconstants
+      "vld            $vr4,  %4,    0             \n\t"  // load shuff
+      "vld            $vr5,  %4,    16            \n\t"
+      "vld            $vr6,  %4,    32            \n\t"
+      "vld            $vr7,  %4,    48            \n\t"
+      "1:                                         \n\t"
+      "vld            $vr8,  %0,    0             \n\t"
+      "vld            $vr9,  %0,    16            \n\t"
+      "vld            $vr10, %0,    32            \n\t"  // load 16 pixels of RGB
+      "vor.v          $vr12, $vr3,  $vr3          \n\t"
+      "vor.v          $vr13, $vr3,  $vr3          \n\t"
+      "addi.d         %2,    %2,    -16           \n\t"  // 16 processed per loop.
+      "vshuf.b        $vr14, $vr9,  $vr8,  $vr4   \n\t"
+      "vshuf.b        $vr15, $vr9,  $vr10, $vr5   \n\t"
+      "vshuf.b        $vr16, $vr9,  $vr8,  $vr6   \n\t"
+      "vshuf.b        $vr17, $vr9,  $vr10, $vr7   \n\t"
+      "vmaddwev.h.bu  $vr12, $vr16, $vr1          \n\t"  //G
+      "vmaddwev.h.bu  $vr13, $vr17, $vr1          \n\t"
+      "vmaddwev.h.bu  $vr12, $vr14, $vr0          \n\t"  //B
+      "vmaddwev.h.bu  $vr13, $vr15, $vr0          \n\t"
+      "vmaddwod.h.bu  $vr12, $vr14, $vr2          \n\t"  //R
+      "vmaddwod.h.bu  $vr13, $vr15, $vr2          \n\t"
+      "addi.d         %0,    %0,    48            \n\t"
+      "vpickod.b      $vr10, $vr13, $vr12         \n\t"
+      "vst            $vr10, %1,    0             \n\t"
+      "addi.d         %1,    %1,    16            \n\t"
+      "bnez           %2,    1b                   \n\t"
+      : "+&r"(src_rgba),    // %0
+        "+&r"(dst_y),       // %1
+        "+&r"(width)        // %2
+      : "r"(rgbconstants),  // %3
+        "r"(shuff)          // %4
+      : "memory"
+    );
+}
+
+void RGB24ToYJRow_LSX(const uint8_t* src_rgb24, uint8_t* dst_yj, int width) {
+  RGBToYMatrixRow_LSX(src_rgb24, dst_yj, width, &kRgb24JPEGConstants);
+}
+
+void RAWToYJRow_LSX(const uint8_t* src_raw, uint8_t* dst_yj, int width) {
+  RGBToYMatrixRow_LSX(src_raw, dst_yj, width, &kRawJPEGConstants);
+}
+
+void RGB24ToYRow_LSX(const uint8_t* src_rgb24, uint8_t* dst_y, int width) {
+  RGBToYMatrixRow_LSX(src_rgb24, dst_y, width, &kRgb24I601Constants);
+}
+
+void RAWToYRow_LSX(const uint8_t* src_raw, uint8_t* dst_y, int width) {
+  RGBToYMatrixRow_LSX(src_raw, dst_y, width, &kRawI601Constants);
+}
+
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv