libyuv:loongson optimize compare/row/scale/rotate files with mmi.

Currently, libyuv supports MIPS SIMD Arch(MSA),
but libyuv does not supports MultiMedia Instruction(MMI)(such as loongson3a platform).

In order to improve performance of libyuv on loongson3a platform,
this provides optimize 98 functions with mmi.

BUG=libyuv:804

Change-Id: I8947626009efad769b3103a867363ece25d79629
Reviewed-on: https://chromium-review.googlesource.com/1122064
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>

5 files changed, 715 insertions, 2 deletions

diff --git a/include/libyuv/compare_row.h b/include/libyuv/compare_row.h
index e81f7455..e95b9d93 100644
--- a/include/libyuv/compare_row.h
+++ b/include/libyuv/compare_row.h
@@ -84,6 +84,11 @@ extern "C" {
 #define HAS_SUMSQUAREERROR_MSA
 #endif
 
+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
+#define HAS_HAMMINGDISTANCE_MMI
+#define HAS_SUMSQUAREERROR_MMI
+#endif
+
 uint32_t HammingDistance_C(const uint8_t* src_a,
                            const uint8_t* src_b,
                            int count);
@@ -102,7 +107,9 @@ uint32_t HammingDistance_NEON(const uint8_t* src_a,
 uint32_t HammingDistance_MSA(const uint8_t* src_a,
                              const uint8_t* src_b,
                              int count);
-
+uint32_t HammingDistance_MMI(const uint8_t* src_a,
+                             const uint8_t* src_b,
+                             int count);
 uint32_t SumSquareError_C(const uint8_t* src_a,
                           const uint8_t* src_b,
                           int count);
@@ -118,6 +125,9 @@ uint32_t SumSquareError_NEON(const uint8_t* src_a,
 uint32_t SumSquareError_MSA(const uint8_t* src_a,
                             const uint8_t* src_b,
                             int count);
+uint32_t SumSquareError_MMI(const uint8_t* src_a,
+                            const uint8_t* src_b,
+                            int count);
 
 uint32_t HashDjb2_C(const uint8_t* src, int count, uint32_t seed);
 uint32_t HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed);
diff --git a/include/libyuv/cpu_id.h b/include/libyuv/cpu_id.h
index 0229cb5e..b01cd25c 100644
--- a/include/libyuv/cpu_id.h
+++ b/include/libyuv/cpu_id.h
@@ -48,6 +48,7 @@ static const int kCpuHasAVX512VPOPCNTDQ = 0x100000;
 // These flags are only valid on MIPS processors.
 static const int kCpuHasMIPS = 0x200000;
 static const int kCpuHasMSA = 0x400000;
+static const int kCpuHasMMI = 0x800000;
 
 // Optional init function. TestCpuFlag does an auto-init.
 // Returns cpu_info flags.
diff --git a/include/libyuv/rotate_row.h b/include/libyuv/rotate_row.h
index 5edc0fcf..022293ee 100644..100755
--- a/include/libyuv/rotate_row.h
+++ b/include/libyuv/rotate_row.h
@@ -60,6 +60,11 @@ extern "C" {
 #define HAS_TRANSPOSEUVWX16_MSA
 #endif
 
+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
+#define HAS_TRANSPOSEWX8_MMI
+#define HAS_TRANSPOSEUVWX8_MMI
+#endif
+
 void TransposeWxH_C(const uint8_t* src,
                     int src_stride,
                     uint8_t* dst,
@@ -87,6 +92,11 @@ void TransposeWx8_SSSE3(const uint8_t* src,
                         uint8_t* dst,
                         int dst_stride,
                         int width);
+void TransposeWx8_MMI(const uint8_t* src,
+                      int src_stride,
+                      uint8_t* dst,
+                      int dst_stride,
+                      int width);
 void TransposeWx8_Fast_SSSE3(const uint8_t* src,
                              int src_stride,
                              uint8_t* dst,
@@ -108,6 +118,11 @@ void TransposeWx8_Any_SSSE3(const uint8_t* src,
                             uint8_t* dst,
                             int dst_stride,
                             int width);
+void TransposeWx8_Any_MMI(const uint8_t* src,
+                          int src_stride,
+                          uint8_t* dst,
+                          int dst_stride,
+                          int width);
 void TransposeWx8_Fast_Any_SSSE3(const uint8_t* src,
                                  int src_stride,
                                  uint8_t* dst,
@@ -156,6 +171,13 @@ void TransposeUVWx8_NEON(const uint8_t* src,
                          uint8_t* dst_b,
                          int dst_stride_b,
                          int width);
+void TransposeUVWx8_MMI(const uint8_t* src,
+                        int src_stride,
+                        uint8_t* dst_a,
+                        int dst_stride_a,
+                        uint8_t* dst_b,
+                        int dst_stride_b,
+                        int width);
 void TransposeUVWx16_MSA(const uint8_t* src,
                          int src_stride,
                          uint8_t* dst_a,
@@ -178,6 +200,13 @@ void TransposeUVWx8_Any_NEON(const uint8_t* src,
                              uint8_t* dst_b,
                              int dst_stride_b,
                              int width);
+void TransposeUVWx8_Any_MMI(const uint8_t* src,
+                            int src_stride,
+                            uint8_t* dst_a,
+                            int dst_stride_a,
+                            uint8_t* dst_b,
+                            int dst_stride_b,
+                            int width);
 void TransposeUVWx16_Any_MSA(const uint8_t* src,
                              int src_stride,
                              uint8_t* dst_a,
diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index 1468f4b9..06ab3756 100644..100755
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -483,6 +483,81 @@ extern "C" {
 #define HAS_YUY2TOYROW_MSA
 #endif
 
+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
+#define HAS_ABGRTOUVROW_MMI
+#define HAS_ABGRTOYROW_MMI
+#define HAS_ARGB1555TOARGBROW_MMI
+#define HAS_ARGB1555TOUVROW_MMI
+#define HAS_ARGB1555TOYROW_MMI
+#define HAS_ARGB4444TOARGBROW_MMI
+#define HAS_ARGB4444TOUVROW_MMI
+#define HAS_ARGB4444TOYROW_MMI
+#define HAS_ARGBADDROW_MMI
+#define HAS_ARGBATTENUATEROW_MMI
+#define HAS_ARGBBLENDROW_MMI
+#define HAS_ARGBCOLORMATRIXROW_MMI
+#define HAS_ARGBCOPYALPHAROW_MMI
+#define HAS_ARGBCOPYYTOALPHAROW_MMI
+#define HAS_ARGBEXTRACTALPHAROW_MMI
+#define HAS_ARGBGRAYROW_MMI
+#define HAS_ARGBMIRRORROW_MMI
+#define HAS_ARGBMULTIPLYROW_MMI
+#define HAS_ARGBSEPIAROW_MMI
+#define HAS_ARGBSHADEROW_MMI
+#define HAS_ARGBSHUFFLEROW_MMI
+#define HAS_ARGBSUBTRACTROW_MMI
+#define HAS_ARGBTOARGB1555ROW_MMI
+#define HAS_ARGBTOARGB4444ROW_MMI
+#define HAS_ARGBTORAWROW_MMI
+#define HAS_ARGBTORGB24ROW_MMI
+#define HAS_ARGBTORGB565DITHERROW_MMI
+#define HAS_ARGBTORGB565ROW_MMI
+#define HAS_ARGBTOUV444ROW_MMI
+#define HAS_ARGBTOUVJROW_MMI
+#define HAS_ARGBTOUVROW_MMI
+#define HAS_ARGBTOYJROW_MMI
+#define HAS_ARGBTOYROW_MMI
+#define HAS_BGRATOUVROW_MMI
+#define HAS_BGRATOYROW_MMI
+#define HAS_BLENDPLANEROW_MMI
+#define HAS_COMPUTECUMULATIVESUMROW_MMI
+#define HAS_CUMULATIVESUMTOAVERAGEROW_MMI
+#define HAS_HALFFLOATROW_MMI
+#define HAS_I400TOARGBROW_MMI
+#define HAS_I422TOUYVYROW_MMI
+#define HAS_I422TOYUY2ROW_MMI
+#define HAS_INTERPOLATEROW_MMI
+#define HAS_J400TOARGBROW_MMI
+#define HAS_MERGERGBROW_MMI
+#define HAS_MERGEUVROW_MMI
+#define HAS_MIRRORROW_MMI
+#define HAS_MIRRORUVROW_MMI
+#define HAS_RAWTOARGBROW_MMI
+#define HAS_RAWTORGB24ROW_MMI
+#define HAS_RAWTOUVROW_MMI
+#define HAS_RAWTOYROW_MMI
+#define HAS_RGB24TOARGBROW_MMI
+#define HAS_RGB24TOUVROW_MMI
+#define HAS_RGB24TOYROW_MMI
+#define HAS_RGB565TOARGBROW_MMI
+#define HAS_RGB565TOUVROW_MMI
+#define HAS_RGB565TOYROW_MMI
+#define HAS_RGBATOUVROW_MMI
+#define HAS_RGBATOYROW_MMI
+#define HAS_SOBELROW_MMI
+#define HAS_SOBELTOPLANEROW_MMI
+#define HAS_SOBELXROW_MMI
+#define HAS_SOBELXYROW_MMI
+#define HAS_SOBELYROW_MMI
+#define HAS_SPLITRGBROW_MMI
+#define HAS_SPLITUVROW_MMI
+#define HAS_UYVYTOUVROW_MMI
+#define HAS_UYVYTOYROW_MMI
+#define HAS_YUY2TOUV422ROW_MMI
+#define HAS_YUY2TOUVROW_MMI
+#define HAS_YUY2TOYROW_MMI
+#endif
+
 #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
 #if defined(VISUALC_HAS_AVX2)
 #define SIMD_ALIGNED(var) __declspec(align(32)) var
@@ -837,6 +912,8 @@ 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_y, 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_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
+void ARGBToYJRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
 void ARGBToUV444Row_NEON(const uint8_t* src_argb,
                          uint8_t* dst_u,
                          uint8_t* dst_v,
@@ -855,6 +932,15 @@ void ARGBToUVRow_MSA(const uint8_t* src_argb0,
                      uint8_t* dst_u,
                      uint8_t* dst_v,
                      int width);
+void ARGBToUV444Row_MMI(const uint8_t* src_argb,
+                        uint8_t* dst_u,
+                        uint8_t* dst_v,
+                        int width);
+void ARGBToUVRow_MMI(const uint8_t* src_argb0,
+                     int src_stride_argb,
+                     uint8_t* dst_u,
+                     uint8_t* dst_v,
+                     int width);
 void ARGBToUVJRow_NEON(const uint8_t* src_argb,
                        int src_stride_argb,
                        uint8_t* dst_u,
@@ -940,6 +1026,51 @@ void ARGB1555ToUVRow_MSA(const uint8_t* src_argb1555,
                          uint8_t* dst_u,
                          uint8_t* dst_v,
                          int width);
+void ARGBToUVJRow_MMI(const uint8_t* src_rgb0,
+                      int src_stride_rgb,
+                      uint8_t* dst_u,
+                      uint8_t* dst_v,
+                      int width);
+void BGRAToUVRow_MMI(const uint8_t* src_rgb0,
+                     int src_stride_rgb,
+                     uint8_t* dst_u,
+                     uint8_t* dst_v,
+                     int width);
+void ABGRToUVRow_MMI(const uint8_t* src_rgb0,
+                     int src_stride_rgb,
+                     uint8_t* dst_u,
+                     uint8_t* dst_v,
+                     int width);
+void RGBAToUVRow_MMI(const uint8_t* src_rgb0,
+                     int src_stride_rgb,
+                     uint8_t* dst_u,
+                     uint8_t* dst_v,
+                     int width);
+void RGB24ToUVRow_MMI(const uint8_t* src_rgb0,
+                      int src_stride_rgb,
+                      uint8_t* dst_u,
+                      uint8_t* dst_v,
+                      int width);
+void RAWToUVRow_MMI(const uint8_t* src_rgb0,
+                    int src_stride_rgb,
+                    uint8_t* dst_u,
+                    uint8_t* dst_v,
+                    int width);
+void RGB565ToUVRow_MMI(const uint8_t* src_rgb565,
+                       int src_stride_rgb565,
+                       uint8_t* dst_u,
+                       uint8_t* dst_v,
+                       int width);
+void ARGB1555ToUVRow_MMI(const uint8_t* src_argb1555,
+                         int src_stride_argb1555,
+                         uint8_t* dst_u,
+                         uint8_t* dst_v,
+                         int width);
+void ARGB4444ToUVRow_MMI(const uint8_t* src_argb4444,
+                          int src_stride_argb4444,
+                          uint8_t* dst_u,
+                          uint8_t* dst_v,
+                          int width);
 void BGRAToYRow_NEON(const uint8_t* src_bgra, uint8_t* dst_y, int width);
 void ABGRToYRow_NEON(const uint8_t* src_abgr, uint8_t* dst_y, int width);
 void RGBAToYRow_NEON(const uint8_t* src_rgba, uint8_t* dst_y, int width);
@@ -959,6 +1090,17 @@ void RGB24ToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width);
 void RAWToYRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width);
 void RGB565ToYRow_MSA(const uint8_t* src_rgb565, uint8_t* dst_y, int width);
 void ARGB1555ToYRow_MSA(const uint8_t* src_argb1555, uint8_t* dst_y, int width);
+void BGRAToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
+void ABGRToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
+void RGBAToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
+void RGB24ToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
+void RAWToYRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width);
+void RGB565ToYRow_MMI(const uint8_t* src_rgb565, uint8_t* dst_y, int width);
+void ARGB1555ToYRow_MMI(const uint8_t* src_argb1555, uint8_t* dst_y, int width);
+void ARGB4444ToYRow_MMI(const uint8_t* src_argb4444,
+                         uint8_t* dst_y,
+                         int width);
+
 void ARGBToYRow_C(const uint8_t* src_argb0, uint8_t* dst_y, int width);
 void ARGBToYJRow_C(const uint8_t* src_argb0, uint8_t* dst_y, int width);
 void BGRAToYRow_C(const uint8_t* src_argb0, uint8_t* dst_y, int width);
@@ -1001,6 +1143,20 @@ void RGB565ToYRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void ARGB1555ToYRow_Any_MSA(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             int width);
+void BGRAToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ABGRToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RGBAToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ARGBToYJRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ARGBToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RGB24ToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RAWToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RGB565ToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ARGB1555ToYRow_Any_MMI(const uint8_t* src_ptr,
+                            uint8_t* dst_ptr,
+                            int width);
+void ARGB4444ToYRow_Any_MMI(const uint8_t* src_ptr,
+                             uint8_t* dst_ptr,
+                             int width);
 
 void ARGBToUVRow_AVX2(const uint8_t* src_argb0,
                       int src_stride_argb,
@@ -1090,6 +1246,15 @@ void ARGBToUVRow_Any_MSA(const uint8_t* src_ptr,
                          uint8_t* dst_u,
                          uint8_t* dst_v,
                          int width);
+void ARGBToUV444Row_Any_MMI(const uint8_t* src_ptr,
+                            uint8_t* dst_u,
+                            uint8_t* dst_v,
+                            int width);
+void ARGBToUVRow_Any_MMI(const uint8_t* src_ptr,
+                         int src_stride_ptr,
+                         uint8_t* dst_u,
+                         uint8_t* dst_v,
+                         int width);
 void ARGBToUVJRow_Any_NEON(const uint8_t* src_ptr,
                            int src_stride_ptr,
                            uint8_t* dst_u,
@@ -1175,6 +1340,51 @@ void ARGB1555ToUVRow_Any_MSA(const uint8_t* src_ptr,
                              uint8_t* dst_u,
                              uint8_t* dst_v,
                              int width);
+void ARGBToUVJRow_Any_MMI(const uint8_t* src_ptr,
+                          int src_stride_ptr,
+                          uint8_t* dst_u,
+                          uint8_t* dst_v,
+                          int width);
+void BGRAToUVRow_Any_MMI(const uint8_t* src_ptr,
+                         int src_stride_ptr,
+                         uint8_t* dst_u,
+                         uint8_t* dst_v,
+                         int width);
+void ABGRToUVRow_Any_MMI(const uint8_t* src_ptr,
+                         int src_stride_ptr,
+                         uint8_t* dst_u,
+                         uint8_t* dst_v,
+                         int width);
+void RGBAToUVRow_Any_MMI(const uint8_t* src_ptr,
+                         int src_stride_ptr,
+                         uint8_t* dst_u,
+                         uint8_t* dst_v,
+                         int width);
+void RGB24ToUVRow_Any_MMI(const uint8_t* src_ptr,
+                          int src_stride_ptr,
+                          uint8_t* dst_u,
+                          uint8_t* dst_v,
+                          int width);
+void RAWToUVRow_Any_MMI(const uint8_t* src_ptr,
+                        int src_stride_ptr,
+                        uint8_t* dst_u,
+                        uint8_t* dst_v,
+                        int width);
+void RGB565ToUVRow_Any_MMI(const uint8_t* src_ptr,
+                           int src_stride_ptr,
+                           uint8_t* dst_u,
+                           uint8_t* dst_v,
+                           int width);
+void ARGB1555ToUVRow_Any_MMI(const uint8_t* src_ptr,
+                             int src_stride_ptr,
+                             uint8_t* dst_u,
+                             uint8_t* dst_v,
+                             int width);
+void ARGB4444ToUVRow_Any_MMI(const uint8_t* src_ptr,
+                              int src_stride_ptr,
+                              uint8_t* dst_u,
+                              uint8_t* dst_v,
+                              int width);
 void ARGBToUVRow_C(const uint8_t* src_rgb0,
                    int src_stride_rgb,
                    uint8_t* dst_u,
@@ -1254,12 +1464,14 @@ void MirrorRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
 void MirrorRow_SSSE3(const uint8_t* src, uint8_t* dst, int width);
 void MirrorRow_NEON(const uint8_t* src, uint8_t* dst, int width);
 void MirrorRow_MSA(const uint8_t* src, uint8_t* dst, int width);
+void MirrorRow_MMI(const uint8_t* src, uint8_t* dst, int width);
 void MirrorRow_C(const uint8_t* src, uint8_t* dst, int width);
 void MirrorRow_Any_AVX2(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void MirrorRow_Any_SSSE3(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void MirrorRow_Any_SSE2(const uint8_t* src, uint8_t* dst, int width);
 void MirrorRow_Any_NEON(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void MirrorRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void MirrorRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 
 void MirrorUVRow_SSSE3(const uint8_t* src,
                        uint8_t* dst_u,
@@ -1273,6 +1485,10 @@ void MirrorUVRow_MSA(const uint8_t* src_uv,
                      uint8_t* dst_u,
                      uint8_t* dst_v,
                      int width);
+void MirrorUVRow_MMI(const uint8_t* src_uv,
+                     uint8_t* dst_u,
+                     uint8_t* dst_v,
+                     int width);
 void MirrorUVRow_C(const uint8_t* src_uv,
                    uint8_t* dst_u,
                    uint8_t* dst_v,
@@ -1282,6 +1498,7 @@ void ARGBMirrorRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
 void ARGBMirrorRow_SSE2(const uint8_t* src, uint8_t* dst, int width);
 void ARGBMirrorRow_NEON(const uint8_t* src, uint8_t* dst, int width);
 void ARGBMirrorRow_MSA(const uint8_t* src, uint8_t* dst, int width);
+void ARGBMirrorRow_MMI(const uint8_t* src, uint8_t* dst, int width);
 void ARGBMirrorRow_C(const uint8_t* src, uint8_t* dst, int width);
 void ARGBMirrorRow_Any_AVX2(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
@@ -1293,6 +1510,7 @@ void ARGBMirrorRow_Any_NEON(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             int width);
 void ARGBMirrorRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ARGBMirrorRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 
 void SplitUVRow_C(const uint8_t* src_uv,
                   uint8_t* dst_u,
@@ -1314,6 +1532,10 @@ void SplitUVRow_MSA(const uint8_t* src_uv,
                     uint8_t* dst_u,
                     uint8_t* dst_v,
                     int width);
+void SplitUVRow_MMI(const uint8_t* src_uv,
+                    uint8_t* dst_u,
+                    uint8_t* dst_v,
+                    int width);
 void SplitUVRow_Any_SSE2(const uint8_t* src_ptr,
                          uint8_t* dst_u,
                          uint8_t* dst_v,
@@ -1330,6 +1552,10 @@ void SplitUVRow_Any_MSA(const uint8_t* src_ptr,
                         uint8_t* dst_u,
                         uint8_t* dst_v,
                         int width);
+void SplitUVRow_Any_MMI(const uint8_t* src_ptr,
+                        uint8_t* dst_u,
+                        uint8_t* dst_v,
+                        int width);
 
 void MergeUVRow_C(const uint8_t* src_u,
                   const uint8_t* src_v,
@@ -1351,6 +1577,10 @@ void MergeUVRow_MSA(const uint8_t* src_u,
                     const uint8_t* src_v,
                     uint8_t* dst_uv,
                     int width);
+void MergeUVRow_MMI(const uint8_t* src_u,
+                    const uint8_t* src_v,
+                    uint8_t* dst_uv,
+                    int width);
 void MergeUVRow_Any_SSE2(const uint8_t* y_buf,
                          const uint8_t* uv_buf,
                          uint8_t* dst_ptr,
@@ -1367,6 +1597,10 @@ void MergeUVRow_Any_MSA(const uint8_t* y_buf,
                         const uint8_t* uv_buf,
                         uint8_t* dst_ptr,
                         int width);
+void MergeUVRow_Any_MMI(const uint8_t* y_buf,
+                        const uint8_t* uv_buf,
+                        uint8_t* dst_ptr,
+                        int width);
 
 void SplitRGBRow_C(const uint8_t* src_rgb,
                    uint8_t* dst_r,
@@ -1383,6 +1617,11 @@ void SplitRGBRow_NEON(const uint8_t* src_rgb,
                       uint8_t* dst_g,
                       uint8_t* dst_b,
                       int width);
+void SplitRGBRow_MMI(const uint8_t* src_rgb,
+                     uint8_t* dst_r,
+                     uint8_t* dst_g,
+                     uint8_t* dst_b,
+                     int width);
 void SplitRGBRow_Any_SSSE3(const uint8_t* src_ptr,
                            uint8_t* dst_r,
                            uint8_t* dst_g,
@@ -1393,6 +1632,11 @@ void SplitRGBRow_Any_NEON(const uint8_t* src_ptr,
                           uint8_t* dst_g,
                           uint8_t* dst_b,
                           int width);
+void SplitRGBRow_Any_MMI(const uint8_t* src_ptr,
+                         uint8_t* dst_r,
+                         uint8_t* dst_g,
+                         uint8_t* dst_b,
+                         int width);
 
 void MergeRGBRow_C(const uint8_t* src_r,
                    const uint8_t* src_g,
@@ -1409,6 +1653,11 @@ void MergeRGBRow_NEON(const uint8_t* src_r,
                       const uint8_t* src_b,
                       uint8_t* dst_rgb,
                       int width);
+void MergeRGBRow_MMI(const uint8_t* src_r,
+                     const uint8_t* src_g,
+                     const uint8_t* src_b,
+                     uint8_t* dst_rgb,
+                     int width);
 void MergeRGBRow_Any_SSSE3(const uint8_t* y_buf,
                            const uint8_t* u_buf,
                            const uint8_t* v_buf,
@@ -1419,6 +1668,11 @@ void MergeRGBRow_Any_NEON(const uint8_t* src_r,
                           const uint8_t* src_b,
                           uint8_t* dst_rgb,
                           int width);
+void MergeRGBRow_Any_MMI(const uint8_t* src_r,
+                         const uint8_t* src_g,
+                         const uint8_t* src_b,
+                         uint8_t* dst_rgb,
+                         int width);
 
 void MergeUVRow_16_C(const uint16_t* src_u,
                      const uint16_t* src_v,
@@ -1497,12 +1751,16 @@ void CopyRow_16_C(const uint16_t* src, uint16_t* dst, int count);
 void ARGBCopyAlphaRow_C(const uint8_t* src, uint8_t* dst, int width);
 void ARGBCopyAlphaRow_SSE2(const uint8_t* src, uint8_t* dst, int width);
 void ARGBCopyAlphaRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
+void ARGBCopyAlphaRow_MMI(const uint8_t* src, uint8_t* dst, int width);
 void ARGBCopyAlphaRow_Any_SSE2(const uint8_t* src_ptr,
                                uint8_t* dst_ptr,
                                int width);
 void ARGBCopyAlphaRow_Any_AVX2(const uint8_t* src_ptr,
                                uint8_t* dst_ptr,
                                int width);
+void ARGBCopyAlphaRow_Any_MMI(const uint8_t* src_ptr,
+                              uint8_t* dst_ptr,
+                              int width);
 
 void ARGBExtractAlphaRow_C(const uint8_t* src_argb, uint8_t* dst_a, int width);
 void ARGBExtractAlphaRow_SSE2(const uint8_t* src_argb,
@@ -1517,6 +1775,9 @@ void ARGBExtractAlphaRow_NEON(const uint8_t* src_argb,
 void ARGBExtractAlphaRow_MSA(const uint8_t* src_argb,
                              uint8_t* dst_a,
                              int width);
+void ARGBExtractAlphaRow_MMI(const uint8_t* src_argb,
+                             uint8_t* dst_a,
+                             int width);
 void ARGBExtractAlphaRow_Any_SSE2(const uint8_t* src_ptr,
                                   uint8_t* dst_ptr,
                                   int width);
@@ -1529,16 +1790,23 @@ void ARGBExtractAlphaRow_Any_NEON(const uint8_t* src_ptr,
 void ARGBExtractAlphaRow_Any_MSA(const uint8_t* src_ptr,
                                  uint8_t* dst_ptr,
                                  int width);
+void ARGBExtractAlphaRow_Any_MMI(const uint8_t* src_ptr,
+                                 uint8_t* dst_ptr,
+                                 int width);
 
 void ARGBCopyYToAlphaRow_C(const uint8_t* src, uint8_t* dst, int width);
 void ARGBCopyYToAlphaRow_SSE2(const uint8_t* src, uint8_t* dst, int width);
 void ARGBCopyYToAlphaRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
+void ARGBCopyYToAlphaRow_MMI(const uint8_t* src, uint8_t* dst, int width);
 void ARGBCopyYToAlphaRow_Any_SSE2(const uint8_t* src_ptr,
                                   uint8_t* dst_ptr,
                                   int width);
 void ARGBCopyYToAlphaRow_Any_AVX2(const uint8_t* src_ptr,
                                   uint8_t* dst_ptr,
                                   int width);
+void ARGBCopyYToAlphaRow_Any_MMI(const uint8_t* src_ptr,
+                                  uint8_t* dst_ptr,
+                                  int width);
 
 void SetRow_C(uint8_t* dst, uint8_t v8, int width);
 void SetRow_MSA(uint8_t* dst, uint8_t v8, int width);
@@ -1576,6 +1844,10 @@ void ARGBShuffleRow_MSA(const uint8_t* src_argb,
                         uint8_t* dst_argb,
                         const uint8_t* shuffler,
                         int width);
+void ARGBShuffleRow_MMI(const uint8_t* src_argb,
+                        uint8_t* dst_argb,
+                        const uint8_t* shuffler,
+                        int width);
 void ARGBShuffleRow_Any_SSSE3(const uint8_t* src_ptr,
                               uint8_t* dst_ptr,
                               const uint8_t* param,
@@ -1592,6 +1864,10 @@ void ARGBShuffleRow_Any_MSA(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             const uint8_t* param,
                             int width);
+void ARGBShuffleRow_Any_MMI(const uint8_t* src_ptr,
+                            uint8_t* dst_ptr,
+                            const uint8_t* param,
+                            int width);
 
 void RGB24ToARGBRow_SSSE3(const uint8_t* src_rgb24,
                           uint8_t* dst_argb,
@@ -1615,28 +1891,40 @@ void RGB24ToARGBRow_NEON(const uint8_t* src_rgb24,
                          uint8_t* dst_argb,
                          int width);
 void RGB24ToARGBRow_MSA(const uint8_t* src_rgb24, uint8_t* dst_argb, int width);
+void RGB24ToARGBRow_MMI(const uint8_t* src_rgb24, uint8_t* dst_argb, int width);
 void RAWToARGBRow_NEON(const uint8_t* src_raw, uint8_t* dst_argb, int width);
 void RAWToARGBRow_MSA(const uint8_t* src_raw, uint8_t* dst_argb, int width);
+void RAWToARGBRow_MMI(const uint8_t* src_raw, uint8_t* dst_argb, int width);
 void RAWToRGB24Row_NEON(const uint8_t* src_raw, uint8_t* dst_rgb24, int width);
 void RAWToRGB24Row_MSA(const uint8_t* src_raw, uint8_t* dst_rgb24, int width);
+void RAWToRGB24Row_MMI(const uint8_t* src_raw, uint8_t* dst_rgb24, int width);
 void RGB565ToARGBRow_NEON(const uint8_t* src_rgb565,
                           uint8_t* dst_argb,
                           int width);
 void RGB565ToARGBRow_MSA(const uint8_t* src_rgb565,
                          uint8_t* dst_argb,
                          int width);
+void RGB565ToARGBRow_MMI(const uint8_t* src_rgb565,
+                         uint8_t* dst_argb,
+                         int width);
 void ARGB1555ToARGBRow_NEON(const uint8_t* src_argb1555,
                             uint8_t* dst_argb,
                             int width);
 void ARGB1555ToARGBRow_MSA(const uint8_t* src_argb1555,
                            uint8_t* dst_argb,
                            int width);
+void ARGB1555ToARGBRow_MMI(const uint8_t* src_argb1555,
+                           uint8_t* dst_argb,
+                           int width);
 void ARGB4444ToARGBRow_NEON(const uint8_t* src_argb4444,
                             uint8_t* dst_argb,
                             int width);
 void ARGB4444ToARGBRow_MSA(const uint8_t* src_argb4444,
                            uint8_t* dst_argb,
                            int width);
+void ARGB4444ToARGBRow_MMI(const uint8_t* src_argb4444,
+                           uint8_t* dst_argb,
+                           int width);
 void RGB24ToARGBRow_C(const uint8_t* src_rgb24, uint8_t* dst_argb, int width);
 void RAWToARGBRow_C(const uint8_t* src_raw, uint8_t* dst_argb, int width);
 void RAWToRGB24Row_C(const uint8_t* src_raw, uint8_t* dst_rgb24, int width);
@@ -1687,24 +1975,35 @@ void RGB24ToARGBRow_Any_NEON(const uint8_t* src_ptr,
 void RGB24ToARGBRow_Any_MSA(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             int width);
+void RGB24ToARGBRow_Any_MMI(const uint8_t* src_ptr,
+                            uint8_t* dst_ptr,
+                            int width);
 void RAWToARGBRow_Any_NEON(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RAWToARGBRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RAWToARGBRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RAWToRGB24Row_Any_NEON(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             int width);
 void RAWToRGB24Row_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void RAWToRGB24Row_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void RGB565ToARGBRow_Any_NEON(const uint8_t* src_ptr,
                               uint8_t* dst_ptr,
                               int width);
 void RGB565ToARGBRow_Any_MSA(const uint8_t* src_ptr,
                              uint8_t* dst_ptr,
                              int width);
+void RGB565ToARGBRow_Any_MMI(const uint8_t* src_ptr,
+                             uint8_t* dst_ptr,
+                             int width);
 void ARGB1555ToARGBRow_Any_NEON(const uint8_t* src_ptr,
                                 uint8_t* dst_ptr,
                                 int width);
 void ARGB1555ToARGBRow_Any_MSA(const uint8_t* src_ptr,
                                uint8_t* dst_ptr,
                                int width);
+void ARGB1555ToARGBRow_Any_MMI(const uint8_t* src_ptr,
+                               uint8_t* dst_ptr,
+                               int width);
 void ARGB4444ToARGBRow_Any_NEON(const uint8_t* src_ptr,
                                 uint8_t* dst_ptr,
                                 int width);
@@ -1712,6 +2011,9 @@ void ARGB4444ToARGBRow_Any_NEON(const uint8_t* src_ptr,
 void ARGB4444ToARGBRow_Any_MSA(const uint8_t* src_ptr,
                                uint8_t* dst_ptr,
                                int width);
+void ARGB4444ToARGBRow_Any_MMI(const uint8_t* src_ptr,
+                               uint8_t* dst_ptr,
+                               int width);
 
 void ARGBToRGB24Row_SSSE3(const uint8_t* src, uint8_t* dst, int width);
 void ARGBToRAWRow_SSSE3(const uint8_t* src, uint8_t* dst, int width);
@@ -1780,6 +2082,20 @@ void ARGBToRGB565DitherRow_MSA(const uint8_t* src_argb,
                                const uint32_t dither4,
                                int width);
 
+void ARGBToRGB24Row_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
+void ARGBToRAWRow_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
+void ARGBToRGB565Row_MMI(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
+void ARGBToARGB1555Row_MMI(const uint8_t* src_argb,
+                           uint8_t* dst_rgb,
+                           int width);
+void ARGBToARGB4444Row_MMI(const uint8_t* src_argb,
+                           uint8_t* dst_rgb,
+                           int width);
+void ARGBToRGB565DitherRow_MMI(const uint8_t* src_argb,
+                               uint8_t* dst_rgb,
+                               const uint32_t dither4,
+                               int width);
+
 void ARGBToRGBARow_C(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
 void ARGBToRGB24Row_C(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
 void ARGBToRAWRow_C(const uint8_t* src_argb, uint8_t* dst_rgb, int width);
@@ -1793,6 +2109,7 @@ void J400ToARGBRow_SSE2(const uint8_t* src_y, uint8_t* dst_argb, int width);
 void J400ToARGBRow_AVX2(const uint8_t* src_y, uint8_t* dst_argb, int width);
 void J400ToARGBRow_NEON(const uint8_t* src_y, uint8_t* dst_argb, int width);
 void J400ToARGBRow_MSA(const uint8_t* src_y, uint8_t* dst_argb, int width);
+void J400ToARGBRow_MMI(const uint8_t* src_y, uint8_t* dst_argb, int width);
 void J400ToARGBRow_C(const uint8_t* src_y, uint8_t* dst_argb, int width);
 void J400ToARGBRow_Any_SSE2(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
@@ -1804,6 +2121,7 @@ void J400ToARGBRow_Any_NEON(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             int width);
 void J400ToARGBRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void J400ToARGBRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 
 void I444ToARGBRow_C(const uint8_t* src_y,
                      const uint8_t* src_u,
@@ -2324,6 +2642,7 @@ void I400ToARGBRow_SSE2(const uint8_t* y_buf, uint8_t* dst_argb, int width);
 void I400ToARGBRow_AVX2(const uint8_t* y_buf, uint8_t* dst_argb, int width);
 void I400ToARGBRow_NEON(const uint8_t* src_y, uint8_t* dst_argb, int width);
 void I400ToARGBRow_MSA(const uint8_t* src_y, uint8_t* dst_argb, int width);
+void I400ToARGBRow_MMI(const uint8_t* src_y, uint8_t* dst_argb, int width);
 void I400ToARGBRow_Any_SSE2(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             int width);
@@ -2334,6 +2653,7 @@ void I400ToARGBRow_Any_NEON(const uint8_t* src_ptr,
                             uint8_t* dst_ptr,
                             int width);
 void I400ToARGBRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void I400ToARGBRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 
 // ARGB preattenuated alpha blend.
 void ARGBBlendRow_SSSE3(const uint8_t* src_argb0,
@@ -2348,6 +2668,10 @@ void ARGBBlendRow_MSA(const uint8_t* src_argb0,
                       const uint8_t* src_argb1,
                       uint8_t* dst_argb,
                       int width);
+void ARGBBlendRow_MMI(const uint8_t* src_argb0,
+                      const uint8_t* src_argb1,
+                      uint8_t* dst_argb,
+                      int width);
 void ARGBBlendRow_C(const uint8_t* src_argb0,
                     const uint8_t* src_argb1,
                     uint8_t* dst_argb,
@@ -2374,6 +2698,16 @@ void BlendPlaneRow_Any_AVX2(const uint8_t* y_buf,
                             const uint8_t* v_buf,
                             uint8_t* dst_ptr,
                             int width);
+void BlendPlaneRow_MMI(const uint8_t* src0,
+                       const uint8_t* src1,
+                       const uint8_t* alpha,
+                       uint8_t* dst,
+                       int width);
+void BlendPlaneRow_Any_MMI(const uint8_t* y_buf,
+                           const uint8_t* u_buf,
+                           const uint8_t* v_buf,
+                           uint8_t* dst_ptr,
+                           int width);
 void BlendPlaneRow_C(const uint8_t* src0,
                      const uint8_t* src1,
                      const uint8_t* alpha,
@@ -2418,6 +2752,14 @@ void ARGBMultiplyRow_Any_MSA(const uint8_t* y_buf,
                              const uint8_t* uv_buf,
                              uint8_t* dst_ptr,
                              int width);
+void ARGBMultiplyRow_MMI(const uint8_t* src_argb0,
+                         const uint8_t* src_argb1,
+                         uint8_t* dst_argb,
+                         int width);
+void ARGBMultiplyRow_Any_MMI(const uint8_t* y_buf,
+                             const uint8_t* uv_buf,
+                             uint8_t* dst_ptr,
+                             int width);
 
 // ARGB add images.
 void ARGBAddRow_C(const uint8_t* src_argb0,
@@ -2456,6 +2798,14 @@ void ARGBAddRow_Any_MSA(const uint8_t* y_buf,
                         const uint8_t* uv_buf,
                         uint8_t* dst_ptr,
                         int width);
+void ARGBAddRow_MMI(const uint8_t* src_argb0,
+                    const uint8_t* src_argb1,
+                    uint8_t* dst_argb,
+                    int width);
+void ARGBAddRow_Any_MMI(const uint8_t* y_buf,
+                        const uint8_t* uv_buf,
+                        uint8_t* dst_ptr,
+                        int width);
 
 // ARGB subtract images. Same API as Blend, but these require
 // pointer and width alignment for SSE2.
@@ -2495,6 +2845,14 @@ void ARGBSubtractRow_Any_MSA(const uint8_t* y_buf,
                              const uint8_t* uv_buf,
                              uint8_t* dst_ptr,
                              int width);
+void ARGBSubtractRow_MMI(const uint8_t* src_argb0,
+                         const uint8_t* src_argb1,
+                         uint8_t* dst_argb,
+                         int width);
+void ARGBSubtractRow_Any_MMI(const uint8_t* y_buf,
+                             const uint8_t* uv_buf,
+                             uint8_t* dst_ptr,
+                             int width);
 
 void ARGBToRGB24Row_Any_SSSE3(const uint8_t* src_ptr,
                               uint8_t* dst_ptr,
@@ -2584,6 +2942,24 @@ void ARGBToRGB565DitherRow_Any_MSA(const uint8_t* src_ptr,
                                    const uint32_t param,
                                    int width);
 
+void ARGBToRGB24Row_Any_MMI(const uint8_t* src_ptr,
+                            uint8_t* dst_ptr,
+                            int width);
+void ARGBToRAWRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void ARGBToRGB565Row_Any_MMI(const uint8_t* src_ptr,
+                             uint8_t* dst_ptr,
+                             int width);
+void ARGBToARGB1555Row_Any_MMI(const uint8_t* src_ptr,
+                               uint8_t* dst_ptr,
+                               int width);
+void ARGBToARGB4444Row_Any_MMI(const uint8_t* src_ptr,
+                               uint8_t* dst_ptr,
+                               int width);
+void ARGBToRGB565DitherRow_Any_MMI(const uint8_t* src_ptr,
+                                   uint8_t* dst_ptr,
+                                   const uint32_t param,
+                                   int width);
+
 void I444ToARGBRow_Any_NEON(const uint8_t* y_buf,
                             const uint8_t* u_buf,
                             const uint8_t* v_buf,
@@ -2770,15 +3146,25 @@ void YUY2ToUV422Row_NEON(const uint8_t* src_yuy2,
                          uint8_t* dst_v,
                          int width);
 void YUY2ToYRow_MSA(const uint8_t* src_yuy2, uint8_t* dst_y, int width);
+void YUY2ToYRow_MMI(const uint8_t* src_yuy2, uint8_t* dst_y, int width);
 void YUY2ToUVRow_MSA(const uint8_t* src_yuy2,
                      int src_stride_yuy2,
                      uint8_t* dst_u,
                      uint8_t* dst_v,
                      int width);
+void YUY2ToUVRow_MMI(const uint8_t* src_yuy2,
+                     int src_stride_yuy2,
+                     uint8_t* dst_u,
+                     uint8_t* dst_v,
+                     int width);
 void YUY2ToUV422Row_MSA(const uint8_t* src_yuy2,
                         uint8_t* dst_u,
                         uint8_t* dst_v,
                         int width);
+void YUY2ToUV422Row_MMI(const uint8_t* src_yuy2,
+                        uint8_t* dst_u,
+                        uint8_t* dst_v,
+                        int width);
 void YUY2ToYRow_C(const uint8_t* src_yuy2, uint8_t* dst_y, int width);
 void YUY2ToUVRow_C(const uint8_t* src_yuy2,
                    int src_stride_yuy2,
@@ -2820,15 +3206,25 @@ void YUY2ToUV422Row_Any_NEON(const uint8_t* src_ptr,
                              uint8_t* dst_v,
                              int width);
 void YUY2ToYRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void YUY2ToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void YUY2ToUVRow_Any_MSA(const uint8_t* src_ptr,
                          int src_stride_ptr,
                          uint8_t* dst_u,
                          uint8_t* dst_v,
                          int width);
+void YUY2ToUVRow_Any_MMI(const uint8_t* src_ptr,
+                         int src_stride_ptr,
+                         uint8_t* dst_u,
+                         uint8_t* dst_v,
+                         int width);
 void YUY2ToUV422Row_Any_MSA(const uint8_t* src_ptr,
                             uint8_t* dst_u,
                             uint8_t* dst_v,
                             int width);
+void YUY2ToUV422Row_Any_MMI(const uint8_t* src_ptr,
+                            uint8_t* dst_u,
+                            uint8_t* dst_v,
+                            int width);
 void UYVYToYRow_AVX2(const uint8_t* src_uyvy, uint8_t* dst_y, int width);
 void UYVYToUVRow_AVX2(const uint8_t* src_uyvy,
                       int stride_uyvy,
@@ -2870,15 +3266,25 @@ void UYVYToUV422Row_NEON(const uint8_t* src_uyvy,
                          uint8_t* dst_v,
                          int width);
 void UYVYToYRow_MSA(const uint8_t* src_uyvy, uint8_t* dst_y, int width);
+void UYVYToYRow_MMI(const uint8_t* src_uyvy, uint8_t* dst_y, int width);
 void UYVYToUVRow_MSA(const uint8_t* src_uyvy,
                      int src_stride_uyvy,
                      uint8_t* dst_u,
                      uint8_t* dst_v,
                      int width);
+void UYVYToUVRow_MMI(const uint8_t* src_uyvy,
+                     int src_stride_uyvy,
+                     uint8_t* dst_u,
+                     uint8_t* dst_v,
+                     int width);
 void UYVYToUV422Row_MSA(const uint8_t* src_uyvy,
                         uint8_t* dst_u,
                         uint8_t* dst_v,
                         int width);
+void UYVYToUV422Row_MMI(const uint8_t* src_uyvy,
+                        uint8_t* dst_u,
+                        uint8_t* dst_v,
+                        int width);
 
 void UYVYToYRow_C(const uint8_t* src_uyvy, uint8_t* dst_y, int width);
 void UYVYToUVRow_C(const uint8_t* src_uyvy,
@@ -2921,15 +3327,25 @@ void UYVYToUV422Row_Any_NEON(const uint8_t* src_ptr,
                              uint8_t* dst_v,
                              int width);
 void UYVYToYRow_Any_MSA(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
+void UYVYToYRow_Any_MMI(const uint8_t* src_ptr, uint8_t* dst_ptr, int width);
 void UYVYToUVRow_Any_MSA(const uint8_t* src_ptr,
                          int src_stride_ptr,
                          uint8_t* dst_u,
                          uint8_t* dst_v,
                          int width);
+void UYVYToUVRow_Any_MMI(const uint8_t* src_ptr,
+                         int src_stride_ptr,
+                         uint8_t* dst_u,
+                         uint8_t* dst_v,
+                         int width);
 void UYVYToUV422Row_Any_MSA(const uint8_t* src_ptr,
                             uint8_t* dst_u,
                             uint8_t* dst_v,
                             int width);
+void UYVYToUV422Row_Any_MMI(const uint8_t* src_ptr,
+                            uint8_t* dst_u,
+                            uint8_t* dst_v,
+                            int width);
 
 void I422ToYUY2Row_C(const uint8_t* src_y,
                      const uint8_t* src_u,
@@ -3006,21 +3422,41 @@ void I422ToYUY2Row_MSA(const uint8_t* src_y,
                        const uint8_t* src_v,
                        uint8_t* dst_yuy2,
                        int width);
+void I422ToYUY2Row_MMI(const uint8_t* src_y,
+                       const uint8_t* src_u,
+                       const uint8_t* src_v,
+                       uint8_t* dst_yuy2,
+                       int width);
 void I422ToUYVYRow_MSA(const uint8_t* src_y,
                        const uint8_t* src_u,
                        const uint8_t* src_v,
                        uint8_t* dst_uyvy,
                        int width);
+void I422ToUYVYRow_MMI(const uint8_t* src_y,
+                       const uint8_t* src_u,
+                       const uint8_t* src_v,
+                       uint8_t* dst_uyvy,
+                       int width);
 void I422ToYUY2Row_Any_MSA(const uint8_t* y_buf,
                            const uint8_t* u_buf,
                            const uint8_t* v_buf,
                            uint8_t* dst_ptr,
                            int width);
+void I422ToYUY2Row_Any_MMI(const uint8_t* y_buf,
+                           const uint8_t* u_buf,
+                           const uint8_t* v_buf,
+                           uint8_t* dst_ptr,
+                           int width);
 void I422ToUYVYRow_Any_MSA(const uint8_t* y_buf,
                            const uint8_t* u_buf,
                            const uint8_t* v_buf,
                            uint8_t* dst_ptr,
                            int width);
+void I422ToUYVYRow_Any_MMI(const uint8_t* y_buf,
+                           const uint8_t* u_buf,
+                           const uint8_t* v_buf,
+                           uint8_t* dst_ptr,
+                           int width);
 
 // Effects related row functions.
 void ARGBAttenuateRow_C(const uint8_t* src_argb, uint8_t* dst_argb, int width);
@@ -3036,6 +3472,9 @@ void ARGBAttenuateRow_NEON(const uint8_t* src_argb,
 void ARGBAttenuateRow_MSA(const uint8_t* src_argb,
                           uint8_t* dst_argb,
                           int width);
+void ARGBAttenuateRow_MMI(const uint8_t* src_argb,
+                          uint8_t* dst_argb,
+                          int width);
 void ARGBAttenuateRow_Any_SSSE3(const uint8_t* src_ptr,
                                 uint8_t* dst_ptr,
                                 int width);
@@ -3048,6 +3487,9 @@ void ARGBAttenuateRow_Any_NEON(const uint8_t* src_ptr,
 void ARGBAttenuateRow_Any_MSA(const uint8_t* src_ptr,
                               uint8_t* dst_ptr,
                               int width);
+void ARGBAttenuateRow_Any_MMI(const uint8_t* src_ptr,
+                              uint8_t* dst_ptr,
+                              int width);
 
 // Inverse table for unattenuate, shared by C and SSE2.
 extern const uint32_t fixed_invtbl8[256];
@@ -3071,11 +3513,13 @@ void ARGBGrayRow_C(const uint8_t* src_argb, uint8_t* dst_argb, int width);
 void ARGBGrayRow_SSSE3(const uint8_t* src_argb, uint8_t* dst_argb, int width);
 void ARGBGrayRow_NEON(const uint8_t* src_argb, uint8_t* dst_argb, int width);
 void ARGBGrayRow_MSA(const uint8_t* src_argb, uint8_t* dst_argb, int width);
+void ARGBGrayRow_MMI(const uint8_t* src_argb, uint8_t* dst_argb, int width);
 
 void ARGBSepiaRow_C(uint8_t* dst_argb, int width);
 void ARGBSepiaRow_SSSE3(uint8_t* dst_argb, int width);
 void ARGBSepiaRow_NEON(uint8_t* dst_argb, int width);
 void ARGBSepiaRow_MSA(uint8_t* dst_argb, int width);
+void ARGBSepiaRow_MMI(uint8_t* dst_argb, int width);
 
 void ARGBColorMatrixRow_C(const uint8_t* src_argb,
                           uint8_t* dst_argb,
@@ -3093,6 +3537,10 @@ void ARGBColorMatrixRow_MSA(const uint8_t* src_argb,
                             uint8_t* dst_argb,
                             const int8_t* matrix_argb,
                             int width);
+void ARGBColorMatrixRow_MMI(const uint8_t* src_argb,
+                            uint8_t* dst_argb,
+                            const int8_t* matrix_argb,
+                            int width);
 
 void ARGBColorTableRow_C(uint8_t* dst_argb,
                          const uint8_t* table_argb,
@@ -3145,6 +3593,10 @@ void ARGBShadeRow_MSA(const uint8_t* src_argb,
                       uint8_t* dst_argb,
                       int width,
                       uint32_t value);
+void ARGBShadeRow_MMI(const uint8_t* src_argb,
+                      uint8_t* dst_argb,
+                      int width,
+                      uint32_t value);
 
 // Used for blur.
 void CumulativeSumToAverageRow_SSE2(const int32_t* topleft,
@@ -3158,6 +3610,11 @@ void ComputeCumulativeSumRow_SSE2(const uint8_t* row,
                                   const int32_t* previous_cumsum,
                                   int width);
 
+void ComputeCumulativeSumRow_MMI(const uint8_t* row,
+                                 int32_t* cumsum,
+                                 const int32_t* previous_cumsum,
+                                 int width);
+
 void CumulativeSumToAverageRow_C(const int32_t* tl,
                                  const int32_t* bl,
                                  int w,
@@ -3208,6 +3665,11 @@ void InterpolateRow_MSA(uint8_t* dst_ptr,
                         ptrdiff_t src_stride,
                         int width,
                         int source_y_fraction);
+void InterpolateRow_MMI(uint8_t* dst_ptr,
+                        const uint8_t* src_ptr,
+                        ptrdiff_t src_stride,
+                        int width,
+                        int source_y_fraction);
 void InterpolateRow_Any_NEON(uint8_t* dst_ptr,
                              const uint8_t* src_ptr,
                              ptrdiff_t src_stride_ptr,
@@ -3228,6 +3690,11 @@ void InterpolateRow_Any_MSA(uint8_t* dst_ptr,
                             ptrdiff_t src_stride_ptr,
                             int width,
                             int source_y_fraction);
+void InterpolateRow_Any_MMI(uint8_t* dst_ptr,
+                            const uint8_t* src_ptr,
+                            ptrdiff_t src_stride_ptr,
+                            int width,
+                            int source_y_fraction);
 
 void InterpolateRow_16_C(uint16_t* dst_ptr,
                          const uint16_t* src_ptr,
@@ -3256,6 +3723,11 @@ void SobelXRow_MSA(const uint8_t* src_y0,
                    const uint8_t* src_y2,
                    uint8_t* dst_sobelx,
                    int width);
+void SobelXRow_MMI(const uint8_t* src_y0,
+                   const uint8_t* src_y1,
+                   const uint8_t* src_y2,
+                   uint8_t* dst_sobelx,
+                   int width);
 void SobelYRow_C(const uint8_t* src_y0,
                  const uint8_t* src_y1,
                  uint8_t* dst_sobely,
@@ -3272,6 +3744,10 @@ void SobelYRow_MSA(const uint8_t* src_y0,
                    const uint8_t* src_y1,
                    uint8_t* dst_sobely,
                    int width);
+void SobelYRow_MMI(const uint8_t* src_y0,
+                   const uint8_t* src_y1,
+                   uint8_t* dst_sobely,
+                   int width);
 void SobelRow_C(const uint8_t* src_sobelx,
                 const uint8_t* src_sobely,
                 uint8_t* dst_argb,
@@ -3288,6 +3764,10 @@ void SobelRow_MSA(const uint8_t* src_sobelx,
                   const uint8_t* src_sobely,
                   uint8_t* dst_argb,
                   int width);
+void SobelRow_MMI(const uint8_t* src_sobelx,
+                  const uint8_t* src_sobely,
+                  uint8_t* dst_argb,
+                  int width);
 void SobelToPlaneRow_C(const uint8_t* src_sobelx,
                        const uint8_t* src_sobely,
                        uint8_t* dst_y,
@@ -3304,6 +3784,10 @@ void SobelToPlaneRow_MSA(const uint8_t* src_sobelx,
                          const uint8_t* src_sobely,
                          uint8_t* dst_y,
                          int width);
+void SobelToPlaneRow_MMI(const uint8_t* src_sobelx,
+                         const uint8_t* src_sobely,
+                         uint8_t* dst_y,
+                         int width);
 void SobelXYRow_C(const uint8_t* src_sobelx,
                   const uint8_t* src_sobely,
                   uint8_t* dst_argb,
@@ -3320,6 +3804,10 @@ void SobelXYRow_MSA(const uint8_t* src_sobelx,
                     const uint8_t* src_sobely,
                     uint8_t* dst_argb,
                     int width);
+void SobelXYRow_MMI(const uint8_t* src_sobelx,
+                    const uint8_t* src_sobely,
+                    uint8_t* dst_argb,
+                    int width);
 void SobelRow_Any_SSE2(const uint8_t* y_buf,
                        const uint8_t* uv_buf,
                        uint8_t* dst_ptr,
@@ -3332,6 +3820,10 @@ void SobelRow_Any_MSA(const uint8_t* y_buf,
                       const uint8_t* uv_buf,
                       uint8_t* dst_ptr,
                       int width);
+void SobelRow_Any_MMI(const uint8_t* y_buf,
+                      const uint8_t* uv_buf,
+                      uint8_t* dst_ptr,
+                      int width);
 void SobelToPlaneRow_Any_SSE2(const uint8_t* y_buf,
                               const uint8_t* uv_buf,
                               uint8_t* dst_ptr,
@@ -3344,6 +3836,10 @@ void SobelToPlaneRow_Any_MSA(const uint8_t* y_buf,
                              const uint8_t* uv_buf,
                              uint8_t* dst_ptr,
                              int width);
+void SobelToPlaneRow_Any_MMI(const uint8_t* y_buf,
+                             const uint8_t* uv_buf,
+                             uint8_t* dst_ptr,
+                             int width);
 void SobelXYRow_Any_SSE2(const uint8_t* y_buf,
                          const uint8_t* uv_buf,
                          uint8_t* dst_ptr,
@@ -3356,6 +3852,10 @@ void SobelXYRow_Any_MSA(const uint8_t* y_buf,
                         const uint8_t* uv_buf,
                         uint8_t* dst_ptr,
                         int width);
+void SobelXYRow_Any_MMI(const uint8_t* y_buf,
+                        const uint8_t* uv_buf,
+                        uint8_t* dst_ptr,
+                        int width);
 
 void ARGBPolynomialRow_C(const uint8_t* src_argb,
                          uint8_t* dst_argb,
diff --git a/include/libyuv/scale_row.h b/include/libyuv/scale_row.h
index 7194ba09..3042136d 100644..100755
--- a/include/libyuv/scale_row.h
+++ b/include/libyuv/scale_row.h
@@ -110,6 +110,24 @@ extern "C" {
 #define HAS_SCALEROWDOWN4_MSA
 #endif
 
+#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
+#define HAS_FIXEDDIV1_MIPS
+#define HAS_FIXEDDIV_MIPS
+#define HAS_SCALEARGBCOLS_MMI
+#define HAS_SCALEARGBCOLSUP2_MMI
+#define HAS_SCALEARGBFILTERCOLS_MMI
+#define HAS_SCALEARGBROWDOWN2_MMI
+#define HAS_SCALEARGBROWDOWNEVEN_MMI
+#define HAS_SCALEROWDOWN2_MMI
+#define HAS_SCALEROWDOWN4_MMI
+#define HAS_SCALEADDROW_MMI
+#define HAS_SCALEADDROW_16_MMI
+#define HAS_SCALEROWDOWN2_16_MMI
+#define HAS_SCALEROWDOWN4_16_MMI
+#define HAS_SCALECOLS_MMI
+#define HAS_SCALECOLS_16_MMI
+#endif
+
 // Scale ARGB vertically with bilinear interpolation.
 void ScalePlaneVertical(int src_height,
                         int dst_width,
@@ -147,12 +165,17 @@ enum FilterMode ScaleFilterReduce(int src_width,
 // Divide num by div and return as 16.16 fixed point result.
 int FixedDiv_C(int num, int div);
 int FixedDiv_X86(int num, int div);
+int FixedDiv_MIPS(int num, int div);
 // Divide num - 1 by div - 1 and return as 16.16 fixed point result.
 int FixedDiv1_C(int num, int div);
 int FixedDiv1_X86(int num, int div);
+int FixedDiv1_MIPS(int num, int div);
 #ifdef HAS_FIXEDDIV_X86
 #define FixedDiv FixedDiv_X86
 #define FixedDiv1 FixedDiv1_X86
+#elif defined HAS_FIXEDDIV_MIPS
+#define FixedDiv FixedDiv_MIPS
+#define FixedDiv1 FixedDiv1_MIPS
 #else
 #define FixedDiv FixedDiv_C
 #define FixedDiv1 FixedDiv1_C
@@ -569,6 +592,26 @@ void ScaleARGBCols_Any_MSA(uint8_t* dst_ptr,
                            int dst_width,
                            int x,
                            int dx);
+void ScaleARGBFilterCols_MMI(uint8_t* dst_argb,
+                             const uint8_t* src_argb,
+                             int dst_width,
+                             int x,
+                             int dx);
+void ScaleARGBCols_MMI(uint8_t* dst_argb,
+                       const uint8_t* src_argb,
+                       int dst_width,
+                       int x,
+                       int dx);
+void ScaleARGBFilterCols_Any_MMI(uint8_t* dst_ptr,
+                                 const uint8_t* src_ptr,
+                                 int dst_width,
+                                 int x,
+                                 int dx);
+void ScaleARGBCols_Any_MMI(uint8_t* dst_ptr,
+                           const uint8_t* src_ptr,
+                           int dst_width,
+                           int x,
+                           int dx);
 
 // ARGB Row functions
 void ScaleARGBRowDown2_SSE2(const uint8_t* src_argb,
@@ -607,6 +650,18 @@ void ScaleARGBRowDown2Box_MSA(const uint8_t* src_argb,
                               ptrdiff_t src_stride,
                               uint8_t* dst_argb,
                               int dst_width);
+void ScaleARGBRowDown2_MMI(const uint8_t* src_argb,
+                           ptrdiff_t src_stride,
+                           uint8_t* dst_argb,
+                           int dst_width);
+void ScaleARGBRowDown2Linear_MMI(const uint8_t* src_argb,
+                                 ptrdiff_t src_stride,
+                                 uint8_t* dst_argb,
+                                 int dst_width);
+void ScaleARGBRowDown2Box_MMI(const uint8_t* src_argb,
+                              ptrdiff_t src_stride,
+                              uint8_t* dst_argb,
+                              int dst_width);
 void ScaleARGBRowDown2_Any_SSE2(const uint8_t* src_ptr,
                                 ptrdiff_t src_stride,
                                 uint8_t* dst_ptr,
@@ -643,7 +698,18 @@ void ScaleARGBRowDown2Box_Any_MSA(const uint8_t* src_ptr,
                                   ptrdiff_t src_stride,
                                   uint8_t* dst_ptr,
                                   int dst_width);
-
+void ScaleARGBRowDown2_Any_MMI(const uint8_t* src_ptr,
+                               ptrdiff_t src_stride,
+                               uint8_t* dst_ptr,
+                               int dst_width);
+void ScaleARGBRowDown2Linear_Any_MMI(const uint8_t* src_ptr,
+                                     ptrdiff_t src_stride,
+                                     uint8_t* dst_ptr,
+                                     int dst_width);
+void ScaleARGBRowDown2Box_Any_MMI(const uint8_t* src_ptr,
+                                  ptrdiff_t src_stride,
+                                  uint8_t* dst_ptr,
+                                  int dst_width);
 void ScaleARGBRowDownEven_SSE2(const uint8_t* src_argb,
                                ptrdiff_t src_stride,
                                int src_stepx,
@@ -674,6 +740,16 @@ void ScaleARGBRowDownEvenBox_MSA(const uint8_t* src_argb,
                                  int src_stepx,
                                  uint8_t* dst_argb,
                                  int dst_width);
+void ScaleARGBRowDownEven_MMI(const uint8_t* src_argb,
+                              ptrdiff_t src_stride,
+                              int32_t src_stepx,
+                              uint8_t* dst_argb,
+                              int dst_width);
+void ScaleARGBRowDownEvenBox_MMI(const uint8_t* src_argb,
+                                 ptrdiff_t src_stride,
+                                 int src_stepx,
+                                 uint8_t* dst_argb,
+                                 int dst_width);
 void ScaleARGBRowDownEven_Any_SSE2(const uint8_t* src_ptr,
                                    ptrdiff_t src_stride,
                                    int src_stepx,
@@ -704,6 +780,16 @@ void ScaleARGBRowDownEvenBox_Any_MSA(const uint8_t* src_ptr,
                                      int src_stepx,
                                      uint8_t* dst_ptr,
                                      int dst_width);
+void ScaleARGBRowDownEven_Any_MMI(const uint8_t* src_ptr,
+                                  ptrdiff_t src_stride,
+                                  int32_t src_stepx,
+                                  uint8_t* dst_ptr,
+                                  int dst_width);
+void ScaleARGBRowDownEvenBox_Any_MMI(const uint8_t* src_ptr,
+                                     ptrdiff_t src_stride,
+                                     int src_stepx,
+                                     uint8_t* dst_ptr,
+                                     int dst_width);
 
 // ScaleRowDown2Box also used by planar functions
 // NEON downscalers with interpolation.
@@ -936,6 +1022,93 @@ void ScaleRowDown34_1_Box_Any_MSA(const uint8_t* src_ptr,
                                   uint8_t* dst_ptr,
                                   int dst_width);
 
+void ScaleRowDown2_MMI(const uint8_t* src_ptr,
+                       ptrdiff_t src_stride,
+                       uint8_t* dst,
+                       int dst_width);
+void ScaleRowDown2_16_MMI(const uint16_t* src_ptr,
+                          ptrdiff_t src_stride,
+                          uint16_t* dst,
+                          int dst_width);
+void ScaleRowDown2Linear_MMI(const uint8_t* src_ptr,
+                             ptrdiff_t src_stride,
+                             uint8_t* dst,
+                             int dst_width);
+void ScaleRowDown2Linear_16_MMI(const uint16_t* src_ptr,
+                                ptrdiff_t src_stride,
+                                uint16_t* dst,
+                                int dst_width);
+void ScaleRowDown2Box_MMI(const uint8_t* src_ptr,
+                          ptrdiff_t src_stride,
+                          uint8_t* dst,
+                          int dst_width);
+void ScaleRowDown2Box_16_MMI(const uint16_t* src_ptr,
+                             ptrdiff_t src_stride,
+                             uint16_t* dst,
+                             int dst_width);
+void ScaleRowDown2Box_Odd_MMI(const uint8_t* src_ptr,
+                              ptrdiff_t src_stride,
+                              uint8_t* dst,
+                              int dst_width);
+void ScaleRowDown4_MMI(const uint8_t* src_ptr,
+                       ptrdiff_t src_stride,
+                       uint8_t* dst,
+                       int dst_width);
+void ScaleRowDown4_16_MMI(const uint16_t* src_ptr,
+                          ptrdiff_t src_stride,
+                          uint16_t* dst,
+                          int dst_width);
+void ScaleRowDown4Box_MMI(const uint8_t* src_ptr,
+                          ptrdiff_t src_stride,
+                          uint8_t* dst,
+                          int dst_width);
+void ScaleRowDown4Box_16_MMI(const uint16_t* src_ptr,
+                             ptrdiff_t src_stride,
+                             uint16_t* dst,
+                             int dst_width);
+void ScaleAddRow_MMI(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width);
+void ScaleAddRow_16_MMI(const uint16_t* src_ptr,
+                        uint32_t* dst_ptr,
+                        int src_width);
+void ScaleColsUp2_MMI(uint8_t* dst_ptr,
+                      const uint8_t* src_ptr,
+                      int dst_width,
+                      int x,
+                      int dx);
+void ScaleColsUp2_16_MMI(uint16_t* dst_ptr,
+                         const uint16_t* src_ptr,
+                         int dst_width,
+                         int x,
+                         int dx);
+void ScaleARGBColsUp2_MMI(uint8_t* dst_argb,
+                          const uint8_t* src_argb,
+                          int dst_width,
+                          int x,
+                          int dx);
+
+void ScaleRowDown2_Any_MMI(const uint8_t* src_ptr,
+                           ptrdiff_t src_stride,
+                           uint8_t* dst_ptr,
+                           int dst_width);
+void ScaleRowDown2Linear_Any_MMI(const uint8_t* src_ptr,
+                                 ptrdiff_t src_stride,
+                                 uint8_t* dst_ptr,
+                                 int dst_width);
+void ScaleRowDown2Box_Any_MMI(const uint8_t* src_ptr,
+                              ptrdiff_t src_stride,
+                              uint8_t* dst_ptr,
+                              int dst_width);
+void ScaleRowDown4_Any_MMI(const uint8_t* src_ptr,
+                           ptrdiff_t src_stride,
+                           uint8_t* dst_ptr,
+                           int dst_width);
+void ScaleRowDown4Box_Any_MMI(const uint8_t* src_ptr,
+                              ptrdiff_t src_stride,
+                              uint8_t* dst_ptr,
+                              int dst_width);
+void ScaleAddRow_Any_MMI(const uint8_t* src_ptr,
+                         uint16_t* dst_ptr,
+                         int src_width);
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv