Add MSA optimized NV12/21 To RGB row functions

include/libyuv/row.h

 /*
  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS. All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 410 matching lines @@
 #define HAS_RGB24TOARGBROW_MSA
 #define HAS_RAWTOARGBROW_MSA
 #define HAS_ARGB1555TOYROW_MSA
 #define HAS_RGB565TOYROW_MSA
 #define HAS_RGB24TOYROW_MSA
 #define HAS_RAWTOYROW_MSA
 #define HAS_ARGB1555TOUVROW_MSA
 #define HAS_RGB565TOUVROW_MSA
 #define HAS_RGB24TOUVROW_MSA
 #define HAS_RAWTOUVROW_MSA
+#define HAS_NV12TOARGBROW_MSA
+#define HAS_NV12TORGB565ROW_MSA
+#define HAS_NV21TOARGBROW_MSA
+#define HAS_SOBELROW_MSA
+#define HAS_SOBELTOPLANEROW_MSA
+#define HAS_SOBELXYROW_MSA
 #endif
 
 #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
 #if defined(VISUALC_HAS_AVX2)
 #define SIMD_ALIGNED(var) __declspec(align(32)) var
 #else
 #define SIMD_ALIGNED(var) __declspec(align(16)) var
 #endif
 typedef __declspec(align(16)) int16 vec16[8];
 typedef __declspec(align(16)) int32 vec32[4];
@@ skipping 308 matching lines @@
                            const uint8* src_v,
                            uint8* dst_argb4444,
                            const struct YuvConstants* yuvconstants,
                            int width);
 void I422ToARGB1555Row_MSA(const uint8* src_y,
                            const uint8* src_u,
                            const uint8* src_v,
                            uint8* dst_argb1555,
                            const struct YuvConstants* yuvconstants,
                            int width);
+void NV12ToARGBRow_MSA(const uint8* src_y,
+                       const uint8* src_uv,
+                       uint8* dst_argb,
+                       const struct YuvConstants* yuvconstants,
+                       int width);
+void NV12ToRGB565Row_MSA(const uint8* src_y,
+                         const uint8* src_uv,
+                         uint8* dst_rgb565,
+                         const struct YuvConstants* yuvconstants,
+                         int width);
+void NV21ToARGBRow_MSA(const uint8* src_y,
+                       const uint8* src_vu,
+                       uint8* dst_argb,
+                       const struct YuvConstants* yuvconstants,
+                       int width);
 
 void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYRow_Any_AVX2(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYJRow_AVX2(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYJRow_Any_AVX2(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYJRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width);
 void BGRAToYRow_SSSE3(const uint8* src_bgra, uint8* dst_y, int width);
 void ABGRToYRow_SSSE3(const uint8* src_abgr, uint8* dst_y, int width);
 void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int width);
@@ skipping 1552 matching lines @@
                                const uint8* src_v,
                                uint8* dst_argb4444,
                                const struct YuvConstants* yuvconstants,
                                int width);
 void I422ToARGB1555Row_Any_MSA(const uint8* src_y,
                                const uint8* src_u,
                                const uint8* src_v,
                                uint8* dst_argb1555,
                                const struct YuvConstants* yuvconstants,
                                int width);
+void NV12ToARGBRow_Any_MSA(const uint8* src_y,
+                           const uint8* src_uv,
+                           uint8* dst_argb,
+                           const struct YuvConstants* yuvconstants,
+                           int width);
+void NV12ToRGB565Row_Any_MSA(const uint8* src_y,
+                             const uint8* src_uv,
+                             uint8* dst_argb,
+                             const struct YuvConstants* yuvconstants,
+                             int width);
+void NV21ToARGBRow_Any_MSA(const uint8* src_y,
+                           const uint8* src_vu,
+                           uint8* dst_argb,
+                           const struct YuvConstants* yuvconstants,
+                           int width);
 
 void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width);
 void YUY2ToUVRow_AVX2(const uint8* src_yuy2,
                       int stride_yuy2,
                       uint8* dst_u,
                       uint8* dst_v,
                       int width);
 void YUY2ToUV422Row_AVX2(const uint8* src_yuy2,
                          uint8* dst_u,
                          uint8* dst_v,
@@ skipping 469 matching lines @@
                 uint8* dst_argb,
                 int width);
 void SobelRow_SSE2(const uint8* src_sobelx,
                    const uint8* src_sobely,
                    uint8* dst_argb,
                    int width);
 void SobelRow_NEON(const uint8* src_sobelx,
                    const uint8* src_sobely,
                    uint8* dst_argb,
                    int width);
+void SobelRow_MSA(const uint8* src_sobelx,
+                  const uint8* src_sobely,
+                  uint8* dst_argb,
+                  int width);
 void SobelToPlaneRow_C(const uint8* src_sobelx,
                        const uint8* src_sobely,
                        uint8* dst_y,
                        int width);
 void SobelToPlaneRow_SSE2(const uint8* src_sobelx,
                           const uint8* src_sobely,
                           uint8* dst_y,
                           int width);
 void SobelToPlaneRow_NEON(const uint8* src_sobelx,
                           const uint8* src_sobely,
                           uint8* dst_y,
                           int width);
+void SobelToPlaneRow_MSA(const uint8* src_sobelx,
+                         const uint8* src_sobely,
+                         uint8* dst_y,
+                         int width);
 void SobelXYRow_C(const uint8* src_sobelx,
                   const uint8* src_sobely,
                   uint8* dst_argb,
                   int width);
 void SobelXYRow_SSE2(const uint8* src_sobelx,
                      const uint8* src_sobely,
                      uint8* dst_argb,
                      int width);
 void SobelXYRow_NEON(const uint8* src_sobelx,
                      const uint8* src_sobely,
                      uint8* dst_argb,
                      int width);
+void SobelXYRow_MSA(const uint8* src_sobelx,
+                    const uint8* src_sobely,
+                    uint8* dst_argb,
+                    int width);
 void SobelRow_Any_SSE2(const uint8* src_sobelx,
                        const uint8* src_sobely,
                        uint8* dst_argb,
                        int width);
 void SobelRow_Any_NEON(const uint8* src_sobelx,
                        const uint8* src_sobely,
                        uint8* dst_argb,
                        int width);
+void SobelRow_Any_MSA(const uint8* src_sobelx,
+                      const uint8* src_sobely,
+                      uint8* dst_argb,
+                      int width);
 void SobelToPlaneRow_Any_SSE2(const uint8* src_sobelx,
                               const uint8* src_sobely,
                               uint8* dst_y,
                               int width);
 void SobelToPlaneRow_Any_NEON(const uint8* src_sobelx,
                               const uint8* src_sobely,
                               uint8* dst_y,
                               int width);
+void SobelToPlaneRow_Any_MSA(const uint8* src_sobelx,
+                             const uint8* src_sobely,
+                             uint8* dst_y,
+                             int width);
 void SobelXYRow_Any_SSE2(const uint8* src_sobelx,
                          const uint8* src_sobely,
                          uint8* dst_argb,
                          int width);
 void SobelXYRow_Any_NEON(const uint8* src_sobelx,
                          const uint8* src_sobely,
                          uint8* dst_argb,
                          int width);
+void SobelXYRow_Any_MSA(const uint8* src_sobelx,
+                        const uint8* src_sobely,
+                        uint8* dst_argb,
+                        int width);
 
 void ARGBPolynomialRow_C(const uint8* src_argb,
                          uint8* dst_argb,
                          const float* poly,
                          int width);
 void ARGBPolynomialRow_SSE2(const uint8* src_argb,
                             uint8* dst_argb,
                             const float* poly,
                             int width);
 void ARGBPolynomialRow_AVX2(const uint8* src_argb,
@@ skipping 44 matching lines @@
                                  int width,
                                  const uint8* luma,
                                  uint32 lumacoeff);
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif
 
 #endif  // INCLUDE_LIBYUV_ROW_H_