Libyuv MIPS DSPR2 optimizations.

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 346 matching lines @@
 // The following are available on Mips platforms:
 #if !defined(LIBYUV_DISABLE_MIPS) && defined(__mips__) && \
     (_MIPS_SIM == _MIPS_SIM_ABI32) && (__mips_isa_rev < 6)
 #define HAS_COPYROW_MIPS
 #if defined(__mips_dsp) && (__mips_dsp_rev >= 2)
 #define HAS_I422TOARGBROW_DSPR2
 #define HAS_INTERPOLATEROW_DSPR2
 #define HAS_MIRRORROW_DSPR2
 #define HAS_MIRRORUVROW_DSPR2
 #define HAS_SPLITUVROW_DSPR2
+#define HAS_RGB24TOARGBROW_DSPR2
+#define HAS_RAWTOARGBROW_DSPR2
+#define HAS_RGB565TOARGBROW_DSPR2
+#define HAS_ARGB1555TOARGBROW_DSPR2
+#define HAS_ARGB4444TOARGBROW_DSPR2
+#define HAS_I444TOARGBROW_DSPR2
+#define HAS_I422TOARGB4444ROW_DSPR2
+#define HAS_I422TOARGB1555ROW_DSPR2
+#define HAS_NV12TOARGBROW_DSPR2
+#define HAS_BGRATOUVROW_DSPR2
+#define HAS_BGRATOYROW_DSPR2
+#define HAS_ABGRTOUVROW_DSPR2
+#define HAS_ARGBTOYROW_DSPR2
+#define HAS_ABGRTOYROW_DSPR2
+#define HAS_RGBATOUVROW_DSPR2
+#define HAS_RGBATOYROW_DSPR2
+#define HAS_ARGBTOUVROW_DSPR2
 #endif
 #endif
 
 #if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
 #define HAS_ARGBMIRRORROW_MSA
 #define HAS_I422TOUYVYROW_MSA
 #define HAS_I422TOYUY2ROW_MSA
 #define HAS_MIRRORROW_MSA
 #define HAS_UYVYTOUVROW_MSA
 #define HAS_UYVYTOYROW_MSA
@@ skipping 276 matching lines @@
                         const struct YuvConstants* yuvconstants,
                         int width);
 void YUY2ToARGBRow_NEON(const uint8* src_yuy2,
                         uint8* dst_argb,
                         const struct YuvConstants* yuvconstants,
                         int width);
 void UYVYToARGBRow_NEON(const uint8* src_uyvy,
                         uint8* dst_argb,
                         const struct YuvConstants* yuvconstants,
                         int width);
+void I444ToARGBRow_DSPR2(const uint8* src_y,
+                         const uint8* src_u,
+                         const uint8* src_v,
+                         uint8* dst_argb,
+                         const struct YuvConstants* yuvconstants,
+                         int width);
+void I422ToARGB4444Row_DSPR2(const uint8* src_y,
+                             const uint8* src_u,
+                             const uint8* src_v,
+                             uint8* dst_argb4444,
+                             const struct YuvConstants* yuvconstants,
+                             int width);
+void I422ToARGB1555Row_DSPR2(const uint8* src_y,
+                             const uint8* src_u,
+                             const uint8* src_v,
+                             uint8* dst_argb1555,
+                             const struct YuvConstants* yuvconstants,
+                             int width);
+void NV12ToARGBRow_DSPR2(const uint8* src_y,
+                         const uint8* src_uv,
+                         uint8* dst_argb,
+                         const struct YuvConstants* yuvconstants,
+                         int width);
+
 void I422ToARGBRow_MSA(const uint8* src_y,
                        const uint8* src_u,
                        const uint8* src_v,
                        uint8* dst_argb,
                        const struct YuvConstants* yuvconstants,
                        int width);
 void I422ToRGBARow_MSA(const uint8* src_y,
                        const uint8* src_u,
                        const uint8* src_v,
                        uint8* dst_rgba,
@@ skipping 109 matching lines @@
                           uint8* dst_v,
                           int width);
 void BGRAToYRow_NEON(const uint8* src_bgra, uint8* dst_y, int width);
 void ABGRToYRow_NEON(const uint8* src_abgr, uint8* dst_y, int width);
 void RGBAToYRow_NEON(const uint8* src_rgba, uint8* dst_y, int width);
 void RGB24ToYRow_NEON(const uint8* src_rgb24, uint8* dst_y, int width);
 void RAWToYRow_NEON(const uint8* src_raw, uint8* dst_y, int width);
 void RGB565ToYRow_NEON(const uint8* src_rgb565, uint8* dst_y, int width);
 void ARGB1555ToYRow_NEON(const uint8* src_argb1555, uint8* dst_y, int width);
 void ARGB4444ToYRow_NEON(const uint8* src_argb4444, uint8* dst_y, int width);
+void BGRAToUVRow_DSPR2(const uint8* src_bgra,
+                       int src_stride_bgra,
+                       uint8* dst_u,
+                       uint8* dst_v,
+                       int width);
+void BGRAToYRow_DSPR2(const uint8* src_bgra, uint8* dst_y, int width);
+void ABGRToUVRow_DSPR2(const uint8* src_abgr,
+                       int src_stride_abgr,
+                       uint8* dst_u,
+                       uint8* dst_v,
+                       int width);
+void ARGBToYRow_DSPR2(const uint8* src_argb, uint8* dst_y, int width);
+void ABGRToYRow_DSPR2(const uint8* src_abgr, uint8* dst_y, int width);
+void RGBAToUVRow_DSPR2(const uint8* src_rgba,
+                       int src_stride_rgba,
+                       uint8* dst_u,
+                       uint8* dst_v,
+                       int width);
+void RGBAToYRow_DSPR2(const uint8* src_rgba, uint8* dst_y, int width);
+void ARGBToUVRow_DSPR2(const uint8* src_argb,
+                       int src_stride_argb,
+                       uint8* dst_u,
+                       uint8* dst_v,
+                       int width);
 void ARGBToYRow_C(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYJRow_C(const uint8* src_argb, uint8* dst_y, int width);
 void BGRAToYRow_C(const uint8* src_bgra, uint8* dst_y, int width);
 void ABGRToYRow_C(const uint8* src_abgr, uint8* dst_y, int width);
 void RGBAToYRow_C(const uint8* src_rgba, uint8* dst_y, int width);
 void RGB24ToYRow_C(const uint8* src_rgb24, uint8* dst_y, int width);
 void RAWToYRow_C(const uint8* src_raw, uint8* dst_y, int width);
 void RGB565ToYRow_C(const uint8* src_rgb565, uint8* dst_y, int width);
 void ARGB1555ToYRow_C(const uint8* src_argb1555, uint8* dst_y, int width);
 void ARGB4444ToYRow_C(const uint8* src_argb4444, uint8* dst_y, int width);
 void ARGBToYRow_Any_SSSE3(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYJRow_Any_SSSE3(const uint8* src_argb, uint8* dst_y, int width);
 void BGRAToYRow_Any_SSSE3(const uint8* src_bgra, uint8* dst_y, int width);
 void ABGRToYRow_Any_SSSE3(const uint8* src_abgr, uint8* dst_y, int width);
 void RGBAToYRow_Any_SSSE3(const uint8* src_rgba, uint8* dst_y, int width);
 void RGB24ToYRow_Any_SSSE3(const uint8* src_rgb24, uint8* dst_y, int width);
 void RAWToYRow_Any_SSSE3(const uint8* src_raw, uint8* dst_y, int width);
 void ARGBToYRow_Any_NEON(const uint8* src_argb, uint8* dst_y, int width);
 void ARGBToYJRow_Any_NEON(const uint8* src_argb, uint8* dst_y, int width);
 void BGRAToYRow_Any_NEON(const uint8* src_bgra, uint8* dst_y, int width);
 void ABGRToYRow_Any_NEON(const uint8* src_abgr, uint8* dst_y, int width);
 void RGBAToYRow_Any_NEON(const uint8* src_rgba, uint8* dst_y, int width);
 void RGB24ToYRow_Any_NEON(const uint8* src_rgb24, uint8* dst_y, int width);
 void RAWToYRow_Any_NEON(const uint8* src_raw, uint8* dst_y, int width);
 void RGB565ToYRow_Any_NEON(const uint8* src_rgb565, uint8* dst_y, int width);
 void ARGB1555ToYRow_Any_NEON(const uint8* src_argb1555,
                              uint8* dst_y,
                              int width);
+void BGRAToYRow_Any_DSPR2(const uint8* src_bgra, uint8* dst_y, int width);
+void ARGBToYRow_Any_DSPR2(const uint8* src_argb, uint8* dst_y, int width);
+void ABGRToYRow_Any_DSPR2(const uint8* src_abgr, uint8* dst_y, int width);
+void RGBAToYRow_Any_DSPR2(const uint8* src_rgba, uint8* dst_y, int width);
 void ARGB4444ToYRow_Any_NEON(const uint8* src_argb4444,
                              uint8* dst_y,
                              int width);
 void ARGBToYRow_Any_MSA(const uint8* src_argb, uint8* dst_y, int width);
 
 void ARGBToUVRow_AVX2(const uint8* src_argb,
                       int src_stride_argb,
                       uint8* dst_u,
                       uint8* dst_v,
                       int width);
@@ skipping 118 matching lines @@
 void ARGB1555ToUVRow_Any_NEON(const uint8* src_argb1555,
                               int src_stride_argb1555,
                               uint8* dst_u,
                               uint8* dst_v,
                               int width);
 void ARGB4444ToUVRow_Any_NEON(const uint8* src_argb4444,
                               int src_stride_argb4444,
                               uint8* dst_u,
                               uint8* dst_v,
                               int width);
+void BGRAToUVRow_Any_DSPR2(const uint8* src_bgra,
+                           int src_stride_bgra,
+                           uint8* dst_u,
+                           uint8* dst_v,
+                           int width);
+void ABGRToUVRow_Any_DSPR2(const uint8* src_abgr,
+                           int src_stride_abgr,
+                           uint8* dst_u,
+                           uint8* dst_v,
+                           int width);
+void RGBAToUVRow_Any_DSPR2(const uint8* src_rgba,
+                           int src_stride_rgba,
+                           uint8* dst_u,
+                           uint8* dst_v,
+                           int width);
+void ARGBToUVRow_Any_DSPR2(const uint8* src_argb,
+                           int src_stride_argb,
+                           uint8* dst_u,
+                           uint8* dst_v,
+                           int width);
 void ARGBToUVRow_C(const uint8* src_argb,
                    int src_stride_argb,
                    uint8* dst_u,
+                   uint8* dst_v,
+                   int width);
+void ARGBToUVJRow_C(const uint8* src_argb,
+                    int src_stride_argb,
+                    uint8* dst_u,
+                    uint8* dst_v,
+                    int width);
+void ARGBToUVRow_C(const uint8* src_argb,
+                   int src_stride_argb,
+                   uint8* dst_u,
                    uint8* dst_v,
                    int width);
 void ARGBToUVJRow_C(const uint8* src_argb,
                     int src_stride_argb,
                     uint8* dst_u,
                     uint8* dst_v,
                     int width);
 void BGRAToUVRow_C(const uint8* src_bgra,
                    int src_stride_bgra,
                    uint8* dst_u,
@@ skipping 273 matching lines @@
 void RGB24ToARGBRow_NEON(const uint8* src_rgb24, uint8* dst_argb, int width);
 void RAWToARGBRow_NEON(const uint8* src_raw, uint8* dst_argb, int width);
 void RAWToRGB24Row_NEON(const uint8* src_raw, uint8* dst_rgb24, int width);
 void RGB565ToARGBRow_NEON(const uint8* src_rgb565, uint8* dst_argb, int width);
 void ARGB1555ToARGBRow_NEON(const uint8* src_argb1555,
                             uint8* dst_argb,
                             int width);
 void ARGB4444ToARGBRow_NEON(const uint8* src_argb4444,
                             uint8* dst_argb,
                             int width);
+void RGB24ToARGBRow_DSPR2(const uint8* src_rgb24, uint8* dst_argb, int width);
+void RAWToARGBRow_DSPR2(const uint8* src_raw, uint8* dst_argb, int width);
+void RGB565ToARGBRow_DSPR2(const uint8* src_rgb565, uint8* dst_argb, int width);
+void ARGB1555ToARGBRow_DSPR2(const uint8* src_argb1555,
+                             uint8* dst_argb,
+                             int width);
+void ARGB4444ToARGBRow_DSPR2(const uint8* src_argb4444,
+                             uint8* dst_argb,
+                             int width);
 void ARGB4444ToARGBRow_MSA(const uint8* src_argb4444,
                            uint8* dst_argb,
                            int width);
 void RGB24ToARGBRow_C(const uint8* src_rgb24, uint8* dst_argb, int width);
 void RAWToARGBRow_C(const uint8* src_raw, uint8* dst_argb, int width);
 void RAWToRGB24Row_C(const uint8* src_raw, uint8* dst_rgb24, int width);
 void RGB565ToARGBRow_C(const uint8* src_rgb, uint8* dst_argb, int width);
 void ARGB1555ToARGBRow_C(const uint8* src_argb, uint8* dst_argb, int width);
 void ARGB4444ToARGBRow_C(const uint8* src_argb, uint8* dst_argb, int width);
 void RGB24ToARGBRow_Any_SSSE3(const uint8* src_rgb24,
@@ skipping 28 matching lines @@
 void RAWToRGB24Row_Any_NEON(const uint8* src_raw, uint8* dst_rgb24, int width);
 void RGB565ToARGBRow_Any_NEON(const uint8* src_rgb565,
                               uint8* dst_argb,
                               int width);
 void ARGB1555ToARGBRow_Any_NEON(const uint8* src_argb1555,
                                 uint8* dst_argb,
                                 int width);
 void ARGB4444ToARGBRow_Any_NEON(const uint8* src_argb4444,
                                 uint8* dst_argb,
                                 int width);
+void RGB24ToARGBRow_Any_DSPR2(const uint8* src_rgb24,
+                              uint8* dst_argb,
+                              int width);
+void RAWToARGBRow_Any_DSPR2(const uint8* src_raw, uint8* dst_argb, int width);
+void RGB565ToARGBRow_Any_DSPR2(const uint8* src_rgb565,
+                               uint8* dst_argb,
+                               int width);
+void ARGB1555ToARGBRow_Any_DSPR2(const uint8* src_argb1555,
+                                 uint8* dst_argb,
+                                 int width);
+void ARGB4444ToARGBRow_Any_DSPR2(const uint8* src_argb4444,
+                                 uint8* dst_argb,
+                                 int width);
+
 void ARGB4444ToARGBRow_Any_MSA(const uint8* src_argb4444,
                                uint8* dst_argb,
                                int width);
 
 void ARGBToRGB24Row_SSSE3(const uint8* src_argb, uint8* dst_rgb, int width);
 void ARGBToRAWRow_SSSE3(const uint8* src_argb, uint8* dst_rgb, int width);
 void ARGBToRGB565Row_SSE2(const uint8* src_argb, uint8* dst_rgb, int width);
 void ARGBToARGB1555Row_SSE2(const uint8* src_argb, uint8* dst_rgb, int width);
 void ARGBToARGB4444Row_SSE2(const uint8* src_argb, uint8* dst_rgb, int width);
 
@@ skipping 723 matching lines @@
                               const struct YuvConstants* yuvconstants,
                               int width);
 void YUY2ToARGBRow_Any_NEON(const uint8* src_yuy2,
                             uint8* dst_argb,
                             const struct YuvConstants* yuvconstants,
                             int width);
 void UYVYToARGBRow_Any_NEON(const uint8* src_uyvy,
                             uint8* dst_argb,
                             const struct YuvConstants* yuvconstants,
                             int width);
+void I444ToARGBRow_Any_DSPR2(const uint8* src_y,
+                             const uint8* src_u,
+                             const uint8* src_v,
+                             uint8* dst_argb,
+                             const struct YuvConstants* yuvconstants,
+                             int width);
+void I422ToARGB4444Row_Any_DSPR2(const uint8* src_y,
+                                 const uint8* src_u,
+                                 const uint8* src_v,
+                                 uint8* dst_argb,
+                                 const struct YuvConstants* yuvconstants,
+                                 int width);
+void I422ToARGBRow_Any_DSPR2(const uint8* src_y,
+                             const uint8* src_u,
+                             const uint8* src_v,
+                             uint8* dst_argb,
+                             const struct YuvConstants* yuvconstants,
+                             int width);
 void I422ToARGBRow_DSPR2(const uint8* src_y,
                          const uint8* src_u,
                          const uint8* src_v,
                          uint8* dst_argb,
                          const struct YuvConstants* yuvconstants,
                          int width);
+void I422ToARGB1555Row_Any_DSPR2(const uint8* src_y,
+                                 const uint8* src_u,
+                                 const uint8* src_v,
+                                 uint8* dst_argb,
+                                 const struct YuvConstants* yuvconstants,
+                                 int width);
+void I411ToARGBRow_Any_DSPR2(const uint8* src_y,
+                             const uint8* src_u,
+                             const uint8* src_v,
+                             uint8* dst_argb,
+                             const struct YuvConstants* yuvconstants,
+                             int width);
+void NV12ToARGBRow_Any_DSPR2(const uint8* src_y,
+                             const uint8* src_uv,
+                             uint8* dst_argb,
+                             const struct YuvConstants* yuvconstants,
+                             int width);
 void I422ToARGBRow_DSPR2(const uint8* src_y,
                          const uint8* src_u,
                          const uint8* src_v,
                          uint8* dst_argb,
                          const struct YuvConstants* yuvconstants,
                          int width);
 void I422ToARGBRow_Any_MSA(const uint8* src_y,
                            const uint8* src_u,
                            const uint8* src_v,
                            uint8* dst_argb,
@@ skipping 630 matching lines @@
                                  int width,
                                  const uint8* luma,
                                  uint32 lumacoeff);
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif
 
 #endif  // INCLUDE_LIBYUV_ROW_H_