Libyuv MIPS DSPR2 optimizations.

source/convert.cc

 /*
  *  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 561 matching lines @@
   }
 #endif
 #if defined(HAS_ARGBTOUVROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ARGBToUVRow = ARGBToUVRow_Any_NEON;
     if (IS_ALIGNED(width, 16)) {
       ARGBToUVRow = ARGBToUVRow_NEON;
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2)) {
+    ARGBToYRow = ARGBToYRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_DSPR2;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_MSA)
   if (TestCpuFlag(kCpuHasMSA)) {
     ARGBToYRow = ARGBToYRow_Any_MSA;
     if (IS_ALIGNED(width, 16)) {
       ARGBToYRow = ARGBToYRow_MSA;
     }
   }
 #endif
+#if defined(HAS_ARGBTOUVROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2)) {
+    ARGBToUVRow = ARGBToUVRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToUVRow = ARGBToUVRow_DSPR2;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOUVROW_MSA)
   if (TestCpuFlag(kCpuHasMSA)) {
     ARGBToUVRow = ARGBToUVRow_Any_MSA;
     if (IS_ALIGNED(width, 32)) {
       ARGBToUVRow = ARGBToUVRow_MSA;
     }
   }
 #endif
 
   for (y = 0; y < height - 1; y += 2) {
@@ skipping 57 matching lines @@
   }
 #endif
 #if defined(HAS_BGRATOUVROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     BGRAToUVRow = BGRAToUVRow_Any_NEON;
     if (IS_ALIGNED(width, 16)) {
       BGRAToUVRow = BGRAToUVRow_NEON;
     }
   }
 #endif
+#if defined(HAS_BGRATOYROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2)) {
+    BGRAToYRow = BGRAToYRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 8)) {
+      BGRAToYRow = BGRAToYRow_DSPR2;
+    }
+  }
+#endif
+#if defined(HAS_BGRATOUVROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2)) {
+    BGRAToUVRow = BGRAToUVRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 16)) {
+      BGRAToUVRow = BGRAToUVRow_DSPR2;
+    }
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     BGRAToUVRow(src_bgra, src_stride_bgra, dst_u, dst_v, width);
     BGRAToYRow(src_bgra, dst_y, width);
     BGRAToYRow(src_bgra + src_stride_bgra, dst_y + dst_stride_y, width);
     src_bgra += src_stride_bgra * 2;
     dst_y += dst_stride_y * 2;
     dst_u += dst_stride_u;
     dst_v += dst_stride_v;
   }
@@ skipping 49 matching lines @@
   }
 #endif
 #if defined(HAS_ABGRTOUVROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     ABGRToUVRow = ABGRToUVRow_Any_NEON;
     if (IS_ALIGNED(width, 16)) {
       ABGRToUVRow = ABGRToUVRow_NEON;
     }
   }
 #endif
+#if defined(HAS_ABGRTOYROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2)) {
+    ABGRToYRow = ABGRToYRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 8)) {
+      ABGRToYRow = ABGRToYRow_DSPR2;
+    }
+  }
+#endif
+#if defined(HAS_ABGRTOUVROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2)) {
+    ABGRToUVRow = ABGRToUVRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 16)) {
+      ABGRToUVRow = ABGRToUVRow_DSPR2;
+    }
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     ABGRToUVRow(src_abgr, src_stride_abgr, dst_u, dst_v, width);
     ABGRToYRow(src_abgr, dst_y, width);
     ABGRToYRow(src_abgr + src_stride_abgr, dst_y + dst_stride_y, width);
     src_abgr += src_stride_abgr * 2;
     dst_y += dst_stride_y * 2;
     dst_u += dst_stride_u;
     dst_v += dst_stride_v;
   }
@@ skipping 49 matching lines @@
   }
 #endif
 #if defined(HAS_RGBATOUVROW_NEON)
   if (TestCpuFlag(kCpuHasNEON)) {
     RGBAToUVRow = RGBAToUVRow_Any_NEON;
     if (IS_ALIGNED(width, 16)) {
       RGBAToUVRow = RGBAToUVRow_NEON;
     }
   }
 #endif
+#if defined(HAS_RGBATOYROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2)) {
+    RGBAToYRow = RGBAToYRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 8)) {
+      RGBAToYRow = RGBAToYRow_DSPR2;
+    }
+  }
+#endif
+#if defined(HAS_RGBATOUVROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2)) {
+    RGBAToUVRow = RGBAToUVRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 16)) {
+      RGBAToUVRow = RGBAToUVRow_DSPR2;
+    }
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     RGBAToUVRow(src_rgba, src_stride_rgba, dst_u, dst_v, width);
     RGBAToYRow(src_rgba, dst_y, width);
     RGBAToYRow(src_rgba + src_stride_rgba, dst_y + dst_stride_y, width);
     src_rgba += src_stride_rgba * 2;
     dst_y += dst_stride_y * 2;
     dst_u += dst_stride_u;
     dst_v += dst_stride_v;
   }
@@ skipping 193 matching lines @@
 #if defined(HAS_ARGBTOYROW_AVX2) && defined(HAS_ARGBTOUVROW_AVX2)
     if (TestCpuFlag(kCpuHasAVX2)) {
       ARGBToUVRow = ARGBToUVRow_Any_AVX2;
       ARGBToYRow = ARGBToYRow_Any_AVX2;
       if (IS_ALIGNED(width, 32)) {
         ARGBToUVRow = ARGBToUVRow_AVX2;
         ARGBToYRow = ARGBToYRow_AVX2;
       }
     }
 #endif
+#if defined(HAS_RAWTOARGBROW_DSPR2)
+    if (TestCpuFlag(kCpuHasDSPR2)) {
+      RAWToARGBRow = RAWToARGBRow_Any_DSPR2;
+      if (IS_ALIGNED(width, 4)) {
+        RAWToARGBRow = RAWToARGBRow_DSPR2;
+      }
+    }
+#endif
     {
       // Allocate 2 rows of ARGB.
       const int kRowSize = (width * 4 + 31) & ~31;
       align_buffer_64(row, kRowSize * 2);
 #endif
 
   for (y = 0; y < height - 1; y += 2) {
 #if defined(HAS_RAWTOYROW_NEON)
     RAWToUVRow(src_raw, src_stride_raw, dst_u, dst_v, width);
     RAWToYRow(src_raw, dst_y, width);
@@ skipping 108 matching lines @@
 #if defined(HAS_ARGBTOYROW_AVX2) && defined(HAS_ARGBTOUVROW_AVX2)
       if (TestCpuFlag(kCpuHasAVX2)) {
         ARGBToUVRow = ARGBToUVRow_Any_AVX2;
         ARGBToYRow = ARGBToYRow_Any_AVX2;
         if (IS_ALIGNED(width, 32)) {
           ARGBToUVRow = ARGBToUVRow_AVX2;
           ARGBToYRow = ARGBToYRow_AVX2;
         }
       }
 #endif
+#if defined(HAS_RGB565TOARGBROW_DSPR2)
+      if (TestCpuFlag(kCpuHasDSPR2)) {
+        RGB565ToARGBRow = RGB565ToARGBRow_Any_DSPR2;
+        if (IS_ALIGNED(width, 8)) {
+          RGB565ToARGBRow = RGB565ToARGBRow_DSPR2;
+        }
+      }
+#endif
       {
         // Allocate 2 rows of ARGB.
         const int kRowSize = (width * 4 + 31) & ~31;
         align_buffer_64(row, kRowSize * 2);
 #endif
 
   for (y = 0; y < height - 1; y += 2) {
 #if defined(HAS_RGB565TOYROW_NEON)
     RGB565ToUVRow(src_rgb565, src_stride_rgb565, dst_u, dst_v, width);
     RGB565ToYRow(src_rgb565, dst_y, width);
@@ skipping 389 matching lines @@
     dst_u += dst_stride_u;
     dst_v += dst_stride_v;
   }
   return 0;
 }
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif