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>

1 files changed, 226 insertions, 0 deletions

diff --git a/source/convert_from_argb.cc b/source/convert_from_argb.cc
index c8d91252..1b070c10 100644..100755
--- a/source/convert_from_argb.cc
+++ b/source/convert_from_argb.cc
@@ -76,6 +76,14 @@ int ARGBToI444(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOUV444ROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToUV444Row = ARGBToUV444Row_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToUV444Row = ARGBToUV444Row_MMI;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOYROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ARGBToYRow = ARGBToYRow_Any_SSSE3;
@@ -108,6 +116,14 @@ int ARGBToI444(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYRow = ARGBToYRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToUV444Row(src_argb, dst_u, dst_v, width);
@@ -208,6 +224,23 @@ int ARGBToI422(const uint8_t* src_argb,
   }
 #endif
 
+#if defined(HAS_ARGBTOYROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYRow = ARGBToYRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_MMI;
+    }
+  }
+#endif
+#if defined(HAS_ARGBTOUVROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToUVRow = ARGBToUVRow_MMI;
+    }
+  }
+#endif
+
   for (y = 0; y < height; ++y) {
     ARGBToUVRow(src_argb, 0, dst_u, dst_v, width);
     ARGBToYRow(src_argb, dst_y, width);
@@ -298,6 +331,22 @@ int ARGBToNV12(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYRow = ARGBToYRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_MMI;
+    }
+  }
+#endif
+#if defined(HAS_ARGBTOUVROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToUVRow = ARGBToUVRow_MMI;
+    }
+  }
+#endif
 #if defined(HAS_MERGEUVROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     MergeUVRow_ = MergeUVRow_Any_SSE2;
@@ -330,6 +379,14 @@ int ARGBToNV12(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_MERGEUVROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    MergeUVRow_ = MergeUVRow_Any_MMI;
+    if (IS_ALIGNED(halfwidth, 8)) {
+      MergeUVRow_ = MergeUVRow_MMI;
+    }
+  }
+#endif
   {
     // Allocate a rows of uv.
     align_buffer_64(row_u, ((halfwidth + 31) & ~31) * 2);
@@ -434,6 +491,23 @@ int ARGBToNV21(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYRow = ARGBToYRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_MMI;
+    }
+  }
+#endif
+#if defined(HAS_ARGBTOUVROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToUVRow = ARGBToUVRow_MMI;
+    }
+  }
+#endif
+
 #if defined(HAS_MERGEUVROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     MergeUVRow_ = MergeUVRow_Any_SSE2;
@@ -466,6 +540,14 @@ int ARGBToNV21(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_MERGEUVROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    MergeUVRow_ = MergeUVRow_Any_MMI;
+    if (IS_ALIGNED(halfwidth, 8)) {
+      MergeUVRow_ = MergeUVRow_MMI;
+    }
+  }
+#endif
   {
     // Allocate a rows of uv.
     align_buffer_64(row_u, ((halfwidth + 31) & ~31) * 2);
@@ -575,6 +657,22 @@ int ARGBToYUY2(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYRow = ARGBToYRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_MMI;
+    }
+  }
+#endif
+#if defined(HAS_ARGBTOUVROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToUVRow = ARGBToUVRow_MMI;
+    }
+  }
+#endif
 #if defined(HAS_I422TOYUY2ROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     I422ToYUY2Row = I422ToYUY2Row_Any_SSE2;
@@ -607,6 +705,14 @@ int ARGBToYUY2(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_I422TOYUY2ROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    I422ToYUY2Row = I422ToYUY2Row_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      I422ToYUY2Row = I422ToYUY2Row_MMI;
+    }
+  }
+#endif
 
   {
     // Allocate a rows of yuv.
@@ -712,6 +818,22 @@ int ARGBToUYVY(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYRow = ARGBToYRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_MMI;
+    }
+  }
+#endif
+#if defined(HAS_ARGBTOUVROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToUVRow = ARGBToUVRow_Any_MMI;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToUVRow = ARGBToUVRow_MMI;
+    }
+  }
+#endif
 #if defined(HAS_I422TOUYVYROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     I422ToUYVYRow = I422ToUYVYRow_Any_SSE2;
@@ -744,6 +866,14 @@ int ARGBToUYVY(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_I422TOUYVYROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    I422ToUYVYRow = I422ToUYVYRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      I422ToUYVYRow = I422ToUYVYRow_MMI;
+    }
+  }
+#endif
 
   {
     // Allocate a rows of yuv.
@@ -821,6 +951,14 @@ int ARGBToI400(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYRow = ARGBToYRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToYRow(src_argb, dst_y, width);
@@ -911,6 +1049,14 @@ int ARGBToRGB24(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTORGB24ROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToRGB24Row = ARGBToRGB24Row_Any_MMI;
+    if (IS_ALIGNED(width, 4)) {
+      ARGBToRGB24Row = ARGBToRGB24Row_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToRGB24Row(src_argb, dst_rgb24, width);
@@ -977,6 +1123,14 @@ int ARGBToRAW(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTORAWROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToRAWRow = ARGBToRAWRow_Any_MMI;
+    if (IS_ALIGNED(width, 4)) {
+      ARGBToRAWRow = ARGBToRAWRow_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToRAWRow(src_argb, dst_raw, width);
@@ -1047,6 +1201,14 @@ int ARGBToRGB565Dither(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTORGB565DITHERROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_Any_MMI;
+    if (IS_ALIGNED(width, 4)) {
+      ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToRGB565DitherRow(src_argb, dst_rgb565,
@@ -1116,6 +1278,14 @@ int ARGBToRGB565(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTORGB565ROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToRGB565Row = ARGBToRGB565Row_Any_MMI;
+    if (IS_ALIGNED(width, 4)) {
+      ARGBToRGB565Row = ARGBToRGB565Row_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToRGB565Row(src_argb, dst_rgb565, width);
@@ -1182,6 +1352,14 @@ int ARGBToARGB1555(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOARGB1555ROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToARGB1555Row = ARGBToARGB1555Row_Any_MMI;
+    if (IS_ALIGNED(width, 4)) {
+      ARGBToARGB1555Row = ARGBToARGB1555Row_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToARGB1555Row(src_argb, dst_argb1555, width);
@@ -1248,6 +1426,14 @@ int ARGBToARGB4444(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOARGB4444ROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToARGB4444Row = ARGBToARGB4444Row_Any_MMI;
+    if (IS_ALIGNED(width, 4)) {
+      ARGBToARGB4444Row = ARGBToARGB4444Row_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToARGB4444Row(src_argb, dst_argb4444, width);
@@ -1424,6 +1610,14 @@ int ARGBToJ420(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYJRow = ARGBToYJRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYJRow = ARGBToYJRow_MMI;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOUVJROW_MSA)
   if (TestCpuFlag(kCpuHasMSA)) {
     ARGBToUVJRow = ARGBToUVJRow_Any_MSA;
@@ -1432,6 +1626,14 @@ int ARGBToJ420(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOUVJROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToUVJRow = ARGBToUVJRow_Any_MMI;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToUVJRow = ARGBToUVJRow_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height - 1; y += 2) {
     ARGBToUVJRow(src_argb, src_stride_argb, dst_u, dst_v, width);
@@ -1525,6 +1727,14 @@ int ARGBToJ422(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYJRow = ARGBToYJRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYJRow = ARGBToYJRow_MMI;
+    }
+  }
+#endif
 #if defined(HAS_ARGBTOUVJROW_MSA)
   if (TestCpuFlag(kCpuHasMSA)) {
     ARGBToUVJRow = ARGBToUVJRow_Any_MSA;
@@ -1533,6 +1743,14 @@ int ARGBToJ422(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOUVJROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToUVJRow = ARGBToUVJRow_Any_MMI;
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToUVJRow = ARGBToUVJRow_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToUVJRow(src_argb, 0, dst_u, dst_v, width);
@@ -1602,6 +1820,14 @@ int ARGBToJ400(const uint8_t* src_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTOYJROW_MMI)
+  if (TestCpuFlag(kCpuHasMMI)) {
+    ARGBToYJRow = ARGBToYJRow_Any_MMI;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYJRow = ARGBToYJRow_MMI;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     ARGBToYJRow(src_argb, dst_yj, width);