Add MSA optimized remaining scale row functions

source/scale.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 89 matching lines @@
     }
   }
 #endif
 #if defined(HAS_SCALEROWDOWN2_DSPR2)
   if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_ptr, 4) &&
       IS_ALIGNED(src_stride, 4) && IS_ALIGNED(row_stride, 4) &&
       IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) {
     ScaleRowDown2 = filtering ? ScaleRowDown2Box_DSPR2 : ScaleRowDown2_DSPR2;
   }
 #endif
+#if defined(HAS_SCALEROWDOWN2_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    ScaleRowDown2 =
+        filtering == kFilterNone
+            ? ScaleRowDown2_Any_MSA
+            : (filtering == kFilterLinear ? ScaleRowDown2Linear_Any_MSA
+                                          : ScaleRowDown2Box_Any_MSA);
+    if (IS_ALIGNED(dst_width, 32)) {
+      ScaleRowDown2 = filtering == kFilterNone ? ScaleRowDown2_MSA
+                                               : (filtering == kFilterLinear
+                                                      ? ScaleRowDown2Linear_MSA
+                                                      : ScaleRowDown2Box_MSA);
+    }
+  }
+#endif
 
   if (filtering == kFilterLinear) {
     src_stride = 0;
   }
   // TODO(fbarchard): Loop through source height to allow odd height.
   for (y = 0; y < dst_height; ++y) {
     ScaleRowDown2(src_ptr, src_stride, dst_ptr, dst_width);
     src_ptr += row_stride;
     dst_ptr += dst_stride;
   }
@@ skipping 105 matching lines @@
     }
   }
 #endif
 #if defined(HAS_SCALEROWDOWN4_DSPR2)
   if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(row_stride, 4) &&
       IS_ALIGNED(src_ptr, 4) && IS_ALIGNED(src_stride, 4) &&
       IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) {
     ScaleRowDown4 = filtering ? ScaleRowDown4Box_DSPR2 : ScaleRowDown4_DSPR2;
   }
 #endif
+#if defined(HAS_SCALEROWDOWN4_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    ScaleRowDown4 =
+        filtering ? ScaleRowDown4Box_Any_MSA : ScaleRowDown4_Any_MSA;
+    if (IS_ALIGNED(dst_width, 16)) {
+      ScaleRowDown4 = filtering ? ScaleRowDown4Box_MSA : ScaleRowDown4_MSA;
+    }
+  }
+#endif
 
   if (filtering == kFilterLinear) {
     src_stride = 0;
   }
   for (y = 0; y < dst_height; ++y) {
     ScaleRowDown4(src_ptr, src_stride, dst_ptr, dst_width);
     src_ptr += row_stride;
     dst_ptr += dst_stride;
   }
 }
@@ skipping 315 matching lines @@
       IS_ALIGNED(dst_ptr, 4) && IS_ALIGNED(dst_stride, 4)) {
     if (!filtering) {
       ScaleRowDown38_3 = ScaleRowDown38_DSPR2;
       ScaleRowDown38_2 = ScaleRowDown38_DSPR2;
     } else {
       ScaleRowDown38_3 = ScaleRowDown38_3_Box_DSPR2;
       ScaleRowDown38_2 = ScaleRowDown38_2_Box_DSPR2;
     }
   }
 #endif
+#if defined(HAS_SCALEROWDOWN38_MSA)
+  if (TestCpuFlag(kCpuHasMSA)) {
+    if (!filtering) {
+      ScaleRowDown38_3 = ScaleRowDown38_Any_MSA;
+      ScaleRowDown38_2 = ScaleRowDown38_Any_MSA;
+    } else {
+      ScaleRowDown38_3 = ScaleRowDown38_3_Box_Any_MSA;
+      ScaleRowDown38_2 = ScaleRowDown38_2_Box_Any_MSA;
+    }
+    if (dst_width % 12 == 0) {
+      if (!filtering) {
+        ScaleRowDown38_3 = ScaleRowDown38_MSA;
+        ScaleRowDown38_2 = ScaleRowDown38_MSA;
+      } else {
+        ScaleRowDown38_3 = ScaleRowDown38_3_Box_MSA;
+        ScaleRowDown38_2 = ScaleRowDown38_2_Box_MSA;
+      }
+    }
+  }
+#endif
 
   for (y = 0; y < dst_height - 2; y += 3) {
     ScaleRowDown38_3(src_ptr, filter_stride, dst_ptr, dst_width);
     src_ptr += src_stride * 3;
     dst_ptr += dst_stride;
     ScaleRowDown38_3(src_ptr, filter_stride, dst_ptr, dst_width);
     src_ptr += src_stride * 3;
     dst_ptr += dst_stride;
     ScaleRowDown38_2(src_ptr, filter_stride, dst_ptr, dst_width);
     src_ptr += src_stride * 2;
@@ skipping 255 matching lines @@
     }
 #endif
 #if defined(HAS_SCALEADDROW_NEON)
     if (TestCpuFlag(kCpuHasNEON)) {
       ScaleAddRow = ScaleAddRow_Any_NEON;
       if (IS_ALIGNED(src_width, 16)) {
         ScaleAddRow = ScaleAddRow_NEON;
       }
     }
 #endif
+#if defined(HAS_SCALEADDROW_MSA)
+    if (TestCpuFlag(kCpuHasMSA)) {
+      ScaleAddRow = ScaleAddRow_Any_MSA;
+      if (IS_ALIGNED(src_width, 16)) {
+        ScaleAddRow = ScaleAddRow_MSA;
+      }
+    }
+#endif
 
     for (j = 0; j < dst_height; ++j) {
       int boxheight;
       int iy = y >> 16;
       const uint8* src = src_ptr + iy * src_stride;
       y += dy;
       if (y > max_y) {
         y = max_y;
       }
       boxheight = MIN1((y >> 16) - iy);
@@ skipping 915 matching lines @@
   return I420Scale(src_y, src_width, src_u, src_halfwidth, src_v, src_halfwidth,
                    src_width, src_height, dst_y, dst_width, dst_u,
                    dst_halfwidth, dst_v, dst_halfwidth, dst_width, aheight,
                    interpolate ? kFilterBox : kFilterNone);
 }
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif