avx2 interpolate use 8 bit

source/row_neon.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 2241 matching lines @@
     "+r"(width)        // %2
   :
   : "cc", "memory", "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "q8"
   );
 }
 
 // Bilinear filter 16x2 -> 16x1
 void InterpolateRow_NEON(uint8* dst_ptr,
                          const uint8* src_ptr, ptrdiff_t src_stride,
                          int dst_width, int source_y_fraction) {
-  int y1_fraction = source_y_fraction >> 1;
+  int y1_fraction = source_y_fraction;
   asm volatile (
     "cmp        %4, #0                         \n"
     "beq        100f                           \n"
     "add        %2, %1                         \n"
-    "cmp        %4, #64                        \n"
+    "cmp        %4, #128                       \n"
     "beq        50f                            \n"
 
     "vdup.8     d5, %4                         \n"
-    "rsb        %4, #128                       \n"
+    "rsb        %4, #256                       \n"
     "vdup.8     d4, %4                         \n"
     // General purpose row blend.
   "1:                                          \n"
     MEMACCESS(1)
     "vld1.8     {q0}, [%1]!                    \n"
     MEMACCESS(2)
     "vld1.8     {q1}, [%2]!                    \n"
     "subs       %3, %3, #16                    \n"
     "vmull.u8   q13, d0, d4                    \n"
     "vmull.u8   q14, d1, d4                    \n"
     "vmlal.u8   q13, d2, d5                    \n"
     "vmlal.u8   q14, d3, d5                    \n"
-    "vrshrn.u16 d0, q13, #7                    \n"
-    "vrshrn.u16 d1, q14, #7                    \n"
+    "vrshrn.u16 d0, q13, #8                    \n"
+    "vrshrn.u16 d1, q14, #8                    \n"
     MEMACCESS(0)
     "vst1.8     {q0}, [%0]!                    \n"
     "bgt        1b                             \n"
     "b          99f                            \n"
 
     // Blend 50 / 50.
   "50:                                         \n"
     MEMACCESS(1)
     "vld1.8     {q0}, [%1]!                    \n"
     MEMACCESS(2)
@@ skipping 583 matching lines @@
     "r"(6)             // %5
   : "cc", "memory", "q0", "q1"  // Clobber List
   );
 }
 #endif  // defined(__ARM_NEON__) && !defined(__aarch64__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif