libvpx: Pull from upstream

source/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_avx2.c

 /*
  *  Copyright (c) 2010 The WebM 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 93 matching lines @@
   for (i = output_height; i > 1; i-=2) {
     // load the 2 strides of source
     srcReg32b1 = _mm256_castsi128_si256(
                  _mm_loadu_si128((__m128i *)(src_ptr-3)));
     srcReg32b1 = _mm256_inserti128_si256(srcReg32b1,
                  _mm_loadu_si128((__m128i *)
                  (src_ptr+src_pixels_per_line-3)), 1);
 
     // filter the source buffer
     srcRegFilt32b1_1= _mm256_shuffle_epi8(srcReg32b1, filt1Reg);
-    srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt2Reg);
+    srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt4Reg);
 
     // multiply 2 adjacent elements with the filter and add the result
     srcRegFilt32b1_1 = _mm256_maddubs_epi16(srcRegFilt32b1_1, firstFilters);
-    srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, secondFilters);
+    srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, forthFilters);
 
     // add and saturate the results together
     srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, srcRegFilt32b2);
 
     // filter the source buffer
-    srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b1, filt4Reg);
+    srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b1, filt2Reg);
     srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt3Reg);
 
     // multiply 2 adjacent elements with the filter and add the result
-    srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, forthFilters);
+    srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, secondFilters);
     srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, thirdFilters);
 
     // add and saturate the results together
     srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1,
                        _mm256_min_epi16(srcRegFilt32b3, srcRegFilt32b2));
 
     // reading 2 strides of the next 16 bytes
     // (part of it was being read by earlier read)
     srcReg32b2 = _mm256_castsi128_si256(
                  _mm_loadu_si128((__m128i *)(src_ptr+5)));
     srcReg32b2 = _mm256_inserti128_si256(srcReg32b2,
                  _mm_loadu_si128((__m128i *)
                  (src_ptr+src_pixels_per_line+5)), 1);
 
     // add and saturate the results together
     srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1,
                        _mm256_max_epi16(srcRegFilt32b3, srcRegFilt32b2));
 
     // filter the source buffer
     srcRegFilt32b2_1 = _mm256_shuffle_epi8(srcReg32b2, filt1Reg);
-    srcRegFilt32b2 = _mm256_shuffle_epi8(srcReg32b2, filt2Reg);
+    srcRegFilt32b2 = _mm256_shuffle_epi8(srcReg32b2, filt4Reg);
 
     // multiply 2 adjacent elements with the filter and add the result
     srcRegFilt32b2_1 = _mm256_maddubs_epi16(srcRegFilt32b2_1, firstFilters);
-    srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, secondFilters);
+    srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, forthFilters);
 
     // add and saturate the results together
     srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, srcRegFilt32b2);
 
     // filter the source buffer
-    srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b2, filt4Reg);
+    srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b2, filt2Reg);
     srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b2, filt3Reg);
 
     // multiply 2 adjacent elements with the filter and add the result
-    srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, forthFilters);
+    srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, secondFilters);
     srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, thirdFilters);
 
     // add and saturate the results together
     srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1,
                        _mm256_min_epi16(srcRegFilt32b3, srcRegFilt32b2));
     srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1,
                        _mm256_max_epi16(srcRegFilt32b3, srcRegFilt32b2));
 
 
     srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, addFilterReg64);
@@ skipping 27 matching lines @@
   if (i > 0) {
     __m128i srcReg1, srcReg2, srcRegFilt1_1, srcRegFilt2_1;
     __m128i srcRegFilt2, srcRegFilt3;
 
     srcReg1 = _mm_loadu_si128((__m128i *)(src_ptr-3));
 
     // filter the source buffer
     srcRegFilt1_1 = _mm_shuffle_epi8(srcReg1,
                     _mm256_castsi256_si128(filt1Reg));
     srcRegFilt2 = _mm_shuffle_epi8(srcReg1,
-                  _mm256_castsi256_si128(filt2Reg));
+                  _mm256_castsi256_si128(filt4Reg));
 
     // multiply 2 adjacent elements with the filter and add the result
     srcRegFilt1_1 = _mm_maddubs_epi16(srcRegFilt1_1,
                     _mm256_castsi256_si128(firstFilters));
     srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
-                  _mm256_castsi256_si128(secondFilters));
+                  _mm256_castsi256_si128(forthFilters));
 
     // add and saturate the results together
     srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, srcRegFilt2);
 
     // filter the source buffer
     srcRegFilt3= _mm_shuffle_epi8(srcReg1,
-                 _mm256_castsi256_si128(filt4Reg));
+                 _mm256_castsi256_si128(filt2Reg));
     srcRegFilt2= _mm_shuffle_epi8(srcReg1,
                  _mm256_castsi256_si128(filt3Reg));
 
     // multiply 2 adjacent elements with the filter and add the result
     srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3,
-                  _mm256_castsi256_si128(forthFilters));
+                  _mm256_castsi256_si128(secondFilters));
     srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
                   _mm256_castsi256_si128(thirdFilters));
 
     // add and saturate the results together
     srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
                     _mm_min_epi16(srcRegFilt3, srcRegFilt2));
 
     // reading the next 16 bytes
     // (part of it was being read by earlier read)
     srcReg2 = _mm_loadu_si128((__m128i *)(src_ptr+5));
 
     // add and saturate the results together
     srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
                     _mm_max_epi16(srcRegFilt3, srcRegFilt2));
 
     // filter the source buffer
     srcRegFilt2_1 = _mm_shuffle_epi8(srcReg2,
                     _mm256_castsi256_si128(filt1Reg));
     srcRegFilt2 = _mm_shuffle_epi8(srcReg2,
-                  _mm256_castsi256_si128(filt2Reg));
+                  _mm256_castsi256_si128(filt4Reg));
 
     // multiply 2 adjacent elements with the filter and add the result
     srcRegFilt2_1 = _mm_maddubs_epi16(srcRegFilt2_1,
                     _mm256_castsi256_si128(firstFilters));
     srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
-                  _mm256_castsi256_si128(secondFilters));
+                  _mm256_castsi256_si128(forthFilters));
 
     // add and saturate the results together
     srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, srcRegFilt2);
 
     // filter the source buffer
     srcRegFilt3 = _mm_shuffle_epi8(srcReg2,
-                  _mm256_castsi256_si128(filt4Reg));
+                  _mm256_castsi256_si128(filt2Reg));
     srcRegFilt2 = _mm_shuffle_epi8(srcReg2,
                   _mm256_castsi256_si128(filt3Reg));
 
     // multiply 2 adjacent elements with the filter and add the result
     srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3,
-                  _mm256_castsi256_si128(forthFilters));
+                  _mm256_castsi256_si128(secondFilters));
     srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
                   _mm256_castsi256_si128(thirdFilters));
 
     // add and saturate the results together
     srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
                     _mm_min_epi16(srcRegFilt3, srcRegFilt2));
     srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
                     _mm_max_epi16(srcRegFilt3, srcRegFilt2));
 
 
@@ skipping 255 matching lines @@
 
     // shrink to 8 bit each 16 bits, the first lane contain the first
     // convolve result and the second lane contain the second convolve
     // result
     srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt3);
 
     // save 16 bytes
     _mm_store_si128((__m128i*)output_ptr, srcRegFilt1);
   }
 }