libvpx: Pull from upstream

source/libvpx/vp9/common/x86/vp9_idct_intrin_sse2.h

 /*
  *  Copyright (c) 2014 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 97 matching lines @@
   in[15]  = _mm_load_si128((const __m128i *)(input + 15 * 16));
 }
 
 #define RECON_AND_STORE(dest, in_x) \
   {                                                     \
      __m128i d0 = _mm_loadl_epi64((__m128i *)(dest)); \
       d0 = _mm_unpacklo_epi8(d0, zero); \
       d0 = _mm_add_epi16(in_x, d0); \
       d0 = _mm_packus_epi16(d0, d0); \
       _mm_storel_epi64((__m128i *)(dest), d0); \
-      dest += stride; \
   }
 
 static INLINE void write_buffer_8x16(uint8_t *dest, __m128i *in, int stride) {
   const __m128i final_rounding = _mm_set1_epi16(1<<5);
   const __m128i zero = _mm_setzero_si128();
   // Final rounding and shift
   in[0] = _mm_adds_epi16(in[0], final_rounding);
   in[1] = _mm_adds_epi16(in[1], final_rounding);
   in[2] = _mm_adds_epi16(in[2], final_rounding);
   in[3] = _mm_adds_epi16(in[3], final_rounding);
@@ skipping 20 matching lines @@
   in[7] = _mm_srai_epi16(in[7], 6);
   in[8] = _mm_srai_epi16(in[8], 6);
   in[9] = _mm_srai_epi16(in[9], 6);
   in[10] = _mm_srai_epi16(in[10], 6);
   in[11] = _mm_srai_epi16(in[11], 6);
   in[12] = _mm_srai_epi16(in[12], 6);
   in[13] = _mm_srai_epi16(in[13], 6);
   in[14] = _mm_srai_epi16(in[14], 6);
   in[15] = _mm_srai_epi16(in[15], 6);
 
-  RECON_AND_STORE(dest, in[0]);
-  RECON_AND_STORE(dest, in[1]);
-  RECON_AND_STORE(dest, in[2]);
-  RECON_AND_STORE(dest, in[3]);
-  RECON_AND_STORE(dest, in[4]);
-  RECON_AND_STORE(dest, in[5]);
-  RECON_AND_STORE(dest, in[6]);
-  RECON_AND_STORE(dest, in[7]);
-  RECON_AND_STORE(dest, in[8]);
-  RECON_AND_STORE(dest, in[9]);
-  RECON_AND_STORE(dest, in[10]);
-  RECON_AND_STORE(dest, in[11]);
-  RECON_AND_STORE(dest, in[12]);
-  RECON_AND_STORE(dest, in[13]);
-  RECON_AND_STORE(dest, in[14]);
-  RECON_AND_STORE(dest, in[15]);
+  RECON_AND_STORE(dest +  0 * stride, in[0]);
+  RECON_AND_STORE(dest +  1 * stride, in[1]);
+  RECON_AND_STORE(dest +  2 * stride, in[2]);
+  RECON_AND_STORE(dest +  3 * stride, in[3]);
+  RECON_AND_STORE(dest +  4 * stride, in[4]);
+  RECON_AND_STORE(dest +  5 * stride, in[5]);
+  RECON_AND_STORE(dest +  6 * stride, in[6]);
+  RECON_AND_STORE(dest +  7 * stride, in[7]);
+  RECON_AND_STORE(dest +  8 * stride, in[8]);
+  RECON_AND_STORE(dest +  9 * stride, in[9]);
+  RECON_AND_STORE(dest + 10 * stride, in[10]);
+  RECON_AND_STORE(dest + 11 * stride, in[11]);
+  RECON_AND_STORE(dest + 12 * stride, in[12]);
+  RECON_AND_STORE(dest + 13 * stride, in[13]);
+  RECON_AND_STORE(dest + 14 * stride, in[14]);
+  RECON_AND_STORE(dest + 15 * stride, in[15]);
 }