libvpx: Pull from upstream

source/libvpx/vp9/encoder/x86/vp9_dct_sse2.c

 /*
  *  Copyright (c) 2012 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 694 matching lines @@
   in[1] = _mm_slli_epi16(in[1], 2);
   in[2] = _mm_slli_epi16(in[2], 2);
   in[3] = _mm_slli_epi16(in[3], 2);
   in[4] = _mm_slli_epi16(in[4], 2);
   in[5] = _mm_slli_epi16(in[5], 2);
   in[6] = _mm_slli_epi16(in[6], 2);
   in[7] = _mm_slli_epi16(in[7], 2);
 }
 
 // right shift and rounding
-static INLINE void right_shift_8x8(__m128i *res, int const bit) {
-  const __m128i kOne = _mm_set1_epi16(1);
-  const int bit_m02 = bit - 2;
+static INLINE void right_shift_8x8(__m128i *res, const int bit) {
   __m128i sign0 = _mm_srai_epi16(res[0], 15);
   __m128i sign1 = _mm_srai_epi16(res[1], 15);
   __m128i sign2 = _mm_srai_epi16(res[2], 15);
   __m128i sign3 = _mm_srai_epi16(res[3], 15);
   __m128i sign4 = _mm_srai_epi16(res[4], 15);
   __m128i sign5 = _mm_srai_epi16(res[5], 15);
   __m128i sign6 = _mm_srai_epi16(res[6], 15);
   __m128i sign7 = _mm_srai_epi16(res[7], 15);
 
-  if (bit_m02 >= 0) {
-    __m128i k_const_rounding = _mm_slli_epi16(kOne, bit_m02);
-    res[0] = _mm_add_epi16(res[0], k_const_rounding);
-    res[1] = _mm_add_epi16(res[1], k_const_rounding);
-    res[2] = _mm_add_epi16(res[2], k_const_rounding);
-    res[3] = _mm_add_epi16(res[3], k_const_rounding);
-    res[4] = _mm_add_epi16(res[4], k_const_rounding);
-    res[5] = _mm_add_epi16(res[5], k_const_rounding);
-    res[6] = _mm_add_epi16(res[6], k_const_rounding);
-    res[7] = _mm_add_epi16(res[7], k_const_rounding);
+  if (bit == 2) {
+    const __m128i const_rounding = _mm_set1_epi16(1);
+    res[0] = _mm_add_epi16(res[0], const_rounding);
+    res[1] = _mm_add_epi16(res[1], const_rounding);
+    res[2] = _mm_add_epi16(res[2], const_rounding);
+    res[3] = _mm_add_epi16(res[3], const_rounding);
+    res[4] = _mm_add_epi16(res[4], const_rounding);
+    res[5] = _mm_add_epi16(res[5], const_rounding);
+    res[6] = _mm_add_epi16(res[6], const_rounding);
+    res[7] = _mm_add_epi16(res[7], const_rounding);
   }
 
   res[0] = _mm_sub_epi16(res[0], sign0);
   res[1] = _mm_sub_epi16(res[1], sign1);
   res[2] = _mm_sub_epi16(res[2], sign2);
   res[3] = _mm_sub_epi16(res[3], sign3);
   res[4] = _mm_sub_epi16(res[4], sign4);
   res[5] = _mm_sub_epi16(res[5], sign5);
   res[6] = _mm_sub_epi16(res[6], sign6);
   res[7] = _mm_sub_epi16(res[7], sign7);
 
-  res[0] = _mm_srai_epi16(res[0], bit);
-  res[1] = _mm_srai_epi16(res[1], bit);
-  res[2] = _mm_srai_epi16(res[2], bit);
-  res[3] = _mm_srai_epi16(res[3], bit);
-  res[4] = _mm_srai_epi16(res[4], bit);
-  res[5] = _mm_srai_epi16(res[5], bit);
-  res[6] = _mm_srai_epi16(res[6], bit);
-  res[7] = _mm_srai_epi16(res[7], bit);
+  if (bit == 1) {
+    res[0] = _mm_srai_epi16(res[0], 1);
+    res[1] = _mm_srai_epi16(res[1], 1);
+    res[2] = _mm_srai_epi16(res[2], 1);
+    res[3] = _mm_srai_epi16(res[3], 1);
+    res[4] = _mm_srai_epi16(res[4], 1);
+    res[5] = _mm_srai_epi16(res[5], 1);
+    res[6] = _mm_srai_epi16(res[6], 1);
+    res[7] = _mm_srai_epi16(res[7], 1);
+  } else {
+    res[0] = _mm_srai_epi16(res[0], 2);
+    res[1] = _mm_srai_epi16(res[1], 2);
+    res[2] = _mm_srai_epi16(res[2], 2);
+    res[3] = _mm_srai_epi16(res[3], 2);
+    res[4] = _mm_srai_epi16(res[4], 2);
+    res[5] = _mm_srai_epi16(res[5], 2);
+    res[6] = _mm_srai_epi16(res[6], 2);
+    res[7] = _mm_srai_epi16(res[7], 2);
+  }
 }
 
 // write 8x8 array
 static INLINE void write_buffer_8x8(tran_low_t *output, __m128i *res,
                                     int stride) {
   store_output(&res[0], (output + 0 * stride));
   store_output(&res[1], (output + 1 * stride));
   store_output(&res[2], (output + 2 * stride));
   store_output(&res[3], (output + 3 * stride));
   store_output(&res[4], (output + 4 * stride));
@@ skipping 1641 matching lines @@
 
 #define FDCT32x32_2D vp9_highbd_fdct32x32_sse2
 #define FDCT32x32_HIGH_PRECISION 1
 #include "vp9/encoder/x86/vp9_dct32x32_sse2.c" // NOLINT
 #undef  FDCT32x32_2D
 #undef  FDCT32x32_HIGH_PRECISION
 
 #undef  DCT_HIGH_BIT_DEPTH
 
 #endif  // CONFIG_VP9_HIGHBITDEPTH