media/base/simd/vector_math_sse.cc

   1 // Copyright 2013 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 #include "media/base/vector_math_testing.h"
   6
   7 #include <xmmintrin.h>  // NOLINT
   8
   9 namespace media {
  10 namespace vector_math {
  11
  12 void FMUL_SSE(const float src[], float scale, int len, float dest[]) {
  13   const int rem = len % 4;
  14   const int last_index = len - rem;
  15   __m128 m_scale = _mm_set_ps1(scale);
  16   for (int i = 0; i < last_index; i += 4)
  17     _mm_store_ps(dest + i, _mm_mul_ps(_mm_load_ps(src + i), m_scale));
  18
  19   // Handle any remaining values that wouldn't fit in an SSE pass.
  20   for (int i = last_index; i < len; ++i)
  21     dest[i] = src[i] * scale;
  22 }
  23
  24 void FMAC_SSE(const float src[], float scale, int len, float dest[]) {
  25   const int rem = len % 4;
  26   const int last_index = len - rem;
  27   __m128 m_scale = _mm_set_ps1(scale);
  28   for (int i = 0; i < last_index; i += 4) {
  29     _mm_store_ps(dest + i, _mm_add_ps(_mm_load_ps(dest + i),
  30                  _mm_mul_ps(_mm_load_ps(src + i), m_scale)));
  31   }
  32
  33   // Handle any remaining values that wouldn't fit in an SSE pass.
  34   for (int i = last_index; i < len; ++i)
  35     dest[i] += src[i] * scale;
  36 }
  37
  38 }  // namespace vector_math
  39 }  // namespace media