cobalt/media/base/simd/filter_yuv_sse2.cc

// Copyright (c) 2011 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#if defined(_MSC_VER)
#include <intrin.h>
#else
#include <emmintrin.h>
#include <mmintrin.h>
#endif

#include "cobalt/media/base/simd/filter_yuv.h"
#include "starboard/types.h"

namespace cobalt {
namespace media {

void FilterYUVRows_SSE2(uint8_t* dest, const uint8_t* src0, const uint8_t* src1,
                        int width, uint8_t fraction) {
  int pixel = 0;

  // Process the unaligned bytes first.
  int unaligned_width = (16 - (reinterpret_cast<uintptr_t>(dest) & 15)) & 15;
  while (pixel < width && pixel < unaligned_width) {
    dest[pixel] =
        (src0[pixel] * (256 - fraction) + src1[pixel] * fraction) >> 8;
    ++pixel;
  }

  __m128i zero = _mm_setzero_si128();
  __m128i src1_fraction = _mm_set1_epi16(fraction);
  __m128i src0_fraction = _mm_set1_epi16(256 - fraction);
  const __m128i* src0_128 = reinterpret_cast<const __m128i*>(src0 + pixel);
  const __m128i* src1_128 = reinterpret_cast<const __m128i*>(src1 + pixel);
  __m128i* dest128 = reinterpret_cast<__m128i*>(dest + pixel);
  __m128i* end128 = reinterpret_cast<__m128i*>(
      reinterpret_cast<uintptr_t>(dest + width) & ~15);

  while (dest128 < end128) {
    __m128i src0 = _mm_loadu_si128(src0_128);
    __m128i src1 = _mm_loadu_si128(src1_128);
    __m128i src2 = _mm_unpackhi_epi8(src0, zero);
    __m128i src3 = _mm_unpackhi_epi8(src1, zero);
    src0 = _mm_unpacklo_epi8(src0, zero);
    src1 = _mm_unpacklo_epi8(src1, zero);
    src0 = _mm_mullo_epi16(src0, src0_fraction);
    src1 = _mm_mullo_epi16(src1, src1_fraction);
    src2 = _mm_mullo_epi16(src2, src0_fraction);
    src3 = _mm_mullo_epi16(src3, src1_fraction);
    src0 = _mm_add_epi16(src0, src1);
    src2 = _mm_add_epi16(src2, src3);
    src0 = _mm_srli_epi16(src0, 8);
    src2 = _mm_srli_epi16(src2, 8);
    src0 = _mm_packus_epi16(src0, src2);
    *dest128++ = src0;
    ++src0_128;
    ++src1_128;
    pixel += 16;
  }

  while (pixel < width) {
    dest[pixel] =
        (src0[pixel] * (256 - fraction) + src1[pixel] * fraction) >> 8;
    ++pixel;
  }
}

}  // namespace media
}  // namespace cobalt