1// Copyright (c) 2011 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#if defined(_MSC_VER)
6#include <intrin.h>
7#else
8#include <mmintrin.h>
9#include <emmintrin.h>
10#endif
11
12#include "media/base/simd/filter_yuv.h"
13
14namespace media {
15
16void FilterYUVRows_SSE2(uint8* dest,
17                        const uint8* src0,
18                        const uint8* src1,
19                        int width,
20                        int fraction) {
21  int pixel = 0;
22
23  // Process the unaligned bytes first.
24  int unaligned_width =
25      (16 - (reinterpret_cast<uintptr_t>(dest) & 15)) & 15;
26  while (pixel < width && pixel < unaligned_width) {
27    dest[pixel] = (src0[pixel] * (256 - fraction) +
28                   src1[pixel] * fraction) >> 8;
29    ++pixel;
30  }
31
32  __m128i zero = _mm_setzero_si128();
33  __m128i src1_fraction = _mm_set1_epi16(fraction);
34  __m128i src0_fraction = _mm_set1_epi16(256 - fraction);
35  const __m128i* src0_128 =
36      reinterpret_cast<const __m128i*>(src0 + pixel);
37  const __m128i* src1_128 =
38      reinterpret_cast<const __m128i*>(src1 + pixel);
39  __m128i* dest128 = reinterpret_cast<__m128i*>(dest + pixel);
40  __m128i* end128 = reinterpret_cast<__m128i*>(
41      reinterpret_cast<uintptr_t>(dest + width) & ~15);
42
43  while (dest128 < end128) {
44    __m128i src0 = _mm_loadu_si128(src0_128);
45    __m128i src1 = _mm_loadu_si128(src1_128);
46    __m128i src2 = _mm_unpackhi_epi8(src0, zero);
47    __m128i src3 = _mm_unpackhi_epi8(src1, zero);
48    src0 = _mm_unpacklo_epi8(src0, zero);
49    src1 = _mm_unpacklo_epi8(src1, zero);
50    src0 = _mm_mullo_epi16(src0, src0_fraction);
51    src1 = _mm_mullo_epi16(src1, src1_fraction);
52    src2 = _mm_mullo_epi16(src2, src0_fraction);
53    src3 = _mm_mullo_epi16(src3, src1_fraction);
54    src0 = _mm_add_epi16(src0, src1);
55    src2 = _mm_add_epi16(src2, src3);
56    src0 = _mm_srli_epi16(src0, 8);
57    src2 = _mm_srli_epi16(src2, 8);
58    src0 = _mm_packus_epi16(src0, src2);
59    *dest128++ = src0;
60    ++src0_128;
61    ++src1_128;
62    pixel += 16;
63  }
64
65  while (pixel < width) {
66    dest[pixel] = (src0[pixel] * (256 - fraction) +
67                   src1[pixel] * fraction) >> 8;
68    ++pixel;
69  }
70}
71
72}  // namespace media
73