1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/gfx/skia/trunk/src/opts/SkBlurImage_opts_SSE2.cpp Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,117 @@
1.4 +/*
1.5 + * Copyright 2013 The Android Open Source Project
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license that can be
1.8 + * found in the LICENSE file.
1.9 + */
1.10 +
1.11 +
1.12 +#include "SkBitmap.h"
1.13 +#include "SkColorPriv.h"
1.14 +#include "SkBlurImage_opts_SSE2.h"
1.15 +#include "SkRect.h"
1.16 +
1.17 +#include <emmintrin.h>
1.18 +
1.19 +namespace {
1.20 +
1.21 +enum BlurDirection {
1.22 + kX, kY
1.23 +};
1.24 +
1.25 +/**
1.26 + * Helper function to spread the components of a 32-bit integer into the
1.27 + * lower 8 bits of each 32-bit element of an SSE register.
1.28 + */
1.29 +
1.30 +inline __m128i expand(int a) {
1.31 + const __m128i zero = _mm_setzero_si128();
1.32 +
1.33 + // 0 0 0 0 0 0 0 0 0 0 0 0 A R G B
1.34 + __m128i result = _mm_cvtsi32_si128(a);
1.35 +
1.36 + // 0 0 0 0 0 0 0 0 0 A 0 R 0 G 0 B
1.37 + result = _mm_unpacklo_epi8(result, zero);
1.38 +
1.39 + // 0 0 0 A 0 0 0 R 0 0 0 G 0 0 0 B
1.40 + return _mm_unpacklo_epi16(result, zero);
1.41 +}
1.42 +
1.43 +template<BlurDirection srcDirection, BlurDirection dstDirection>
1.44 +void SkBoxBlur_SSE2(const SkPMColor* src, int srcStride, SkPMColor* dst, int kernelSize,
1.45 + int leftOffset, int rightOffset, int width, int height)
1.46 +{
1.47 + const int rightBorder = SkMin32(rightOffset + 1, width);
1.48 + const int srcStrideX = srcDirection == kX ? 1 : srcStride;
1.49 + const int dstStrideX = dstDirection == kX ? 1 : height;
1.50 + const int srcStrideY = srcDirection == kX ? srcStride : 1;
1.51 + const int dstStrideY = dstDirection == kX ? width : 1;
1.52 + const __m128i scale = _mm_set1_epi32((1 << 24) / kernelSize);
1.53 + const __m128i half = _mm_set1_epi32(1 << 23);
1.54 + const __m128i zero = _mm_setzero_si128();
1.55 + for (int y = 0; y < height; ++y) {
1.56 + __m128i sum = zero;
1.57 + const SkPMColor* p = src;
1.58 + for (int i = 0; i < rightBorder; ++i) {
1.59 + sum = _mm_add_epi32(sum, expand(*p));
1.60 + p += srcStrideX;
1.61 + }
1.62 +
1.63 + const SkPMColor* sptr = src;
1.64 + SkColor* dptr = dst;
1.65 + for (int x = 0; x < width; ++x) {
1.66 +#if 0
1.67 + // In SSE4.1, this would be
1.68 + __m128i result = _mm_mullo_epi32(sum, scale);
1.69 +#else
1.70 + // But SSE2 has no PMULLUD, so we must do AG and RB separately.
1.71 + __m128i tmp1 = _mm_mul_epu32(sum, scale);
1.72 + __m128i tmp2 = _mm_mul_epu32(_mm_srli_si128(sum, 4),
1.73 + _mm_srli_si128(scale, 4));
1.74 + __m128i result = _mm_unpacklo_epi32(_mm_shuffle_epi32(tmp1, _MM_SHUFFLE(0,0,2,0)),
1.75 + _mm_shuffle_epi32(tmp2, _MM_SHUFFLE(0,0,2,0)));
1.76 +#endif
1.77 + // sumA*scale+.5 sumB*scale+.5 sumG*scale+.5 sumB*scale+.5
1.78 + result = _mm_add_epi32(result, half);
1.79 +
1.80 + // 0 0 0 A 0 0 0 R 0 0 0 G 0 0 0 B
1.81 + result = _mm_srli_epi32(result, 24);
1.82 +
1.83 + // 0 0 0 0 0 0 0 0 0 A 0 R 0 G 0 B
1.84 + result = _mm_packs_epi32(result, zero);
1.85 +
1.86 + // 0 0 0 0 0 0 0 0 0 0 0 0 A R G B
1.87 + result = _mm_packus_epi16(result, zero);
1.88 + *dptr = _mm_cvtsi128_si32(result);
1.89 + if (x >= leftOffset) {
1.90 + SkColor l = *(sptr - leftOffset * srcStrideX);
1.91 + sum = _mm_sub_epi32(sum, expand(l));
1.92 + }
1.93 + if (x + rightOffset + 1 < width) {
1.94 + SkColor r = *(sptr + (rightOffset + 1) * srcStrideX);
1.95 + sum = _mm_add_epi32(sum, expand(r));
1.96 + }
1.97 + sptr += srcStrideX;
1.98 + if (srcDirection == kY) {
1.99 + _mm_prefetch(reinterpret_cast<const char*>(sptr + (rightOffset + 1) * srcStrideX),
1.100 + _MM_HINT_T0);
1.101 + }
1.102 + dptr += dstStrideX;
1.103 + }
1.104 + src += srcStrideY;
1.105 + dst += dstStrideY;
1.106 + }
1.107 +}
1.108 +
1.109 +} // namespace
1.110 +
1.111 +bool SkBoxBlurGetPlatformProcs_SSE2(SkBoxBlurProc* boxBlurX,
1.112 + SkBoxBlurProc* boxBlurY,
1.113 + SkBoxBlurProc* boxBlurXY,
1.114 + SkBoxBlurProc* boxBlurYX) {
1.115 + *boxBlurX = SkBoxBlur_SSE2<kX, kX>;
1.116 + *boxBlurY = SkBoxBlur_SSE2<kY, kY>;
1.117 + *boxBlurXY = SkBoxBlur_SSE2<kX, kY>;
1.118 + *boxBlurYX = SkBoxBlur_SSE2<kY, kX>;
1.119 + return true;
1.120 +}
