| OLD | NEW |
|---|
| (Empty) |
| 1 /* | |
| 2 * Copyright 2014 The Android Open Source Project | |
| 3 * | |
| 4 * Use of this source code is governed by a BSD-style license that can be | |
| 5 * found in the LICENSE file. | |
| 6 */ | |
| 7 | |
| 8 #include "SkBitmap.h" | |
| 9 #include "SkBlurImage_opts_SSE4.h" | |
| 10 #include "SkColorPriv.h" | |
| 11 #include "SkRect.h" | |
| 12 | |
| 13 /* With the exception of the Android framework we always build the SSE4 function
s | |
| 14 * and enable the caller to determine SSE4 support. However, for the Android fr
amework, | |
| 15 * if the device does not support SSE4x then the compiler will not supply the re
quired | |
| 16 * -msse4* option needed to build this file, so instead we provide a stub implem
entation. | |
| 17 */ | |
| 18 #if !defined(SK_BUILD_FOR_ANDROID_FRAMEWORK) || SK_CPU_SSE_LEVEL >= SK_CPU_SSE_L
EVEL_SSE41 | |
| 19 | |
| 20 #include <smmintrin.h> | |
| 21 | |
| 22 namespace { | |
| 23 enum BlurDirection { | |
| 24 kX, kY | |
| 25 }; | |
| 26 | |
| 27 /* Helper function to spread the components of a 32-bit integer into the | |
| 28 * lower 8 bits of each 32-bit element of an SSE register. | |
| 29 */ | |
| 30 inline __m128i expand(int a) { | |
| 31 const __m128i zero = _mm_setzero_si128(); | |
| 32 | |
| 33 // 0 0 0 0 0 0 0 0 0 0 0 0 A R G B | |
| 34 __m128i result = _mm_cvtsi32_si128(a); | |
| 35 | |
| 36 // 0 0 0 0 0 0 0 0 0 A 0 R 0 G 0 B | |
| 37 result = _mm_unpacklo_epi8(result, zero); | |
| 38 | |
| 39 // 0 0 0 A 0 0 0 R 0 0 0 G 0 0 0 B | |
| 40 return _mm_unpacklo_epi16(result, zero); | |
| 41 } | |
| 42 | |
| 43 template<BlurDirection srcDirection, BlurDirection dstDirection> | |
| 44 void SkBoxBlur_SSE4(const SkPMColor* src, int srcStride, SkPMColor* dst, int ker
nelSize, | |
| 45 int leftOffset, int rightOffset, int width, int height) | |
| 46 { | |
| 47 const int rightBorder = SkMin32(rightOffset + 1, width); | |
| 48 const int srcStrideX = srcDirection == kX ? 1 : srcStride; | |
| 49 const int dstStrideX = dstDirection == kX ? 1 : height; | |
| 50 const int srcStrideY = srcDirection == kX ? srcStride : 1; | |
| 51 const int dstStrideY = dstDirection == kX ? width : 1; | |
| 52 const __m128i scale = _mm_set1_epi32((1 << 24) / kernelSize); | |
| 53 const __m128i half = _mm_set1_epi32(1 << 23); | |
| 54 const __m128i zero = _mm_setzero_si128(); | |
| 55 for (int y = 0; y < height; ++y) { | |
| 56 __m128i sum = zero; | |
| 57 const SkPMColor* p = src; | |
| 58 for (int i = 0; i < rightBorder; ++i) { | |
| 59 sum = _mm_add_epi32(sum, expand(*p)); | |
| 60 p += srcStrideX; | |
| 61 } | |
| 62 | |
| 63 const SkPMColor* sptr = src; | |
| 64 SkColor* dptr = dst; | |
| 65 for (int x = 0; x < width; ++x) { | |
| 66 __m128i result = _mm_mullo_epi32(sum, scale); | |
| 67 | |
| 68 // sumA*scale+.5 sumB*scale+.5 sumG*scale+.5 sumB*scale+.5 | |
| 69 result = _mm_add_epi32(result, half); | |
| 70 | |
| 71 // 0 0 0 A 0 0 0 R 0 0 0 G 0 0 0 B | |
| 72 result = _mm_srli_epi32(result, 24); | |
| 73 | |
| 74 // 0 0 0 0 0 0 0 0 0 A 0 R 0 G 0 B | |
| 75 result = _mm_packs_epi32(result, zero); | |
| 76 | |
| 77 // 0 0 0 0 0 0 0 0 0 0 0 0 A R G B | |
| 78 result = _mm_packus_epi16(result, zero); | |
| 79 *dptr = _mm_cvtsi128_si32(result); | |
| 80 if (x >= leftOffset) { | |
| 81 SkColor l = *(sptr - leftOffset * srcStrideX); | |
| 82 sum = _mm_sub_epi32(sum, expand(l)); | |
| 83 } | |
| 84 if (x + rightOffset + 1 < width) { | |
| 85 SkColor r = *(sptr + (rightOffset + 1) * srcStrideX); | |
| 86 sum = _mm_add_epi32(sum, expand(r)); | |
| 87 } | |
| 88 sptr += srcStrideX; | |
| 89 if (srcDirection == kY) { | |
| 90 _mm_prefetch(reinterpret_cast<const char*>(sptr + (rightOffset +
1) * srcStrideX), | |
| 91 _MM_HINT_T0); | |
| 92 } | |
| 93 dptr += dstStrideX; | |
| 94 } | |
| 95 src += srcStrideY; | |
| 96 dst += dstStrideY; | |
| 97 } | |
| 98 } | |
| 99 | |
| 100 } // namespace | |
| 101 | |
| 102 bool SkBoxBlurGetPlatformProcs_SSE4(SkBoxBlurProc* boxBlurX, | |
| 103 SkBoxBlurProc* boxBlurY, | |
| 104 SkBoxBlurProc* boxBlurXY, | |
| 105 SkBoxBlurProc* boxBlurYX) { | |
| 106 *boxBlurX = SkBoxBlur_SSE4<kX, kX>; | |
| 107 *boxBlurY = SkBoxBlur_SSE4<kY, kY>; | |
| 108 *boxBlurXY = SkBoxBlur_SSE4<kX, kY>; | |
| 109 *boxBlurYX = SkBoxBlur_SSE4<kY, kX>; | |
| 110 return true; | |
| 111 } | |
| 112 | |
| 113 #else // !defined(SK_BUILD_FOR_ANDROID_FRAMEWORK) || SK_CPU_SSE_LEVEL >= SK_CPU_
SSE_LEVEL_SSE41 | |
| 114 | |
| 115 bool SkBoxBlurGetPlatformProcs_SSE4(SkBoxBlurProc* boxBlurX, | |
| 116 SkBoxBlurProc* boxBlurY, | |
| 117 SkBoxBlurProc* boxBlurXY, | |
| 118 SkBoxBlurProc* boxBlurYX) { | |
| 119 sk_throw(); | |
| 120 } | |
| 121 | |
| 122 | |
| 123 #endif | |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 375503003:
Revert of Add SSE4 version of BlurImage optimizations. (Closed)
Base URL: https://skia.googlesource.com/skia.git@master