diff options
-rw-r--r-- | bench/BlurImageFilterBench.cpp | 4 | ||||
-rw-r--r-- | gyp/opts.gyp | 3 | ||||
-rw-r--r-- | src/effects/SkBlurImageFilter.cpp | 28 | ||||
-rw-r--r-- | src/opts/SkBlurImage_opts.h | 19 | ||||
-rw-r--r-- | src/opts/SkBlurImage_opts_SSE2.cpp | 115 | ||||
-rw-r--r-- | src/opts/SkBlurImage_opts_SSE2.h | 12 | ||||
-rw-r--r-- | src/opts/SkBlurImage_opts_none.cpp | 14 | ||||
-rw-r--r-- | src/opts/opts_check_SSE2.cpp | 14 |
8 files changed, 189 insertions, 20 deletions
diff --git a/bench/BlurImageFilterBench.cpp b/bench/BlurImageFilterBench.cpp index f6bc0e4ccc..dc3edb03d7 100644 --- a/bench/BlurImageFilterBench.cpp +++ b/bench/BlurImageFilterBench.cpp @@ -84,6 +84,10 @@ private: typedef SkBenchmark INHERITED; }; +DEF_BENCH(return new BlurImageFilterBench(BLUR_SIGMA_LARGE, 0, false);) +DEF_BENCH(return new BlurImageFilterBench(BLUR_SIGMA_SMALL, 0, false);) +DEF_BENCH(return new BlurImageFilterBench(0, BLUR_SIGMA_LARGE, false);) +DEF_BENCH(return new BlurImageFilterBench(0, BLUR_SIGMA_SMALL, false);) DEF_BENCH(return new BlurImageFilterBench(BLUR_SIGMA_SMALL, BLUR_SIGMA_SMALL, true);) DEF_BENCH(return new BlurImageFilterBench(BLUR_SIGMA_SMALL, BLUR_SIGMA_SMALL, false);) DEF_BENCH(return new BlurImageFilterBench(BLUR_SIGMA_LARGE, BLUR_SIGMA_LARGE, true);) diff --git a/gyp/opts.gyp b/gyp/opts.gyp index 6e6af474ed..4fc0f34ebf 100644 --- a/gyp/opts.gyp +++ b/gyp/opts.gyp @@ -51,6 +51,7 @@ '../src/opts/SkBitmapFilter_opts_SSE2.cpp', '../src/opts/SkBlitRow_opts_SSE2.cpp', '../src/opts/SkBlitRect_opts_SSE2.cpp', + '../src/opts/SkBlurImage_opts_SSE2.cpp', '../src/opts/SkMorphology_opts_SSE2.cpp', '../src/opts/SkUtils_opts_SSE2.cpp', '../src/opts/SkXfermode_opts_none.cpp', @@ -78,6 +79,7 @@ '../src/opts/SkBlitMask_opts_arm.cpp', '../src/opts/SkBlitRow_opts_arm.cpp', '../src/opts/SkBlitRow_opts_arm.h', + '../src/opts/SkBlurImage_opts_none.cpp', '../src/opts/SkMorphology_opts_none.cpp', '../src/opts/SkXfermode_opts_arm.cpp', ], @@ -102,6 +104,7 @@ '../src/opts/SkBitmapProcState_opts_none.cpp', '../src/opts/SkBlitMask_opts_none.cpp', '../src/opts/SkBlitRow_opts_none.cpp', + '../src/opts/SkBlurImage_opts_none.cpp', '../src/opts/SkMorphology_opts_none.cpp', '../src/opts/SkUtils_opts_none.cpp', '../src/opts/SkXfermode_opts_none.cpp', diff --git a/src/effects/SkBlurImageFilter.cpp b/src/effects/SkBlurImageFilter.cpp index c04c258829..852c0afad3 100644 --- a/src/effects/SkBlurImageFilter.cpp +++ b/src/effects/SkBlurImageFilter.cpp @@ -10,6 +10,7 @@ #include "SkColorPriv.h" #include "SkFlattenableBuffers.h" #include "SkGpuBlurUtils.h" +#include "SkBlurImage_opts.h" #if SK_SUPPORT_GPU #include "GrContext.h" #include "SkImageFilterUtils.h" @@ -125,26 +126,6 @@ static void boxBlur(const SkPMColor* src, int srcStride, SkPMColor* dst, int ker } } -static void boxBlurX(const SkPMColor* src, int srcStride, SkPMColor* dst, int kernelSize, - int leftOffset, int rightOffset, int width, int height) -{ - boxBlur<kX, kX>(src, srcStride, dst, kernelSize, leftOffset, rightOffset, width, height); -} - -#ifndef SK_DISABLE_BLUR_DIVISION_OPTIMIZATION -static void boxBlurXY(const SkPMColor* src, int srcStride, SkPMColor* dst, int kernelSize, - int leftOffset, int rightOffset, int width, int height) -{ - boxBlur<kX, kY>(src, srcStride, dst, kernelSize, leftOffset, rightOffset, width, height); -} -#endif - -static void boxBlurY(const SkPMColor* src, int srcStride, SkPMColor* dst, int kernelSize, - int topOffset, int bottomOffset, int width, int height) -{ - boxBlur<kY, kY>(src, srcStride, dst, kernelSize, topOffset, bottomOffset, width, height); -} - static void getBox3Params(SkScalar s, int *kernelSize, int* kernelSize3, int *lowOffset, int *highOffset) { @@ -212,6 +193,13 @@ bool SkBlurImageFilter::onFilterImage(Proxy* proxy, SkPMColor* d = dst->getAddr32(0, 0); int w = dstBounds.width(), h = dstBounds.height(); int sw = src.rowBytesAsPixels(); + SkBoxBlurProc boxBlurX, boxBlurY, boxBlurXY; + if (!SkBoxBlurGetPlatformProcs(&boxBlurX, &boxBlurY, &boxBlurXY)) { + boxBlurX = boxBlur<kX, kX>; + boxBlurY = boxBlur<kY, kY>; + boxBlurXY = boxBlur<kX, kY>; + } + if (kernelSizeX > 0 && kernelSizeY > 0) { #ifndef SK_DISABLE_BLUR_DIVISION_OPTIMIZATION boxBlurX(s, sw, t, kernelSizeX, lowOffsetX, highOffsetX, w, h); diff --git a/src/opts/SkBlurImage_opts.h b/src/opts/SkBlurImage_opts.h new file mode 100644 index 0000000000..a8fb209353 --- /dev/null +++ b/src/opts/SkBlurImage_opts.h @@ -0,0 +1,19 @@ +/* + * Copyright 2013 The Android Open Source Project + * + * Use of this source code is governed by a BSD-style license that can be + * found in the LICENSE file. + */ + +#ifndef SkBlurImage_opts_DEFINED +#define SkBlurImage_opts_DEFINED + +#include "SkColorPriv.h" + +typedef void (*SkBoxBlurProc)(const SkPMColor* src, int srcStride, SkPMColor* dst, int kernelSize, + int leftOffset, int rightOffset, int width, int height); + +bool SkBoxBlurGetPlatformProcs(SkBoxBlurProc* boxBlurX, + SkBoxBlurProc* boxBlurY, + SkBoxBlurProc* boxBlurXY); +#endif diff --git a/src/opts/SkBlurImage_opts_SSE2.cpp b/src/opts/SkBlurImage_opts_SSE2.cpp new file mode 100644 index 0000000000..97cd2fdb59 --- /dev/null +++ b/src/opts/SkBlurImage_opts_SSE2.cpp @@ -0,0 +1,115 @@ +/* + * Copyright 2013 The Android Open Source Project + * + * Use of this source code is governed by a BSD-style license that can be + * found in the LICENSE file. + */ + + +#include "SkBitmap.h" +#include "SkColorPriv.h" +#include "SkBlurImage_opts_SSE2.h" +#include "SkRect.h" + +#include <emmintrin.h> + +namespace { + +enum BlurDirection { + kX, kY +}; + +/** + * Helper function to spread the components of a 32-bit integer into the + * lower 8 bits of each 32-bit element of an SSE register. + */ + +inline __m128i expand(int a) { + const __m128i zero = _mm_setzero_si128(); + + // 0 0 0 0 0 0 0 0 0 0 0 0 A R G B + __m128i result = _mm_cvtsi32_si128(a); + + // 0 0 0 0 0 0 0 0 0 A 0 R 0 G 0 B + result = _mm_unpacklo_epi8(result, zero); + + // 0 0 0 A 0 0 0 R 0 0 0 G 0 0 0 B + return _mm_unpacklo_epi16(result, zero); +} + +template<BlurDirection srcDirection, BlurDirection dstDirection> +void SkBoxBlur_SSE2(const SkPMColor* src, int srcStride, SkPMColor* dst, int kernelSize, + int leftOffset, int rightOffset, int width, int height) +{ + const int rightBorder = SkMin32(rightOffset + 1, width); + const int srcStrideX = srcDirection == kX ? 1 : srcStride; + const int dstStrideX = dstDirection == kX ? 1 : height; + const int srcStrideY = srcDirection == kX ? srcStride : 1; + const int dstStrideY = dstDirection == kX ? width : 1; + const __m128i scale = _mm_set1_epi32((1 << 24) / kernelSize); + const __m128i half = _mm_set1_epi32(1 << 23); + const __m128i zero = _mm_setzero_si128(); + for (int y = 0; y < height; ++y) { + __m128i sum = zero; + const SkPMColor* p = src; + for (int i = 0; i < rightBorder; ++i) { + sum = _mm_add_epi32(sum, expand(*p)); + p += srcStrideX; + } + + const SkPMColor* sptr = src; + SkColor* dptr = dst; + for (int x = 0; x < width; ++x) { +#if 0 + // In SSE4.1, this would be + __m128i result = _mm_mullo_epi32(sum, scale); +#else + // But SSE2 has no PMULLUD, so we must do AG and RB separately. + __m128i tmp1 = _mm_mul_epu32(sum, scale); + __m128i tmp2 = _mm_mul_epu32(_mm_srli_si128(sum, 4), + _mm_srli_si128(scale, 4)); + __m128i result = _mm_unpacklo_epi32(_mm_shuffle_epi32(tmp1, _MM_SHUFFLE(0,0,2,0)), + _mm_shuffle_epi32(tmp2, _MM_SHUFFLE(0,0,2,0))); +#endif + // sumA*scale+.5 sumB*scale+.5 sumG*scale+.5 sumB*scale+.5 + result = _mm_add_epi32(result, half); + + // 0 0 0 A 0 0 0 R 0 0 0 G 0 0 0 B + result = _mm_srli_epi32(result, 24); + + // 0 0 0 0 0 0 0 0 0 A 0 R 0 G 0 B + result = _mm_packs_epi32(result, zero); + + // 0 0 0 0 0 0 0 0 0 0 0 0 A R G B + result = _mm_packus_epi16(result, zero); + *dptr = _mm_cvtsi128_si32(result); + if (x >= leftOffset) { + SkColor l = *(sptr - leftOffset * srcStrideX); + sum = _mm_sub_epi32(sum, expand(l)); + } + if (x + rightOffset + 1 < width) { + SkColor r = *(sptr + (rightOffset + 1) * srcStrideX); + sum = _mm_add_epi32(sum, expand(r)); + } + sptr += srcStrideX; + if (srcDirection == kY) { + _mm_prefetch(reinterpret_cast<const char*>(sptr + (rightOffset + 1) * srcStrideX), + _MM_HINT_T0); + } + dptr += dstStrideX; + } + src += srcStrideY; + dst += dstStrideY; + } +} + +} // namespace + +bool SkBoxBlurGetPlatformProcs_SSE2(SkBoxBlurProc* boxBlurX, + SkBoxBlurProc* boxBlurY, + SkBoxBlurProc* boxBlurXY) { + *boxBlurX = SkBoxBlur_SSE2<kX, kX>; + *boxBlurY = SkBoxBlur_SSE2<kY, kY>; + *boxBlurXY = SkBoxBlur_SSE2<kX, kY>; + return true; +} diff --git a/src/opts/SkBlurImage_opts_SSE2.h b/src/opts/SkBlurImage_opts_SSE2.h new file mode 100644 index 0000000000..4dd804aa1a --- /dev/null +++ b/src/opts/SkBlurImage_opts_SSE2.h @@ -0,0 +1,12 @@ +/* + * Copyright 2013 The Android Open Source Project + * + * Use of this source code is governed by a BSD-style license that can be + * found in the LICENSE file. + */ + +#include "SkBlurImage_opts.h" + +bool SkBoxBlurGetPlatformProcs_SSE2(SkBoxBlurProc* boxBlurX, + SkBoxBlurProc* boxBlurY, + SkBoxBlurProc* boxBlurXY); diff --git a/src/opts/SkBlurImage_opts_none.cpp b/src/opts/SkBlurImage_opts_none.cpp new file mode 100644 index 0000000000..dbdac8a950 --- /dev/null +++ b/src/opts/SkBlurImage_opts_none.cpp @@ -0,0 +1,14 @@ +/* + * Copyright 2013 The Android Open Source Project + * + * Use of this source code is governed by a BSD-style license that can be + * found in the LICENSE file. + */ + +#include "SkBlurImage_opts.h" + +bool SkBoxBlurGetPlatformProcs(SkBoxBlurProc* boxBlurX, + SkBoxBlurProc* boxBlurY, + SkBoxBlurProc* boxBlurXY) { + return false; +} diff --git a/src/opts/opts_check_SSE2.cpp b/src/opts/opts_check_SSE2.cpp index b40ca9d4b8..efbdbc1850 100644 --- a/src/opts/opts_check_SSE2.cpp +++ b/src/opts/opts_check_SSE2.cpp @@ -12,6 +12,7 @@ #include "SkBlitRow.h" #include "SkBlitRect_opts_SSE2.h" #include "SkBlitRow_opts_SSE2.h" +#include "SkBlurImage_opts_SSE2.h" #include "SkUtils_opts_SSE2.h" #include "SkUtils.h" #include "SkMorphology_opts.h" @@ -268,6 +269,19 @@ SkMorphologyProc SkMorphologyGetPlatformProc(SkMorphologyProcType type) { } } +bool SkBoxBlurGetPlatformProcs(SkBoxBlurProc* boxBlurX, + SkBoxBlurProc* boxBlurY, + SkBoxBlurProc* boxBlurXY) { +#ifdef SK_DISABLE_BLUR_DIVISION_OPTIMIZATION + return false; +#else + if (!cachedHasSSE2()) { + return false; + } + return SkBoxBlurGetPlatformProcs_SSE2(boxBlurX, boxBlurY, boxBlurXY); +#endif +} + SkBlitRow::ColorRectProc PlatformColorRectProcFactory(); // suppress warning SkBlitRow::ColorRectProc PlatformColorRectProcFactory() { |