/* * Copyright 2016 Google Inc. * * Use of this source code is governed by a BSD-style license that can be * found in the LICENSE file. */ #include "../jumper/SkJumper.h" #include "SkArenaAlloc.h" #include "SkBlendModePriv.h" #include "SkBlitter.h" #include "SkColor.h" #include "SkColorFilter.h" #include "SkColorSpacePriv.h" #include "SkColorSpaceXformer.h" #include "SkColorSpaceXformSteps.h" #include "SkOpts.h" #include "SkPM4f.h" #include "SkPM4fPriv.h" #include "SkRasterPipeline.h" #include "SkShader.h" #include "SkShaderBase.h" #include "SkTo.h" #include "SkUtils.h" class SkRasterPipelineBlitter final : public SkBlitter { public: // This is our common entrypoint for creating the blitter once we've sorted out shaders. static SkBlitter* Create(const SkPixmap&, const SkPaint&, SkArenaAlloc*, const SkRasterPipeline& shaderPipeline, SkShaderBase::Context*, bool is_opaque, bool is_constant); SkRasterPipelineBlitter(SkPixmap dst, SkBlendMode blend, SkArenaAlloc* alloc, SkShaderBase::Context* burstCtx) : fDst(dst) , fBlend(blend) , fAlloc(alloc) , fBurstCtx(burstCtx) , fColorPipeline(alloc) {} void blitH (int x, int y, int w) override; void blitAntiH (int x, int y, const SkAlpha[], const int16_t[]) override; void blitAntiH2(int x, int y, U8CPU a0, U8CPU a1) override; void blitAntiV2(int x, int y, U8CPU a0, U8CPU a1) override; void blitMask (const SkMask&, const SkIRect& clip) override; void blitRect (int x, int y, int width, int height) override; void blitV (int x, int y, int height, SkAlpha alpha) override; private: void append_load_dst(SkRasterPipeline*) const; void append_store (SkRasterPipeline*) const; // If we have an burst context, use it to fill our shader buffer. void burst_shade(int x, int y, int w); SkPixmap fDst; SkBlendMode fBlend; SkArenaAlloc* fAlloc; SkShaderBase::Context* fBurstCtx; SkRasterPipeline fColorPipeline; SkJumper_MemoryCtx fShaderOutput = {nullptr,0}, // Possibly updated each call to burst_shade(). fDstPtr = {nullptr,0}, // Always points to the top-left of fDst. fMaskPtr = {nullptr,0}; // Updated each call to blitMask(). // We may be able to specialize blitH() or blitRect() into a memset. bool fCanMemsetInBlitRect = false; uint64_t fMemsetColor = 0; // Big enough for largest dst format, F16. // Built lazily on first use. std::function fBlitRect, fBlitAntiH, fBlitMaskA8, fBlitMaskLCD16; // These values are pointed to by the blit pipelines above, // which allows us to adjust them from call to call. float fCurrentCoverage = 0.0f; float fDitherRate = 0.0f; std::vector fShaderBuffer; typedef SkBlitter INHERITED; }; SkBlitter* SkCreateRasterPipelineBlitter(const SkPixmap& dst, const SkPaint& paint, const SkMatrix& ctm, SkArenaAlloc* alloc) { // For legacy/SkColorSpaceXformCanvas to keep working, // we need to sometimes still need to distinguish null dstCS from sRGB. #if 0 SkColorSpace* dstCS = dst.colorSpace() ? dst.colorSpace() : sk_srgb_singleton(); #else SkColorSpace* dstCS = dst.colorSpace(); #endif SkPM4f paintColor = premul_in_dst_colorspace(paint.getColor(), dstCS); auto shader = as_SB(paint.getShader()); SkRasterPipeline_<256> shaderPipeline; if (!shader) { // Having no shader makes things nice and easy... just use the paint color. shaderPipeline.append_constant_color(alloc, paintColor); bool is_opaque = paintColor.a() == 1.0f, is_constant = true; return SkRasterPipelineBlitter::Create(dst, paint, alloc, shaderPipeline, nullptr, is_opaque, is_constant); } bool is_opaque = shader->isOpaque() && paintColor.a() == 1.0f; bool is_constant = shader->isConstant(); // Check whether the shader prefers to run in burst mode. if (auto* burstCtx = shader->makeBurstPipelineContext( SkShaderBase::ContextRec(paint, ctm, nullptr, SkShaderBase::ContextRec::kPM4f_DstType, dstCS), alloc)) { return SkRasterPipelineBlitter::Create(dst, paint, alloc, shaderPipeline, burstCtx, is_opaque, is_constant); } if (shader->appendStages({&shaderPipeline, alloc, dstCS, paint, nullptr, ctm})) { if (paintColor.a() != 1.0f) { shaderPipeline.append(SkRasterPipeline::scale_1_float, alloc->make(paintColor.a())); } return SkRasterPipelineBlitter::Create(dst, paint, alloc, shaderPipeline, nullptr, is_opaque, is_constant); } // The shader has opted out of drawing anything. return alloc->make(); } SkBlitter* SkCreateRasterPipelineBlitter(const SkPixmap& dst, const SkPaint& paint, const SkRasterPipeline& shaderPipeline, bool is_opaque, SkArenaAlloc* alloc) { bool is_constant = false; // If this were the case, it'd be better to just set a paint color. return SkRasterPipelineBlitter::Create(dst, paint, alloc, shaderPipeline, nullptr, is_opaque, is_constant); } SkBlitter* SkRasterPipelineBlitter::Create(const SkPixmap& dst, const SkPaint& paint, SkArenaAlloc* alloc, const SkRasterPipeline& shaderPipeline, SkShaderBase::Context* burstCtx, bool is_opaque, bool is_constant) { auto blitter = alloc->make(dst, paint.getBlendMode(), alloc, burstCtx); // Our job in this factory is to fill out the blitter's color pipeline. // This is the common front of the full blit pipelines, each constructed lazily on first use. // The full blit pipelines handle reading and writing the dst, blending, coverage, dithering. auto colorPipeline = &blitter->fColorPipeline; // Let's get the shader in first. if (burstCtx) { colorPipeline->append(SkRasterPipeline::load_f32, &blitter->fShaderOutput); } else { colorPipeline->extend(shaderPipeline); } // If there's a color filter it comes next. if (auto colorFilter = paint.getColorFilter()) { colorFilter->appendStages(colorPipeline, dst.colorSpace(), alloc, is_opaque); is_opaque = is_opaque && (colorFilter->getFlags() & SkColorFilter::kAlphaUnchanged_Flag); } // Not all formats make sense to dither (think, F16). We set their dither rate // to zero. We need to decide if we're going to dither now to keep is_constant accurate. if (paint.isDither()) { switch (dst.info().colorType()) { default: blitter->fDitherRate = 0.0f; break; case kARGB_4444_SkColorType: blitter->fDitherRate = 1/15.0f; break; case kRGB_565_SkColorType: blitter->fDitherRate = 1/63.0f; break; case kGray_8_SkColorType: case kRGB_888x_SkColorType: case kRGBA_8888_SkColorType: case kBGRA_8888_SkColorType: blitter->fDitherRate = 1/255.0f; break; case kRGB_101010x_SkColorType: case kRGBA_1010102_SkColorType: blitter->fDitherRate = 1/1023.0f; break; } // TODO: for constant colors, we could try to measure the effect of dithering, and if // it has no value (i.e. all variations result in the same 32bit color, then we // could disable it (for speed, by not adding the stage). } is_constant = is_constant && (blitter->fDitherRate == 0.0f); // We're logically done here. The code between here and return blitter is all optimization. // A pipeline that's still constant here can collapse back into a constant color. if (is_constant) { SkPM4f constantColor; SkJumper_MemoryCtx constantColorPtr = { &constantColor, 0 }; colorPipeline->append(SkRasterPipeline::store_f32, &constantColorPtr); colorPipeline->run(0,0,1,1); colorPipeline->reset(); colorPipeline->append_constant_color(alloc, constantColor); is_opaque = constantColor.a() == 1.0f; } // We can strength-reduce SrcOver into Src when opaque. if (is_opaque && blitter->fBlend == SkBlendMode::kSrcOver) { blitter->fBlend = SkBlendMode::kSrc; } // When we're drawing a constant color in Src mode, we can sometimes just memset. // (The previous two optimizations help find more opportunities for this one.) if (is_constant && blitter->fBlend == SkBlendMode::kSrc && blitter->fDst.shiftPerPixel() <= 3 /*TODO: F32*/) { // Run our color pipeline all the way through to produce what we'd memset when we can. // Not all blits can memset, so we need to keep colorPipeline too. SkRasterPipeline_<256> p; p.extend(*colorPipeline); blitter->fDstPtr = SkJumper_MemoryCtx{&blitter->fMemsetColor, 0}; blitter->append_store(&p); p.run(0,0,1,1); blitter->fCanMemsetInBlitRect = true; } blitter->fDstPtr = SkJumper_MemoryCtx{ blitter->fDst.writable_addr(), blitter->fDst.rowBytesAsPixels(), }; return blitter; } void SkRasterPipelineBlitter::append_load_dst(SkRasterPipeline* p) const { const void* ctx = &fDstPtr; switch (fDst.info().colorType()) { default: break; case kGray_8_SkColorType: p->append(SkRasterPipeline::load_g8_dst, ctx); break; case kAlpha_8_SkColorType: p->append(SkRasterPipeline::load_a8_dst, ctx); break; case kRGB_565_SkColorType: p->append(SkRasterPipeline::load_565_dst, ctx); break; case kARGB_4444_SkColorType: p->append(SkRasterPipeline::load_4444_dst, ctx); break; case kBGRA_8888_SkColorType: p->append(SkRasterPipeline::load_bgra_dst, ctx); break; case kRGBA_8888_SkColorType: p->append(SkRasterPipeline::load_8888_dst, ctx); break; case kRGBA_1010102_SkColorType: p->append(SkRasterPipeline::load_1010102_dst, ctx); break; case kRGBA_F16_SkColorType: p->append(SkRasterPipeline::load_f16_dst, ctx); break; case kRGBA_F32_SkColorType: p->append(SkRasterPipeline::load_f32_dst, ctx); break; case kRGB_888x_SkColorType: p->append(SkRasterPipeline::load_8888_dst, ctx); p->append(SkRasterPipeline::force_opaque_dst ); break; case kRGB_101010x_SkColorType: p->append(SkRasterPipeline::load_1010102_dst, ctx); p->append(SkRasterPipeline::force_opaque_dst ); break; } if (fDst.info().alphaType() == kUnpremul_SkAlphaType) { p->append(SkRasterPipeline::premul_dst); } } void SkRasterPipelineBlitter::append_store(SkRasterPipeline* p) const { if (fDst.info().alphaType() == kUnpremul_SkAlphaType) { p->append(SkRasterPipeline::unpremul); } if (fDitherRate > 0.0f) { p->append(SkRasterPipeline::dither, &fDitherRate); } const void* ctx = &fDstPtr; switch (fDst.info().colorType()) { default: break; case kGray_8_SkColorType: p->append(SkRasterPipeline::luminance_to_alpha); p->append(SkRasterPipeline::store_a8, ctx); break; case kAlpha_8_SkColorType: p->append(SkRasterPipeline::store_a8, ctx); break; case kRGB_565_SkColorType: p->append(SkRasterPipeline::store_565, ctx); break; case kARGB_4444_SkColorType: p->append(SkRasterPipeline::store_4444, ctx); break; case kBGRA_8888_SkColorType: p->append(SkRasterPipeline::store_bgra, ctx); break; case kRGBA_8888_SkColorType: p->append(SkRasterPipeline::store_8888, ctx); break; case kRGBA_1010102_SkColorType: p->append(SkRasterPipeline::store_1010102, ctx); break; case kRGBA_F16_SkColorType: p->append(SkRasterPipeline::store_f16, ctx); break; case kRGBA_F32_SkColorType: p->append(SkRasterPipeline::store_f32, ctx); break; case kRGB_888x_SkColorType: p->append(SkRasterPipeline::force_opaque ); p->append(SkRasterPipeline::store_8888, ctx); break; case kRGB_101010x_SkColorType: p->append(SkRasterPipeline::force_opaque ); p->append(SkRasterPipeline::store_1010102, ctx); break; } } void SkRasterPipelineBlitter::burst_shade(int x, int y, int w) { SkASSERT(fBurstCtx); if (w > SkToInt(fShaderBuffer.size())) { fShaderBuffer.resize(w); } fBurstCtx->shadeSpan4f(x,y, fShaderBuffer.data(), w); // We'll be reading from fShaderOutput.pixels + x, so back up by x. fShaderOutput = SkJumper_MemoryCtx{ fShaderBuffer.data() - x, 0 }; } void SkRasterPipelineBlitter::blitH(int x, int y, int w) { this->blitRect(x,y,w,1); } void SkRasterPipelineBlitter::blitRect(int x, int y, int w, int h) { if (fCanMemsetInBlitRect) { for (int ylimit = y+h; y < ylimit; y++) { switch (fDst.shiftPerPixel()) { case 0: memset (fDst.writable_addr8 (x,y), fMemsetColor, w); break; case 1: sk_memset16(fDst.writable_addr16(x,y), fMemsetColor, w); break; case 2: sk_memset32(fDst.writable_addr32(x,y), fMemsetColor, w); break; case 3: sk_memset64(fDst.writable_addr64(x,y), fMemsetColor, w); break; default: SkASSERT(false); break; } } return; } if (!fBlitRect) { SkRasterPipeline p(fAlloc); p.extend(fColorPipeline); if (fBlend == SkBlendMode::kSrcOver && (fDst.info().colorType() == kRGBA_8888_SkColorType || fDst.info().colorType() == kBGRA_8888_SkColorType) && !fDst.colorSpace() && fDst.info().alphaType() != kUnpremul_SkAlphaType && fDitherRate == 0.0f) { auto stage = fDst.info().colorType() == kRGBA_8888_SkColorType ? SkRasterPipeline::srcover_rgba_8888 : SkRasterPipeline::srcover_bgra_8888; p.append(stage, &fDstPtr); } else { if (fBlend != SkBlendMode::kSrc) { this->append_load_dst(&p); SkBlendMode_AppendStages(fBlend, &p); } this->append_store(&p); } fBlitRect = p.compile(); } if (fBurstCtx) { // We can only burst shade one row at a time. for (int ylimit = y+h; y < ylimit; y++) { this->burst_shade(x,y,w); fBlitRect(x,y, w,1); } } else { // If not bursting we can blit the entire rect at once. fBlitRect(x,y,w,h); } } void SkRasterPipelineBlitter::blitAntiH(int x, int y, const SkAlpha aa[], const int16_t runs[]) { if (!fBlitAntiH) { SkRasterPipeline p(fAlloc); p.extend(fColorPipeline); if (SkBlendMode_ShouldPreScaleCoverage(fBlend, /*rgb_coverage=*/false)) { p.append(SkRasterPipeline::scale_1_float, &fCurrentCoverage); this->append_load_dst(&p); SkBlendMode_AppendStages(fBlend, &p); } else { this->append_load_dst(&p); SkBlendMode_AppendStages(fBlend, &p); p.append(SkRasterPipeline::lerp_1_float, &fCurrentCoverage); } this->append_store(&p); fBlitAntiH = p.compile(); } for (int16_t run = *runs; run > 0; run = *runs) { switch (*aa) { case 0x00: break; case 0xff: this->blitH(x,y,run); break; default: fCurrentCoverage = *aa * (1/255.0f); if (fBurstCtx) { this->burst_shade(x,y,run); } fBlitAntiH(x,y,run,1); } x += run; runs += run; aa += run; } } void SkRasterPipelineBlitter::blitAntiH2(int x, int y, U8CPU a0, U8CPU a1) { SkIRect clip = {x,y, x+2,y+1}; uint8_t coverage[] = { (uint8_t)a0, (uint8_t)a1 }; SkMask mask; mask.fImage = coverage; mask.fBounds = clip; mask.fRowBytes = 2; mask.fFormat = SkMask::kA8_Format; this->blitMask(mask, clip); } void SkRasterPipelineBlitter::blitAntiV2(int x, int y, U8CPU a0, U8CPU a1) { SkIRect clip = {x,y, x+1,y+2}; uint8_t coverage[] = { (uint8_t)a0, (uint8_t)a1 }; SkMask mask; mask.fImage = coverage; mask.fBounds = clip; mask.fRowBytes = 1; mask.fFormat = SkMask::kA8_Format; this->blitMask(mask, clip); } void SkRasterPipelineBlitter::blitV(int x, int y, int height, SkAlpha alpha) { SkIRect clip = {x,y, x+1,y+height}; SkMask mask; mask.fImage = α mask.fBounds = clip; mask.fRowBytes = 0; // so we reuse the 1 "row" for all of height mask.fFormat = SkMask::kA8_Format; this->blitMask(mask, clip); } void SkRasterPipelineBlitter::blitMask(const SkMask& mask, const SkIRect& clip) { if (mask.fFormat == SkMask::kBW_Format) { // TODO: native BW masks? return INHERITED::blitMask(mask, clip); } // We'll use the first (A8) plane of any mask and ignore the other two, just like Ganesh. SkMask::Format effectiveMaskFormat = mask.fFormat == SkMask::k3D_Format ? SkMask::kA8_Format : mask.fFormat; // Lazily build whichever pipeline we need, specialized for each mask format. if (effectiveMaskFormat == SkMask::kA8_Format && !fBlitMaskA8) { SkRasterPipeline p(fAlloc); p.extend(fColorPipeline); if (SkBlendMode_ShouldPreScaleCoverage(fBlend, /*rgb_coverage=*/false)) { p.append(SkRasterPipeline::scale_u8, &fMaskPtr); this->append_load_dst(&p); SkBlendMode_AppendStages(fBlend, &p); } else { this->append_load_dst(&p); SkBlendMode_AppendStages(fBlend, &p); p.append(SkRasterPipeline::lerp_u8, &fMaskPtr); } this->append_store(&p); fBlitMaskA8 = p.compile(); } if (effectiveMaskFormat == SkMask::kLCD16_Format && !fBlitMaskLCD16) { SkRasterPipeline p(fAlloc); p.extend(fColorPipeline); if (SkBlendMode_ShouldPreScaleCoverage(fBlend, /*rgb_coverage=*/true)) { // Somewhat unusually, scale_565 needs dst loaded first. this->append_load_dst(&p); p.append(SkRasterPipeline::scale_565, &fMaskPtr); SkBlendMode_AppendStages(fBlend, &p); } else { this->append_load_dst(&p); SkBlendMode_AppendStages(fBlend, &p); p.append(SkRasterPipeline::lerp_565, &fMaskPtr); } this->append_store(&p); fBlitMaskLCD16 = p.compile(); } std::function* blitter = nullptr; // Update fMaskPtr to point "into" this current mask, but lined up with fDstPtr at (0,0). // This sort of trickery upsets UBSAN (pointer-overflow) so we do our math in uintptr_t. // mask.fRowBytes is a uint32_t, which would break our addressing math on 64-bit builds. size_t rowBytes = mask.fRowBytes; switch (effectiveMaskFormat) { case SkMask::kA8_Format: fMaskPtr.stride = rowBytes; fMaskPtr.pixels = (void*)((uintptr_t)mask.fImage - mask.fBounds.left() * (size_t)1 - mask.fBounds.top() * rowBytes); blitter = &fBlitMaskA8; break; case SkMask::kLCD16_Format: fMaskPtr.stride = rowBytes / 2; fMaskPtr.pixels = (void*)((uintptr_t)mask.fImage - mask.fBounds.left() * (size_t)2 - mask.fBounds.top() * rowBytes); blitter = &fBlitMaskLCD16; break; default: return; } SkASSERT(blitter); if (fBurstCtx) { // We can only burst shade one row at a time. int x = clip.left(); for (int y = clip.top(); y < clip.bottom(); y++) { this->burst_shade(x,y,clip.width()); (*blitter)(x,y, clip.width(),1); } } else { // If not bursting we can blit the entire mask at once. (*blitter)(clip.left(),clip.top(), clip.width(),clip.height()); } }