diff options
author | Brian Osman <brianosman@google.com> | 2017-08-30 10:02:10 -0400 |
---|---|---|
committer | Skia Commit-Bot <skia-commit-bot@chromium.org> | 2017-08-30 15:27:42 +0000 |
commit | f9810666bd40db8fb1650e6c727c1a83b8090136 (patch) | |
tree | 0d346bbb44d61c7da4c53655620e890f60758ffc /src/gpu/GrSoftwarePathRenderer.cpp | |
parent | e750391c349ff1fa1c179fbf2f5af27fb4405cef (diff) |
Threaded generation of software paths
Re-land of: https://skia-review.googlesource.com/36560
All information needed by the thread is captured by the prepare
callback object, the lambda captures a pointer to that, and does the
mask render. Once it's done, it signals the semaphore (also owned by the
callback). The callback defers the semaphore wait even longer (into the
ASAP upload), so the odds of waiting for the thread are REALLY low.
Also did a bunch of cleanup along the way, and put in some trace markers
so we can monitor how well this is working.
Traces of a GM that includes GPU and SW path rendering (path-reverse):
Original:
https://screenshot.googleplex.com/f5BG3901tQg.png
Threaded, with wait in the callback (notice pre flush callback blocking):
https://screenshot.googleplex.com/htOSZFE2s04.png
Current version, with wait deferred to ASAP upload function:
https://screenshot.googleplex.com/GHjD0U3C34q.png
Bug: skia:
Change-Id: Idb92f385590749f41328a9aec65b2a93f4775079
Reviewed-on: https://skia-review.googlesource.com/40775
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Diffstat (limited to 'src/gpu/GrSoftwarePathRenderer.cpp')
-rw-r--r-- | src/gpu/GrSoftwarePathRenderer.cpp | 135 |
1 files changed, 121 insertions, 14 deletions
diff --git a/src/gpu/GrSoftwarePathRenderer.cpp b/src/gpu/GrSoftwarePathRenderer.cpp index 27ba8e6436..e76af89933 100644 --- a/src/gpu/GrSoftwarePathRenderer.cpp +++ b/src/gpu/GrSoftwarePathRenderer.cpp @@ -8,9 +8,16 @@ #include "GrSoftwarePathRenderer.h" #include "GrAuditTrail.h" #include "GrClip.h" +#include "GrContextPriv.h" #include "GrGpuResourcePriv.h" +#include "GrOpFlushState.h" +#include "GrOpList.h" #include "GrResourceProvider.h" #include "GrSWMaskHelper.h" +#include "SkMakeUnique.h" +#include "SkSemaphore.h" +#include "SkTaskGroup.h" +#include "SkTraceEvent.h" #include "ops/GrDrawOp.h" #include "ops/GrRectOpFactory.h" @@ -145,10 +152,84 @@ void GrSoftwarePathRenderer::DrawToTargetWithShapeMask( maskMatrix.preConcat(viewMatrix); paint.addCoverageFragmentProcessor(GrSimpleTextureEffect::Make( std::move(proxy), nullptr, maskMatrix, GrSamplerParams::kNone_FilterMode)); - renderTargetContext->addDrawOp(clip, - GrRectOpFactory::MakeNonAAFillWithLocalMatrix( - std::move(paint), SkMatrix::I(), invert, dstRect, - GrAAType::kNone, &userStencilSettings)); + DrawNonAARect(renderTargetContext, std::move(paint), userStencilSettings, clip, SkMatrix::I(), + dstRect, invert); +} + +static sk_sp<GrTextureProxy> make_deferred_mask_texture_proxy(GrContext* context, SkBackingFit fit, + int width, int height) { + GrSurfaceDesc desc; + desc.fOrigin = kTopLeft_GrSurfaceOrigin; + desc.fWidth = width; + desc.fHeight = height; + desc.fConfig = kAlpha_8_GrPixelConfig; + + sk_sp<GrSurfaceContext> sContext = + context->contextPriv().makeDeferredSurfaceContext(desc, fit, SkBudgeted::kYes); + if (!sContext || !sContext->asTextureProxy()) { + return nullptr; + } + return sContext->asTextureProxyRef(); +} + +namespace { + +class GrMaskUploaderPrepareCallback : public GrPrepareCallback { +public: + GrMaskUploaderPrepareCallback(sk_sp<GrTextureProxy> proxy, const SkIRect& maskBounds, + const SkMatrix& viewMatrix, const GrShape& shape, GrAA aa) + : fProxy(std::move(proxy)) + , fMaskBounds(maskBounds) + , fViewMatrix(viewMatrix) + , fShape(shape) + , fAA(aa) + , fWaited(false) {} + + ~GrMaskUploaderPrepareCallback() override { + if (!fWaited) { + // This can happen if our owning op list fails to instantiate (so it never prepares) + fPixelsReady.wait(); + } + } + + void operator()(GrOpFlushState* flushState) override { + TRACE_EVENT0("skia", "Mask Uploader Pre Flush Callback"); + auto uploadMask = [this](GrDrawOp::WritePixelsFn& writePixelsFn) { + TRACE_EVENT0("skia", "Mask Upload"); + this->fPixelsReady.wait(); + this->fWaited = true; + // If the worker thread was unable to allocate pixels, this check will fail, and we'll + // end up drawing with an uninitialized mask texture, but at least we won't crash. + if (this->fPixels.addr()) { + writePixelsFn(this->fProxy.get(), 0, 0, + this->fPixels.width(), this->fPixels.height(), + kAlpha_8_GrPixelConfig, + this->fPixels.addr(), this->fPixels.rowBytes()); + } + }; + flushState->addASAPUpload(std::move(uploadMask)); + } + + SkAutoPixmapStorage* getPixels() { return &fPixels; } + SkSemaphore* getSemaphore() { return &fPixelsReady; } + const SkIRect& getMaskBounds() const { return fMaskBounds; } + const SkMatrix* getViewMatrix() const { return &fViewMatrix; } + const GrShape& getShape() const { return fShape; } + GrAA getAA() const { return fAA; } + +private: + // NOTE: This ref cnt isn't thread safe! + sk_sp<GrTextureProxy> fProxy; + SkAutoPixmapStorage fPixels; + SkSemaphore fPixelsReady; + + SkIRect fMaskBounds; + SkMatrix fViewMatrix; + GrShape fShape; + GrAA fAA; + bool fWaited; +}; + } //////////////////////////////////////////////////////////////////////////////// @@ -205,11 +286,6 @@ bool GrSoftwarePathRenderer::onDrawPath(const DrawPathArgs& args) { } GrUniqueKey maskKey; - struct KeyData { - SkScalar fFractionalTranslateX; - SkScalar fFractionalTranslateY; - }; - if (useCache) { // We require the upper left 2x2 of the matrix to match exactly for a cache hit. SkScalar sx = args.fViewMatrix->get(SkMatrix::kMScaleX); @@ -240,8 +316,6 @@ bool GrSoftwarePathRenderer::onDrawPath(const DrawPathArgs& args) { builder[4] = fracX | (fracY >> 8); args.fShape->writeUnstyledKey(&builder[5]); #endif - // FIXME: Doesn't the key need to consider whether we're using AA or not? In practice that - // should always be true, though. } sk_sp<GrTextureProxy> proxy; @@ -251,9 +325,42 @@ bool GrSoftwarePathRenderer::onDrawPath(const DrawPathArgs& args) { if (!proxy) { SkBackingFit fit = useCache ? SkBackingFit::kExact : SkBackingFit::kApprox; GrAA aa = GrAAType::kCoverage == args.fAAType ? GrAA::kYes : GrAA::kNo; - proxy = GrSWMaskHelper::DrawShapeMaskToTexture(args.fContext, *args.fShape, - *boundsForMask, aa, - fit, args.fViewMatrix); + + SkTaskGroup* taskGroup = args.fContext->contextPriv().getTaskGroup(); + if (taskGroup) { + proxy = make_deferred_mask_texture_proxy(args.fContext, fit, + boundsForMask->width(), + boundsForMask->height()); + if (!proxy) { + return false; + } + + auto uploader = skstd::make_unique<GrMaskUploaderPrepareCallback>( + proxy, *boundsForMask, *args.fViewMatrix, *args.fShape, aa); + GrMaskUploaderPrepareCallback* uploaderRaw = uploader.get(); + + auto drawAndUploadMask = [uploaderRaw] { + TRACE_EVENT0("skia", "Threaded SW Mask Render"); + GrSWMaskHelper helper(uploaderRaw->getPixels()); + if (helper.init(uploaderRaw->getMaskBounds(), uploaderRaw->getViewMatrix())) { + helper.drawShape(uploaderRaw->getShape(), SkRegion::kReplace_Op, + uploaderRaw->getAA(), 0xFF); + } else { + SkDEBUGFAIL("Unable to allocate SW mask."); + } + uploaderRaw->getSemaphore()->signal(); + }; + taskGroup->add(std::move(drawAndUploadMask)); + args.fRenderTargetContext->getOpList()->addPrepareCallback(std::move(uploader)); + } else { + GrSWMaskHelper helper; + if (!helper.init(*boundsForMask, args.fViewMatrix)) { + return false; + } + helper.drawShape(*args.fShape, SkRegion::kReplace_Op, aa, 0xFF); + proxy = helper.toTextureProxy(args.fContext, fit); + } + if (!proxy) { return false; } |