src/gpu/GrContext.cpp

/*
 * Copyright 2011 Google Inc.
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */


#include "GrContext.h"

#include "effects/GrSingleTextureEffect.h"
#include "effects/GrConfigConversionEffect.h"

#include "GrAARectRenderer.h"
#include "GrBufferAllocPool.h"
#include "GrGpu.h"
#include "GrDrawTargetCaps.h"
#include "GrIndexBuffer.h"
#include "GrInOrderDrawBuffer.h"
#include "GrOvalRenderer.h"
#include "GrPathRenderer.h"
#include "GrPathUtils.h"
#include "GrResourceCache.h"
#include "GrSoftwarePathRenderer.h"
#include "GrStencilBuffer.h"
#include "GrTextStrike.h"
#include "SkRTConf.h"
#include "SkRRect.h"
#include "SkStrokeRec.h"
#include "SkTLazy.h"
#include "SkTLS.h"
#include "SkTrace.h"

// It can be useful to set this to false to test whether a bug is caused by using the
// InOrderDrawBuffer, to compare performance of using/not using InOrderDrawBuffer, or to make
// debugging simpler.
SK_CONF_DECLARE(bool, c_Defer, "gpu.deferContext", true,
                "Defers rendering in GrContext via GrInOrderDrawBuffer.");

#define BUFFERED_DRAW (c_Defer ? kYes_BufferedDraw : kNo_BufferedDraw)

#ifdef SK_DEBUG
    // change this to a 1 to see notifications when partial coverage fails
    #define GR_DEBUG_PARTIAL_COVERAGE_CHECK 0
#else
    #define GR_DEBUG_PARTIAL_COVERAGE_CHECK 0
#endif

static const size_t MAX_RESOURCE_CACHE_COUNT = GR_DEFAULT_RESOURCE_CACHE_COUNT_LIMIT;
static const size_t MAX_RESOURCE_CACHE_BYTES = GR_DEFAULT_RESOURCE_CACHE_MB_LIMIT * 1024 * 1024;

static const size_t DRAW_BUFFER_VBPOOL_BUFFER_SIZE = 1 << 15;
static const int DRAW_BUFFER_VBPOOL_PREALLOC_BUFFERS = 4;

static const size_t DRAW_BUFFER_IBPOOL_BUFFER_SIZE = 1 << 11;
static const int DRAW_BUFFER_IBPOOL_PREALLOC_BUFFERS = 4;

#define ASSERT_OWNED_RESOURCE(R) SkASSERT(!(R) || (R)->getContext() == this)

// Glorified typedef to avoid including GrDrawState.h in GrContext.h
class GrContext::AutoRestoreEffects : public GrDrawState::AutoRestoreEffects {};

class GrContext::AutoCheckFlush {
public:
    AutoCheckFlush(GrContext* context) : fContext(context) { SkASSERT(NULL != context); }

    ~AutoCheckFlush() {
        if (fContext->fFlushToReduceCacheSize) {
            fContext->flush();
        }
    }

private:
    GrContext* fContext;
};

GrContext* GrContext::Create(GrBackend backend, GrBackendContext backendContext) {
    GrContext* context = SkNEW(GrContext);
    if (context->init(backend, backendContext)) {
        return context;
    } else {
        context->unref();
        return NULL;
    }
}

GrContext::GrContext() {
    fDrawState = NULL;
    fGpu = NULL;
    fClip = NULL;
    fPathRendererChain = NULL;
    fSoftwarePathRenderer = NULL;
    fTextureCache = NULL;
    fFontCache = NULL;
    fDrawBuffer = NULL;
    fDrawBufferVBAllocPool = NULL;
    fDrawBufferIBAllocPool = NULL;
    fFlushToReduceCacheSize = false;
    fAARectRenderer = NULL;
    fOvalRenderer = NULL;
    fViewMatrix.reset();
    fMaxTextureSizeOverride = 1 << 20;
}

bool GrContext::init(GrBackend backend, GrBackendContext backendContext) {
    SkASSERT(NULL == fGpu);

    fGpu = GrGpu::Create(backend, backendContext, this);
    if (NULL == fGpu) {
        return false;
    }

    fDrawState = SkNEW(GrDrawState);
    fGpu->setDrawState(fDrawState);

    fTextureCache = SkNEW_ARGS(GrResourceCache,
                               (MAX_RESOURCE_CACHE_COUNT,
                                MAX_RESOURCE_CACHE_BYTES));
    fTextureCache->setOverbudgetCallback(OverbudgetCB, this);

    fFontCache = SkNEW_ARGS(GrFontCache, (fGpu));

    fLastDrawWasBuffered = kNo_BufferedDraw;

    fAARectRenderer = SkNEW(GrAARectRenderer);
    fOvalRenderer = SkNEW(GrOvalRenderer);

    fDidTestPMConversions = false;

    this->setupDrawBuffer();

    return true;
}

GrContext::~GrContext() {
    if (NULL == fGpu) {
        return;
    }

    this->flush();

    for (int i = 0; i < fCleanUpData.count(); ++i) {
        (*fCleanUpData[i].fFunc)(this, fCleanUpData[i].fInfo);
    }

    // Since the gpu can hold scratch textures, give it a chance to let go
    // of them before freeing the texture cache
    fGpu->purgeResources();

    delete fTextureCache;
    fTextureCache = NULL;
    delete fFontCache;
    delete fDrawBuffer;
    delete fDrawBufferVBAllocPool;
    delete fDrawBufferIBAllocPool;

    fAARectRenderer->unref();
    fOvalRenderer->unref();

    fGpu->unref();
    SkSafeUnref(fPathRendererChain);
    SkSafeUnref(fSoftwarePathRenderer);
    fDrawState->unref();
}

void GrContext::contextLost() {
    this->contextDestroyed();
    this->setupDrawBuffer();
}

void GrContext::contextDestroyed() {
    // abandon first to so destructors
    // don't try to free the resources in the API.
    fGpu->abandonResources();

    // a path renderer may be holding onto resources that
    // are now unusable
    SkSafeSetNull(fPathRendererChain);
    SkSafeSetNull(fSoftwarePathRenderer);

    delete fDrawBuffer;
    fDrawBuffer = NULL;

    delete fDrawBufferVBAllocPool;
    fDrawBufferVBAllocPool = NULL;

    delete fDrawBufferIBAllocPool;
    fDrawBufferIBAllocPool = NULL;

    fAARectRenderer->reset();
    fOvalRenderer->reset();

    fTextureCache->purgeAllUnlocked();

    fFontCache->freeAll();
    fGpu->markContextDirty();
}

void GrContext::resetContext(uint32_t state) {
    fGpu->markContextDirty(state);
}

void GrContext::freeGpuResources() {
    this->flush();

    fGpu->purgeResources();

    fAARectRenderer->reset();
    fOvalRenderer->reset();

    fTextureCache->purgeAllUnlocked();
    fFontCache->freeAll();
    // a path renderer may be holding onto resources
    SkSafeSetNull(fPathRendererChain);
    SkSafeSetNull(fSoftwarePathRenderer);
}

size_t GrContext::getGpuTextureCacheBytes() const {
  return fTextureCache->getCachedResourceBytes();
}

////////////////////////////////////////////////////////////////////////////////

GrTexture* GrContext::findAndRefTexture(const GrTextureDesc& desc,
                                        const GrCacheID& cacheID,
                                        const GrTextureParams* params) {
    GrResourceKey resourceKey = GrTexture::ComputeKey(fGpu, params, desc, cacheID);
    GrResource* resource = fTextureCache->find(resourceKey);
    SkSafeRef(resource);
    return static_cast<GrTexture*>(resource);
}

bool GrContext::isTextureInCache(const GrTextureDesc& desc,
                                 const GrCacheID& cacheID,
                                 const GrTextureParams* params) const {
    GrResourceKey resourceKey = GrTexture::ComputeKey(fGpu, params, desc, cacheID);
    return fTextureCache->hasKey(resourceKey);
}

void GrContext::addStencilBuffer(GrStencilBuffer* sb) {
    ASSERT_OWNED_RESOURCE(sb);

    GrResourceKey resourceKey = GrStencilBuffer::ComputeKey(sb->width(),
                                                            sb->height(),
                                                            sb->numSamples());
    fTextureCache->addResource(resourceKey, sb);
}

GrStencilBuffer* GrContext::findStencilBuffer(int width, int height,
                                              int sampleCnt) {
    GrResourceKey resourceKey = GrStencilBuffer::ComputeKey(width,
                                                            height,
                                                            sampleCnt);
    GrResource* resource = fTextureCache->find(resourceKey);
    return static_cast<GrStencilBuffer*>(resource);
}

static void stretchImage(void* dst,
                         int dstW,
                         int dstH,
                         void* src,
                         int srcW,
                         int srcH,
                         size_t bpp) {
    GrFixed dx = (srcW << 16) / dstW;
    GrFixed dy = (srcH << 16) / dstH;

    GrFixed y = dy >> 1;

    size_t dstXLimit = dstW*bpp;
    for (int j = 0; j < dstH; ++j) {
        GrFixed x = dx >> 1;
        void* srcRow = (uint8_t*)src + (y>>16)*srcW*bpp;
        void* dstRow = (uint8_t*)dst + j*dstW*bpp;
        for (size_t i = 0; i < dstXLimit; i += bpp) {
            memcpy((uint8_t*) dstRow + i,
                   (uint8_t*) srcRow + (x>>16)*bpp,
                   bpp);
            x += dx;
        }
        y += dy;
    }
}

namespace {

// position + local coordinate
extern const GrVertexAttrib gVertexAttribs[] = {
    {kVec2f_GrVertexAttribType, 0,               kPosition_GrVertexAttribBinding},
    {kVec2f_GrVertexAttribType, sizeof(GrPoint), kLocalCoord_GrVertexAttribBinding}
};

};

// The desired texture is NPOT and tiled but that isn't supported by
// the current hardware. Resize the texture to be a POT
GrTexture* GrContext::createResizedTexture(const GrTextureDesc& desc,
                                           const GrCacheID& cacheID,
                                           void* srcData,
                                           size_t rowBytes,
                                           bool filter) {
    SkAutoTUnref<GrTexture> clampedTexture(this->findAndRefTexture(desc, cacheID, NULL));
    if (NULL == clampedTexture) {
        clampedTexture.reset(this->createTexture(NULL, desc, cacheID, srcData, rowBytes));

        if (NULL == clampedTexture) {
            return NULL;
        }
    }

    GrTextureDesc rtDesc = desc;
    rtDesc.fFlags =  rtDesc.fFlags |
                     kRenderTarget_GrTextureFlagBit |
                     kNoStencil_GrTextureFlagBit;
    rtDesc.fWidth  = GrNextPow2(desc.fWidth);
    rtDesc.fHeight = GrNextPow2(desc.fHeight);

    GrTexture* texture = fGpu->createTexture(rtDesc, NULL, 0);

    if (NULL != texture) {
        GrDrawTarget::AutoStateRestore asr(fGpu, GrDrawTarget::kReset_ASRInit);
        GrDrawState* drawState = fGpu->drawState();
        drawState->setRenderTarget(texture->asRenderTarget());

        // if filtering is not desired then we want to ensure all
        // texels in the resampled image are copies of texels from
        // the original.
        GrTextureParams params(SkShader::kClamp_TileMode, filter ? GrTextureParams::kBilerp_FilterMode :
                                                                   GrTextureParams::kNone_FilterMode);
        drawState->addColorTextureEffect(clampedTexture, SkMatrix::I(), params);

        drawState->setVertexAttribs<gVertexAttribs>(SK_ARRAY_COUNT(gVertexAttribs));

        GrDrawTarget::AutoReleaseGeometry arg(fGpu, 4, 0);

        if (arg.succeeded()) {
            GrPoint* verts = (GrPoint*) arg.vertices();
            verts[0].setIRectFan(0, 0, texture->width(), texture->height(), 2 * sizeof(GrPoint));
            verts[1].setIRectFan(0, 0, 1, 1, 2 * sizeof(GrPoint));
            fGpu->drawNonIndexed(kTriangleFan_GrPrimitiveType, 0, 4);
        }
    } else {
        // TODO: Our CPU stretch doesn't filter. But we create separate
        // stretched textures when the texture params is either filtered or
        // not. Either implement filtered stretch blit on CPU or just create
        // one when FBO case fails.

        rtDesc.fFlags = kNone_GrTextureFlags;
        // no longer need to clamp at min RT size.
        rtDesc.fWidth  = GrNextPow2(desc.fWidth);
        rtDesc.fHeight = GrNextPow2(desc.fHeight);
        size_t bpp = GrBytesPerPixel(desc.fConfig);
        SkAutoSMalloc<128*128*4> stretchedPixels(bpp * rtDesc.fWidth * rtDesc.fHeight);
        stretchImage(stretchedPixels.get(), rtDesc.fWidth, rtDesc.fHeight,
                     srcData, desc.fWidth, desc.fHeight, bpp);

        size_t stretchedRowBytes = rtDesc.fWidth * bpp;

        SkDEBUGCODE(GrTexture* texture = )fGpu->createTexture(rtDesc, stretchedPixels.get(),
                                                              stretchedRowBytes);
        SkASSERT(NULL != texture);
    }

    return texture;
}

GrTexture* GrContext::createTexture(const GrTextureParams* params,
                                    const GrTextureDesc& desc,
                                    const GrCacheID& cacheID,
                                    void* srcData,
                                    size_t rowBytes,
                                    GrResourceKey* cacheKey) {
    SK_TRACE_EVENT0("GrContext::createTexture");

    GrResourceKey resourceKey = GrTexture::ComputeKey(fGpu, params, desc, cacheID);

    GrTexture* texture;
    if (GrTexture::NeedsResizing(resourceKey)) {
        texture = this->createResizedTexture(desc, cacheID,
                                             srcData, rowBytes,
                                             GrTexture::NeedsBilerp(resourceKey));
    } else {
        texture= fGpu->createTexture(desc, srcData, rowBytes);
    }

    if (NULL != texture) {
        // Adding a resource could put us overbudget. Try to free up the
        // necessary space before adding it.
        fTextureCache->purgeAsNeeded(1, texture->sizeInBytes());
        fTextureCache->addResource(resourceKey, texture);

        if (NULL != cacheKey) {
            *cacheKey = resourceKey;
        }
    }

    return texture;
}

static GrTexture* create_scratch_texture(GrGpu* gpu,
                                         GrResourceCache* textureCache,
                                         const GrTextureDesc& desc) {
    GrTexture* texture = gpu->createTexture(desc, NULL, 0);
    if (NULL != texture) {
        GrResourceKey key = GrTexture::ComputeScratchKey(texture->desc());
        // Adding a resource could put us overbudget. Try to free up the
        // necessary space before adding it.
        textureCache->purgeAsNeeded(1, texture->sizeInBytes());
        // Make the resource exclusive so future 'find' calls don't return it
        textureCache->addResource(key, texture, GrResourceCache::kHide_OwnershipFlag);
    }
    return texture;
}

GrTexture* GrContext::lockAndRefScratchTexture(const GrTextureDesc& inDesc, ScratchTexMatch match) {

    SkASSERT((inDesc.fFlags & kRenderTarget_GrTextureFlagBit) ||
             !(inDesc.fFlags & kNoStencil_GrTextureFlagBit));

    // Renderable A8 targets are not universally supported (e.g., not on ANGLE)
    SkASSERT(this->isConfigRenderable(kAlpha_8_GrPixelConfig, inDesc.fSampleCnt > 0) ||
             !(inDesc.fFlags & kRenderTarget_GrTextureFlagBit) ||
             (inDesc.fConfig != kAlpha_8_GrPixelConfig));

    if (!fGpu->caps()->reuseScratchTextures() &&
        !(inDesc.fFlags & kRenderTarget_GrTextureFlagBit)) {
        // If we're never recycling this texture we can always make it the right size
        return create_scratch_texture(fGpu, fTextureCache, inDesc);
    }

    GrTextureDesc desc = inDesc;

    if (kApprox_ScratchTexMatch == match) {
        // bin by pow2 with a reasonable min
        static const int MIN_SIZE = 16;
        desc.fWidth  = GrMax(MIN_SIZE, GrNextPow2(desc.fWidth));
        desc.fHeight = GrMax(MIN_SIZE, GrNextPow2(desc.fHeight));
    }

    GrResource* resource = NULL;
    int origWidth = desc.fWidth;
    int origHeight = desc.fHeight;

    do {
        GrResourceKey key = GrTexture::ComputeScratchKey(desc);
        // Ensure we have exclusive access to the texture so future 'find' calls don't return it
        resource = fTextureCache->find(key, GrResourceCache::kHide_OwnershipFlag);
        if (NULL != resource) {
            resource->ref();
            break;
        }
        if (kExact_ScratchTexMatch == match) {
            break;
        }
        // We had a cache miss and we are in approx mode, relax the fit of the flags.

        // We no longer try to reuse textures that were previously used as render targets in
        // situations where no RT is needed; doing otherwise can confuse the video driver and
        // cause significant performance problems in some cases.
        if (desc.fFlags & kNoStencil_GrTextureFlagBit) {
            desc.fFlags = desc.fFlags & ~kNoStencil_GrTextureFlagBit;
        } else {
            break;
        }

    } while (true);

    if (NULL == resource) {
        desc.fFlags = inDesc.fFlags;
        desc.fWidth = origWidth;
        desc.fHeight = origHeight;
        resource = create_scratch_texture(fGpu, fTextureCache, desc);
    }

    return static_cast<GrTexture*>(resource);
}

void GrContext::addExistingTextureToCache(GrTexture* texture) {

    if (NULL == texture) {
        return;
    }

    // This texture should already have a cache entry since it was once
    // attached
    SkASSERT(NULL != texture->getCacheEntry());

    // Conceptually, the cache entry is going to assume responsibility
    // for the creation ref. Assert refcnt == 1.
    SkASSERT(texture->unique());

    if (fGpu->caps()->reuseScratchTextures() || NULL != texture->asRenderTarget()) {
        // Since this texture came from an AutoScratchTexture it should
        // still be in the exclusive pile. Recycle it.
        fTextureCache->makeNonExclusive(texture->getCacheEntry());
        this->purgeCache();
    } else if (texture->getDeferredRefCount() <= 0) {
        // When we aren't reusing textures we know this scratch texture
        // will never be reused and would be just wasting time in the cache
        fTextureCache->makeNonExclusive(texture->getCacheEntry());
        fTextureCache->deleteResource(texture->getCacheEntry());
    } else {
        // In this case (fDeferredRefCount > 0) but the cache is the only
        // one holding a real ref. Mark the object so when the deferred
        // ref count goes to 0 the texture will be deleted (remember
        // in this code path scratch textures aren't getting reused).
        texture->setNeedsDeferredUnref();
    }
}


void GrContext::unlockScratchTexture(GrTexture* texture) {
    ASSERT_OWNED_RESOURCE(texture);
    SkASSERT(NULL != texture->getCacheEntry());

    // If this is a scratch texture we detached it from the cache
    // while it was locked (to avoid two callers simultaneously getting
    // the same texture).
    if (texture->getCacheEntry()->key().isScratch()) {
        if (fGpu->caps()->reuseScratchTextures() || NULL != texture->asRenderTarget()) {
            fTextureCache->makeNonExclusive(texture->getCacheEntry());
            this->purgeCache();
        } else if (texture->unique() && texture->getDeferredRefCount() <= 0) {
            // Only the cache now knows about this texture. Since we're never
            // reusing scratch textures (in this code path) it would just be
            // wasting time sitting in the cache.
            fTextureCache->makeNonExclusive(texture->getCacheEntry());
            fTextureCache->deleteResource(texture->getCacheEntry());
        } else {
            // In this case (fRefCnt > 1 || defRefCnt > 0) but we don't really
            // want to readd it to the cache (since it will never be reused).
            // Instead, give up the cache's ref and leave the decision up to
            // addExistingTextureToCache once its ref count reaches 0. For
            // this to work we need to leave it in the exclusive list.
            texture->setFlag((GrTextureFlags) GrTexture::kReturnToCache_FlagBit);
            // Give up the cache's ref to the texture
            texture->unref();
        }
    }
}

void GrContext::purgeCache() {
    if (NULL != fTextureCache) {
        fTextureCache->purgeAsNeeded();
    }
}

bool GrContext::OverbudgetCB(void* data) {
    SkASSERT(NULL != data);

    GrContext* context = reinterpret_cast<GrContext*>(data);

    // Flush the InOrderDrawBuffer to possibly free up some textures
    context->fFlushToReduceCacheSize = true;

    return true;
}


GrTexture* GrContext::createUncachedTexture(const GrTextureDesc& descIn,
                                            void* srcData,
                                            size_t rowBytes) {
    GrTextureDesc descCopy = descIn;
    return fGpu->createTexture(descCopy, srcData, rowBytes);
}

void GrContext::getTextureCacheLimits(int* maxTextures,
                                      size_t* maxTextureBytes) const {
    fTextureCache->getLimits(maxTextures, maxTextureBytes);
}

void GrContext::setTextureCacheLimits(int maxTextures, size_t maxTextureBytes) {
    fTextureCache->setLimits(maxTextures, maxTextureBytes);
}

int GrContext::getMaxTextureSize() const {
    return GrMin(fGpu->caps()->maxTextureSize(), fMaxTextureSizeOverride);
}

int GrContext::getMaxRenderTargetSize() const {
    return fGpu->caps()->maxRenderTargetSize();
}

int GrContext::getMaxSampleCount() const {
    return fGpu->caps()->maxSampleCount();
}

///////////////////////////////////////////////////////////////////////////////

GrTexture* GrContext::wrapBackendTexture(const GrBackendTextureDesc& desc) {
    return fGpu->wrapBackendTexture(desc);
}

GrRenderTarget* GrContext::wrapBackendRenderTarget(const GrBackendRenderTargetDesc& desc) {
    return fGpu->wrapBackendRenderTarget(desc);
}

///////////////////////////////////////////////////////////////////////////////

bool GrContext::supportsIndex8PixelConfig(const GrTextureParams* params,
                                          int width, int height) const {
    const GrDrawTargetCaps* caps = fGpu->caps();
    if (!caps->eightBitPaletteSupport()) {
        return false;
    }

    bool isPow2 = GrIsPow2(width) && GrIsPow2(height);

    if (!isPow2) {
        bool tiled = NULL != params && params->isTiled();
        if (tiled && !caps->npotTextureTileSupport()) {
            return false;
        }
    }
    return true;
}


////////////////////////////////////////////////////////////////////////////////

void GrContext::clear(const SkIRect* rect,
                      const GrColor color,
                      bool canIgnoreRect,
                      GrRenderTarget* target) {
    AutoRestoreEffects are;
    AutoCheckFlush acf(this);
    this->prepareToDraw(NULL, BUFFERED_DRAW, &are, &acf)->clear(rect, color,
                                                                canIgnoreRect, target);
}

void GrContext::drawPaint(const GrPaint& origPaint) {
    // set rect to be big enough to fill the space, but not super-huge, so we
    // don't overflow fixed-point implementations
    SkRect r;
    r.setLTRB(0, 0,
              SkIntToScalar(getRenderTarget()->width()),
              SkIntToScalar(getRenderTarget()->height()));
    SkMatrix inverse;
    SkTCopyOnFirstWrite<GrPaint> paint(origPaint);
    AutoMatrix am;

    // We attempt to map r by the inverse matrix and draw that. mapRect will
    // map the four corners and bound them with a new rect. This will not
    // produce a correct result for some perspective matrices.
    if (!this->getMatrix().hasPerspective()) {
        if (!fViewMatrix.invert(&inverse)) {
            GrPrintf("Could not invert matrix\n");
            return;
        }
        inverse.mapRect(&r);
    } else {
        if (!am.setIdentity(this, paint.writable())) {
            GrPrintf("Could not invert matrix\n");
            return;
        }
    }
    // by definition this fills the entire clip, no need for AA
    if (paint->isAntiAlias()) {
        paint.writable()->setAntiAlias(false);
    }
    this->drawRect(*paint, r);
}

#ifdef SK_DEVELOPER
void GrContext::dumpFontCache() const {
    fFontCache->dump();
}
#endif

////////////////////////////////////////////////////////////////////////////////

/*  create a triangle strip that strokes the specified triangle. There are 8
 unique vertices, but we repreat the last 2 to close up. Alternatively we
 could use an indices array, and then only send 8 verts, but not sure that
 would be faster.
 */
static void setStrokeRectStrip(GrPoint verts[10], SkRect rect,
                               SkScalar width) {
    const SkScalar rad = SkScalarHalf(width);
    rect.sort();

    verts[0].set(rect.fLeft + rad, rect.fTop + rad);
    verts[1].set(rect.fLeft - rad, rect.fTop - rad);
    verts[2].set(rect.fRight - rad, rect.fTop + rad);
    verts[3].set(rect.fRight + rad, rect.fTop - rad);
    verts[4].set(rect.fRight - rad, rect.fBottom - rad);
    verts[5].set(rect.fRight + rad, rect.fBottom + rad);
    verts[6].set(rect.fLeft + rad, rect.fBottom - rad);
    verts[7].set(rect.fLeft - rad, rect.fBottom + rad);
    verts[8] = verts[0];
    verts[9] = verts[1];
}

static bool isIRect(const SkRect& r) {
    return SkScalarIsInt(r.fLeft)  && SkScalarIsInt(r.fTop) &&
           SkScalarIsInt(r.fRight) && SkScalarIsInt(r.fBottom);
}

static bool apply_aa_to_rect(GrDrawTarget* target,
                             const SkRect& rect,
                             SkScalar strokeWidth,
                             const SkMatrix& combinedMatrix,
                             SkRect* devBoundRect,
                             bool* useVertexCoverage) {
    // we use a simple coverage ramp to do aa on axis-aligned rects
    // we check if the rect will be axis-aligned, and the rect won't land on
    // integer coords.

    // we are keeping around the "tweak the alpha" trick because
    // it is our only hope for the fixed-pipe implementation.
    // In a shader implementation we can give a separate coverage input
    // TODO: remove this ugliness when we drop the fixed-pipe impl
    *useVertexCoverage = false;
    if (!target->getDrawState().canTweakAlphaForCoverage()) {
        if (target->shouldDisableCoverageAAForBlend()) {
#ifdef SK_DEBUG
            //GrPrintf("Turning off AA to correctly apply blend.\n");
#endif
            return false;
        } else {
            *useVertexCoverage = true;
        }
    }
    const GrDrawState& drawState = target->getDrawState();
    if (drawState.getRenderTarget()->isMultisampled()) {
        return false;
    }

    if (0 == strokeWidth && target->willUseHWAALines()) {
        return false;
    }

#if defined(SHADER_AA_FILL_RECT) || !defined(IGNORE_ROT_AA_RECT_OPT)
    if (strokeWidth >= 0) {
#endif
        if (!combinedMatrix.preservesAxisAlignment()) {
            return false;
        }

#if defined(SHADER_AA_FILL_RECT) || !defined(IGNORE_ROT_AA_RECT_OPT)
    } else {
        if (!combinedMatrix.preservesRightAngles()) {
            return false;
        }
    }
#endif

    combinedMatrix.mapRect(devBoundRect, rect);

    if (strokeWidth < 0) {
        return !isIRect(*devBoundRect);
    } else {
        return true;
    }
}

static inline bool rect_contains_inclusive(const SkRect& rect, const SkPoint& point) {
    return point.fX >= rect.fLeft && point.fX <= rect.fRight &&
           point.fY >= rect.fTop && point.fY <= rect.fBottom;
}

void GrContext::drawRect(const GrPaint& paint,
                         const SkRect& rect,
                         const SkStrokeRec* stroke,
                         const SkMatrix* matrix) {
    SK_TRACE_EVENT0("GrContext::drawRect");

    AutoRestoreEffects are;
    AutoCheckFlush acf(this);
    GrDrawTarget* target = this->prepareToDraw(&paint, BUFFERED_DRAW, &are, &acf);

    SkScalar width = stroke == NULL ? -1 : stroke->getWidth();
    SkMatrix combinedMatrix = target->drawState()->getViewMatrix();
    if (NULL != matrix) {
        combinedMatrix.preConcat(*matrix);
    }

    // Check if this is a full RT draw and can be replaced with a clear. We don't bother checking
    // cases where the RT is fully inside a stroke.
    if (width < 0) {
        SkRect rtRect;
        target->getDrawState().getRenderTarget()->getBoundsRect(&rtRect);
        SkRect clipSpaceRTRect = rtRect;
        bool checkClip = false;
        if (NULL != this->getClip()) {
            checkClip = true;
            clipSpaceRTRect.offset(SkIntToScalar(this->getClip()->fOrigin.fX),
                                   SkIntToScalar(this->getClip()->fOrigin.fY));
        }
        // Does the clip contain the entire RT?
        if (!checkClip || target->getClip()->fClipStack->quickContains(clipSpaceRTRect)) {
            SkMatrix invM;
            if (!combinedMatrix.invert(&invM)) {
                return;
            }
            // Does the rect bound the RT?
            SkPoint srcSpaceRTQuad[4];
            invM.mapRectToQuad(srcSpaceRTQuad, rtRect);
            if (rect_contains_inclusive(rect, srcSpaceRTQuad[0]) &&
                rect_contains_inclusive(rect, srcSpaceRTQuad[1]) &&
                rect_contains_inclusive(rect, srcSpaceRTQuad[2]) &&
                rect_contains_inclusive(rect, srcSpaceRTQuad[3])) {
                // Will it blend?
                GrColor clearColor;
                if (paint.isOpaqueAndConstantColor(&clearColor)) {
                    target->clear(NULL, clearColor, true);
                    return;
                }
            }
        }
    }

    SkRect devBoundRect;
    bool useVertexCoverage;
    bool needAA = paint.isAntiAlias() &&
                  !target->getDrawState().getRenderTarget()->isMultisampled();
    bool doAA = needAA && apply_aa_to_rect(target, rect, width, combinedMatrix, &devBoundRect,
                                           &useVertexCoverage);
    if (doAA) {
        GrDrawState::AutoViewMatrixRestore avmr;
        if (!avmr.setIdentity(target->drawState())) {
            return;
        }
        if (width >= 0) {
            fAARectRenderer->strokeAARect(this->getGpu(), target, rect,
                                          combinedMatrix, devBoundRect,
                                          stroke, useVertexCoverage);
        } else {
            // filled AA rect
            fAARectRenderer->fillAARect(this->getGpu(), target,
                                        rect, combinedMatrix, devBoundRect,
                                        useVertexCoverage);
        }
        return;
    }

    if (width >= 0) {
        // TODO: consider making static vertex buffers for these cases.
        // Hairline could be done by just adding closing vertex to
        // unitSquareVertexBuffer()

        static const int worstCaseVertCount = 10;
        target->drawState()->setDefaultVertexAttribs();
        GrDrawTarget::AutoReleaseGeometry geo(target, worstCaseVertCount, 0);

        if (!geo.succeeded()) {
            GrPrintf("Failed to get space for vertices!\n");
            return;
        }

        GrPrimitiveType primType;
        int vertCount;
        GrPoint* vertex = geo.positions();

        if (width > 0) {
            vertCount = 10;
            primType = kTriangleStrip_GrPrimitiveType;
            setStrokeRectStrip(vertex, rect, width);
        } else {
            // hairline
            vertCount = 5;
            primType = kLineStrip_GrPrimitiveType;
            vertex[0].set(rect.fLeft, rect.fTop);
            vertex[1].set(rect.fRight, rect.fTop);
            vertex[2].set(rect.fRight, rect.fBottom);
            vertex[3].set(rect.fLeft, rect.fBottom);
            vertex[4].set(rect.fLeft, rect.fTop);
        }

        GrDrawState::AutoViewMatrixRestore avmr;
        if (NULL != matrix) {
            GrDrawState* drawState = target->drawState();
            avmr.set(drawState, *matrix);
        }

        target->drawNonIndexed(primType, 0, vertCount);
    } else {
        // filled BW rect
        target->drawSimpleRect(rect, matrix);
    }
}

void GrContext::drawRectToRect(const GrPaint& paint,
                               const SkRect& dstRect,
                               const SkRect& localRect,
                               const SkMatrix* dstMatrix,
                               const SkMatrix* localMatrix) {
    SK_TRACE_EVENT0("GrContext::drawRectToRect");
    AutoRestoreEffects are;
    AutoCheckFlush acf(this);
    GrDrawTarget* target = this->prepareToDraw(&paint, BUFFERED_DRAW, &are, &acf);

    target->drawRect(dstRect, dstMatrix, &localRect, localMatrix);
}

namespace {

extern const GrVertexAttrib gPosUVColorAttribs[] = {
    {kVec2f_GrVertexAttribType,  0, kPosition_GrVertexAttribBinding },
    {kVec2f_GrVertexAttribType,  sizeof(GrPoint), kLocalCoord_GrVertexAttribBinding },
    {kVec4ub_GrVertexAttribType, 2*sizeof(GrPoint), kColor_GrVertexAttribBinding}
};

extern const GrVertexAttrib gPosColorAttribs[] = {
    {kVec2f_GrVertexAttribType,  0, kPosition_GrVertexAttribBinding},
    {kVec4ub_GrVertexAttribType, sizeof(GrPoint), kColor_GrVertexAttribBinding},
};

static void set_vertex_attributes(GrDrawState* drawState,
                                  const GrPoint* texCoords,
                                  const GrColor* colors,
                                  int* colorOffset,
                                  int* texOffset) {
    *texOffset = -1;
    *colorOffset = -1;

    if (NULL != texCoords && NULL != colors) {
        *texOffset = sizeof(GrPoint);
        *colorOffset = 2*sizeof(GrPoint);
        drawState->setVertexAttribs<gPosUVColorAttribs>(3);
    } else if (NULL != texCoords) {
        *texOffset = sizeof(GrPoint);
        drawState->setVertexAttribs<gPosUVColorAttribs>(2);
    } else if (NULL != colors) {
        *colorOffset = sizeof(GrPoint);
        drawState->setVertexAttribs<gPosColorAttribs>(2);
    } else {
        drawState->setVertexAttribs<gPosColorAttribs>(1);
    }
}

};

void GrContext::drawVertices(const GrPaint& paint,
                             GrPrimitiveType primitiveType,
                             int vertexCount,
                             const GrPoint positions[],
                             const GrPoint texCoords[],
                             const GrColor colors[],
                             const uint16_t indices[],
                             int indexCount) {
    SK_TRACE_EVENT0("GrContext::drawVertices");

    AutoRestoreEffects are;
    AutoCheckFlush acf(this);
    GrDrawTarget::AutoReleaseGeometry geo; // must be inside AutoCheckFlush scope

    GrDrawTarget* target = this->prepareToDraw(&paint, BUFFERED_DRAW, &are, &acf);

    GrDrawState* drawState = target->drawState();

    int colorOffset = -1, texOffset = -1;
    set_vertex_attributes(drawState, texCoords, colors, &colorOffset, &texOffset);

    size_t vertexSize = drawState->getVertexSize();
    if (sizeof(GrPoint) != vertexSize) {
        if (!geo.set(target, vertexCount, 0)) {
            GrPrintf("Failed to get space for vertices!\n");
            return;
        }
        void* curVertex = geo.vertices();

        for (int i = 0; i < vertexCount; ++i) {
            *((GrPoint*)curVertex) = positions[i];

            if (texOffset >= 0) {
                *(GrPoint*)((intptr_t)curVertex + texOffset) = texCoords[i];
            }
            if (colorOffset >= 0) {
                *(GrColor*)((intptr_t)curVertex + colorOffset) = colors[i];
            }
            curVertex = (void*)((intptr_t)curVertex + vertexSize);
        }
    } else {
        target->setVertexSourceToArray(positions, vertexCount);
    }

    // we don't currently apply offscreen AA to this path. Need improved
    // management of GrDrawTarget's geometry to avoid copying points per-tile.

    if (NULL != indices) {
        target->setIndexSourceToArray(indices, indexCount);
        target->drawIndexed(primitiveType, 0, 0, vertexCount, indexCount);
        target->resetIndexSource();
    } else {
        target->drawNonIndexed(primitiveType, 0, vertexCount);
    }
}

///////////////////////////////////////////////////////////////////////////////

void GrContext::drawRRect(const GrPaint& paint,
                          const SkRRect& rect,
                          const SkStrokeRec& stroke) {
    if (rect.isEmpty()) {
       return;
    }

    AutoRestoreEffects are;
    AutoCheckFlush acf(this);
    GrDrawTarget* target = this->prepareToDraw(&paint, BUFFERED_DRAW, &are, &acf);

    if (!fOvalRenderer->drawSimpleRRect(target, this, paint.isAntiAlias(), rect, stroke)) {
        SkPath path;
        path.addRRect(rect);
        this->internalDrawPath(target, paint.isAntiAlias(), path, stroke);
    }
}

///////////////////////////////////////////////////////////////////////////////

void GrContext::drawOval(const GrPaint& paint,
                         const SkRect& oval,
                         const SkStrokeRec& stroke) {
    if (oval.isEmpty()) {
       return;
    }

    AutoRestoreEffects are;
    AutoCheckFlush acf(this);
    GrDrawTarget* target = this->prepareToDraw(&paint, BUFFERED_DRAW, &are, &acf);

    if (!fOvalRenderer->drawOval(target, this, paint.isAntiAlias(), oval, stroke)) {
        SkPath path;
        path.addOval(oval);
        this->internalDrawPath(target, paint.isAntiAlias(), path, stroke);
    }
}

// Can 'path' be drawn as a pair of filled nested rectangles?
static bool is_nested_rects(GrDrawTarget* target,
                            const SkPath& path,
                            const SkStrokeRec& stroke,
                            SkRect rects[2],
                            bool* useVertexCoverage) {
    SkASSERT(stroke.isFillStyle());

    if (path.isInverseFillType()) {
        return false;
    }

    const GrDrawState& drawState = target->getDrawState();

    // TODO: this restriction could be lifted if we were willing to apply
    // the matrix to all the points individually rather than just to the rect
    if (!drawState.getViewMatrix().preservesAxisAlignment()) {
        return false;
    }

    *useVertexCoverage = false;
    if (!target->getDrawState().canTweakAlphaForCoverage()) {
        if (target->shouldDisableCoverageAAForBlend()) {
            return false;
        } else {
            *useVertexCoverage = true;
        }
    }

    SkPath::Direction dirs[2];
    if (!path.isNestedRects(rects, dirs)) {
        return false;
    }

    if (SkPath::kWinding_FillType == path.getFillType() && dirs[0] == dirs[1]) {
        // The two rects need to be wound opposite to each other
        return false;
    }

    // Right now, nested rects where the margin is not the same width
    // all around do not render correctly
    const SkScalar* outer = rects[0].asScalars();
    const SkScalar* inner = rects[1].asScalars();

    SkScalar margin = SkScalarAbs(outer[0] - inner[0]);
    for (int i = 1; i < 4; ++i) {
        SkScalar temp = SkScalarAbs(outer[i] - inner[i]);
        if (!SkScalarNearlyEqual(margin, temp)) {
            return false;
        }
    }

    return true;
}

void GrContext::drawPath(const GrPaint& paint, const SkPath& path, const SkStrokeRec& stroke) {

    if (path.isEmpty()) {
       if (path.isInverseFillType()) {
           this->drawPaint(paint);
       }
       return;
    }

    // Note that internalDrawPath may sw-rasterize the path into a scratch texture.
    // Scratch textures can be recycled after they are returned to the texture
    // cache. This presents a potential hazard for buffered drawing. However,
    // the writePixels that uploads to the scratch will perform a flush so we're
    // OK.
    AutoRestoreEffects are;
    AutoCheckFlush acf(this);
    GrDrawTarget* target = this->prepareToDraw(&paint, BUFFERED_DRAW, &are, &acf);
    GrDrawState* drawState = target->drawState();

    bool useCoverageAA = paint.isAntiAlias() && !drawState->getRenderTarget()->isMultisampled();

    if (useCoverageAA && stroke.getWidth() < 0 && !path.isConvex()) {
        // Concave AA paths are expensive - try to avoid them for special cases
        bool useVertexCoverage;
        SkRect rects[2];

        if (is_nested_rects(target, path, stroke, rects, &useVertexCoverage)) {
            SkMatrix origViewMatrix = drawState->getViewMatrix();
            GrDrawState::AutoViewMatrixRestore avmr;
            if (!avmr.setIdentity(target->drawState())) {
                return;
            }

            fAARectRenderer->fillAANestedRects(this->getGpu(), target,
                                               rects,
                                               origViewMatrix,
                                               useVertexCoverage);
            return;
        }
    }

    SkRect ovalRect;
    bool isOval = path.isOval(&ovalRect);

    if (!isOval || path.isInverseFillType()
        || !fOvalRenderer->drawOval(target, this, paint.isAntiAlias(), ovalRect, stroke)) {
        this->internalDrawPath(target, paint.isAntiAlias(), path, stroke);
    }
}

void GrContext::internalDrawPath(GrDrawTarget* target, bool useAA, const SkPath& path,
                                 const SkStrokeRec& origStroke) {
    SkASSERT(!path.isEmpty());

    // An Assumption here is that path renderer would use some form of tweaking
    // the src color (either the input alpha or in the frag shader) to implement
    // aa. If we have some future driver-mojo path AA that can do the right
    // thing WRT to the blend then we'll need some query on the PR.
    bool useCoverageAA = useAA &&
        !target->getDrawState().getRenderTarget()->isMultisampled() &&
        !target->shouldDisableCoverageAAForBlend();


    GrPathRendererChain::DrawType type =
        useCoverageAA ? GrPathRendererChain::kColorAntiAlias_DrawType :
                        GrPathRendererChain::kColor_DrawType;

    const SkPath* pathPtr = &path;
    SkTLazy<SkPath> tmpPath;
    SkTCopyOnFirstWrite<SkStrokeRec> stroke(origStroke);

    // Try a 1st time without stroking the path and without allowing the SW renderer
    GrPathRenderer::AutoClearPath acp(this->getPathRenderer(*pathPtr, *stroke, 
                                      target, false, type,
                                      pathPtr->getFillType()));

    if (NULL == acp.renderer()) {
        if (!GrPathRenderer::IsStrokeHairlineOrEquivalent(*stroke, this->getMatrix(), NULL)) {
            // It didn't work the 1st time, so try again with the stroked path
            if (stroke->applyToPath(tmpPath.init(), *pathPtr)) {
                pathPtr = tmpPath.get();
                stroke.writable()->setFillStyle();
                if (pathPtr->isEmpty()) {
                    return;
                }
            }
        }

        // This time, allow SW renderer
        acp.set(this->getPathRenderer(*pathPtr, *stroke, target, true, type,
                                      pathPtr->getFillType()));
    }

    if (NULL == acp.renderer()) {
#ifdef SK_DEBUG
        GrPrintf("Unable to find path renderer compatible with path.\n");
#endif
        return;
    }

    acp->drawPath(*stroke, target, useCoverageAA);
}

////////////////////////////////////////////////////////////////////////////////

void GrContext::flush(int flagsBitfield) {
    if (NULL == fDrawBuffer) {
        return;
    }

    if (kDiscard_FlushBit & flagsBitfield) {
        fDrawBuffer->reset();
    } else {
        fDrawBuffer->flush();
    }
    fFlushToReduceCacheSize = false;
}

bool GrContext::writeTexturePixels(GrTexture* texture,
                                   int left, int top, int width, int height,
                                   GrPixelConfig config, const void* buffer, size_t rowBytes,
                                   uint32_t flags) {
    SK_TRACE_EVENT0("GrContext::writeTexturePixels");
    ASSERT_OWNED_RESOURCE(texture);

    if ((kUnpremul_PixelOpsFlag & flags) || !fGpu->canWriteTexturePixels(texture, config)) {
        if (NULL != texture->asRenderTarget()) {
            return this->writeRenderTargetPixels(texture->asRenderTarget(),
                                                 left, top, width, height,
                                                 config, buffer, rowBytes, flags);
        } else {
            return false;
        }
    }

    if (!(kDontFlush_PixelOpsFlag & flags)) {
        this->flush();
    }

    return fGpu->writeTexturePixels(texture, left, top, width, height,
                                    config, buffer, rowBytes);
}

bool GrContext::readTexturePixels(GrTexture* texture,
                                  int left, int top, int width, int height,
                                  GrPixelConfig config, void* buffer, size_t rowBytes,
                                  uint32_t flags) {
    SK_TRACE_EVENT0("GrContext::readTexturePixels");
    ASSERT_OWNED_RESOURCE(texture);

    GrRenderTarget* target = texture->asRenderTarget();
    if (NULL != target) {
        return this->readRenderTargetPixels(target,
                                            left, top, width, height,
                                            config, buffer, rowBytes,
                                            flags);
    } else {
        // TODO: make this more efficient for cases where we're reading the entire
        //       texture, i.e., use GetTexImage() instead

        // create scratch rendertarget and read from that
        GrAutoScratchTexture ast;
        GrTextureDesc desc;
        desc.fFlags = kRenderTarget_GrTextureFlagBit;
        desc.fWidth = width;
        desc.fHeight = height;
        desc.fConfig = config;
        desc.fOrigin = kTopLeft_GrSurfaceOrigin;
        ast.set(this, desc, kExact_ScratchTexMatch);
        GrTexture* dst = ast.texture();
        if (NULL != dst && NULL != (target = dst->asRenderTarget())) {
            this->copyTexture(texture, target, NULL);
            return this->readRenderTargetPixels(target,
                                                left, top, width, height,
                                                config, buffer, rowBytes,
                                                flags);
        }

        return false;
    }
}

#include "SkConfig8888.h"

namespace {
/**
 * Converts a GrPixelConfig to a SkCanvas::Config8888. Only byte-per-channel
 * formats are representable as Config8888 and so the function returns false
 * if the GrPixelConfig has no equivalent Config8888.
 */
bool grconfig_to_config8888(GrPixelConfig config,
                            bool unpremul,
                            SkCanvas::Config8888* config8888) {
    switch (config) {
        case kRGBA_8888_GrPixelConfig:
            if (unpremul) {
                *config8888 = SkCanvas::kRGBA_Unpremul_Config8888;
            } else {
                *config8888 = SkCanvas::kRGBA_Premul_Config8888;
            }
            return true;
        case kBGRA_8888_GrPixelConfig:
            if (unpremul) {
                *config8888 = SkCanvas::kBGRA_Unpremul_Config8888;
            } else {
                *config8888 = SkCanvas::kBGRA_Premul_Config8888;
            }
            return true;
        default:
            return false;
    }
}

// It returns a configuration with where the byte position of the R & B components are swapped in
// relation to the input config. This should only be called with the result of
// grconfig_to_config8888 as it will fail for other configs.
SkCanvas::Config8888 swap_config8888_red_and_blue(SkCanvas::Config8888 config8888) {
    switch (config8888) {
        case SkCanvas::kBGRA_Premul_Config8888:
            return SkCanvas::kRGBA_Premul_Config8888;
        case SkCanvas::kBGRA_Unpremul_Config8888:
            return SkCanvas::kRGBA_Unpremul_Config8888;
        case SkCanvas::kRGBA_Premul_Config8888:
            return SkCanvas::kBGRA_Premul_Config8888;
        case SkCanvas::kRGBA_Unpremul_Config8888:
            return SkCanvas::kBGRA_Unpremul_Config8888;
        default:
            GrCrash("Unexpected input");
            return SkCanvas::kBGRA_Unpremul_Config8888;;
    }
}
}

bool GrContext::readRenderTargetPixels(GrRenderTarget* target,
                                       int left, int top, int width, int height,
                                       GrPixelConfig dstConfig, void* buffer, size_t rowBytes,
                                       uint32_t flags) {
    SK_TRACE_EVENT0("GrContext::readRenderTargetPixels");
    ASSERT_OWNED_RESOURCE(target);

    if (NULL == target) {
        target = fRenderTarget.get();
        if (NULL == target) {
            return false;
        }
    }

    if (!(kDontFlush_PixelOpsFlag & flags)) {
        this->flush();
    }

    // Determine which conversions have to be applied: flipY, swapRAnd, and/or unpremul.

    // If fGpu->readPixels would incur a y-flip cost then we will read the pixels upside down. We'll
    // either do the flipY by drawing into a scratch with a matrix or on the cpu after the read.
    bool flipY = fGpu->readPixelsWillPayForYFlip(target, left, top,
                                                 width, height, dstConfig,
                                                 rowBytes);
    // We ignore the preferred config if it is different than our config unless it is an R/B swap.
    // In that case we'll perform an R and B swap while drawing to a scratch texture of the swapped
    // config. Then we will call readPixels on the scratch with the swapped config. The swaps during
    // the draw cancels out the fact that we call readPixels with a config that is R/B swapped from
    // dstConfig.
    GrPixelConfig readConfig = dstConfig;
    bool swapRAndB = false;
    if (GrPixelConfigSwapRAndB(dstConfig) ==
        fGpu->preferredReadPixelsConfig(dstConfig, target->config())) {
        readConfig = GrPixelConfigSwapRAndB(readConfig);
        swapRAndB = true;
    }

    bool unpremul = SkToBool(kUnpremul_PixelOpsFlag & flags);

    if (unpremul && !GrPixelConfigIs8888(dstConfig)) {
        // The unpremul flag is only allowed for these two configs.
        return false;
    }

    // If the src is a texture and we would have to do conversions after read pixels, we instead
    // do the conversions by drawing the src to a scratch texture. If we handle any of the
    // conversions in the draw we set the corresponding bool to false so that we don't reapply it
    // on the read back pixels.
    GrTexture* src = target->asTexture();
    GrAutoScratchTexture ast;
    if (NULL != src && (swapRAndB || unpremul || flipY)) {
        // Make the scratch a render target because we don't have a robust readTexturePixels as of
        // yet. It calls this function.
        GrTextureDesc desc;
        desc.fFlags = kRenderTarget_GrTextureFlagBit;
        desc.fWidth = width;
        desc.fHeight = height;
        desc.fConfig = readConfig;
        desc.fOrigin = kTopLeft_GrSurfaceOrigin;

        // When a full read back is faster than a partial we could always make the scratch exactly
        // match the passed rect. However, if we see many different size rectangles we will trash
        // our texture cache and pay the cost of creating and destroying many textures. So, we only
        // request an exact match when the caller is reading an entire RT.
        ScratchTexMatch match = kApprox_ScratchTexMatch;
        if (0 == left &&
            0 == top &&
            target->width() == width &&
            target->height() == height &&
            fGpu->fullReadPixelsIsFasterThanPartial()) {
            match = kExact_ScratchTexMatch;
        }
        ast.set(this, desc, match);
        GrTexture* texture = ast.texture();
        if (texture) {
            // compute a matrix to perform the draw
            SkMatrix textureMatrix;
            textureMatrix.setTranslate(SK_Scalar1 *left, SK_Scalar1 *top);
            textureMatrix.postIDiv(src->width(), src->height());

            SkAutoTUnref<const GrEffectRef> effect;
            if (unpremul) {
                effect.reset(this->createPMToUPMEffect(src, swapRAndB, textureMatrix));
                if (NULL != effect) {
                    unpremul = false; // we no longer need to do this on CPU after the read back.
                }
            }
            // If we failed to create a PM->UPM effect and have no other conversions to perform then
            // there is no longer any point to using the scratch.
            if (NULL != effect || flipY || swapRAndB) {
                if (!effect) {
                    effect.reset(GrConfigConversionEffect::Create(
                                                    src,
                                                    swapRAndB,
                                                    GrConfigConversionEffect::kNone_PMConversion,
                                                    textureMatrix));
                }
                swapRAndB = false; // we will handle the swap in the draw.

                // We protect the existing geometry here since it may not be
                // clear to the caller that a draw operation (i.e., drawSimpleRect)
                // can be invoked in this method
                GrDrawTarget::AutoGeometryAndStatePush agasp(fGpu, GrDrawTarget::kReset_ASRInit);
                GrDrawState* drawState = fGpu->drawState();
                SkASSERT(effect);
                drawState->addColorEffect(effect);

                drawState->setRenderTarget(texture->asRenderTarget());
                SkRect rect = SkRect::MakeWH(SkIntToScalar(width), SkIntToScalar(height));
                fGpu->drawSimpleRect(rect, NULL);
                // we want to read back from the scratch's origin
                left = 0;
                top = 0;
                target = texture->asRenderTarget();
            }
        }
    }
    if (!fGpu->readPixels(target,
                          left, top, width, height,
                          readConfig, buffer, rowBytes)) {
        return false;
    }
    // Perform any conversions we weren't able to perform using a scratch texture.
    if (unpremul || swapRAndB) {
        // These are initialized to suppress a warning
        SkCanvas::Config8888 srcC8888 = SkCanvas::kNative_Premul_Config8888;
        SkCanvas::Config8888 dstC8888 = SkCanvas::kNative_Premul_Config8888;

        SkDEBUGCODE(bool c8888IsValid =) grconfig_to_config8888(dstConfig, false, &srcC8888);
        grconfig_to_config8888(dstConfig, unpremul, &dstC8888);

        if (swapRAndB) {
            SkASSERT(c8888IsValid); // we should only do r/b swap on 8888 configs
            srcC8888 = swap_config8888_red_and_blue(srcC8888);
        }
        SkASSERT(c8888IsValid);
        uint32_t* b32 = reinterpret_cast<uint32_t*>(buffer);
        SkConvertConfig8888Pixels(b32, rowBytes, dstC8888,
                                  b32, rowBytes, srcC8888,
                                  width, height);
    }
    return true;
}

void GrContext::resolveRenderTarget(GrRenderTarget* target) {
    SkASSERT(target);
    ASSERT_OWNED_RESOURCE(target);
    // In the future we may track whether there are any pending draws to this
    // target. We don't today so we always perform a flush. We don't promise
    // this to our clients, though.
    this->flush();
    fGpu->resolveRenderTarget(target);
}

void GrContext::copyTexture(GrTexture* src, GrRenderTarget* dst, const SkIPoint* topLeft) {
    if (NULL == src || NULL == dst) {
        return;
    }
    ASSERT_OWNED_RESOURCE(src);

    // Writes pending to the source texture are not tracked, so a flush
    // is required to ensure that the copy captures the most recent contents
    // of the source texture. See similar behavior in
    // GrContext::resolveRenderTarget.
    this->flush();

    GrDrawTarget::AutoStateRestore asr(fGpu, GrDrawTarget::kReset_ASRInit);
    GrDrawState* drawState = fGpu->drawState();
    drawState->setRenderTarget(dst);
    SkMatrix sampleM;
    sampleM.setIDiv(src->width(), src->height());
    SkIRect srcRect = SkIRect::MakeWH(dst->width(), dst->height());
    if (NULL != topLeft) {
        srcRect.offset(*topLeft);
    }
    SkIRect srcBounds = SkIRect::MakeWH(src->width(), src->height());
    if (!srcRect.intersect(srcBounds)) {
        return;
    }
    sampleM.preTranslate(SkIntToScalar(srcRect.fLeft), SkIntToScalar(srcRect.fTop));
    drawState->addColorTextureEffect(src, sampleM);
    SkRect dstR = SkRect::MakeWH(SkIntToScalar(srcRect.width()), SkIntToScalar(srcRect.height()));
    fGpu->drawSimpleRect(dstR, NULL);
}

bool GrContext::writeRenderTargetPixels(GrRenderTarget* target,
                                        int left, int top, int width, int height,
                                        GrPixelConfig srcConfig,
                                        const void* buffer,
                                        size_t rowBytes,
                                        uint32_t flags) {
    SK_TRACE_EVENT0("GrContext::writeRenderTargetPixels");
    ASSERT_OWNED_RESOURCE(target);

    if (NULL == target) {
        target = fRenderTarget.get();
        if (NULL == target) {
            return false;
        }
    }

    // TODO: when underlying api has a direct way to do this we should use it (e.g. glDrawPixels on
    // desktop GL).

    // We will always call some form of writeTexturePixels and we will pass our flags on to it.
    // Thus, we don't perform a flush here since that call will do it (if the kNoFlush flag isn't
    // set.)

    // If the RT is also a texture and we don't have to premultiply then take the texture path.
    // We expect to be at least as fast or faster since it doesn't use an intermediate texture as
    // we do below.

#if !defined(SK_BUILD_FOR_MAC)
    // At least some drivers on the Mac get confused when glTexImage2D is called on a texture
    // attached to an FBO. The FBO still sees the old image. TODO: determine what OS versions and/or
    // HW is affected.
    if (NULL != target->asTexture() && !(kUnpremul_PixelOpsFlag & flags) &&
        fGpu->canWriteTexturePixels(target->asTexture(), srcConfig)) {
        return this->writeTexturePixels(target->asTexture(),
                                        left, top, width, height,
                                        srcConfig, buffer, rowBytes, flags);
    }
#endif

    // We ignore the preferred config unless it is a R/B swap of the src config. In that case
    // we will upload the original src data to a scratch texture but we will spoof it as the swapped
    // config. This scratch will then have R and B swapped. We correct for this by swapping again
    // when drawing the scratch to the dst using a conversion effect.
    bool swapRAndB = false;
    GrPixelConfig writeConfig = srcConfig;
    if (GrPixelConfigSwapRAndB(srcConfig) ==
        fGpu->preferredWritePixelsConfig(srcConfig, target->config())) {
        writeConfig = GrPixelConfigSwapRAndB(srcConfig);
        swapRAndB = true;
    }

    GrTextureDesc desc;
    desc.fWidth = width;
    desc.fHeight = height;
    desc.fConfig = writeConfig;
    GrAutoScratchTexture ast(this, desc);
    GrTexture* texture = ast.texture();
    if (NULL == texture) {
        return false;
    }

    SkAutoTUnref<const GrEffectRef> effect;
    SkMatrix textureMatrix;
    textureMatrix.setIDiv(texture->width(), texture->height());

    // allocate a tmp buffer and sw convert the pixels to premul
    SkAutoSTMalloc<128 * 128, uint32_t> tmpPixels(0);

    if (kUnpremul_PixelOpsFlag & flags) {
        if (!GrPixelConfigIs8888(srcConfig)) {
            return false;
        }
        effect.reset(this->createUPMToPMEffect(texture, swapRAndB, textureMatrix));
        // handle the unpremul step on the CPU if we couldn't create an effect to do it.
        if (NULL == effect) {
            SkCanvas::Config8888 srcConfig8888, dstConfig8888;
            SkDEBUGCODE(bool success = )
            grconfig_to_config8888(srcConfig, true, &srcConfig8888);
            SkASSERT(success);
            SkDEBUGCODE(success = )
            grconfig_to_config8888(srcConfig, false, &dstConfig8888);
            SkASSERT(success);
            const uint32_t* src = reinterpret_cast<const uint32_t*>(buffer);
            tmpPixels.reset(width * height);
            SkConvertConfig8888Pixels(tmpPixels.get(), 4 * width, dstConfig8888,
                                      src, rowBytes, srcConfig8888,
                                      width, height);
            buffer = tmpPixels.get();
            rowBytes = 4 * width;
        }
    }
    if (NULL == effect) {
        effect.reset(GrConfigConversionEffect::Create(texture,
                                                      swapRAndB,
                                                      GrConfigConversionEffect::kNone_PMConversion,
                                                      textureMatrix));
    }

    if (!this->writeTexturePixels(texture,
                                  0, 0, width, height,
                                  writeConfig, buffer, rowBytes,
                                  flags & ~kUnpremul_PixelOpsFlag)) {
        return false;
    }

    // writeRenderTargetPixels can be called in the midst of drawing another
    // object (e.g., when uploading a SW path rendering to the gpu while
    // drawing a rect) so preserve the current geometry.
    SkMatrix matrix;
    matrix.setTranslate(SkIntToScalar(left), SkIntToScalar(top));
    GrDrawTarget::AutoGeometryAndStatePush agasp(fGpu, GrDrawTarget::kReset_ASRInit, &matrix);
    GrDrawState* drawState = fGpu->drawState();
    SkASSERT(effect);
    drawState->addColorEffect(effect);

    drawState->setRenderTarget(target);

    fGpu->drawSimpleRect(SkRect::MakeWH(SkIntToScalar(width), SkIntToScalar(height)), NULL);
    return true;
}
////////////////////////////////////////////////////////////////////////////////

GrDrawTarget* GrContext::prepareToDraw(const GrPaint* paint,
                                       BufferedDraw buffered,
                                       AutoRestoreEffects* are,
                                       AutoCheckFlush* acf) {
    // All users of this draw state should be freeing up all effects when they're done.
    // Otherwise effects that own resources may keep those resources alive indefinitely.
    SkASSERT(0 == fDrawState->numColorStages() && 0 == fDrawState->numCoverageStages());

    if (kNo_BufferedDraw == buffered && kYes_BufferedDraw == fLastDrawWasBuffered) {
        fDrawBuffer->flush();
        fLastDrawWasBuffered = kNo_BufferedDraw;
    }
    ASSERT_OWNED_RESOURCE(fRenderTarget.get());
    if (NULL != paint) {
        SkASSERT(NULL != are);
        SkASSERT(NULL != acf);
        are->set(fDrawState);
        fDrawState->setFromPaint(*paint, fViewMatrix, fRenderTarget.get());
#if GR_DEBUG_PARTIAL_COVERAGE_CHECK
        if ((paint->hasMask() || 0xff != paint->fCoverage) &&
            !fGpu->canApplyCoverage()) {
            GrPrintf("Partial pixel coverage will be incorrectly blended.\n");
        }
#endif
    } else {
        fDrawState->reset(fViewMatrix);
        fDrawState->setRenderTarget(fRenderTarget.get());
    }
    GrDrawTarget* target;
    if (kYes_BufferedDraw == buffered) {
        fLastDrawWasBuffered = kYes_BufferedDraw;
        target = fDrawBuffer;
    } else {
        SkASSERT(kNo_BufferedDraw == buffered);
        fLastDrawWasBuffered = kNo_BufferedDraw;
        target = fGpu;
    }
    fDrawState->setState(GrDrawState::kClip_StateBit, NULL != fClip &&
                                                     !fClip->fClipStack->isWideOpen());
    target->setClip(fClip);
    SkASSERT(fDrawState == target->drawState());
    return target;
}

/*
 * This method finds a path renderer that can draw the specified path on
 * the provided target.
 * Due to its expense, the software path renderer has split out so it can
 * can be individually allowed/disallowed via the "allowSW" boolean.
 */
GrPathRenderer* GrContext::getPathRenderer(const SkPath& path,
                                           const SkStrokeRec& stroke,
                                           const GrDrawTarget* target,
                                           bool allowSW,
                                           GrPathRendererChain::DrawType drawType,
                                           SkPath::FillType fillType,
                                           GrPathRendererChain::StencilSupport* stencilSupport) {

    if (NULL == fPathRendererChain) {
        fPathRendererChain = SkNEW_ARGS(GrPathRendererChain, (this));
    }

    GrPathRenderer* pr = fPathRendererChain->getPathRenderer(path,
                                                             stroke,
                                                             target,
                                                             drawType,
                                                             fillType,
                                                             stencilSupport);

    if (NULL == pr && allowSW) {
        if (NULL == fSoftwarePathRenderer) {
            fSoftwarePathRenderer = SkNEW_ARGS(GrSoftwarePathRenderer, (this));
        }
        pr = fSoftwarePathRenderer;
        pr->setPath(path, fillType);
    }

    return pr;
}

////////////////////////////////////////////////////////////////////////////////
bool GrContext::isConfigRenderable(GrPixelConfig config, bool withMSAA) const {
    return fGpu->caps()->isConfigRenderable(config, withMSAA);
}

void GrContext::setupDrawBuffer() {
    SkASSERT(NULL == fDrawBuffer);
    SkASSERT(NULL == fDrawBufferVBAllocPool);
    SkASSERT(NULL == fDrawBufferIBAllocPool);

    fDrawBufferVBAllocPool =
        SkNEW_ARGS(GrVertexBufferAllocPool, (fGpu, false,
                                    DRAW_BUFFER_VBPOOL_BUFFER_SIZE,
                                    DRAW_BUFFER_VBPOOL_PREALLOC_BUFFERS));
    fDrawBufferIBAllocPool =
        SkNEW_ARGS(GrIndexBufferAllocPool, (fGpu, false,
                                   DRAW_BUFFER_IBPOOL_BUFFER_SIZE,
                                   DRAW_BUFFER_IBPOOL_PREALLOC_BUFFERS));

    fDrawBuffer = SkNEW_ARGS(GrInOrderDrawBuffer, (fGpu,
                                                   fDrawBufferVBAllocPool,
                                                   fDrawBufferIBAllocPool));

    fDrawBuffer->setDrawState(fDrawState);
}

GrDrawTarget* GrContext::getTextTarget() {
    return this->prepareToDraw(NULL, BUFFERED_DRAW, NULL, NULL);
}

const GrIndexBuffer* GrContext::getQuadIndexBuffer() const {
    return fGpu->getQuadIndexBuffer();
}

namespace {
void test_pm_conversions(GrContext* ctx, int* pmToUPMValue, int* upmToPMValue) {
    GrConfigConversionEffect::PMConversion pmToUPM;
    GrConfigConversionEffect::PMConversion upmToPM;
    GrConfigConversionEffect::TestForPreservingPMConversions(ctx, &pmToUPM, &upmToPM);
    *pmToUPMValue = pmToUPM;
    *upmToPMValue = upmToPM;
}
}

const GrEffectRef* GrContext::createPMToUPMEffect(GrTexture* texture,
                                                  bool swapRAndB,
                                                  const SkMatrix& matrix) {
    if (!fDidTestPMConversions) {
        test_pm_conversions(this, &fPMToUPMConversion, &fUPMToPMConversion);
        fDidTestPMConversions = true;
    }
    GrConfigConversionEffect::PMConversion pmToUPM =
        static_cast<GrConfigConversionEffect::PMConversion>(fPMToUPMConversion);
    if (GrConfigConversionEffect::kNone_PMConversion != pmToUPM) {
        return GrConfigConversionEffect::Create(texture, swapRAndB, pmToUPM, matrix);
    } else {
        return NULL;
    }
}

const GrEffectRef* GrContext::createUPMToPMEffect(GrTexture* texture,
                                                  bool swapRAndB,
                                                  const SkMatrix& matrix) {
    if (!fDidTestPMConversions) {
        test_pm_conversions(this, &fPMToUPMConversion, &fUPMToPMConversion);
        fDidTestPMConversions = true;
    }
    GrConfigConversionEffect::PMConversion upmToPM =
        static_cast<GrConfigConversionEffect::PMConversion>(fUPMToPMConversion);
    if (GrConfigConversionEffect::kNone_PMConversion != upmToPM) {
        return GrConfigConversionEffect::Create(texture, swapRAndB, upmToPM, matrix);
    } else {
        return NULL;
    }
}

GrPath* GrContext::createPath(const SkPath& inPath, const SkStrokeRec& stroke) {
    SkASSERT(fGpu->caps()->pathRenderingSupport());

    // TODO: now we add to fTextureCache. This should change to fResourceCache.
    GrResourceKey resourceKey = GrPath::ComputeKey(inPath, stroke);
    GrPath* path = static_cast<GrPath*>(fTextureCache->find(resourceKey));
    if (NULL != path && path->isEqualTo(inPath, stroke)) {
        path->ref();
    } else {
        path = fGpu->createPath(inPath, stroke);
        fTextureCache->purgeAsNeeded(1, path->sizeInBytes());
        fTextureCache->addResource(resourceKey, path);
    }
    return path;
}

///////////////////////////////////////////////////////////////////////////////
#if GR_CACHE_STATS
void GrContext::printCacheStats() const {
    fTextureCache->printStats();
}
#endif