gpu/dawn/GrDawnGpu.cpp

/*
 * Copyright 2019 Google Inc.
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */

#include "src/gpu/dawn/GrDawnGpu.h"

#include "include/gpu/GrBackendSemaphore.h"
#include "include/gpu/GrBackendSurface.h"
#include "include/gpu/GrContextOptions.h"
#include "include/gpu/GrDirectContext.h"
#include "src/core/SkConvertPixels.h"
#include "src/gpu/GrDataUtils.h"
#include "src/gpu/GrDirectContextPriv.h"
#include "src/gpu/GrGeometryProcessor.h"
#include "src/gpu/GrGpuResourceCacheAccess.h"
#include "src/gpu/GrPipeline.h"
#include "src/gpu/GrRenderTarget.h"
#include "src/gpu/GrSemaphore.h"
#include "src/gpu/GrStencilSettings.h"
#include "src/gpu/GrTexture.h"
#include "src/gpu/dawn/GrDawnAttachment.h"
#include "src/gpu/dawn/GrDawnBuffer.h"
#include "src/gpu/dawn/GrDawnCaps.h"
#include "src/gpu/dawn/GrDawnOpsRenderPass.h"
#include "src/gpu/dawn/GrDawnProgramBuilder.h"
#include "src/gpu/dawn/GrDawnRenderTarget.h"
#include "src/gpu/dawn/GrDawnTexture.h"
#include "src/gpu/dawn/GrDawnUtil.h"

#include "src/core/SkAutoMalloc.h"
#include "src/core/SkMipmap.h"
#include "src/sksl/SkSLCompiler.h"

#if !defined(SK_BUILD_FOR_WIN)
#include <unistd.h>
#endif // !defined(SK_BUILD_FOR_WIN)

static const int kMaxRenderPipelineEntries = 1024;

namespace {

class Fence {
public:
    Fence(const wgpu::Device& device, const wgpu::Fence& fence)
      : fDevice(device), fFence(fence), fCalled(false) {
        fFence.OnCompletion(0, callback, this);
    }

    static void callback(WGPUFenceCompletionStatus status, void* userData) {
        Fence* fence = static_cast<Fence*>(userData);
        fence->fCalled = true;
    }

    bool check() {
        fDevice.Tick();
        return fCalled;
    }

    wgpu::Fence fence() { return fFence; }

private:
    wgpu::Device            fDevice;
    wgpu::Fence             fFence;
    bool                    fCalled;
};

}

static wgpu::FilterMode to_dawn_filter_mode(GrSamplerState::Filter filter) {
    switch (filter) {
        case GrSamplerState::Filter::kNearest:
            return wgpu::FilterMode::Nearest;
        case GrSamplerState::Filter::kLinear:
            return wgpu::FilterMode::Linear;
        default:
            SkASSERT(!"unsupported filter mode");
            return wgpu::FilterMode::Nearest;
    }
}

static wgpu::FilterMode to_dawn_mipmap_mode(GrSamplerState::MipmapMode mode) {
    switch (mode) {
        case GrSamplerState::MipmapMode::kNone:
            // Fall-through (Dawn does not have an equivalent for "None")
        case GrSamplerState::MipmapMode::kNearest:
            return wgpu::FilterMode::Nearest;
        case GrSamplerState::MipmapMode::kLinear:
            return wgpu::FilterMode::Linear;
        default:
            SkASSERT(!"unsupported filter mode");
            return wgpu::FilterMode::Nearest;
    }
}

static wgpu::AddressMode to_dawn_address_mode(GrSamplerState::WrapMode wrapMode) {
    switch (wrapMode) {
        case GrSamplerState::WrapMode::kClamp:
            return wgpu::AddressMode::ClampToEdge;
        case GrSamplerState::WrapMode::kRepeat:
            return wgpu::AddressMode::Repeat;
        case GrSamplerState::WrapMode::kMirrorRepeat:
            return wgpu::AddressMode::MirrorRepeat;
        case GrSamplerState::WrapMode::kClampToBorder:
            SkASSERT(!"unsupported address mode");
    }
    SkASSERT(!"unsupported address mode");
    return wgpu::AddressMode::ClampToEdge;
}

sk_sp<GrGpu> GrDawnGpu::Make(const wgpu::Device& device,
                             const GrContextOptions& options, GrDirectContext* direct) {
    if (!device) {
        return nullptr;
    }

    return sk_sp<GrGpu>(new GrDawnGpu(direct, options, device));
}

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

GrDawnGpu::GrDawnGpu(GrDirectContext* direct, const GrContextOptions& options,
                     const wgpu::Device& device)
        : INHERITED(direct)
        , fDevice(device)
        , fQueue(device.GetDefaultQueue())
        , fUniformRingBuffer(this, wgpu::BufferUsage::Uniform)
        , fStagingBufferManager(this)
        , fRenderPipelineCache(kMaxRenderPipelineEntries)
        , fFinishCallbacks(this) {
    fCaps.reset(new GrDawnCaps(options));
    fCompiler.reset(new SkSL::Compiler(fCaps->shaderCaps()));
}

GrDawnGpu::~GrDawnGpu() {
    this->waitOnAllBusyStagingBuffers();
}

void GrDawnGpu::disconnect(DisconnectType type) {
    if (DisconnectType::kCleanup == type) {
        this->waitOnAllBusyStagingBuffers();
    }
    fStagingBufferManager.reset();
    fQueue = nullptr;
    fDevice = nullptr;
    INHERITED::disconnect(type);
}

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

GrOpsRenderPass* GrDawnGpu::onGetOpsRenderPass(
        GrRenderTarget* rt,
        GrAttachment*,
        GrSurfaceOrigin origin,
        const SkIRect& bounds,
        const GrOpsRenderPass::LoadAndStoreInfo& colorInfo,
        const GrOpsRenderPass::StencilLoadAndStoreInfo& stencilInfo,
        const SkTArray<GrSurfaceProxy*, true>& sampledProxies,
        GrXferBarrierFlags renderPassXferBarriers) {
    fOpsRenderPass.reset(new GrDawnOpsRenderPass(this, rt, origin, colorInfo, stencilInfo));
    return fOpsRenderPass.get();
}

///////////////////////////////////////////////////////////////////////////////
sk_sp<GrGpuBuffer> GrDawnGpu::onCreateBuffer(size_t size, GrGpuBufferType type,
                                             GrAccessPattern accessPattern, const void* data) {
    sk_sp<GrGpuBuffer> b(new GrDawnBuffer(this, size, type, accessPattern));
    if (data && b) {
        b->updateData(data, size);
    }
    return b;
}

////////////////////////////////////////////////////////////////////////////////
bool GrDawnGpu::onWritePixels(GrSurface* surface, int left, int top, int width, int height,
                              GrColorType surfaceColorType, GrColorType srcColorType,
                              const GrMipLevel texels[], int mipLevelCount,
                              bool prepForTexSampling) {
    GrDawnTexture* texture = static_cast<GrDawnTexture*>(surface->asTexture());
    if (!texture) {
        return false;
    }
    this->uploadTextureData(srcColorType, texels, mipLevelCount,
                            SkIRect::MakeXYWH(left, top, width, height), texture->texture());
    if (mipLevelCount < texture->maxMipmapLevel() + 1) {
        texture->markMipmapsDirty();
    }
    return true;
}

bool GrDawnGpu::onTransferPixelsTo(GrTexture* texture, int left, int top, int width, int height,
                                   GrColorType textureColorType, GrColorType bufferColorType,
                                   GrGpuBuffer* transferBuffer, size_t bufferOffset,
                                   size_t rowBytes) {
    SkASSERT(!"unimplemented");
    return false;
}

bool GrDawnGpu::onTransferPixelsFrom(GrSurface* surface, int left, int top, int width, int height,
                                     GrColorType surfaceColorType, GrColorType bufferColorType,
                                     GrGpuBuffer* transferBuffer, size_t offset) {
    SkASSERT(!"unimplemented");
    return false;
}

////////////////////////////////////////////////////////////////////////////////
sk_sp<GrTexture> GrDawnGpu::onCreateTexture(SkISize dimensions,
                                            const GrBackendFormat& backendFormat,
                                            GrRenderable renderable,
                                            int renderTargetSampleCnt,
                                            SkBudgeted budgeted,
                                            GrProtected,
                                            int mipLevelCount,
                                            uint32_t levelClearMask) {
    if (levelClearMask) {
        return nullptr;
    }

    wgpu::TextureFormat format;
    if (!backendFormat.asDawnFormat(&format)) {
        return nullptr;
    }

    GrMipmapStatus mipmapStatus =
        mipLevelCount > 1 ? GrMipmapStatus::kDirty : GrMipmapStatus::kNotAllocated;

    return GrDawnTexture::Make(this, dimensions, format, renderable, renderTargetSampleCnt,
                               budgeted, mipLevelCount, mipmapStatus);
}

sk_sp<GrTexture> GrDawnGpu::onCreateCompressedTexture(SkISize dimensions, const GrBackendFormat&,
                                                      SkBudgeted, GrMipmapped, GrProtected,
                                                      const void* data, size_t dataSize) {
    SkASSERT(!"unimplemented");
    return nullptr;
}

sk_sp<GrTexture> GrDawnGpu::onWrapBackendTexture(const GrBackendTexture& backendTex,
                                                 GrWrapOwnership ownership,
                                                 GrWrapCacheable cacheable,
                                                 GrIOType ioType) {
    GrDawnTextureInfo info;
    if (!backendTex.getDawnTextureInfo(&info)) {
        return nullptr;
    }

    SkISize dimensions = { backendTex.width(), backendTex.height() };
    return GrDawnTexture::MakeWrapped(this, dimensions, GrRenderable::kNo, 1, cacheable, ioType,
                                      info);
}

sk_sp<GrTexture> GrDawnGpu::onWrapCompressedBackendTexture(const GrBackendTexture& backendTex,
                                                           GrWrapOwnership ownership,
                                                           GrWrapCacheable cacheable) {
    return nullptr;
}

sk_sp<GrTexture> GrDawnGpu::onWrapRenderableBackendTexture(const GrBackendTexture& tex,
                                                           int sampleCnt,
                                                           GrWrapOwnership,
                                                           GrWrapCacheable cacheable) {
    GrDawnTextureInfo info;
    if (!tex.getDawnTextureInfo(&info) || !info.fTexture) {
        return nullptr;
    }

    SkISize dimensions = { tex.width(), tex.height() };
    sampleCnt = this->caps()->getRenderTargetSampleCount(sampleCnt, tex.getBackendFormat());
    if (sampleCnt < 1) {
        return nullptr;
    }

    sk_sp<GrTexture> result = GrDawnTexture::MakeWrapped(this, dimensions, GrRenderable::kYes,
                                                         sampleCnt, cacheable, kRW_GrIOType, info);
    result->markMipmapsDirty();
    return result;
}

sk_sp<GrRenderTarget> GrDawnGpu::onWrapBackendRenderTarget(const GrBackendRenderTarget& rt) {
    GrDawnRenderTargetInfo info;
    if (!rt.getDawnRenderTargetInfo(&info) || !info.fTextureView) {
        return nullptr;
    }

    SkISize dimensions = { rt.width(), rt.height() };
    int sampleCnt = 1;
    return GrDawnRenderTarget::MakeWrapped(this, dimensions, sampleCnt, info);
}

sk_sp<GrAttachment> GrDawnGpu::makeStencilAttachmentForRenderTarget(const GrRenderTarget* rt,
                                                                    SkISize dimensions,
                                                                    int numStencilSamples) {
    fStats.incStencilAttachmentCreates();
    return GrDawnAttachment::MakeStencil(this, dimensions, numStencilSamples);
}

GrBackendTexture GrDawnGpu::onCreateBackendTexture(SkISize dimensions,
                                                   const GrBackendFormat& backendFormat,
                                                   GrRenderable renderable,
                                                   GrMipmapped mipMapped,
                                                   GrProtected isProtected) {
    wgpu::TextureFormat format;
    if (!backendFormat.asDawnFormat(&format)) {
        return GrBackendTexture();
    }

    wgpu::TextureDescriptor desc;
    desc.usage =
        wgpu::TextureUsage::Sampled |
        wgpu::TextureUsage::CopySrc |
        wgpu::TextureUsage::CopyDst;

    if (GrRenderable::kYes == renderable) {
        desc.usage |= wgpu::TextureUsage::OutputAttachment;
    }

    int numMipLevels = 1;
    if (mipMapped == GrMipmapped::kYes) {
        numMipLevels = SkMipmap::ComputeLevelCount(dimensions.width(), dimensions.height()) + 1;
    }

    desc.size.width = dimensions.width();
    desc.size.height = dimensions.height();
    desc.size.depth = 1;
    desc.format = format;
    desc.mipLevelCount = numMipLevels;

    wgpu::Texture tex = this->device().CreateTexture(&desc);

    GrDawnTextureInfo info;
    info.fTexture = tex;
    info.fFormat = desc.format;
    info.fLevelCount = desc.mipLevelCount;
    return GrBackendTexture(dimensions.width(), dimensions.height(), info);
}

void GrDawnGpu::uploadTextureData(GrColorType srcColorType, const GrMipLevel texels[],
                                  int mipLevelCount, const SkIRect& rect,
                                  wgpu::Texture texture) {
    uint32_t x = rect.x();
    uint32_t y = rect.y();
    uint32_t width = rect.width();
    uint32_t height = rect.height();

    for (int i = 0; i < mipLevelCount; i++) {
        const void* src = texels[i].fPixels;
        size_t srcRowBytes = texels[i].fRowBytes;
        SkColorType colorType = GrColorTypeToSkColorType(srcColorType);
        size_t trimRowBytes = width * SkColorTypeBytesPerPixel(colorType);
        size_t dstRowBytes = GrDawnRoundRowBytes(trimRowBytes);
        size_t size = dstRowBytes * height;
        GrStagingBufferManager::Slice slice =
                this->stagingBufferManager()->allocateStagingBufferSlice(size);
        SkRectMemcpy(slice.fOffsetMapPtr, dstRowBytes, src, srcRowBytes, trimRowBytes, height);

        wgpu::BufferCopyView srcBuffer = {};
        srcBuffer.buffer = static_cast<GrDawnBuffer*>(slice.fBuffer)->get();
        srcBuffer.layout.offset = slice.fOffset;
        srcBuffer.layout.bytesPerRow = dstRowBytes;
        srcBuffer.layout.rowsPerImage = height;

        wgpu::TextureCopyView dstTexture;
        dstTexture.texture = texture;
        dstTexture.mipLevel = i;
        dstTexture.origin = {x, y, 0};

        wgpu::Extent3D copySize = {width, height, 1};
        this->getCopyEncoder().CopyBufferToTexture(&srcBuffer, &dstTexture, &copySize);
        x /= 2;
        y /= 2;
        width = std::max(1u, width / 2);
        height = std::max(1u, height / 2);
    }
}

bool GrDawnGpu::onUpdateBackendTexture(const GrBackendTexture& backendTexture,
                                       sk_sp<GrRefCntedCallback> finishedCallback,
                                       const BackendTextureData* data) {
    GrDawnTextureInfo info;
    SkAssertResult(backendTexture.getDawnTextureInfo(&info));

    size_t bpp = GrDawnBytesPerBlock(info.fFormat);
    size_t baseLayerSize = bpp * backendTexture.width() * backendTexture.height();
    const void* pixels;
    SkAutoMalloc defaultStorage(baseLayerSize);
    if (data && data->type() == BackendTextureData::Type::kPixmaps) {
        SkTDArray<GrMipLevel> texels;
        GrColorType colorType = SkColorTypeToGrColorType(data->pixmap(0).colorType());
        int numMipLevels = info.fLevelCount;
        texels.append(numMipLevels);
        for (int i = 0; i < numMipLevels; ++i) {
            texels[i] = {data->pixmap(i).addr(), data->pixmap(i).rowBytes()};
        }
        SkIRect dstRect = SkIRect::MakeSize(backendTexture.dimensions());
        this->uploadTextureData(colorType, texels.begin(), texels.count(), dstRect, info.fTexture);
        return true;
    }
    pixels = defaultStorage.get();
    GrColorType colorType;
    if (!GrDawnFormatToGrColorType(info.fFormat, &colorType)) {
        return false;
    }
    SkISize size{backendTexture.width(), backendTexture.height()};
    GrImageInfo imageInfo(colorType, kUnpremul_SkAlphaType, nullptr, size);
    GrClearImage(imageInfo, defaultStorage.get(), bpp * backendTexture.width(), data->color());
    wgpu::Device device = this->device();
    wgpu::CommandEncoder copyEncoder = this->getCopyEncoder();
    int w = backendTexture.width(), h = backendTexture.height();
    for (uint32_t i = 0; i < info.fLevelCount; i++) {
        size_t origRowBytes = bpp * w;
        size_t rowBytes = GrDawnRoundRowBytes(origRowBytes);
        size_t size = rowBytes * h;
        GrStagingBufferManager::Slice stagingBuffer =
                this->stagingBufferManager()->allocateStagingBufferSlice(size);
        if (rowBytes == origRowBytes) {
            memcpy(stagingBuffer.fOffsetMapPtr, pixels, size);
        } else {
            const char* src = static_cast<const char*>(pixels);
            char* dst = static_cast<char*>(stagingBuffer.fOffsetMapPtr);
            for (int row = 0; row < h; row++) {
                memcpy(dst, src, origRowBytes);
                dst += rowBytes;
                src += origRowBytes;
            }
        }
        wgpu::BufferCopyView srcBuffer = {};
        srcBuffer.buffer = static_cast<GrDawnBuffer*>(stagingBuffer.fBuffer)->get();
        srcBuffer.layout.offset = stagingBuffer.fOffset;
        srcBuffer.layout.bytesPerRow = rowBytes;
        srcBuffer.layout.rowsPerImage = h;
        wgpu::TextureCopyView dstTexture;
        dstTexture.texture = info.fTexture;
        dstTexture.mipLevel = i;
        dstTexture.origin = {0, 0, 0};
        wgpu::Extent3D copySize = {(uint32_t)w, (uint32_t)h, 1};
        copyEncoder.CopyBufferToTexture(&srcBuffer, &dstTexture, &copySize);
        w = std::max(1, w / 2);
        h = std::max(1, h / 2);
    }
    return true;
}

GrBackendTexture GrDawnGpu::onCreateCompressedBackendTexture(
        SkISize dimensions, const GrBackendFormat&, GrMipmapped, GrProtected) {
    return {};
}

bool GrDawnGpu::onUpdateCompressedBackendTexture(const GrBackendTexture&,
                                                 sk_sp<GrRefCntedCallback> finishedCallback,
                                                 const BackendTextureData*) {
    return false;
}

void GrDawnGpu::deleteBackendTexture(const GrBackendTexture& tex) {
    GrDawnTextureInfo info;
    if (tex.getDawnTextureInfo(&info)) {
        info.fTexture = nullptr;
    }
}

bool GrDawnGpu::compile(const GrProgramDesc&, const GrProgramInfo&) {
    return false;
}

#if GR_TEST_UTILS
bool GrDawnGpu::isTestingOnlyBackendTexture(const GrBackendTexture& tex) const {
    GrDawnTextureInfo info;
    if (!tex.getDawnTextureInfo(&info)) {
        return false;
    }

    return info.fTexture.Get();
}

GrBackendRenderTarget GrDawnGpu::createTestingOnlyBackendRenderTarget(SkISize dimensions,
                                                                      GrColorType colorType,
                                                                      int sampleCnt,
                                                                      GrProtected isProtected) {
    if (dimensions.width()  > this->caps()->maxTextureSize() ||
        dimensions.height() > this->caps()->maxTextureSize()) {
        return {};
    }

    // We don't support MSAA in this backend yet.
    if (sampleCnt != 1) {
        return {};
    }

    if (isProtected == GrProtected::kYes) {
        return {};
    }

    wgpu::TextureFormat format;
    if (!GrColorTypeToDawnFormat(colorType, &format)) {
        return {};
    }

    wgpu::TextureDescriptor desc;
    desc.usage =
        wgpu::TextureUsage::CopySrc |
        wgpu::TextureUsage::OutputAttachment;

    desc.size.width = dimensions.width();
    desc.size.height = dimensions.height();
    desc.size.depth = 1;
    desc.format = format;

    wgpu::Texture tex = this->device().CreateTexture(&desc);

    GrDawnRenderTargetInfo info;
    info.fTextureView = tex.CreateView();
    info.fFormat = desc.format;
    info.fLevelCount = desc.mipLevelCount;

    return GrBackendRenderTarget(dimensions.width(), dimensions.height(), 1, 0, info);
}

void GrDawnGpu::deleteTestingOnlyBackendRenderTarget(const GrBackendRenderTarget& rt) {
    GrDawnRenderTargetInfo info;
    if (rt.getDawnRenderTargetInfo(&info)) {
        info.fTextureView = nullptr;
    }
}

void GrDawnGpu::testingOnly_flushGpuAndSync() {
    this->submitToGpu(true);
}

#endif

void GrDawnGpu::addFinishedProc(GrGpuFinishedProc finishedProc,
                                GrGpuFinishedContext finishedContext) {
    fFinishCallbacks.add(finishedProc, finishedContext);
}

void GrDawnGpu::checkForCompletedStagingBuffers() {
    // We expect all the buffer maps to trigger in order of submission so we bail after the first
    // non finished map since we always push new busy buffers to the back of our list.
    while (!fBusyStagingBuffers.empty() && fBusyStagingBuffers.front()->isMapped()) {
        fBusyStagingBuffers.pop_front();
    }
}

void GrDawnGpu::waitOnAllBusyStagingBuffers() {
    while (!fBusyStagingBuffers.empty()) {
        fDevice.Tick();
        this->checkForCompletedStagingBuffers();
    }
}

void GrDawnGpu::takeOwnershipOfBuffer(sk_sp<GrGpuBuffer> buffer) {
    fSubmittedStagingBuffers.push_back(std::move(buffer));
}


static void callback(WGPUFenceCompletionStatus status, void* userData) {
    *static_cast<bool*>(userData) = true;
}

bool GrDawnGpu::onSubmitToGpu(bool syncCpu) {
    this->flushCopyEncoder();
    if (!fCommandBuffers.empty()) {
        fQueue.Submit(fCommandBuffers.size(), &fCommandBuffers.front());
        fCommandBuffers.clear();
    }

    this->moveStagingBuffersToBusyAndMapAsync();
    if (syncCpu) {
        wgpu::FenceDescriptor desc;
        wgpu::Fence fence = fQueue.CreateFence(&desc);
        bool called = false;
        fence.OnCompletion(0, callback, &called);
        while (!called) {
            fDevice.Tick();
        }
        fFinishCallbacks.callAll(true);
    }

    this->checkForCompletedStagingBuffers();

    return true;
}

static wgpu::Texture get_dawn_texture_from_surface(GrSurface* src) {
    if (auto t = static_cast<GrDawnTexture*>(src->asTexture())) {
        return t->texture();
    } else {
        return nullptr;
    }
}

bool GrDawnGpu::onCopySurface(GrSurface* dst,
                              GrSurface* src,
                              const SkIRect& srcRect,
                              const SkIPoint& dstPoint) {
    wgpu::Texture srcTexture = get_dawn_texture_from_surface(src);
    wgpu::Texture dstTexture = get_dawn_texture_from_surface(dst);
    if (!srcTexture || !dstTexture) {
        return false;
    }

    uint32_t width = srcRect.width(), height = srcRect.height();

    wgpu::TextureCopyView srcTextureView, dstTextureView;
    srcTextureView.texture = srcTexture;
    srcTextureView.origin = {(uint32_t) srcRect.x(), (uint32_t) srcRect.y(), 0};
    dstTextureView.texture = dstTexture;
    dstTextureView.origin = {(uint32_t) dstPoint.x(), (uint32_t) dstPoint.y(), 0};

    wgpu::Extent3D copySize = {width, height, 1};
    this->getCopyEncoder().CopyTextureToTexture(&srcTextureView, &dstTextureView, &copySize);
    return true;
}

static void callback(WGPUBufferMapAsyncStatus status, void* userdata) {
    *static_cast<bool*>(userdata) = true;
}

bool GrDawnGpu::onReadPixels(GrSurface* surface, int left, int top, int width, int height,
                             GrColorType surfaceColorType, GrColorType dstColorType, void* buffer,
                             size_t rowBytes) {
    wgpu::Texture tex = get_dawn_texture_from_surface(surface);

    if (!tex || 0 == rowBytes) {
        return false;
    }
    size_t origRowBytes = rowBytes;
    int origSizeInBytes = origRowBytes * height;
    rowBytes = GrDawnRoundRowBytes(rowBytes);
    int sizeInBytes = rowBytes * height;

    wgpu::BufferDescriptor desc;
    desc.usage = wgpu::BufferUsage::CopyDst | wgpu::BufferUsage::MapRead;
    desc.size = sizeInBytes;

    wgpu::Buffer buf = device().CreateBuffer(&desc);

    wgpu::TextureCopyView srcTexture;
    srcTexture.texture = tex;
    srcTexture.origin = {(uint32_t) left, (uint32_t) top, 0};

    wgpu::BufferCopyView dstBuffer = {};
    dstBuffer.buffer = buf;
    dstBuffer.layout.offset = 0;
    dstBuffer.layout.bytesPerRow = rowBytes;
    dstBuffer.layout.rowsPerImage = height;

    wgpu::Extent3D copySize = {(uint32_t) width, (uint32_t) height, 1};
    this->getCopyEncoder().CopyTextureToBuffer(&srcTexture, &dstBuffer, &copySize);
    this->submitToGpu(true);

    bool mapped = false;
    buf.MapAsync(wgpu::MapMode::Read, 0, 0, callback, &mapped);
    while (!mapped) {
        device().Tick();
    }
    const void* readPixelsPtr = buf.GetConstMappedRange();

    if (rowBytes == origRowBytes) {
        memcpy(buffer, readPixelsPtr, origSizeInBytes);
    } else {
        const char* src = static_cast<const char*>(readPixelsPtr);
        char* dst = static_cast<char*>(buffer);
        for (int row = 0; row < height; row++) {
            memcpy(dst, src, origRowBytes);
            dst += origRowBytes;
            src += rowBytes;
        }
    }
    buf.Unmap();
    return true;
}

bool GrDawnGpu::onRegenerateMipMapLevels(GrTexture* tex) {
    this->flushCopyEncoder();
    GrDawnTexture* src = static_cast<GrDawnTexture*>(tex);
    int srcWidth = tex->width();
    int srcHeight = tex->height();

    // SkMipmap doesn't include the base level in the level count so we have to add 1
    uint32_t levelCount = SkMipmap::ComputeLevelCount(tex->width(), tex->height()) + 1;

    // Create a temporary texture for mipmap generation, then copy to source.
    // We have to do this even for renderable textures, since GrDawnRenderTarget currently only
    // contains a view, not a texture.
    wgpu::TextureDescriptor texDesc;
    texDesc.usage = wgpu::TextureUsage::Sampled |
                    wgpu::TextureUsage::CopySrc |
                    wgpu::TextureUsage::OutputAttachment;
    texDesc.size.width = (tex->width() + 1) / 2;
    texDesc.size.height = (tex->height() + 1) / 2;
    texDesc.size.depth = 1;
    texDesc.mipLevelCount = levelCount - 1;
    texDesc.format = src->format();
    wgpu::Texture dstTexture = fDevice.CreateTexture(&texDesc);

    const char* vs =
        "layout(location = 0) out float2 texCoord;\n"
        "float2 positions[4] = float2[4](float2(-1.0, 1.0),\n"
                                        "float2(1.0, 1.0),\n"
                                        "float2(-1.0, -1.0),\n"
                                        "float2(1.0, -1.0));\n"
        "float2 texCoords[4] = float2[4](float2(0.0, 0.0),\n"
                                        "float2(1.0, 0.0),\n"
                                        "float2(0.0, 1.0),\n"
                                        "float2(1.0, 1.0));\n"
        "void main() {\n"
        "    sk_Position = float4(positions[sk_VertexID], 0.0, 1.0);\n"
        "    texCoord = texCoords[sk_VertexID];\n"
        "}\n";
    SkSL::String vsSPIRV =
        this->SkSLToSPIRV(vs, SkSL::Program::kVertex_Kind, false, 0, nullptr);

    const char* fs =
        "layout(set = 0, binding = 0) uniform sampler samp;\n"
        "layout(set = 0, binding = 1) uniform texture2D tex;\n"
        "layout(location = 0) in float2 texCoord;\n"
        "void main() {\n"
        "    sk_FragColor = sample(makeSampler2D(tex, samp), texCoord);\n"
        "}\n";
    SkSL::String fsSPIRV =
        this->SkSLToSPIRV(fs, SkSL::Program::kFragment_Kind, false, 0, nullptr);

    wgpu::ProgrammableStageDescriptor vsDesc;
    vsDesc.module = this->createShaderModule(vsSPIRV);
    vsDesc.entryPoint = "main";

    wgpu::ProgrammableStageDescriptor fsDesc;
    fsDesc.module = this->createShaderModule(fsSPIRV);
    fsDesc.entryPoint = "main";

    wgpu::VertexStateDescriptor vertexStateDesc;
    vertexStateDesc.indexFormat = wgpu::IndexFormat::Uint32;

    wgpu::ColorStateDescriptor csDesc;
    csDesc.format = static_cast<GrDawnTexture*>(tex)->format();

    wgpu::RenderPipelineDescriptor renderPipelineDesc;
    renderPipelineDesc.vertexStage = vsDesc;
    renderPipelineDesc.fragmentStage = &fsDesc;
    renderPipelineDesc.vertexState = &vertexStateDesc;
    renderPipelineDesc.primitiveTopology = wgpu::PrimitiveTopology::TriangleStrip;
    renderPipelineDesc.colorStateCount = 1;
    renderPipelineDesc.colorStates = &csDesc;
    wgpu::RenderPipeline pipeline = fDevice.CreateRenderPipeline(&renderPipelineDesc);

    wgpu::BindGroupLayout bgl = pipeline.GetBindGroupLayout(0);
    wgpu::TextureViewDescriptor srcViewDesc;
    srcViewDesc.mipLevelCount = 1;
    wgpu::TextureView srcView = src->texture().CreateView(&srcViewDesc);
    wgpu::SamplerDescriptor samplerDesc;
    samplerDesc.minFilter = wgpu::FilterMode::Linear;
    wgpu::Sampler sampler = fDevice.CreateSampler(&samplerDesc);
    wgpu::CommandEncoder commandEncoder = fDevice.CreateCommandEncoder();
    for (uint32_t mipLevel = 0; mipLevel < texDesc.mipLevelCount; mipLevel++) {
        int dstWidth = std::max(1, srcWidth / 2);
        int dstHeight = std::max(1, srcHeight / 2);
        wgpu::TextureViewDescriptor dstViewDesc;
        dstViewDesc.format = static_cast<GrDawnTexture*>(tex)->format();
        dstViewDesc.dimension = wgpu::TextureViewDimension::e2D;
        dstViewDesc.baseMipLevel = mipLevel;
        dstViewDesc.mipLevelCount = 1;
        wgpu::TextureView dstView = dstTexture.CreateView(&dstViewDesc);
        wgpu::BindGroupEntry bge[2];
        bge[0].binding = 0;
        bge[0].sampler = sampler;
        bge[1].binding = 1;
        bge[1].textureView = srcView;
        wgpu::BindGroupDescriptor bgDesc;
        bgDesc.layout = bgl;
        bgDesc.entryCount = 2;
        bgDesc.entries = bge;
        wgpu::BindGroup bindGroup = fDevice.CreateBindGroup(&bgDesc);
        wgpu::RenderPassColorAttachmentDescriptor colorAttachment;
        colorAttachment.attachment = dstView;
        colorAttachment.clearColor = { 0.0f, 0.0f, 0.0f, 0.0f };
        colorAttachment.loadOp = wgpu::LoadOp::Load;
        colorAttachment.storeOp = wgpu::StoreOp::Store;
        wgpu::RenderPassColorAttachmentDescriptor* colorAttachments = { &colorAttachment };
        wgpu::RenderPassDescriptor renderPassDesc;
        renderPassDesc.colorAttachmentCount = 1;
        renderPassDesc.colorAttachments = colorAttachments;
        wgpu::RenderPassEncoder rpe = commandEncoder.BeginRenderPass(&renderPassDesc);
        rpe.SetPipeline(pipeline);
        rpe.SetBindGroup(0, bindGroup);
        rpe.Draw(4, 1, 0, 0);
        rpe.EndPass();

        wgpu::Extent3D copySize = {(uint32_t)dstWidth, (uint32_t)dstHeight, 1};
        wgpu::TextureCopyView srcCopyView;
        srcCopyView.texture = dstTexture;
        srcCopyView.mipLevel = mipLevel;
        wgpu::TextureCopyView dstCopyView;
        dstCopyView.mipLevel = mipLevel + 1;
        dstCopyView.texture = src->texture();
        commandEncoder.CopyTextureToTexture(&srcCopyView, &dstCopyView, &copySize);

        srcHeight = dstHeight;
        srcWidth = dstWidth;
        srcView = dstView;
    }
    fCommandBuffers.push_back(commandEncoder.Finish());
    return true;
}

void GrDawnGpu::submit(GrOpsRenderPass* renderPass) {
    this->flushCopyEncoder();
    static_cast<GrDawnOpsRenderPass*>(renderPass)->submit();
}

GrFence SK_WARN_UNUSED_RESULT GrDawnGpu::insertFence() {
    wgpu::FenceDescriptor desc;
    wgpu::Fence fence = fQueue.CreateFence(&desc);
    return reinterpret_cast<GrFence>(new Fence(fDevice, fence));
}

bool GrDawnGpu::waitFence(GrFence fence) {
    return reinterpret_cast<Fence*>(fence)->check();
}

void GrDawnGpu::deleteFence(GrFence fence) const {
    delete reinterpret_cast<Fence*>(fence);
}

std::unique_ptr<GrSemaphore> SK_WARN_UNUSED_RESULT GrDawnGpu::makeSemaphore(bool isOwned) {
    SkASSERT(!"unimplemented");
    return nullptr;
}

std::unique_ptr<GrSemaphore> GrDawnGpu::wrapBackendSemaphore(
        const GrBackendSemaphore& semaphore,
        GrResourceProvider::SemaphoreWrapType wrapType,
        GrWrapOwnership ownership) {
    SkASSERT(!"unimplemented");
    return nullptr;
}

void GrDawnGpu::insertSemaphore(GrSemaphore* semaphore) {
    SkASSERT(!"unimplemented");
}

void GrDawnGpu::waitSemaphore(GrSemaphore* semaphore) {
    SkASSERT(!"unimplemented");
}

void GrDawnGpu::checkFinishProcs() {
    fFinishCallbacks.check();
}

std::unique_ptr<GrSemaphore> GrDawnGpu::prepareTextureForCrossContextUsage(GrTexture* texture) {
    SkASSERT(!"unimplemented");
    return nullptr;
}

sk_sp<GrDawnProgram> GrDawnGpu::getOrCreateRenderPipeline(
        GrRenderTarget* rt,
        const GrProgramInfo& programInfo) {
    GrProgramDesc desc = this->caps()->makeDesc(rt, programInfo);
    if (!desc.isValid()) {
        return nullptr;
    }

    if (sk_sp<GrDawnProgram>* program = fRenderPipelineCache.find(desc)) {
        return *program;
    }

    wgpu::TextureFormat colorFormat;
    SkAssertResult(programInfo.backendFormat().asDawnFormat(&colorFormat));

    wgpu::TextureFormat stencilFormat = wgpu::TextureFormat::Depth24PlusStencil8;
    bool hasDepthStencil = rt->getStencilAttachment() != nullptr;

    sk_sp<GrDawnProgram> program = GrDawnProgramBuilder::Build(
        this, rt, programInfo, colorFormat,
        hasDepthStencil, stencilFormat, &desc);
    fRenderPipelineCache.insert(desc, program);
    return program;
}

wgpu::Sampler GrDawnGpu::getOrCreateSampler(GrSamplerState samplerState) {
    auto i = fSamplers.find(samplerState);
    if (i != fSamplers.end()) {
        return i->second;
    }
    wgpu::SamplerDescriptor desc;
    desc.addressModeU = to_dawn_address_mode(samplerState.wrapModeX());
    desc.addressModeV = to_dawn_address_mode(samplerState.wrapModeY());
    desc.addressModeW = wgpu::AddressMode::ClampToEdge;
    desc.magFilter = desc.minFilter = to_dawn_filter_mode(samplerState.filter());
    desc.mipmapFilter = to_dawn_mipmap_mode(samplerState.mipmapMode());
    wgpu::Sampler sampler = device().CreateSampler(&desc);
    fSamplers.insert(std::pair<GrSamplerState, wgpu::Sampler>(samplerState, sampler));
    return sampler;
}

GrDawnRingBuffer::Slice GrDawnGpu::allocateUniformRingBufferSlice(int size) {
    return fUniformRingBuffer.allocate(size);
}

void GrDawnGpu::appendCommandBuffer(wgpu::CommandBuffer commandBuffer) {
    if (commandBuffer) {
        fCommandBuffers.push_back(commandBuffer);
    }
}

wgpu::CommandEncoder GrDawnGpu::getCopyEncoder() {
    if (!fCopyEncoder) {
        fCopyEncoder = fDevice.CreateCommandEncoder();
    }
    return fCopyEncoder;
}

void GrDawnGpu::flushCopyEncoder() {
    if (fCopyEncoder) {
        fCommandBuffers.push_back(fCopyEncoder.Finish());
        fCopyEncoder = nullptr;
    }
}

void GrDawnGpu::moveStagingBuffersToBusyAndMapAsync() {
    for (size_t i = 0; i < fSubmittedStagingBuffers.size(); ++i) {
        GrDawnBuffer* buffer = static_cast<GrDawnBuffer*>(fSubmittedStagingBuffers[i].get());
        buffer->mapWriteAsync();
        fBusyStagingBuffers.push_back(std::move(fSubmittedStagingBuffers[i]));
    }
    fSubmittedStagingBuffers.clear();
}

SkSL::String GrDawnGpu::SkSLToSPIRV(const char* shaderString, SkSL::Program::Kind kind, bool flipY,
                                    uint32_t rtHeightOffset, SkSL::Program::Inputs* inputs) {
    SkSL::Program::Settings settings;
    settings.fFlipY = flipY;
    settings.fRTHeightOffset = rtHeightOffset;
    settings.fRTHeightBinding = 0;
    settings.fRTHeightSet = 0;
    std::unique_ptr<SkSL::Program> program = this->shaderCompiler()->convertProgram(
        kind,
        shaderString,
        settings);
    if (!program) {
        SkDebugf("SkSL error:\n%s\n", this->shaderCompiler()->errorText().c_str());
        SkASSERT(false);
        return "";
    }
    if (inputs) {
        *inputs = program->fInputs;
    }
    SkSL::String code;
    if (!this->shaderCompiler()->toSPIRV(*program, &code)) {
        return "";
    }
    return code;
}

wgpu::ShaderModule GrDawnGpu::createShaderModule(const SkSL::String& spirvSource) {
    wgpu::ShaderModuleSPIRVDescriptor desc;
    desc.codeSize = spirvSource.size() / 4;
    desc.code = reinterpret_cast<const uint32_t*>(spirvSource.c_str());

    wgpu::ShaderModuleDescriptor smDesc;
    smDesc.nextInChain = &desc;

    return fDevice.CreateShaderModule(&smDesc);
}