src/gpu/graphite/DrawList.cpp

/*
 * Copyright 2021 Google LLC
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */
#include "src/gpu/graphite/DrawList.h"

#include "include/core/SkTypes.h"
#include "include/gpu/graphite/Recorder.h"
#include "src/core/SkSafetyChecks.h"
#include "src/core/SkTraceEvent.h"
#include "src/gpu/graphite/DrawPass.h"
#include "src/gpu/graphite/DrawWriter.h"
#include "src/gpu/graphite/KeyContext.h"
#include "src/gpu/graphite/RecorderPriv.h"
#include "src/gpu/graphite/Renderer.h"
#include "src/gpu/graphite/geom/Geometry.h"

namespace skgpu::graphite {

std::pair<DrawParams*, Layer*> DrawList::recordDraw(
        const Renderer* renderer,
        const Transform& localToDevice,
        const Geometry& geometry,
        const Clip& clip,
        DrawOrder ordering,
        UniquePaintParamsID paintID,
        SkEnumBitMask<DstUsage> dstUsage,
        BarrierType barrierBeforeDraws,
        PipelineDataGatherer* gatherer,
        StorageContext* storageContext,
        const StrokeStyle* stroke,
        Layer*) {

    SkASSERT(localToDevice.valid());
    SkASSERT(!geometry.isEmpty() && !clip.drawBounds().isEmptyNegativeOrNaN());
    SkASSERT(!(renderer->depthStencilFlags() & DepthStencilFlags::kStencil) ||
             ordering.stencilIndex() != DrawOrder::kUnassigned);

    // TODO: Add validation that the renderer's expected shape type and stroke params match provided
    const Draw& draw = fDraws.emplace_back(renderer,
                                           this->deduplicateTransform(localToDevice),
                                           geometry,
                                           clip,
                                           ordering,
                                           barrierBeforeDraws,
                                           stroke);

    fRenderStepCount += renderer->numRenderSteps();
    // Create a sort key for every render step in this draw
    for (int stepIndex = 0; stepIndex < draw.renderer()->numRenderSteps(); ++stepIndex) {
        const RenderStep* const step = draw.renderer()->steps()[stepIndex];
        const bool performsShading = step->performsShading() && paintID.isValid();

        if (storageContext && step->storageUniformStride() > 0) {
            storageContext->recordAlignment(step->storageUniformStride(),
                                            step->storageUniformAlignment());
        }

        gatherer->markOffsetAndAlign(performsShading, step->uniformAlignment());

        GraphicsPipelineCache::Index pipelineIndex = fPipelineCache.insert(
                { step->renderStepID(), performsShading ?
                                        paintID : UniquePaintParamsID::Invalid()});

        step->writeUniformsAndTextures(draw.drawParams(), gatherer);

        auto [combinedUniforms, combinedTextures] =
                gatherer->endCombinedData(performsShading);

        UniformDataCache::Index uniformIndex = combinedUniforms ?
                fUniformDataCache.insert(combinedUniforms) : UniformDataCache::kInvalidIndex;
        TextureDataCache::Index textureBindingIndex = combinedTextures ?
                fTextureDataCache.insert(combinedTextures) : TextureDataCache::kInvalidIndex;

        fSortKeys.push_back({&draw, stepIndex, pipelineIndex, uniformIndex, textureBindingIndex});
        gatherer->rewindForRenderStep();
    }

    fPassBounds.join(clip.drawBounds());
    fRequiresMSAA |= renderer->requiresMSAA();
    fDepthStencilFlags |= renderer->depthStencilFlags();
    if (dstUsage & DstUsage::kDstReadRequired) {
        // For paints that read from the dst, update the bounds. It may later be determined that the
        // DstReadStrategy does not require them, but they are inexpensive to track.
        fDstReadBounds.join(clip.drawBounds());
    }

#if defined(SK_DEBUG)
    if (geometry.isCoverageMaskShape()) {
        fCoverageMaskShapeDrawCount++;
    }
#endif

    return {nullptr, nullptr};
}

std::unique_ptr<DrawPass> DrawList::snapDrawPass(Recorder* recorder,
                                                 StorageContext* storageContext,
                                                 DrawContext* drawContext,
                                                 sk_sp<TextureProxy> target,
                                                 const SkImageInfo& targetInfo,
                                                 DstReadStrategy dstReadStrategy) {
    // NOTE: This assert is here to ensure SortKey is as tightly packed as possible. Any change to
    // its size should be done with care and good reason. The performance of sorting the keys is
    // heavily tied to the total size.
    //
    // At 24 bytes (current), sorting is about 30% slower than if SortKey could be packed into just
    // 16 bytes. There are several ways this could be done if necessary:
    //  - Restricting the max draw count to 16k (14-bits) and only using a single index to refer to
    //    the uniform data => 8 bytes of key, 8 bytes of pointer.
    //  - Restrict the max draw count to 32k (15-bits), use a single uniform index, and steal the
    //    4 low bits from the Draw* pointer since it's 16 byte aligned.
    //  - Compact the Draw* to an index into the original collection, although that has extra
    //    indirection and does not work as well with SkTBlockList.
    // In pseudo tests, manipulating the pointer or having to mask out indices was about 15% slower
    // than an 8 byte key and unmodified pointer.
    static_assert(sizeof(SortKey) == SkAlignTo(16 + sizeof(void*), alignof(SortKey)));

    // TODO: Explore sorting algorithms; in all likelihood this will be mostly sorted already, so
    // algorithms that approach O(n) in that condition may be favorable. Alternatively, could
    // explore radix sort that is always O(n). Brief testing suggested std::sort was faster than
    // std::stable_sort and SkTQSort on my [ml]'s Windows desktop. Also worth considering in-place
    // vs. algorithms that require an extra O(n) storage.
    // TODO: It's not strictly necessary, but would a stable sort be useful or just end up hiding
    // bugs in the DrawOrder determination code?
    std::sort(fSortKeys.begin(), fSortKeys.end());

    TRACE_EVENT0_ALWAYS("skia.gpu", "Snap DrawPass");
    TRACE_EVENT_INSTANT1("skia.gpu", "DrawList Stats", TRACE_EVENT_SCOPE_THREAD,
                         "draw count", fDraws.count());

    // The DrawList is converted directly into the DrawPass' data structures, but once the DrawPass
    // is returned from Make(), it is considered immutable.
    std::unique_ptr<DrawPass> drawPass(new DrawPass(target,
                                                    {fLoadOp, StoreOp::kStore},
                                                    fClearColor));

    DrawBufferManager* bufferMgr = recorder->priv().drawBufferManager();
    DrawWriter drawWriter(&drawPass->fCommandList, bufferMgr);
    GraphicsPipelineCache::Index lastPipeline = GraphicsPipelineCache::kInvalidIndex;
    const SkIRect targetBounds = SkIRect::MakeSize(targetInfo.dimensions());
    SkIRect lastScissor = targetBounds;

    SkASSERT(drawPass->fTarget->isFullyLazy() ||
             SkIRect::MakeSize(drawPass->fTarget->dimensions()).contains(lastScissor));
    drawPass->fCommandList.setScissor(lastScissor);

    const Caps* caps = recorder->priv().caps();
    const bool useStorageBuffers = caps->storageBufferSupport();
    UniformTracker uniformTracker(useStorageBuffers);

    if (storageContext) {
        storageContext->finalizePrecachedStorageData();
    }

    // TODO(b/372953722): Remove this forced binding command behavior once dst copies are always
    // bound separately from the rest of the textures.
    const bool rebindTexturesOnPipelineChange = dstReadStrategy == DstReadStrategy::kTextureCopy;
    // Keep track of the prior draw's PaintOrder. If the current draw requires barriers and there
    // is no pipeline or state change, then we must compare the current and prior draw's PaintOrders
    // to determine if the draws overlap. If they do, we must inject a flush between them such that
    // the barrier addition and draw commands are ordered correctly.
    CompressedPaintersOrder priorDrawPaintOrder {};

    // Accumulate rough pixel area touched by each pipeline as we iterate the SortKeys
    drawPass->fPipelineDrawAreas.push_back_n(fPipelineCache.count(), 0.f);

    TextureTracker textureBindingTracker(&fTextureDataCache);
    for (const DrawList::SortKey& key : fSortKeys) {
        const DrawList::Draw& draw = key.draw();
        const RenderStep& renderStep = key.renderStep();

        const bool pipelineChange = key.pipelineIndex() != lastPipeline;
        drawPass->fPipelineDrawAreas[key.pipelineIndex()] +=
                draw.drawParams().drawBounds().area();

        const bool uniformBindingChange = uniformTracker.writeUniforms(
                fUniformDataCache, bufferMgr, key.uniformIndex());

        // TODO(b/372953722): The Dawn and Vulkan CommandBuffer implementations currently append any
        // dst copy to the texture bind group/descriptor set automatically when processing a
        // BindTexturesAndSamplers call because they use a single group to contain all textures.
        // However, from the DrawPass POV, we can run into the scenario where two pipelines have the
        // same textures+samplers except one requires a dst-copy and the other does not. In this
        // case we wouldn't necessarily insert a new command when the pipeline changed and then
        // end up with layout validation errors.
        const bool textureBindingsChange = textureBindingTracker.setCurrentTextureBindings(
                key.textureBindingIndex()) ||
                (rebindTexturesOnPipelineChange && pipelineChange &&
                 key.textureBindingIndex() != TextureDataCache::kInvalidIndex);

        std::optional<SkIRect> newScissor =
                renderStep.getScissor(draw.drawParams(), lastScissor, targetBounds);

        const bool stateChange = uniformBindingChange  || textureBindingsChange ||
                                 newScissor.has_value();

        // Update DrawWriter *before* we actually change any state so that accumulated draws from
        // the previous state use the proper state.
        if (pipelineChange) {
            drawWriter.newPipelineState(renderStep.primitiveType(),
                                        renderStep.staticDataStride(),
                                        renderStep.appendDataStride(draw.drawParams()),
                                        renderStep.getRenderStateFlags(),
                                        draw.drawParams().barrierBeforeDraws());
        } else if (stateChange) {
            drawWriter.newDynamicState();
        } else if (draw.drawParams().barrierBeforeDraws() != BarrierType::kNone &&
                   priorDrawPaintOrder != draw.drawParams().order().paintOrder()) {
            // Even if there is no pipeline or state change, we must consider whether a
            // DrawPassCommand to add barriers must be inserted before any draw commands. If so,
            // then determine if the current and prior draws overlap (ie, their PaintOrders are
            // unequal). If so, perform a flush() to make sure the draw and add barrier commands are
            // appended to the command list in the proper order.
            drawWriter.flush();
        }

        // Make state changes before accumulating new draw data
        if (pipelineChange) {
            drawPass->fCommandList.bindGraphicsPipeline(key.pipelineIndex());
            lastPipeline = key.pipelineIndex();
        }
        if (stateChange) {
            if (uniformBindingChange) {
                uniformTracker.bindUniforms(UniformSlot::kCombinedUniforms, &drawPass->fCommandList);
            }
            if (textureBindingsChange) {
                textureBindingTracker.bindTextures(&drawPass->fCommandList);
            }
            if (newScissor.has_value()) {
                drawPass->fCommandList.setScissor(*newScissor);
                lastScissor = *newScissor;
            }
        }

        uint32_t uniformSsboIndex = useStorageBuffers ? uniformTracker.ssboIndex() : 0;
        renderStep.writeVertices(&drawWriter, storageContext, draw.drawParams(), uniformSsboIndex);

        if (bufferMgr->hasMappingFailed()) {
            SKIA_LOG_W("Failed to write necessary vertex/instance data for DrawPass, dropping!");
            this->reset(LoadOp::kLoad);
            return nullptr;
        }

        // Update priorDrawPaintOrder value before iterating to analyze the next draw.
        priorDrawPaintOrder = draw.drawParams().order().paintOrder();
    }
    // Finish recording draw calls for any collected data still pending at end of the loop
    drawWriter.flush();

    if (bufferMgr->hasMappingFailed()) {
        SKIA_LOG_W("Failed to write necessary vertex/instance data for DrawPass, dropping!");
        this->reset(LoadOp::kLoad);
        return nullptr;
    }

    if (storageContext) {
        const bool hasStorageData = !storageContext->isEmpty();
        auto storageResult = storageContext->finalize(recorder, drawContext);
        if (hasStorageData && !storageResult) SK_UNLIKELY {
            SKIA_LOG_W("Failed to write Storage Data for Draw pass, dropping!");
            this->reset(LoadOp::kLoad);
            return nullptr;
        }
        if (storageResult) {
            drawPass->setStorageResult(std::move(*storageResult));
        }
    }

    drawPass->fBounds = fPassBounds.roundOut().asSkIRect();
    drawPass->fPipelineDescs   = fPipelineCache.detach();
    drawPass->fSampledTextures = fTextureDataCache.detachTextures();

    TRACE_EVENT_INSTANT2("skia.gpu",
                         "DrawPass Stats",
                         TRACE_EVENT_SCOPE_THREAD,
                         "# commands", drawPass->fCommandList.count(),
                         "# textures", drawPass->fSampledTextures.size());

    this->reset(LoadOp::kLoad);

    return drawPass;
}

void DrawList::reset(LoadOp op, SkColor4f clearColor) {
    SK_SCOPED_DISABLE_PARTITION_ALLOC_SAFETY_CHECKS;
    fDraws.reset();
    fSortKeys.clear();
    DrawListBase::reset(op, clearColor);
}

} // namespace skgpu::graphite