src/gpu/graphite/GpuWorkSubmission.cpp

/*
 * Copyright 2022 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/GpuWorkSubmission.h"

#include "src/gpu/RefCntedCallback.h"
#include "src/gpu/graphite/Buffer.h"
#include "src/gpu/graphite/CommandBuffer.h"
#include "src/gpu/graphite/QueueManager.h"

namespace skgpu::graphite {

GpuWorkSubmission::GpuWorkSubmission(std::unique_ptr<CommandBuffer> cmdBuffer,
                                     QueueManager* queueManager)
        : GpuWorkSubmission(std::move(cmdBuffer), /*protectedCmdBuffer=*/nullptr, queueManager) {}

GpuWorkSubmission::GpuWorkSubmission(std::unique_ptr<CommandBuffer> cmdBuffer,
                                     std::unique_ptr<CommandBuffer> protectedCmdBuffer,
                                     QueueManager* queueManager)
        : fCommandBuffer(std::move(cmdBuffer))
        , fProtectedCommandBuffer(std::move(protectedCmdBuffer))
        , fQueueManager(queueManager) {
    // Protected command buffers cannot write to unprotected host-accessible memory,
    // so they will never have buffers to async map.
    SkASSERT(!fProtectedCommandBuffer ||
             fProtectedCommandBuffer->buffersToAsyncMapOnSubmit().empty());
    if (fCommandBuffer) {
        SkSpan<const sk_sp<Buffer>> buffers = fCommandBuffer->buffersToAsyncMapOnSubmit();
        if (!buffers.empty()) {
            fOutstandingAsyncMapCounter = sk_make_sp<SkRefCnt>();
            for (auto& buffer : buffers) {
                SkASSERT(!buffer->isUnmappable());
                fOutstandingAsyncMapCounter->ref();
                buffer->asyncMap([](GpuFinishedContext c, CallbackResult) {
                                     static_cast<SkRefCnt*>(c)->unref();
                                 }, fOutstandingAsyncMapCounter.get());
            }
        }
    }
}

GpuWorkSubmission::~GpuWorkSubmission() {
    auto cleanup = [this](std::unique_ptr<CommandBuffer>& commandBuffer) {
        if (commandBuffer) {
            commandBuffer->resetCommandBuffer();
            fQueueManager->returnCommandBuffer(std::move(commandBuffer));
        }
    };
    this->callFinishedProcs(/*success=*/true);
    cleanup(fCommandBuffer);
    cleanup(fProtectedCommandBuffer);
}

bool GpuWorkSubmission::isFinished(const SharedContext* sharedContext) {
    return this->onIsFinished(sharedContext) &&
           (!fOutstandingAsyncMapCounter || fOutstandingAsyncMapCounter->unique());
}

void GpuWorkSubmission::waitUntilFinished(const SharedContext* sharedContext) {
    this->onWaitUntilFinished(sharedContext);
    if (fOutstandingAsyncMapCounter) {
        while (!fOutstandingAsyncMapCounter->unique()) {
            fQueueManager->tick();
        }
    }
}

void GpuWorkSubmission::addFinishedProc(sk_sp<RefCntedCallback> finishedProc) {
    if (finishedProc) {
        fFinishedProcs.push_back(std::move(finishedProc));
    }
}

void GpuWorkSubmission::addFinishedProcs(
        skia_private::TArray<sk_sp<RefCntedCallback>> finishedProcs) {
    for (int i = 0; i < finishedProcs.size(); ++i) {
        if (finishedProcs[i]) {
            fFinishedProcs.push_back(std::move(finishedProcs[i]));
        }
    }
}

std::optional<GpuStats> GpuWorkSubmission::gpuStats() const {
    std::optional<GpuStats> unprotectedStats =
            fCommandBuffer ? fCommandBuffer->gpuStats() : std::nullopt;
    std::optional<GpuStats> protectedStats =
            fProtectedCommandBuffer ? fProtectedCommandBuffer->gpuStats() : std::nullopt;
    if (unprotectedStats && protectedStats) {
        GpuStats stats;
        stats.elapsedTime = unprotectedStats->elapsedTime + protectedStats->elapsedTime;
        stats.numOcclusionPassSamples
            = unprotectedStats->numOcclusionPassSamples + protectedStats->numOcclusionPassSamples;
        return stats;
    } else if (unprotectedStats) {
        return unprotectedStats;
    } else {
        return protectedStats;
    }
}

void GpuWorkSubmission::callFinishedProcs(bool success) {
    if (!success) {
        for (int i = 0; i < fFinishedProcs.size(); ++i) {
            fFinishedProcs[i]->setFailureResult();
        }
    } else {
        if (auto stats = this->gpuStats()) {
            for (int i = 0; i < fFinishedProcs.size(); ++i) {
                if (fFinishedProcs[i]->receivesGpuStats()) {
                    fFinishedProcs[i]->setStats(*stats);
                }
            }
        }
    }
    fFinishedProcs.clear();
}

} // namespace skgpu::graphite