dawn-cmake/src/dawn/native/ComputePassEncoder.cpp

// Copyright 2018 The Dawn Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "dawn/native/ComputePassEncoder.h"

#include "dawn/native/BindGroup.h"
#include "dawn/native/BindGroupLayout.h"
#include "dawn/native/Buffer.h"
#include "dawn/native/CommandEncoder.h"
#include "dawn/native/CommandValidation.h"
#include "dawn/native/Commands.h"
#include "dawn/native/ComputePipeline.h"
#include "dawn/native/Device.h"
#include "dawn/native/InternalPipelineStore.h"
#include "dawn/native/ObjectType_autogen.h"
#include "dawn/native/PassResourceUsageTracker.h"
#include "dawn/native/QuerySet.h"
#include "dawn/native/utils/WGPUHelpers.h"

namespace dawn::native {

namespace {

ResultOrError<ComputePipelineBase*> GetOrCreateIndirectDispatchValidationPipeline(
    DeviceBase* device) {
    InternalPipelineStore* store = device->GetInternalPipelineStore();

    if (store->dispatchIndirectValidationPipeline != nullptr) {
        return store->dispatchIndirectValidationPipeline.Get();
    }

    // TODO(https://crbug.com/dawn/1108): Propagate validation feedback from this
    // shader in various failure modes.
    // Type 'bool' cannot be used in storage class 'uniform' as it is non-host-shareable.
    Ref<ShaderModuleBase> shaderModule;
    DAWN_TRY_ASSIGN(shaderModule, utils::CreateShaderModule(device, R"(
                struct UniformParams {
                    maxComputeWorkgroupsPerDimension: u32,
                    clientOffsetInU32: u32,
                    enableValidation: u32,
                    duplicateNumWorkgroups: u32,
                }

                struct IndirectParams {
                    data: array<u32>
                }

                struct ValidatedParams {
                    data: array<u32>
                }

                @group(0) @binding(0) var<uniform> uniformParams: UniformParams;
                @group(0) @binding(1) var<storage, read_write> clientParams: IndirectParams;
                @group(0) @binding(2) var<storage, read_write> validatedParams: ValidatedParams;

                @compute @workgroup_size(1, 1, 1)
                fn main() {
                    for (var i = 0u; i < 3u; i = i + 1u) {
                        var numWorkgroups = clientParams.data[uniformParams.clientOffsetInU32 + i];
                        if (uniformParams.enableValidation > 0u &&
                            numWorkgroups > uniformParams.maxComputeWorkgroupsPerDimension) {
                            numWorkgroups = 0u;
                        }
                        validatedParams.data[i] = numWorkgroups;

                        if (uniformParams.duplicateNumWorkgroups > 0u) {
                             validatedParams.data[i + 3u] = numWorkgroups;
                        }
                    }
                }
            )"));

    Ref<BindGroupLayoutBase> bindGroupLayout;
    DAWN_TRY_ASSIGN(bindGroupLayout,
                    utils::MakeBindGroupLayout(
                        device,
                        {
                            {0, wgpu::ShaderStage::Compute, wgpu::BufferBindingType::Uniform},
                            {1, wgpu::ShaderStage::Compute, kInternalStorageBufferBinding},
                            {2, wgpu::ShaderStage::Compute, wgpu::BufferBindingType::Storage},
                        },
                        /* allowInternalBinding */ true));

    Ref<PipelineLayoutBase> pipelineLayout;
    DAWN_TRY_ASSIGN(pipelineLayout, utils::MakeBasicPipelineLayout(device, bindGroupLayout));

    ComputePipelineDescriptor computePipelineDescriptor = {};
    computePipelineDescriptor.layout = pipelineLayout.Get();
    computePipelineDescriptor.compute.module = shaderModule.Get();
    computePipelineDescriptor.compute.entryPoint = "main";

    DAWN_TRY_ASSIGN(store->dispatchIndirectValidationPipeline,
                    device->CreateComputePipeline(&computePipelineDescriptor));

    return store->dispatchIndirectValidationPipeline.Get();
}

}  // namespace

ComputePassEncoder::ComputePassEncoder(DeviceBase* device,
                                       const ComputePassDescriptor* descriptor,
                                       CommandEncoder* commandEncoder,
                                       EncodingContext* encodingContext)
    : ProgrammableEncoder(device, descriptor->label, encodingContext),
      mCommandEncoder(commandEncoder) {
    TrackInDevice();
}

// static
Ref<ComputePassEncoder> ComputePassEncoder::Create(DeviceBase* device,
                                                   const ComputePassDescriptor* descriptor,
                                                   CommandEncoder* commandEncoder,
                                                   EncodingContext* encodingContext) {
    return AcquireRef(new ComputePassEncoder(device, descriptor, commandEncoder, encodingContext));
}

ComputePassEncoder::ComputePassEncoder(DeviceBase* device,
                                       CommandEncoder* commandEncoder,
                                       EncodingContext* encodingContext,
                                       ErrorTag errorTag)
    : ProgrammableEncoder(device, encodingContext, errorTag), mCommandEncoder(commandEncoder) {}

// static
Ref<ComputePassEncoder> ComputePassEncoder::MakeError(DeviceBase* device,
                                                      CommandEncoder* commandEncoder,
                                                      EncodingContext* encodingContext) {
    return AcquireRef(
        new ComputePassEncoder(device, commandEncoder, encodingContext, ObjectBase::kError));
}

void ComputePassEncoder::DestroyImpl() {
    // Ensure that the pass has exited. This is done for passes only since validation requires
    // they exit before destruction while bundles do not.
    mEncodingContext->EnsurePassExited(this);
}

ObjectType ComputePassEncoder::GetType() const {
    return ObjectType::ComputePassEncoder;
}

void ComputePassEncoder::APIEnd() {
    if (mEncodingContext->TryEncode(
            this,
            [&](CommandAllocator* allocator) -> MaybeError {
                if (IsValidationEnabled()) {
                    DAWN_TRY(ValidateProgrammableEncoderEnd());
                }

                allocator->Allocate<EndComputePassCmd>(Command::EndComputePass);

                return {};
            },
            "encoding %s.End().", this)) {
        mEncodingContext->ExitComputePass(this, mUsageTracker.AcquireResourceUsage());
    }
}

void ComputePassEncoder::APIEndPass() {
    GetDevice()->EmitDeprecationWarning("endPass() has been deprecated. Use end() instead.");
    APIEnd();
}

void ComputePassEncoder::APIDispatch(uint32_t workgroupCountX,
                                     uint32_t workgroupCountY,
                                     uint32_t workgroupCountZ) {
    GetDevice()->EmitDeprecationWarning(
        "dispatch() has been deprecated. Use dispatchWorkgroups() instead.");
    APIDispatchWorkgroups(workgroupCountX, workgroupCountY, workgroupCountZ);
}

void ComputePassEncoder::APIDispatchWorkgroups(uint32_t workgroupCountX,
                                               uint32_t workgroupCountY,
                                               uint32_t workgroupCountZ) {
    mEncodingContext->TryEncode(
        this,
        [&](CommandAllocator* allocator) -> MaybeError {
            if (IsValidationEnabled()) {
                DAWN_TRY(mCommandBufferState.ValidateCanDispatch());

                uint32_t workgroupsPerDimension =
                    GetDevice()->GetLimits().v1.maxComputeWorkgroupsPerDimension;

                DAWN_INVALID_IF(workgroupCountX > workgroupsPerDimension,
                                "Dispatch workgroup count X (%u) exceeds max compute "
                                "workgroups per dimension (%u).",
                                workgroupCountX, workgroupsPerDimension);

                DAWN_INVALID_IF(workgroupCountY > workgroupsPerDimension,
                                "Dispatch workgroup count Y (%u) exceeds max compute "
                                "workgroups per dimension (%u).",
                                workgroupCountY, workgroupsPerDimension);

                DAWN_INVALID_IF(workgroupCountZ > workgroupsPerDimension,
                                "Dispatch workgroup count Z (%u) exceeds max compute "
                                "workgroups per dimension (%u).",
                                workgroupCountZ, workgroupsPerDimension);
            }

            // Record the synchronization scope for Dispatch, which is just the current
            // bindgroups.
            AddDispatchSyncScope();

            DispatchCmd* dispatch = allocator->Allocate<DispatchCmd>(Command::Dispatch);
            dispatch->x = workgroupCountX;
            dispatch->y = workgroupCountY;
            dispatch->z = workgroupCountZ;

            return {};
        },
        "encoding %s.DispatchWorkgroups(%u, %u, %u).", this, workgroupCountX, workgroupCountY,
        workgroupCountZ);
}

ResultOrError<std::pair<Ref<BufferBase>, uint64_t>>
ComputePassEncoder::TransformIndirectDispatchBuffer(Ref<BufferBase> indirectBuffer,
                                                    uint64_t indirectOffset) {
    DeviceBase* device = GetDevice();

    const bool shouldDuplicateNumWorkgroups =
        device->ShouldDuplicateNumWorkgroupsForDispatchIndirect(
            mCommandBufferState.GetComputePipeline());
    if (!IsValidationEnabled() && !shouldDuplicateNumWorkgroups) {
        return std::make_pair(indirectBuffer, indirectOffset);
    }

    // Save the previous command buffer state so it can be restored after the
    // validation inserts additional commands.
    CommandBufferStateTracker previousState = mCommandBufferState;

    auto* const store = device->GetInternalPipelineStore();

    Ref<ComputePipelineBase> validationPipeline;
    DAWN_TRY_ASSIGN(validationPipeline, GetOrCreateIndirectDispatchValidationPipeline(device));

    Ref<BindGroupLayoutBase> layout;
    DAWN_TRY_ASSIGN(layout, validationPipeline->GetBindGroupLayout(0));

    uint32_t storageBufferOffsetAlignment = device->GetLimits().v1.minStorageBufferOffsetAlignment;

    // Let the offset be the indirectOffset, aligned down to |storageBufferOffsetAlignment|.
    const uint32_t clientOffsetFromAlignedBoundary = indirectOffset % storageBufferOffsetAlignment;
    const uint64_t clientOffsetAlignedDown = indirectOffset - clientOffsetFromAlignedBoundary;
    const uint64_t clientIndirectBindingOffset = clientOffsetAlignedDown;

    // Let the size of the binding be the additional offset, plus the size.
    const uint64_t clientIndirectBindingSize =
        kDispatchIndirectSize + clientOffsetFromAlignedBoundary;

    // Neither 'enableValidation' nor 'duplicateNumWorkgroups' can be declared as 'bool' as
    // currently in WGSL type 'bool' cannot be used in storage class 'uniform' as 'it is
    // non-host-shareable'.
    struct UniformParams {
        uint32_t maxComputeWorkgroupsPerDimension;
        uint32_t clientOffsetInU32;
        uint32_t enableValidation;
        uint32_t duplicateNumWorkgroups;
    };

    // Create a uniform buffer to hold parameters for the shader.
    Ref<BufferBase> uniformBuffer;
    {
        UniformParams params;
        params.maxComputeWorkgroupsPerDimension =
            device->GetLimits().v1.maxComputeWorkgroupsPerDimension;
        params.clientOffsetInU32 = clientOffsetFromAlignedBoundary / sizeof(uint32_t);
        params.enableValidation = static_cast<uint32_t>(IsValidationEnabled());
        params.duplicateNumWorkgroups = static_cast<uint32_t>(shouldDuplicateNumWorkgroups);

        DAWN_TRY_ASSIGN(uniformBuffer,
                        utils::CreateBufferFromData(device, wgpu::BufferUsage::Uniform, {params}));
    }

    // Reserve space in the scratch buffer to hold the validated indirect params.
    ScratchBuffer& scratchBuffer = store->scratchIndirectStorage;
    const uint64_t scratchBufferSize =
        shouldDuplicateNumWorkgroups ? 2 * kDispatchIndirectSize : kDispatchIndirectSize;
    DAWN_TRY(scratchBuffer.EnsureCapacity(scratchBufferSize));
    Ref<BufferBase> validatedIndirectBuffer = scratchBuffer.GetBuffer();

    Ref<BindGroupBase> validationBindGroup;
    ASSERT(indirectBuffer->GetUsage() & kInternalStorageBuffer);
    DAWN_TRY_ASSIGN(validationBindGroup,
                    utils::MakeBindGroup(device, layout,
                                         {
                                             {0, uniformBuffer},
                                             {1, indirectBuffer, clientIndirectBindingOffset,
                                              clientIndirectBindingSize},
                                             {2, validatedIndirectBuffer, 0, scratchBufferSize},
                                         },
                                         UsageValidationMode::Internal));

    // Issue commands to validate the indirect buffer.
    APISetPipeline(validationPipeline.Get());
    APISetBindGroup(0, validationBindGroup.Get());
    APIDispatchWorkgroups(1);

    // Restore the state.
    RestoreCommandBufferState(std::move(previousState));

    // Return the new indirect buffer and indirect buffer offset.
    return std::make_pair(std::move(validatedIndirectBuffer), uint64_t(0));
}

void ComputePassEncoder::APIDispatchIndirect(BufferBase* indirectBuffer, uint64_t indirectOffset) {
    GetDevice()->EmitDeprecationWarning(
        "dispatchIndirect() has been deprecated. Use dispatchWorkgroupsIndirect() instead.");
    APIDispatchWorkgroupsIndirect(indirectBuffer, indirectOffset);
}

void ComputePassEncoder::APIDispatchWorkgroupsIndirect(BufferBase* indirectBuffer,
                                                       uint64_t indirectOffset) {
    mEncodingContext->TryEncode(
        this,
        [&](CommandAllocator* allocator) -> MaybeError {
            if (IsValidationEnabled()) {
                DAWN_TRY(GetDevice()->ValidateObject(indirectBuffer));
                DAWN_TRY(ValidateCanUseAs(indirectBuffer, wgpu::BufferUsage::Indirect));
                DAWN_TRY(mCommandBufferState.ValidateCanDispatch());

                DAWN_INVALID_IF(indirectOffset % 4 != 0,
                                "Indirect offset (%u) is not a multiple of 4.", indirectOffset);

                DAWN_INVALID_IF(
                    indirectOffset >= indirectBuffer->GetSize() ||
                        indirectOffset + kDispatchIndirectSize > indirectBuffer->GetSize(),
                    "Indirect offset (%u) and dispatch size (%u) exceeds the indirect buffer "
                    "size (%u).",
                    indirectOffset, kDispatchIndirectSize, indirectBuffer->GetSize());
            }

            SyncScopeUsageTracker scope;
            scope.BufferUsedAs(indirectBuffer, wgpu::BufferUsage::Indirect);
            mUsageTracker.AddReferencedBuffer(indirectBuffer);
            // TODO(crbug.com/dawn/1166): If validation is enabled, adding |indirectBuffer|
            // is needed for correct usage validation even though it will only be bound for
            // storage. This will unecessarily transition the |indirectBuffer| in
            // the backend.

            Ref<BufferBase> indirectBufferRef = indirectBuffer;

            // Get applied indirect buffer with necessary changes on the original indirect
            // buffer. For example,
            // - Validate each indirect dispatch with a single dispatch to copy the indirect
            //   buffer params into a scratch buffer if they're valid, and otherwise zero them
            //   out.
            // - Duplicate all the indirect dispatch parameters to support @num_workgroups on
            //   D3D12.
            // - Directly return the original indirect dispatch buffer if we don't need any
            //   transformations on it.
            // We could consider moving the validation earlier in the pass after the last
            // last point the indirect buffer was used with writable usage, as well as batch
            // validation for multiple dispatches into one, but inserting commands at
            // arbitrary points in the past is not possible right now.
            DAWN_TRY_ASSIGN(std::tie(indirectBufferRef, indirectOffset),
                            TransformIndirectDispatchBuffer(indirectBufferRef, indirectOffset));

            // If we have created a new scratch dispatch indirect buffer in
            // TransformIndirectDispatchBuffer(), we need to track it in mUsageTracker.
            if (indirectBufferRef.Get() != indirectBuffer) {
                // |indirectBufferRef| was replaced with a scratch buffer. Add it to the
                // synchronization scope.
                scope.BufferUsedAs(indirectBufferRef.Get(), wgpu::BufferUsage::Indirect);
                mUsageTracker.AddReferencedBuffer(indirectBufferRef.Get());
            }

            AddDispatchSyncScope(std::move(scope));

            DispatchIndirectCmd* dispatch =
                allocator->Allocate<DispatchIndirectCmd>(Command::DispatchIndirect);
            dispatch->indirectBuffer = std::move(indirectBufferRef);
            dispatch->indirectOffset = indirectOffset;
            return {};
        },
        "encoding %s.DispatchWorkgroupsIndirect(%s, %u).", this, indirectBuffer, indirectOffset);
}

void ComputePassEncoder::APISetPipeline(ComputePipelineBase* pipeline) {
    mEncodingContext->TryEncode(
        this,
        [&](CommandAllocator* allocator) -> MaybeError {
            if (IsValidationEnabled()) {
                DAWN_TRY(GetDevice()->ValidateObject(pipeline));
            }

            mCommandBufferState.SetComputePipeline(pipeline);

            SetComputePipelineCmd* cmd =
                allocator->Allocate<SetComputePipelineCmd>(Command::SetComputePipeline);
            cmd->pipeline = pipeline;

            return {};
        },
        "encoding %s.SetPipeline(%s).", this, pipeline);
}

void ComputePassEncoder::APISetBindGroup(uint32_t groupIndexIn,
                                         BindGroupBase* group,
                                         uint32_t dynamicOffsetCount,
                                         const uint32_t* dynamicOffsets) {
    mEncodingContext->TryEncode(
        this,
        [&](CommandAllocator* allocator) -> MaybeError {
            BindGroupIndex groupIndex(groupIndexIn);

            if (IsValidationEnabled()) {
                DAWN_TRY(
                    ValidateSetBindGroup(groupIndex, group, dynamicOffsetCount, dynamicOffsets));
            }

            mUsageTracker.AddResourcesReferencedByBindGroup(group);
            RecordSetBindGroup(allocator, groupIndex, group, dynamicOffsetCount, dynamicOffsets);
            mCommandBufferState.SetBindGroup(groupIndex, group, dynamicOffsetCount, dynamicOffsets);

            return {};
        },
        "encoding %s.SetBindGroup(%u, %s, %u, ...).", this, groupIndexIn, group,
        dynamicOffsetCount);
}

void ComputePassEncoder::APIWriteTimestamp(QuerySetBase* querySet, uint32_t queryIndex) {
    mEncodingContext->TryEncode(
        this,
        [&](CommandAllocator* allocator) -> MaybeError {
            if (IsValidationEnabled()) {
                DAWN_TRY(ValidateTimestampQuery(GetDevice(), querySet, queryIndex));
            }

            mCommandEncoder->TrackQueryAvailability(querySet, queryIndex);

            WriteTimestampCmd* cmd =
                allocator->Allocate<WriteTimestampCmd>(Command::WriteTimestamp);
            cmd->querySet = querySet;
            cmd->queryIndex = queryIndex;

            return {};
        },
        "encoding %s.WriteTimestamp(%s, %u).", this, querySet, queryIndex);
}

void ComputePassEncoder::AddDispatchSyncScope(SyncScopeUsageTracker scope) {
    PipelineLayoutBase* layout = mCommandBufferState.GetPipelineLayout();
    for (BindGroupIndex i : IterateBitSet(layout->GetBindGroupLayoutsMask())) {
        scope.AddBindGroup(mCommandBufferState.GetBindGroup(i));
    }
    mUsageTracker.AddDispatch(scope.AcquireSyncScopeUsage());
}

void ComputePassEncoder::RestoreCommandBufferState(CommandBufferStateTracker state) {
    // Encode commands for the backend to restore the pipeline and bind groups.
    if (state.HasPipeline()) {
        APISetPipeline(state.GetComputePipeline());
    }
    for (BindGroupIndex i(0); i < kMaxBindGroupsTyped; ++i) {
        BindGroupBase* bg = state.GetBindGroup(i);
        if (bg != nullptr) {
            const std::vector<uint32_t>& offsets = state.GetDynamicOffsets(i);
            if (offsets.empty()) {
                APISetBindGroup(static_cast<uint32_t>(i), bg);
            } else {
                APISetBindGroup(static_cast<uint32_t>(i), bg, offsets.size(), offsets.data());
            }
        }
    }

    // Restore the frontend state tracking information.
    mCommandBufferState = std::move(state);
}

CommandBufferStateTracker* ComputePassEncoder::GetCommandBufferStateTrackerForTesting() {
    return &mCommandBufferState;
}

}  // namespace dawn::native