compute-runtime/opencl/source/command_queue/enqueue_kernel.h

/*
 * Copyright (C) 2018-2023 Intel Corporation
 *
 * SPDX-License-Identifier: MIT
 *
 */

#pragma once
#include "shared/source/command_stream/command_stream_receiver.h"

#include "opencl/source/built_ins/builtins_dispatch_builder.h"
#include "opencl/source/command_queue/command_queue_hw.h"
#include "opencl/source/command_queue/gpgpu_walker.h"

#include <new>

namespace NEO {

template <typename GfxFamily>
cl_int CommandQueueHw<GfxFamily>::enqueueKernel(
    Kernel *pKernel,
    cl_uint workDim,
    const size_t *globalWorkOffsetIn,
    const size_t *globalWorkSizeIn,
    const size_t *localWorkSizeIn,
    cl_uint numEventsInWaitList,
    const cl_event *eventWaitList,
    cl_event *event) {

    if (workDim > device->getDeviceInfo().maxWorkItemDimensions) {
        return CL_INVALID_WORK_DIMENSION;
    }

    size_t region[3] = {1, 1, 1};
    size_t globalWorkOffset[3] = {0, 0, 0};
    size_t workGroupSize[3] = {1, 1, 1};
    size_t enqueuedLocalWorkSize[3] = {0, 0, 0};

    auto &kernel = *pKernel;
    const auto &kernelInfo = kernel.getKernelInfo();

    if (!kernel.isPatched()) {
        if (event) {
            *event = nullptr;
        }

        return CL_INVALID_KERNEL_ARGS;
    }

    if (kernel.isUsingSharedObjArgs()) {
        kernel.resetSharedObjectsPatchAddresses();
    }

    bool haveRequiredWorkGroupSize = false;

    if (kernelInfo.kernelDescriptor.kernelAttributes.requiredWorkgroupSize[0] != 0) {
        haveRequiredWorkGroupSize = true;
    }

    size_t remainder = 0;
    size_t totalWorkItems = 1u;
    const size_t *localWkgSizeToPass = localWorkSizeIn ? workGroupSize : nullptr;
    size_t reqdWorkgroupSize[3] = {};

    for (auto i = 0u; i < workDim; i++) {
        region[i] = globalWorkSizeIn ? globalWorkSizeIn[i] : 0;
        globalWorkOffset[i] = globalWorkOffsetIn
                                  ? globalWorkOffsetIn[i]
                                  : 0;

        if (localWorkSizeIn) {
            if (haveRequiredWorkGroupSize) {
                if (kernelInfo.kernelDescriptor.kernelAttributes.requiredWorkgroupSize[i] != localWorkSizeIn[i]) {
                    return CL_INVALID_WORK_GROUP_SIZE;
                }
            }
            if (localWorkSizeIn[i] == 0) {
                return CL_INVALID_WORK_GROUP_SIZE;
            }
            if (kernel.getAllowNonUniform()) {
                workGroupSize[i] = std::min(localWorkSizeIn[i], std::max(static_cast<size_t>(1), globalWorkSizeIn[i]));
            } else {
                workGroupSize[i] = localWorkSizeIn[i];
            }
            enqueuedLocalWorkSize[i] = localWorkSizeIn[i];
            totalWorkItems *= localWorkSizeIn[i];
        }

        remainder += region[i] % workGroupSize[i];
    }

    if (remainder != 0 && !kernel.getAllowNonUniform()) {
        return CL_INVALID_WORK_GROUP_SIZE;
    }

    if (haveRequiredWorkGroupSize) {
        reqdWorkgroupSize[0] = kernelInfo.kernelDescriptor.kernelAttributes.requiredWorkgroupSize[0];
        reqdWorkgroupSize[1] = kernelInfo.kernelDescriptor.kernelAttributes.requiredWorkgroupSize[1];
        reqdWorkgroupSize[2] = kernelInfo.kernelDescriptor.kernelAttributes.requiredWorkgroupSize[2];
        localWkgSizeToPass = reqdWorkgroupSize;
    }

    NullSurface s;
    Surface *surfaces[] = {&s};

    if (context->isProvidingPerformanceHints()) {
        if (kernel.hasPrintfOutput()) {
            context->providePerformanceHint(CL_CONTEXT_DIAGNOSTICS_LEVEL_NEUTRAL_INTEL, PRINTF_DETECTED_IN_KERNEL, kernelInfo.kernelDescriptor.kernelMetadata.kernelName.c_str());
        }
    }

    if (kernelInfo.builtinDispatchBuilder != nullptr) {
        cl_int err = kernelInfo.builtinDispatchBuilder->validateDispatch(&kernel, workDim, Vec3<size_t>(region), Vec3<size_t>(workGroupSize), Vec3<size_t>(globalWorkOffset));
        if (err != CL_SUCCESS)
            return err;
    }

    DBG_LOG(PrintDispatchParameters, "Kernel: ", kernelInfo.kernelDescriptor.kernelMetadata.kernelName,
            ", LWS: ", localWorkSizeIn ? localWorkSizeIn[0] : 0, ", ", localWorkSizeIn ? localWorkSizeIn[1] : 0, ", ", localWorkSizeIn ? localWorkSizeIn[2] : 0,
            ", GWS: ", globalWorkSizeIn[0], ", ", globalWorkSizeIn[1], ", ", globalWorkSizeIn[2],
            ", SIMD: ", kernelInfo.getMaxSimdSize());

    if (totalWorkItems > kernel.getMaxKernelWorkGroupSize()) {
        return CL_INVALID_WORK_GROUP_SIZE;
    }

    return enqueueHandler<CL_COMMAND_NDRANGE_KERNEL>(
        surfaces,
        false,
        &kernel,
        workDim,
        globalWorkOffset,
        region,
        localWkgSizeToPass,
        enqueuedLocalWorkSize,
        numEventsInWaitList,
        eventWaitList,
        event);
}

} // namespace NEO