compute-runtime/unit_tests/command_queue/ioq_task_tests_mt.cpp

/*
 * Copyright (C) 2017-2019 Intel Corporation
 *
 * SPDX-License-Identifier: MIT
 *
 */

#include "unit_tests/command_queue/enqueue_fixture.h"
#include "unit_tests/fixtures/hello_world_fixture.h"
#include "unit_tests/fixtures/image_fixture.h"
#include "unit_tests/mocks/mock_buffer.h"

using namespace NEO;

typedef HelloWorldTest<HelloWorldFixtureFactory> IOQTaskTestsMt;

TEST_F(IOQTaskTestsMt, enqueueReadBuffer_blockingAndBlockedOnUserEvent) {
    auto buffer = std::unique_ptr<Buffer>(BufferHelper<>::create());

    auto alignedReadPtr = alignedMalloc(BufferDefaults::sizeInBytes, MemoryConstants::cacheLineSize);
    ASSERT_NE(nullptr, alignedReadPtr);

    auto userEvent = clCreateUserEvent(pContext, &retVal);
    EXPECT_EQ(CL_SUCCESS, retVal);

    auto previousTaskLevel = pCmdQ->taskLevel;
    auto previousTaskCount = pCmdQ->taskCount;

    std::thread t([=]() {
        Event *ev = castToObject<Event>(userEvent);
        while (ev->peekHasChildEvents() == false) {
            // active wait for VirtualEvent (which is added after queue is blocked)
        }
        auto ret = clSetUserEventStatus(userEvent, CL_COMPLETE);
        ASSERT_EQ(CL_SUCCESS, ret);
    });

    buffer->forceDisallowCPUCopy = true; // no task level incrasing when cpu copy
    retVal = EnqueueReadBufferHelper<>::enqueueReadBuffer(pCmdQ,
                                                          buffer.get(),
                                                          CL_TRUE,
                                                          0,
                                                          BufferDefaults::sizeInBytes,
                                                          alignedReadPtr,
                                                          nullptr,
                                                          1,
                                                          &userEvent,
                                                          nullptr);
    EXPECT_EQ(CL_SUCCESS, retVal);

    EXPECT_LT(previousTaskCount, pCmdQ->taskCount);
    EXPECT_LT(previousTaskLevel, pCmdQ->taskLevel);

    t.join();

    retVal = clReleaseEvent(userEvent);
    EXPECT_EQ(CL_SUCCESS, retVal);

    alignedFree(alignedReadPtr);
}

TEST_F(IOQTaskTestsMt, enqueueMarker_blockedOnUserEvent) {

    auto userEvent = clCreateUserEvent(pContext, &retVal);
    EXPECT_EQ(CL_SUCCESS, retVal);

    std::thread t([=]() {
        Event *ev = castToObject<Event>(userEvent);
        while (ev->peekHasChildEvents() == false) {
            // active wait for VirtualEvent (which is added after queue is blocked)
        }
        auto ret = clSetUserEventStatus(userEvent, CL_COMPLETE);
        ASSERT_EQ(CL_SUCCESS, ret);
    });

    retVal = pCmdQ->enqueueMarkerWithWaitList(
        1,
        &userEvent,
        nullptr);
    EXPECT_EQ(CL_SUCCESS, retVal);

    t.join();

    retVal = clReleaseEvent(userEvent);
    EXPECT_EQ(CL_SUCCESS, retVal);
}

TEST_F(IOQTaskTestsMt, enqueueMapBuffer) {
    AlignedBuffer alignedBuffer;

    auto userEvent = clCreateUserEvent(pContext, &retVal);
    EXPECT_EQ(CL_SUCCESS, retVal);

    cl_event outputEvent = nullptr;
    void *mappedPtr = pCmdQ->enqueueMapBuffer(&alignedBuffer, false, CL_MAP_READ, 0, alignedBuffer.getSize(), 1, &userEvent, &outputEvent, retVal);

    EXPECT_EQ(CL_SUCCESS, retVal);

    const int32_t numThreads = 20;

    std::thread threads[numThreads];
    std::thread threadUnblocking;

    cl_event ouputEventsFromThreads[numThreads];
    void *mappedPtrs[numThreads];

    for (int32_t i = 0; i < numThreads; i++) {
        threads[i] = std::thread([&](int32_t index) {
            cl_int errCode = CL_SUCCESS;
            cl_int success = CL_SUCCESS;
            mappedPtrs[index] = pCmdQ->enqueueMapBuffer(&alignedBuffer, false, CL_MAP_READ, 0, alignedBuffer.getSize(), 0, nullptr, &ouputEventsFromThreads[index], errCode);
            EXPECT_EQ(success, errCode);
        },
                                 i);

        if (i == numThreads / 2) {
            threadUnblocking = std::thread([=]() {
                auto ret = clSetUserEventStatus(userEvent, CL_COMPLETE);
                EXPECT_EQ(CL_SUCCESS, ret);
            });
        }
    }

    cl_int errCode = clWaitForEvents(1, &outputEvent);
    EXPECT_EQ(CL_SUCCESS, errCode);

    cl_int eventStatus = 0;
    errCode = clGetEventInfo(outputEvent, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int), &eventStatus, nullptr);
    EXPECT_EQ(CL_SUCCESS, errCode);
    EXPECT_EQ(CL_COMPLETE, eventStatus);

    for (int32_t i = 0; i < numThreads; i++) {
        threads[i].join();
        cl_int errCode = clWaitForEvents(1, &ouputEventsFromThreads[i]);
        EXPECT_EQ(CL_SUCCESS, errCode);
        errCode = clGetEventInfo(outputEvent, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int), &eventStatus, nullptr);
        EXPECT_EQ(CL_SUCCESS, errCode);
        EXPECT_EQ(CL_COMPLETE, eventStatus);
    }

    threadUnblocking.join();
    retVal = clReleaseEvent(userEvent);

    for (int32_t i = 0; i < numThreads; i++) {
        pCmdQ->enqueueUnmapMemObject(&alignedBuffer, mappedPtrs[i], 0, nullptr, nullptr);
        retVal = clReleaseEvent(ouputEventsFromThreads[i]);
        EXPECT_EQ(CL_SUCCESS, retVal);
    }

    pCmdQ->enqueueUnmapMemObject(&alignedBuffer, mappedPtr, 0, nullptr, nullptr);

    retVal = clReleaseEvent(outputEvent);
    EXPECT_EQ(CL_SUCCESS, retVal);
}

TEST_F(IOQTaskTestsMt, enqueueMapImage) {
    auto image = std::unique_ptr<Image>(ImageHelper<Image1dDefaults>::create(context));

    auto userEvent = clCreateUserEvent(pContext, &retVal);
    EXPECT_EQ(CL_SUCCESS, retVal);

    cl_event outputEvent = nullptr;
    const size_t origin[] = {0, 0, 0};
    const size_t region[] = {1, 1, 1};

    void *mappedPtr = pCmdQ->enqueueMapImage(image.get(), false, CL_MAP_READ, origin, region, nullptr, nullptr, 1, &userEvent, &outputEvent, retVal);

    EXPECT_EQ(CL_SUCCESS, retVal);

    const int32_t numThreads = 20;

    std::thread threads[numThreads];
    std::thread threadUnblocking;

    cl_event ouputEventsFromThreads[numThreads];
    void *mappedPtrs[numThreads];

    for (int32_t i = 0; i < numThreads; i++) {
        threads[i] = std::thread([&](int32_t index) {
            cl_int errCode = CL_SUCCESS;
            cl_int success = CL_SUCCESS;
            mappedPtrs[index] = pCmdQ->enqueueMapImage(image.get(), false, CL_MAP_READ, origin, region, nullptr, nullptr, 0, nullptr, &ouputEventsFromThreads[index], errCode);
            EXPECT_EQ(success, errCode);
        },
                                 i);

        if (i == numThreads / 2) {
            threadUnblocking = std::thread([=]() {
                auto ret = clSetUserEventStatus(userEvent, CL_COMPLETE);
                EXPECT_EQ(CL_SUCCESS, ret);
            });
        }
    }

    cl_int errCode = clWaitForEvents(1, &outputEvent);
    EXPECT_EQ(CL_SUCCESS, errCode);

    cl_int eventStatus = 0;
    errCode = clGetEventInfo(outputEvent, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int), &eventStatus, nullptr);
    EXPECT_EQ(CL_SUCCESS, errCode);
    EXPECT_EQ(CL_COMPLETE, eventStatus);

    for (int32_t i = 0; i < numThreads; i++) {
        threads[i].join();
        cl_int errCode = clWaitForEvents(1, &ouputEventsFromThreads[i]);
        EXPECT_EQ(CL_SUCCESS, errCode);
        errCode = clGetEventInfo(outputEvent, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int), &eventStatus, nullptr);
        EXPECT_EQ(CL_SUCCESS, errCode);
        EXPECT_EQ(CL_COMPLETE, eventStatus);
    }

    threadUnblocking.join();
    retVal = clReleaseEvent(userEvent);

    for (int32_t i = 0; i < numThreads; i++) {
        pCmdQ->enqueueUnmapMemObject(image.get(), mappedPtrs[i], 0, nullptr, nullptr);
        retVal = clReleaseEvent(ouputEventsFromThreads[i]);
        EXPECT_EQ(CL_SUCCESS, retVal);
    }

    pCmdQ->enqueueUnmapMemObject(image.get(), mappedPtr, 0, nullptr, nullptr);

    retVal = clReleaseEvent(outputEvent);
    EXPECT_EQ(CL_SUCCESS, retVal);
}