Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add multithread enqueue blit OOQ test

Browse Source 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk
2022-03-02 13:42:34 +00:00
committed by Compute-Runtime-Automation
parent b41f088fe9
commit 0919cad885
2 changed files with 76 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
opencl/test/unit_test/mt_tests/command_queue/ooq_task_tests_mt.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright (C) 2018-2021 Intel Corporation
* Copyright (C) 2018-2022 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@ -8,6 +8,8 @@
#include "opencl/test/unit_test/command_queue/enqueue_fixture.h"
#include "opencl/test/unit_test/fixtures/hello_world_fixture.h"
#include <future>
using namespace NEO;
struct OOQFixtureFactory : public HelloWorldFixtureFactory {
@ -89,4 +91,76 @@ TEST_F(OOQTaskTestsMt, GivenBlockedOnUserEventWhenEnqueingMarkerThenSuccessIsRet
retVal = clReleaseEvent(userEvent);
EXPECT_EQ(CL_SUCCESS, retVal);
}
TEST_F(OOQTaskTestsMt, givenBlitterWhenEnqueueCopyAndKernelUsingMultipleThreadsThenSuccessReturned) {
auto hwInfo = *defaultHwInfo;
hwInfo.capabilityTable.blitterOperationsSupported = true;
REQUIRE_FULL_BLITTER_OR_SKIP(&hwInfo);
DebugManagerStateRestore restorer;
DebugManager.flags.EnableBlitterForEnqueueOperations.set(1);
DebugManager.flags.DoCpuCopyOnReadBuffer.set(0);
DebugManager.flags.DoCpuCopyOnWriteBuffer.set(0);
constexpr uint32_t numThreads = 32;
std::atomic_uint32_t barrier = numThreads;
std::array<std::future<void>, numThreads> threads;
auto device = MockClDevice::createWithNewExecutionEnvironment<MockDevice>(&hwInfo, rootDeviceIndex);
MockClDevice clDevice(device);
auto cmdQ = createCommandQueue(&clDevice, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE);
EXPECT_EQ(cmdQ->taskCount, 0u);
EXPECT_EQ(cmdQ->getGpgpuCommandStreamReceiver().peekTaskCount(), 0u);
EXPECT_EQ(cmdQ->peekBcsTaskCount(aub_stream::EngineType::ENGINE_BCS), 0u);
auto buffer = std::unique_ptr<Buffer>(BufferHelper<>::create());
for (auto &thread : threads) {
thread = std::async(std::launch::async, [&]() {
auto alignedReadPtr = alignedMalloc(BufferDefaults::sizeInBytes, MemoryConstants::cacheLineSize);
barrier.fetch_sub(1u);
while (barrier.load() != 0u) {
std::this_thread::yield();
}
auto retVal = EnqueueWriteBufferHelper<>::enqueueWriteBuffer(cmdQ,
buffer.get(),
CL_TRUE,
0,
BufferDefaults::sizeInBytes,
alignedReadPtr,
nullptr,
0,
nullptr,
nullptr);
EXPECT_EQ(CL_SUCCESS, retVal);
size_t workSize[] = {64};
retVal = EnqueueKernelHelper<>::enqueueKernel(cmdQ, KernelFixture::pKernel, 1, nullptr, workSize, workSize, 0, nullptr, nullptr);
EXPECT_EQ(CL_SUCCESS, retVal);
retVal = EnqueueReadBufferHelper<>::enqueueReadBuffer(cmdQ,
buffer.get(),
CL_TRUE,
0,
BufferDefaults::sizeInBytes,
alignedReadPtr,
nullptr,
0,
nullptr,
nullptr);
EXPECT_EQ(CL_SUCCESS, retVal);
alignedFree(alignedReadPtr);
});
}
for (auto &thread : threads) {
thread.get();
}
EXPECT_NE(cmdQ->taskCount, 0u);
EXPECT_NE(cmdQ->getGpgpuCommandStreamReceiver().peekTaskCount(), 0u);
EXPECT_EQ(cmdQ->peekBcsTaskCount(aub_stream::EngineType::ENGINE_BCS), 2 * numThreads);
clReleaseCommandQueue(cmdQ);
}