Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-11-10 05:49:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update isCooperativeDispatchSupported

Browse Source 
Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
This commit is contained in:
Filip Hazubski
2021-08-23 17:42:53 +00:00
committed by Compute-Runtime-Automation
parent 6b5a5d481e
commit 3d6d4acda2
11 changed files with 77 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
opencl/source/api/api.cpp
View File

@@ -5949,7 +5949,7 @@ cl_int CL_API_CALL clEnqueueNDCountKernelINTEL(cl_command_queue commandQueue,
auto &hwHelper = HwHelper::get(hardwareInfo.platform.eRenderCoreFamily);
auto engineGroupType = hwHelper.getEngineGroupType(pCommandQueue->getGpgpuEngine().getEngineType(),
pCommandQueue->getGpgpuEngine().getEngineUsage(), hardwareInfo);
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType)) {
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType, hardwareInfo)) {
retVal = CL_INVALID_COMMAND_QUEUE;
return retVal;
}

3
opencl/source/kernel/kernel.cpp
View File

@@ -1041,9 +1041,6 @@ uint32_t Kernel::getMaxWorkGroupCount(const cl_uint workDim, const size_t *local
auto engineGroupType = hwHelper.getEngineGroupType(commandQueue->getGpgpuEngine().getEngineType(),
commandQueue->getGpgpuEngine().getEngineUsage(), hardwareInfo);
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType)) {
return 0;
}
const auto &kernelDescriptor = kernelInfo.kernelDescriptor;
auto dssCount = hardwareInfo.gtSystemInfo.DualSubSliceCount;

14
opencl/test/unit_test/command_queue/enqueue_kernel_1_tests.cpp
View File

@@ -311,7 +311,7 @@ TEST_F(clEnqueueNDCountKernelTests, GivenQueueIncapableWhenEnqueuingNDCountKerne
auto &hwHelper = HwHelper::get(::defaultHwInfo->platform.eRenderCoreFamily);
auto engineGroupType = hwHelper.getEngineGroupType(pCommandQueue->getGpgpuEngine().getEngineType(),
pCommandQueue->getGpgpuEngine().getEngineUsage(), *::defaultHwInfo);
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType)) {
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType, *::defaultHwInfo)) {
GTEST_SKIP();
}
@@ -342,10 +342,10 @@ TEST_F(EnqueueKernelTest, givenKernelWhenAllArgsAreSetThenClEnqueueNDCountKernel
cl_int retVal = CL_SUCCESS;
CommandQueue *pCmdQ2 = createCommandQueue(pClDevice);
HwHelper &hwHelper = HwHelper::get(pClDevice->getDevice().getHardwareInfo().platform.eRenderCoreFamily);
HwHelper &hwHelper = HwHelper::get(hardwareInfo.platform.eRenderCoreFamily);
auto engineGroupType = hwHelper.getEngineGroupType(pCmdQ2->getGpgpuEngine().getEngineType(),
pCmdQ2->getGpgpuEngine().getEngineUsage(), hardwareInfo);
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType)) {
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType, hardwareInfo)) {
pCmdQ2->getGpgpuEngine().osContext = pCmdQ2->getDevice().getEngine(aub_stream::ENGINE_CCS, EngineUsage::LowPriority).osContext;
}
@@ -390,10 +390,10 @@ TEST_F(EnqueueKernelTest, givenKernelWhenNotAllArgsAreSetButSetKernelArgIsCalled
cl_int retVal = CL_SUCCESS;
CommandQueue *pCmdQ2 = createCommandQueue(pClDevice);
HwHelper &hwHelper = HwHelper::get(pClDevice->getDevice().getHardwareInfo().platform.eRenderCoreFamily);
HwHelper &hwHelper = HwHelper::get(hardwareInfo.platform.eRenderCoreFamily);
auto engineGroupType = hwHelper.getEngineGroupType(pCmdQ2->getGpgpuEngine().getEngineType(),
pCmdQ2->getGpgpuEngine().getEngineUsage(), hardwareInfo);
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType)) {
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType, hardwareInfo)) {
pCmdQ2->getGpgpuEngine().osContext = pCmdQ2->getDevice().getEngine(aub_stream::ENGINE_CCS, EngineUsage::LowPriority).osContext;
}
@@ -438,10 +438,10 @@ TEST_F(EnqueueKernelTest, givenKernelWhenSetKernelArgIsCalledForEachArgButAtLeas
cl_int retVal = CL_SUCCESS;
CommandQueue *pCmdQ2 = createCommandQueue(pClDevice);
HwHelper &hwHelper = HwHelper::get(pClDevice->getDevice().getHardwareInfo().platform.eRenderCoreFamily);
HwHelper &hwHelper = HwHelper::get(hardwareInfo.platform.eRenderCoreFamily);
auto engineGroupType = hwHelper.getEngineGroupType(pCmdQ2->getGpgpuEngine().getEngineType(),
pCmdQ2->getGpgpuEngine().getEngineUsage(), hardwareInfo);
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType)) {
if (!hwHelper.isCooperativeDispatchSupported(engineGroupType, hardwareInfo)) {
pCmdQ2->getGpgpuEngine().osContext = pCmdQ2->getDevice().getEngine(aub_stream::ENGINE_CCS, EngineUsage::LowPriority).osContext;
}

14
opencl/test/unit_test/command_queue/sync_buffer_handler_tests.cpp
View File

@@ -16,6 +16,8 @@
#include "opencl/test/unit_test/mocks/mock_platform.h"
#include "test.h"
#include "engine_node.h"
using namespace NEO;
class MockSyncBufferHandler : public SyncBufferHandler {
@@ -69,6 +71,9 @@ class SyncBufferHandlerTest : public SyncBufferEnqueueHandlerTest {
kernel->executionType = KernelExecutionType::Concurrent;
commandQueue = reinterpret_cast<MockCommandQueue *>(new MockCommandQueueHw<FamilyType>(context, pClDevice, 0));
hwHelper = &HwHelper::get(pClDevice->getHardwareInfo().platform.eRenderCoreFamily);
if (hwHelper->isCooperativeEngineSupported(pClDevice->getHardwareInfo())) {
commandQueue->gpgpuEngine = &pClDevice->getEngine(aub_stream::EngineType::ENGINE_CCS, EngineUsage::Cooperative);
}
}
template <typename FamilyType>
@@ -93,7 +98,7 @@ class SyncBufferHandlerTest : public SyncBufferEnqueueHandlerTest {
bool isCooperativeDispatchSupported() {
auto engineGroupType = hwHelper->getEngineGroupType(commandQueue->getGpgpuEngine().getEngineType(),
commandQueue->getGpgpuEngine().getEngineUsage(), hardwareInfo);
return hwHelper->isCooperativeDispatchSupported(engineGroupType);
return hwHelper->isCooperativeDispatchSupported(engineGroupType, pDevice->getHardwareInfo());
}
const cl_uint workDim = 1;
@@ -116,9 +121,10 @@ HWTEST_TEMPLATED_F(SyncBufferHandlerTest, GivenAllocateSyncBufferPatchAndConcurr
EXPECT_EQ(workItemsCount, syncBufferHandler->usedBufferSize);
commandQueue->flush();
EXPECT_EQ(syncBufferHandler->graphicsAllocation->getTaskCount(
pDevice->getUltCommandStreamReceiver<FamilyType>().getOsContext().getContextId()),
pDevice->getUltCommandStreamReceiver<FamilyType>().latestSentTaskCount);
auto pCsr = commandQueue->getGpgpuEngine().commandStreamReceiver;
EXPECT_EQ(syncBufferHandler->graphicsAllocation->getTaskCount(pCsr->getOsContext().getContextId()),
static_cast<UltCommandStreamReceiver<FamilyType> *>(pCsr)->latestSentTaskCount);
}
HWTEST_TEMPLATED_F(SyncBufferHandlerTest, GivenConcurrentKernelWithoutAllocateSyncBufferPatchWhenEnqueuingConcurrentKernelThenSyncBufferIsNotCreated) {

1
opencl/test/unit_test/mocks/mock_command_queue.h
View File

@@ -27,6 +27,7 @@ class MockCommandQueue : public CommandQueue {
using CommandQueue::gpgpuEngine;
using CommandQueue::isCopyOnly;
using CommandQueue::obtainNewTimestampPacketNodes;
using CommandQueue::overrideEngine;
using CommandQueue::queueCapabilities;
using CommandQueue::queueFamilyIndex;
using CommandQueue::queueFamilySelected;