intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-19 16:24:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix event submit time for enqueue without kernel

Browse Source 
Change-Id: Ibb9ee7312a2c9618d91ebca03b0e0eb98cc551ee
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
This commit is contained in:
Bartosz Dunajski
2020-08-28 17:39:45 +02:00
committed by sys_ocldev
parent 4e53e4e257
commit ffc218d8a2
4 changed files with 44 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
opencl/source/command_queue/enqueue_common.h
View File

@@ -932,6 +932,13 @@ CompletionStamp CommandQueueHw<GfxFamily>::enqueueCommandWithoutKernel(
flushGpgpuCsr = false;
}
if (eventBuilder.getEvent() && isProfilingEnabled()) {
TimeStampData submitTimeStamp;
getDevice().getOSTime()->getCpuGpuTime(&submitTimeStamp);
eventBuilder.getEvent()->setSubmitTimeStamp(&submitTimeStamp);
}
if (flushGpgpuCsr) {
if (timestampPacketContainer) {
timestampPacketContainer->makeResident(getGpgpuCommandStreamReceiver());
@@ -943,13 +950,6 @@ CompletionStamp CommandQueueHw<GfxFamily>::enqueueCommandWithoutKernel(
surface->makeResident(getGpgpuCommandStreamReceiver());
}
TimeStampData submitTimeStamp;
if (eventBuilder.getEvent() && isProfilingEnabled() && getGpgpuCommandStreamReceiver().peekTimestampPacketWriteEnabled()) {
this->getDevice().getOSTime()->getCpuGpuTime(&submitTimeStamp);
eventBuilder.getEvent()->setSubmitTimeStamp(&submitTimeStamp);
eventBuilder.getEvent()->getTimestampPacketNodes()->makeResident(getGpgpuCommandStreamReceiver());
}
DispatchFlags dispatchFlags(
{}, //csrDependencies
&timestampPacketDependencies.barrierNodes, //barrierTimestampPacketNodes

30
opencl/test/unit_test/command_queue/blit_enqueue_tests.cpp
View File

@@ -1243,6 +1243,36 @@ HWTEST_TEMPLATED_F(BlitEnqueueWithDisabledGpgpuSubmissionTests, givenCacheFlushR
EXPECT_EQ(2u, gpgpuCsr->peekTaskCount());
}
HWTEST_TEMPLATED_F(BlitEnqueueWithDisabledGpgpuSubmissionTests, givenProfilingEnabledWhenSubmittingWithoutFlushToGpgpuThenSetSubmitTime) {
auto mockCommandQueue = static_cast<MockCommandQueueHw<FamilyType> *>(commandQueue.get());
EXPECT_EQ(EnqueueProperties::Operation::None, mockCommandQueue->latestSentEnqueueType);
DebugManager.flags.ForceGpgpuSubmissionForBcsEnqueue.set(-1);
mockCommandQueue->overrideIsCacheFlushForBcsRequired.enabled = true;
mockCommandQueue->overrideIsCacheFlushForBcsRequired.returnValue = true;
mockCommandQueue->setProfilingEnabled();
auto buffer = createBuffer(1, false);
buffer->forceDisallowCPUCopy = true;
int hostPtr = 0;
cl_event clEvent;
commandQueue->enqueueWriteBuffer(buffer.get(), false, 0, 1, &hostPtr, nullptr, 0, nullptr, &clEvent);
EXPECT_EQ(EnqueueProperties::Operation::Blit, mockCommandQueue->latestSentEnqueueType);
EXPECT_EQ(0u, gpgpuCsr->peekTaskCount());
auto event = castToObject<Event>(clEvent);
uint64_t submitTime = 0;
event->getEventProfilingInfo(CL_PROFILING_COMMAND_SUBMIT, sizeof(submitTime), &submitTime, nullptr);
EXPECT_NE(0u, submitTime);
clReleaseEvent(clEvent);
}
HWTEST_TEMPLATED_F(BlitEnqueueWithDisabledGpgpuSubmissionTests, givenCacheFlushNotRequiredWhenDoingBcsCopyThenDontSubmitToGpgpu) {
auto mockCommandQueue = static_cast<MockCommandQueueHw<FamilyType> *>(commandQueue.get());
EXPECT_EQ(EnqueueProperties::Operation::None, mockCommandQueue->latestSentEnqueueType);

6
opencl/test/unit_test/command_queue/enqueue_command_without_kernel_tests.cpp
View File

@@ -100,7 +100,7 @@ HWTEST_F(EnqueueHandlerTimestampEnabledTest, givenProflingAndTimeStampPacketsEna
using EnqueueHandlerTimestampDisabledTest = EnqueueHandlerTimestampTest<false>;
HWTEST_F(EnqueueHandlerTimestampDisabledTest, givenProflingEnabledTimeStampPacketsDisabledWhenEnqueueCommandWithoutKernelThenSubmitTimeStampIsNotSet) {
HWTEST_F(EnqueueHandlerTimestampDisabledTest, givenProflingEnabledTimeStampPacketsDisabledWhenEnqueueCommandWithoutKernelThenSubmitTimeStampIsSet) {
cl_queue_properties properties[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
std::unique_ptr<MockCommandQueueHw<FamilyType>> mockCmdQ(new MockCommandQueueHw<FamilyType>(context, pClDevice, properties));
@@ -122,8 +122,8 @@ HWTEST_F(EnqueueHandlerTimestampDisabledTest, givenProflingEnabledTimeStampPacke
mockCmdQ->enqueueCommandWithoutKernel(surfaces, 1, mockCmdQ->getCS(0), 0, blocking, enqueueProperties, timestampPacketDependencies,
eventsRequest, eventBuilder, 0);
EXPECT_EQ(ev->submitTimeStamp.CPUTimeinNS, 0u);
EXPECT_EQ(ev->submitTimeStamp.GPUTimeStamp, 0u);
EXPECT_NE(ev->submitTimeStamp.CPUTimeinNS, 0u);
EXPECT_NE(ev->submitTimeStamp.GPUTimeStamp, 0u);
delete ev;
}

4
opencl/test/unit_test/mocks/mock_command_queue.h
View File

@@ -209,6 +209,10 @@ class MockCommandQueueHw : public CommandQueueHw<GfxFamily> {
commandQueueProperties |= CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE;
}
void setProfilingEnabled() {
commandQueueProperties |= CL_QUEUE_PROFILING_ENABLE;
}
LinearStream &getCS(size_t minRequiredSize) override {
requestedCmdStreamSize = minRequiredSize;
return CommandQueue::getCS(minRequiredSize);