intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 22:12:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: skip queue state check when waiting for latest IOQ TSP

Browse Source 
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2023-08-17 13:15:00 +00:00
committed by Compute-Runtime-Automation
parent 73ffc56938
commit 59233d9597
3 changed files with 67 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

55
opencl/test/unit_test/api/cl_enqueue_wait_for_events_tests.inl
View File

@@ -15,6 +15,7 @@
#include "opencl/source/event/user_event.h"
#include "opencl/source/gtpin/gtpin_defs.h"
#include "opencl/test/unit_test/mocks/mock_event.h"
#include "opencl/test/unit_test/mocks/mock_kernel.h"
#include "cl_api_tests.h"
@@ -154,6 +155,60 @@ HWTEST_F(clEnqueueWaitForEventsTests, givenOoqWhenWaitingForEventThenCallWaitFor
EXPECT_TRUE(commandQueueHw.latestWaitForTimestampsStatus);
}
struct clEnqueueWaitForTimestampsTests : public clEnqueueWaitForEventsTests {
void SetUp() override {
DebugManager.flags.EnableTimestampWaitForQueues.set(4);
DebugManager.flags.EnableTimestampWaitForEvents.set(4);
DebugManager.flags.EnableTimestampPacket.set(1);
clEnqueueWaitForEventsTests::SetUp();
}
DebugManagerStateRestore restore;
};
HWTEST_F(clEnqueueWaitForTimestampsTests, givenIoqWhenWaitingForLatestEventThenDontCheckQueueCompletion) {
MockCommandQueueHw<FamilyType> commandQueueHw(pContext, pDevice, nullptr);
MockKernelWithInternals kernel(*pDevice);
cl_event event0, event1;
const size_t gws[] = {1, 1, 1};
commandQueueHw.enqueueKernel(kernel.mockKernel, 1, nullptr, gws, nullptr, 0, nullptr, &event0);
commandQueueHw.enqueueKernel(kernel.mockKernel, 1, nullptr, gws, nullptr, 0, nullptr, &event1);
auto eventObj0 = castToObjectOrAbort<Event>(event0);
auto eventObj1 = castToObjectOrAbort<Event>(event1);
auto node0 = eventObj0->getTimestampPacketNodes()->peekNodes()[0];
auto node1 = eventObj1->getTimestampPacketNodes()->peekNodes()[0];
auto contextEnd0 = ptrOffset(node0->getCpuBase(), node0->getContextEndOffset());
auto contextEnd1 = ptrOffset(node1->getCpuBase(), node1->getContextEndOffset());
*reinterpret_cast<typename FamilyType::TimestampPacketType *>(contextEnd0) = 0;
*reinterpret_cast<typename FamilyType::TimestampPacketType *>(contextEnd1) = 0;
EXPECT_EQ(0u, commandQueueHw.isCompletedCalled);
EXPECT_EQ(CL_SUCCESS, clEnqueueWaitForEvents(&commandQueueHw, 1, &event0));
EXPECT_EQ(1u, commandQueueHw.isCompletedCalled);
EXPECT_EQ(CL_SUCCESS, clEnqueueWaitForEvents(&commandQueueHw, 1, &event1));
EXPECT_EQ(1u, commandQueueHw.isCompletedCalled);
commandQueueHw.setOoqEnabled();
EXPECT_EQ(CL_SUCCESS, clEnqueueWaitForEvents(&commandQueueHw, 1, &event0));
EXPECT_EQ(2u, commandQueueHw.isCompletedCalled);
EXPECT_EQ(CL_SUCCESS, clEnqueueWaitForEvents(&commandQueueHw, 1, &event1));
EXPECT_EQ(3u, commandQueueHw.isCompletedCalled);
clReleaseEvent(event0);
clReleaseEvent(event1);
}
HWTEST_F(clEnqueueWaitForEventsTests, givenAlreadyCompletedEventWhenWaitForCompletionThenCheckGpuStateOnce) {
auto &ultCsr = pDevice->getUltCommandStreamReceiver<FamilyType>();
auto csrTagAddress = ultCsr.getTagAddress();