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

Enable task count update from wait

Browse Source 
Signed-off-by: Jobczyk, Lukasz <lukasz.jobczyk@intel.com>
This commit is contained in:
Jobczyk, Lukasz
2022-03-25 13:00:53 +00:00
committed by Compute-Runtime-Automation
parent 022eb054e6
commit d77a6cbe4b
14 changed files with 116 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
opencl/source/command_queue/enqueue_common.h
View File

@ -771,7 +771,7 @@ CompletionStamp CommandQueueHw<GfxFamily>::enqueueNonBlocked(
blocking, //blocking
shouldFlushDC(commandType, printfHandler) || allocNeedsFlushDC, //dcFlush
multiDispatchInfo.usesSlm(), //useSLM
true, //guardCommandBufferWithPipeControl
!getGpgpuCommandStreamReceiver().isUpdateTagFromWaitEnabled(), //guardCommandBufferWithPipeControl
commandType == CL_COMMAND_NDRANGE_KERNEL, //GSBA32BitRequired
requiresCoherency, //requiresCoherency
(QueuePriority::LOW == priority), //lowPriority
@ -1008,7 +1008,7 @@ CompletionStamp CommandQueueHw<GfxFamily>::enqueueCommandWithoutKernel(
blocking, //blocking
false, //dcFlush
false, //useSLM
true, //guardCommandBufferWithPipeControl
!getGpgpuCommandStreamReceiver().isUpdateTagFromWaitEnabled(), //guardCommandBufferWithPipeControl
false, //GSBA32BitRequired
false, //requiresCoherency
false, //lowPriority

6
opencl/source/helpers/task_information.cpp
View File

@ -70,7 +70,7 @@ CompletionStamp &CommandMapUnmap::submit(uint32_t taskLevel, bool terminated) {
true, //blocking
true, //dcFlush
false, //useSLM
true, //guardCommandBufferWithPipeControl
!commandQueue.getGpgpuCommandStreamReceiver().isUpdateTagFromWaitEnabled(), //guardCommandBufferWithPipeControl
false, //GSBA32BitRequired
false, //requiresCoherency
commandQueue.getPriority() == QueuePriority::LOW, //lowPriority
@ -199,7 +199,7 @@ CompletionStamp &CommandComputeKernel::submit(uint32_t taskLevel, bool terminate
true, //blocking
flushDC, //dcFlush
slmUsed, //useSLM
true, //guardCommandBufferWithPipeControl
!commandQueue.getGpgpuCommandStreamReceiver().isUpdateTagFromWaitEnabled(), //guardCommandBufferWithPipeControl
commandType == CL_COMMAND_NDRANGE_KERNEL, //GSBA32BitRequired
requiresCoherency, //requiresCoherency
commandQueue.getPriority() == QueuePriority::LOW, //lowPriority
@ -357,7 +357,7 @@ CompletionStamp &CommandWithoutKernel::submit(uint32_t taskLevel, bool terminate
true, //blocking
false, //dcFlush
false, //useSLM
true, //guardCommandBufferWithPipeControl
!commandStreamReceiver.isUpdateTagFromWaitEnabled(), //guardCommandBufferWithPipeControl
false, //GSBA32BitRequired
false, //requiresCoherency
commandQueue.getPriority() == QueuePriority::LOW, //lowPriority

18
opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_2_tests.cpp
View File

@ -102,24 +102,8 @@ HWTEST_F(CommandStreamReceiverFlushTaskTests, GivenEnableUpdateTaskFromWaitWhenN
// Parse command list
parseCommands<FamilyType>(commandStreamTask, 0);
auto pipeControlExpected = MemorySynchronizationCommands<FamilyType>::getDcFlushEnable(true, *defaultHwInfo);
auto itorPC = find<PIPE_CONTROL *>(cmdList.begin(), cmdList.end());
if (pipeControlExpected) {
EXPECT_NE(cmdList.end(), itorPC);
if (UnitTestHelper<FamilyType>::isPipeControlWArequired(pDevice->getHardwareInfo())) {
itorPC++;
itorPC = find<PIPE_CONTROL *>(itorPC, cmdList.end());
EXPECT_NE(cmdList.end(), itorPC);
}
// Verify that the dcFlushEnabled bit is set in PC
auto pCmdWA = reinterpret_cast<PIPE_CONTROL *>(*itorPC);
EXPECT_EQ(MemorySynchronizationCommands<FamilyType>::getDcFlushEnable(true, *defaultHwInfo), pCmdWA->getDcFlushEnable());
} else {
EXPECT_EQ(cmdList.end(), itorPC);
}
EXPECT_EQ(cmdList.end(), itorPC);
buffer->release();
}

33
opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_3_tests.cpp
View File

@ -1032,7 +1032,7 @@ HWTEST_F(CommandStreamReceiverFlushTaskTests, givenCsrInBatchingModeWithOutOfOrd
EXPECT_EQ(MemorySynchronizationCommands<FamilyType>::getDcFlushEnable(true, *defaultHwInfo), pipeControl->getDcFlushEnable());
}
HWTEST_F(CommandStreamReceiverFlushTaskTests, givenUpdateTaskCountFromWaitSetWhenFlushTaskThenThereIsNoPipeControlForUpdateTaskCount) {
HWTEST_F(CommandStreamReceiverFlushTaskTests, givenUpdateTaskCountFromWaitSetAndGuardCommandBufferWithPipeControlWhenFlushTaskThenThereIsPipeControlForUpdateTaskCount) {
DebugManagerStateRestore restorer;
DebugManager.flags.UpdateTaskCountFromWait.set(3);
@ -1061,6 +1061,37 @@ HWTEST_F(CommandStreamReceiverFlushTaskTests, givenUpdateTaskCountFromWaitSetWhe
parseCommands<FamilyType>(commandStream);
auto itorPipeControl = find<typename FamilyType::PIPE_CONTROL *>(cmdList.begin(), cmdList.end());
EXPECT_NE(itorPipeControl, cmdList.end());
}
HWTEST_F(CommandStreamReceiverFlushTaskTests, givenUpdateTaskCountFromWaitSetWhenFlushTaskThenThereIsNoPipeControlForUpdateTaskCount) {
DebugManagerStateRestore restorer;
DebugManager.flags.UpdateTaskCountFromWait.set(3);
CommandQueueHw<FamilyType> commandQueue(nullptr, pClDevice, 0, false);
auto &commandStream = commandQueue.getCS(4096u);
auto mockCsr = new MockCsrHw2<FamilyType>(*pDevice->executionEnvironment, pDevice->getRootDeviceIndex(), pDevice->getDeviceBitfield());
pDevice->resetCommandStreamReceiver(mockCsr);
mockCsr->useNewResourceImplicitFlush = false;
mockCsr->useGpuIdleImplicitFlush = false;
mockCsr->overrideDispatchPolicy(DispatchMode::BatchedDispatch);
DispatchFlags dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
dispatchFlags.preemptionMode = PreemptionHelper::getDefaultPreemptionMode(pDevice->getHardwareInfo());
mockCsr->flushTask(commandStream,
0,
dsh,
ioh,
ssh,
taskLevel,
dispatchFlags,
*pDevice);
parseCommands<FamilyType>(commandStream);
auto itorPipeControl = find<typename FamilyType::PIPE_CONTROL *>(cmdList.begin(), cmdList.end());
EXPECT_EQ(itorPipeControl, cmdList.end());
}

6
opencl/test/unit_test/helpers/hw_helper_tests_pvc_and_later.cpp
View File

@ -86,6 +86,12 @@ HWTEST2_F(HwHelperTestPvcAndLater, GivenVariousValuesWhenCallingGetBarriersCount
EXPECT_EQ(32u, hwHelper.getBarriersCountFromHasBarriers(7u));
}
HWTEST2_F(HwHelperTestPvcAndLater, givenHwHelperWhenCheckIsUpdateTaskCountFromWaitSupportedThenReturnsTrue, IsAtLeastXeHpcCore) {
auto &hwHelper = HwHelper::get(hardwareInfo.platform.eRenderCoreFamily);
EXPECT_TRUE(hwHelper.isUpdateTaskCountFromWaitSupported());
}
HWTEST2_F(HwHelperTestPvcAndLater, givenCooperativeContextSupportedWhenGetEngineInstancesThenReturnCorrectAmountOfCooperativeCcs, IsAtLeastXeHpcCore) {
HardwareInfo hwInfo = *defaultHwInfo;
hwInfo.gtSystemInfo.CCSInfo.NumberOfCCSEnabled = 2;