intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-31 20:13:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: add missing dependency wait for in-order appendWaitOnEvents

Browse Source 
Related-To: NEO-7966

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2023-08-04 09:47:12 +00:00
committed by Compute-Runtime-Automation
parent 30734fa844
commit a7e0594fd9
2 changed files with 24 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -2131,10 +2131,6 @@ inline ze_result_t CommandListCoreFamily<gfxCoreFamily>::addEventsToCmdList(uint
if (hasInOrderDependencies) {
CommandListCoreFamily<gfxCoreFamily>::appendWaitOnInOrderDependency(this->inOrderDependencyCounterAllocation, this->inOrderDependencyCounter, this->inOrderAllocationOffset, relaxedOrderingAllowed);
if (NEO::EncodeUserInterruptHelper::isOperationAllowed(NEO::EncodeUserInterruptHelper::afterSemaphoreMask)) {
NEO::EnodeUserInterrupt<GfxFamily>::encode(*commandContainer.getCommandStream());
}
}
if (numWaitEvents > 0) {
@@ -2210,6 +2206,10 @@ void CommandListCoreFamily<gfxCoreFamily>::appendWaitOnInOrderDependency(NEO::Gr
gpuAddress += sizeof(uint64_t);
}
if (NEO::EncodeUserInterruptHelper::isOperationAllowed(NEO::EncodeUserInterruptHelper::afterSemaphoreMask)) {
NEO::EnodeUserInterrupt<GfxFamily>::encode(*commandContainer.getCommandStream());
}
}
template <GFXCORE_FAMILY gfxCoreFamily>
@@ -2232,6 +2232,10 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWaitOnEvents(uint32_t nu
callId = neoDevice->getRootDeviceEnvironment().tagsManager->currentCallCount;
}
if (signalInOrderCompletion) {
CommandListCoreFamily<gfxCoreFamily>::appendWaitOnInOrderDependency(this->inOrderDependencyCounterAllocation, this->inOrderDependencyCounter, this->inOrderAllocationOffset, relaxedOrderingAllowed);
}
bool dcFlushRequired = false;
if (this->dcFlushSupport) {
@@ -2395,12 +2399,9 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWriteGlobalTimestamp(
uint64_t *dstptr, ze_event_handle_t hSignalEvent,
uint32_t numWaitEvents, ze_event_handle_t *phWaitEvents) {
if (numWaitEvents > 0) {
if (phWaitEvents) {
CommandListCoreFamily<gfxCoreFamily>::appendWaitOnEvents(numWaitEvents, phWaitEvents, false, true, false);
} else {
return ZE_RESULT_ERROR_INVALID_ARGUMENT;
}
ze_result_t ret = addEventsToCmdList(numWaitEvents, phWaitEvents, false, true);
if (ret != ZE_RESULT_SUCCESS) {
return ret;
}
Event *signalEvent = nullptr;

15
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp
View File

@@ -1831,6 +1831,7 @@ HWTEST2_F(InOrderCmdListTests, givenCopyOnlyInOrderModeWhenProgrammingCopyRegion
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenProgrammingAppendWaitOnEventsThenSignalSyncAllocation, IsAtLeastXeHpCore) {
using MI_STORE_DATA_IMM = typename FamilyType::MI_STORE_DATA_IMM;
using MI_SEMAPHORE_WAIT = typename FamilyType::MI_SEMAPHORE_WAIT;
auto immCmdList = createImmCmdList<gfxCoreFamily>();
@@ -1841,6 +1842,7 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenProgrammingAppendWaitOnEvents
auto eventHandle = events[0]->toHandle();
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, eventHandle, 0, nullptr, launchParams, false);
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
auto offset = cmdStream->getUsed();
@@ -1851,14 +1853,23 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenProgrammingAppendWaitOnEvents
ptrOffset(cmdStream->getCpuBase(), offset),
(cmdStream->getUsed() - offset)));
auto sdiItor = find<MI_STORE_DATA_IMM *>(cmdList.begin(), cmdList.end());
auto semaphoreItor = find<MI_SEMAPHORE_WAIT *>(cmdList.begin(), cmdList.end());
ASSERT_NE(cmdList.end(), semaphoreItor);
auto semaphoreCmd = genCmdCast<MI_SEMAPHORE_WAIT *>(*semaphoreItor);
ASSERT_NE(nullptr, semaphoreCmd);
EXPECT_EQ(2u, semaphoreCmd->getSemaphoreDataDword());
EXPECT_EQ(immCmdList->inOrderDependencyCounterAllocation->getGpuAddress(), semaphoreCmd->getSemaphoreGraphicsAddress());
auto sdiItor = find<MI_STORE_DATA_IMM *>(semaphoreItor, cmdList.end());
ASSERT_NE(cmdList.end(), sdiItor);
auto sdiCmd = genCmdCast<MI_STORE_DATA_IMM *>(*sdiItor);
EXPECT_EQ(immCmdList->inOrderDependencyCounterAllocation->getGpuAddress(), sdiCmd->getAddress());
EXPECT_EQ(0u, sdiCmd->getStoreQword());
EXPECT_EQ(2u, sdiCmd->getDataDword0());
EXPECT_EQ(3u, sdiCmd->getDataDword0());
}
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenProgrammingCounterWithOverflowThenHandleItCorrectly, IsAtLeastXeHpCore) {