intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: RelaxedOrdering handling in BcsSplit path

Browse Source 
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2024-03-05 14:01:46 +00:00
committed by Compute-Runtime-Automation
parent 8b4bca89bc
commit 7bb23818e6
2 changed files with 13 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
level_zero/core/source/device/bcs_split.h
View File

@@ -74,6 +74,7 @@ struct BcsSplit {
NEO::TransferDirection direction,
std::function<ze_result_t(T, K, size_t, ze_event_handle_t)> appendCall) {
ze_result_t result = ZE_RESULT_SUCCESS;
const bool hasStallingCmds = !hasRelaxedOrderingDependencies;
auto markerEventIndexRet = this->events.obtainForSplit(Context::fromHandle(cmdList->getCmdListContext()), MemoryConstants::pageSize64k / sizeof(typename CommandListCoreFamilyImmediate<gfxCoreFamily>::GfxFamily::TimestampPacketType));
if (!markerEventIndexRet.has_value()) {
@@ -120,7 +121,7 @@ struct BcsSplit {
result = appendCall(localDstPtr, localSrcPtr, localSize, eventHandle);
if (cmdList->flushTaskSubmissionEnabled()) {
cmdList->executeCommandListImmediateWithFlushTaskImpl(performMigration, false, hasRelaxedOrderingDependencies, false, cmdQsForSplit[i]);
cmdList->executeCommandListImmediateWithFlushTaskImpl(performMigration, hasStallingCmds, hasRelaxedOrderingDependencies, false, cmdQsForSplit[i]);
} else {
cmdList->executeCommandListImmediateImpl(performMigration, cmdQsForSplit[i]);
}

11
level_zero/core/test/unit_tests/xe_hpc_core/test_cmdqueue_xe_hpc_core.cpp
View File

@@ -984,6 +984,17 @@ HWTEST2_F(CommandQueueCommandsXeHpc, givenFlushTaskSubmissionEnabledAndSplitBcsC
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[3])->getCsr()->peekTaskCount(), 1u);
EXPECT_TRUE(ultCsr->latestFlushedBatchBuffer.hasRelaxedOrderingDependencies);
commandList0->appendMemoryCopy(dstPtr, srcPtr, size, nullptr, 0, nullptr, false, false);
EXPECT_TRUE(ultCsr->latestFlushedBatchBuffer.hasRelaxedOrderingDependencies);
EXPECT_FALSE(ultCsr->latestFlushedBatchBuffer.hasStallingCmds);
directSubmission->relaxedOrderingEnabled = false;
commandList0->appendMemoryCopy(dstPtr, srcPtr, size, nullptr, 0, nullptr, false, false);
EXPECT_FALSE(ultCsr->latestFlushedBatchBuffer.hasRelaxedOrderingDependencies);
EXPECT_TRUE(ultCsr->latestFlushedBatchBuffer.hasStallingCmds);
context->freeMem(srcPtr);
context->freeMem(dstPtr);
}