intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-21 01:04:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix BCS split events handling in RelaxedOrdering mode

Browse Source 
Related-To: NEO-7458

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2023-02-06 10:11:45 +00:00
committed by Compute-Runtime-Automation
parent 2fb9eba861
commit 756a9a9aaa
2 changed files with 70 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
level_zero/core/source/device/bcs_split.h
View File

@@ -93,7 +93,7 @@ struct BcsSplit {
engineCount--;
}
cmdList->addEventsToCmdList(static_cast<uint32_t>(this->cmdQs.size()), eventHandles.data(), true);
cmdList->addEventsToCmdList(static_cast<uint32_t>(this->cmdQs.size()), eventHandles.data(), hasRelaxedOrderingDependencies);
cmdList->appendEventForProfilingAllWalkers(this->events.marker[markerEventIndex], false, true);
if (hSignalEvent) {

69
level_zero/core/test/unit_tests/xe_hpc_core/test_cmdqueue_xe_hpc_core.cpp
View File

@@ -556,6 +556,75 @@ HWTEST2_F(CommandQueueCommandsXeHpc, givenFlushTaskSubmissionEnabledAndSplitBcsC
context->freeMem(dstPtr);
}
HWTEST2_F(CommandQueueCommandsXeHpc, givenRelaxedOrderingNotAllowedWhenDispatchSplitThenUseSemaphores, IsXeHpcCore) {
using MI_FLUSH_DW = typename FamilyType::MI_FLUSH_DW;
DebugManagerStateRestore restorer;
DebugManager.flags.SplitBcsCopy.set(1);
ze_result_t returnValue;
auto hwInfo = *NEO::defaultHwInfo;
hwInfo.featureTable.ftrBcsInfo = 0b111111111;
hwInfo.capabilityTable.blitterOperationsSupported = true;
auto testNeoDevice = NEO::MockDevice::createWithNewExecutionEnvironment<NEO::MockDevice>(&hwInfo);
auto testL0Device = std::unique_ptr<L0::Device>(L0::Device::create(driverHandle.get(), testNeoDevice, false, &returnValue));
ze_command_queue_desc_t desc = {};
desc.ordinal = static_cast<uint32_t>(testNeoDevice->getEngineGroupIndexFromEngineGroupType(NEO::EngineGroupType::Copy));
std::unique_ptr<L0::CommandList> commandList0(CommandList::createImmediate(productFamily,
testL0Device.get(),
&desc,
false,
NEO::EngineGroupType::Copy,
returnValue));
ASSERT_NE(nullptr, commandList0);
constexpr size_t alignment = 4096u;
constexpr size_t size = 8 * MemoryConstants::megaByte;
void *srcPtr;
void *dstPtr;
ze_device_mem_alloc_desc_t deviceDesc = {};
context->allocDeviceMem(device->toHandle(),
&deviceDesc,
size, alignment, &srcPtr);
ze_host_mem_alloc_desc_t hostDesc = {};
context->allocHostMem(&hostDesc, size, alignment, &dstPtr);
auto ultCsr = static_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(commandList0->csr);
ultCsr->recordFlusheBatchBuffer = true;
EXPECT_EQ(1u, ultCsr->getNumClients());
auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(*ultCsr);
ultCsr->directSubmission.reset(directSubmission);
auto result = commandList0->appendMemoryCopy(dstPtr, srcPtr, size, nullptr, 0, nullptr, false);
ASSERT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[0])->getTaskCount(), 0u);
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[1])->getTaskCount(), 0u);
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[2])->getTaskCount(), 0u);
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[3])->getTaskCount(), 0u);
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[0])->getCsr()->peekTaskCount(), 1u);
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[1])->getCsr()->peekTaskCount(), 1u);
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[2])->getCsr()->peekTaskCount(), 1u);
EXPECT_EQ(static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[3])->getCsr()->peekTaskCount(), 1u);
EXPECT_FALSE(ultCsr->latestFlushedBatchBuffer.hasRelaxedOrderingDependencies);
uint32_t semaphoresFound = 0;
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(cmdList, commandList0->commandContainer.getCommandStream()->getCpuBase(), commandList0->commandContainer.getCommandStream()->getUsed()));
for (auto &cmd : cmdList) {
if (genCmdCast<typename FamilyType::MI_SEMAPHORE_WAIT *>(cmd)) {
semaphoresFound++;
}
}
EXPECT_EQ(4u, semaphoresFound);
context->freeMem(srcPtr);
context->freeMem(dstPtr);
}
HWTEST2_F(CommandQueueCommandsXeHpc, givenSplitBcsCopyAndImmediateCommandListWhenAppendingMemoryCopyThenSuccessIsReturned, IsXeHpcCore) {
using MI_FLUSH_DW = typename FamilyType::MI_FLUSH_DW;