intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-10 15:12:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: Stop direct submission before removing host ptrs

Browse Source 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk
2024-05-28 10:15:39 +00:00
committed by Compute-Runtime-Automation
parent 387ec34207
commit 7e6a08098b
6 changed files with 73 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
level_zero/core/source/cmdlist/cmdlist.cpp
View File

@@ -52,10 +52,30 @@ void CommandList::storePrintfKernel(Kernel *kernel) {
void CommandList::removeHostPtrAllocations() {
auto memoryManager = device ? device->getNEODevice()->getMemoryManager() : nullptr;
bool restartDirectSubmission = !this->hostPtrMap.empty() && memoryManager && device->getNEODevice()->getRootDeviceEnvironment().getProductHelper().restartDirectSubmissionForHostptrFree();
if (restartDirectSubmission) {
const auto &engines = memoryManager->getRegisteredEngines(device->getRootDeviceIndex());
for (const auto &engine : engines) {
auto lock = engine.commandStreamReceiver->obtainUniqueOwnership();
engine.commandStreamReceiver->stopDirectSubmission(false);
}
}
for (auto &allocation : hostPtrMap) {
UNRECOVERABLE_IF(memoryManager == nullptr);
memoryManager->freeGraphicsMemory(allocation.second);
}
if (restartDirectSubmission) {
const auto &engines = memoryManager->getRegisteredEngines(device->getRootDeviceIndex());
for (const auto &engine : engines) {
if (engine.commandStreamReceiver->isAnyDirectSubmissionEnabled()) {
engine.commandStreamReceiver->flushTagUpdate();
}
}
}
hostPtrMap.clear();
}

42
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_3.cpp
View File

@@ -314,6 +314,48 @@ HWTEST2_F(CommandListCreate,
device->getNEODevice()->getMemoryManager()->freeSystemMemory(cmdListHostBuffer);
}
HWTEST2_F(CommandListCreate,
givenCmdListHostPointerUsedWhenRemoveHostPtrAllocationThenStopDirectSubmissions,
IsAtLeastSkl) {
const auto &engines = device->getNEODevice()->getMemoryManager()->getRegisteredEngines(device->getRootDeviceIndex());
for (const auto &engine : engines) {
auto ultCsr = static_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(engine.commandStreamReceiver);
ultCsr->callFlushTagUpdate = false;
ultCsr->directSubmissionAvailable = true;
ultCsr->callBaseStopDirectSubmission = false;
}
auto commandList = std::make_unique<::L0::ult::CommandListCoreFamily<gfxCoreFamily>>();
commandList->initialize(device, NEO::EngineGroupType::renderCompute, 0u);
size_t cmdListHostPtrSize = MemoryConstants::pageSize;
void *cmdListHostBuffer = device->getNEODevice()->getMemoryManager()->allocateSystemMemory(cmdListHostPtrSize, cmdListHostPtrSize);
AlignedAllocationData outData = commandList->getAlignedAllocationData(device, cmdListHostBuffer, cmdListHostPtrSize, false);
ASSERT_NE(nullptr, outData.alloc);
for (const auto &engine : engines) {
auto ultCsr = static_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(engine.commandStreamReceiver);
EXPECT_FALSE(ultCsr->stopDirectSubmissionCalled);
EXPECT_FALSE(ultCsr->flushTagUpdateCalled);
}
commandList->removeHostPtrAllocations();
for (const auto &engine : engines) {
auto ultCsr = static_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(engine.commandStreamReceiver);
if (device->getNEODevice()->getRootDeviceEnvironment().getProductHelper().restartDirectSubmissionForHostptrFree()) {
EXPECT_TRUE(ultCsr->stopDirectSubmissionCalled);
EXPECT_TRUE(ultCsr->flushTagUpdateCalled);
} else {
EXPECT_FALSE(ultCsr->stopDirectSubmissionCalled);
EXPECT_FALSE(ultCsr->flushTagUpdateCalled);
}
}
device->getNEODevice()->getMemoryManager()->freeSystemMemory(cmdListHostBuffer);
}
using PlatformSupport = IsWithinProducts<IGFX_SKYLAKE, IGFX_DG1>;
HWTEST2_F(CommandListCreate,
givenCommandListWhenMemoryCopyRegionHavingHostMemoryWithSignalAndWaitScopeEventsUsingRenderEngineThenPipeControlsWithDcFlushIsFound,