intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-07 12:42:54 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: remove doubled memory prefetch operation when executing command list

Browse Source 
Related-To: NEO-10356

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2025-04-04 10:13:07 +00:00
committed by Compute-Runtime-Automation
parent ecf8a07d26
commit 58fe89e116
3 changed files with 50 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
View File

@@ -432,12 +432,13 @@ template <GFXCORE_FAMILY gfxCoreFamily>
inline ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommandListImmediateWithFlushTaskImpl(bool performMigration, bool hasStallingCmds, bool hasRelaxedOrderingDependencies, NEO::AppendOperations appendOperation,
bool requireTaskCountUpdate, CommandQueue *cmdQ,
MutexLock *outerLock) {
auto cmdQImp = static_cast<CommandQueueImp *>(cmdQ);
this->commandContainer.removeDuplicatesFromResidencyContainer();
auto commandStream = this->commandContainer.getCommandStream();
size_t commandStreamStart = this->cmdListCurrentStartOffset;
if (appendOperation == NEO::AppendOperations::cmdList && this->dispatchCmdListBatchBufferAsPrimary) {
auto cmdListStartCmdBufferStream = reinterpret_cast<CommandQueueImp *>(cmdQ)->getStartingCmdBuffer();
auto cmdListStartCmdBufferStream = cmdQImp->getStartingCmdBuffer();
// check if queue starting stream is the same as immediate,
// if they are the same - immediate command list buffer has preamble in it including jump from immediate to regular cmdlist - proceed normal
// if not - regular cmdlist is the starting command buffer - no queue preamble or waiting commands
@@ -447,15 +448,17 @@ inline ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommand
}
}
auto csr = static_cast<CommandQueueImp *>(cmdQ)->getCsr();
auto csr = cmdQImp->getCsr();
auto lockCSR = outerLock != nullptr ? std::move(*outerLock) : csr->obtainUniqueOwnership();
if (NEO::ApiSpecificConfig::isSharedAllocPrefetchEnabled()) {
auto svmAllocMgr = this->device->getDriverHandle()->getSvmAllocsManager();
svmAllocMgr->prefetchSVMAllocs(*this->device->getNEODevice(), *csr);
}
if (appendOperation != NEO::AppendOperations::cmdList) {
if (NEO::ApiSpecificConfig::isSharedAllocPrefetchEnabled()) {
auto svmAllocMgr = this->device->getDriverHandle()->getSvmAllocsManager();
svmAllocMgr->prefetchSVMAllocs(*this->device->getNEODevice(), *csr);
}
cmdQ->registerCsrClient();
cmdQ->registerCsrClient();
}
std::unique_lock<std::mutex> lockForIndirect;
if (this->hasIndirectAllocationsAllowed()) {
@@ -501,7 +504,6 @@ inline ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommand
return ZE_RESULT_ERROR_OUT_OF_DEVICE_MEMORY;
}
auto cmdQImp = static_cast<CommandQueueImp *>(cmdQ);
cmdQImp->clearHeapContainer();
// save offset from immediate stream - even when not used to dispatch commands, can be used for epilogue

38
level_zero/core/test/unit_tests/xe_hpc_core/test_cmdlist_xe_hpc_core.cpp
View File

@@ -223,6 +223,44 @@ HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenForceMemoryPrefetchForKmdMigra
commandQueue->destroy();
}
HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenForceMemoryPrefetchForKmdMigratedSharedAllocationsWhenExecutingCommandListsOnImmediateCommandListThenMemoryPrefetchIsCalledOnce, IsXeHpcCore) {
DebugManagerStateRestore restore;
debugManager.flags.UseKmdMigration.set(true);
debugManager.flags.ForceMemoryPrefetchForKmdMigratedSharedAllocations.set(true);
debugManager.flags.EnableBOChunkingPrefetch.set(false);
size_t size = 10;
size_t alignment = 1u;
void *ptr = nullptr;
ze_device_mem_alloc_desc_t deviceDesc = {};
ze_host_mem_alloc_desc_t hostDesc = {};
auto result = context->allocSharedMem(device->toHandle(), &deviceDesc, &hostDesc, size, alignment, &ptr);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_NE(nullptr, ptr);
ze_result_t returnValue;
ze_command_queue_desc_t queueDesc = {};
ze_command_list_handle_t commandListHandle = CommandList::create(productFamily, device, NEO::EngineGroupType::compute, 0u, returnValue, false)->toHandle();
auto commandList = CommandList::fromHandle(commandListHandle);
commandList->close();
auto commandListImmediate = CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::compute, result);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
result = commandListImmediate->appendCommandLists(1, &commandListHandle, nullptr, 0, nullptr);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
auto mockMemoryManager = reinterpret_cast<NEO::MockMemoryManager *>(neoDevice->getMemoryManager());
EXPECT_TRUE(mockMemoryManager->setMemPrefetchCalled);
EXPECT_EQ(1u, mockMemoryManager->setMemPrefetchCalledCount);
context->freeMem(ptr);
commandList->destroy();
commandListImmediate->destroy();
}
HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenNoForceMemoryPrefetchForKmdMigratedSharedAllocationsAndNoEnableBOChunkingPrefetchWhenExecutingCommandListsOnCommandQueueThenMemoryPrefetchIsNotCalled, IsXeHpcCore) {
DebugManagerStateRestore restore;
debugManager.flags.UseKmdMigration.set(true);