intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-25 05:24:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: extend regkey to force prefetch of shared memory for cmd list immediate

Browse Source 
Related-To: NEO-7841

Signed-off-by: Milczarek, Slawomir <slawomir.milczarek@intel.com>
This commit is contained in:
Milczarek, Slawomir
2023-06-01 05:53:21 +00:00
committed by Compute-Runtime-Automation
parent d7332b8043
commit ac3a2b6756
2 changed files with 39 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
View File

@@ -229,6 +229,11 @@ inline ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommand
auto csr = static_cast<CommandQueueImp *>(cmdQ)->getCsr();
auto lockCSR = csr->obtainUniqueOwnership();
if (NEO::DebugManager.flags.ForceMemoryPrefetchForKmdMigratedSharedAllocations.get()) {
auto svmAllocMgr = this->device->getDriverHandle()->getSvmAllocsManager();
svmAllocMgr->prefetchSVMAllocs(*this->device->getNEODevice(), *csr);
}
if (cmdQ->getClientId() == CommandQueue::clientNotRegistered) {
cmdQ->setClientId(csr->registerClient());
}

39
level_zero/core/test/unit_tests/xe_hpc_core/test_cmdlist_xe_hpc_core.cpp
View File

@@ -121,17 +121,13 @@ HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenForceMemoryPrefetchForKmdMigra
DebugManager.flags.UseKmdMigration.set(true);
DebugManager.flags.ForceMemoryPrefetchForKmdMigratedSharedAllocations.set(true);
auto pCommandList = std::make_unique<WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>>();
auto result = pCommandList->initialize(device, NEO::EngineGroupType::Compute, 0u);
ASSERT_EQ(ZE_RESULT_SUCCESS, result);
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 = {};
result = context->allocSharedMem(device->toHandle(), &deviceDesc, &hostDesc, size, alignment, &ptr);
auto result = context->allocSharedMem(device->toHandle(), &deviceDesc, &hostDesc, size, alignment, &ptr);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_NE(nullptr, ptr);
@@ -155,6 +151,39 @@ HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenForceMemoryPrefetchForKmdMigra
commandQueue->destroy();
}
HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenForceMemoryPrefetchForKmdMigratedSharedAllocationsWhenExecutingCommandListImmediateWithFlushTaskThenMemoryPrefetchIsCalled, IsXeHpcCore) {
DebugManagerStateRestore restore;
DebugManager.flags.UseKmdMigration.set(true);
DebugManager.flags.ForceMemoryPrefetchForKmdMigratedSharedAllocations.set(true);
auto pCommandList = std::make_unique<WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>>();
auto result = pCommandList->initialize(device, NEO::EngineGroupType::Compute, 0u);
ASSERT_EQ(ZE_RESULT_SUCCESS, result);
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 = {};
result = context->allocSharedMem(device->toHandle(), &deviceDesc, &hostDesc, size, alignment, &ptr);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_NE(nullptr, ptr);
const ze_command_queue_desc_t desc = {};
ze_result_t returnValue;
std::unique_ptr<L0::CommandList> commandList(CommandList::createImmediate(productFamily, device, &desc, false, NEO::EngineGroupType::RenderCompute, returnValue));
auto &commandListImmediate = static_cast<MockCommandListImmediate<gfxCoreFamily> &>(*commandList);
result = commandListImmediate.executeCommandListImmediateWithFlushTask(false, false, false);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
auto mockMemoryManager = reinterpret_cast<NEO::MockMemoryManager *>(neoDevice->getMemoryManager());
EXPECT_TRUE(mockMemoryManager->setMemPrefetchCalled);
context->freeMem(ptr);
}
HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenAppendMemoryPrefetchForKmdMigratedSharedAllocationsWhenPrefetchApiIsCalledThenRequestMemoryPrefetch, IsXeHpcCore) {
DebugManagerStateRestore restore;