intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-22 10:17:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add memory prefetch support for cmd list copy-only

Browse Source 
Related-To: NEO-6740

Signed-off-by: Milczarek, Slawomir <slawomir.milczarek@intel.com>
This commit is contained in:
Milczarek, Slawomir
2022-12-05 15:12:17 +00:00
committed by Compute-Runtime-Automation
parent 4adba15dbb
commit 4fd21cf59c
2 changed files with 82 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
level_zero/core/source/cmdqueue/cmdqueue_hw.inl
View File

@@ -251,6 +251,7 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandListsCopyOnly(
auto commandList = CommandList::fromHandle(phCommandLists[i]);
this->programOneCmdListBatchBufferStart(commandList, child);
this->mergeOneCmdListPipelinedState(commandList);
this->prefetchMemoryToDeviceAssociatedWithCmdList(commandList);
}
this->migrateSharedAllocationsIfRequested(ctx.isMigrationRequested, phCommandLists[0]);

81
level_zero/core/test/unit_tests/xe_hpc_core/test_cmdlist_xe_hpc_core.cpp
View File

@@ -389,6 +389,87 @@ HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenAppendMemoryPrefetchForKmdMigr
commandQueue->destroy();
}
HWTEST2_F(CommandListStatePrefetchXeHpcCore, givenAppendMemoryPrefetchForKmdMigratedSharedAllocationsSetWhenPrefetchApiIsCalledForUnifiedSharedMemoryOnCmdListCopyOnlyThenCallMigrateAllocationsToGpu, IsXeHpcCore) {
using GfxFamily = typename NEO::GfxFamilyMapper<gfxCoreFamily>::GfxFamily;
using POSTSYNC_DATA = typename FamilyType::POSTSYNC_DATA;
using WALKER_TYPE = typename FamilyType::WALKER_TYPE;
DebugManagerStateRestore restore;
DebugManager.flags.AppendMemoryPrefetchForKmdMigratedSharedAllocations.set(1);
DebugManager.flags.UseKmdMigration.set(1);
neoDevice->deviceBitfield = 0b001;
auto memoryManager = static_cast<MockMemoryManager *>(device->getDriverHandle()->getMemoryManager());
memoryManager->prefetchManager.reset(new MockPrefetchManager());
createKernel();
ze_result_t returnValue;
ze_command_queue_desc_t queueDesc = {};
ze_command_list_handle_t commandListHandle = CommandList::createImmediate(productFamily, device, &queueDesc, false, NEO::EngineGroupType::Copy, returnValue)->toHandle();
auto commandList = CommandList::fromHandle(commandListHandle);
auto commandQueue = CommandQueue::create(productFamily, device, neoDevice->getDefaultEngine().commandStreamReceiver, &queueDesc, commandList->isCopyOnly(), false, returnValue);
ze_event_pool_desc_t eventPoolDesc = {};
eventPoolDesc.count = 1;
eventPoolDesc.flags = ZE_EVENT_POOL_FLAG_KERNEL_TIMESTAMP;
ze_event_desc_t eventDesc = {};
eventDesc.index = 0;
auto eventPool = std::unique_ptr<EventPool>(EventPool::create(driverHandle.get(), context, 0, nullptr, &eventPoolDesc, returnValue));
EXPECT_EQ(ZE_RESULT_SUCCESS, returnValue);
auto event = std::unique_ptr<Event>(Event::create<uint32_t>(eventPool.get(), &eventDesc, device));
size_t size = 10;
size_t alignment = 1u;
void *srcPtr = nullptr;
void *dstPtr = 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, &srcPtr);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_NE(nullptr, srcPtr);
result = context->allocSharedMem(device->toHandle(), &deviceDesc, &hostDesc, size, alignment, &dstPtr);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_NE(nullptr, dstPtr);
result = commandList->appendMemoryPrefetch(srcPtr, size);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
result = commandList->appendMemoryPrefetch(dstPtr, size);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
auto prefetchManager = static_cast<MockPrefetchManager *>(memoryManager->prefetchManager.get());
EXPECT_EQ(2u, commandList->getPrefetchContext().allocations.size());
result = commandList->appendMemoryCopy(dstPtr, srcPtr, size, event->toHandle(), 0, nullptr);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
result = commandQueue->executeCommandLists(1, &commandListHandle, nullptr, true);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_TRUE(memoryManager->setMemPrefetchCalled);
EXPECT_EQ(0u, memoryManager->memPrefetchSubDeviceId);
EXPECT_TRUE(prefetchManager->migrateAllocationsToGpuCalled);
EXPECT_EQ(2u, commandList->getPrefetchContext().allocations.size());
commandList->reset();
EXPECT_TRUE(prefetchManager->removeAllocationsCalled);
EXPECT_EQ(0u, commandList->getPrefetchContext().allocations.size());
context->freeMem(srcPtr);
context->freeMem(dstPtr);
commandList->destroy();
commandQueue->destroy();
}
using CommandListEventFenceTestsXeHpcCore = Test<ModuleFixture>;
HWTEST2_F(CommandListEventFenceTestsXeHpcCore, givenCommandListWithProfilingEventAfterCommandWhenRevId03ThenMiFenceIsAdded, IsXeHpcCore) {