intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-05 09:09:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Avoid implicit sync for async mode immediate copy queue

Browse Source 
Related-To: LOCI-1988

Signed-off-by: Aravind Gopalakrishnan <aravind.gopalakrishnan@intel.com>
This commit is contained in:
Aravind Gopalakrishnan
2022-04-25 23:49:47 +00:00
committed by Compute-Runtime-Automation
parent ced22d45e9
commit 31b2433b2f
10 changed files with 158 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
level_zero/core/source/cmdlist/cmdlist.h
View File

@@ -257,6 +257,7 @@ struct CommandList : _ze_command_list_handle_t {
uint32_t partitionCount = 1;
bool isFlushTaskSubmissionEnabled = false;
bool isSyncModeQueue = false;
bool isTbxMode = false;
bool commandListSLMEnabled = false;
bool requiresQueueUncachedMocs = false;

22
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -117,12 +117,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::initialize(Device *device, NEO
this->partitionCount = static_cast<uint32_t>(this->device->getNEODevice()->getDeviceBitfield().count());
}
if (this->cmdListType == CommandListType::TYPE_IMMEDIATE && !isCopyOnly() && !isInternal()) {
const auto &hwInfo = device->getHwInfo();
this->isFlushTaskSubmissionEnabled = NEO::HwHelper::get(hwInfo.platform.eRenderCoreFamily).isPlatformFlushTaskEnabled(hwInfo);
if (NEO::DebugManager.flags.EnableFlushTaskSubmission.get() != -1) {
this->isFlushTaskSubmissionEnabled = !!NEO::DebugManager.flags.EnableFlushTaskSubmission.get();
}
if (this->isFlushTaskSubmissionEnabled) {
commandContainer.setFlushTaskUsedForImmediate(this->isFlushTaskSubmissionEnabled);
}
@@ -149,17 +144,22 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::executeCommandListImmediate(bo
this->close();
ze_command_list_handle_t immediateHandle = this->toHandle();
this->commandContainer.removeDuplicatesFromResidencyContainer();
const auto commandListExecutionResult = this->cmdQImmediate->executeCommandLists(1, &immediateHandle, nullptr, performMigration);
if (commandListExecutionResult == ZE_RESULT_ERROR_DEVICE_LOST) {
return commandListExecutionResult;
}
const auto synchronizationResult = this->cmdQImmediate->synchronize(std::numeric_limits<uint64_t>::max());
if (synchronizationResult == ZE_RESULT_ERROR_DEVICE_LOST) {
return synchronizationResult;
}
if (this->isCopyOnly() && !this->isSyncModeQueue && !this->isTbxMode) {
this->commandContainer.currentLinearStreamStartOffset = this->commandContainer.getCommandStream()->getUsed();
} else {
const auto synchronizationResult = this->cmdQImmediate->synchronize(std::numeric_limits<uint64_t>::max());
if (synchronizationResult == ZE_RESULT_ERROR_DEVICE_LOST) {
return synchronizationResult;
}
this->reset();
this->reset();
}
return ZE_RESULT_SUCCESS;
}

8
level_zero/core/source/cmdlist/cmdlist_imp.cpp
View File

@@ -114,6 +114,13 @@ CommandList *CommandList::createImmediate(uint32_t productFamily, Device *device
commandList->internalUsage = internalUsage;
commandList->cmdListType = CommandListType::TYPE_IMMEDIATE;
commandList->isSyncModeQueue = (desc->mode == ZE_COMMAND_QUEUE_MODE_SYNCHRONOUS);
if (!(NEO::EngineGroupType::Copy == engineType) && !internalUsage) {
const auto &hwInfo = device->getHwInfo();
commandList->isFlushTaskSubmissionEnabled = NEO::HwHelper::get(hwInfo.platform.eRenderCoreFamily).isPlatformFlushTaskEnabled(hwInfo);
if (NEO::DebugManager.flags.EnableFlushTaskSubmission.get() != -1) {
commandList->isFlushTaskSubmissionEnabled = !!NEO::DebugManager.flags.EnableFlushTaskSubmission.get();
}
}
returnValue = commandList->initialize(device, engineType, desc->flags);
if (returnValue != ZE_RESULT_SUCCESS) {
commandList->destroy();
@@ -130,6 +137,7 @@ CommandList *CommandList::createImmediate(uint32_t productFamily, Device *device
commandList->cmdQImmediate = commandQueue;
commandList->csr = csr;
commandList->isTbxMode = (csr->getType() == NEO::CommandStreamReceiverType::CSR_TBX) || (csr->getType() == NEO::CommandStreamReceiverType::CSR_TBX_WITH_AUB);
commandList->commandListPreemptionMode = device->getDevicePreemptionMode();
return commandList;
}

7
level_zero/core/source/cmdqueue/cmdqueue_hw.inl
View File

@@ -369,6 +369,7 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandLists(
auto commandList = CommandList::fromHandle(phCommandLists[i]);
auto &cmdBufferAllocations = commandList->commandContainer.getCmdBufferAllocations();
auto cmdBufferCount = cmdBufferAllocations.size();
bool immediateMode = (commandList->cmdListType == CommandList::CommandListType::TYPE_IMMEDIATE) ? true : false;
auto commandListPreemption = commandList->getCommandListPreemptionMode();
if (statePreemption != commandListPreemption) {
@@ -413,7 +414,11 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandLists(
for (size_t iter = 0; iter < cmdBufferCount; iter++) {
auto allocation = cmdBufferAllocations[iter];
NEO::EncodeBatchBufferStartOrEnd<GfxFamily>::programBatchBufferStart(&child, allocation->getGpuAddress(), true);
uint64_t startOffset = allocation->getGpuAddress();
if (immediateMode && (iter == (cmdBufferCount - 1))) {
startOffset = ptrOffset(allocation->getGpuAddress(), commandList->commandContainer.currentLinearStreamStartOffset);
}
NEO::EncodeBatchBufferStartOrEnd<GfxFamily>::programBatchBufferStart(&child, startOffset, true);
}
printfFunctionContainer.insert(printfFunctionContainer.end(),