intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: Submit every marker with dependencies

Browse Source 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk
2025-08-28 15:51:30 +00:00
committed by Compute-Runtime-Automation
parent 9b2802af0e
commit 9d91e85f5c
14 changed files with 55 additions and 88 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
opencl/source/command_queue/command_queue.cpp
View File

@@ -1210,6 +1210,11 @@ bool CommandQueue::isBlockedCommandStreamRequired(uint32_t commandType, const Ev
return false;
}
bool CommandQueue::isDependenciesFlushForMarkerRequired(const EventsRequest &eventsRequest) const {
return this->getGpgpuCommandStreamReceiver().peekTimestampPacketWriteEnabled() && !this->isOOQEnabled() &&
(eventsRequest.outEvent || eventsRequest.numEventsInWaitList > 0);
}
void CommandQueue::storeProperties(const cl_queue_properties *properties) {
if (properties) {
for (size_t i = 0; properties[i] != 0; i += 2) {

2
opencl/source/command_queue/command_queue.h
View File

@@ -458,6 +458,8 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
virtual void obtainTaskLevelAndBlockedStatus(TaskCountType &taskLevel, cl_uint &numEventsInWaitList, const cl_event *&eventWaitList, bool &blockQueueStatus, unsigned int commandType){};
bool isBlockedCommandStreamRequired(uint32_t commandType, const EventsRequest &eventsRequest, bool blockedQueue, bool isMarkerWithProfiling) const;
bool isDependenciesFlushForMarkerRequired(const EventsRequest &eventsRequest) const;
MOCKABLE_VIRTUAL void obtainNewTimestampPacketNodes(size_t numberOfNodes, TimestampPacketContainer &previousNodes, bool clearAllDependencies, CommandStreamReceiver &csr);
void storeProperties(const cl_queue_properties *properties);
void processProperties(const cl_queue_properties *properties);

6
opencl/source/command_queue/enqueue_common.h
View File

@@ -161,8 +161,6 @@ cl_int CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
EventBuilder eventBuilder;
setupEvent(eventBuilder, event, commandType);
const bool isFlushWithPostSyncWrite = isFlushForProfilingRequired(commandType) && ((eventBuilder.getEvent() && eventBuilder.getEvent()->isProfilingEnabled()) || multiDispatchInfo.peekBuiltinOpParams().bcsSplit);
std::unique_ptr<KernelOperation> blockedCommandsData;
std::unique_ptr<PrintfHandler> printfHandler;
TakeOwnershipWrapper<CommandQueueHw<GfxFamily>> queueOwnership(*this);
@@ -190,6 +188,8 @@ cl_int CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
CsrDependencies csrDeps;
BlitPropertiesContainer blitPropertiesContainer;
const bool isFlushWithPostSyncWrite = isFlushForProfilingRequired(commandType) && ((eventBuilder.getEvent() && eventBuilder.getEvent()->isProfilingEnabled()) || multiDispatchInfo.peekBuiltinOpParams().bcsSplit || this->isDependenciesFlushForMarkerRequired(eventsRequest));
if (this->context->getRootDeviceIndices().size() > 1) {
eventsRequest.fillCsrDependenciesForRootDevices(csrDeps, computeCommandStreamReceiver);
}
@@ -198,7 +198,7 @@ cl_int CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
const auto &hwInfo = this->getDevice().getHardwareInfo();
auto &productHelper = getDevice().getProductHelper();
bool canUsePipeControlInsteadOfSemaphoresForOnCsrDependencies = false;
bool isNonStallingIoqBarrier = (CL_COMMAND_BARRIER == commandType) && !isOOQEnabled() && (debugManager.flags.OptimizeIoqBarriersHandling.get() != 0);
bool isNonStallingIoqBarrier = isFlushForProfilingRequired(commandType) && !isOOQEnabled() && (debugManager.flags.OptimizeIoqBarriersHandling.get() != 0);
const bool isNonStallingIoqBarrierWithDependencies = isNonStallingIoqBarrier && (eventsRequest.numEventsInWaitList > 0);
if (computeCommandStreamReceiver.peekTimestampPacketWriteEnabled()) {