intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 14:02:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature(ocl) use tags to synchronize multi root device events

Browse Source 
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
This commit is contained in:
Maciej Plewka
2023-01-19 16:11:39 +00:00
committed by Compute-Runtime-Automation
parent a27834ff7d
commit fa4830036a
56 changed files with 683 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
opencl/source/helpers/properties_helper.cpp
View File

@@ -26,7 +26,6 @@ namespace NEO {
void flushDependentCsr(CommandStreamReceiver &dependentCsr, CsrDependencies &csrDeps) {
auto csrOwnership = dependentCsr.obtainUniqueOwnership();
dependentCsr.updateTagFromWait();
csrDeps.taskCountContainer.push_back({dependentCsr.peekTaskCount(), reinterpret_cast<uint64_t>(dependentCsr.getTagAddress())});
}
void EventsRequest::fillCsrDependenciesForTimestampPacketContainer(CsrDependencies &csrDeps, CommandStreamReceiver &currentCsr, CsrDependencies::DependenciesType depsType) const {
@@ -68,23 +67,22 @@ void EventsRequest::fillCsrDependenciesForTimestampPacketContainer(CsrDependenci
}
}
void EventsRequest::fillCsrDependenciesForTaskCountContainer(CsrDependencies &csrDeps, CommandStreamReceiver &currentCsr) const {
void EventsRequest::fillCsrDependenciesForRootDevices(CsrDependencies &csrDeps, CommandStreamReceiver &currentCsr) const {
for (cl_uint i = 0; i < this->numEventsInWaitList; i++) {
auto event = castToObjectOrAbort<Event>(this->eventWaitList[i]);
if (event->isUserEvent() || CompletionStamp::notReady == event->peekTaskCount()) {
continue;
}
if (event->getCommandQueue() && event->getCommandQueue()->getDevice().getRootDeviceIndex() != currentCsr.getRootDeviceIndex()) {
auto timestampPacketContainer = event->getMultiRootDeviceTimestampPacketNodes();
if (!timestampPacketContainer || timestampPacketContainer->peekNodes().empty()) {
continue;
}
auto &dependentCsr = event->getCommandQueue()->getGpgpuCommandStreamReceiver();
if (!dependentCsr.isLatestTaskCountFlushed()) {
flushDependentCsr(dependentCsr, csrDeps);
} else {
csrDeps.taskCountContainer.push_back({event->peekTaskCount(), reinterpret_cast<uint64_t>(dependentCsr.getTagAddress())});
}
auto graphicsAllocation = event->getCommandQueue()->getGpgpuCommandStreamReceiver().getTagsMultiAllocation()->getGraphicsAllocation(currentCsr.getRootDeviceIndex());
currentCsr.getResidencyAllocations().push_back(graphicsAllocation);
csrDeps.multiRootTimeStampSyncContainer.push_back(timestampPacketContainer);
}
}
}