intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-23 11:03:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Optimize timestamp packet dependencies

Browse Source 
- Clear dependencies even if last engine changed
- Do no program semaphore waiting for blit when blit is submitted with gpgpu
- Track barrier timestamps to correctly synchronize blits in OOQ

Related-To: NEO-6444
Signed-off-by: Maciej Dziuban <maciej.dziuban@intel.com>
This commit is contained in:
Maciej Dziuban
2021-11-23 16:23:18 +00:00
committed by Compute-Runtime-Automation
parent f20236c7f2
commit e3bb526067
11 changed files with 534 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
opencl/source/command_queue/command_queue.cpp
View File

@@ -681,7 +681,6 @@ void CommandQueue::updateBcsTaskCount(aub_stream::EngineType bcsEngineType, uint
uint32_t CommandQueue::peekBcsTaskCount(aub_stream::EngineType bcsEngineType) const {
const CopyEngineState &state = bcsStates[EngineHelpers::getBcsIndex(bcsEngineType)];
DEBUG_BREAK_IF(!state.isValid());
return state.taskCount;
}
@@ -706,10 +705,6 @@ void CommandQueue::obtainNewTimestampPacketNodes(size_t numberOfNodes, Timestamp
previousNodes.swapNodes(*timestampPacketContainer);
if ((previousNodes.peekNodes().size() > 0) && (previousNodes.peekNodes()[0]->getAllocator() != allocator)) {
clearAllDependencies = false;
}
if (clearAllDependencies) {
previousNodes.moveNodesToNewContainer(*deferredTimestampPackets);
}
@@ -1007,4 +1002,61 @@ void CommandQueue::waitForAllEngines(bool blockedQueue, PrintfHandler *printfHan
}
}
void CommandQueue::setupBarrierTimestampForBcsEngines(aub_stream::EngineType engineType, TimestampPacketDependencies &timestampPacketDependencies) {
if (!getGpgpuCommandStreamReceiver().isStallingCommandsOnNextFlushRequired()) {
return;
}
// Ensure we have exactly 1 barrier node.
if (timestampPacketDependencies.barrierNodes.peekNodes().empty()) {
timestampPacketDependencies.barrierNodes.add(getGpgpuCommandStreamReceiver().getTimestampPacketAllocator()->getTag());
}
if (isOOQEnabled()) {
// Barrier node will be signalled on gpgpuCsr. Save it for later use on blitters.
for (auto currentBcsIndex = 0u; currentBcsIndex < bcsTimestampPacketContainers.size(); currentBcsIndex++) {
const auto currentBcsEngineType = EngineHelpers::mapBcsIndexToEngineType(currentBcsIndex, true);
if (currentBcsEngineType == engineType) {
// Node is already added to barrierNodes for this engine, no need to save it.
continue;
}
// Save latest timestamp (override previous, if any).
TimestampPacketContainer newContainer{};
newContainer.assignAndIncrementNodesRefCounts(timestampPacketDependencies.barrierNodes);
bcsTimestampPacketContainers[currentBcsIndex].lastBarrierToWaitFor.swapNodes(newContainer);
}
}
}
void CommandQueue::processBarrierTimestampForBcsEngine(aub_stream::EngineType bcsEngineType, TimestampPacketDependencies &blitDependencies) {
BcsTimestampPacketContainers &bcsContainers = bcsTimestampPacketContainers[EngineHelpers::getBcsIndex(bcsEngineType)];
bcsContainers.lastBarrierToWaitFor.moveNodesToNewContainer(blitDependencies.barrierNodes);
}
void CommandQueue::setLastBcsPacket(aub_stream::EngineType bcsEngineType) {
if (isOOQEnabled()) {
TimestampPacketContainer dummyContainer{};
dummyContainer.assignAndIncrementNodesRefCounts(*this->timestampPacketContainer);
BcsTimestampPacketContainers &bcsContainers = bcsTimestampPacketContainers[EngineHelpers::getBcsIndex(bcsEngineType)];
bcsContainers.lastSignalledPacket.swapNodes(dummyContainer);
}
}
void CommandQueue::fillCsrDependenciesWithLastBcsPackets(CsrDependencies &csrDeps) {
for (BcsTimestampPacketContainers &bcsContainers : bcsTimestampPacketContainers) {
if (bcsContainers.lastSignalledPacket.peekNodes().empty()) {
continue;
}
csrDeps.timestampPacketContainer.push_back(&bcsContainers.lastSignalledPacket);
}
}
void CommandQueue::clearLastBcsPackets() {
for (BcsTimestampPacketContainers &bcsContainers : bcsTimestampPacketContainers) {
bcsContainers.lastSignalledPacket.moveNodesToNewContainer(*deferredTimestampPackets);
}
}
} // namespace NEO