Skip to content

GPU: Fix handling of non critical errors in double-pipeline mode #14870

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
davidrohr merged 1 commit into from
Nov 28, 2025
+12 −14
Merged

GPU: Fix handling of non critical errors in double-pipeline mode #14870

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 9 additions & 12 deletions GPU/GPUTracking/Base/GPUReconstruction.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@ struct GPUReconstructionPipelineQueue {
} // namespace

struct GPUReconstructionPipelineContext {
std::queue<GPUReconstructionPipelineQueue*> queue;
std::queue<GPUReconstructionPipelineQueue*> pipelineQueue;
std::mutex mutex;
std::condition_variable cond;
bool terminate = false;
Expand Down Expand Up @@ -1089,13 +1089,13 @@ void GPUReconstruction::RunPipelineWorker()
while (!terminate) {
{
std::unique_lock<std::mutex> lk(mPipelineContext->mutex);
mPipelineContext->cond.wait(lk, [this] { return this->mPipelineContext->queue.size() > 0; });
mPipelineContext->cond.wait(lk, [this] { return this->mPipelineContext->pipelineQueue.size() > 0; });
}
GPUReconstructionPipelineQueue* q;
{
std::lock_guard<std::mutex> lk(mPipelineContext->mutex);
q = mPipelineContext->queue.front();
mPipelineContext->queue.pop();
q = mPipelineContext->pipelineQueue.front();
mPipelineContext->pipelineQueue.pop();
}
if (q->op == 1) {
terminate = 1;
Expand Down Expand Up @@ -1132,26 +1132,23 @@ int32_t GPUReconstruction::EnqueuePipeline(bool terminate)
if (rec->mPipelineContext->terminate) {
throw std::runtime_error("Must not enqueue work after termination request");
}
rec->mPipelineContext->queue.push(q);
rec->mPipelineContext->pipelineQueue.push(q);
rec->mPipelineContext->terminate = terminate;
rec->mPipelineContext->cond.notify_one();
}
q->c.wait(lkdone, [&q]() { return q->done; });
if (q->retVal) {
if (terminate || (q->retVal && (q->retVal != 3 || !GetProcessingSettings().ignoreNonFatalGPUErrors))) {
return q->retVal;
}
if (terminate) {
return 0;
} else {
return mChains[0]->FinalizePipelinedProcessing();
}
int32_t retVal2 = mChains[0]->FinalizePipelinedProcessing();
return retVal2 ? retVal2 : q->retVal;
}

GPUChain* GPUReconstruction::GetNextChainInQueue()
{
GPUReconstruction* rec = mMaster ? mMaster : this;
std::lock_guard<std::mutex> lk(rec->mPipelineContext->mutex);
return rec->mPipelineContext->queue.size() && rec->mPipelineContext->queue.front()->op == 0 ? rec->mPipelineContext->queue.front()->chain : nullptr;
return rec->mPipelineContext->pipelineQueue.size() && rec->mPipelineContext->pipelineQueue.front()->op == 0 ? rec->mPipelineContext->pipelineQueue.front()->chain : nullptr;
}

void GPUReconstruction::PrepareEvent() // TODO: Clean this up, this should not be called from chainTracking but before
Expand Down
5 changes: 3 additions & 2 deletions GPU/GPUTracking/Global/GPUChainTrackingCompression.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,8 +62,9 @@ int32_t GPUChainTracking::RunTPCCompression()
#ifdef GPUCA_TPC_GEOMETRY_O2
if (mPipelineFinalizationCtx && GetProcessingSettings().doublePipelineClusterizer) {
SynchronizeEventAndRelease(mEvents->single);
((GPUChainTracking*)GetNextChainInQueue())->RunTPCClusterizer_prepare(false);
((GPUChainTracking*)GetNextChainInQueue())->mCFContext->ptrClusterNativeSave = processorsShadow()->ioPtrs.clustersNative;
auto* foreignChain = (GPUChainTracking*)GetNextChainInQueue();
foreignChain->RunTPCClusterizer_prepare(false);
foreignChain->mCFContext->ptrClusterNativeSave = processorsShadow()->ioPtrs.clustersNative;
}
#endif
SynchronizeStream(0);
Expand Down