Skip to content

GPU: Fix segfault with --noEvents option, remove an unnecessary library, and improve/fix a couple of debug messages #14963

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

Open
davidrohr wants to merge 3 commits into
base: dev
Choose a base branch
from
+22 −18
Open

GPU: Fix segfault with --noEvents option, remove an unnecessary library, and improve/fix a couple of debug messages #14963

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1485341
GPU CUDA: Do not link against nvrtc library, which is not used
davidrohr Jan 16, 2026
f11d5f6
GPU: Fix direct memory allocation debug message
davidrohr Jan 16, 2026
b927376
GPU: Fix crash with --noEvents option, and improve some debug messages
davidrohr Jan 16, 2026
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
2 changes: 1 addition & 1 deletion GPU/GPUTracking/Base/GPUReconstruction.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -764,7 +764,7 @@ void* GPUReconstruction::AllocateDirectMemory(size_t size, int32_t type)
}
UpdateMaxMemoryUsed();
if (GetProcessingSettings().allocDebugLevel >= 2) {
std::cout << "Allocated (unmanaged " << (type == GPUMemoryResource::MEMORY_GPU ? "gpu" : "host") << "): " << size << " - available: " << ptrDiff(poolend, pool) << "\n";
std::cout << "Allocated (unmanaged " << ((type & GPUMemoryResource::MEMORY_GPU) ? "gpu" : "host") << "): " << size << " - available: " << ptrDiff(poolend, pool) << "\n";
}
return retVal;
}
Expand Down
2 changes: 1 addition & 1 deletion GPU/GPUTracking/Base/cuda/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -148,7 +148,7 @@ if (onnxruntime_FOUND)
endif()

# Setting target architecture and adding GPU libraries
target_link_libraries(${targetName} PRIVATE cuda cudart nvrtc)
target_link_libraries(${targetName} PRIVATE cuda cudart)
set_target_cuda_arch(${targetName})
#target_link_options(${targetName} PRIVATE "LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/version_script.ld")
#set_target_properties(${targetName} PROPERTIES LINK_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/version_script.ld)
Expand Down
36 changes: 20 additions & 16 deletions GPU/GPUTracking/Standalone/Benchmark/standalone.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -173,7 +173,7 @@ int32_t ReadConfiguration(int argc, char** argv)
return 1;
}
if (configStandalone.proc.doublePipeline && (configStandalone.runs < 4 || !configStandalone.outputcontrolmem)) {
printf("Double pipeline mode needs at least 3 runs per event and external output. To cycle though multiple events, use --preloadEvents and --runs n for n iterations round-robin\n");
printf("Double pipeline mode needs at least 4 runs per event and external output. To cycle though multiple events, use --preloadEvents and --runs n for n iterations round-robin\n");
return 1;
}
if (configStandalone.TF.bunchSim && configStandalone.TF.nMerge) {
Expand Down Expand Up @@ -297,7 +297,8 @@ int32_t SetupReconstruction()
printf("Error reading event config file\n");
return 1;
}
printf("Read event settings from dir %s (solenoidBz: %f, constBz %d, maxTimeBin %d)\n", eventsDir.c_str(), rec->GetGRPSettings().solenoidBzNominalGPU, (int32_t)rec->GetGRPSettings().constBz, rec->GetGRPSettings().grpContinuousMaxTimeBin);
const char* tmptext = configStandalone.noEvents ? "Using default event settings, no event dir loaded" : "Read event settings from dir ";
printf("%s%s (solenoidBz: %f, constBz %d, maxTimeBin %d)\n", tmptext, configStandalone.noEvents ? "" : eventsDir.c_str(), rec->GetGRPSettings().solenoidBzNominalGPU, (int32_t)rec->GetGRPSettings().constBz, rec->GetGRPSettings().grpContinuousMaxTimeBin);
if (configStandalone.testSyncAsync) {
recAsync->ReadSettings(eventsDir.c_str());
}
Expand Down Expand Up @@ -781,13 +782,17 @@ int32_t main(int argc, char** argv)

srand(configStandalone.seed);

for (nEventsInDirectory = 0; true; nEventsInDirectory++) {
std::ifstream in;
in.open((eventsDir + GPUCA_EVDUMP_FILE "." + std::to_string(nEventsInDirectory) + ".dump").c_str(), std::ifstream::binary);
if (in.fail()) {
break;
nEventsInDirectory = 0;
if (!configStandalone.noEvents) {
while (true) {
std::ifstream in;
in.open((eventsDir + GPUCA_EVDUMP_FILE "." + std::to_string(nEventsInDirectory) + ".dump").c_str(), std::ifstream::binary);
if (in.fail()) {
break;
}
in.close();
nEventsInDirectory++;
}
in.close();
}

if (configStandalone.TF.bunchSim || configStandalone.TF.nMerge) {
Expand Down Expand Up @@ -824,11 +829,7 @@ int32_t main(int argc, char** argv)
fflush(stdout);
for (int32_t i = 0; i < nEvents - configStandalone.StartEvent; i++) {
LoadEvent(configStandalone.StartEvent + i, i);
if (configStandalone.proc.debugLevel >= 2) {
printf("Loading event %d\n", i);
} else {
printf(" %d", i);
}
printf(configStandalone.proc.debugLevel >= 2 ? "Loading event %d\n" : " %d", i + configStandalone.StartEvent);
fflush(stdout);
}
printf("\n");
Expand Down Expand Up @@ -856,7 +857,7 @@ int32_t main(int argc, char** argv)
if (iEvent != configStandalone.StartEvent) {
printf("\n");
}
if (configStandalone.noEvents == false && !configStandalone.preloadEvents) {
if (!configStandalone.noEvents && !configStandalone.preloadEvents) {
HighResTimer timerLoad;
timerLoad.Start();
if (LoadEvent(iEvent, 0)) {
Expand Down Expand Up @@ -889,12 +890,14 @@ int32_t main(int argc, char** argv)
}
printf("Loading time: %'d us\n", (int32_t)(1000000 * timerLoad.GetCurrentElapsedTime()));
}
printf("Processing Event %d\n", iEvent);

nIteration.store(0);
nIterationEnd.store(0);
double pipelineWalltime = 1.;
if (configStandalone.proc.doublePipeline) {
if (configStandalone.noEvents) {
printf("No processing, no events loaded\n");
} else if (configStandalone.proc.doublePipeline) {
printf(configStandalone.preloadEvents ? "Processing Events %d to %d in Pipeline\n" : "Processing Event %d in Pipeline %d times\n", iEvent, configStandalone.preloadEvents ? std::min(iEvent + configStandalone.runs - 1, nEvents - 1) : configStandalone.runs);
HighResTimer timerPipeline;
if (configStandalone.proc.debugLevel < 2 && (RunBenchmark(rec, chainTracking, 1, iEvent, &nTracksTotal, &nClustersTotal) || RunBenchmark(recPipeline, chainTrackingPipeline, 2, iEvent, &nTracksTotal, &nClustersTotal))) {
goto breakrun;
Expand All @@ -907,6 +910,7 @@ int32_t main(int argc, char** argv)
pipelineWalltime = timerPipeline.GetElapsedTime() / (configStandalone.runs - 2);
printf("Pipeline wall time: %f, %d iterations, %f per event\n", timerPipeline.GetElapsedTime(), configStandalone.runs - 2, pipelineWalltime);
} else {
printf("Processing Event %d\n", iEvent);
if (RunBenchmark(rec, chainTracking, configStandalone.runs, iEvent, &nTracksTotal, &nClustersTotal)) {
goto breakrun;
}
Expand Down