Skip to content

NN clusterizer: Improving speed of NN input-filling kernel #14749

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 6 commits into from
Oct 27, 2025
Merged

NN clusterizer: Improving speed of NN input-filling kernel #14749

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
9c8984c
Improve GPU filling kernel speed
ChSonnabend Oct 17, 2025
a075c43
Adjusting parameter bounds and additional GPU kernel optimizations
ChSonnabend Oct 18, 2025
587c3e6
Adding back if statement for early exit
ChSonnabend Oct 18, 2025
6e43257
const'ing + fixing CPU kernel
ChSonnabend Oct 18, 2025
bb795c4
Remiving print statements
ChSonnabend Oct 19, 2025
f7cdc0b
Fixing CI build issue
ChSonnabend Oct 27, 2025
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/Definitions/GPUDefParametersDefaults.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -482,7 +482,7 @@

#define GPUCA_LB_GPUTPCNNClusterizerKernels_runCfClusterizer GPUCA_LB_GPUTPCNNClusterizerKernels
#define GPUCA_LB_GPUTPCNNClusterizerKernels_fillInputNNCPU GPUCA_LB_GPUTPCNNClusterizerKernels
#define GPUCA_LB_GPUTPCNNClusterizerKernels_fillInputNNGPU GPUCA_LB_GPUTPCNNClusterizerKernels
#define GPUCA_LB_GPUTPCNNClusterizerKernels_fillInputNNGPU 1024
#define GPUCA_LB_GPUTPCNNClusterizerKernels_determineClass1Labels GPUCA_LB_GPUTPCNNClusterizerKernels
#define GPUCA_LB_GPUTPCNNClusterizerKernels_determineClass2Labels GPUCA_LB_GPUTPCNNClusterizerKernels
#define GPUCA_LB_GPUTPCNNClusterizerKernels_publishClass1Regression GPUCA_LB_GPUTPCNNClusterizerKernels
Expand Down
2 changes: 1 addition & 1 deletion GPU/GPUTracking/Global/GPUChainTrackingClusterizer.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1035,7 +1035,7 @@ int32_t GPUChainTracking::RunTPCClusterizer(bool synchronizeOutput)
// Filling the data
if (mRec->IsGPU() || GetProcessingSettings().nn.nnClusterizerForceGpuInputFill) {
// Fills element by element of each input matrix -> better parallelizability, but worse on CPU due to unnecessary computations
runKernel<GPUTPCNNClusterizerKernels, GPUTPCNNClusterizerKernels::fillInputNNGPU>({GetGrid(iSize * clustererNNShadow.mNnClusterizerRowTimeSizeFull, lane), krnlRunRangeNone}, iSector, clustererNNShadow.mNnInferenceInputDType, propagateMCLabels, batchStart);
runKernel<GPUTPCNNClusterizerKernels, GPUTPCNNClusterizerKernels::fillInputNNGPU>({GetGrid(iSize * clustererNNShadow.mNnClusterizerRowTimeSizeThreads , lane), krnlRunRangeNone}, iSector, clustererNNShadow.mNnInferenceInputDType, propagateMCLabels, batchStart);
} else {
// Fills the whole input matrix at once -> better performance on CPU, but worse parallelizability
runKernel<GPUTPCNNClusterizerKernels, GPUTPCNNClusterizerKernels::fillInputNNCPU>({GetGrid(iSize, lane), krnlRunRangeNone}, iSector, clustererNNShadow.mNnInferenceInputDType, propagateMCLabels, batchStart);
Expand Down
1 change: 1 addition & 0 deletions GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizer.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,7 @@ class GPUTPCNNClusterizer : public GPUProcessor
uint32_t mNnClusterizerPadTimeSize = 0;
uint32_t mNnClusterizerRowTimeSize = 0;
uint32_t mNnClusterizerRowTimeSizeFull = 0;
uint32_t mNnClusterizerRowTimeSizeThreads = 0;

// Boundary lookup table
// int32_t mBoundaryMapSizeRow = 0;
Expand Down
1 change: 1 addition & 0 deletions GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizerHost.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -98,6 +98,7 @@ void GPUTPCNNClusterizerHost::initClusterizer(const GPUSettingsProcessingNNclust
clustererNN.mNnClusterizerPadTimeSize = clustererNN.mNnClusterizerFullPadSize * clustererNN.mNnClusterizerFullTimeSize;
clustererNN.mNnClusterizerRowTimeSize = clustererNN.mNnClusterizerFullRowSize * clustererNN.mNnClusterizerFullTimeSize;
clustererNN.mNnClusterizerRowTimeSizeFull = clustererNN.mNnClusterizerRowTimeSize + (settings.nnClusterizerAddIndexData ? 3 : 0);
clustererNN.mNnClusterizerRowTimeSizeThreads = clustererNN.mNnClusterizerRowTimeSize + (settings.nnClusterizerAddIndexData ? 1 : 0);
clustererNN.mNnClusterizerElementSize = clustererNN.mNnClusterizerChargeArraySize + (settings.nnClusterizerAddIndexData ? 3 : 0);
// clustererNN.mBoundaryMapSizeRow = 3 * clustererNN.mNnClusterizerSizeInputRow + o2::tpc::constants::MAXGLOBALPADROW;
// clustererNN.mBoundaryPadding = 11; // padding on each side to account for pad_offset. N=11 since then mIsBoundary = 24320 ~< (1.5 x 2^14 = 24576) && N must be bigger than (NPads[row(end_iroc + 1)] - NPads[row(end_iroc)])/2 (=6) for pad_offset to work
Expand Down
Loading