NN clusterizer: Improve filling kernel speed by ChSonnabend · Pull Request #14510 · AliceO2Group/AliceO2

ChSonnabend · 2025-07-13T23:18:16Z

Improves GPU kernel speed by ~30%

github-actions · 2025-07-13T23:18:24Z

REQUEST FOR PRODUCTION RELEASES:
To request your PR to be included in production software, please add the corresponding labels called "async-" to your PR. Add the labels directly (if you have the permissions) or add a comment of the form (note that labels are separated by a ",")

+async-label <label1>, <label2>, !<label3> ...

This will add <label1> and <label2> and removes <label3>.

The following labels are available
async-2023-pbpb-apass4
async-2023-pp-apass4
async-2024-pp-apass1
async-2022-pp-apass7
async-2024-pp-cpass0
async-2024-PbPb-apass1
async-2024-ppRef-apass1
async-2024-PbPb-apass2
async-2023-PbPb-apass5

Please consider the following formatting changes to AliceO2Group#14510

alibuild · 2025-07-14T00:40:25Z

Error while checking build/O2/fullCI_slc9 for 560f77f at 2025-07-14 02:40:

## sw/BUILD/e634489fc902e8cb213d4d883e791c07a9571dc8-latest/log
/sw/SOURCES/O2/14510-slc9_x86-64/0/GPU/GPUTracking/DataTypes/GPUTPCGeometry.h:114:98: error: array subscript 152 is above array bounds of 'const uint8_t [152]' {aka 'const unsigned char [152]'} [-Werror=array-bounds=]
make[2]: *** [GPU/GPUTracking/CMakeFiles/GPUTracking.dir/build.make:1161: GPU/GPUTracking/CMakeFiles/GPUTracking.dir/TPCClusterFinder/GPUTPCNNClusterizerHost.cxx.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [CMakeFiles/Makefile2:606: GPU/GPUTracking/CMakeFiles/GPUTracking.dir/all] Error 2
make: *** [Makefile:136: all] Error 2


## sw/BUILD/O2-latest/log
/sw/SOURCES/O2/14510-slc9_x86-64/0/GPU/GPUTracking/DataTypes/GPUTPCGeometry.h:114:98: error: array subscript 152 is above array bounds of 'const uint8_t [152]' {aka 'const unsigned char [152]'} [-Werror=array-bounds=]
make[2]: *** [GPU/GPUTracking/CMakeFiles/GPUTracking.dir/build.make:1161: GPU/GPUTracking/CMakeFiles/GPUTracking.dir/TPCClusterFinder/GPUTPCNNClusterizerHost.cxx.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [CMakeFiles/Makefile2:606: GPU/GPUTracking/CMakeFiles/GPUTracking.dir/all] Error 2
make: *** [Makefile:136: all] Error 2

Full log here.

ChSonnabend · 2025-07-14T06:47:23Z

@drohr, Build error seems unrelated and not sure what exactly it would fail at... The maximum value of uint8 would be 255 if I am not mistaking

davidrohr · 2025-07-14T06:58:14Z

I don't think the error is unrelated. It tells you that you are accessing gputpcgeometry_internal::mNPads[152] which is out of bounds.

ChSonnabend · 2025-07-14T07:17:35Z

I don't think the error is unrelated. It tells you that you are accessing gputpcgeometry_internal::mNPads[152] which is out of bounds.

Indeed, sorry for the noise. Not sure how I didn't spot that.

Please consider the following formatting changes to AliceO2Group#14510

…r coallesced access

… for-loop over row dimension as access is somewhat coalsced too

Please consider the following formatting changes to AliceO2Group#14510

ChSonnabend · 2025-07-16T15:52:35Z

These are the best improvements I could find so far. @drohr can merge for now once the CI is green, please?

alibuild · 2025-07-16T15:54:28Z

Error while checking build/O2/fullCI_slc9 for 069a7e9 at 2025-07-17 16:09:

## sw/BUILD/O2-latest/log
/sw/SOURCES/O2/14510-slc9_x86-64/0/GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizerKernels.cxx:158:74: error: comparison of integer expressions of different signedness: 'int32_t' {aka 'int'} and 'uint32_t' {aka 'unsigned int'} [-Werror=sign-compare]
/sw/SOURCES/O2/14510-slc9_x86-64/0/GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizerKernels.cxx:188:32: error: comparison of integer expressions of different signedness: 'int32_t' {aka 'int'} and 'uint32_t' {aka 'unsigned int'} [-Werror=sign-compare]
make[2]: *** [GPU/GPUTracking/CMakeFiles/GPUTracking.dir/build.make:1133: GPU/GPUTracking/CMakeFiles/GPUTracking.dir/TPCClusterFinder/GPUTPCNNClusterizerKernels.cxx.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [CMakeFiles/Makefile2:606: GPU/GPUTracking/CMakeFiles/GPUTracking.dir/all] Error 2
make: *** [Makefile:136: all] Error 2

Full log here.

drohr · 2025-07-16T16:13:01Z

These are the best improvements I could find so far. @drohr can merge for now once the CI is green, please?

Please ping @davidrohr and not @drohr over and over. We might share the same nice name, but we are not the same person :-)

ChSonnabend · 2025-07-16T16:20:39Z

@davidrohr Ready to merge once the CI is green. Not sure what is currently going on with the build container, but it builds locally

alibuild · 2025-07-18T00:58:14Z

Error while checking build/O2/fullCI_slc9 for 4949b55 at 2025-07-18 02:58:

## sw/BUILD/O2-latest/log
/sw/SOURCES/O2/14510-slc9_x86-64/0/GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizerKernels.cxx:68:67: error: no matching function for call to 'o2::gpu::GPUCommonMath::Min(uint32_t, o2::gpu::tpccf::SizeT)'
ninja: build stopped: subcommand failed.

Full log here.

alibuild · 2025-07-18T08:26:40Z

Error while checking build/O2/fullCI_slc9 for f4729fd at 2025-07-18 10:26:

## sw/BUILD/O2-latest/log
/sw/SOURCES/O2/14510-slc9_x86-64/0/GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizerKernels.cxx:176:74: error: comparison of integer expressions of different signedness: 'int32_t' {aka 'int'} and 'uint32_t' {aka 'unsigned int'} [-Werror=sign-compare]
/sw/SOURCES/O2/14510-slc9_x86-64/0/GPU/GPUTracking/TPCClusterFinder/GPUTPCNNClusterizerKernels.cxx:206:32: error: comparison of integer expressions of different signedness: 'int32_t' {aka 'int'} and 'uint32_t' {aka 'unsigned int'} [-Werror=sign-compare]
make[2]: *** [GPU/GPUTracking/CMakeFiles/GPUTracking.dir/build.make:1133: GPU/GPUTracking/CMakeFiles/GPUTracking.dir/TPCClusterFinder/GPUTPCNNClusterizerKernels.cxx.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [CMakeFiles/Makefile2:606: GPU/GPUTracking/CMakeFiles/GPUTracking.dir/all] Error 2
make: *** [Makefile:136: all] Error 2

Full log here.

* First version of lookup tables * Simplifying computations + bug-fixes * Fixes for indexing and offsets * Adjusting CPU kernel * Please consider the following formatting changes * Fix for row-number access * Please consider the following formatting changes * Improve kernel speed by ~15%. Next test: for-loop in pad direction for coallesced access * IMproving kernel speed by 30% compared to original version. Next try: for-loop over row dimension as access is somewhat coalsced too * Please consider the following formatting changes * Minor improvements for MC handling * Beautifications to trigger the CI * Compile-fix * Fix int32_t error in fullCI build --------- Co-authored-by: ALICE Action Bot <alibuild@cern.ch>

ChSonnabend added 4 commits July 12, 2025 03:22

First version of lookup tables

290ba0e

Simplifying computations + bug-fixes

ed459ea

Fixes for indexing and offsets

1ac8c2c

Adjusting CPU kernel

fdf6ef8

ChSonnabend requested a review from davidrohr as a code owner July 13, 2025 23:18

Please consider the following formatting changes

7604fab

alibuild mentioned this pull request Jul 13, 2025

Please consider the following formatting changes to #14510 ChSonnabend/AliceO2#27

Merged

Merge pull request #27 from alibuild/alibot-cleanup-14510

560f77f

Please consider the following formatting changes to AliceO2Group#14510

Fix for row-number access

0c2ac6a

Please consider the following formatting changes

33c8278

alibuild mentioned this pull request Jul 14, 2025

Please consider the following formatting changes to #14510 ChSonnabend/AliceO2#28

Merged

Merge pull request #28 from alibuild/alibot-cleanup-14510

9a7fa44

Please consider the following formatting changes to AliceO2Group#14510

ChSonnabend marked this pull request as draft July 14, 2025 07:56

Improve kernel speed by ~15%. Next test: for-loop in pad direction fo…

eecb8a8

…r coallesced access

alibuild mentioned this pull request Jul 15, 2025

Please consider the following formatting changes to #14510 ChSonnabend/AliceO2#29

Merged

ChSonnabend and others added 4 commits July 16, 2025 09:58

IMproving kernel speed by 30% compared to original version. Next try:…

3447927

… for-loop over row dimension as access is somewhat coalsced too

Please consider the following formatting changes

6ef38a2

Merge pull request #29 from alibuild/alibot-cleanup-14510

24b1d04

Please consider the following formatting changes to AliceO2Group#14510

Minor improvements for MC handling

069a7e9

ChSonnabend changed the title ~~NN clusterizer: Change to Lookup table~~ NN clusterizer: Improve filling kernel speed Jul 16, 2025

ChSonnabend marked this pull request as ready for review July 16, 2025 15:51

Beautifications to trigger the CI

4949b55

Compile-fix

f4729fd

Fix int32_t error in fullCI build

2af6060

davidrohr merged commit da3a178 into AliceO2Group:dev Jul 18, 2025
11 checks passed

ChSonnabend deleted the nn_clusterizer_lookup_boundary branch July 18, 2025 21:49

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

NN clusterizer: Improve filling kernel speed#14510

NN clusterizer: Improve filling kernel speed#14510
davidrohr merged 17 commits intoAliceO2Group:devfrom
ChSonnabend:nn_clusterizer_lookup_boundary

ChSonnabend commented Jul 13, 2025

Uh oh!

github-actions bot commented Jul 13, 2025

Uh oh!

alibuild commented Jul 14, 2025

Uh oh!

ChSonnabend commented Jul 14, 2025

Uh oh!

davidrohr commented Jul 14, 2025

Uh oh!

ChSonnabend commented Jul 14, 2025

Uh oh!

ChSonnabend commented Jul 16, 2025

Uh oh!

alibuild commented Jul 16, 2025 •

edited

Loading

Uh oh!

drohr commented Jul 16, 2025

Uh oh!

ChSonnabend commented Jul 16, 2025

Uh oh!

alibuild commented Jul 18, 2025

Uh oh!

alibuild commented Jul 18, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

4 participants

Conversation

ChSonnabend commented Jul 13, 2025

Uh oh!

github-actions bot commented Jul 13, 2025

Uh oh!

alibuild commented Jul 14, 2025

Uh oh!

ChSonnabend commented Jul 14, 2025

Uh oh!

davidrohr commented Jul 14, 2025

Uh oh!

ChSonnabend commented Jul 14, 2025

Uh oh!

ChSonnabend commented Jul 16, 2025

Uh oh!

alibuild commented Jul 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

drohr commented Jul 16, 2025

Uh oh!

ChSonnabend commented Jul 16, 2025

Uh oh!

alibuild commented Jul 18, 2025

Uh oh!

alibuild commented Jul 18, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

4 participants

alibuild commented Jul 16, 2025 •

edited

Loading