Login

Alice Weekly Meeting: Software for Hardware Accelerators / PDP-SRC

→ Europe/Zurich
ALICE GPU Meeting
Zoom Meeting ID
61230224927
Host
David Rohr
Useful links
Join via phone
Zoom URL
    • 11:20 → 11:40
      Discussion 20m
      Speakers: David Rohr (CERN), Ole Schmidt (CERN)

      Color code: (critical, news during the meeting: green, news from this week: blue, news from last week: purple, no news: black)

      High priority Framework issues:

      • Fix dropping lifetime::timeframe for good: Still pending: problem with CCDB objects getting lost by DPL leading to "Dropping lifetime::timeframe", saw at least one occation during SW validation.
        • Checked with Ernst, we still get such errors at EOR, but these are false errors. Giulio has suppressed them in O2/dev. Need to wait for SW update at P2.
        • Giulio: Related problem with multi-output proxy - to be checked - Status?
      • Start / Stop / Start: 2 problems on O2 side left:
          • All processes are crashing randomly (usually ~2 out of >10k) when restarting. Stack trace hints to FMQ. https://its.cern.ch/jira/browse/O2-4639
          • TPC ITS matching QC crashing accessing CCDB objects. Not clear if same problem as above, or a problem in the task itself:
      • Stabilize calibration / fix EoS: New scheme: https://its.cern.ch/jira/browse/O2-4308: Still WiP.
      • Fix problem with ccdb-populator: no idea yet - since Ole left, someone else will have to take care.

       

      Other framework tickets:

      • Will from now on be followed up in serparate minutes..

       

      Global calibration topics:

      • TPC IDC and SAC workflow issues to be reevaluated with new O2 at restart of data taking. Cannot reproduce the problems any more.

       

      Sync reconstruction

      • Had a problem with COSMIC runs crashing due to bug in ITS code. Will create COSMIC REPLAY data sets as well, so this can be tested in staging.
      • Asked Ernst if he can from now on prepare SYNTHETIC and REPLAY data sets. He is looking into it.

       

      Async reconstruction

      • Remaining oscilation problem: GPUs get sometimes stalled for a long time up to 2 minutes. Checking 2 things:
        • does the situation get better without GPU monitoring? --> Inconclusive
        • We can use increased GPU processes priority as a mitigation, but doesn't fully fix the issue.
      • ḾI100 GPU stuck problem will only be addressed after AMD has fixed the operation with the latest official ROCm stack.
      • Limiting factor for pp workflow is now the TPC time series, which is to slow and creates backpressure (costs ~20% performance on EPNs). Enabled multi-threading as recommended by Matthias - need to check if it works.

       

      EPN major topics:

      • Fast movement of nodes between async / online without EPN expert intervention.
        • 2 goals I would like to set for the final solution:
          • It should not be needed to stop the SLURM schedulers when moving nodes, there should be no limitation for ongoing runs at P2 and ongoing async jobs.
          • We must not lose which nodes are marked as bad while moving.
      • Interface to change SHM memory sizes when no run is ongoing. Otherwise we cannot tune the workflow for both Pb-Pb and pp: https://alice.its.cern.ch/jira/browse/EPN-250
        • Lubos to provide interface to querry current EPN SHM settings - ETA July 2023, Status?
      • Improve DataDistribution file replay performance, currently cannot do faster than 0.8 Hz, cannot test MI100 EPN in Pb-Pb at nominal rate, and cannot test pp workflow for 100 EPNs in FST since DD injects TFs too slowly. https://alice.its.cern.ch/jira/browse/EPN-244 NO ETA
      • DataDistribution distributes data round-robin in absense of backpressure, but it would be better to do it based on buffer utilization, and give more data to MI100 nodes. Now, we are driving the MI50 nodes at 100% capacity with backpressure, and then only backpressured TFs go on MI100 nodes. This increases the memory pressure on the MI50 nodes, which is anyway a critical point. https://alice.its.cern.ch/jira/browse/EPN-397
      • TfBuilders should stop in ERROR when they lose connection.
      • Allow epn user and grid user to set nice level of processes: https://its.cern.ch/jira/browse/EPN-349
      • Improve core dumps: https://its.cern.ch/jira/browse/EPN-487
        • A GPU process crashed tonight, but I cannot find the core dump. I believe what happened is that ODC took down the collection killing all processes producing core dumps, and only the last dumps were kept.
        • Changes to suppress core dumps for SIGABRT seems to work, last incidept yielded core dumps. Waiting to understand the other minor issues before closing.
      • Tentative time for ALMA9 deployment: december 2024.
        • First test node with ALMA9 / ROCm 6.2 installed.
        • Idea is to make sure everything works on that node, create a separate build container to have new RPMs in parallel to test in staging, so we can easily switch.

       

      Other EPN topics:

       

      Full system test issues:

      Topology generation:

      • Should test to deploy topology with DPL driver, to have the remote GUI available.
        • DPL driver needs to implement FMQ state machine. Postponed until YETS issues solved.
      • 2 Occucances where the git repository in the topology cache was corrupted. Not really clear how this can happen, also not reproducible. Was solved by wiping the cache. Will add a check to the topology scripts to check for a corrupt repository, and in that case delete it and check it out anew.

       

      AliECS related topics:

      • Extra env var field still not multi-line by default.

       

      GPU ROCm / compiler topics:

      • Compilation failure due to missing symbols when compiling with -O0. Similar problem found by Matteo, being debugged.. Sent a reproducer to AMD.
      • Internal compiler error with LOG(...) macro: we have a workaround, AMD has a reproducer, waiting for a fix.
      • New miscompilation for ROCm 6.1
        • Found a workaround for us.
        • Provided a standalone reproducer to AMD.
        • Fix will certainly not make it into ROCm 6.2.1, but hopefully into 6.3, so that this won't hit us again.
      • ROCm 6.2
        • Was released last weekend, unfortunately a workaround, which we need was removed, while the fix didn't make it into the release, so we cannot use this release (or we'd need to port the workaround and compile the ROCm LLVM ourselves).
        • Waiting for 6.2.1, which will hopefully contain the fix.
      • EPNs prepared test nodes with ALMA9 / ROCm 6.2.
        • We cannot use them for O2 with GPU yet, but want to make sure everything builds, test new compilers, and Christian can check if that solves his ML framework issues.
      • Bumping GCC:
        • Bumping to GCC 13 needs undoing the Geant4 revert, bumping the GPU build container to new custom container with CUDA 12.6 and again custom ROCm compiler, and there is one missing problem not yet understood that ROCm fails to compile with GCC 13/14 headers on RHEL (while working on my laptop).
        • For GCC 14.2 in the future, needs also bumping ROOT, either fixing compile warnings on ONNXRuntime or bumping ONNXRuntime, bumning jason-c switching from autoconf to CMake, and waiting for new CUDA > 12.6 supporting GCC 14.
          • Made all this and disabled CUDA support in a test on the EPN, so Christian can use GCC 14 for ML clustering tests.
          • On my laptop hacked CUDA to use GCC 13 headers but compiling with GCC 14, which also works.
      • Bumping LLVM:
        • LVM bumped to 17, needed bumping arrow, adaptong CodeChecker, and fixing GCC includes (was incorrectly using the system GCC for includes instead of alidist GCC, which among others means clang could not compile C++14 code).
        • GCC header fix leads to a problem with relocation, Giulio / Sergio looking how to fix it.
        • Attempting to bump to LLVM 18, but this will require fixing a problem with arrow / O2Physics.
        • No need anymore to go to 18 for GPUs, required fixes were backported.

       

      TPC GPU Processing

      • Bug in TPC QC with MC embedding, TPC QC does not respect sourceID of MC labels, so confuses tracks of signal and of background events.
      • New problem with bogus values in TPC fast transformation map still pending. Sergey is investigating, but waiting for input from Alex. Ruben reported that he still sees such bogus values.
      • Status of cluster error parameterizations
        • No progress yet on newly requested debug streamers.
        • Waiting for TPC to check PR with full cluster errors during seeding.
      • TPC reported a problem with laser runs. In case of bad data (tpc data outside of the triggered drift time), GPUs sometimes can get stuck, so apparently the skipping of bad data is not fully working. Recorded some laser raw data to check. Started debugging yesterday, but ran into problem with slow EPN NFS. Meanwhile fixed by EPN.
      • Fixed bug that FMQ SHM message for occupancy map was allocated 4x too large (sizeof(int) vs sizeof(char) problem, and not 32hbf vs 128hbf confusion as originally anticipated).
      • Improved GPU / TPC configuration dump, no same syntax in standalone test and in O2, making it easily comparable.
      • Added a mode to the standalone benchmark to impose the default O2 settings, for better reproducibility.
      • Problem in GPU code since "char" on aarch64 linux is unsigned, not signed. Will probably move all of the GPU folder to using int8 ... uint64. For now, Giulio switched failing variables from "char" to "signed char"

      TPC processing performance regression:

      • Final solution: merging transformation maps on the fly into a single flat object:
        • Ruben has created a PR with a first version of the flat object. Still needs to be adapted when the map format will be changed. Also need to check it for GPU performance.
        • Discussion with Sergey in TPC SCD meeting today how to proceed.
      • Temporary mitigation with RTC:
        • Still stuck due to GPU RTC problems.

       

      General GPU Processing

      • Pending problems with using GPU RTC at P2:
        • /tmp is inside the slurm container and wiped afterwards. Fixed by using /var/tmp.
        • RTC is started from one of the GPU processes, which has a NUMA pinning to one NUMA domain, thus it uses only half of the CPU cores. Need to extend the CPU pinning for the compilation subprocesses.
        • RTC compiles for the architecturs of the original build, which is currently MI50/MI100, i.e. all nodes compile twice, which takes extra time. Need to add an option to select an architecture, and the topology generation must put in the setting for MI50 / MI100 architectures.
        • AMD compiler leaves stale temp folders, can avoid this by setting $TMPDIR.
        • RTC compilation fails in an online run since headers (e.g. <cmath>) are not found - Not understood yet.

       

      Oncalls:

      • All weeks covered this year.
    • 11:40 → 11:45
      Following up JIRA tickets 5m
      Speaker: Ernst Hellbar (CERN)

      Overview of low-priority framework issues

    • 11:45 → 11:50
      TPC ML Clustering 5m
      Speaker: Christian Sonnabend (CERN, Heidelberg University (DE))

      MIGraphX

      • Tested in O2: fails to aquire gpu as target device for model compilation. Model is loaded successfully on CPU though
      • Tested standalone MIGraphX compilation (with alienv enter GCC-Toolchain/latest for gcc14) and here it succeded

       

      ONNXRuntime

      • Compilation works now
      • Need to figure out how to set Ort::MemoryInfo correctly to target GPU memory

       

      Data for the neural network

      • Created training data for the NN from real data using native clusters and track paths
      • Need to adjust data sampler but will submit jobs again later today
    • 11:50 → 11:55
      ITS Tracking 5m
      Speaker: Matteo Concas (CERN)
      • SMatrix GPU implementation testing: follow-up to Maksym's work for the single matrix
        • Tests now focus on SMatrix<double, N=3>, which is what is used on DCA fitter. The new tolerance is 1e-8.
        • Added benchmark for timing.
        • Extended also to 1e6 simultaneous inversions.
      • Porting of the DCAFitterN class on GPU: https://github.com/AliceO2Group/AliceO2/pull/13365
        • Applied some of the usual massage to make the DCAFitterN.h code ROOT and STD library agnostic.

        • This does not compile on GPU. Notably, there are currently three main issues:

          • When I try to create a library with, e.g. nvcc, it does not realise GPUCA_GPUCODE_DEVICE should be defined in the device code (see DCAFitterN.cu).

          • Equivalent GPU-portable versions of ROOT::Math::Dot and ROOT::Math::Similarity functions are already present in the SMatrixGPU.h file but are not yet connected to the aliasing in Cartesian.h.
          • I am probably missing some detail in the usage of the forward-declared classes in GPUROOTSMatrixFwd.h since I end up having incomplete types when I try to compile any o2::GPU::gpustd::array<SVectorGPU<float, 3>>.
      • ITS GPU tracking resumed: https://github.com/AliceO2Group/AliceO2/pull/13383/files
        • HIP tracking results diverge now, even in deterministic mode, for no apparent reason investigating.
        • CUDA~CPU are still consistent in deterministic mode
    • 11:55 → 12:15
      TPC Track Model Decoding on GPU 20m
      Speaker: Gabriele Cimador (Universita e INFN Trieste (IT))
Powered by Indico v3.3.10-pre