[GCC13] Missing intrinsics 2 #44645

iarspider · 2024-04-06T11:33:42Z

When testing fixes for gcc13, two more cases of identifier "__builtin_ia32_cmpccxadd" is undefined were detected:

>> Compiling alpaka/cuda src/RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.dev.cc
/cvmfs/cms-ib.cern.ch/sw/x86_64/nweek-02831/el9_amd64_gcc13/external/gcc/13.2.0-1b0a3367d04f48f01ad3ccf40e55475c/lib/gcc/x86_64-redhat-linux-gnu/13.2.0/include/cmpccxaddintrin.h(63): error: identifier "__builtin_ia32_cmpccxadd" is undefined
    return __builtin_ia32_cmpccxadd (__A, __B, __C, __D);
           ^

/cvmfs/cms-ib.cern.ch/sw/x86_64/nweek-02831/el9_amd64_gcc13/external/gcc/13.2.0-1b0a3367d04f48f01ad3ccf40e55475c/lib/gcc/x86_64-redhat-linux-gnu/13.2.0/include/cmpccxaddintrin.h(71): error: identifier "__builtin_ia32_cmpccxadd64" is undefined
    return __builtin_ia32_cmpccxadd64 (__A, __B, __C, __D);
           ^

2 errors detected in the compilation of "src/RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.dev.cc".

and

>> Compiling alpaka/cuda edm plugin src/RecoParticleFlow/PFRecHitProducer/plugins/alpaka/PFRecHitTopologyESProducer.cc
/cvmfs/cms-ib.cern.ch/sw/x86_64/nweek-02831/el9_amd64_gcc13/external/gcc/13.2.0-1b0a3367d04f48f01ad3ccf40e55475c/lib/gcc/x86_64-redhat-linux-gnu/13.2.0/include/cmpccxaddintrin.h(63): error: identifier "__builtin_ia32_cmpccxadd" is undefined
    return __builtin_ia32_cmpccxadd (__A, __B, __C, __D);
           ^

/cvmfs/cms-ib.cern.ch/sw/x86_64/nweek-02831/el9_amd64_gcc13/external/gcc/13.2.0-1b0a3367d04f48f01ad3ccf40e55475c/lib/gcc/x86_64-redhat-linux-gnu/13.2.0/include/cmpccxaddintrin.h(71): error: identifier "__builtin_ia32_cmpccxadd64" is undefined
    return __builtin_ia32_cmpccxadd64 (__A, __B, __C, __D);
           ^

2 errors detected in the compilation of "src/RecoParticleFlow/PFRecHitProducer/plugins/alpaka/PFRecHitProducerKernel.dev.cc".

Similar problem was fixed in #44595 by @makortel and @dan131riley .

The text was updated successfully, but these errors were encountered:

iarspider · 2024-04-06T11:33:46Z

assign heterogeneous,core

cmsbuild · 2024-04-06T11:33:59Z

New categories assigned: heterogeneous,core

@Dr15Jones,@fwyzard,@makortel,@makortel,@smuzaffar you have been requested to review this Pull request/Issue and eventually sign? Thanks

cmsbuild · 2024-04-06T11:34:01Z

cms-bot internal usage

cmsbuild · 2024-04-06T11:34:01Z

A new Issue was created by @iarspider.

@sextonkennedy, @smuzaffar, @Dr15Jones, @rappoccio, @makortel, @antoniovilela can you please review it and eventually sign/assign? Thanks.

cms-bot commands are listed here

fwyzard · 2024-04-08T11:47:42Z

According to the release notes, __builtin_ia32_cmpccxadd and __builtin_ia32_cmpccxadd64 should now be supported by CUDA 12.4.1 .

fwyzard · 2024-04-08T11:53:18Z

See cms-sw/cmsdist#9126 .

iarspider · 2024-04-08T11:56:23Z

@fwyzard thanks!

In case that doesn't fix the issue, here is include chain that leads to tbb:

RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.dev.cc
RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.h
RecoParticleFlow/PFRecHitProducer/interface/PFRecHitTopologyRecord.h
Geometry/Records/interface/CaloGeometryRecord.h
FWCore/Framework/interface/EventSetupRecordImplementation.h
FWCore/Framework/interface/EventSetupRecord.h
FWCore/Framework/interface/EventSetupRecordImpl.h
FWCore/Concurrency/interface/WaitingTaskHolder.h
oneapi/tbb/task_group.h

fwyzard · 2024-04-08T12:30:43Z

thanks @iarspider, we should probably try to fix the include chain anyway.

fwyzard · 2024-04-08T12:43:47Z

Moving

#include "RecoParticleFlow/PFRecHitProducer/interface/PFRecHitTopologyRecord.h"

from RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.h to RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducer.cc should fix the issue.

iarspider · 2024-04-08T12:44:50Z

Moving
#include "RecoParticleFlow/PFRecHitProducer/interface/PFRecHitTopologyRecord.h"
from RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.h to RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducer.cc should fix the issue.

Thanks! Could you please open a PR with this change?

fwyzard · 2024-04-08T12:48:19Z

#44656

iarspider · 2024-04-09T09:04:04Z

@fwyzard looks like there are more cases like this: 1, 2. I will check the include chains for these two.

fwyzard · 2024-04-09T09:15:58Z

@makortel , if we want to avoid compiling (some of) the framework headers with nvcc, would it make sense to add something like

#ifdef __NVCC__
#error Framework headers should not be compiled by the device compiler
#endif

to FWCore/Concurrency/interface/WaitingTaskHolder.h and similar headers ?

iarspider · 2024-04-09T09:29:06Z

For RecoParticleFlow/PFRecHitProducer/plugins/alpaka/PFRecHitTopologyESProducer.cc: list - a bit long this time, since I've traced all paths (but can also be not 100% accurate if some includes are guarded with #ifdef).

iarspider · 2024-04-09T09:34:42Z

For RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu:

RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu -> RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.h -> CalibFormats/SiStripObjects/interface/SiStripClusterizerConditionsGPU.h -> HeterogeneousCore/CUDACore/interface/ESProduct.h -> HeterogeneousCore/CUDAUtilities/interface/EventCache.h -> FWCore/Utilities/interface/ReusableObjectHolder.h -> oneapi/tbb/concurrent_queue.h
RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu -> CalibFormats/SiStripObjects/interface/SiStripClusterizerConditionsGPU.h -> HeterogeneousCore/CUDACore/interface/ESProduct.h -> HeterogeneousCore/CUDAUtilities/interface/EventCache.h -> FWCore/Utilities/interface/ReusableObjectHolder.h -> oneapi/tbb/concurrent_queue.h

(this is the only tbb header used by this file, the PFRecHitTopologyESProducer.cc referenced several other tbb headers)

fwyzard · 2024-04-09T10:09:41Z

For RecoParticleFlow/PFRecHitProducer/plugins/alpaka/PFRecHitTopologyESProducer.cc: list - a bit long this time, since I've traced all paths (but can also be not 100% accurate if some includes are guarded with #ifdef).

RecoParticleFlow/PFRecHitProducer/plugins/alpaka/PFRecHitTopologyESProducer.cc should not be compiled by nvcc.

fwyzard · 2024-04-09T10:10:06Z

For RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu:

RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu -> RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.h -> CalibFormats/SiStripObjects/interface/SiStripClusterizerConditionsGPU.h -> HeterogeneousCore/CUDACore/interface/ESProduct.h -> HeterogeneousCore/CUDAUtilities/interface/EventCache.h -> FWCore/Utilities/interface/ReusableObjectHolder.h -> oneapi/tbb/concurrent_queue.h
RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu -> CalibFormats/SiStripObjects/interface/SiStripClusterizerConditionsGPU.h -> HeterogeneousCore/CUDACore/interface/ESProduct.h -> HeterogeneousCore/CUDAUtilities/interface/EventCache.h -> FWCore/Utilities/interface/ReusableObjectHolder.h -> oneapi/tbb/concurrent_queue.h

(this is the only tbb header used by this file, the PFRecHitTopologyESProducer.cc referenced several other tbb headers)

Unfortunately this does not seem trivial to disentangle.

iarspider · 2024-04-09T10:26:59Z

I have just realized that I have processed wrong file, sorry.

…

On Tue, Apr 9, 2024, 12:10 Andrea Bocci ***@***.***> wrote: For RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu : RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu -> RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.h -> CalibFormats/SiStripObjects/interface/SiStripClusterizerConditionsGPU.h -> HeterogeneousCore/CUDACore/interface/ESProduct.h -> HeterogeneousCore/CUDAUtilities/interface/EventCache.h -> FWCore/Utilities/interface/ReusableObjectHolder.h -> oneapi/tbb/concurrent_queue.h RecoLocalTracker/SiStripClusterizer/plugins/SiStripRawToClusterGPUKernel.cu -> CalibFormats/SiStripObjects/interface/SiStripClusterizerConditionsGPU.h -> HeterogeneousCore/CUDACore/interface/ESProduct.h -> HeterogeneousCore/CUDAUtilities/interface/EventCache.h -> FWCore/Utilities/interface/ReusableObjectHolder.h -> oneapi/tbb/concurrent_queue.h (this is the only tbb header used by this file, the PFRecHitTopologyESProducer.cc referenced several other tbb headers) Unfortunately this does not seem trivial to disentangle. — Reply to this email directly, view it on GitHub <#44645 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAE3NORURKT667HQB5PWYK3Y4O5BJAVCNFSM6AAAAABF2L43OWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANBUGYZTOOJYG4> . You are receiving this because you were mentioned.Message ID: ***@***.***>

iarspider · 2024-04-09T11:12:34Z

Corrected dependency chain for PFRecHitProducerKernel.dev.cc: https://gist.github.com/iarspider/f655770228e41eebbf38618f836870bd . All chains stem from RecoParticleFlow/PFRecHitProducer/interface/PFRecHitTopologyRecord.h and RecoParticleFlow/PFRecHitProducer/interface/PFRecHitParamsRecord.h

fwyzard · 2024-04-09T16:00:22Z

Can you try to edit RecoParticleFlow/PFRecHitProducer/plugins/alpaka/CalorimeterDefinitions.h and replace the include of the differente Records with forward declarations ?

dan131riley · 2024-04-22T13:13:31Z

@makortel , if we want to avoid compiling (some of) the framework headers with nvcc, would it make sense to add something like
#ifdef __NVCC__
#error Framework headers should not be compiled by the device compiler
#endif
to FWCore/Concurrency/interface/WaitingTaskHolder.h and similar headers ?

Trying this in 14_1_0_pre2, I get hits in

                 from src/RecoLocalCalo/HcalRecProducers/src/MahiGPU.cu:11:
                 from src/RecoParticleFlow/PFRecHitProducer/plugins/alpaka/PFRecHitProducerKernel.dev.cc:8:
                 from src/RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.dev.cc:9:

Full chains:

                 from /mnt/data1/dsr/tmp/CMSSW_14_1_0_pre2/src/FWCore/Framework/interface/DependentRecordImplementation.h:26,
                 from /mnt/data1/dsr/tmp/CMSSW_14_1_0_pre2/src/CondFormats/DataRecord/interface/HcalPedestalWidthsRcd.h:5,
                 from /mnt/data1/dsr/tmp/CMSSW_14_1_0_pre2/src/CondFormats/DataRecord/interface/HcalCombinedRecordsGPU.h:4,
                 from src/RecoLocalCalo/HcalRecProducers/src/DeclsForKernels.h:11,
                 from src/RecoLocalCalo/HcalRecProducers/src/SimpleAlgoGPU.h:4,
                 from src/RecoLocalCalo/HcalRecProducers/src/MahiGPU.cu:11:

                 from /mnt/data1/dsr/tmp/CMSSW_14_1_0_pre2/src/RecoParticleFlow/PFRecHitProducer/interface/PFRecHitParamsRecord.h:4,
                 from src/RecoParticleFlow/PFRecHitProducer/plugins/alpaka/CalorimeterDefinitions.h:12,
                 from src/RecoParticleFlow/PFRecHitProducer/plugins/alpaka/PFRecHitProducerKernel.h:6,
                 from src/RecoParticleFlow/PFRecHitProducer/plugins/alpaka/PFRecHitProducerKernel.dev.cc:8:

                 from /mnt/data1/dsr/tmp/CMSSW_14_1_0_pre2/src/FWCore/Framework/interface/DependentRecordImplementation.h:26,
                 from /mnt/data1/dsr/tmp/CMSSW_14_1_0_pre2/src/RecoParticleFlow/PFRecHitProducer/interface/PFRecHitTopologyRecord.h:4,
                 from /mnt/data1/dsr/tmp/CMSSW_14_1_0_pre2/src/RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.h:11,
                 from src/RecoParticleFlow/PFClusterProducer/plugins/alpaka/PFClusterSoAProducerKernel.dev.cc:9:

fwyzard · 2024-04-24T07:41:29Z

either #44682 or #44828 should take care of the alpaka ParticleFlow ones.

iarspider · 2024-05-03T09:59:40Z

@cmsbuild please close

cmsbuild added core-pending pending-signatures heterogeneous-pending labels Apr 6, 2024

fwyzard mentioned this issue Apr 8, 2024

Move EventSetup record dependency to non-device source file #44656

Merged

iarspider mentioned this issue Apr 10, 2024

Do not include EventSetup record definitions in device code #44682

Merged

fwyzard mentioned this issue Apr 29, 2024

Do not include EventSetup record definitions in device code [14.0.x] #44871

Merged

cmsbuild closed this as completed May 3, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[GCC13] Missing intrinsics 2 #44645

[GCC13] Missing intrinsics 2 #44645

iarspider commented Apr 6, 2024

iarspider commented Apr 6, 2024

cmsbuild commented Apr 6, 2024

cmsbuild commented Apr 6, 2024 •

edited

Loading

cmsbuild commented Apr 6, 2024

fwyzard commented Apr 8, 2024

fwyzard commented Apr 8, 2024

iarspider commented Apr 8, 2024

fwyzard commented Apr 8, 2024

fwyzard commented Apr 8, 2024

iarspider commented Apr 8, 2024

fwyzard commented Apr 8, 2024

iarspider commented Apr 9, 2024 •

edited

Loading

fwyzard commented Apr 9, 2024

iarspider commented Apr 9, 2024 •

edited

Loading

iarspider commented Apr 9, 2024 •

edited

Loading

fwyzard commented Apr 9, 2024

fwyzard commented Apr 9, 2024

iarspider commented Apr 9, 2024 via email

iarspider commented Apr 9, 2024

fwyzard commented Apr 9, 2024

dan131riley commented Apr 22, 2024

fwyzard commented Apr 24, 2024 •

edited

Loading

iarspider commented May 3, 2024

[GCC13] Missing intrinsics 2 #44645

[GCC13] Missing intrinsics 2 #44645

Comments

iarspider commented Apr 6, 2024

iarspider commented Apr 6, 2024

cmsbuild commented Apr 6, 2024

cmsbuild commented Apr 6, 2024 • edited Loading

cmsbuild commented Apr 6, 2024

fwyzard commented Apr 8, 2024

fwyzard commented Apr 8, 2024

iarspider commented Apr 8, 2024

fwyzard commented Apr 8, 2024

fwyzard commented Apr 8, 2024

iarspider commented Apr 8, 2024

fwyzard commented Apr 8, 2024

iarspider commented Apr 9, 2024 • edited Loading

fwyzard commented Apr 9, 2024

iarspider commented Apr 9, 2024 • edited Loading

iarspider commented Apr 9, 2024 • edited Loading

fwyzard commented Apr 9, 2024

fwyzard commented Apr 9, 2024

iarspider commented Apr 9, 2024 via email

iarspider commented Apr 9, 2024

fwyzard commented Apr 9, 2024

dan131riley commented Apr 22, 2024

fwyzard commented Apr 24, 2024 • edited Loading

iarspider commented May 3, 2024

cmsbuild commented Apr 6, 2024 •

edited

Loading

iarspider commented Apr 9, 2024 •

edited

Loading

iarspider commented Apr 9, 2024 •

edited

Loading

iarspider commented Apr 9, 2024 •

edited

Loading

fwyzard commented Apr 24, 2024 •

edited

Loading