[runtime_cxxmodules] Enable on AArch64

[hahnjo]

It was disabled in commit a67863d33a ("Disable modules on aarch64 due to ODR violation") in 2019. I cannot reproduce these problems on lxplus-arm, so try to turn it back on.

[github-actions[bot]]

Test Results

    18 files      18 suites   4d 2h 21m 3s ⏱️  2 662 tests  2 662 ✅ 0 💤 0 ❌ 46 198 runs  46 198 ✅ 0 💤 0 ❌

Results for commit b0121698.

:recycle: This comment has been updated with latest results.

[dpiparo]

I propose to merge it once we have the ARM nodes online to be able to test immediately, would this be ok?

[hahnjo]

Yes, I'm waiting for the AArch64 node in our CI so we can test there, and then (after) I'd still like to ask CMS to run their tests.

[hahnjo]

@aandvalenzuela @smuzaffar if you have some cycles, can you test this change with CMSSW on AArch64? This should align the configurations with x86_64 to also enable runtime_cxxmodules by default

[smuzaffar]

https://github.com/cms-sw/root/pull/212 is running cmssw aarch64 tests

[smuzaffar]

Hi, most of cmssw tests passed but for few relvals we get runtime errors like [a]

[a] https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-7b6638/42123/runTheMatrix-results/140.063_RunZeroBias2022D/step3_RunZeroBias2022D.log

cling JIT session error: In graph cling-module-926-jitted-objectbuffer, section .text._ZNK4reco10HitPattern23numberOfLostTrackerHitsENS0_11HitCategoryE: relocation target "_ZN4reco10HitPattern16missingHitFilterEt" at address 0x4000968500f0 is out of range of Page21 fixup at 0x4001a7270114 (_ZNK4reco10HitPattern23numberOfLostTrackerHitsENS0_11HitCategoryE, 0x4001a727010c + 0x8)
----- Begin Fatal Exception 11-Oct-2024 15:08:51 CEST-----------------------
An exception of category 'FatalRootError' occurred while
   [0] Processing  Event run: 357735 lumi: 53 event: 87840020 stream: 0
   [1] Running path 'dqmoffline_1_step'
   [2] Prefetching for module NanoAODDQM/'nanoDQM'
   [3] Prefetching for module SimplePATTauFlatTableProducer/'boostedTauTable'
   [4] Prefetching for module PATObjectCrossLinker/'linkedObjects'
   [5] Prefetching for module PATMuonRefSelector/'finalMuons'
   [6] Prefetching for module PATMuonUserDataEmbedder/'slimmedMuonsWithUserData'
   [7] Calling method for module EvaluateMuonMVAID/'muonMVAID'
   Additional Info:
      [a] Fatal Root Error: @SUB=TClingCallFunc::make_wrapper
Failed to compile
  ==== SOURCE BEGIN ====
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wformat-security"
__attribute__((used)) __attribute__((annotate("__cling__ptrcheck(off)")))
extern "C" void __cf_365(void* obj, int nargs, void** args, void* ret)
{
   if (ret) {
      new (ret) (double) (((const reco::TrackBase*)obj)->validFraction());
      return;
   }
   else {
      (void)(((const reco::TrackBase*)obj)->validFraction());
      return;
   }
}
#pragma clang diagnostic pop
  ==== SOURCE END ====

----- End Fatal Exception -------------------------------------------------
Another exception was caught while trying to clean up files after the primary fatal exception.

[hahnjo]

Hi, most of cmssw tests passed but for few relvals we get runtime errors [...]

Thanks for testing! This needs debugging (likely after CHEP)...

[hahnjo]

Revisiting this PR before the end of the year: Thanks to the clear error message and some guess work, I managed to reproduce the issue in a standalone ROOT session:

root [0] struct A { static void f() {} void take_f(void (*fp)()) { fp(); } void pass_f() { take_f(f); } void call_f() { f(); } };
root [1] A::f()
root [2] #include <sys/mman.h>
root [3] for (int i = 0; i < 1024 * 1024; i++) { mmap(nullptr, 8192, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, -1, 0); }
root [4] A a;
root [5] a.call_f()
root [6] a.pass_f()
cling JIT session error: In graph cling-module-16-jitted-objectbuffer, section .text._ZN11__cling_N501A6pass_fEv: relocation target "_ZN11__cling_N501A1fEv" at address 0xffffa18a4034 is out of range of Page21 fixup at 0xfffd88b60044 (_ZN11__cling_N501A6pass_fEv, 0xfffd88b6003c + 0x8)

Currently still investigating why this happens with runtime_cxxmodules but not without...

[hahnjo]

Alright, the issue with the reproducer in https://github.com/root-project/root/pull/16401#issuecomment-2532084281 is understood and fixed. Let's hope that this is also fixes the CMS relvals - @aandvalenzuela @smuzaffar could you maybe run the tests again? Thanks in advance for all your help!

[smuzaffar]

Alright, the issue with the reproducer in #16401 (comment) is understood and fixed. Let's hope that this is also fixes the CMS relvals - @aandvalenzuela @smuzaffar could you maybe run the tests again? Thanks in advance for all your help!

cmssw tests started via https://github.com/cms-sw/root/pull/215

[smuzaffar]

cmssw tests for aarch64 look good

[hahnjo]

cmssw tests for aarch64 look good

Fantastic, thank you! Unfortunately our macOS nodes are not happy, so I'll need to push a slightly fixed version. I don't think we strictly need to test with full CMSSW once more...

[hahnjo]

@vgvassilev FYI you already approved this three months ago, and now that the issue in CMS is fixed I plan to land this soon

[vgvassilev]

Go ahead.