Closed valassi closed 3 months ago
I mark this as a major issue and pin it, because this affects all SM results in user code. As pointed out by Olivier in PR #852, there is a problem in cudacpp with iconfig-channel mappings. IMO this must be fixed before the release.
This issue #856 (LHE color mismatch in gg_ttgg for iconfig=104) can now be reproduced in the CI if rotxxx is fixed. In the PR #857 which fixes rotxxx with volatile, I configured the tmad test for gg_ttgg to use iconfig=104. I get the following failure https://github.com/madgraph5/madgraph4gpu/actions/runs/9698902245/job/26766683190
*** (2-none) Compare MADEVENT_CPP xQUICK events.lhe to MADEVENT_FORTRAN events.lhe reference (including colors and helicities) ***
ERROR! events.lhe.none.QUICK and events.lhe.ref.QUICK differ!
diff /home/runner/work/madgraph4gpu/madgraph4gpu/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/events.lhe.none.QUICK /home/runner/work/madgraph4gpu/madgraph4gpu/epochX/cudacpp/gg_ttgg.mad/SubProcesses/P1_gg_ttxgg/events.lhe.ref.QUICK | head -20
4c4
< 21 -1 0 0 501 503 -0.00000000000E+00 -0.00000000000E+00 -0.33445071051E+03 0.33445071051E+03 0.00000000000E+00 0. 1.
---
> 21 -1 0 0 502 503 -0.00000000000E+00 -0.00000000000E+00 -0.33445071051E+03 0.33445071051E+03 0.00000000000E+00 0. 1.
6c6
< -6 1 1 2 0 502 -0.96800713603E+02 0.45396286052E+02 0.12624284574E+03 0.23936887233E+03 0.17300000000E+03 0. 1.
---
> -6 1 1 2 0 504 -0.96800713603E+02 0.45396286052E+02 0.12624284574E+03 0.23936887233E+03 0.17300000000E+03 0. 1.
8c8
< 21 1 1 2 505 504 -0.56352823282E+01 -0.25774621670E+02 -0.66514459789E+01 0.27208992315E+02 0.00000000000E+00 0. 1.
---
> 21 1 1 2 505 501 -0.56352823282E+01 -0.25774621670E+02 -0.66514459789E+01 0.27208992315E+02 0.00000000000E+00 0. 1.
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
[testsuite_oneprocess.sh] tmad_test (gg_ttgg.mad) finished with status=1 (NOT OK) at Thu Jun 27 15:00:19 UTC 2024
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
[testsuite_oneprocess.sh] tmad_test (gg_ttgg.mad) FPTYPE=d: issue will not be bypassed, test has FAILED
A tentative fix for this issue is in https://github.com/mg5amcnlo/mg5amcnlo/pull/116 This was initially meant to be merged to madgraph4gpu in PR #877, but this will not happen. An equivalent patch in cudacpp was also developed as an alternative, but will also not be enabled in 877. (The association to #877, which had been added at some point, has therefore been removed).
This remains a major pending issue in my opinion.
Marking as reopened in the sense that it was NOT fixed in #877.
Marking as closed because it was fixed by Olivier in #880.
Unpinning the issue as this was finally fixed
This is a followup to #855.
If I fix the SIGFPE in rotxxx by adding volatile in fortran code, the crash is avoided but then there is an LHE file mismatch
I think that this is due to the issue identified by Olivier in WIP PR #852, namely channel/iconfig mapping issues.
But this is an INDEPENDENT issue from the SIGFPE crash (even ifit is bizarre that both happen only for specific iconfig choices).