Closed valassi closed 8 months ago
Is your issue is this?
Command "generate_events run_01" interrupted with error:
InvalidCardEdition : value 'fortran' for entry 'cudacpp_backend' is not valid. Preserving previous value: 'CPP'.
allowed values are Fortan, CPP, CUDA
Please report this bug on https://bugs.launchpad.net/mg5amcnlo
More information is found in 'ME5_debug'.
Please attach this file to your report.
I'm looking about that issue (not sure if this is mg5 or the plugin), I can not do any run as long as this is not solved (and therefore can not run on #801), so I'm investigating it for the moment
UPDATE: that one was easy to fix (it is obvious but my brain hide to issue for me for so so long ....)
Hi Olivier, sorry NO! I was wrong.
Code generation succeeds if I update mg5amcnlo to gpucpp on top of the master of madgraph4gpu. I will test this here.
Instead I have the impression that code generation fails if I update mg5amcnlo to gpucpp on top of the jt774 branch (PR #801) of madgraph4gpu. So it seems that there is an interplay of my/Jorgen's changes for HIP and your changes for mg5amcnlo? I need to check better, sorry.
If the above is confirmed, I suggest:
Sounds ok? But let me check before I say stupid things. It will take me time to run all tests however, will merge this tomorrow at the earliest
Is your issue is this?
Command "generate_events run_01" interrupted with error: InvalidCardEdition : value 'fortran' for entry 'cudacpp_backend' is not valid. Preserving previous value: 'CPP'. allowed values are Fortan, CPP, CUDA Please report this bug on https://bugs.launchpad.net/mg5amcnlo More information is found in 'ME5_debug'. Please attach this file to your report.
I'm looking about that issue (not sure if this is mg5 or the plugin), I can not do any run as long as this is not solved (and therefore can not run on #801), so I'm investigating it for the moment
UPDATE: that one was easy to fix (it is obvious but my brain hide to issue for me for so so long ....)
No my issue was different... but again, I think only in jt774 ie PR 801
Ok, I have pushed the fix for the mistake that I face directly in jt774. For the update of the current submodule, I do not use it personally, so this is not urgent for me.
Hi Olivier, thanks. Ok thanks for confirming that this PR #811 and the gpucpp update is not urgent.
I will look at the jt774 instead then, thanks for adding the fix directly there.
Instead I have the impression that code generation fails if I update mg5amcnlo to gpucpp on top of the jt774 branch (PR #801) of madgraph4gpu. So it seems that there is an interplay of my/Jorgen's changes for HIP and your changes for mg5amcnlo? I need to check better, sorry.
Ok not sure anymore why/where I saw an exception in reset_simd. Now code generation with this update seems ok also in jt774? But I will postpone this anyway. I will do PR #801 WITHOUT this update. And convert this to WIP (and postponed)
Looks like we hit "could not check for binary extension: HTTP 403: API rate limit exceeded for installation ID 13042647. " which might have trigger some false alarm in the recent commit/PR... I guess we do too much work for github ;-)
Hi Olivier, in the end it was easier for me to do this one first.
I upgraded mg5amcnlo to the current (well yesterday's) gpucpp branch, have regenerated the code and run all tests.
If this is ok for you I would go ahead and merge.
Apart from generated code and test logs, there are no changes other than that in MG5aMC/mg5amcnlo.
It should be a trivial change, but still please approve it ;-)
Thanks Andrea
Thanks a lot Olivier, now merging
Hi @oliviermattelaer as discussed in PR #801.
This WIP PR updates the mg5amcnlo submodule to the latest commit in branch gpucpp.
Note: code generation now fails (well it is actually treatcards in my case) because reset_simd in banner.py throws an exception. It would be enough to change that exception to a warning for the time being, an dthings would be better.
Or otherwise we can take the time to understand better what goes on.