search

madgraph5 / madgraph4gpu

GPU development for the Madgraph5_aMC@NLO event generator software package
28 stars 33 forks source link

Fix 825 #850

Closed oliviermattelaer closed 1 month ago

oliviermattelaer commented 1 month ago

This does fix #825 where the problem was that the unary handling was not propagated within the aloha routine. This patch include such handling now but I'm a bit worried about other processes that can be impacted. This PR (via the CI) is a simple way to test many processes to spot potential issue, if it goes trough, I will anyway have to check the corner case, I'm worried about...

oliviermattelaer commented 1 month ago

@roiser, ok this is good to be squashed and merged.

Olivier

valassi commented 1 month ago

Hi @oliviermattelaer I had a look since I saw that you posted this.

I confirm that #825 seems fixed: the cross section from fortran and cuda/c++ now matches in susy_gg_tt.

Conversely for #826 I still see that there is NO cross section for susy_gg_t1t1. So I am a bit surprised when you say that you do see a cross section. Maybe it depends on the number of events and the exact setup? Anyway, I agree with you that this needs more investigation.

One point, I see that this patch also changes the Parameters.h files in other processes including eemumu. Is this expected? Anyway, I will run a few tests on all processes and see what this gives.

Just in case, this is in https://github.com/valassi/madgraph4gpu/tree/susy (but I will not make a PR out of this, at least not yet)

Thanks! Andrea

oliviermattelaer commented 1 month ago

Hi Andrea,

in Parameters.h this sounds surprising (I do not see any of those files changed in your repo actually) But in all process/model, the file HelAmps_sm.h is going to be modified since the issue has nothing specific to the MSSM model and we could --maybe we should-- have detected in the SM has well.

For #826, I will "first" try to fix the missmatch that I do observe on my laptop, and then we might need to work together to understand why your machine is just crashing (if it is still happens at that stage obviously).

valassi commented 1 month ago

Hi Andrea,

in Parameters.h this sounds surprising

OPS! SORRY! Yes I meant that HelAmps.h changed, not Paramaters.h!

valassi commented 1 month ago

For me this is good to go and should be merged. However it will then also need an update of refence files for runTest (#859), which is in my PR #860. So I suggest,

valassi commented 1 month ago

I merged my #860. As I expected, github automatically closed this one as merged too. Thanks Olivier!