Problem for CKKW merging algorithm (for the "CMS Drell Yan" process)

[oliviermattelaer]

crash in the pure fortran vectorised code

generate p p > ta+ ta-
add process p p > ta+ ta- j
#add process p p > ta+ ta- j j
output PROC_sm_2 --vector_size=2
launch
set ickkw 1
set mc_grouped_subproc F

Findings so far:

• issue is the behaviour of the findmt function
• difference with vector_size is the value of iconfig when reaching that routine.
  • When running G2 the vectorised code hase iconfig set to 1 and not to 2

[oliviermattelaer]

Fixed issue: mg5amcnlo/mg5amcnlo@ef334c5
@roiser @valassi

[valassi]

Thanks Olivier! For reference, this is relevant to the CMS process suggested by Sapta

They use these cards
https://github.com/cms-sw/genproductions/tree/mg29x/bin/GenValidation/updated_validation_cards/dy_mlm_mll50

And they generate the gridpack with this code:
https://github.com/cms-sw/genproductions/blob/master/bin/MadGraph5_aMCatNLO/gridpack_generation.sh

[roiser]

Hi @oliviermattelaer , thanks, was this the issue in "cluster.f" we discussed earlier? cheers

[valassi]

For the record, I confirm that this is fixed in commit mg5amcnlo/mg5amcnlo@6d2182c which I am now using in madgraph4gpu (see MR #654, in 57d8ec3), while I can reproduce an error after launch in mg5amcnlo/mg5amcnlo@c1d10f4

So this is definitely fixed