"Invalid instruction" when passing code to SLURM cluster

Where did you compile it ? And on which partition did you executed it?

"Invalid instructions" seems that you are using invalid instruction for the cpu that you are using.
Lemaitre3 has two partitions with two different type of cpu.
The debug partition has older cpu which does not support avx512 type of instructions so if you want to run within that partition then you have to reduce compilation flag or compile directly on those nodes

Cheers,

Olivier

> On 25 Feb 2021, at 18:01, matteo maltoni <email address hidden> wrote:
>
> New question #695742 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/695742
>
> Hi!
>
> I'm trying to submit to the lemaitre3 cluster a python3 code which computes matrix elements for four-jet events, but the job fails due to an "invalid instruction" in my code, which I paste at the end of the post. The code runs normally both on my laptop and on the front-end of the cluster.
>
> Can you give me any hints about what's wrong?
>
> Here's my code:
>
>
> import sys
> sys.path.append ('/home/users/m/m/mmaltoni/MG5_aMC_v2_9_2')
> sys.path.append ('/home/users/m/m/mmaltoni/MG5_aMC_v2_9_2/HEPTools/lhapdf6_py3/bin')
> sys.path.append ('/home/users/m/m/mmaltoni/MG5_aMC_v2_9_2/Standalone_4j/SubProcesses')
> import allmatrix2py
> import madgraph.various.lhe_parser as lhe_parser
> import madgraph.various.misc as misc
> import myfunctions as myf
> from itertools import permutations
> import time
> import numpy as np
>
> allmatrix2py.initialise ('/home/users/m/m/mmaltoni/MG5_aMC_v2_9_2/bin/PROC_TopEffTh_0/Cards/param_card.dat')
> lhe = lhe_parser.EventFile ("/home/users/m/m/mmaltoni/MG5_aMC_v2_9_2/bin/PROC_TopEffTh_0/Events/run_02/unweighted_events.lhe.gz")
> lhapdf = misc.import_python_lhapdf ("/home/users/m/m/mmaltoni/MG5_aMC_v2_9_2/HEPTools/lhapdf6_py3/bin/lhapdf-config")
>
> # Getting a PDF member object
> pdf = lhapdf.mkPDF("NNPDF23_lo_as_0130_qed", 0)
> ELHC = 6.5e3
> FactScale = 5e2
>
> sphercut = 0.36
>
> #-------main
>
> pdg_list = [list of all the possible subprocesses' pdg lists]
>
> me_pos = 0
> me_neg = 0
>
> sigma_meas = 0.
> sigma_meas_sph = 0.
>
> nev = 0
> start_t = time.time ()
>
> for event in lhe:
> nev += 1
> if nev%1000. == 0:
> elapsed = (time.time () - start_t)/60
> minu, sec = myf.time_conversion (elapsed)
> print (nev, minu, 'min', sec, 'sec')
> weight = myf.math.fabs (event.wgt)
> p = []
> p_final = []
> pdgl = []
> #pT = []
>
> for particle in event:
> mom = [particle.E, particle.px, particle.py, particle.pz]
> p.append (mom)
> pdgl.append (particle.pdg)
> if particle.status == 1:
> p_final.append (mom)
>
> spher = myf.transv_spher (p_final)
>
> P = myf.invert_momenta (p)
>
> #calculating the weighted matrix element
> ans_int = 0.
>
> for pdgl in pdg_list:
> for pdgl_in in permutations (pdgl[:2]):
> pdfw = pdf.xfxQ(pdgl_in[0], p[0][0]/ELHC, FactScale)*pdf.xfxQ(pdgl_in[1], p[1][0]/ELHC, FactScale)
> ans = 0
> for pdgl_out in permutations (pdgl[2:]):
> pdgl_f = pdgl_in + pdgl_out
> ans += allmatrix2py.smatrixhel (pdgl_f, 1, P, .13, 0., -1)
> ans_int += (ans * pdfw) / 6.
>
> #Counting events
> sigma_meas += event.wgt * np.sign (ans_int)
> sigma_meas_sph += event.wgt * np.sign (spher - sphercut)
>
>
> print ('sigma meas:', sigma_meas)
> print ('sigma spher:', sigma_meas_sph)
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Hi Olivier,

Thank you for your answer.
I executed the script in my $HOME directory on the cluster, allowing both partitions; you can find my script below (sbatch script_name.sh). This is the error I get:

Task ID: 1
/var/spool/slurmd/job69700787/slurm_script: line 20: 11911 Istruzione non consentita python3 $HOME/MG5_aMC_v2_9_2/Standalone_4j/SubProcesses/matrix.py

How can I reduce compilation flag (for lhapdf, for instance)?

Cheers,

Matteo

P.S. Here's my script:

#!/bin/bash
# Submission script for Lemaitre3
#SBATCH --job-name=fourjet
#SBATCH --array=1-10
#SBATCH --time=02:00:00 # hh:mm:ss
#
#SBATCH --ntasks=1
#SBATCH --mem-per-cpu=2625 # megabytes
#SBATCH --partition=batch,debug
#
#SBATCH --<email address hidden>
#SBATCH --mail-type=END,FAIL
#
#SBATCH --comment=fourjet
#SBATCH --output=fourjet.out

echo "Task ID: $SLURM_ARRAY_TASK_ID"

module load Python/3.7.4-GCCcore-8.3.0
python3 $HOME/MG5_aMC_v2_9_2/Standalone_4j/SubProcesses/matrix.py

I apologise for the Italian, I totally forgot to translate. Some online forums suggest for it a translation like "Invalid" or "Illegal instruction".

Anyway, asking for the batch partition only solved my problem: the job ran on the cluster with no issues.

Cheers,

Matteo

Thanks Olivier Mattelaer, that solved my question.