search

madgraph5 / madgraph4gpu

GPU development for the Madgraph5_aMC@NLO event generator software package
30 stars 32 forks source link

LHE file mismatch fortran vs cudacpp in tlau (for multi backend builds, possibly for normal builds too) - also: "fail to reach target" #993

Open valassi opened 2 months ago

valassi commented 2 months ago

I just had a quick first look at lhe event files produced in multi backend tests.

In dy+3j just a few occasional events are the same, but most are different. The problem is also that the weight is completely different. This is from the 500 events in PR #948

diff events_fortran.lhe events_cppnone.lhe | more
450c450
< #  Integrated weight (pb)  :       142.38107118268414
---
> #  Integrated weight (pb)  :       3199.366020218372
455c455
< 1.423811e+02 6.637245e-01 1.423811e+02 0
---
> 3.199366e+03 1.661125e+01 3.199366e+03 0

Will need to understand this.

valassi commented 1 month ago

I added 'fail to reach target' in the title, I saw this in dy+4j: what is this? And is this related to what I observed above in dd+3j?

In dy+4jets

[avalassi@itscrd90 gcc11/usr] /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tlau/fromgridpacks> ls -l pp_dy4j.mad/*/*
-rw-r--r--. 1 avalassi zg  19851 Sep 14 14:36 pp_dy4j.mad/cpp512y/events.lhe.gz
-rw-r--r--. 1 avalassi zg 114963 Sep 14 14:36 pp_dy4j.mad/cpp512y/output.txt
-rw-r--r--. 1 avalassi zg  19851 Sep 14 15:06 pp_dy4j.mad/cpp512z/events.lhe.gz
-rw-r--r--. 1 avalassi zg 114963 Sep 14 15:06 pp_dy4j.mad/cpp512z/output.txt
-rw-r--r--. 1 avalassi zg  19851 Sep 14 14:07 pp_dy4j.mad/cppavx2/events.lhe.gz
-rw-r--r--. 1 avalassi zg 114963 Sep 14 14:07 pp_dy4j.mad/cppavx2/output.txt
-rw-r--r--. 1 avalassi zg  19851 Sep 14 12:42 pp_dy4j.mad/cppnone/events.lhe.gz
-rw-r--r--. 1 avalassi zg 114986 Sep 14 12:42 pp_dy4j.mad/cppnone/output.txt
-rw-r--r--. 1 avalassi zg  19851 Sep 14 13:37 pp_dy4j.mad/cppsse4/events.lhe.gz
-rw-r--r--. 1 avalassi zg 114978 Sep 14 13:37 pp_dy4j.mad/cppsse4/output.txt
-rw-r--r--. 1 avalassi zg  19851 Sep 14 11:00 pp_dy4j.mad/cuda/events.lhe.gz
-rw-r--r--. 1 avalassi zg 115208 Sep 14 11:00 pp_dy4j.mad/cuda/output.txt
-rw-r--r--. 1 avalassi zg  35494 Sep 14 10:46 pp_dy4j.mad/fortran/events.lhe.gz
-rw-r--r--. 1 avalassi zg 104982 Sep 14 10:46 pp_dy4j.mad/fortran/output.txt

egrep '(events :|fail)' pp_dy4j.mad/*/*txt
pp_dy4j.mad/cpp512y/output.txt:INFO: fail to reach target 100 
pp_dy4j.mad/cpp512y/output.txt:     Nb of events :  44
pp_dy4j.mad/cpp512z/output.txt:INFO: fail to reach target 100 
pp_dy4j.mad/cpp512z/output.txt:     Nb of events :  44
pp_dy4j.mad/cppavx2/output.txt:INFO: fail to reach target 100 
pp_dy4j.mad/cppavx2/output.txt:     Nb of events :  44
pp_dy4j.mad/cppnone/output.txt:INFO: fail to reach target 100 
pp_dy4j.mad/cppnone/output.txt:     Nb of events :  44
pp_dy4j.mad/cppsse4/output.txt:INFO: fail to reach target 100 
pp_dy4j.mad/cppsse4/output.txt:     Nb of events :  44
pp_dy4j.mad/cuda/output.txt:INFO: fail to reach target 100 
pp_dy4j.mad/cuda/output.txt:     Nb of events :  44
pp_dy4j.mad/fortran/output.txt:     Nb of events :  100