Profiling

[Juergen Reuter]

Profiling is necessary for large LHC 2->6 processes, in the SM as well as especially in BSM models. I am using specifically the following process, testable with the svn tree.

model = NMSSM_Hgg

alias tau_b = e3:E3:b:B
alias proton = u:d:s:c:U:D:S:C:g
alias fjet = proton:tau_b

process pp_4t = proton, proton => fjet, fjet, e3, E3, e3, E3

compile

?slha_read_decays = true
read_slha("nmssm.slha")

cuts = all Pt >= 10 GeV [fjet]
       and all -5 <= Eta <= 5 [fjet]
       and all Dist > 0.4 [fjet,fjet]

sqrts = 14 TeV
beams = p,p => lhapdf

integrate (pp_4t) { iterations = 5:50000,15:100000 }

The code generation takes roughly order of an hour, compilation and phase space generation 1-2 days. The first iteration is now running 2 days, still waiting for the first number.

[kilian]

To give profiling a chance, I fixed the configure option --enable-fc-profiling in -r2995, also added --enable-fc-static which is needed here. Note that for useful profiling, one has to create a standalone executable

process = ...
compile as "..."

and then run this executable in place of whizard.

For a simpler 2->6 process I verified that most of the running time is in the matrix element, as it should be, so the only likely speedup would be parallelization. But I'll check the present example as well.

[Juergen Reuter]

What is the status here?

[kilian]

After some mangling with static executables, profiling works, and I close the ticket. Individual profile results may be attached to the tickets where they are useful, e.g., #389, #380.

Here's a howto:

• configure Whizard with --enable-fc-profiling
• make a static executable, say whizard-static

  compile as "whizard-static"
  

• run this executable instead of whizard

This should set -pg flags wherever needed, and the run produces gmon.out, which can be interpreted as usual:

gprof whizard-static > gprof.out