sPHENIX tracking discussion

[pinkenburg <pinkenburg AT bnl.gov>]

Hi folks,

I promised yesterday to send the location of the profiling plots for our 
tracking. Under
/phenix/scratch/pinkenbu/sPhenix/haiwang/callgrind
you find 3 reports, tpcclustering.png shows a single hijing event run up 
to the tpc clustering (which does take a substantial fraction of cpu all 
by itself). The code spends > 10% of it's time creating poisson 
distributed random numbers using TRandom3 and 20% calculating 
exponential. That might be a good start. I have a vague recollection 
that TRandoms implementation of random distributions lets a lot to be 
desired, one could try to replace it by the gsl equivalent. Also using 
strings as keys in a map is a bad idea if code should run fast.

pythia.png and hijing.png are from tracking a single pythia and hijing 
event. Sadly no obvious hot spots which would give us a factor of 10.

Mike put a howto in our sPHENIX wiki from which you can just cut and paste:

https://wiki.bnl.gov/sPHENIX/index.php/Tools#callgrind

The only caveat is that annotation right now requires a private build of 
the libraries. By default it looks for the location of the sources 
(/home/phnxbld/...) which are not available on our interactive machines. 
There's probably a command line option I haven't found yet.

Be aware that it runs under valgrind which can take significant time 
(the central hijing event took about 24 hours to process - and this was 
without the G4 part)

Chris


--
*************************************************************

Christopher H. Pinkenburg       ;    pinkenburg AT bnl.gov
                                ;    http://www.phenix.bnl.gov/~pinkenbu

Brookhaven National Laboratory  ;    phone: (631) 344-5692
Physics Department Bldg 510 C   ;    fax:   (631) 344-3253
Upton, NY 11973-5000

*************************************************************