This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[gomp] omp performance question


2006/10/30, François-Xavier Coudert <fxcoudert@gmail.com>:
> single threaded ( FCFLAGS=-O1) timings on x64_64, dual CPU (dual core
> each), gave:
> real    64m36.502s
> user    64m36.886s
> sys     0m0.040s
>
> same machine, OMP enabled  (FCFLAGS="-O1 -fopenmp"):
> real    67m16.611s
> user    112m22.885s
> sys     25m44.641s

What is OMP_NUM_THREADS set to?

The numbers above use the default implementation without setting OMP_NUM_THREADS explicitely. Letting OMP_NUM_THREADS=4, i.e. the number of cores available, results are:

OMP_NUM_THREADS=4, OMP_DYNAMIC=[DEFAULT]
real    59m47.223s
user    91m33.127s
sys     24m35.856s

OMP_NUM_THREADS=4, OMP_DYNAMIC=FALSE
real    57m36.233s
user    91m59.685s
sys     26m31.735s


Some more timings (same machine): OMP_NUM_THREADS=4, OMP_DYNAMIC=TRUE IN PROGRESS (ps -efL shows four threads running)

OMP_NUM_THREADS=8, OMP_DYNAMIC=TRUE
real    168m46.983s
user    97m53.903s
sys     113m0.108s
(ps -efL showed three threads actually running)

OMP_NUM_THREADS=8, OMP_DYNAMIC=FALSE
real    85m54.649s
user    125m20.442s
sys     48m15.253s

I'll try to add some numbers from ia64 (8CPU) if I can get exclusive hold of it.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]