def profile_memsize_nthreads(
executable,
nthreads=[1, 2, 4],
gridsize=np.arange(200, 800 + 1, 200),
timecmd="time",
nsteps=1000,
fplu=7,
double=False,
verbose=False,
showplot=False,
figname="",
):
"""Record the time the execution of a programme takes"""
NN = gridsize
timefmt = "%e"
# iterate over nthreads
time = np.zeros((len(nthreads), len(NN)))
for iN, N in enumerate(NN):
executable_with_args = executable + " -n {0}".format(N)
for inthr, OMP_NUM_THREADS in enumerate(nthreads):
print "\nNow running on ", OMP_NUM_THREADS, " threads"
# set environment variable
os.environ["OMP_NUM_THREADS"] = str(OMP_NUM_THREADS)
# get execution time
time[inthr, iN] = timeomp.get_execution_time(
executable_with_args, timecmd=timecmd, timefmt=timefmt, verbose=verbose
)[0]
# write data to file
data = dict(nthreads=nthreads, time=time, NN=NN, nsteps=nsteps, fplu=fplu, double=double)
with open("scaling_memsize_nthreads.pkl", "wb") as fout:
pickle.dump(data, fout)
# plot data
if showplot or figname:
plot.plot_flops_vs_memsize(savefig=bool(figname), figname=figname, **data)
def profile_nthreads(executable,nthreads=[],timecmd="time",timefmt="%e %U %S",
labels=None,verbose=False,savefig=False):
"""Record the time the execution of a programme takes"""
# iterate over nthreads
times = np.zeros((len(nthreads),len(timefmt.split())))
for i,omp_num_threads in enumerate(nthreads):
print '\nNow running on ',omp_num_threads,' threads'
os.environ['OMP_NUM_THREADS'] = str(omp_num_threads)
times[i,:] = timeomp.get_execution_time(executable,timecmd=timecmd,
timefmt=timefmt,labels=labels,
verbose=verbose)
data = dict(nthreads=nthreads,times=times)
# write output to file
with open('scaling_nthreads.pkl','w') as fout:
pickle.dump(data,fout)
plot.plot_times_vs_nthread(labels=labels,**data)
