def test_run():
run_script.run(script,
workdir=workdir,
ncount=ncount,
ntotalthreads=int(1e6 / 512 * 256),
threads_per_block=256)
return
def test_component_long(ncount = 1e6):
instr = os.path.join(thisdir, "src_mon_instrument.py")
outdir = 'out.psd_monitor'
ncount = int(ncount)
run_script.run(instr, outdir, ncount=ncount, monitor_factory=psd_monitor)
if interactive:
import histogram.hdf as hh
hist = hh.load(os.path.join(outdir, "psd.h5"))
from histogram import plot as plotHist
plotHist(hist)
return
def run(instrument_script,
ncount,
niters,
mpi=False,
nodes=1,
acc_run=False,
skip_ncounts=[],
**kwds):
outdir = 'out.debug-check-speed'
if os.path.exists(outdir):
shutil.rmtree(outdir)
# TODO: this stdout file gets overwritten on each call, move or use append?
stdout = open("./check-speed-stdout.txt", "w")
sys.stderr = stdout
avg_times = []
for n in ncount:
if n in skip_ncounts:
# insert element to preserve ordering
avg_times.append(np.nan)
continue
print(" Running '{}' with n={}".format(instrument_script, n))
times = []
# redirect script output to reduce terminal clutter
for iter in range(niters + 1):
sys.stdout = stdout
buffer_size = int(n) if n > 1e6 else int(1e6)
if buffer_size > 1e9:
buffer_size = int(1e9)
# get the runtime for the script. Note: this probably isn't the
# best timing, since it will include the overhead of launching the
# script and any file IO done inside the script, but there is no
# current way to time individual components.
# TODO: the stdout could be parsed to pull the process timings for acc components
# TODO: there is still mcvine component output, likely from a subprocess in which this redirection has no effect
if mpi:
# run script in MPI mode
# NOTE: with mcvine=0.44, pyyaml 5.3 must be used to avoid yaml load error. This is fixed in newer versions of mcvine.
time_before = time.time_ns()
run_script.run_mpi(instrument_script,
outdir,
nodes=nodes,
buffer_size=buffer_size,
ncount=n,
overwrite_datafiles=True,
**kwds)
time_after = time.time_ns()
elif acc_run:
# use accelerated run script
time_before = time.time_ns()
run_acc_script.run(instrument_script,
outdir,
buffer_size=buffer_size,
ncount=n,
overwrite_datafiles=True,
**kwds)
time_after = time.time_ns()
else:
# use default run script (mcvine, single node)
time_before = time.time_ns()
run_script.run1(instrument_script,
outdir,
buffer_size=buffer_size,
ncount=n,
overwrite_datafiles=True,
**kwds)
time_after = time.time_ns()
# skip first run
if iter < 1:
continue
delta = time_after - time_before
times.append(delta)
sys.stdout = sys.__stdout__
time_avg = np.sum(np.asarray(times)) / len(times)
avg_times.append(time_avg)
print(" TIME = {} ms ({} s)".format(time_avg * 1e-6, time_avg * 1e-9))
sys.stdout = sys.__stdout__
sys.stderr = sys.__stderr__
stdout.close()
return avg_times
def test_run():
run_script.run(script, workdir, ncount=ncount)
return