def main():
# Look for parallel in args
parser = argparse.ArgumentParser()
parser.add_argument('--parallel', action = "store_true")
args = parser.parse_args()
# What to run
argdicts = {
"-driver" : ["llg"],
"-ts" : ["midpoint-bdf", "bdf2"],
"-ms-method" : ["implicit", "decoupled", "disabled"],
"-solver" : "gmres",
"-matrix-type" : "som",
"-prec" : "som-main-exact",
"-tmax" : 100,
"-tol" : 0.01,
"-mp-pred" : "ebdf3",
# "-initial-m" : "exactly_z",
}
# Where it's going to end up
base_outdir = os.path.abspath(pjoin(os.path.dirname(__file__), "Validation"))
# Run
err_codes, outdirs = mm.run_sweep(argdicts, base_outdir,
serial_mode=not args.parallel)
# Get data
datasets = list(map(mm.parse_run, outdirs))
# Check things
ran = all((e == 0 for e in err_codes))
t1 = all([tests.check_ndt_less_than(d, 80) for d in datasets
if d['-ms-method'] != "decoupled"])
t2 = all([tests.check_ndt_less_than(d, 115) for d in datasets
if d['-ms-method'] == "decoupled"])
# Decoupled ms introduces wiggles which put an effective cap on the step size.
if all([ran, t1, t2]):
return 0
else:
return 1
def main():
# What to run
argdicts = {
"-driver" : "ode",
"-exact" : ["sin", "poly3", "poly2", "stiff_test"],
"-ts" : ["bdf2", "midpoint-bdf", "tr", "bdf1"],
"-tmax" : 10,
"-tol" : 1e-4,
"-disable-mm-opt" : 1,
"-always-write-trace" : 1, # Otherwise we get wrong ndts by counting len(dts)
}
# ??ds not sure why we need "-disable-mm-opt",
# but there's something stupid going on with mass matrix storage
# Where it's going to end up
base_outdir = os.path.abspath(pjoin(os.path.dirname(__file__), "Validation"))
# Run
err_codes, outdirs = mm.run_sweep(argdicts, base_outdir)
# Get data
datasets = list(filter(lambda d: d is not None, map(mm.parse_run, outdirs)))
# Check things ran
test_results = []
test_results.append(all([e == 0 for e in err_codes]))
# Use bdf2's nsteps as a maximum, tr and imr are more accurate than
# bdf2 so this should be true (unless bdf2's numerical damping has
# kicked in and caused it to jump to a steady state too soon), (we
# assume that the order of data is preserved here so that bdf_data[n]
# is the same exact solution as imr_data[n] etc.).
bdf2_data = [d for d in datasets if d['-ts'] == 'bdf2']
for ts in argdicts['-ts']:
# bdf1 sucks (first order) so it needs far more steps, do it
# manually.
if ts == "bdf1":
max_err = 0.4
max_steps = [550, 3800, 1050, 70]
else:
max_err = 0.07
max_steps = [1.3*len(d['times']) for d in bdf2_data]
ts_data = [d for d in datasets if d['-ts'] == ts]
# Check errors are small
test_results.append(all([tests.check_error_norm(d, max_err) for d in ts_data]))
# Check all the runs:
nsteps_ok = [tests.check_ndt_less_than(d, m)
for d, m in zip(ts_data, max_steps)]
test_results.append(all(nsteps_ok))
if all(test_results):
return 0
else:
return 1
