diff --git a/examples/example_utils.py b/examples/example_utils.py
index 11f7742a8f8555706e1e45348e7dd95129790838..b684387a4d076ce7d60f0c39f1f4a08cde78fa32 100644
--- a/examples/example_utils.py
+++ b/examples/example_utils.py
@@ -94,6 +94,8 @@ class HysopArgParser(argparse.ArgumentParser):
os.environ[target] = '0'
elif isinstance(value, str):
os.environ[target] = value
+ elif isinstance(value, int):
+ os.environ[target] = str(value)
else:
msg='Invalid value of type {}.'.format(type(value))
raise TypeError(msg)
@@ -148,7 +150,6 @@ class HysopArgParser(argparse.ArgumentParser):
args = self.parse_args()
args.__class__ = HysopNamespace
- self._check_fft_args(args)
self._check_threading_args(args)
self._setup_hysop_env(args)
@@ -400,6 +401,21 @@ class HysopArgParser(argparse.ArgumentParser):
simu.add_argument('-lcfl', '--lagrangian-cfl', type=float, default=None,
dest='lcfl',
help='Specify LCFL for adaptive time stepping.')
+ simu.add_argument('-stopi', '--stop-at-initialization', default=False, action='store_true',
+ dest='stop_at_initialization',
+ help='Stop execution before problem initialization.')
+ simu.add_argument('-stopd', '--stop-at-discretization', default=False, action='store_true',
+ dest='stop_at_discretization',
+ help='Stop execution before problem discretization.')
+ simu.add_argument('-stopwp', '--stop-at-work-properties', default=False, action='store_true',
+ dest='stop_at_work_properties',
+ help='Stop execution before problem work properties retrieval.')
+ simu.add_argument('-stopwa', '--stop-at-work-allocation', default=False, action='store_true',
+ dest='stop_at_work_allocation',
+ help='Stop execution before problem work properties allocation (allocation of temporary buffers).')
+ simu.add_argument('-stops', '--stop-at-setup', default=False, action='store_true',
+ dest='stop_at_setup',
+ help='Stop execution before problem setup.')
simu.add_argument('-dr', '--dry-run', default=False, action='store_true',
dest='dry_run',
help='Stop execution before the first simulation iteration.')
@@ -409,7 +425,8 @@ class HysopArgParser(argparse.ArgumentParser):
self._check_default(args, ('tstart', 'tend'), float)
self._check_default(args, 'dt', float, allow_none=True)
self._check_default(args, 'nb_iter', int, allow_none=True)
- self._check_default(args, 'dry_run', bool, allow_none=False)
+ self._check_default(args, ('dry_run', 'stop_at_initialization', 'stop_at_discretization', 'stop_at_setup',
+ 'stop_at_work_properties', 'stop_at_work_allocation'), bool, allow_none=False)
self._check_positive(args, 'dt', strict=True, allow_none=True)
self._check_positive(args, 'nb_iter', strict=True, allow_none=True)
self._check_positive(args, 'max_iter', strict=True, allow_none=True)
@@ -481,13 +498,13 @@ class HysopArgParser(argparse.ArgumentParser):
threading = self.add_argument_group('threading parameters')
msg = "Enable threads for backends that supports it (Numba and FFTW) by setting HYSOP_ENABLE_THREADS. "
msg += "Disabling threading will limit all threading backends to one thread and set numba default backend to 'cpu' instead of 'parallel'."
- threading.add_argument('--enable-threading', type=str, default=True,
+ threading.add_argument('--enable-threading', type=str, default='1',
dest='enable_threading',
help=msg)
- msg="If this parameter is set to 'physical', the maximum number of threads will be set to the number of physical cores available to the process (taking into account the cpuset). "
- msg+="If set to 'logical', logical cores will be chosen instead. Else this parameter expects a positive integer."
- msg+='Set the default maximum usable threads for threading backends (OpenMP, MKL) and operator backends using threads (Numba, FFTW). '
+ msg='Set the default maximum usable threads for threading backends (OpenMP, MKL) and operator backends using threads (Numba, FFTW). '
msg+='This parameter will set HYSOP_MAX_THREADS and does not affect the OpenCL backend.'
+ msg+="If this parameter is set to 'physical', the maximum number of threads will be set to the number of physical cores available to the process (taking into account the cpuset). "
+ msg+="If set to 'logical', logical cores will be chosen instead. Else this parameter expects a positive integer."
msg+='If --enable-threads is set to False, this parameter is ignored and HYSOP_MAX_THREADS will be set to 1.'
threading.add_argument('--max-threads', type=str, default='physical',
dest='max_threads',
@@ -498,15 +515,21 @@ class HysopArgParser(argparse.ArgumentParser):
threading.add_argument('--mkl-threads', type=str, default=None,
dest='mkl_threads',
help='This parameter will set MKL_NUM_THREADS to a custom value (overrides --max-threads).')
- threading.add_argument('--fftw-threads', type=str, default=None,
- dest='fftw_threads',
- help='This parameter will set FFTW_NUM_THREADS to a custom value (overrides --max-threads).')
threading.add_argument('--numba-threads', type=str, default=None,
dest='numba_threads',
help='This parameter will set NUMBA_NUM_THREADS to a custom value (overrides --max-threads).')
threading.add_argument('--numba-threading-layer', type=str, default='workqueue',
dest='numba_threading_layer',
help="This parameter will set NUMBA_THREADING_LAYER to a custom value ('workqueue' is available on all platforms, but not 'omp' and 'tbb'). Use 'numba -s' to list available numba threading layers.")
+ threading.add_argument('--fftw-threads', type=str, default=None,
+ dest='fftw_threads',
+ help='This parameter will set HYSOP_FFTW_NUM_THREADS to a custom value (overrides --max-threads).')
+ threading.add_argument('--fftw-planner-effort', type=str, default='FFTW_ESTIMATE',
+ dest='fftw_planner_effort',
+ help='Set default planning effort for FFTW plans. The actual number of threads used by FFTW may depend on the planning step. This parameter will set HYSOP_FFTW_PLANNER_EFFORT.')
+ threading.add_argument('--fftw-planner-timelimit', type=str, default='FFTW_NO_TIMELIMIT',
+ dest='fftw_planner_timelimit',
+ help='Set an approximate upper bound in seconds for FFTW planning. This parameter will set HYSOP_FFTW_PLANNER_TIMELIMIT.')
return threading
@@ -548,8 +571,10 @@ class HysopArgParser(argparse.ArgumentParser):
return opencl
def _check_threading_args(self, args):
- self._check_default(args, ('enable_threading', 'max_threads', 'numba_threading_layer'), str, allow_none=False)
- self._check_default(args, ('openmp_threads', 'mkl_threads', 'numba_threads'), str, allow_none=True)
+ self._check_default(args, ('enable_threading', 'max_threads', 'numba_threading_layer',
+ 'fftw_planner_effort', 'fftw_planner_timelimit'), str, allow_none=False)
+ self._check_default(args, ('openmp_threads', 'mkl_threads', 'numba_threads', 'fftw_threads'),
+ str, allow_none=True)
args.enable_threading = self._convert_bool('enable_threading', args.enable_threading)
if args.enable_threading:
@@ -557,9 +582,12 @@ class HysopArgParser(argparse.ArgumentParser):
else:
args.max_threads = 1
for argname in ('openmp_threads', 'mkl_threads', 'numba_threads', 'fftw_threads'):
- setattr(args, argname, self._convert_threads(argname, getattr(args, argname), default=args.max_threads))
-
- args.numba_threading_layer = self._convert_numba_threading_layer('numba_threading_layer', args.numba_threading_layer)
+ setattr(args, argname, self._convert_threads(argname, getattr(args, argname),
+ default=args.max_threads))
+ args.numba_threading_layer = self._convert_numba_threading_layer('numba_threading_layer',
+ args.numba_threading_layer)
+ args.fftw_planner_effort = self._convert_fftw_planner_effort('fftw_planner_effort',
+ args.fftw_planner_effort)
def _check_opencl_args(self, args):
self._check_default(args, ('cl_platform_id', 'cl_device_id'), int, allow_none=True)
@@ -974,17 +1002,17 @@ class HysopArgParser(argparse.ArgumentParser):
msg='Uknown tracing module \'{}\'.'.format(module)
self.error(msg)
- set_env('ENABLE_THREADING', args.enable_threading)
- set_env('MAX_THREADS', args.max_threads)
- set_env('OMP_NUM_THREADS', args.openmp_threads, False)
- set_env('MKL_NUM_THREADS', args.mkl_threads, False)
- set_env('NUMBA_NUM_THREADS', args.numba_threads, False)
- set_env('NUMBA_THREADING_LAYER', args.numba_threading_layer, False
- set_env('FFTW_NUM_THREADS', args.fftw_threads, False)
- if (args.fftw_planner_effort is not None):
- set_env('FFTW_PLANNER_EFFORT', args.fftw_planner_effort, False)
- if (args.fftw_planner_timelimite is not None):
- set_env('FFTW_PLANNER_TIMELIMIT', args.fftw_planner_timelimite, False)
+ self.set_env('ENABLE_THREADING', args.enable_threading, True)
+ self.set_env('MAX_THREADS', args.max_threads, True)
+ self.set_env('FFTW_NUM_THREADS', args.fftw_threads, True)
+ self.set_env('FFTW_PLANNER_EFFORT', args.fftw_planner_effort, True)
+ self.set_env('FFTW_PLANNER_TIMELIMIT', args.fftw_planner_timelimit, True)
+
+ # those environment variables are not part of HySoP
+ self.set_env('OMP_NUM_THREADS', args.openmp_threads, False)
+ self.set_env('MKL_NUM_THREADS', args.mkl_threads, False)
+ self.set_env('NUMBA_NUM_THREADS', args.numba_threads, False)
+ self.set_env('NUMBA_THREADING_LAYER', args.numba_threading_layer, False)
def _setup_parameters(self, args):
pass
@@ -1054,6 +1082,19 @@ class HysopArgParser(argparse.ArgumentParser):
msg = msg.format(argname, val)
self.error(msg)
return val
+
+ def _convert_fftw_planner_effort(self, argname, val):
+ values = {
+ 'FFTW_ESTIMATE': 'FFTW_ESTIMATE',
+ 'FFTW_MEASURE': 'FFTW_MEASURE',
+ 'FFTW_PATIENT': 'FFTW_PATIENT',
+ 'FFTW_EXHAUSTIVE': 'FFTW_EXHAUSTIVE',
+ 'fftw_estimate': 'FFTW_ESTIMATE',
+ 'fftw_measure': 'FFTW_MEASURE',
+ 'fftw_patient': 'FFTW_PATIENT',
+ 'fftw_exhaustive': 'FFTW_EXHAUSTIVE',
+ }
+ self._check_convert(argname, val, values)
def _convert_numba_threading_layer(self, argname, val):
values = {
diff --git a/examples/particles_above_salt/particles_above_salt_bc.py b/examples/particles_above_salt/particles_above_salt_bc.py
index 456bd7166c0b6ed6b0f744481d32af2badb1147f..9b1cb78b7fa2a9ff2b47ce259c780eb5881066cc 100644
--- a/examples/particles_above_salt/particles_above_salt_bc.py
+++ b/examples/particles_above_salt/particles_above_salt_bc.py
@@ -295,7 +295,7 @@ def compute(args):
compute_mean_fields,
min_max_U, min_max_W,
adapt_dt)
- problem.build()
+ problem.build(args)
# If a visu_rank was provided, and show_graph was set,
# display the graph on the given process rank.
diff --git a/hysop/__init__.py.in b/hysop/__init__.py.in
index 61931fee295e5c13d9ff5a3948186a78b0e0de0d..fa99e48793562d7372ba4e9afd05373752840a56 100644
--- a/hysop/__init__.py.in
+++ b/hysop/__init__.py.in
@@ -38,11 +38,11 @@ __VERBOSE__ = get_env('VERBOSE', ("@VERBOSE@" is "ON"))
__DEBUG__ = get_env('DEBUG', ("@DEBUG@" is "ON"))
__PROFILE__ = get_env('PROFILE', ("@PROFILE@" is "ON"))
-__TRACE_CALLS__ = get_env('TRACE_CALLS', False)
-__TRACE_WARNINGS__ = get_env('TRACE_WARNINGS', False)
+__TRACE_CALLS__ = get_env('TRACE_CALLS', False)
+__TRACE_WARNINGS__ = get_env('TRACE_WARNINGS', False)
__TRACE_MEMALLOCS__ = get_env('TRACE_MEMALLOC', False)
-__TRACE_KERNELS__ = get_env('TRACE_KERNELS', False)
-__KERNEL_DEBUG__ = get_env('KERNEL_DEBUG', False)
+__TRACE_KERNELS__ = get_env('TRACE_KERNELS', False)
+__KERNEL_DEBUG__ = get_env('KERNEL_DEBUG', False)
__BACKTRACE_BIG_MEMALLOCS__ = get_env('BACKTRACE_BIG_MEMALLOCS', False)
__TEST_ALL_OPENCL_PLATFORMS__ = get_env('TEST_ALL_OPENCL_PLATFORMS', False)
@@ -60,7 +60,7 @@ __DEFAULT_NUMBA_TARGET__ = ('parallel' if __ENABLE_THREADING__ else 'cpu')
# FFTW
__FFTW_NUM_THREADS__ = get_env('FFTW_NUM_THREADS', __MAX_THREADS__)
__FFTW_PLANNER_EFFORT__ = get_env('FFTW_PLANNER_EFFORT', 'FFTW_ESTIMATE')
-__FFTW_PLANNER_TIMELIMIT__ = get_env('FFTW_PLANNER_TIMELIMIT', None)
+__FFTW_PLANNER_TIMELIMIT__ = get_env('FFTW_PLANNER_TIMELIMIT', 'FFTW_NO_TIMELIMIT')
# OpenCL
__DEFAULT_PLATFORM_ID__ = int(get_env('DEFAULT_PLATFORM_ID', @OPENCL_DEFAULT_OPENCL_PLATFORM_ID@))
@@ -171,12 +171,20 @@ msg_threads = \
NUMBA_THREADING_LAYER: {}
NUMBA_NUM_THREADS: {}
--------------------------------
+ FFTW_NUM_THREADS: {}
+ FFTW_PLANNER_EFFORT: {}
+ FFTW_PLANNER_TIMELIMIT: {}
+ --------------------------------
'''.format(
- __ENABLE_THREADING__, __MAX_THREADS__,
+ __ENABLE_THREADING__,
+ __MAX_THREADS__,
os.environ['OMP_NUM_THREADS'],
os.environ['MKL_NUM_THREADS'],
__DEFAULT_NUMBA_TARGET__,
os.environ['NUMBA_THREADING_LAYER'],
- os.environ['NUMBA_NUM_THREADS'])
+ os.environ['NUMBA_NUM_THREADS'],
+ __FFTW_NUM_THREADS__,
+ __FFTW_PLANNER_EFFORT__,
+ __FFTW_PLANNER_TIMELIMIT__)
mprint(msg_threads)
diff --git a/hysop/backend/device/opencl/clpeak.py b/hysop/backend/device/opencl/clpeak.py
index 657305c8a9d31b49d957af2f0c4e413de0010720..1cbba49a1b34c66119a67918164e8f8ef94324ae 100644
--- a/hysop/backend/device/opencl/clpeak.py
+++ b/hysop/backend/device/opencl/clpeak.py
@@ -14,7 +14,7 @@ from hysop.backend.hardware.hwinfo import HardwareStatistics
class ClPeakInfo(object):
__FNULL = open(os.devnull, 'w')
- __CMD_TIME_OUT = 30 # 30s timeout for clpeak calls
+ __CMD_TIME_OUT = 60 # 60s timeout for clpeak calls
__clpeak_bandwidth_units = {
'bps': 1e00,
'kbps': 1e03,
diff --git a/hysop/problem.py b/hysop/problem.py
index 094d16b7cd6642d5cfe4baf74a69f3999d5ee101..44269d90ee9cba66e062df76ba7cc29eba098173 100644
--- a/hysop/problem.py
+++ b/hysop/problem.py
@@ -1,5 +1,5 @@
-import datetime
+import datetime, sys
from hysop.constants import Backend, MemoryOrdering
from hysop.tools.string_utils import vprint_banner
from hysop.tools.contexts import Timer
@@ -19,22 +19,43 @@ class Problem(ComputationalGraph):
self.push_nodes(*ops)
@debug
- def build(self, allow_subbuffers=False):
+ def build(self, args=None, allow_subbuffers=False):
with Timer() as tm:
- vprint('\nInitializing problem...')
- self.initialize(outputs_are_inputs=True, topgraph_method=None)
- vprint('\nDiscretizing problem...')
- self.discretize()
- vprint('\nGetting work properties...')
- work = self.get_work_properties()
- vprint('\nAllocating work...')
- work.allocate(allow_subbuffers=allow_subbuffers)
- vprint('\nSetting up problem...')
- self.setup(work)
+ msg = self.build_problem(args=args, allow_subbuffers=allow_subbuffers)
+ if msg:
+ msg='Problem {} achieved, exiting !'.format(msg)
+ vprint_banner(msg)
+ sys.exit(0)
msg=' Problem building took {} ({}s) '
msg=msg.format(datetime.timedelta(seconds=round(tm.interval)),
tm.interval)
vprint_banner(msg, spacing=True, at_border=2)
+
+ def build_problem(self, args, allow_subbuffers):
+ if (args is not None) and args.stop_at_initialization:
+ return 'initialization'
+ vprint('\nInitializing problem...')
+ self.initialize(outputs_are_inputs=True, topgraph_method=None)
+
+ if (args is not None) and args.stop_at_discretization:
+ return 'discretization'
+ vprint('\nDiscretizing problem...')
+ self.discretize()
+
+ if (args is not None) and args.stop_at_work_properties:
+ return 'work properties retrieval'
+ vprint('\nGetting work properties...')
+ work = self.get_work_properties()
+
+ if (args is not None) and args.stop_at_work_allocation:
+ return 'work allocation'
+ vprint('\nAllocating work...')
+ work.allocate(allow_subbuffers=allow_subbuffers)
+
+ if (args is not None) and args.stop_at_setup:
+ return 'setup'
+ vprint('\nSetting up problem...')
+ self.setup(work)
def discretize(self):
super(Problem, self).discretize()