diff --git a/src/pmill-768.py b/src/PMillennium768MultiNodeWithVtuneAndAllinea.py
similarity index 58%
rename from src/pmill-768.py
rename to src/PMillennium768MultiNodeWithVtuneAndAllinea.py
index e0f732c42c97974cdcc8949dd324577edb94b9f1..056df19377131181b5e4a140d95b187d7b543960 100644
--- a/src/pmill-768.py
+++ b/src/PMillennium768MultiNodeWithVtuneAndAllinea.py
@@ -3,18 +3,19 @@ import reframe.core.launchers.mpi
import common
-
-@rfm.parameterized_test(*([tasks, threads, iteration, launcher]
- for tasks in [2]
- for threads in list(reversed([56]))
+# 4 ranks and 64 threads were found to be an optimal way to use a
+# node on DINE for the pmill 768 benchmark
+@rfm.parameterized_test(*([tasks, tasks_per_node, threads, iteration, launcher]
+ for tasks in [16]
+ for tasks_per_node in [4]
+ for threads in list(reversed([64]))
for iteration in [1]
- for launcher in ['mpirun','perf-report']))
-class PMillenniumTest(rfm.RegressionTest):
- def __init__(self, num_tasks, num_threads, iteration, launcher):
- ic_dir = common.setup(self, launcher, num_tasks, num_tasks_per_node=num_tasks)
-
+ for launcher in ['profiler:Vtune','profiler:Allinea']))
+class PMillennium768MultiNode(rfm.RegressionTest):
+ def __init__(self, num_tasks, tasks_per_node, num_threads, iteration, launcher):
+ ic_dir = common.setup(self, launcher, num_tasks, tasks_per_node)
+
test_dir = 'swiftsim/examples/PMillennium/PMillennium-768'
- self.cpus_per_task = 32 # added
print(f'Running with threads : <{num_threads}>')
self.keep_files = [test_dir]
@@ -24,23 +25,22 @@ class PMillenniumTest(rfm.RegressionTest):
'--with-parmetis'
]
self.prerun_cmds = [f'pushd {test_dir}']
- num_iterations = 5
-
+ num_steps = 10
+ self.time_limit = '3h'
- if launcher == 'mpirun':
+ if launcher == 'profiler:Vtune':
self.executable = 'aps'
self.executable_opts = ['--collection-mode=mpi,omp', '../../swift_mpi']
- elif launcher == 'perf-report':
+ elif launcher == 'profiler:Allinea':
self.executable = 'perf-report'
self.executable_opts = ['--mpi=intel-mpi', '../../swift_mpi']
- self.time_limit = '2h40m'
self.executable_opts += [
'--cosmology',
'--self-gravity',
'-v', '1',
f'--threads={num_threads}',
- '-n', f'{num_iterations}',
+ '-n', f'{num_steps}',
'-P', 'Restarts:enable:0',
f'-PInitialConditions:file_name:{ic_dir}/pmillenium/PMill-768.hdf5',
'p-mill-768.yml'
diff --git a/src/common.py b/src/common.py
index 2ccd53e8f39170044a21a554ae87a7807618c20e..9bd875a3ff5f6026cbefd6eca4cd335dc0540c0f 100644
--- a/src/common.py
+++ b/src/common.py
@@ -2,11 +2,10 @@ import reframe as rfm
import reframe.utility.sanity as sn
-
def setup(test, launcher, num_tasks, num_tasks_per_node=1):
test.time_limit = '1h'
- test.num_tasks = 1 # num_tasks
- test.num_tasks_per_node = 1
+ test.num_tasks = num_tasks
+ test.num_tasks_per_node = num_tasks_per_node
test.valid_prog_environs = ['*']
@@ -28,26 +27,31 @@ def setup(test, launcher, num_tasks, num_tasks_per_node=1):
if rfm.utility.osext.osuser() == 'dc-turn5':
if test.current_system.name == 'cosma7':
ic_dir = '/cosma7/data/ds007/dc-turn5/swift_initial_conditions'
- if launcher == 'mpirun':
+ if launcher == 'profiler:Vtune':
test.valid_systems = ['cosma7:cpu_multi_node_aps']
- elif launcher == 'perf-report':
+ elif launcher == 'profiler:Allinea':
test.valid_systems = ['cosma7:cpu_multi_node_perf_report']
elif test.current_system.name == 'dine':
ic_dir = '/cosma5/data/durham/dc-turn5/swift_initial_conditions'
- if launcher == 'mpirun':
+ if launcher == 'profiler:Vtune':
test.valid_systems = ['dine:cpu_multi_node_aps']
- elif launcher == 'perf-report':
+ elif launcher == 'profiler:Allinea':
test.valid_systems = ['dine:cpu_multi_node_perf_report']
- elif launcher == 'scalasca':
+ elif launcher == 'profiler:ScoreP':
+ # ScoreP only works on with non-mpi version of SWIFT
+ # because MPI_THREAD_MULTIPLE (needed by SWIFT) is not supported
test.valid_systems = ['dine:cpu_single_node']
else:
raise ValueError(f'Need to handle {test.current_system.name} for {rfm.utility.osext.osuser()}')
+ ''' TODO: ADD FILE PATHS
elif rfm.utility.osext.osuser() == 'dc-fraw1':
if test.current_system.name == 'cosma7' or test.current_system.name == 'dine':
ic_dir = '/cosma5/data/do008/dc-fraw1/swift_initial_conditions'
else:
raise ValueError(f'Need to handle {test.current_system.name} for {rfm.utility.osext.osuser()}')
+ '''
else:
raise ValueError(f'Need to handle {test.current_system.name} for {rfm.utility.osext.osuser()}')
+
return ic_dir
diff --git a/src/eagle_6_scorep.py b/src/eagle6SingleNodeWithScoreP.py
similarity index 100%
rename from src/eagle_6_scorep.py
rename to src/eagle6SingleNodeWithScoreP.py
diff --git a/src/sodshock_3d.py b/src/sodshock3dMultiNodeWithVtuneAndAllinea.py
similarity index 59%
rename from src/sodshock_3d.py
rename to src/sodshock3dMultiNodeWithVtuneAndAllinea.py
index 73fd08972cf81165f7f3466cbb1b16b5d7123112..ab9564a0f0fb7d2b9b1553b397475ee219816ab3 100644
--- a/src/sodshock_3d.py
+++ b/src/sodshock3dMultiNodeWithVtuneAndAllinea.py
@@ -3,18 +3,19 @@ import reframe.core.launchers.mpi
import common
-
-@rfm.parameterized_test(*([tasks, threads, iteration, launcher]
- for tasks in [2]
- for threads in list(reversed([32]))
+# 4 ranks and 64 threads were found to be an optimal way to use a
+# node on DINE for the pmill 768 benchmark
+@rfm.parameterized_test(*([tasks, tasks_per_node, threads, iteration, launcher]
+ for tasks in [16]
+ for tasks_per_node in [4]
+ for threads in list(reversed([64]))
for iteration in [1]
- for launcher in ['mpirun', 'perf-report', 'scalasca']))
-class SodShock3dTest(rfm.RegressionTest):
- def __init__(self, num_tasks, num_threads, iteration, launcher):
- ic_dir = common.setup(self, launcher, num_tasks, num_tasks_per_node=num_tasks)
+ for launcher in ['profiler:Vtune', 'profiler:Allinea']))
+class SodShock3dTestMultiNode(rfm.RegressionTest):
+ def __init__(self, num_tasks, tasks_per_node, num_threads, iteration, launcher):
+ ic_dir = common.setup(self, launcher, num_tasks, tasks_per_node)
test_dir = 'swiftsim/examples/HydroTests/SodShock_3D'
- self.cpus_per_task = 32 # added
print(f'Running with threads : <{num_threads}>')
self.keep_files = [test_dir]
@@ -26,18 +27,14 @@ class SodShock3dTest(rfm.RegressionTest):
self.prerun_cmds = [f'pushd {test_dir}']
num_iterations = 10000
- if launcher == 'mpirun':
- self.executable = 'aps'
+ if launcher == 'profiler:Vtune':
+ self.executable = 'aps'
self.executable_opts = ['--collection-mode=mpi,omp', '../../swift_mpi',
'-P', 'Snapshots:time_first:999999999']
- elif launcher == 'perf-report':
+ elif launcher == 'profiler:Allinea':
self.executable = 'perf-report'
self.executable_opts = ['--mpi=intel-mpi', '../../swift_mpi']
self.time_limit = '2h40m'
- elif launcher == 'scalasca':
- self.build_system.make_opts = ['CC="scorep --user --thread=pthread icc"']
- self.executable = '../../swift'
- self.executable_opts = ['-P', 'Snapshots:time_first:999999999']
self.executable_opts += [
'--hydro',
diff --git a/src/sodshock_scalasca.py b/src/sodshock3dSingleNodeWithScoreP.py
similarity index 63%
rename from src/sodshock_scalasca.py
rename to src/sodshock3dSingleNodeWithScoreP.py
index df4d1e14c92fe582f082da780cff6c51a0fc3fe0..21d62b2ec45cdaffb55610eed9e4a2a35e3e1121 100644
--- a/src/sodshock_scalasca.py
+++ b/src/sodshock3dSingleNodeWithScoreP.py
@@ -3,42 +3,38 @@ import reframe.core.launchers.mpi
import common
-
+# ScoreP is only compatible with SWIFT if we with no MPI
@rfm.parameterized_test(*([tasks, threads, iteration, launcher]
- for tasks in [2]
- for threads in list(reversed([32]))
+ for tasks in [1]
+ for threads in list(reversed([64]))
for iteration in [1]
- for launcher in ['scalasca']))
-class SodShock3dTest(rfm.RegressionTest):
+ for launcher in ['profiler:ScoreP']))
+class SodShock3dTestSingleNode(rfm.RegressionTest):
def __init__(self, num_tasks, num_threads, iteration, launcher):
ic_dir = common.setup(self, launcher, num_tasks, num_tasks_per_node=num_tasks)
test_dir = 'swiftsim/examples/HydroTests/SodShock_3D'
- # self.cpus_per_task = 32 # added
print(f'Running with threads : <{num_threads}>')
self.keep_files = [test_dir]
self.build_system.config_opts = [
- '--disable-ipo', #temporary disable - support is forthcoming
+ '--disable-ipo', #temporary disable - ScoreP support is forthcoming
'--with-tbbmalloc',
'--with-parmetis'
]
self.prerun_cmds = [f'pushd {test_dir}']
- num_iterations = 1
+ num_steps = 10000
- # self.build_system.options = ['CC=scorep mpicc', 'CFLAGS=-fopenmp'] #Added for scalasca
- #self.build_system.make_opts = ['CC="scorep --user --thread=pthread mpicc"'] #change mpicc to icc for non-mpi
self.build_system.make_opts = ['CC="scorep --user --thread=pthread icc"']
- if launcher == 'scalasca':
- self.executable = '../../swift'
+ self.executable = '../../swift'
self.executable_opts += [
'-P', 'Snapshots:time_first:999999999',
'--hydro',
'-v', '1',
f'--threads={num_threads}',
- '-n', f'{num_iterations}',
+ '-n', f'{num_steps}',
'-P', 'Restarts:enable:0',
'-P', f'InitialConditions:file_name:{ic_dir}/sodshock/glassCube_64.hdf5',
'sodShock.yml'