diff --git a/apps/benchmarks/UniformGridGenerated/CMakeLists.txt b/apps/benchmarks/UniformGridGenerated/CMakeLists.txt
index 50d7a1b028c687a3c77fb8c2fee885e564866e2e..368f9dc9e33a11708918bb566c2bfee5bcf3e8b2 100644
--- a/apps/benchmarks/UniformGridGenerated/CMakeLists.txt
+++ b/apps/benchmarks/UniformGridGenerated/CMakeLists.txt
@@ -10,8 +10,8 @@ waLBerla_python_file_generates(UniformGridGenerated.py
GenDefines.h)
-#foreach(config trt smagorinsky mrt3 mrt entropic_kbc_n4 cumulant )
-foreach(config trt trt_nt_split trt_nt trt_split)
+foreach(config trt smagorinsky smagorinsky_d3q27 mrt mrt_d3q27 entropic entropic_kbc_n4 cumulant cumulant_d3q27
+ trt_nt_split trt_nt trt_split trt_aa1 trt_aa2 trt_aa3 trt_aa4 trt_aa1_d3q27 trt_aa2_d3q27 trt_aa3_d3q27 trt_aa4_d3q27 )
waLBerla_add_executable ( NAME UniformGridBenchmarkGenerated_${config}
FILES UniformGridGenerated.cpp Pinning.cpp UniformGridGenerated.py
DEPENDS blockforest boundary core domain_decomposition field geometry timeloop vtk gui
diff --git a/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.py b/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.py
index bf6736d79e9e2a246af2751ba2622252ae8edb23..6b0eed38063234a8aeedb2e7229fec54986f52a1 100644
--- a/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.py
+++ b/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.py
@@ -18,6 +18,7 @@ options_dict = {
'trt': {
'method': 'trt',
'stencil': 'D3Q19',
+ 'maxwellian_moments': False,
'compressible': False,
'relaxation_rate': omega,
'opts': {"two_field_split": False, "two_field_nt_stores": False}
@@ -26,6 +27,7 @@ options_dict = {
'method': 'trt',
'stencil': 'D3Q19',
'compressible': False,
+ 'maxwellian_moments': False,
'relaxation_rate': omega,
'opts': {"two_field_split": True, "two_field_nt_stores": True}
},
@@ -33,6 +35,7 @@ options_dict = {
'method': 'trt',
'stencil': 'D3Q19',
'compressible': False,
+ 'maxwellian_moments': False,
'relaxation_rate': omega,
'opts': {"two_field_split": False, "two_field_nt_stores": True}
},
@@ -40,12 +43,46 @@ options_dict = {
'method': 'trt',
'stencil': 'D3Q19',
'compressible': False,
+ 'maxwellian_moments': False,
'relaxation_rate': omega,
'opts': {"two_field_split": True, "two_field_nt_stores": False}
},
+ 'trt_aa1': {
+ 'method': 'trt',
+ 'stencil': 'D3Q19',
+ 'compressible': False,
+ 'relaxation_rate': omega,
+ 'maxwellian_moments': False,
+ 'opts': {"aa_even_split": True, "aa_odd_split": False}
+ },
+ 'trt_aa2': {
+ 'method': 'trt',
+ 'stencil': 'D3Q19',
+ 'compressible': False,
+ 'relaxation_rate': omega,
+ 'maxwellian_moments': False,
+ 'opts': {"aa_even_split": False, "aa_odd_split": True}
+ },
+ 'trt_aa3': {
+ 'method': 'trt',
+ 'stencil': 'D3Q19',
+ 'compressible': False,
+ 'relaxation_rate': omega,
+ 'maxwellian_moments': False,
+ 'opts': {"aa_even_split": True, "aa_odd_split": True}
+ },
+ 'trt_aa4': {
+ 'method': 'trt',
+ 'stencil': 'D3Q19',
+ 'compressible': False,
+ 'relaxation_rate': omega,
+ 'maxwellian_moments': False,
+ 'opts': {"aa_even_split": False, "aa_odd_split": False}
+ },
'mrt': {
'method': 'mrt',
'stencil': 'D3Q19',
+ 'maxwellian_moments': False,
'relaxation_rates': [omega, 1.3, 1.4, 1.2, 1.1, 1.15, 1.234],
},
'mrt_full': {
@@ -112,10 +149,10 @@ with CodeGeneration() as ctx:
'aa_odd_cse_pdfs': False,
'aa_odd_cse_global': True,
- 'aa_odd_split': True,
+ 'aa_odd_split': False,
'aa_odd_nt_stores': False,
- 'compiled_in_boundaries': False,
+ 'compiled_in_boundaries': True,
}
config_name = ctx.config
noopt = False
@@ -158,10 +195,14 @@ with CodeGeneration() as ctx:
((0, 0, 1), UBB([0.05, 0, 0])),
((0, 0, -1), NoSlip()),
))
- cr_even = create_lb_collision_rule(stencil="D3Q19", compressible=False, optimization={'cse_global': opts['aa_even_cse_global'],
- 'cse_pdfs': opts['aa_even_cse_pdfs']})
- cr_odd = create_lb_collision_rule(stencil="D3Q19", compressible=False, optimization={'cse_global': opts['aa_odd_cse_global'],
- 'cse_pdfs': opts['aa_odd_cse_pdfs']})
+ cr_even = create_lb_collision_rule(**options, optimization={'cse_global': opts['aa_even_cse_global'],
+ 'cse_pdfs': opts['aa_even_cse_pdfs'],
+ # 'split': opts['aa_even_split']
+ })
+ cr_odd = create_lb_collision_rule(**options, optimization={'cse_global': opts['aa_odd_cse_global'],
+ 'cse_pdfs': opts['aa_odd_cse_pdfs'],
+ #'split': opts['aa_odd_split']
+ })
update_rule_aa_even = update_rule_with_push_boundaries(cr_even, pdfs, boundaries, AAEvenTimeStepAccessor, AAOddTimeStepAccessor.read)
update_rule_aa_odd = update_rule_with_push_boundaries(cr_odd, pdfs, boundaries, AAOddTimeStepAccessor, AAEvenTimeStepAccessor.read)
else:
diff --git a/apps/benchmarks/UniformGridGenerated/params.py b/apps/benchmarks/UniformGridGenerated/params.py
index 450e2acdda5a8301244cad7c3f657e4d8a821418..761eae184ee5adcde2a12ebd02b0682a1bbb8279 100644
--- a/apps/benchmarks/UniformGridGenerated/params.py
+++ b/apps/benchmarks/UniformGridGenerated/params.py
@@ -58,7 +58,7 @@ def domain_decomposition_func_z_strong_scaling(processes, threads, block_size):
class BenchmarkScenario:
def __init__(self, block_size=(256, 128, 128), direct_comm=True,
time_step_mode='aa', two_field_kernel_type='generated',
- domain_decomposition_func=domain_decomposition_func_z, pinning="", f_padding=0,
+ domain_decomposition_func=domain_decomposition_func_z, pinning="auto", f_padding=0,
db_file_name='uniform_grid_gen.sqlite'):
self.block_size = block_size
self.direct_comm = direct_comm
@@ -68,13 +68,17 @@ class BenchmarkScenario:
self.threads = int(os.environ['OMP_NUM_THREADS'])
self.processes = wlb.mpi.numProcesses()
self.db_file_name = db_file_name
+
+ if pinning == 'auto':
+ pinning = ",".join([str(i) for i in range(self.threads)])
+
self.pinning = pinning
self.f_padding = f_padding
@wlb.member_callback
def config(self, **kwargs):
- time_steps_for_128_cubed = 10
- time_steps = int(128**3 / prod(self.block_size) * time_steps_for_128_cubed)
+ time_steps_for_128_cubed = 30
+ time_steps = int(128**3 / prod(self.block_size) * time_steps_for_128_cubed * self.threads)
time_steps = max(10, time_steps)
cfg = {
'DomainSetup': {
@@ -183,14 +187,25 @@ def weak_scaling():
def trt_two_field_benchmark():
scenarios = wlb.ScenarioManager()
common = {'block_size': (300, 100, 100),
- 'time_step_mode': 'two_field',
+ 'time_step_mode': 'twoField',
'direct_comm': False,
'domain_decomposition_func': domain_decomposition_func_z_strong_scaling,
+ 'db_file_name': 'trt_two_field_bench.sqlite',
}
scenarios.add(BenchmarkScenario(**common, two_field_kernel_type='manualGeneric'))
scenarios.add(BenchmarkScenario(**common, two_field_kernel_type='manualD3Q19'))
scenarios.add(BenchmarkScenario(**common, two_field_kernel_type='generated'))
+def trt_single_field_benchmark():
+ scenarios = wlb.ScenarioManager()
+ common = {'block_size': (300, 100, 100),
+ 'direct_comm': False,
+ 'domain_decomposition_func': domain_decomposition_func_z_strong_scaling,
+ 'db_file_name': 'trt_single_field_benchmark.sqlite',
+ }
+ for time_step_mode in ['aa', 'aaKernelOnly']:
+ scenarios.add(BenchmarkScenario(**common, time_step_mode=time_step_mode))
+
def padding_test():
scenarios = wlb.ScenarioManager()
@@ -205,6 +220,7 @@ def padding_test():
continue
scenarios.add(sc)
-
+#trt_single_field_benchmark()
+trt_two_field_benchmark()
#single_node_benchmark()
-padding_test()
+#padding_test()