aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--Docs/source/running_cpp/parameters.rst28
-rw-r--r--Examples/Modules/laser_injection_from_file/inputs.2d_test_txye2
-rw-r--r--Examples/Physics_applications/laser_ion/inputs6
-rw-r--r--Examples/Physics_applications/plasma_mirror/inputs_2d2
-rw-r--r--Examples/Tests/reduced_diags/inputs_loadbalancecosts2
-rw-r--r--Python/pywarpx/picmi.py14
-rw-r--r--Source/WarpX.cpp40
-rw-r--r--Tools/LibEnsemble/sim/inputs2
-rw-r--r--Tools/PerformanceTests/automated_test_5_loadimbalance2
9 files changed, 68 insertions, 30 deletions
diff --git a/Docs/source/running_cpp/parameters.rst b/Docs/source/running_cpp/parameters.rst
index 9e38ab133..f50d59d41 100644
--- a/Docs/source/running_cpp/parameters.rst
+++ b/Docs/source/running_cpp/parameters.rst
@@ -207,7 +207,7 @@ Distribution across MPI ranks and parallelization
When using mesh refinement, this number applies to the subdomains
of the coarsest level, but also to any of the finer level.
-* ``warpx.load_balance_intervals`` (`string`) optional (default `0`)
+* ``algo.load_balance_intervals`` (`string`) optional (default `0`)
Using the `Intervals parser`_ syntax, this string defines the timesteps at which
WarpX should try to redistribute the work across MPI ranks, in order to have
better load balancing.
@@ -220,27 +220,35 @@ Distribution across MPI ranks and parallelization
This relies on each MPI rank handling several (in fact many) subdomains
(see ``max_grid_size``).
-* ``warpx.load_balance_with_sfc`` (`0` or `1`) optional (default `0`)
- If this is `1`: use a Space-Filling Curve (SFC) algorithm in order to
- perform load-balancing of the simulation.
- If this is `0`: the Knapsack algorithm is used instead.
-
-* ``warpx.load_balance_efficiency_ratio_threshold`` (`float`) optional (default `1.1`)
+* ``algo.load_balance_efficiency_ratio_threshold`` (`float`) optional (default `1.1`)
Controls whether to adopt a proposed distribution mapping computed during a load balance.
If the the ratio of the proposed to current distribution mapping *efficiency* (i.e.,
average cost per MPI process; efficiency is a number in the range [0, 1]) is greater
than the threshold value, the proposed distribution mapping is adopted. The suggested
- range of values is ``warpx.load_balance_efficiency_ratio_threshold >= 1``, which ensures
+ range of values is ``algo.load_balance_efficiency_ratio_threshold >= 1``, which ensures
that the new distribution mapping is adopted only if doing so would improve the load
balance efficiency. The higher the threshold value, the more conservative is the criterion
for adoption of a proposed distribution; for example, with
- ``warpx.load_balance_efficiency_ratio_threshold = 1``, the proposed distribution is
+ ``algo.load_balance_efficiency_ratio_threshold = 1``, the proposed distribution is
adopted *any* time the proposed distribution improves load balancing; if instead
- ``warpx.load_balance_efficiency_ratio_threshold = 2``, the proposed distribution is
+ ``algo.load_balance_efficiency_ratio_threshold = 2``, the proposed distribution is
adopted only if doing so would yield a 100% to the load balance efficiency (with this
threshold value, if the current efficiency is ``0.45``, the new distribution would only be
adopted if the proposed efficiency were greater than ``0.9``).
+* ``algo.load_balance_with_sfc`` (`0` or `1`) optional (default `0`)
+ If this is `1`: use a Space-Filling Curve (SFC) algorithm in order to
+ perform load-balancing of the simulation.
+ If this is `0`: the Knapsack algorithm is used instead.
+
+* ``algo.load_balance_knapsack_factor`` (`float`) optional (default `1.24`)
+ Controls the maximum number of boxes that can be assigned to a rank during
+ load balance when using the 'knapsack' policy for update of the distribution
+ mapping; the maximum is
+ `load_balance_knapsack_factor*(average number of boxes per rank)`.
+ For example, if there are 4 boxes per rank and `load_balance_knapsack_factor=2`,
+ no more than 8 boxes can be assigned to any rank.
+
* ``algo.load_balance_costs_update`` (`Heuristic` or `Timers`) optional (default `Timers`)
If this is `Heuristic`: load balance costs are updated according to a measure of
particles and cells assigned to each box of the domain. The cost :math:`c` is
diff --git a/Examples/Modules/laser_injection_from_file/inputs.2d_test_txye b/Examples/Modules/laser_injection_from_file/inputs.2d_test_txye
index e28da0180..b332af822 100644
--- a/Examples/Modules/laser_injection_from_file/inputs.2d_test_txye
+++ b/Examples/Modules/laser_injection_from_file/inputs.2d_test_txye
@@ -21,9 +21,9 @@ interpolation.noy = 3
interpolation.noz = 3
warpx.cfl = 0.98
warpx.do_dynamic_scheduling = 0
-warpx.load_balance_intervals = -1
warpx.use_filter = 0
algo.maxwell_solver = ckc
+algo.load_balance_intervals = -1
#################################
############# LASER #############
diff --git a/Examples/Physics_applications/laser_ion/inputs b/Examples/Physics_applications/laser_ion/inputs
index 3012076ff..bac7d5508 100644
--- a/Examples/Physics_applications/laser_ion/inputs
+++ b/Examples/Physics_applications/laser_ion/inputs
@@ -45,9 +45,9 @@ amr.blocking_factor = 64
warpx.do_dynamic_scheduling = 0
# load balancing
-warpx.load_balance_intervals = 10
-warpx.load_balance_with_sfc = 0
-#warpx.load_balance_efficiency_ratio_threshold = 1.1
+algo.load_balance_intervals = 10
+algo.load_balance_with_sfc = 0
+#algo.load_balance_efficiency_ratio_threshold = 1.1
algo.load_balance_costs_update = Heuristic
#algo.costs_heuristic_particles_wt = 0.9
#algo.costs_heuristic_cells_wt = 0.1
diff --git a/Examples/Physics_applications/plasma_mirror/inputs_2d b/Examples/Physics_applications/plasma_mirror/inputs_2d
index 6808ba36c..dfba91ecb 100644
--- a/Examples/Physics_applications/plasma_mirror/inputs_2d
+++ b/Examples/Physics_applications/plasma_mirror/inputs_2d
@@ -27,8 +27,8 @@ my_constants.zp2 = 24.e-6
my_constants.zc2 = 24.05545177444479562e-6
warpx.cfl = 1.0
warpx.do_dynamic_scheduling = 0
-warpx.load_balance_intervals = 66
warpx.use_filter = 1
+algo.load_balance_intervals = 66
#################################
############ PLASMA #############
diff --git a/Examples/Tests/reduced_diags/inputs_loadbalancecosts b/Examples/Tests/reduced_diags/inputs_loadbalancecosts
index 98b87422f..a71c456d0 100644
--- a/Examples/Tests/reduced_diags/inputs_loadbalancecosts
+++ b/Examples/Tests/reduced_diags/inputs_loadbalancecosts
@@ -24,7 +24,7 @@ warpx.use_filter = 1
algo.maxwell_solver = yee # or ckc
# Load balancing
-warpx.load_balance_intervals = 2
+algo.load_balance_intervals = 2
#algo.load_balance_costs_update = Timers
# Interpolation
diff --git a/Python/pywarpx/picmi.py b/Python/pywarpx/picmi.py
index 1c5d9f373..21b84cc34 100644
--- a/Python/pywarpx/picmi.py
+++ b/Python/pywarpx/picmi.py
@@ -666,7 +666,12 @@ class Simulation(picmistandard.PICMI_Simulation):
self.serialize_ics = kw.pop('warpx_serialize_ics', None)
self.do_dynamic_scheduling = kw.pop('warpx_do_dynamic_scheduling', None)
self.load_balance_intervals = kw.pop('warpx_load_balance_intervals', None)
+ self.load_balance_efficiency_ratio_threshold = kw.pop('warpx_load_balance_efficiency_ratio_threshold', None)
self.load_balance_with_sfc = kw.pop('warpx_load_balance_with_sfc', None)
+ self.load_balance_knapsack_factor = kw.pop('warpx_load_balance_knapsack_factor', None)
+ self.load_balance_costs_update = kw.pop('warpx_load_balance_costs_update', None)
+ self.costs_heuristic_particles_wt = kw.pop('warpx_costs_heuristic_particles_wt', None)
+ self.costs_heuristic_cells_wt = kw.pop('warpx_costs_heuristic_cells_wt', None)
self.use_fdtd_nci_corr = kw.pop('warpx_use_fdtd_nci_corr', None)
self.inputs_initialized = False
@@ -690,13 +695,18 @@ class Simulation(picmistandard.PICMI_Simulation):
pywarpx.algo.charge_deposition = self.charge_deposition_algo
pywarpx.algo.field_gathering = self.field_gathering_algo
pywarpx.algo.particle_pusher = self.particle_pusher_algo
+ pywarpx.algo.load_balance_intervals = self.load_balance_intervals
+ pywarpx.algo.load_balance_efficiency_ratio_threshold = self.load_balance_efficiency_ratio_threshold
+ pywarpx.algo.load_balance_with_sfc = self.load_balance_with_sfc
+ pywarpx.algo.load_balance_knapsack_factor = self.load_balance_knapsack_factor
+ pywarpx.algo.load_balance_costs_update = self.load_balance_costs_update
+ pywarpx.algo.costs_heuristic_particles_wt = self.costs_heuristic_particles_wt
+ pywarpx.algo.costs_heuristic_cells_wt = self.costs_heuristic_cells_wt
pywarpx.warpx.use_filter = self.use_filter
pywarpx.warpx.serialize_ics = self.serialize_ics
pywarpx.warpx.do_dynamic_scheduling = self.do_dynamic_scheduling
- pywarpx.warpx.load_balance_intervals = self.load_balance_intervals
- pywarpx.warpx.load_balance_with_sfc = self.load_balance_with_sfc
pywarpx.particles.use_fdtd_nci_corr = self.use_fdtd_nci_corr
diff --git a/Source/WarpX.cpp b/Source/WarpX.cpp
index e6c192834..e688d2b73 100644
--- a/Source/WarpX.cpp
+++ b/Source/WarpX.cpp
@@ -612,14 +612,6 @@ WarpX::ReadParameters ()
fine_tag_hi = RealVect{hi};
}
- std::vector<std::string> load_balance_intervals_string_vec = {"0"};
- pp.queryarr("load_balance_intervals", load_balance_intervals_string_vec);
- load_balance_intervals = IntervalsParser(load_balance_intervals_string_vec);
- pp.query("load_balance_with_sfc", load_balance_with_sfc);
- pp.query("load_balance_knapsack_factor", load_balance_knapsack_factor);
- queryWithParser(pp, "load_balance_efficiency_ratio_threshold",
- load_balance_efficiency_ratio_threshold);
-
pp.query("do_dynamic_scheduling", do_dynamic_scheduling);
pp.query("do_nodal", do_nodal);
@@ -668,11 +660,21 @@ WarpX::ReadParameters ()
// Use same shape factors in all directions, for gathering
galerkin_interpolation = false;
}
- load_balance_costs_update_algo = GetAlgorithmInteger(pp, "load_balance_costs_update");
+
em_solver_medium = GetAlgorithmInteger(pp, "em_solver_medium");
if (em_solver_medium == MediumForEM::Macroscopic ) {
macroscopic_solver_algo = GetAlgorithmInteger(pp,"macroscopic_sigma_method");
}
+
+ // Load balancing parameters
+ std::vector<std::string> load_balance_intervals_string_vec = {"0"};
+ pp.queryarr("load_balance_intervals", load_balance_intervals_string_vec);
+ load_balance_intervals = IntervalsParser(load_balance_intervals_string_vec);
+ pp.query("load_balance_with_sfc", load_balance_with_sfc);
+ pp.query("load_balance_knapsack_factor", load_balance_knapsack_factor);
+ queryWithParser(pp, "load_balance_efficiency_ratio_threshold",
+ load_balance_efficiency_ratio_threshold);
+ load_balance_costs_update_algo = GetAlgorithmInteger(pp, "load_balance_costs_update");
queryWithParser(pp, "costs_heuristic_cells_wt", costs_heuristic_cells_wt);
queryWithParser(pp, "costs_heuristic_particles_wt", costs_heuristic_particles_wt);
}
@@ -895,7 +897,25 @@ WarpX::BackwardCompatibility ()
}
if (ppw.queryarr("load_balance_int", backward_strings)){
amrex::Abort("warpx.load_balance_int is no longer a valid option. "
- "Please use the renamed option warpx.load_balance_intervals instead.");
+ "Please use the renamed option algo.load_balance_intervals instead.");
+ }
+ if (ppw.queryarr("load_balance_intervals", backward_strings)){
+ amrex::Abort("warpx.load_balance_intervals is no longer a valid option. "
+ "Please use the renamed option algo.load_balance_intervals instead.");
+ }
+
+ amrex::Real backward_Real;
+ if (ppw.query("load_balance_efficiency_ratio_threshold", backward_Real)){
+ amrex::Abort("warpx.load_balance_efficiency_ratio_threshold is not supported anymore. "
+ "Please use the renamed option algo.load_balance_efficiency_ratio_threshold.");
+ }
+ if (ppw.query("load_balance_with_sfc", backward_int)){
+ amrex::Abort("warpx.load_balance_with_sfc is not supported anymore. "
+ "Please use the renamed option algo.load_balance_with_sfc.");
+ }
+ if (ppw.query("load_balance_knapsack_factor", backward_Real)){
+ amrex::Abort("warpx.load_balance_knapsack_factor is not supported anymore. "
+ "Please use the renamed option algo.load_balance_knapsack_factor.");
}
if (ppw.queryarr("override_sync_int", backward_strings)){
amrex::Abort("warpx.override_sync_int is no longer a valid option. "
diff --git a/Tools/LibEnsemble/sim/inputs b/Tools/LibEnsemble/sim/inputs
index 098f84922..ccc71128d 100644
--- a/Tools/LibEnsemble/sim/inputs
+++ b/Tools/LibEnsemble/sim/inputs
@@ -31,6 +31,7 @@ algo.charge_deposition = standard
algo.field_gathering = momentum-conserving
algo.particle_pusher = vay
algo.maxwell_solver = ckc
+algo.load_balance_intervals = -1
interpolation.nox = 3
interpolation.noy = 3
interpolation.noz = 3
@@ -39,7 +40,6 @@ warpx.filter_npass_each_dir = 1 4
warpx.cfl = .9999
warpx.do_pml = 0
warpx.do_dynamic_scheduling = 1
-warpx.load_balance_intervals = -1
# Moving window
warpx.do_moving_window = 1
warpx.moving_window_dir = z
diff --git a/Tools/PerformanceTests/automated_test_5_loadimbalance b/Tools/PerformanceTests/automated_test_5_loadimbalance
index ead3cad28..9dff80507 100644
--- a/Tools/PerformanceTests/automated_test_5_loadimbalance
+++ b/Tools/PerformanceTests/automated_test_5_loadimbalance
@@ -15,7 +15,7 @@ geometry.prob_lo = -20.e-6 -20.e-6 -20.e-6 # physical domain
geometry.prob_hi = 20.e-6 20.e-6 20.e-6
warpx.verbose = 1
-warpx.load_balance_intervals = -5
+algo.load_balance_intervals = -5
interpolation.nox = 3
interpolation.noy = 3
interpolation.noz = 3
generated by cgit v1.2.3 (git 2.46.0) at 2025-08-17 11:54:43 +0000