codegen: fix some constant types for vectorization and correct field layout for FluctuatingMRT

apps/benchmarks/UniformGridGPU/UniformGridGPU.py

@@ -157,7 +157,7 @@ with CodeGeneration() as ctx:
 
     # getter & setter
     setter_assignments = macroscopic_values_setter(lb_method, velocity=velocity_field.center_vector,
-                                                   pdfs=pdfs.center_vector, density=1)
+                                                   pdfs=pdfs.center_vector, density=1.0)
     getter_assignments = macroscopic_values_getter(lb_method, velocity=velocity_field.center_vector,
                                                    pdfs=pdfs.center_vector, density=None)
     generate_sweep(ctx, 'UniformGridGPU_MacroSetter', setter_assignments)

apps/benchmarks/UniformGridGPU/UniformGridGPU_AA.py

@@ -106,7 +106,7 @@ with CodeGeneration() as ctx:
 
     # getter & setter
     setter_assignments = macroscopic_values_setter(lb_method, velocity=velocity_field.center_vector,
-                                                   pdfs=pdfs.center_vector, density=1)
+                                                   pdfs=pdfs.center_vector, density=1.0)
     getter_assignments = macroscopic_values_getter(lb_method, velocity=velocity_field.center_vector,
                                                    pdfs=pdfs.center_vector, density=None)
     generate_sweep(ctx, 'UniformGridGPU_AA_MacroSetter', setter_assignments)

apps/benchmarks/UniformGridGenerated/UniformGridGenerated.py

@@ -26,7 +26,7 @@ options_dict = {
     'mrt': {
         'method': 'mrt',
         'stencil': 'D3Q19',
-        'relaxation_rates': [0, omega, 1.3, 1.4, omega, 1.2, 1.1, 1.15, 1.234, 1.4235, 1.242, 1.2567, 0.9, 0.7],
+        'relaxation_rates': [0.0, omega, 1.3, 1.4, omega, 1.2, 1.1, 1.15, 1.234, 1.4235, 1.242, 1.2567, 0.9, 0.7],
     },
     'mrt_full': {
         'method': 'mrt',
@@ -171,7 +171,7 @@ with CodeGeneration() as ctx:
                    cpu_openmp=openmp_enabled, ghost_layers=1)
 
     setter_assignments = macroscopic_values_setter(update_rule_two_field.method, velocity=velocity_field.center_vector,
-                                                   pdfs=pdfs.center_vector, density=1)
+                                                   pdfs=pdfs.center_vector, density=1.0)
     getter_assignments = macroscopic_values_getter(update_rule_two_field.method, velocity=velocity_field.center_vector,
                                                    pdfs=pdfs.center_vector, density=None)
     generate_sweep(ctx, 'GenMacroSetter', setter_assignments, cpu_openmp=openmp_enabled)

apps/showcases/PhaseFieldAllenCahn/CPU/multiphase_codegen.py

@@ -17,6 +17,7 @@ from lbmpy.phasefield_allen_cahn.force_model import MultiphaseForceModel
 
 import numpy as np
 import sympy as sp
+import pystencils as ps
 
 stencil_phase = get_stencil("D3Q19")
 stencil_hydro = get_stencil("D3Q27")
@@ -125,8 +126,9 @@ phase_field_LB_step = create_lb_update_rule(lb_method=method_phase,
                                             kernel_type='stream_pull_collide')
 
 phase_field_LB_step.set_main_assignments_from_dict({**phase_field_LB_step.main_assignments_dict, **{C.center: sum_h}})
+subexp = [ps.Assignment(a.lhs, float(a.rhs)) if a.rhs == 0 else a for a in phase_field_LB_step.subexpressions]
 phase_field_LB_step = AssignmentCollection(main_assignments=phase_field_LB_step.main_assignments,
-                                           subexpressions=phase_field_LB_step.subexpressions)
+                                           subexpressions=subexp)
 phase_field_LB_step = sympy_cse(phase_field_LB_step)
 
 # ---------------------------------------------------------------------------------------------------------

tests/lbm/codegen/FluctuatingMRT.cpp

@@ -68,10 +68,10 @@ int main( int argc, char ** argv )
    const double remainingTimeLoggerFrequency = parameters.getParameter< double >( "remainingTimeLoggerFrequency", 3.0 ); // in seconds
 
    // create fields
-   BlockDataID forceFieldId = field::addToStorage<VectorField_T>( blocks, "Force", real_t( 0.0 ));
+   BlockDataID forceFieldId = field::addToStorage<VectorField_T>( blocks, "Force", real_t( 0.0 ), field::fzyx);
 
    LatticeModel_T latticeModel = LatticeModel_T( forceFieldId, omega, 0, omega, omega, seed, temperature, uint_t(0) );
-   BlockDataID pdfFieldId = lbm::addPdfFieldToStorage( blocks, "pdf field", latticeModel, initialVelocity, real_t(1) );
+   BlockDataID pdfFieldId = lbm::addPdfFieldToStorage( blocks, "pdf field", latticeModel, initialVelocity, real_t(1), uint_t(1), field::fzyx );
    BlockDataID flagFieldId = field::addFlagFieldToStorage< FlagField_T >( blocks, "flag field" );
 
    // create and initialize flag field

tests/lbm/codegen/FluctuatingMRT.py

@@ -9,7 +9,7 @@ from lbmpy_walberla import generate_lattice_model
 with CodeGeneration() as ctx:
     omega_shear = sp.symbols("omega_shear")
     temperature = sp.symbols("temperature")
-    force_field, vel_field = ps.fields("force(3), velocity(3): [3D]", layout='zyxf')
+    force_field, vel_field = ps.fields("force(3), velocity(3): [3D]", layout='fzyx')
 
     def rr_getter(moment_group):
         is_shear = [is_shear_moment(m, 3) for m in moment_group]
@@ -49,4 +49,7 @@ with CodeGeneration() as ctx:
         optimization={'cse_global': True}
     )
 
-    generate_lattice_model(ctx, 'FluctuatingMRT_LatticeModel', collision_rule)
+    params = {}
+    if ctx.optimize_for_localhost:
+        params['cpu_vectorize_info'] = {'assume_inner_stride_one': True, 'assume_aligned': True}
+    generate_lattice_model(ctx, 'FluctuatingMRT_LatticeModel', collision_rule, field_layout='fzyx', **params)

tests/lbm/codegen/GeneratedOutflowBC.py

@@ -38,7 +38,7 @@ method = create_lb_method(**options)
 
 # getter & setter
 setter_assignments = macroscopic_values_setter(method, velocity=velocity_field.center_vector,
-                                               pdfs=pdfs, density=1)
+                                               pdfs=pdfs, density=1.0)
 
 # opt = {'instruction_set': 'sse', 'assume_aligned': True, 'nontemporal': False, 'assume_inner_stride_one': True}