Switch LB to pull scheme (un-optimized)

maintainer/walberla_kernels/generate_lb_kernels.py

@@ -33,7 +33,7 @@
 import lbmpy.enums
 
 import lbmpy_walberla
-import lbmpy_espresso
+import lbmpy_walberla.additional_data_handler
 
 import lees_edwards
 import relaxation_rates
@@ -52,7 +52,7 @@
     target = ps.Target.CPU
 
 # Make sure we have the correct versions of the required dependencies
-for module, requirement in [(ps, "==1.2"), (lbmpy, "==1.2")]:
+for module, requirement in [(ps, "==1.2"), (lbmpy, "==1.3.1")]:
     assert pkg_resources.packaging.specifiers.SpecifierSet(requirement).contains(module.__version__), \
         f"{module.__name__} version {module.__version__} doesn't match requirement {requirement}"
 
@@ -105,14 +105,6 @@ def paramlist(parameters, keys):
         force_model=lbmpy.forcemodels.Schiller(force_field.center_vector)
     )
 
-    # generate stream kernels
-    for params, target_suffix in paramlist(parameters, ("GPU", "CPU", "AVX")):
-        pystencils_espresso.generate_stream_sweep(
-            ctx,
-            method,
-            f"StreamSweep{precision_prefix}{target_suffix}",
-            params)
-
     # generate initial densities
     for params, target_suffix in paramlist(parameters, (default_key,)):
         pystencils_walberla.codegen.generate_sweep(
@@ -121,6 +113,14 @@ def paramlist(parameters, keys):
             pystencils_espresso.generate_setters(ctx, method, params),
             **params)
 
+    # generate velocity field update kernel
+    for params, target_suffix in paramlist(parameters, (default_key,)):
+        pystencils_walberla.codegen.generate_sweep(
+            ctx,
+            f"VelFieldUpdate{precision_prefix}{target_suffix}",
+            pystencils_espresso.generate_vel_field_update(ctx, method, params),
+            **params)
+
     # generate unthermalized Lees-Edwards collision rule
     le_config = lbmpy.LBMConfig(stencil=stencil,
                                 method=lbmpy.Method.TRT,
@@ -129,7 +129,7 @@ def paramlist(parameters, keys):
                                 zero_centered=False,
                                 force_model=lbmpy.ForceModel.GUO,
                                 force=force_field.center_vector,
-                                kernel_type="collide_only")
+                                kernel_type="default_stream_collide")
     lbm_opt = lbmpy.LBMOptimisation(symbolic_field=fields["pdfs"])
     le_collision_rule_unthermalized = lbmpy.create_lb_update_rule(
         lbm_config=le_config,
@@ -138,11 +138,11 @@ def paramlist(parameters, keys):
         config, le_collision_rule_unthermalized,
         fields["pdfs"], stencil, 1)  # shear_dir_normal y
     for params, target_suffix in paramlist(parameters, ("GPU", "CPU", "AVX")):
-        pystencils_espresso.generate_collision_sweep(
+        pystencils_espresso.generate_streaming_collision_sweep(
             ctx,
             le_config,
             le_collision_rule_unthermalized,
-            f"CollideSweep{precision_prefix}LeesEdwards{target_suffix}",
+            f"StreamCollideSweep{precision_prefix}LeesEdwards{target_suffix}",
             params
         )
 
@@ -159,11 +159,11 @@ def paramlist(parameters, keys):
                       "double_precision": ctx.double_accuracy}
     )
     for params, target_suffix in paramlist(parameters, ("GPU", "CPU", "AVX")):
-        pystencils_espresso.generate_collision_sweep(
+        pystencils_espresso.generate_streaming_collision_sweep(
             ctx,
             method,
             collision_rule_thermalized,
-            f"CollideSweep{precision_prefix}Thermalized{target_suffix}",
+            f"StreamCollideSweep{precision_prefix}Thermalized{target_suffix}",
             params
         )
 
@@ -184,17 +184,17 @@ def paramlist(parameters, keys):
         )
 
     # boundary conditions
-    ubb_dynamic = lbmpy_espresso.UBB(
-        lambda *args: None, dim=3, data_type=config.data_type.default_factory())
-    ubb_data_handler = lbmpy_espresso.BounceBackSlipVelocityUBB(
-        method.stencil, ubb_dynamic)
-
+    ubb = lbmpy.boundaries.boundaryconditions.UBB(
+        lambda *args: None, dim=3, data_type=config.data_type.default_factory(), adapt_velocity_to_force=True)
+    ubb_data_handler = lbmpy_walberla.additional_data_handler.default_additional_data_handler(
+        ubb, method, fields["pdfs"], target=target) 
     for _, target_suffix in paramlist(parameters, ("GPU", "CPU")):
         lbmpy_walberla.generate_boundary(
-            ctx, f"Dynamic_UBB_{precision_suffix}{target_suffix}", ubb_dynamic,
+            ctx, f"Dynamic_UBB_{precision_suffix}{target_suffix}", ubb,
             method, additional_data_handler=ubb_data_handler,
-            streaming_pattern="push", target=target)
+            streaming_pattern="pull", target=target)
 
+        # replace Walberla-specific real_t by float/double.
         with open(f"Dynamic_UBB_{precision_suffix}{target_suffix}.h", "r+") as f:
             content = f.read()
             f.seek(0)

maintainer/walberla_kernels/pystencils_espresso.py

@@ -110,7 +110,7 @@ def generate_config(ctx, params):
     return pystencils_walberla.codegen.config_from_context(ctx, **params)
 
 
-def generate_collision_sweep(
+def generate_streaming_collision_sweep(
         ctx, lb_method, collision_rule, class_name, params):
     config = generate_config(ctx, params)
 
@@ -122,7 +122,7 @@ def generate_collision_sweep(
         collision_rule,
         fields['pdfs'],
         fields['pdfs_tmp'],
-        lbmpy.fieldaccess.CollideOnlyInplaceAccessor())
+        lbmpy.fieldaccess.StreamPullTwoFieldsAccessor())
     collide_ast = ps.create_kernel(
         collide_update_rule, config=config, **params)
     collide_ast.function_name = 'kernel_collide'
@@ -131,24 +131,6 @@ def generate_collision_sweep(
         ctx, class_name, collide_ast, **params)
 
 
-def generate_stream_sweep(ctx, lb_method, class_name, params):
-    config = generate_config(ctx, params)
-
-    # Symbols for PDF (twice, due to double buffering)
-    fields = generate_fields(config, lb_method.stencil)
-
-    # Generate stream kernel
-    stream_update_rule = lbmpy.updatekernels.create_stream_pull_with_output_kernel(
-        lb_method, fields['pdfs'], fields['pdfs_tmp'],
-        output={'velocity': fields['velocity']})
-    stream_ast = ps.create_kernel(stream_update_rule, config=config, **params)
-    stream_ast.function_name = 'kernel_stream'
-    stream_ast.assumed_inner_stride_one = True
-    pystencils_walberla.codegen.generate_sweep(
-        ctx, class_name, stream_ast,
-        field_swaps=[(fields['pdfs'], fields['pdfs_tmp'])], **params)
-
-
 def generate_setters(ctx, lb_method, params):
     config = generate_config(ctx, params)
     fields = generate_fields(config, lb_method.stencil)
@@ -160,3 +142,15 @@ def generate_setters(ctx, lb_method, params):
         fields['velocity'].center_vector,
         fields['pdfs'].center_vector)
     return pdfs_setter
+
+
+def generate_vel_field_update(ctx, lb_method, params):
+    config = generate_config(ctx, params)
+    fields = generate_fields(config, lb_method.stencil)
+
+    vel_updater = lbmpy.macroscopic_value_kernels.macroscopic_values_getter(
+        lb_method,
+        None,  # denisty
+        fields['velocity'].center_vector,
+        fields['pdfs'].center_vector)
+    return vel_updater

src/walberla_bridge/src/electrokinetics/generated_kernels/AdvectiveFluxKernel_double_precision.cpp

@@ -18,7 +18,7 @@
 //! \\author lbmpy
 //======================================================================================================================
 
-// kernel generated with pystencils v1.2, lbmpy v1.2, lbmpy_walberla/pystencils_walberla from waLBerla commit ref: a839fac6ef7d0c58e7710e4d50490e9dd7146b4a
+// kernel generated with pystencils v1.2, lbmpy v1.3.1, lbmpy_walberla/pystencils_walberla from waLBerla commit 065ce5f311850371a97ac4766f47dbb5ca8424ba
 
 #include <cmath>
 
@@ -1625,8 +1625,8 @@ static FUNC_PREFIX void advectivefluxkernel_double_precision_advectivefluxkernel
 } // namespace internal_5255e1c780a944d646f270232511968b
 
 void AdvectiveFluxKernel_double_precision::run(IBlock *block) {
-  auto u = block->getData<field::GhostLayerField<double, 3>>(uID);
   auto rho = block->getData<field::GhostLayerField<double, 1>>(rhoID);
+  auto u = block->getData<field::GhostLayerField<double, 3>>(uID);
   auto j = block->getData<field::GhostLayerField<double, 13>>(jID);
 
   WALBERLA_ASSERT_GREATER_EQUAL(-1, -int_c(j->nrOfGhostLayers()));
@@ -1664,8 +1664,8 @@ void AdvectiveFluxKernel_double_precision::runOnCellInterval(const shared_ptr<St
   if (ci.empty())
     return;
 
-  auto u = block->getData<field::GhostLayerField<double, 3>>(uID);
   auto rho = block->getData<field::GhostLayerField<double, 1>>(rhoID);
+  auto u = block->getData<field::GhostLayerField<double, 3>>(uID);
   auto j = block->getData<field::GhostLayerField<double, 13>>(jID);
 
   WALBERLA_ASSERT_GREATER_EQUAL(ci.xMin() - 1, -int_c(j->nrOfGhostLayers()));

src/walberla_bridge/src/electrokinetics/generated_kernels/AdvectiveFluxKernel_double_precision.h

@@ -17,9 +17,9 @@
 //! \\author pystencils
 //======================================================================================================================
 
-// kernel generated with pystencils v1.2, lbmpy v1.2,
-// lbmpy_walberla/pystencils_walberla from waLBerla commit ref:
-// a839fac6ef7d0c58e7710e4d50490e9dd7146b4a
+// kernel generated with pystencils v1.2, lbmpy v1.3.1,
+// lbmpy_walberla/pystencils_walberla from waLBerla commit
+// 065ce5f311850371a97ac4766f47dbb5ca8424ba
 
 #pragma once
 #include "core/DataTypes.h"

src/walberla_bridge/src/electrokinetics/generated_kernels/AdvectiveFluxKernel_single_precision.cpp

@@ -18,7 +18,7 @@
 //! \\author lbmpy
 //======================================================================================================================
 
-// kernel generated with pystencils v1.2, lbmpy v1.2, lbmpy_walberla/pystencils_walberla from waLBerla commit ref: a839fac6ef7d0c58e7710e4d50490e9dd7146b4a
+// kernel generated with pystencils v1.2, lbmpy v1.3.1, lbmpy_walberla/pystencils_walberla from waLBerla commit 065ce5f311850371a97ac4766f47dbb5ca8424ba
 
 #include <cmath>
 
@@ -1626,8 +1626,8 @@ static FUNC_PREFIX void advectivefluxkernel_single_precision_advectivefluxkernel
 
 void AdvectiveFluxKernel_single_precision::run(IBlock *block) {
   auto u = block->getData<field::GhostLayerField<float, 3>>(uID);
-  auto j = block->getData<field::GhostLayerField<float, 13>>(jID);
   auto rho = block->getData<field::GhostLayerField<float, 1>>(rhoID);
+  auto j = block->getData<field::GhostLayerField<float, 13>>(jID);
 
   WALBERLA_ASSERT_GREATER_EQUAL(-1, -int_c(j->nrOfGhostLayers()));
   float *RESTRICT const _data_j = j->dataAt(-1, -1, -1, 0);
@@ -1665,8 +1665,8 @@ void AdvectiveFluxKernel_single_precision::runOnCellInterval(const shared_ptr<St
     return;
 
   auto u = block->getData<field::GhostLayerField<float, 3>>(uID);
-  auto j = block->getData<field::GhostLayerField<float, 13>>(jID);
   auto rho = block->getData<field::GhostLayerField<float, 1>>(rhoID);
+  auto j = block->getData<field::GhostLayerField<float, 13>>(jID);
 
   WALBERLA_ASSERT_GREATER_EQUAL(ci.xMin() - 1, -int_c(j->nrOfGhostLayers()));
   WALBERLA_ASSERT_GREATER_EQUAL(ci.yMin() - 1, -int_c(j->nrOfGhostLayers()));

src/walberla_bridge/src/electrokinetics/generated_kernels/AdvectiveFluxKernel_single_precision.h

@@ -17,9 +17,9 @@
 //! \\author pystencils
 //======================================================================================================================
 
-// kernel generated with pystencils v1.2, lbmpy v1.2,
-// lbmpy_walberla/pystencils_walberla from waLBerla commit ref:
-// a839fac6ef7d0c58e7710e4d50490e9dd7146b4a
+// kernel generated with pystencils v1.2, lbmpy v1.3.1,
+// lbmpy_walberla/pystencils_walberla from waLBerla commit
+// 065ce5f311850371a97ac4766f47dbb5ca8424ba
 
 #pragma once
 #include "core/DataTypes.h"

src/walberla_bridge/src/electrokinetics/generated_kernels/ContinuityKernel_double_precision.cpp

@@ -18,7 +18,7 @@
 //! \\author lbmpy
 //======================================================================================================================
 
-// kernel generated with pystencils v1.2, lbmpy v1.2, lbmpy_walberla/pystencils_walberla from waLBerla commit ref: a839fac6ef7d0c58e7710e4d50490e9dd7146b4a
+// kernel generated with pystencils v1.2, lbmpy v1.3.1, lbmpy_walberla/pystencils_walberla from waLBerla commit 065ce5f311850371a97ac4766f47dbb5ca8424ba
 
 #include <cmath>
 

src/walberla_bridge/src/electrokinetics/generated_kernels/ContinuityKernel_double_precision.h

@@ -17,9 +17,9 @@
 //! \\author pystencils
 //======================================================================================================================
 
-// kernel generated with pystencils v1.2, lbmpy v1.2,
-// lbmpy_walberla/pystencils_walberla from waLBerla commit ref:
-// a839fac6ef7d0c58e7710e4d50490e9dd7146b4a
+// kernel generated with pystencils v1.2, lbmpy v1.3.1,
+// lbmpy_walberla/pystencils_walberla from waLBerla commit
+// 065ce5f311850371a97ac4766f47dbb5ca8424ba
 
 #pragma once
 #include "core/DataTypes.h"

src/walberla_bridge/src/electrokinetics/generated_kernels/ContinuityKernel_single_precision.cpp

@@ -18,7 +18,7 @@
 //! \\author lbmpy
 //======================================================================================================================
 
-// kernel generated with pystencils v1.2, lbmpy v1.2, lbmpy_walberla/pystencils_walberla from waLBerla commit ref: a839fac6ef7d0c58e7710e4d50490e9dd7146b4a
+// kernel generated with pystencils v1.2, lbmpy v1.3.1, lbmpy_walberla/pystencils_walberla from waLBerla commit 065ce5f311850371a97ac4766f47dbb5ca8424ba
 
 #include <cmath>
 
@@ -108,8 +108,8 @@ static FUNC_PREFIX void continuitykernel_single_precision_continuitykernel_singl
 } // namespace internal_990034b4e4dd57d2802b4bcb5f716e46
 
 void ContinuityKernel_single_precision::run(IBlock *block) {
-  auto j = block->getData<field::GhostLayerField<float, 13>>(jID);
   auto rho = block->getData<field::GhostLayerField<float, 1>>(rhoID);
+  auto j = block->getData<field::GhostLayerField<float, 13>>(jID);
 
   WALBERLA_ASSERT_GREATER_EQUAL(-1, -int_c(j->nrOfGhostLayers()));
   float *RESTRICT const _data_j = j->dataAt(-1, -1, -1, 0);
@@ -140,8 +140,8 @@ void ContinuityKernel_single_precision::runOnCellInterval(const shared_ptr<Struc
   if (ci.empty())
     return;
 
-  auto j = block->getData<field::GhostLayerField<float, 13>>(jID);
   auto rho = block->getData<field::GhostLayerField<float, 1>>(rhoID);
+  auto j = block->getData<field::GhostLayerField<float, 13>>(jID);
 
   WALBERLA_ASSERT_GREATER_EQUAL(ci.xMin() - 1, -int_c(j->nrOfGhostLayers()));
   WALBERLA_ASSERT_GREATER_EQUAL(ci.yMin() - 1, -int_c(j->nrOfGhostLayers()));

src/walberla_bridge/src/electrokinetics/generated_kernels/ContinuityKernel_single_precision.h

@@ -17,9 +17,9 @@
 //! \\author pystencils
 //======================================================================================================================
 
-// kernel generated with pystencils v1.2, lbmpy v1.2,
-// lbmpy_walberla/pystencils_walberla from waLBerla commit ref:
-// a839fac6ef7d0c58e7710e4d50490e9dd7146b4a
+// kernel generated with pystencils v1.2, lbmpy v1.3.1,
+// lbmpy_walberla/pystencils_walberla from waLBerla commit
+// 065ce5f311850371a97ac4766f47dbb5ca8424ba
 
 #pragma once
 #include "core/DataTypes.h"