diff --git a/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.cpp b/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.cpp
index bcb87d151db8a169f86757acdaf9d68486fe560e..b4275d395532f9e90a985ccec1986be1b44d7154 100644
--- a/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.cpp
+++ b/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.cpp
@@ -155,7 +155,7 @@ int main( int argc, char **argv )
auto remainingTimeLoggerFrequency = parameters.getParameter< double >( "remainingTimeLoggerFrequency", -1.0 ); // in seconds
if (remainingTimeLoggerFrequency > 0) {
- auto logger = timing::RemainingTimeLogger( timeLoop.getNrOfTimeSteps() * outerIterations, remainingTimeLoggerFrequency );
+ auto logger = timing::RemainingTimeLogger( timeLoop.getNrOfTimeSteps() * uint_c( outerIterations ), remainingTimeLoggerFrequency );
timeLoop.addFuncAfterTimeStep( logger, "remaining time logger" );
}
diff --git a/python/lbmpy_walberla/templates/LatticeModel.tmpl.cpp b/python/lbmpy_walberla/templates/LatticeModel.tmpl.cpp
index 2b2065cc7600e4d230d7d21f801e8e1044bb7701..b10bc3948026ad1b46b0dfd0db8eba1c86ecdfef 100644
--- a/python/lbmpy_walberla/templates/LatticeModel.tmpl.cpp
+++ b/python/lbmpy_walberla/templates/LatticeModel.tmpl.cpp
@@ -66,7 +66,7 @@ void {{class_name}}::Sweep::streamCollide( IBlock * block, const uint_t numberOf
{{stream_collide_kernel|generate_block_data_to_field_extraction(parameters=['pdfs', 'pdfs_tmp'])|indent(4)}}
auto & lm = dynamic_cast< lbm::PdfField<{{class_name}}> * > (pdfs)->latticeModel();
- WALBERLA_ASSERT_EQUAL( *(lm.blockId), block->getId() );
+ WALBERLA_ASSERT_EQUAL( *(lm.blockId_), block->getId() );
{{stream_collide_kernel|generate_refs_for_kernel_parameters(prefix='lm.', parameters_to_ignore=['pdfs', 'pdfs_tmp'])|indent(4) }}
{{stream_collide_kernel|generate_call('cell_idx_c(numberOfGhostLayersToInclude)')|indent(4)}}
@@ -78,7 +78,7 @@ void {{class_name}}::Sweep::collide( IBlock * block, const uint_t numberOfGhostL
{{collide_kernel|generate_block_data_to_field_extraction(parameters=['pdfs'])|indent(4)}}
auto & lm = dynamic_cast< lbm::PdfField<{{class_name}}> * > (pdfs)->latticeModel();
- WALBERLA_ASSERT_EQUAL( *(lm.blockId), block->getId() );
+ WALBERLA_ASSERT_EQUAL( *(lm.blockId_), block->getId() );
{{collide_kernel|generate_refs_for_kernel_parameters(prefix='lm.', parameters_to_ignore=['pdfs', 'pdfs_tmp'])|indent(4) }}
{{collide_kernel|generate_call('cell_idx_c(numberOfGhostLayersToInclude)')|indent(4)}}
diff --git a/python/lbmpy_walberla/templates/LatticeModel.tmpl.h b/python/lbmpy_walberla/templates/LatticeModel.tmpl.h
index be160e99e89d596f86511a97db314e35aeaaca18..df188fa32081427ae5ea4a05e81f7fe5b63ea1d4 100644
--- a/python/lbmpy_walberla/templates/LatticeModel.tmpl.h
+++ b/python/lbmpy_walberla/templates/LatticeModel.tmpl.h
@@ -134,17 +134,17 @@ public:
private:
void configureBlock(IBlock * block, StructuredBlockStorage * storage)
{
- {{stream_collide_kernel|generate_block_data_to_field_extraction(lmIgnores+lmOffsets, no_declarations=True)|indent(8)}}
+ {{stream_collide_kernel|generate_block_data_to_field_extraction(lmIgnores+lmOffsets, no_declarations=True, update_member=True)|indent(8)}}
{% if need_block_offsets[0] -%}
- block_offset_0 = uint32_c(storage->getBlockCellBB(*block).xMin());
+ block_offset_0_ = uint32_c(storage->getBlockCellBB(*block).xMin());
{% endif -%}
{%- if need_block_offsets[1] -%}
- block_offset_1 = uint32_c(storage->getBlockCellBB(*block).yMin());
+ block_offset_1_ = uint32_c(storage->getBlockCellBB(*block).yMin());
{% endif -%}
{%- if need_block_offsets[2] -%}
- block_offset_2 = uint32_c(storage->getBlockCellBB(*block).zMin());
+ block_offset_2_ = uint32_c(storage->getBlockCellBB(*block).zMin());
{% endif %}
- blockId = &block->getId();
+ blockId_ = &block->getId();
{% if refinement_scaling_info -%}
const uint_t targetLevel = block->getBlockStorage().getLevel(*block);
@@ -182,7 +182,7 @@ private:
// Updated by configureBlock:
{{stream_collide_kernel|generate_block_data_to_field_extraction(lmIgnores, declarations_only=True)|indent(4)}}
- const IBlockID * blockId;
+ const IBlockID * blockId_;
uint_t currentLevel;
// Backend classes can access private members:
diff --git a/python/lbmpy_walberla/walberla_lbm_generation.py b/python/lbmpy_walberla/walberla_lbm_generation.py
index 05e98a96fe31c1fd001c400f696b4e7c143a762c..7e32867b05b1d8b9bd15b4da09b5c4d20d3cb3cd 100644
--- a/python/lbmpy_walberla/walberla_lbm_generation.py
+++ b/python/lbmpy_walberla/walberla_lbm_generation.py
@@ -68,6 +68,15 @@ def __lattice_model(generation_context, class_name, lb_method, stream_collide_as
if refinement_scaling:
refinement_scaling_info = [ (e0,e1,expression_to_code(e2, '', dtype=dtype)) for e0,e1,e2 in refinement_scaling.scaling_info ]
+ # append '_' to entries since they are used as members
+ for i in range(len(refinement_scaling_info)):
+ updated_entry = (refinement_scaling_info[i][0],
+ refinement_scaling_info[i][1].replace(refinement_scaling_info[i][1],
+ refinement_scaling_info[i][1] + '_'),
+ refinement_scaling_info[i][2].replace(refinement_scaling_info[i][1],
+ refinement_scaling_info[i][1] + '_'),
+ )
+ refinement_scaling_info[i] = updated_entry
else:
refinement_scaling_info = None
@@ -147,8 +156,6 @@ def generate_lattice_model(generation_context, class_name, collision_rule, refin
refinement_scaling)
-
-
class RefinementScaling:
level_scale_factor = sp.Symbol("level_scale_factor")
@@ -217,18 +224,27 @@ def stencil_switch_statement(stencil, values):
def field_and_symbol_substitute(expr, variable_prefix="lm.", variables_without_prefix=[]):
variables_without_prefix = [v.name if isinstance(v, sp.Symbol) else v for v in variables_without_prefix]
substitutions = {}
+ # check for member access
+ if variable_prefix.endswith('.'):
+ postfix = '_'
+ else:
+ postfix = ''
for sym in expr.atoms(sp.Symbol):
if isinstance(sym, Field.Access):
fa = sym
prefix = "" if fa.field.name in variables_without_prefix else variable_prefix
+ if prefix.endswith('.'):
+ postfix2 = '_'
+ else:
+ postfix2 = ''
if fa.field.index_dimensions == 0:
- substitutions[fa] = sp.Symbol("%s%s->get(x,y,z)" % (prefix, fa.field.name))
+ substitutions[fa] = sp.Symbol("%s%s->get(x,y,z)" % (prefix, fa.field.name + postfix2))
else:
assert fa.field.index_dimensions == 1, "walberla supports only 0 or 1 index dimensions"
- substitutions[fa] = sp.Symbol("%s%s->get(x,y,z,%s)" % (prefix, fa.field.name, fa.index[0]))
+ substitutions[fa] = sp.Symbol("%s%s->get(x,y,z,%s)" % (prefix, fa.field.name + postfix2, fa.index[0]))
else:
if sym.name not in variables_without_prefix:
- substitutions[sym] = sp.Symbol(variable_prefix + sym.name)
+ substitutions[sym] = sp.Symbol(variable_prefix + sym.name + postfix)
return expr.subs(substitutions)
@@ -244,10 +260,19 @@ def expression_to_code(expr, variable_prefix="lm.", variables_without_prefix=[],
"""
return cpp_printer.doprint(type_expr(field_and_symbol_substitute(expr, variable_prefix, variables_without_prefix),dtype=dtype))
+
def type_expr(eq, dtype):
- eq=type_all_numbers(eq,dtype=dtype)
+ def recurse(expr):
+ for i in range(len(expr.args)):
+ if expr.args[i] == sp.Rational or expr.args[i] == sp.Float:
+ expr.args[i] = type_all_numbers(expr.args[i], dtype=dtype)
+ else:
+ recurse(expr.args[i])
+
+ recurse(eq)
return eq.subs({s: TypedSymbol(s.name, dtype) for s in eq.atoms(sp.Symbol)})
+
def equations_to_code(equations, variable_prefix="lm.", variables_without_prefix=[], dtype="double"):
if isinstance(equations, AssignmentCollection):
diff --git a/python/pystencils_walberla/jinja_filters.py b/python/pystencils_walberla/jinja_filters.py
index 174b9dfecbaf6bd0c1df4d391eca5cb1a5e7c7c3..eb209237bfdf54a64bee209c8d1a50e9d122e06f 100644
--- a/python/pystencils_walberla/jinja_filters.py
+++ b/python/pystencils_walberla/jinja_filters.py
@@ -1,5 +1,6 @@
import jinja2
import sympy as sp
+import re
from pystencils import TypedSymbol
from pystencils.backends.cbackend import generate_c
@@ -90,7 +91,7 @@ def generate_definition(kernel_info, target='cpu'):
def field_extraction_code(field, is_temporary, declaration_only=False,
- no_declaration=False, is_gpu=False):
+ no_declaration=False, is_gpu=False, update_member=False):
"""Returns code string for getting a field pointer.
This can happen in two ways: either the field is extracted from a walberla block, or a temporary field to swap is
@@ -102,9 +103,10 @@ def field_extraction_code(field, is_temporary, declaration_only=False,
declaration_only: only create declaration instead of the full code
no_declaration: create the extraction code, and assume that declarations are elsewhere
is_gpu: if the field is a GhostLayerField or a GpuField
+ update_member: specify if function is used inside a constructor; add _ to members
"""
- # Determine size of f coordinate which is a template parameter
+# Determine size of f coordinate which is a template parameter
f_size = get_field_fsize(field)
field_name = field.name
dtype = get_base_type(field.dtype)
@@ -114,15 +116,18 @@ def field_extraction_code(field, is_temporary, declaration_only=False,
dtype = get_base_type(field.dtype)
field_type = make_field_type(dtype, f_size, is_gpu)
if declaration_only:
- return "%s * %s;" % (field_type, field_name)
+ return "%s * %s_;" % (field_type, field_name)
else:
prefix = "" if no_declaration else "auto "
- return "%s%s = block->uncheckedFastGetData< %s >(%sID);" % (prefix, field_name, field_type, field_name)
+ if update_member:
+ return "%s%s_ = block->uncheckedFastGetData< %s >(%sID);" % (prefix, field_name, field_type, field_name)
+ else:
+ return "%s%s = block->uncheckedFastGetData< %s >(%sID);" % (prefix, field_name, field_type, field_name)
else:
assert field_name.endswith('_tmp')
original_field_name = field_name[:-len('_tmp')]
if declaration_only:
- return "%s * %s;" % (field_type, field_name)
+ return "%s * %s_;" % (field_type, field_name)
else:
declaration = "{type} * {tmp_field_name};".format(type=field_type, tmp_field_name=field_name)
tmp_field_str = temporary_fieldTemplate.format(original_field_name=original_field_name,
@@ -132,7 +137,7 @@ def field_extraction_code(field, is_temporary, declaration_only=False,
@jinja2.contextfilter
def generate_block_data_to_field_extraction(ctx, kernel_info, parameters_to_ignore=(), parameters=None,
- declarations_only=False, no_declarations=False):
+ declarations_only=False, no_declarations=False, update_member=False):
"""Generates code that extracts all required fields of a kernel from a walberla block storage."""
if parameters is not None:
assert parameters_to_ignore == ()
@@ -154,8 +159,12 @@ def generate_block_data_to_field_extraction(ctx, kernel_info, parameters_to_igno
'no_declaration': no_declarations,
'is_gpu': ctx['target'] == 'gpu',
}
- result = "\n".join(field_extraction_code(field=field, is_temporary=False, **args) for field in normal_fields) + "\n"
- result += "\n".join(field_extraction_code(field=field, is_temporary=True, **args) for field in temporary_fields)
+ result = "\n".join(
+ field_extraction_code(field=field, is_temporary=False, update_member=update_member, **args) for field in
+ normal_fields) + "\n"
+ result += "\n".join(
+ field_extraction_code(field=field, is_temporary=True, update_member=update_member, **args) for field in
+ temporary_fields)
return result
@@ -163,7 +172,7 @@ def generate_refs_for_kernel_parameters(kernel_info, prefix, parameters_to_ignor
symbols = {p.field_name for p in kernel_info.parameters if p.is_field_pointer}
symbols.update(p.symbol.name for p in kernel_info.parameters if not p.is_field_parameter)
symbols.difference_update(parameters_to_ignore)
- return "\n".join("auto & %s = %s%s;" % (s, prefix, s) for s in symbols)
+ return "\n".join("auto & %s = %s%s_;" % (s, prefix, s) for s in symbols)
@jinja2.contextfilter
@@ -256,6 +265,7 @@ def generate_call(ctx, kernel_info, ghost_layers_to_include=0, cell_interval=Non
kernel_call_lines.append("const %s %s = %s;" % (type_str, param.symbol.name, shape))
call_parameters = ", ".join([p.symbol.name for p in ast_params])
+
if not is_cpu:
if not spatial_shape_symbols:
spatial_shape_symbols = [p.symbol for p in ast_params if p.is_field_shape]
@@ -295,7 +305,7 @@ def generate_constructor_initializer_list(kernel_info, parameters_to_ignore=None
if param.is_field_pointer and param.field_name not in parameters_to_ignore:
parameter_initializer_list.append("%sID(%sID_)" % (param.field_name, param.field_name))
elif not param.is_field_parameter and param.symbol.name not in parameters_to_ignore:
- parameter_initializer_list.append("%s(%s_)" % (param.symbol.name, param.symbol.name))
+ parameter_initializer_list.append("%s_(%s)" % (param.symbol.name, param.symbol.name))
return ", ".join(parameter_initializer_list)
@@ -314,8 +324,8 @@ def generate_constructor_parameters(kernel_info, parameters_to_ignore=None):
if param.is_field_pointer and param.field_name not in parameters_to_ignore:
parameter_list.append("BlockDataID %sID_" % (param.field_name, ))
elif not param.is_field_parameter and param.symbol.name not in parameters_to_ignore:
- parameter_list.append("%s %s_" % (param.symbol.dtype, param.symbol.name,))
- varying_parameters = ["%s %s_" % e for e in varying_parameters]
+ parameter_list.append("%s %s" % (param.symbol.dtype, param.symbol.name,))
+ varying_parameters = ["%s %s" % e for e in varying_parameters]
return ", ".join(parameter_list + varying_parameters)
@@ -335,7 +345,7 @@ def generate_members(ctx, kernel_info, parameters_to_ignore=(), only_fields=Fals
if param.is_field_pointer and param.field_name not in params_to_skip:
result.append("BlockDataID %sID;" % (param.field_name, ))
elif not param.is_field_parameter and param.symbol.name not in params_to_skip:
- result.append("%s %s;" % (param.symbol.dtype, param.symbol.name,))
+ result.append("%s %s_;" % (param.symbol.dtype, param.symbol.name,))
for field_name in kernel_info.temporary_fields:
f = fields[field_name]
@@ -348,7 +358,7 @@ def generate_members(ctx, kernel_info, parameters_to_ignore=(), only_fields=Fals
result.append(temporary_fieldMemberTemplate.format(type=field_type, original_field_name=original_field_name))
if hasattr(kernel_info, 'varying_parameters'):
- result.extend(["%s %s;" % e for e in kernel_info.varying_parameters])
+ result.extend(["%s %s_;" % e for e in kernel_info.varying_parameters])
return "\n".join(result)
diff --git a/python/pystencils_walberla/templates/Sweep.tmpl.cpp b/python/pystencils_walberla/templates/Sweep.tmpl.cpp
index 8628e40a99beec142dc087831630c6cf189a78ff..ae15b45953455966605ad783ecbf5d2434a6b552 100644
--- a/python/pystencils_walberla/templates/Sweep.tmpl.cpp
+++ b/python/pystencils_walberla/templates/Sweep.tmpl.cpp
@@ -58,6 +58,7 @@ namespace {{namespace}} {
void {{class_name}}::operator()( IBlock * block{%if target is equalto 'gpu'%} , cudaStream_t stream{% endif %} )
{
{{kernel|generate_block_data_to_field_extraction|indent(4)}}
+ {{kernel|generate_refs_for_kernel_parameters(prefix='this->', parameters_to_ignore=['velocity', 'src', 'src_tmp', 'pdfs', 'pdfs_tmp'])|indent(4) }}
{{kernel|generate_call(stream='stream')|indent(4)}}
{{kernel|generate_swaps|indent(4)}}
}
@@ -77,6 +78,7 @@ void {{class_name}}::runOnCellInterval( const shared_ptr<StructuredBlockStorage>
return;
{{kernel|generate_block_data_to_field_extraction|indent(4)}}
+ {{kernel|generate_refs_for_kernel_parameters(prefix='this->', parameters_to_ignore=['velocity', 'src', 'src_tmp', 'pdfs', 'pdfs_tmp'])|indent(4) }}
{{kernel|generate_call(stream='stream', cell_interval='ci')|indent(4)}}
{{kernel|generate_swaps|indent(4)}}
}
diff --git a/tests/lbm/codegen/FluctuatingMRT.cpp b/tests/lbm/codegen/FluctuatingMRT.cpp
index 24e5168feaf6bc8dc1a606f0cafad05a3b4329a2..27021afaa407725d8a7d3a7f20eaf5643b15a155 100644
--- a/tests/lbm/codegen/FluctuatingMRT.cpp
+++ b/tests/lbm/codegen/FluctuatingMRT.cpp
@@ -85,7 +85,7 @@ int main( int argc, char ** argv )
communication.addPackInfo( make_shared< lbm::PdfFieldPackInfo< LatticeModel_T > >( pdfFieldId ) );
// add LBM sweep and communication to time loop
- timeloop.add() << BeforeFunction( [&](){ latticeModel.time_step = uint32_c(timeloop.getCurrentTimeStep()); }, "set RNG counter" )
+ timeloop.add() << BeforeFunction( [&](){ latticeModel.time_step_ = uint32_c(timeloop.getCurrentTimeStep()); }, "set RNG counter" )
<< BeforeFunction( communication, "communication" )
<< Sweep( LatticeModel_T::Sweep( pdfFieldId ), "LB stream & collide" );