diff --git a/pystencils/cpu/kernelcreation.py b/pystencils/cpu/kernelcreation.py
index c84ed8d2b6764c4f67db7c8235fa1acfa3e63752..c28e7d0ad686cad3f1689f1c1e8c67a0ab40def5 100644
--- a/pystencils/cpu/kernelcreation.py
+++ b/pystencils/cpu/kernelcreation.py
@@ -19,7 +19,7 @@ AssignmentOrAstNodeList = List[Union[Assignment, ast.Node]]
def create_kernel(assignments: AssignmentOrAstNodeList, function_name: str = "kernel", type_info='double',
split_groups=(), iteration_slice=None, ghost_layers=None,
- skip_independence_check=False, optimizations=optims_pystencils_cpu) -> KernelFunction:
+ skip_independence_check=False, sympy_optimizations=optims_pystencils_cpu) -> KernelFunction:
"""Creates an abstract syntax tree for a kernel function, by taking a list of update rules.
Loops are created according to the field accesses in the equations.
@@ -55,9 +55,9 @@ def create_kernel(assignments: AssignmentOrAstNodeList, function_name: str = "ke
else:
raise ValueError("Term has to be field access or symbol")
- if optimizations is None:
- optimizations = optims_pystencils_cpu
- assignments = optimize_assignments(assignments, optimizations)
+ if sympy_optimizations is None:
+ sympy_optimizations = optims_pystencils_cpu
+ assignments = optimize_assignments(assignments, sympy_optimizations)
fields_read, fields_written, assignments = add_types(assignments, type_info, not skip_independence_check)
all_fields = fields_read.union(fields_written)
@@ -99,7 +99,7 @@ def create_indexed_kernel(assignments: AssignmentOrAstNodeList,
function_name="kernel",
type_info=None,
coordinate_names=('x', 'y', 'z'),
- optimizations=optims_pystencils_cpu) -> KernelFunction:
+ sympy_optimizations=optims_pystencils_cpu) -> KernelFunction:
"""
Similar to :func:`create_kernel`, but here not all cells of a field are updated but only cells with
coordinates which are stored in an index field. This traversal method can e.g. be used for boundary handling.
@@ -116,10 +116,10 @@ def create_indexed_kernel(assignments: AssignmentOrAstNodeList,
function_name: see documentation of :func:`create_kernel`
coordinate_names: name of the coordinate fields in the struct data type
"""
- if optimizations is None:
- optimizations = optims_pystencils_cpu
+ if sympy_optimizations is None:
+ sympy_optimizations = optims_pystencils_cpu
- assignments = optimize_assignments(assignments, optimizations)
+ assignments = optimize_assignments(assignments, sympy_optimizations)
fields_read, fields_written, assignments = add_types(assignments, type_info, check_independence_condition=False)
all_fields = fields_read.union(fields_written)
diff --git a/pystencils/gpucuda/kernelcreation.py b/pystencils/gpucuda/kernelcreation.py
index 3ac806e6cbc1838e2255b8cd12b057d238c0d5c8..12c7bd3c97dd8f8cc17f2641dbd442b615110082 100644
--- a/pystencils/gpucuda/kernelcreation.py
+++ b/pystencils/gpucuda/kernelcreation.py
@@ -3,18 +3,24 @@ from pystencils.data_types import BasicType, StructType, TypedSymbol
from pystencils.field import Field, FieldType
from pystencils.gpucuda.cudajit import make_python_function
from pystencils.gpucuda.indexing import BlockIndexing
-from pystencils.math_optimizations import optims_pystencils_gpu, optimize_assignments
+from pystencils.math_optimizations import optimize_assignments, optims_pystencils_gpu
from pystencils.transformations import (
add_types, get_base_buffer_index, get_common_shape, parse_base_pointer_info,
resolve_buffer_accesses, resolve_field_accesses, unify_shape_symbols)
-def create_cuda_kernel(assignments, function_name="kernel", type_info=None, indexing_creator=BlockIndexing,
- iteration_slice=None, ghost_layers=None, skip_independence_check=False, optimizations=None):
+def create_cuda_kernel(assignments,
+ function_name="kernel",
+ type_info=None,
+ indexing_creator=BlockIndexing,
+ iteration_slice=None,
+ ghost_layers=None,
+ skip_independence_check=False,
+ sympy_optimizations=None):
- if optimizations is None:
- optimizations = optims_pystencils_gpu
- assignments = optimize_assignments(assignments, optimizations)
+ if sympy_optimizations is None:
+ sympy_optimizations = optims_pystencils_gpu
+ assignments = optimize_assignments(assignments, sympy_optimizations)
fields_read, fields_written, assignments = add_types(assignments, type_info, not skip_independence_check)
all_fields = fields_read.union(fields_written)
@@ -92,11 +98,16 @@ def create_cuda_kernel(assignments, function_name="kernel", type_info=None, inde
return ast
-def created_indexed_cuda_kernel(assignments, index_fields, function_name="kernel", type_info=None,
- coordinate_names=('x', 'y', 'z'), indexing_creator=BlockIndexing, optimizations=None):
- if optimizations is None:
- optimizations = optims_pystencils_gpu
- assignments = optimize_assignments(assignments, optimizations)
+def created_indexed_cuda_kernel(assignments,
+ index_fields,
+ function_name="kernel",
+ type_info=None,
+ coordinate_names=('x', 'y', 'z'),
+ indexing_creator=BlockIndexing,
+ sympy_optimizations=None):
+ if sympy_optimizations is None:
+ sympy_optimizations = optims_pystencils_gpu
+ assignments = optimize_assignments(assignments, sympy_optimizations)
fields_read, fields_written, assignments = add_types(assignments, type_info, check_independence_condition=False)
all_fields = fields_read.union(fields_written)
diff --git a/pystencils/kernelcreation.py b/pystencils/kernelcreation.py
index ae84332cbe6739c34a17eba967bad77d5b402fed..79c3d04d10224b63733bbbc8b88c9ebb093e5ce6 100644
--- a/pystencils/kernelcreation.py
+++ b/pystencils/kernelcreation.py
@@ -15,7 +15,7 @@ from pystencils.transformations import (
def create_kernel(assignments, target='cpu', data_type="double", iteration_slice=None, ghost_layers=None,
skip_independence_check=False,
cpu_openmp=False, cpu_vectorize_info=None, cpu_blocking=None,
- gpu_indexing='block', gpu_indexing_params=MappingProxyType({}), optimizations=None):
+ gpu_indexing='block', gpu_indexing_params=MappingProxyType({}), sympy_optimizations=None):
"""
Creates abstract syntax tree (AST) of kernel, using a list of update equations.
@@ -75,7 +75,7 @@ def create_kernel(assignments, target='cpu', data_type="double", iteration_slice
from pystencils.cpu import add_openmp
ast = create_kernel(assignments, type_info=data_type, split_groups=split_groups,
iteration_slice=iteration_slice, ghost_layers=ghost_layers,
- skip_independence_check=skip_independence_check, optimizations=optimizations)
+ skip_independence_check=skip_independence_check, sympy_optimizations=sympy_optimizations)
omp_collapse = None
if cpu_blocking:
omp_collapse = loop_blocking(ast, cpu_blocking)
@@ -91,15 +91,21 @@ def create_kernel(assignments, target='cpu', data_type="double", iteration_slice
return ast
elif target == 'llvm':
from pystencils.llvm import create_kernel
- ast = create_kernel(assignments, type_info=data_type, split_groups=split_groups,
- iteration_slice=iteration_slice, ghost_layers=ghost_layers, optimizations=optimizations)
+ ast = create_kernel(assignments,
+ type_info=data_type,
+ split_groups=split_groups,
+ iteration_slice=iteration_slice,
+ ghost_layers=ghost_layers,
+ sympy_optimizations=sympy_optimizations)
return ast
elif target == 'gpu':
from pystencils.gpucuda import create_cuda_kernel
ast = create_cuda_kernel(assignments, type_info=data_type,
indexing_creator=indexing_creator_from_params(gpu_indexing, gpu_indexing_params),
- iteration_slice=iteration_slice, ghost_layers=ghost_layers,
- skip_independence_check=skip_independence_check, optimizations=optimizations)
+ iteration_slice=iteration_slice,
+ ghost_layers=ghost_layers,
+ skip_independence_check=skip_independence_check,
+ sympy_optimizations=sympy_optimizations)
return ast
else:
raise ValueError("Unknown target %s. Has to be one of 'cpu', 'gpu' or 'llvm' " % (target,))
@@ -107,7 +113,7 @@ def create_kernel(assignments, target='cpu', data_type="double", iteration_slice
def create_indexed_kernel(assignments, index_fields, target='cpu', data_type="double", coordinate_names=('x', 'y', 'z'),
cpu_openmp=True, gpu_indexing='block', gpu_indexing_params=MappingProxyType({}),
- optimizations=None):
+ sympy_optimizations=None):
"""
Similar to :func:`create_kernel`, but here not all cells of a field are updated but only cells with
coordinates which are stored in an index field. This traversal method can e.g. be used for boundary handling.
@@ -151,7 +157,7 @@ def create_indexed_kernel(assignments, index_fields, target='cpu', data_type="do
from pystencils.cpu import create_indexed_kernel
from pystencils.cpu import add_openmp
ast = create_indexed_kernel(assignments, index_fields=index_fields, type_info=data_type,
- coordinate_names=coordinate_names, optimizations=optimizations)
+ coordinate_names=coordinate_names, sympy_optimizations=sympy_optimizations)
if cpu_openmp:
add_openmp(ast, num_threads=cpu_openmp)
return ast
@@ -162,14 +168,14 @@ def create_indexed_kernel(assignments, index_fields, target='cpu', data_type="do
idx_creator = indexing_creator_from_params(gpu_indexing, gpu_indexing_params)
ast = created_indexed_cuda_kernel(assignments, index_fields, type_info=data_type,
coordinate_names=coordinate_names, indexing_creator=idx_creator,
- optimizations=optimizations)
+ sympy_optimizations=sympy_optimizations)
return ast
else:
raise ValueError("Unknown target %s. Has to be either 'cpu' or 'gpu'" % (target,))
def create_staggered_kernel(staggered_field, expressions, subexpressions=(), target='cpu',
- gpu_exclusive_conditions=False, optimizations=None, **kwargs):
+ gpu_exclusive_conditions=False, sympy_optimizations=None, **kwargs):
"""Kernel that updates a staggered field.
.. image:: /img/staggered_grid.svg
@@ -256,8 +262,7 @@ def create_staggered_kernel(staggered_field, expressions, subexpressions=(), tar
ast = create_kernel(final_assignments,
ghost_layers=ghost_layers,
target=target,
- optimizations=optimizations,
-
+ sympy_optimizations=sympy_optimizations,
**kwargs)
if target == 'cpu':
diff --git a/pystencils/llvm/kernelcreation.py b/pystencils/llvm/kernelcreation.py
index 1bffaa845d982722dbba0d7b87a6dd9d6f86355a..86f26ef6f388d95446f323b5c8965b0d90ff15af 100644
--- a/pystencils/llvm/kernelcreation.py
+++ b/pystencils/llvm/kernelcreation.py
@@ -3,7 +3,7 @@ from pystencils.transformations import insert_casts
def create_kernel(assignments, function_name="kernel", type_info=None, split_groups=(),
- iteration_slice=None, ghost_layers=None, optimizations=None):
+ iteration_slice=None, ghost_layers=None, sympy_optimizations=None):
"""
Creates an abstract syntax tree for a kernel function, by taking a list of update rules.
@@ -32,7 +32,7 @@ def create_kernel(assignments, function_name="kernel", type_info=None, split_gro
split_groups,
iteration_slice,
ghost_layers,
- optimizations=optimizations)
+ sympy_optimizations=sympy_optimizations)
code.body = insert_casts(code.body)
code._compile_function = make_python_function
code._backend = 'llvm'
diff --git a/pystencils_tests/test_sympy_optimizations.py b/pystencils_tests/test_sympy_optimizations.py
index 080205e3aecade863b26c1bd4c7227fe3e111c22..ff3b0fbbdeb26bd79014a89c2b61e13de4173e9e 100644
--- a/pystencils_tests/test_sympy_optimizations.py
+++ b/pystencils_tests/test_sympy_optimizations.py
@@ -48,7 +48,7 @@ def test_do_not_evaluate_constant_terms():
x[0, 0]: -sp.cos(1) + y[0, 0]
})
- ast = pystencils.create_kernel(assignments, target=target, optimizations=optimizations)
+ ast = pystencils.create_kernel(assignments, target=target, sympy_optimizations=optimizations)
code = str(pystencils.show_code(ast))
assert 'cos(' in code
print(code)