New code generation interface for waLBerla

__init__.py

@@ -6,7 +6,7 @@ from .slicing import make_slice
 from .kernelcreation import create_kernel, create_indexed_kernel, create_staggered_kernel
 from .display_utils import show_code, to_dot
 from .simp import AssignmentCollection
-from .assignment import Assignment
+from .assignment import Assignment, assignment_from_stencil
 from .sympyextensions import SymbolCreator
 from .datahandling import create_data_handling
 from .kernel_decorator import kernel
@@ -20,6 +20,7 @@ __all__ = ['Field', 'FieldType', 'fields',
            'show_code', 'to_dot',
            'AssignmentCollection',
            'Assignment',
+           'assignment_from_stencil',
            'SymbolCreator',
            'create_data_handling',
            'kernel',

cpu/vectorization.py

@@ -36,6 +36,9 @@ def vectorize(kernel_ast: ast.KernelFunction, instruction_set: str = 'avx',
                                         depending on the access pattern there might be additional padding
                                         required at the end of the array
     """
+    if instruction_set is None:
+        return
+    
     all_fields = kernel_ast.fields_accessed
     if nontemporal is None or nontemporal is False:
         nontemporal = {}

datahandling/parallel_datahandling.py

@@ -383,4 +383,3 @@ class ParallelDataHandling(DataHandling):
     def load_all(self, directory):
         for field_name, data_name in self._field_name_to_cpu_data_name.items():
             self.blocks.readBlockData(data_name, os.path.join(directory, field_name + ".dat"))
-

gpucuda/indexing.py

@@ -7,7 +7,7 @@ from pystencils.slicing import normalize_slice
 from pystencils.data_types import TypedSymbol, create_type
 from functools import partial
 
-AUTO_BLOCK_SIZE_LIMITING = True
+AUTO_BLOCK_SIZE_LIMITING = False
 
 BLOCK_IDX = [TypedSymbol("blockIdx." + coord, create_type("int")) for coord in ('x', 'y', 'z')]
 THREAD_IDX = [TypedSymbol("threadIdx." + coord, create_type("int")) for coord in ('x', 'y', 'z')]