Fix alignement detection

pystencils/cpu/kernelcreation.py

@@ -76,8 +76,6 @@ def create_kernel(assignments: NodeCollection,
     base_pointer_spec = config.base_pointer_specification
     if base_pointer_spec is None:
         base_pointer_spec = []
-        if config.cpu_vectorize_info and config.cpu_vectorize_info.get('nontemporal'):
-            base_pointer_spec = [['spatialInner0'], ['spatialInner1']] if len(loop_order) >= 2 else [['spatialInner0']]
     base_pointer_info = {field.name: parse_base_pointer_info(base_pointer_spec, loop_order,
                                                              field.spatial_dimensions, field.index_dimensions)
                          for field in fields_without_buffers}

pystencils/cpu/vectorization.py

@@ -140,12 +140,16 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
                                                 strided, keep_loop_stop, assume_sufficient_line_padding,
                                                 default_float_type):
     """Goes over all innermost loops, changes increment to vector width and replaces field accesses by vector type."""
-    vector_width = ast_node.instruction_set['width']
-
-    all_loops = filtered_tree_iteration(ast_node, ast.LoopOverCoordinate, stop_type=ast.SympyAssignment)
+    loop_generator = filtered_tree_iteration(ast_node, ast.LoopOverCoordinate, stop_type=ast.SympyAssignment)
+    all_loops = [loop for loop in loop_generator]
     inner_loops = [loop for loop in all_loops if loop.is_innermost_loop]
     zero_loop_counters = {loop.loop_counter_symbol: 0 for loop in all_loops}
 
+    assert ast_node.instruction_set,\
+        "The ast needs to hold information about the instruction_set for the vectorisation"
+    vector_width = ast_node.instruction_set['width']
+    vector_int_width = ast_node.instruction_set['intwidth']
+
     for loop_node in inner_loops:
         loop_range = loop_node.stop - loop_node.start
 
@@ -175,7 +179,7 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
             base, index = indexed.args
             if loop_counter_symbol in index.atoms(sp.Symbol):
                 loop_counter_is_offset = loop_counter_symbol not in (index - loop_counter_symbol).atoms()
-                aligned_access = (index - loop_counter_symbol).subs(zero_loop_counters) == 0
+                aligned_access = (index - loop_counter_symbol).subs(zero_loop_counters) % vector_width == 0
                 stride = sp.simplify(index.subs({loop_counter_symbol: loop_counter_symbol + 1}) - index)
                 if not loop_counter_is_offset and (not strided or loop_counter_symbol in stride.atoms()):
                     successful = False
@@ -204,7 +208,6 @@ def vectorize_inner_loops_and_adapt_load_stores(ast_node, assume_aligned, nontem
 
         loop_node.step = vector_width
         loop_node.subs(substitutions)
-        vector_int_width = ast_node.instruction_set['intwidth']
         arg_1 = CastFunc(loop_counter_symbol, VectorType(loop_counter_symbol.dtype, vector_int_width))
         arg_2 = CastFunc(tuple(range(vector_int_width if type(vector_int_width) is int else 2)),
                          VectorType(loop_counter_symbol.dtype, vector_int_width))