rework of vectorization

python/dune/perftool/blockstructured/accumulation.py

@@ -11,14 +11,20 @@ import pymbolic.primitives as prim
 
 def name_accumulation_alias(container, accumspace):
     name = container+"_"+accumspace.lfs.name+"_alias"
+    name_tail = container+"_"+accumspace.lfs.name+"_alias_tail"
     k = get_option("number_of_blocks")
     p = accumspace.element.degree()
-    globalarg(name, shape=(k, k, p + 1, p + 1), strides=(p, p * k + 1, 1, k + 1), managed=True)
-    code = "auto {} = &{}.container()({},0);".format(container + "_"+accumspace.lfs.name+"_alias", container, accumspace.lfs.name)
-    instruction(within_inames=frozenset(),
-                code=code,
-                read_variables=frozenset({container}),
-                assignees=frozenset({name}))
+
+    def _add_alias_insn(name):
+        globalarg(name, shape=(k, k, p + 1, p + 1), strides=(p, p * k + 1, 1, k + 1), managed=True)
+        code = "auto {} = &{}.container()({},0);".format(name, container, accumspace.lfs.name)
+        instruction(within_inames=frozenset(),
+                    code=code,
+                    read_variables=frozenset({container}),
+                    assignees=frozenset({name}))
+
+    _add_alias_insn(name)
+    _add_alias_insn(name_tail)
     return name
 
 
@@ -53,5 +59,6 @@ def generate_accumulation_instruction(expr, visitor):
                 expression=prim.Sum((expr_with_weight, assignee)),
                 forced_iname_deps=frozenset(lfs_inames).union(frozenset(quad_inames)),
                 forced_iname_deps_is_final=True,
-                predicates=predicates
+                predicates=predicates,
+                tags=frozenset({'accum'})
                 )