Update all instructions in collect_rotate (independent of processing order)

python/dune/perftool/loopy/transformations/collect_rotate.py

@@ -153,7 +153,6 @@ def collect_vector_data_rotate(knl):
                 if isinstance(insn, lp.Assignment):
                     assignee = prim.Subscript(prim.Variable(arrname), get_quantity_subscripts(insn.assignee) + (prim.Variable('rotate_index'),))
                     new_insns.append(insn.copy(assignee=assignee,
-                                               expression=substitute(insn.expression, replacemap_arr),
                                                depends_on=insn.depends_on.union(frozenset({lp.match.Tagged("sumfact_stage1")})),
                                                )
                                      )
@@ -236,6 +235,8 @@ def collect_vector_data_rotate(knl):
                                                       (vector_indices.get(1), prim.Variable(new_iname)),
                                                       )
 
+    new_insns = [i.copy(expression=substitute(i.expression, replacemap_arr)) for i in new_insns]
+
     other_insns = [i for i in knl.instructions if i.id not in [j.id for j in insns + new_insns]]
 
     #