[bugfix] have the index update instructions depend on the writers

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

@@ -54,6 +54,7 @@ def collect_vector_data_rotate(knl, insns, inames):
     knl = lp.tag_inames(knl, [(new_iname, "vec")])
 
     new_insns = []
+    all_writers = []
 
     #
     # Inspect the given instructions for dependent quantities
@@ -79,6 +80,7 @@ def collect_vector_data_rotate(knl, insns, inames):
         write_match = lp.match.Writes(quantity)
         match = lp.match.And((iname_match, write_match))
         write_insns = lp.find_instructions(knl, match)
+        all_writers.extend([i.id for i in write_insns])
 
         if write_insns:
             arrname = quantity + '_buffered_arr'
@@ -142,7 +144,7 @@ def collect_vector_data_rotate(knl, insns, inames):
                                    prim.Sum((prim.Variable("total_index"), 1)),  # expression
                                    within_inames=common_inames.union(inames),
                                    within_inames_is_final=True,
-                                   depends_on=frozenset([i.id for i in write_insns]).union(frozenset({"assign_total_index"})),
+                                   depends_on=frozenset(all_writers).union(frozenset({"assign_total_index"})),
                                    depends_on_is_final=True,
                                    id="update_total_index",
                                    ))
@@ -162,7 +164,7 @@ def collect_vector_data_rotate(knl, insns, inames):
                                    prim.Remainder(prim.Sum((prim.Variable("rotate_index"), 1)), vec_size),  # expression
                                    within_inames=common_inames.union(inames),
                                    within_inames_is_final=True,
-                                   depends_on=frozenset([i.id for i in write_insns]).union(frozenset({"assign_rotate_index"})),
+                                   depends_on=frozenset(all_writers).union(frozenset({"assign_rotate_index"})),
                                    depends_on_is_final=True,
                                    id="update_rotate_index",
                                    ))
@@ -184,7 +186,7 @@ def collect_vector_data_rotate(knl, insns, inames):
     new_insns.append(lp.CInstruction((),  # iname exprs that the code needs access to
                                      "continue;",  # the code
                                      predicates=frozenset({check}),
-                                     depends_on=frozenset({"update_rotate_index", "update_total_index"}).union(frozenset([i.id for i in write_insns])),
+                                     depends_on=frozenset({"update_rotate_index", "update_total_index"}).union(frozenset(all_writers)),
                                      depends_on_is_final=True,
                                      within_inames=common_inames.union(inames),
                                      within_inames_is_final=True,