While working on PR42108 I noticed that LIM fails to clear range-info
on SSA names whose defining statements it moves, possibly causing
wrong code generation later on.
The following fixes that.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.
Richard.
2014-12-10 Richard Biener <rguent...@suse.de>
* tree-ssa-loop-im.c
(move_computations_dom_walker::before_dom_children): Clear
SSA_NAME_RANGE_INFO on moved stmts.
Index: gcc/tree-ssa-loop-im.c
===================================================================
--- gcc/tree-ssa-loop-im.c (revision 218515)
+++ gcc/tree-ssa-loop-im.c (working copy)
@@ -1232,6 +1286,11 @@ move_computations_dom_walker::before_dom
COND_EXPR, t, arg0, arg1);
todo_ |= TODO_cleanup_cfg;
}
+ if (INTEGRAL_TYPE_P (TREE_TYPE (gimple_assign_lhs (new_stmt)))
+ && (!ALWAYS_EXECUTED_IN (bb)
+ || (ALWAYS_EXECUTED_IN (bb) != level
+ && !flow_loop_nested_p (ALWAYS_EXECUTED_IN (bb), level))))
+ SSA_NAME_RANGE_INFO (gimple_assign_lhs (new_stmt)) = NULL;
gsi_insert_on_edge (loop_preheader_edge (level), new_stmt);
remove_phi_node (&bsi, false);
}
@@ -1291,6 +1350,13 @@ move_computations_dom_walker::before_dom
}
}
gsi_remove (&bsi, false);
+ if (gimple_has_lhs (stmt)
+ && TREE_CODE (gimple_get_lhs (stmt)) == SSA_NAME
+ && INTEGRAL_TYPE_P (TREE_TYPE (gimple_get_lhs (stmt)))
+ && (!ALWAYS_EXECUTED_IN (bb)
+ || !(ALWAYS_EXECUTED_IN (bb) == level
+ || flow_loop_nested_p (ALWAYS_EXECUTED_IN (bb), level))))
+ SSA_NAME_RANGE_INFO (gimple_get_lhs (stmt)) = NULL;
/* In case this is a stmt that is not unconditionally executed
when the target loop header is executed and the stmt may
invoke undefined integer or pointer overflow rewrite it to
