The following should fix PR88448, I've reverted the semantics changing
part of the gimple_assign_set_rhs_* previous change settling for
an alternate fix for the tree-complex.c issue.
Bootstrap & regtest in progress on x86_64-unknown-linux-gnu.
Richard.
2018-12-11 Richard Biener <rguent...@suse.de>
PR middle-end/88448
PR middle-end/88415
* gimple.c (gimple_assign_set_rhs_with_ops): Revert previous
change.
* tree-complex.c (update_complex_assignment): Properly transfer
or clean EH info around gimple_assign_set_rhs_with_ops.
Index: gcc/gimple.c
===================================================================
--- gcc/gimple.c (revision 266975)
+++ gcc/gimple.c (working copy)
@@ -1752,7 +1752,7 @@ gimple_assign_set_rhs_with_ops (gimple_s
if (new_rhs_ops > 2)
gimple_assign_set_rhs3 (stmt, op3);
if (stmt != old_stmt)
- gsi_replace (gsi, stmt, true);
+ gsi_replace (gsi, stmt, false);
}
Index: gcc/tree-complex.c
===================================================================
--- gcc/tree-complex.c (revision 266975)
+++ gcc/tree-complex.c (working copy)
@@ -698,12 +698,11 @@ update_complex_components_on_edge (edge
static void
update_complex_assignment (gimple_stmt_iterator *gsi, tree r, tree i)
{
- gimple *stmt;
-
+ gimple *old_stmt = gsi_stmt (*gsi);
gimple_assign_set_rhs_with_ops (gsi, COMPLEX_EXPR, r, i);
- stmt = gsi_stmt (*gsi);
+ gimple *stmt = gsi_stmt (*gsi);
update_stmt (stmt);
- if (maybe_clean_eh_stmt (stmt))
+ if (maybe_clean_or_replace_eh_stmt (old_stmt, stmt))
bitmap_set_bit (need_eh_cleanup, gimple_bb (stmt)->index);
update_complex_components (gsi, gsi_stmt (*gsi), r, i);
