| Message ID | 5616D1A9.1040202@redhat.com |
|---|---|
| State | New |
| Headers | show |
diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 4286491..64309c1 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,8 @@ +2015-10-08 Jeff Law <law@redhat.com> + + * tree-ssa-dse.c (dse_optimize_stmt): Add missing call to + release_defs. + 2015-10-08 H.J. Lu <hongjiu.lu@intel.com> * config/i386/i386.c (ix86_compute_frame_layout): Round up the diff --git a/gcc/tree-ssa-dse.c b/gcc/tree-ssa-dse.c index ac9c05a..80ebdb6 100644 --- a/gcc/tree-ssa-dse.c +++ b/gcc/tree-ssa-dse.c @@ -271,6 +271,7 @@ dse_optimize_stmt (gimple_stmt_iterator *gsi) /* Remove the dead store. */ if (gsi_remove (gsi, true)) bitmap_set_bit (need_eh_cleanup, gimple_bb (stmt)->index); + release_defs (stmt); } break; }
Another fairly obvious leak, this time in tree-ssa-dse.c where we again call gsi_remove without an associated release_defs. The only thing of note here is this time we have virtual operations on the statement. So before we call gsi_remove and release_defs, we have to call unlink_stmt_vdef to propagate away the uses of the VDEF. tree-ssa-dse properly calls unlink_stmt_vdef, so the right things are happening as far as that's concerned. Like the previous patch, I do have a minimized test, but it's unclear how we'd really want to put this into a testing harness at this point. If we get to an enforced no-leak policy in the manager, then these tests can just drop in as compile tests. Or perhaps as a plug-in test where we can check the state of the name manager. Anyway, bootstrapped and regression tested on x86_64-linux-gnu. Installed on the trunk. Jeff commit edf28250d6a25393684ee0e28f87fafe57183f76 Author: Jeff Law <law@redhat.com> Date: Thu Oct 8 14:25:42 2015 -0600 [PATCH] [2/n] Fix minor SSA_NAME leaks * tree-ssa-dse.c (dse_optimize_stmt): Add missing call to release_defs.