| Message ID | alpine.LNX.2.00.1011241259000.6883@monoid.intra.ispras.ru |
|---|---|
| State | New |
| Headers | show |
On 11/24/2010 05:29 AM, Alexander Monakov wrote: > Hi, > > After the fix for PR 46204, we can enter maybe_tidy_empty_bb with invalid > topological order, but will make no attempt to recompute it (instead the caller, > tidy_control_flow, will do it later). That leads to an ICE in verify_backedges. > > Rather than bringing back some complexity into maybe_tidy_empty_bb to allow it > to recompute toporder again, I'd like to move verify_backedges into the caller > that creates edges between unrelated BBs, thus breaking toporder. > > Bootstrapped and regtested on x86_64-linux, ia64-linux (+ with sel-sched > enabled at -O2), OK for trunk? > > 2010-11-24 Alexander Monakov<amonakov@ispras.ru> > > PR rtl-optimization/46602 > * sel-sched-ir.c (maybe_tidy_empty_bb): Move checking ... > (tidy_control_flow): Here. > > OK, thanks for the patch.
diff --git a/gcc/sel-sched-ir.c b/gcc/sel-sched-ir.c index 71c02c4..231358b 100644 --- a/gcc/sel-sched-ir.c +++ b/gcc/sel-sched-ir.c @@ -3650,10 +3650,6 @@ maybe_tidy_empty_bb (basic_block bb) remove_empty_bb (bb, true); } -#ifdef ENABLE_CHECKING - verify_backedges (); -#endif - return true; } @@ -3735,6 +3731,11 @@ tidy_control_flow (basic_block xbb, bool full_tidying) if (recompute_toporder_p) sel_recompute_toporder (); } + +#ifdef ENABLE_CHECKING + verify_backedges (); +#endif + return changed; }