On Thu, 14 Mar 2019, Richard Biener wrote:
> On Thu, 14 Mar 2019, Jakub Jelinek wrote:
>
> > On Thu, Mar 14, 2019 at 11:23:03AM +0100, Richard Biener wrote:
> > > I am testing the following. There's IMHO also a missed optimization
> > > (for CFG-cleanup?) that when a block does not end up in a call
> > > outgoing abnormal edges can be purged? In this case it is
> > > IPA inlining leaving us with this - the inliner calls
> > > gimple_purge_dead_abnormal_call_edges on the return block
> > > but both cfun->has_nonlocal_label and cfun->calls_setjmp are
> > > false so the function does nothing. This is probably
> > > a premature check there?
> >
> > I think it would be better to keep the invariant that there are no abnormal
> > non-EH edges if both of the cfun bools are false, so that we don't need to
> > do useless work in the 99.9% of cases.
>
> I agree.
>
> > In the inliner case, is it that we haven't yet updated those or something
> > similar (if yes, could we update them earlier)?
>
> It's the edge still being present when DCE recomputes the flag.
>
> > In the DCE case, can't we when clearing the flag schedule a cleanup and only
> > clear it afterwards?
>
> Well, I guess the cleanup would just work iff we'd keep the flag
> check in call_can_make_abnormal_goto but remove the flag check
> in gimple_purge_dead_abnormal_call_edges. Not immediately
> after DCE but at least inlining gets rid of the edge (and the
> abnormal dispatcher) then.
>
> Does that sound reasonable?
>
> Thus like the following, testing in progress.
I've added a testcase.
Bootstrapped and tested on x86_64-unknown-linux-gnu, OK for trunk
or should it wait for GCC10?
Thanks,
Richard.
2019-03-14 Richard Biener <rguent...@suse.de>
PR tree-optimization/89710
* tree-cfg.c (gimple_purge_dead_abnormal_call_edges): Remove
premature check.
* gcc.dg/tree-ssa/pr89710.c: New testcase.
Index: gcc/tree-cfg.c
===================================================================
--- gcc/tree-cfg.c (revision 269678)
+++ gcc/tree-cfg.c (working copy)
@@ -8714,10 +8714,6 @@ gimple_purge_dead_abnormal_call_edges (b
edge_iterator ei;
gimple *stmt = last_stmt (bb);
- if (!cfun->has_nonlocal_label
- && !cfun->calls_setjmp)
- return false;
-
if (stmt && stmt_can_make_abnormal_goto (stmt))
return false;
Index: gcc/testsuite/gcc.dg/tree-ssa/pr89710.c
===================================================================
--- gcc/testsuite/gcc.dg/tree-ssa/pr89710.c (nonexistent)
+++ gcc/testsuite/gcc.dg/tree-ssa/pr89710.c (working copy)
@@ -0,0 +1,34 @@
+/* { dg-do compile } */
+/* { dg-options "-O -fdump-tree-fixup_cfg3" } */
+
+void
+gm (int *);
+
+__attribute__ ((returns_twice)) void
+jg (void)
+{
+}
+
+void
+eb (void)
+{
+ int r6 = 0;
+
+ if (r6 != 0)
+ gm (&r6);
+}
+
+void
+gm (int *r6)
+{
+ jg ();
+
+ for (;;)
+ {
+ eb ();
+ *r6 = 0;
+ }
+}
+
+/* After IPA inlining all abnormal edges and the dispatcher should vanish. */
+/* { dg-final { scan-tree-dump-not "ABNORMAL_DISPATCHER" "fixup_cfg3" } } */
