Some Haifa scheduler bugs

Jeffrey A Law law@hurl.cygnus.com
Mon Sep 1 20:33:00 GMT 1997


  In message <Pine.SOL.3.90.970826140826.2901J-100000@maigret.informatik.rwth-aa
chen.de>you write:
  > > 
  > >   > I'll re-run c-torture with the unmodified scheduler and
  > >   > report all failures.
  > > Please do.  It'll make tracking these things down easier.
  > 
  > Sorry about the delay, our mailserver didn't quite work yesterday.
  > Here are the failures I get without the reemit_notes fix on an
  > i586-linux configuration:
No problem.

I suspect all the cases where the compiler itself is core dumping
is the SCHED_GROUP_P problem in move_insn.
 
However, for execution tests that fail, I wonder how many (if any) are 
caused by -fsched-spec-load-dangerous.  If I remember right that
relies on the OS to support non-faulting loads from zero.

if (!p)
  blah = *p

The load of *p might happen before the code to check that p is
nonzero with -fsched-spec-load-dangerous.

Also note that the -fsched-reverse-{R,S} are for testing purposes
only, they actually pessimize code in an attempt to test the
scheduler more fully.  Long term I expect to rip them out of the
scheduler.

Jeff



More information about the Gcc mailing list