public inbox for gcc@gcc.gnu.org
 help / color / mirror / Atom feed
* ICE on IA-64 when compiling SPEC 2000 176.gcc at -O1
@ 2001-12-05 17:07 Reva Cuthbertson
  2001-12-05 23:19 ` Richard Henderson
  0 siblings, 1 reply; 3+ messages in thread
From: Reva Cuthbertson @ 2001-12-05 17:07 UTC (permalink / raw)
  To: gcc


Hello,

	I am getting an internal compiler error when I
compile the SPEC 2000 benchmark 176.gcc with -O1 on IA-64
with gcc V3.1 on HP-UX and Linux.  The ICE happens in
find_free_reg() in local-alloc.c and it happens because
the born_index > death_index test fails so we call abort().
It looks like the death_index is never set up properly as
it is -1.  Has anyone else seen this?

Thanks!

Reva Cuthbertson
reva@cup.hp.com

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: ICE on IA-64 when compiling SPEC 2000 176.gcc at -O1
  2001-12-05 17:07 ICE on IA-64 when compiling SPEC 2000 176.gcc at -O1 Reva Cuthbertson
@ 2001-12-05 23:19 ` Richard Henderson
  2001-12-06 12:57   ` Reva Cuthbertson
  0 siblings, 1 reply; 3+ messages in thread
From: Richard Henderson @ 2001-12-05 23:19 UTC (permalink / raw)
  To: Reva Cuthbertson; +Cc: gcc

On Wed, Dec 05, 2001 at 05:07:50PM -0800, Reva Cuthbertson wrote:
> ... and it happens because the born_index > death_index test fails so
> we call abort().

This is normally due to register life info being incorrect.

That is, on input to local_alloc some register was reported
to be set once and never die (certainly possible), yet the
register does in fact die.

Can't debug this further without a test case.


r~

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: ICE on IA-64 when compiling SPEC 2000 176.gcc at -O1
  2001-12-05 23:19 ` Richard Henderson
@ 2001-12-06 12:57   ` Reva Cuthbertson
  0 siblings, 0 replies; 3+ messages in thread
From: Reva Cuthbertson @ 2001-12-06 12:57 UTC (permalink / raw)
  To: Richard Henderson; +Cc: gcc

Richard Henderson wrote:

> On Wed, Dec 05, 2001 at 05:07:50PM -0800, Reva Cuthbertson wrote:
> > ... and it happens because the born_index > death_index test fails so
> > we call abort().
>
> This is normally due to register life info being incorrect.
>
> That is, on input to local_alloc some register was reported
> to be set once and never die (certainly possible), yet the
> register does in fact die.
>
> Can't debug this further without a test case.
>
> r~

Richard,

     The problem can be seen when compiling the function
output_signed_leb128() in dwarfout.c in the gcc sources.
The function itself has been commented out on the main
branch so the problem does not happen if you compile
dwarfout.c with -O1.  However, the function output_signed_leb128()
is part of 176.gcc in SPEC 2000.  For convenience, I cut down
output_signed_leb128() into the following test case.  Compile
this with -O1 using gcc 3.1 for IA-64 and the internal compiler
error should occur.

static void
output_signed_leb128 (value)
     long value;
{
  int more;

  do
    {
      unsigned byte = (value & 0x7f);

      value >>= 7;
      if (((value == 0) && ((byte & 0x40) == 0))
          || ((value == -1) && ((byte & 0x40) == 1)))
        more = 0;
    }
  while (more);
}


Thanks!

Reva


^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2001-12-06 20:53 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2001-12-05 17:07 ICE on IA-64 when compiling SPEC 2000 176.gcc at -O1 Reva Cuthbertson
2001-12-05 23:19 ` Richard Henderson
2001-12-06 12:57   ` Reva Cuthbertson

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).