PATCH: Share all CONST_INTs

[Mark Mitchell]

>>>>> "Anthony" == Anthony Green <green@cygnus.com> writes:

    Anthony> Hi Mark -

    Anthony> This patch triggers a failure in jc1.  It fails on x86
    Anthony> Linux while building libgcj.

Shoot -- I had this hunk sitting on my machine for a couple of days
and forgot to check it in.  I bet it will fix the problem.

As a side-note, we really don't want to maintain two different memory
allocation systems indefinitely.  At this point, I'd argue strongly
against accepting a new front-end now that didn't use GC.  I hope that
the Java team will make the conversion to GC soon; it's not really
that much work, and will provide a lot of long-term benefit.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com

2000-04-01  Mark Mitchell  <mark@codesourcery.com>

	* emit-rtl.c (gen_rtx_CONST_INT): Create cached CONST_INTs on the
	permanent obstack.

Index: emit-rtl.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/emit-rtl.c,v
retrieving revision 1.121
diff -c -p -r1.121 emit-rtl.c
*** emit-rtl.c	2000/03/31 08:57:53	1.121
--- emit-rtl.c	2000/04/02 02:47:09
*************** gen_rtx_CONST_INT (mode, arg)
*** 253,259 ****
  				   (hashval_t) arg,
  				   /*insert=*/1);
    if (!*slot)
!     *slot = gen_rtx_raw_CONST_INT (VOIDmode, arg);
  
    return (rtx) *slot;
  }
--- 253,268 ----
  				   (hashval_t) arg,
  				   /*insert=*/1);
    if (!*slot)
!     {
!       if (!ggc_p)
! 	{
! 	  push_obstacks_nochange ();
! 	  end_temporary_allocation ();
! 	}
!       *slot = gen_rtx_raw_CONST_INT (VOIDmode, arg);
!       if (!ggc_p)
! 	pop_obstacks ();
!     }
  
    return (rtx) *slot;
  }