Re: Solaris 2/SPARC Ada bootstrap broken

[Dale Johannesen <dalej at apple dot com>]

On Friday, August 23, 2002, at 02:23 PM, Rainer Orth wrote:

> Dale,
>
> this patch
>
> 	http://gcc.gnu.org/ml/gcc-patches/2002-08/msg00838.html
>
> 2002-08-14  Dale Johannesen  <dalej@apple.com>
>
> 	* explow.c (emit_stack_restore):  Emit memory clobbers
> 	preceding the stack pop, to prevent the scheduler from
> 	moving refs to variable arrays below this pop.
> 	* reload1.c (reload):  Preserve these clobbers for sched2.
> 	* doc/rtl.texi:  Document clobber (mem:BLK (scratch)).
>
> broke Ada bootstrap on sparc-sun-solaris2.5.1/sparc-sun-solaris2.8: I get
> an ICE compiling ada/sem_prag.adb:
>
> +===========================GNAT BUG DETECTED============================
> ==+
> | 3.3 20020814 (experimental) (sparc-sun-solaris2.5.1) GCC error:        
>   |
> | Internal compiler error in reload_combine_note_use, at reload1.c:8962  
>   |
> | Error detected at /vol/gnu/src/gcc/gcc-dated/gcc/ada/sem_prag.adb:3560:
> 7 |
>
> Reverting the patch brings Ada back to bootstrap.  Well, sort of: despite
> several reminders by various people, the basic patch to return Ada
> bootstrap on many platforms has remained unreviewed since June 22nd,
> leaving Ada bootstrap broken for two months ;-(
>
> 	http://gcc.gnu.org/ml/gcc-patches/2002-06/msg01899.html
>
> Thus I need to apply this one and revert your patch.
[back from vacation]
Sorry.  I haven't been able to build Ada here.  The patch fixes a bug which
can cause unreproducible crashes, so I doubt that just backing it out is 
right.
I don't see why the patch should cause this crash.  The code is looking for
uses of pseudoregisters, and the patch only introduces a use of 
stack_pointer_rtx.
That should always be a hard reg, shouldn't it?  Can you show me a dump of 
the RTL
it's complaining about?