GCC build failed with your patch on 2001-01-09T11:35:00Z.

[Alexandre Oliva]

On Jan  9, 2001, Geoff Keating <geoffk@geoffk.org> wrote:

>> But it appears to me that the bug is in rs6000.h.  For a start, 
>> CONST_OK_FOR_LETTER_P(VALUE, 'L') assumes right-shift does
>> sign-extension; that's not necessarily true.

> You are correct here.  I will make a patch.

Thanks.

>> Second, it assumes VALUE must be signed 32-bit value, but, in this
>> case, it isn't.  I haven't investigated where this value comes from,
>> so I can't tell whether it is correct for it to be a positive 64-bit
>> value or if it should have been sign-extended.

> It's always wrong for VALUE to not be a signed 32-bit value here,
> because the mode is SImode.  

> You'll probably find that changing the size of HOST_WIDE_INT will
> trigger many problems, of which this is only one.  I'd recommend you
> fix those problems before making the change.

I can't promise to fix all problems on all 64-bit targets.  I know
there are a number of 64-bit targets that already work with a 64-bit
HOST_WIDE_INT; I was surprised that ppc wasn't one of them.  I'm
pretty sure the problem must be in the ppc back-end, since there are
so many other 64-bit targets that are working correctly.

Unfortunately, I don't see myself having time to pursue ppc issues at
this time.  But I do think the patch is a step in the right direction,
at the very least, because it will make it easier to find other
potential bugs lurking around in the ppc back-end.

Richard, since you approved the patch that auto-detects the need for a
64-bit HOST_WIDE_INT, I'll defer the decision about reverting it or
not to you.  Just let me know...

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me