[parisc-linux] new gcc-default for hppa

[John David Anglin]

> gcc translates this to something like this:
>     /* %r4 is &pty_driver */
>     ldo 28(%r4),%r23
>     ldi 28,%r21
>     ldw,ma 4(%r22),%r20
>     ldw,ma 4(%r22),%r19
>     stw,ma %r20,4(%r23)
>     addib,>= -8,%r21,.-12
>     stw,ma %r19,4(%r23)
>     /* ... */
>     stw %r0,28(%r4)
>     stw %r0,32(%r4)
>     stw %r10,36(%r4)
>     stw %r0,40(%r4)
> 
> When optimizations are turned on, gcc reschedules some of those final 
> stw insns above the structure copy loop. When the structure copy 
> happens, the values get overwritten. ick.
> 
> this was kind of interesting to debug :) but i have no idea how to fix
> it. Dave? :)

I don't either at the moment.  It's memory aliasing problem.  It's
most likely not a problem in the backend.

Can you strip the above down to a simple testcase and file a GCC PR?
I will bump it to high-priority as it is a regression.

-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)