patch: disable string insns on ppc e500

[Geoff Keating]

> Date: Tue, 11 Mar 2003 13:23:04 -0800
> Cc: Geoff Keating <geoffk@geoffk.org>, David Edelsohn <dje@watson.ibm.com>
> From: Aldy Hernandez <aldy@quesejoda.com>
> X-OriginalArrivalTime: 11 Mar 2003 21:23:09.0890 (UTC) FILETIME=[6A38EA20:01C2E814]
> 
> As discussed.  This fixes a gazillion -Os regressions on the e500.
> 
> Applied to mainline.

No, no, no.  TARGET_SPE != CPU_E500.  Please use the right test.

> 2003-03-11  Aldy Hernandez  <aldyh@redhat.com>
> 
> 	* config/rs6000/rs6000.c (rs6000_override_options): Disable string
> 	instructions for e500.
> 
> Index: config/rs6000/rs6000.c
> ===================================================================
> RCS file: /cvs/uberbaum/gcc/config/rs6000/rs6000.c,v
> retrieving revision 1.436
> diff -c -p -r1.436 rs6000.c
> *** config/rs6000/rs6000.c	11 Mar 2003 20:40:53 -0000	1.436
> --- config/rs6000/rs6000.c	11 Mar 2003 21:21:24 -0000
> *************** rs6000_override_options (default_cpu)
> *** 698,703 ****
> --- 698,708 ----
>      SUBSUBTARGET_OVERRIDE_OPTIONS;
>    #endif
> 
> +   /* The e500 does not have string instructions, and we set
> +      MASK_STRING above when optimizing for size.  */
> +   if (TARGET_SPE && (target_flags & MASK_STRING) != 0)
> +     target_flags = target_flags & ~MASK_STRING;
> +
>      /* Handle -m(no-)longcall option.  This is a bit of a cheap hack,
>         using TARGET_OPTIONS to handle a toggle switch, but we're out of
>         bits in target_flags so TARGET_SWITCHES cannot be used.
> 


-- 
- Geoffrey Keating <geoffk@geoffk.org>