Re: [PATCH] Vector return on ARM

[Josh Conner <jconner at apple dot com>]

On May 12, 2005, at 10:34 AM, Josh Conner wrote:

> On May 12, 2005, at 6:21 AM, Julian Brown wrote:
>
>
> > Paul Brook wrote:
> >
> > Yes, Josh's patch doesn't break anything new, it just doesn't fix  
> > that particular vector return bug. I'll see if I can figure out  
> > what's going on with that.
> >
> > Cheers,
> >
> > Julian
>
> This code in output_return_instruction() is making r3 be restored  
> unnecessarily:
>
>                   /* If we can't use ldmib (SA110 bug),
>                      then try to pop r3 instead.  */
>                   if (stack_adjust)
>                     live_regs_mask |= 1 << 3;

Sure enough, there's a conflict of assumptions --  
output_return_instruction() assumes that r3 can be safely clobbered  
(which it does to work around ldmib not behaving properly on SA-110  
devices).  And, use_return_insn() assumes that only r0 will be used  
for return values.

Attached is a patch that updates use_return_insn() so that it  
actually checks which registers are being used for return values  
(regtested on arm-none-elf with strictly better results).  OK for  
mainline?

Do we also want to consider removing support for the SA110  
workaround, or at least making it command-line conditional?  The  
errata is against a fairly old set of rev A devices that have been  
replaced (see http://www.mit.edu/afs/sipb/contrib/doc/specs/ic/cpu/ 
arm/sa110-update.pdf).  The workaround does infer a slight penalty on  
anyone not using the erroneous device.

- Josh
~~~~~~

2005-05-13 Josh Conner <jconner@apple.com>

        * config/arm/arm.c (use_return_insn): Include test of which
        registers are being used to hold return values, to accommodate
        vector return values.

Attachment: epilogue.patch
Description: Binary data