[PATCH v6 2/4] powerpc/64: enhance memcmp() with VMX instruction for long bytes comparision

[Simon Guo]

Hi segher,
On Wed, May 30, 2018 at 05:03:21PM +0800, Simon Guo wrote:
> On Wed, May 30, 2018 at 03:35:40AM -0500, Segher Boessenkool wrote:
> > On Wed, May 30, 2018 at 04:14:02PM +0800, Simon Guo wrote:
> > > Hi Segher,
> > > On Mon, May 28, 2018 at 06:05:59AM -0500, Segher Boessenkool wrote:
> > > > On Fri, May 25, 2018 at 12:07:34PM +0800, wei.guo.simon at gmail.com wrote:
> > > > > +	/* save and restore cr0 */
> > > > > +	mfocrf  r5,64
> > > > > +	EXIT_VMX_OPS
> > > > > +	mtocrf	64,r5
> > > > > +	b	.LcmpAB_lightweight
> > > > 
> > > > That's cr1, not cr0.  You can use mcrf instead, it is cheaper (esp. if
> > > > you have it in a non-volatile CR field before so you need only one, if any).
> > > > 
> > > You are right :) How about using mtcr/mfcr instead, I think they are
> > > fast as well and more readable.
> > 
> > Those are much worse than m[ft]ocrf.
> > 
> > You probably should just shuffle things around so that EXIT_VMX_OPS
> > does not clobber the CR field you need to keep.
> Let me use mcrf then :)

I now felt unformatable to use mcrf like:
mcrf    7,0

since I cannot 100% confident that compiler will not use CR7 or other
CR# in exit_vmx_ops().

Can we switch back to mfocrf/mtocrf with correct CR0 value?
       mfocrf  r5,128
        ...
       mtocrf  128,r5

Thanks,
- Simon