Re: [Qemu-devel] [RFC PATCH 1/6] target/ppc: introduce get_fpr() and set_fpr() helpers for FP register access

[David Gibson <david@gibson.dropbear.id.au>]

[-- Attachment #1: Type: text/plain, Size: 1572 bytes --]

On Mon, Dec 10, 2018 at 12:25:26PM -0600, Richard Henderson wrote:
> On 12/9/18 11:17 PM, David Gibson wrote:
> > On Fri, Dec 07, 2018 at 08:56:30AM +0000, Mark Cave-Ayland wrote:
> >> These helpers allow us to move FP register values to/from the specified TCGv_i64
> >> argument.
> >>
> >> To prevent FP helpers accessing the cpu_fpr array directly, add extra TCG
> >> temporaries as required.
> > 
> > It's not obvious to me why that's a desirable thing.  I'm assuming
> > it's somehow necessary for the stuff later in the series, but I think
> > we need a brief rationale here to explain why this isn't just adding
> > extra reg copies for the sake of it.
> 
> Note that while this introduces extra opcodes, in many cases it does not change
> the number of machine instructions that are generated.  Recall that accessing
> cpu_fpr[N] implies a load from env.  This change makes the load explicit.

I realised that a bit later in looking at the series.  I think a
paraphrasing of the above in the commit message would still be
helpful.

> The change does currently prevent caching cpu_fpr[N] in a host register.  That
> can and will be fixed by optimizing on memory operations instead.  (There is a
> patch that has been outstanding for 13 months to do this.  I intend to finally
> get around to merging it during the 4.0 cycle.)
> 
> 
> r~
> 

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]