Re: [PATCH] rs6000: Add basic support for prefixed and PC-relative addresses

[Michael Meissner <meissner@linux.ibm.com>]

On Thu, May 30, 2019 at 02:43:49PM -0500, Bill Schmidt wrote:
> On 5/30/19 2:20 PM, Segher Boessenkool wrote:
> > On Wed, May 29, 2019 at 10:49:35PM -0500, Bill Schmidt wrote:
> >> 	* config/rs6000/predicates.md (pcrel_address): New
> >> 	define_predicate.
> > Please put that on one line?
> 
> OK.  Emacs in ChangeLog and Fill modes seems to set a line length
> somewhat less than 79.  I generally follow what it tells me, but I can
> fix this.
> >
> >> +      if (LABEL_REF_P (x))
> >> +	output_asm_label (x);
> >> +      else
> >> +	output_addr_const (file, x);
> > Why does LABEL_REF need separate handling here?
> 
> Mike, please respond?

The current code doesn't need special handling.  I don't recall if earlier code
did, or I just assumed output_addr_const did not handle LABEL_REF's.

> > I don't much like penalising scalar single precision float like this.
> > But, this also hurts unaligned lwz...  Do we have statistics on that?
> > Offline, I guess :-)
> 
> Again, I'll defer to Mike on those details (online or offline). ;-)

I have rewritten this section.  In particular rather than basing everything off
of the mode, I have a now have a more targeted approach that knows which
register, mode, and with sign extension requires D/DS/DQ addresses.  Of course
this only works after register allocation.  Before register allocation, we
still have to make a best guess based only on the mode.

-- 
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.ibm.com, phone: +1 (978) 899-4797