Re: [PATCH], PR target/81593, Optimize PowerPC vector sets coming from a vector extracts

[Michael Meissner <meissner at linux dot vnet dot ibm dot com>]

On Thu, Aug 03, 2017 at 10:01:41AM -0500, Segher Boessenkool wrote:
> Hi Mike,
> 
> On Wed, Aug 02, 2017 at 10:28:55AM -0400, Michael Meissner wrote:
> > On Fri, Jul 28, 2017 at 04:08:50PM -0500, Segher Boessenkool wrote:
> > > I think calling this with the rtx elementN args makes this only more
> > > complicated (the function comment doesn't say what they are or what
> > > NULL means, btw).
> 
> You didn't handle the first part of this as far as I see?  It's the
> big complicating issue here.

I am not sure exactly what you are asking for.  This is like the other output
functions that take the rtx insns.

> > +   If ELEMENT1 is null, use the top 64-bit double word of ARG1.  If it is
> > +   non-NULL, it is a 0 or 1 constant that gives the vector element number to
> > +   use for extracting the 64-bit double word from ARG1.
> > +
> > +   If ELEMENT2 is null, use the top 64-bit double word of ARG2.  If it is
> > +   non-NULL, it is a 0 or 1 constant that gives the vector element number to
> > +   use for extracting the 64-bit double word from ARG2.
> > +
> > +   The element number is based on the user element ordering, set by the
> > +   endianess and by the -maltivec={le,be} options.  */
> 
> ("endianness", two n's).
> 
> I don't like using NULL as a magic value at all; it does not simplify
> this interface, it complicates it instead.
>
> Can you move the "which half is high" decision to the callers?

And then essentially there is no need for the function, since each of the 4
concat variants have to have the logic to support big endian, -maltivec=be, and
-maltivec=le.

Let me see what I can do about it.

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