[x86] do not issue invalid vec_select from V2DI

[Richard Henderson]

On Thu, Jan 05, 2006 at 03:29:05PM -0200, Alexandre Oliva wrote:
> Aah.  So how about this for SSE1, then?

Is this coming up with sse1?  I wouldn't think so...

> Even if additional copying (involving reformatting anyway?) would be
> necessary to get the `0' constraint to fit?

Copying does not invoke reformatting, afaik.  As for whether or
not the extra copy is more or less expensive than two reformats,
I don't know.  I would suspect less, but that's just a WAG.

>    movhpd\t{%1, %0|%0, %1}
>    movlpd\t{%H1, %0|%0, %H1}

But for these two, the sse1 insns are smaller, so we should just
use those.  No reason to prefer the "d" insns.


r~