On Wed, Apr 14, 2021 at 02:38:47PM -0500, Segher Boessenkool wrote:
> On Fri, Apr 09, 2021 at 10:43:58AM -0400, Michael Meissner wrote:
> > (Fv mode attribute): Add KFmode and TFmode.
> > (mov<FPMASK:mode><FPMASK2:mode>cc_fpmask): Replace
> > mov<SFDF:mode><SFDF2:mode>cc_p9. Add IEEE 128-bit fp support.
> > (mov<FPMASK:mode><FPMASK2:mode>cc_invert_fpmask): Replace
> > mov<SFDF:mode><SFDF2:mode>cc_invert_p9. Add IEEE 128-bit fp
> > support.
> > (fpmask<mode>): Add IEEE 128-bit fp support. Enable generator to
> > build te RTL.
> > (xxsel<mode>): Add IEEE 128-bit fp support. Enable generator to
> > build te RTL.
>
> > @@ -608,8 +621,13 @@ (define_mode_attr Ff [(SF "f") (DF "d") (DI
> > "d")])
> > ; SF/DF constraint for arithmetic on VSX registers using instructions
> > added in
> > ; ISA 2.06 (power7). This includes instructions that normally target DF
> > mode,
> > ; but are used on SFmode, since internally SFmode values are kept in the
> > DFmode
> > -; format.
> > -(define_mode_attr Fv [(SF "wa") (DF "wa") (DI "wa")])
> > +; format. Also include IEEE 128-bit instructions which are restricted to
> > the
> > +; Altivec registers.
> > +(define_mode_attr Fv [(SF "wa")
> > + (DF "wa")
> > + (DI "wa")
> > + (KF "v")
> > + (TF "v")])
>
> Eww. Please just split the patterns. Fv should just go away, it is
> always "wa" currently. Removing that cascades to more cleanups, which
> is why I haven't done it yet, it takes time.
The problem is you have a combinatorial explosion. Right now, there are two
patterns, one for the normal move, and one for the inverted move. Without
doing a cascaded combination, you would need some 32 patterns to cover all of
the possibilities.
Or you give up on having a conditional move that compares one type and moves a
second:
_Float128 a, b;
double c, d, r;
r = (a == b) ? c : d;
As I recall when I put the original logic in, there were a few places that did
this mixed comparison between SF/DF modes was used in real code.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: [email protected], phone: +1 (978) 899-4797
