[Mesa-dev] [PATCH 07/10] nir: add imprecise flrp optimisation

[Timothy Arceri]

On Sat, 2017-01-07 at 11:14 -0800, Matt Turner wrote:
> On Sat, Jan 7, 2017 at 3:58 AM, Timothy Arceri
> <timothy.arceri at collabora.com> wrote:
> > On BDW:
> > 
> > total instructions in shared programs: 13061890 -> 13061877 (-
> > 0.00%)
> > instructions in affected programs: 2441 -> 2428 (-0.53%)
> > helped: 13
> > HURT: 0
> > 
> > total cycles in shared programs: 256612254 -> 256611784 (-0.00%)
> > cycles in affected programs: 16418 -> 15948 (-2.86%)
> > helped: 10
> > HURT: 2
> > ---
> >  src/compiler/nir/nir_opt_algebraic.py | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/src/compiler/nir/nir_opt_algebraic.py
> > b/src/compiler/nir/nir_opt_algebraic.py
> > index fa9f3f1..7cee378 100644
> > --- a/src/compiler/nir/nir_opt_algebraic.py
> > +++ b/src/compiler/nir/nir_opt_algebraic.py
> > @@ -111,6 +111,7 @@ optimizations = [
> >     (('~flrp', a, b, 1.0), b),
> >     (('~flrp', a, a, b), a),
> >     (('~flrp', 0.0, a, b), ('fmul', a, b)),
> > +   (('~flrp', a, 0.0, c), ('ffma', ('fneg', a), c, a), '!options-
> > >lower_ffma'),
> 
> I don't think we want to be generating ffma directly in
> nir_opt_algebraic?

We already have:

(('~fadd', ('fmul', a, b), c), ('ffma', a, b, c), 'options-
>fuse_ffma'),

Is you concern backends that don't want ffma? The !options->lower_ffma
should avoid  problems there. Or I could switch to using options-
>fuse_ffma (I assume i965 is using it).

> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev