[Mesa-dev] [PATCH 05/14] nir: update opcode definitions for different bit sizes

[Iago Toral]

On Mon, 2016-03-14 at 14:24 -0400, Connor Abbott wrote:
> On Mon, Mar 14, 2016 at 2:10 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
(...)
> >>
> >> -opcode("ldexp", 0, tfloat, [0, 0], [tfloat, tint], "", """
> >> -dst = ldexpf(src0, src1);
> >> +opcode("ldexp", 0, tfloat32, [0, 0], [tfloat32, tint32], "", """
> >> +dst = ldexp(src0, src1);
> >
> >
> > For some things like ldexp, they could easily have a double variant.
> > However, for the moment, it's probably best to lock them to 32-bit until we
> > have a need for it.
> 
> Actually, wait... ARB_gpu_shader_fp64 does add support for ldexp on
> doubles, even if we lower it away before here. Maybe we should make it
> generic too?

Right, we should leave it generic. I've already changed this locally.

Iago