[Mesa-dev] [RFC 31/31] nir: Add a bool to float32 lowering pass

[Jason Ekstrand]

On Mon, Oct 22, 2018 at 6:13 PM Ilia Mirkin <imirkin at alum.mit.edu> wrote:

> On Mon, Oct 22, 2018 at 6:16 PM Jason Ekstrand <jason at jlekstrand.net>
> wrote:
> >
> > This should be useful for drivers that don't support real integers.
> >
> > Cc: Alyssa Rosenzweig <alyssa at rosenzweig.io>
> > ---
> >  src/compiler/Makefile.sources              |   1 +
> >  src/compiler/nir/meson.build               |   1 +
> >  src/compiler/nir/nir_lower_bool_to_float.c | 181 +++++++++++++++++++++
> >  3 files changed, 183 insertions(+)
> >  create mode 100644 src/compiler/nir/nir_lower_bool_to_float.c
> > +   nir_foreach_block(block, impl) {
> > +      nir_foreach_instr_safe(instr, block) {
> > +         switch (instr->type) {
> > +         case nir_instr_type_alu:
> > +            progress |= lower_alu_instr(&b, nir_instr_as_alu(instr));
> > +            break;
> > +
> > +         case nir_instr_type_load_const: {
> > +            nir_load_const_instr *load = nir_instr_as_load_const(instr);
> > +            if (load->def.bit_size == 1) {
> > +               nir_const_value value = load->value;
> > +               for (unsigned i = 0; i < load->def.num_components; i++)
> > +                  load->value.u32[i] = value.b[i] ? NIR_TRUE :
> NIR_FALSE;
> > +               load->def.bit_size = 32;
> > +               progress = true;
> > +            }
> > +            break;
> > +         }
>
> Should this instead rewrite the load_const to a 1.0f / 0.0f?
>

Yup.  As you can tell, this is completely untested.  However, I think it
does make the point that 1-bit bools aren't bad for float-only hardware
which is why I did it.  I've fixed the bugs pointed out so far.

--Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20181022/e73f8743/attachment.html>