[Mesa-dev] [PATCH 1/6] nir/lower_tex: generalize get_texture_size()

[Iago Toral]

On Mon, 2016-12-05 at 20:38 -0800, Kenneth Graunke wrote:
> On Thursday, December 1, 2016 8:53:16 AM PST Iago Toral Quiroga
> wrote:
> > 
> > This was written specifically for RECT samplers. Make it more
> > generic so
> > we can call this from the gradient lowerings too.
> > ---
> >  src/compiler/nir/nir_lower_tex.c | 15 ++++++++++-----
> >  1 file changed, 10 insertions(+), 5 deletions(-)
> > 
> > diff --git a/src/compiler/nir/nir_lower_tex.c
> > b/src/compiler/nir/nir_lower_tex.c
> > index 0efd443..ccca59b 100644
> > --- a/src/compiler/nir/nir_lower_tex.c
> > +++ b/src/compiler/nir/nir_lower_tex.c
> > @@ -154,22 +154,27 @@ get_texture_size(nir_builder *b,
> > nir_tex_instr *tex)
> >  {
> >     b->cursor = nir_before_instr(&tex->instr);
> >  
> > -   /* RECT textures should not be array: */
> > -   assert(!tex->is_array);
> > -
> >     nir_tex_instr *txs;
> >  
> >     txs = nir_tex_instr_create(b->shader, 1);
> >     txs->op = nir_texop_txs;
> > -   txs->sampler_dim = GLSL_SAMPLER_DIM_RECT;
> > +   txs->sampler_dim = tex->sampler_dim;
> > +   txs->is_array = tex->is_array;
> > +   txs->is_shadow = tex->is_shadow;
> > +   txs->is_new_style_shadow = tex->is_new_style_shadow;
> >     txs->texture_index = tex->texture_index;
> > +   txs->texture = (nir_deref_var *)
> > +      nir_copy_deref(txs, &tex->texture->deref);
> > +   txs->sampler_index = tex->sampler_index;
> > +   txs->sampler = (nir_deref_var *)
> > +      nir_copy_deref(txs, &tex->sampler->deref);
> >     txs->dest_type = nir_type_int;
> >  
> >     /* only single src, the lod: */
> >     txs->src[0].src = nir_src_for_ssa(nir_imm_int(b, 0));
> >     txs->src[0].src_type = nir_tex_src_lod;
> >  
> > -   nir_ssa_dest_init(&txs->instr, &txs->dest, 2, 32, NULL);
> > +   nir_ssa_dest_init(&txs->instr, &txs->dest, tex-
> > >coord_components, 32, NULL);
> Hrm.  Won't this make it start returning 3 for array textures?  It
> looks
> like you want 3 for 3D, 1 for 1D, and 2 for everything else...
> 
> I guess you swizzle away the unnecessary components, but another
> option
> would be to just pass in the number of components you wanted, rather
> than emitting moves to scrap them.

Right, we end up only taking the first two components of the result,
but yeah, I can try this too. Thanks!

Iago