[Mesa-dev] [PATCH v2 004/103] i965/vec4/nir: Add bit-size information to types

[Iago Toral]

On Wed, 2016-10-19 at 10:39 -0700, Ian Romanick wrote:
> On 10/19/2016 03:53 AM, Iago Toral wrote:
> > 
> > On Tue, 2016-10-18 at 17:16 -0700, Ian Romanick wrote:
> > > 
> > > On 10/11/2016 02:01 AM, Iago Toral Quiroga wrote:
> > > > 
> > > > @@ -811,7 +811,7 @@
> > > > vec4_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr)
> > > >        if (const_offset) {
> > > >           offset = brw_imm_ud(const_offset->u32[0] & ~15);
> > > >        } else {
> > > > -         offset = get_nir_src(instr->src[1], nir_type_int, 1);
> > > > +         offset = get_nir_src(instr->src[1], nir_type_uint32,
> > > > 1);
> > > Does it matter that this changed form int to uint32?
> > At some point in the past we made brw_type_for_nir_type() only
> > accept
> > bit-sized alu types, so this patch was necessary for things to
> > work.
> > That's not the case now so this patch should not be necessary
> > anymore
> > (a quick run of the fp64 tests shows no regressions without this
> > patch). I guess it does not hurt to add the bit-size information
> > when
> > it is known though, but I think it is safe to drop this if you
> > think it
> > is not worth it.
> Uh... I was just wondering why the sized version is nir_type_uint32
> instead of nir_type_int32 as I would have expected.

Ah, yes, it did not fix anything as far as I know, but since this is an
offset into a buffer I guess it should always be a positive number, so
I changed the type to unsigned. This is also consistent with the use of
brw_imm_ud() for the constant offset case.

Iago