[Bug 89580] Implement a NIR -> vec4 pass

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=89580

--- Comment #15 from Jason Ekstrand <jason at jlekstrand.net> ---
(In reply to Eduardo Lima Mitev from comment #14)
> (In reply to Kenneth Graunke from comment #13)
> > Yeah, that definitely looks bizarre - but I think it's expected.
> > 
> > Fixed-function vertex programs and ARB programs go through prog_to_nir.c
> > (the Mesa IR -> NIR translator), which generates load_uniform intrinsics
> > directly, instead of creating uniform variables then having nir_lower_io
> > lower variable access to intrinsics.  It just sets nir->num_uniforms
> > directly, and brw_nir.c skips the part that inspects uniform variables when
> > dealing with FF/ARB programs.
> > 
> >    /* Set the number of uniforms */
> >    shader->num_uniforms = 4 * c->prog->Parameters->NumParameters;
> > 
> 
> Thanks Jason and Kenneth! I managed to fix my implementation based on your
> comments.
> 
> Just for the sake of clarity, I wonder if it is worth "fixing" the
> inconsistency in nir_shader for these cases, perhaps injecting the missing
> declarations or something like that. Dunno, probably not.

I don't think so.  I think what we want is a longer-term fix that provides some
way to declare in the nir_shader what uniform ranges get used and what to do
with them.  Unfortunately, that's kind of ill-defined at the moment.  I'm
hoping to get around to making it better-defined soon.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20150506/0a15f2fd/attachment.html>