[Mesa-dev] Precision qualifiers in the IR

[Iago Toral]

On Thu, 2015-01-08 at 01:49 -0800, Kenneth Graunke wrote:
> On Thursday, January 08, 2015 10:21:45 AM Iago Toral wrote:
> > Hi,
> > 
> > In GLSL ES when the same uniform is declared in multiple shaders it must
> > have the same precision qualifier. There are a few dEQP tests that fail
> > because this is not checked in Mesa.
> > 
> > I think the right place to add this check would be in
> > cross_validate_globals(), but to do this we would need to access
> > precision qualifiers from an ir_variable (specifically, I think we need
> > this to be in glsl_type so we can do this check with uniform structs
> > too, which is what the tests actually check).
> > 
> > I see precision qualifiers being parsed and stored in the AST, but I
> > don't see where this information is passed to the IR: ir_variable or
> > glsl_type don't have this info, in fact,
> > apply_type_qualifier_to_variable() in ast_to_hir.cpp seems to ignore the
> > ast precision qualifier data completely.
> > 
> > So I am guessing that we should add precision information to the
> > glsl_type, but I wonder if there is a reason why this hasn't been done
> > yet, considering that there seems to be some kind of support for
> > precision qualifiers already, at least in the parser... am I missing
> > something?
> > 
> > Iago
> 
> When we implemented the compiler, nobody really cared about lower precision.
> Desktop GLSL is actually spec'd to parse and ignore the qualifiers - they
> literally are mandated to mean nothing - so they're only useful for ES, and
> we've seen very few ES apps.
> 
> Making lowp, mediump, and highp the same precision is valid, and at the time,
> we were mostly targeting desktop hardware that didn't have a lower precision
> option anyway.  These days, we could potentially use it for half float.
> 
> Hooking it up isn't a bad plan - we just haven't gotten around to it.
> It does seem like we need to at least be tracking it so we can properly
> enforce the rules, and adding that would be a good step towards actually
> doing something with it.
> 
> --Ken

I see, I'll try to add them together with the necessary checks then.
Thanks for the quick reply!

Iago