[Mesa-dev] [PATCH v5 45/70] glsl: atomic counters can be declared as buffer-qualified variables

[Iago Toral]

On Mon, 2015-09-28 at 15:23 +0300, Francisco Jerez wrote:
> Iago Toral <itoral at igalia.com> writes:
> 
> > On Mon, 2015-09-28 at 13:13 +0300, Francisco Jerez wrote:
> >> Iago Toral Quiroga <itoral at igalia.com> writes:
> >> 
> >> > From: Kristian Høgsberg <krh at bitplanet.net>
> >> >
> >> > ---
> >> >  src/glsl/ast_to_hir.cpp | 6 +++---
> >> >  1 file changed, 3 insertions(+), 3 deletions(-)
> >> >
> >> > diff --git a/src/glsl/ast_to_hir.cpp b/src/glsl/ast_to_hir.cpp
> >> > index 566cc87..a364aae 100644
> >> > --- a/src/glsl/ast_to_hir.cpp
> >> > +++ b/src/glsl/ast_to_hir.cpp
> >> > @@ -2789,7 +2789,7 @@ apply_type_qualifier_to_variable(const struct ast_type_qualifier *qual,
> >> >     }
> >> >  
> >> >     if (var->type->contains_atomic()) {
> >> > -      if (var->data.mode == ir_var_uniform) {
> >> > +      if (var->data.mode == ir_var_uniform || var->data.mode == ir_var_shader_storage) {
> >> >           if (var->data.explicit_binding) {
> >> >              unsigned *offset =
> >> >                 &state->atomic_counter_offsets[var->data.binding];
> >> > @@ -2807,8 +2807,8 @@ apply_type_qualifier_to_variable(const struct ast_type_qualifier *qual,
> >> >           }
> >> >        } else if (var->data.mode != ir_var_function_in) {
> >> >           _mesa_glsl_error(loc, state, "atomic counters may only be declared as "
> >> > -                          "function parameters or uniform-qualified "
> >> > -                          "global variables");
> >> > +                          "function parameters, uniform-qualified or "
> >> > +                          "buffer-qualified global variables");
> >> >        }
> >> >     }
> >> >  
> >> 
> >> Spec quote?  Declaring an atomic counter to have buffer storage is
> >> illegal AFAIK.
> >
> > I think you are right, the specs don't seem to include any explicit
> > restrictions affecting the allowed types for buffer variables, however,
> > the description of opaque types seems to be in conflict with that:
> >
> > "The opaque types declare variables that are effectively opaque 
> > handles to other objects. These objects are
> > accessed through built-in functions, not through direct reading or
> > writing of the declared variable.
> > (...)
> > Opaque variables cannot be treated as l-values;(...)"
> >
> > I am thinking that we probably want to revert this patch and instead
> > check that buffer variables do not contain references to any opaque
> > types and produce a compile-time error if that is the case. I'll send
> > patches for this. Maybe the specs should be amended to mention this
> > explicitly too. Ian, what do you think?
> >
> They mention it explicitly AFAICT (section 4.1.7 of the GL 4.5 spec):
> 
>  "[Opaque Types] can only be declared as function parameters or in
>   uniform-qualified variables."

Right, I meant that this is not explicit in the description of buffer
variables.

Iago