On Mon, Apr 30, 2012 at 11:45:50AM -0700, Eric Anholt wrote:
> I'm concerned that this will turn things that shouldn't be constant
> expressions in GLSL (some user-defined function just using its
> (constant) arguments and constants) and treating them as constant
> expressions. Do we have tests for that?
The if (!this->is_builtin) return NULL has not been touched.
> Our variables don't generally have unique names. It looks like from
> your usage, that you could use pointer_hash/compare and get cheaper
> searches plus unique references to variables.
Done that, very nice. Avoids scoping problems with if blocks, which
is a nice bonus.
>
> > + const exec_node *parameter_info = parameters.head;
> >
> > - /* Check if all parameters are constant */
> > - ir_constant *op[3];
> > foreach_list(n, actual_parameters) {
> > - ir_constant *constant = ((ir_rvalue *)
> > n)->constant_expression_value();
> > + ir_constant *constant = ((ir_rvalue *)
> > n)->constant_expression_value(NULL);
> > if (constant == NULL)
> > return NULL;
>
> Why drop the "check if all parameters are constant" code?
The if (constant == NULL) return NULL; is the check, and it is still
there. Only the comment has morphed into "Check constness along the
way." in a higher-level comment block.
OG.
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
