[Mesa-dev] [RFC 01/16] glsl: Add tracking for GLSL precision qualifiers

[Pohjolainen, Topi]

On Mon, May 18, 2015 at 11:19:13AM +0300, Petri Latvala wrote:
> On 05/15/2015 12:39 PM, Topi Pohjolainen wrote:
> >diff --git a/src/glsl/glsl_types.h b/src/glsl/glsl_types.h
> >index 5645dcd..25c4d30 100644
> >--- a/src/glsl/glsl_types.h
> >+++ b/src/glsl/glsl_types.h
> >@@ -100,6 +100,13 @@ enum glsl_matrix_layout {
> >     GLSL_MATRIX_LAYOUT_ROW_MAJOR
> >  };
> >+enum {
> >+   GLSL_PRECISION_NONE = 0,
> >+   GLSL_PRECISION_HIGH,
> >+   GLSL_PRECISION_MEDIUM,
> >+   GLSL_PRECISION_LOW
> >+};
> >+
> >  #ifdef __cplusplus
> >  #include "GL/gl.h"
> >  #include "util/ralloc.h"
> >@@ -768,6 +775,11 @@ struct glsl_struct_field {
> >      * streams (as in ir_variable::stream). -1 otherwise.
> >      */
> >     int stream;
> >+
> >+   /**
> >+    * Precission qualifier
> >+    */
> >+   unsigned precision;
> >  };
> 
> Typo in the comment, "Precision".

Thanks, although I think this is most likely picked out by Iago if he
continues to work on the link-state validation of precision qualifiers.
I included this just for the context sake (see the cover letter). This
wasn't needed by the rest of the series.

> 
> Is there a piglit test for uniforms with different precisions? That should
> be mentioned in the commit log imho.

Tapani and I briefly looked at the existing tests last week and we concluded
that most likely we need a bunch more. However, I'd like to get the
implementation work started first to get better idea what sort of corner cases
are we looking at from Mesa's point of view.