[Mesa-dev] [PATCH 1/9] mesa: Handle common extension checks with more compact code

[Nanley Chery]

On Fri, May 19, 2017 at 05:32:07PM -0700, Ian Romanick wrote:
> On 05/19/2017 05:07 PM, Ilia Mirkin wrote:
> > On Fri, May 19, 2017 at 7:29 PM, Ian Romanick <idr at freedesktop.org> wrote:
> >> I missed that glsl_parser_extras.cpp has its own implementation of the
> >> has_XXX_foo() functions that take the API and version as explicit
> >> parameters.  Your patch predates that change.  There's room for some
> >> modest savings there too.
> > 
> > The main thing about those functions is that they have to be
> > out-of-line, as they're invoked via function pointer. Each one can be
> > cut down a little using your approach of statically looking at the cap
> > restrictions, but I don't think inlining can work. [I added that code
> > to (a) add "permission" checks to enable extensions in glsl and (b)
> > remove the duplication of adding #defines for all the ext names.]
> 
> Right.  I think there's two approaches to try.  The least intrusive is
> to try some of these tricks in the existing functions.  The more
> intrusive would be to generate the data into the table and modify
> _mesa_glsl_extension::compatible_with_state to use the data instead of
> calling the function.
> 
> I'm leaning towards the latter, but I haven't decided how I'd want to
> generate the data.  The method used by other users of
> main/extensions_table.h won't work because we only want a subset of
> extensions.  Hmm...
> 

I'll have to get back to this next week, but please don't feel blocked
on my comments. This may be the latter trick mentioned above, but if it
isn't, a third option is to initialize an array at context creation time
to describe which extensions are supported. That array is then looked up
in the helpers:
https://cgit.freedesktop.org/~nchery/mesa/log/?h=3/ext/init