[Mesa-dev] [PATCH V4 2/6] glsl: assign hidden uniforms their slot id earlier

[Marek Olšák]

On Sun, Oct 11, 2015 at 9:20 AM, Timothy Arceri <t_arceri at yahoo.com.au> wrote:
> On Sat, 2015-10-10 at 18:06 +0200, Marek Olšák wrote:
>> Hi Timothy,
>>
>> One of these 3 commits breaks compilation for Talos shaders with
>> gallium. My piglit patch "glsl-1.30/sampler-bug: ..." contains a
>> minimal test case. I can't say which commit, because Mesa fails to
>> build between them.
>>  It has something to do with uniforms, structures,
>> and samplers.
>>
>> commit dcd9cd03837545055ce2a315e7e8840cc3254d1a
>> Author: Timothy Arceri <t_arceri at yahoo.com.au>
>> Date:   Sun Aug 30 12:50:34 2015 +1000
>>
>>     glsl: store uniform slot id in var location field
>>
>
> Hi Marek,
>
> The piglit test passes on my intel based laptop I can't test on
> anything else until later this week (as I'm traveling) but my guess is
> its something to do with the above patch.
>
> The other two patches shouldn't change anything for gallium drivers
> "glsl: assign hidden uniforms their slot id earlier" just assigns
> hidden uniforms their slot id earlier and there shouldn't be any
> difference once the IR gets to glsl_to_tgsi.
>
> Also "glsl: order indices for samplers inside a struct array" shouldn't
> change things either as in your test the sampler are not inside the
> struct.
>
> There is some code in the glsl_to_tgsi pass that looks like the
> location field would have always been -1 for uniforms other the UBO and
> UBO members maybe this has something to do with the problem now that
> all uniforms now get a non -1 value.
>
>       case ir_var_uniform:
>          entry = new(mem_ctx) variable_storage(var, PROGRAM_UNIFORM,
>                                                var->data.location);
>          this->variables.push_tail(entry);
>          break;
>
> I hope this helps get you started. If you haven't figured it out by
> later in the week than I'll take a look on my desktop once I get home.

The problem is that _mesa_get_sampler_uniform_value returns a sampler
index which is greater than 16. If I allow large sampler indices, it
starts to fail with sampler.file==TGSI_FILE_NULL in the TGSI codegen
later. I didn't investigate further.

Marek