[Mesa-dev] [PATCH 10/12] nir: Store num_direct_uniforms in the nir_shader.

[Jason Ekstrand]

On Wed, Apr 8, 2015 at 12:48 PM, Ian Romanick <idr at freedesktop.org> wrote:
> On 04/08/2015 12:06 AM, Kenneth Graunke wrote:
>> Storing this here is pretty sketchy - I don't know if any driver other
>> than i965 will want to use it.  But this will make it a lot easier to
>> generate NIR code at link time.  We'll probably rework it anyway.
>>
>> Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
>> ---
>>  src/glsl/nir/nir.h                       | 3 +++
>>  src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 5 +++--
>>  2 files changed, 6 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
>> index e844e4d..7d11996 100644
>> --- a/src/glsl/nir/nir.h
>> +++ b/src/glsl/nir/nir.h
>> @@ -1429,6 +1429,9 @@ typedef struct nir_shader {
>>      * access plus one
>>      */
>>     unsigned num_inputs, num_uniforms, num_outputs;
>> +
>> +   /** the number of uniforms that are only accessed directly */
>> +   unsigned num_direct_uniforms;
>>  } nir_shader;
>>
>>  #define nir_foreach_overload(shader, overload)                        \
>> diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
>> index 145a447..034b79a 100644
>> --- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
>> +++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
>> @@ -120,7 +120,7 @@ fs_visitor::emit_nir_code()
>>
>>     if (shader_prog) {
>>        nir_assign_var_locations_scalar_direct_first(nir, &nir->uniforms,
>> -                                                   &num_direct_uniforms,
>> +                                                   &nir->num_direct_uniforms,
>>                                                     &nir->num_uniforms);
>
> Why not just have nir_assign_var_locations_scalar_direct_first modify
> the nir_shader passed in?  That seems more concise.

We could, but I'm not a huge fan of storing it in the nir_shader to
begin with.  As I said in my review, it's a hack I'm willing to go
with for now.  However, since we made variable location assignment
separate from lowering, they are no longer a universal thing.  We
could also potentially have backend-specific assignment passes and I
really don't want to set a precedent of storing this data in
nir_shader.  I'm hoping (maybe foolishly) that we'll come up with a
better place to store them soon.
--Jason

>>     } else {
>>        /* ARB programs generally create a giant array of "uniform" data, and allow
>> @@ -128,7 +128,7 @@ fs_visitor::emit_nir_code()
>>         * analysis, it's all or nothing.  num_direct_uniforms is only useful when
>>         * we have some direct and some indirect access; it doesn't matter here.
>>         */
>> -      num_direct_uniforms = 0;
>> +      nir->num_direct_uniforms = 0;
>>     }
>>     nir_assign_var_locations_scalar(&nir->inputs, &nir->num_inputs);
>>     nir_assign_var_locations_scalar(&nir->outputs, &nir->num_outputs);
>> @@ -343,6 +343,7 @@ void
>>  fs_visitor::nir_setup_uniforms(nir_shader *shader)
>>  {
>>     uniforms = shader->num_uniforms;
>> +   num_direct_uniforms = shader->num_direct_uniforms;
>>
>>     /* We split the uniform register file in half.  The first half is
>>      * entirely direct uniforms.  The second half is indirect.
>>
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev