[Mesa-dev] [PATCH v2 3/4] intel/nir: Split IO arrays into elements

[Jason Ekstrand]

On July 31, 2018 14:54:54 Kenneth Graunke <kenneth at whitecape.org> wrote:

> On Tuesday, July 31, 2018 12:22:02 PM PDT Jason Ekstrand wrote:
>> The NIR nir_lower_io_arrays_to_elements pass attempts to split I/O
>> variables which are arrays or matrices into a sequence of separate
>> variables.  This can help link-time optimization by allowing us to
>> remove varyings at a more granular level.
>>
>> Shader-db results on Kaby Lake:
>>
>> total instructions in shared programs: 15177645 -> 15168494 (-0.06%)
>> instructions in affected programs: 79857 -> 70706 (-11.46%)
>> helped: 392
>> HURT: 0
>> ---
>> src/intel/compiler/brw_nir.c | 4 ++++
>> 1 file changed, 4 insertions(+)
>>
>> diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
>> index 17ccfa48af6..29ad68fdb2a 100644
>> --- a/src/intel/compiler/brw_nir.c
>> +++ b/src/intel/compiler/brw_nir.c
>> @@ -709,6 +709,10 @@ void
>> brw_nir_link_shaders(const struct brw_compiler *compiler,
>>              nir_shader **producer, nir_shader **consumer)
>> {
>> +   nir_lower_io_arrays_to_elements(*producer, *consumer);
>> +   nir_validate_shader(*producer);
>> +   nir_validate_shader(*consumer);
>> +
>> NIR_PASS_V(*producer, nir_remove_dead_variables, nir_var_shader_out);
>> NIR_PASS_V(*consumer, nir_remove_dead_variables, nir_var_shader_in);
>
> This actually passes the test suite?  That's strange.  When I was
> playing with this earlier, it was demoting gl_ClipDistance[1] from
> a float[1] compact array to a single float variable...but leaving it
> marked compact.  var->data.compact = true makes no sense for non-arrays.
>
> I think our clip/cull distance handling assumes that it's an array,
> though.  So...not sure whether the pass should unset compact, or skip
> over compact arrays...

Yes, it passes just fine. I'm not sure exactly how it interacts with clip 
distance lowering though.