[Mesa-dev] Initial GS NIR support for radeonsi

[Timothy Arceri]

On 15/11/17 22:06, Nicolai Hähnle wrote:
> On 10.11.2017 04:13, Timothy Arceri wrote:
>> The support is still WIP but the patches as starting to pile
>> up so thought I'd see if I could land these before continuing.
> 
> Thanks for working on this!
> 
> 
>> Whats missing?
>>
>> Vega support for gs_vtx_offset handling (see patch 3), I don't
>> have one yet for testing so didn't attempt to adapt the code.
> 
> The existing code should work though, right? It's actually not clear to 
> me what patch 3 is really trying to achieve, since the caller provides a 
> plain integer vertex index either way. It's a nice cleanup though.

The nir to llvm path accesses these via an array index (expecting no 
gaps) we just need to update the Vega path to use an array also.

> 
> 
>> Lots of piglit tests still fail. The gs clip distance piglit
>> tests currently cause my system to hang for instance.
> 
> This is usually a mismatch/disagreement between SPI_SHADER_POS_FORMAT 
> and PA_CL_VS_OUT_CNTL: the PA tries to pull more data from the position 
> cache than is allocated there by the SPI, and the hardware gets 
> permanently confused. Probably something in the setup for the GS copy 
> shader needs to be fixed.

Yeah I'm got this fixed now.

> 
> The series looks pretty good already. I'd second Marek's comments, and I 
> think the load_inputs ABI should be reworked to not depend on NIR; I've 
> sent a comment on patch 17 for this.

I've sent a reply I think you missed what I'm trying to work around there :)

> 
> Patches 2-4, 8, 14-16, 18 and 19:

Thanks.

> 
> Reviewed-by: Nicolai Hähnle <nicolai.haehnle at amd.com>
> 
> Cheers,
> Nicolai