Skip to content

nir: set io mask correctly for compact vars.

Dave Airlie requested to merge airlied/mesa:nir-rfc-inputs-read-slots into main

What does this MR do and why?

I started converting llvmpipe to use NIR in most places and fell over a regression with clip dist compact floats,

The patch fixes it, but is it correct I suppose is the question.

nir: set io mask correctly for compact vars.

It needs two changes, one to use the nir helper for compact var, but also one to handle compact derefs like float[1] where the deref->type ends up as float, so glsl_get_length() returns 0.

Edited by Dave Airlie

Merge request reports