08:46 karolherbst: do we have a pass that can lower vec4s to vec2? I kinda think there was something, but...
08:46 karolherbst: I can't find it
08:47 karolherbst: I mean alu operations
08:50 karolherbst: I guess the usual plan is scalarize and hope one cna vectorize it back later
09:47 MrCooper: mahkoh: IIRC mareko said somewhere around 20-30% difference between linear and tiled with AMD GPUs
12:21 mahkoh: MrCooper: do you have a link? I'm wondering if using udmabuf directly on iGPUs might do more harm than good.
22:01 alyssa: karolherbst: nir_lower_alu_width()
22:01 karolherbst: ahh yeah.. forgot to mention that I already found it, but thanks!