Skip to content

Draft: nir: Smarter memcpy lowering

Faith Ekstrand requested to merge gfxstrand/mesa:nir/smarter-memcpy into main

We no longer unroll arbitrarily large memcpys. Also, if we do fall back to emitting a loop, we're a lot smarter about it and try to do a bunch of wide loads first and then fall back to bytes.

Only build tested.

Merge request reports