Skip to content

nir: move the brw mem access lowering pass to nir

Dave Airlie requested to merge airlied/mesa:brw-nir-generic-mem-access into main

This pass has uses for aco/radeonsi/opencl triangle of doom.

  • Handle load_kernel_input
  • Handle 8-bit vec16 correctly.
  • Make less Intel specific by adding a callback.
  • Transition ntt to use this.
Edited by Alyssa Rosenzweig

Merge request reports