Skip to content

microsoft/compiler: Run nir_lower_flrp before converting to DXIL

Jesse Natalie requested to merge jenatali/mesa:dxil-lower-flrp into msclc-d3d12

Looks like algebraic optimizations only support lowering this if it doesn't need to be exact. To be able to lower exact flrp, we need to run the explicit pass.

Without this, I was seeing some flrp instructions trying to run Photoshop workloads. Extracting this from !42 (merged).

Merge request reports