Skip to content

NAK: NIR opcodes that could be supported

I think there are some NIR ALU opcodes that nvidia supports too:

  • has_fmulz: d3d9 performance with dxvk, but I'm was told there are issues with denorms on nvidia hw
  • has_pack_half_2x16_rtz: DXBC's f2f16 is rtz, so dxvk/vkd3d-proton emit a lot of code to avoid inf results. NV rounding mode is per instruction
  • has_find_msb_rev: DXBC/DXIL only has rev versions, should match PTX's bfind.shiftamt
  • has_bfm: should match PTX bmsk.wrap
Edited by Georg Lehmann
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information