Use V_XAD_U32 for xor + add on Vega platforms that support inline assembly
JustinTArthur opened this issue · comments
Unfortunately might take changing the design of some of the macros.
Latest LLVM (e.g. in ROCm) is already making these optimizations.