Glad that's working out for you. Out of curiosity, does this work? vcvt.f32.u32 q0, q0 vrecpe.f32 q0, q0 vcvt.u32.f32 q0, q0, #16
vcvt.f32.u32 q0, q0 vrecpe.f32 q0, q0 vcvt.u32.f32 q0, q0, #16