cupyx.jit.warpsize# cupyx.jit.warpsize = <Data code = "32", type = unsigned int># Returns the number of threads in a warp. In CUDA this is always 32, and in ROCm/HIP always 64. See also numba.cuda.warpsize