Towards numerical benchmark for half-precision floating point arithmetic | IEEE Conference Publication | IEEE Xplore