LIBXSMM: Accelerating Small Matrix Multiplications by Runtime Code Generation | IEEE Conference Publication | IEEE Xplore