Preliminary Results of Autotuning GEMM Kernels for the NVIDIA Kepler Architecture