In this paper, we describe the organization and microarchitecture of MT-MB, a configurable implementation of the Xilinx MicroBlaze soft processor that supports multithreading. Using a suite of synthetic benchmarks, we evaluate five variations of MT-MB and show that multithreading is very effective in hiding the variable latencies associated with custom instructions and custom computational units. Our experimental results show that interleaved and hybrid mul
Roger Moussali, Nabil Ghanem, Mazen A. R. Saghir