An adaptive vector parallel supported transcendental function unit for stream processor

With the stream processors applied widely in high performance computing, the bandwidth in accessing memory has become a more primary bottleneck. Since reducing memory accessing time will improve the effective band width, a stream processor with a transcendental function unit is addressed in this paper, which is based on the Cordic algorithm. The proposed unit works in the stream processor data path with additional instructions, which supports the different vector parallel width adaptively. The whole stream processor which is optimized with the transcendental function unit is implemented in an FPGA platform with 100MHz frequency. The performance is evaluated with different benchmarks, the maximum speed-up ratio is up to 3.2 compared with the primal stream processor.