Open Collections

VIPERS II : a soft-core vector processor with single-copy scratchpad memory

University of British Columbia

Previous work has demonstrated soft-core vector processors in FPGAs can be applied to speed up data-parallel embedded applications, while providing the users an easy-to-use platform to tradeoff performance and area. However, its performance is limited by load and store latencies, requiring extra software design effort to optimize performance. This thesis presents VIPERS II, a new vector ISA and the corresponding microarchitecture, in which the vector processor reads and writes directly to a scratchpad memory instead of the vector register file. With this approach, the load and store operations and their inherent latencies can often be eliminated if the working set of data fits in the vector scratchpad memory. Moreover, with the removal of load/store latencies, the user doesn't have to use loop unrolling to enhance performance, reducing the amount of software effort required and making the vectorized code more compact. The thesis shows the new architecture has the potential to achieve performance similar to that of the unrolled versions of the benchmarks, without actually unrolling the loop. Hardware performance results of VIPERS II demonstrated up to 47x speedup over a Nios II processor with only 13x more resources used.

Text

2010-04-15

Attribution-NonCommercial-NoDerivs 3.0 Unported

http://hdl.handle.net/2429/23595

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/3.0/