A GPU-inspired soft processor for high-throughput acceleration (thesis)
Department of Electrical and Computer Engineering, University of Toronto, Toronto, ON, Canada
University of Toronto, 2008
@phdthesis{kingyens2008gpu,
title={A GPU-inspired soft processor for high-throughput acceleration},
author={Kingyens, J.R.C.},
year={2008},
school={University of Toronto}
}
In this thesis a soft processor programming model and architecture is proposed that is inspired by graphics processing units (GPUs) and well-matched to the strengths of FPGAs, namely highly-parallel and pipelinable computation. The proposed soft processor architecture exploits multithreading, vector operations, and predication to supply a floating-point pipeline of up to 60 stages via hardware support for up to 256 concurrent thread contexts. The key new contributions of this architecture are mechanisms for managing threads and register files that maximize data-level and instruction-level parallelism while overcoming the challenges of port limitations of FPGA block memories, as well as memory and pipeline latency. Through simulation of a system that is (i) programmable via NVIDIA’s high-level Cg language, (ii) supports AMD’s r5xx GPU ISA, and (iii) is realizable on an XtremeData XD1000 FPGA-based accelerator system, it is demonstrated that the proposed soft processor can achieve 100% utilization of the deeply-pipelined floating-point datapath.
May 8, 2011 by hgpu
