论文信息 - Hardware Acceleration with Multi-Threading of Java-Based High Level Synthesis Tool

Hardware Acceleration with Multi-Threading of Java-Based High Level Synthesis Tool

In this research, we attempt to speed up the computational fluid dynamics (CFD) and the convolutional neural network (CNN) using JavaRock-Thrash thread function of the high-level synthesis tool with an FPGA. In the two-dimensional heat equation, by using the thread function of the high-level synthesis tool, up to a 12.13 times speedup compared to single-threaded processing is obtained with multi-threading, up to a 29.0 times speedup against Vivado HLS is achieved. In the convolution process, the process of passing 11 x 11 filters on 2-dimensional data of 33 x 33 described with 484 threads results in a speedup of 78 times compared to the processing time at Vivado HLS.

Hironori Nakajo | Takefumi Miyoshi | Yuto Ishikawa | Keitaro Yanai | Keisuke Koike

[1] Jason Cong,et al. Optimizing FPGA-based Accelerator Design for Deep Convolutional Neural Networks , 2015, FPGA.

[2] Brad L. Hutchings,et al. Sea Cucumber: A Synthesizing Compiler for FPGAs , 2002, FPL.

[3] Maya Gokhale,et al. Stream-oriented FPGA computing in the Streams-C high level language , 2000, Proceedings 2000 IEEE Symposium on Field-Programmable Custom Computing Machines (Cat. No.PR00871).

[4] Jason Cong,et al. Minimizing Computation in Convolutional Neural Networks , 2014, ICANN.

[5] Yoshua Bengio,et al. Gradient-based learning applied to document recognition , 1998, Proc. IEEE.

[6] Yasunori Nagata,et al. A feasibility study on implementing numerical applications on FPGAs using Vivado HLS , 2015 .