A Toolchain for Dynamic Function Off-load on CPU-FPGA Platforms

Abstract

This new toolchain for accelerating application on CPU-FPGA platforms, called Courier-FPGA, extracts runtime information from a running target binary, and re-constructs the function call graph including input-output data. Then, it synthesizes hardware modules on the FPGA and makes software functions on CPU by using Pipeline Generator. The Pipeline Generator also builds a pipeline control program by using Intel Threading Building Block (Intel TBB) to run both hardware modules and software functions in parallel. Finally, Courier-FPGA's Function Off-loader dynamically replaces and off-loads the original functions in the binary by using the built pipeline. Courier-FPGA performs the off-loading without user intervention, source code tweaks or re-compilations of the binary. In our case studies, Courier-FPGA was used to accelerate a histogram-of-gradients (HOG) feature detection program on the Zynq platform. A series of functions were off-loaded, and the program was sped up 3.98 times by using the built pipeline.