AutoDSE: Enabling Software Programmers to Design Efficient FPGA Accelerators

Adopting FPGA as an accelerator in datacenters is becoming mainstream for customized computing, but the fact that FPGAs are hard to program creates a steep learning curve software programmers. Even with help of high-level synthesis (HLS) , designers still have manually perform code reconstruction and cumbersome parameter tuning achieve optimal performance. While many models been leveraged by existing work automate design efficient accelerators, unpredictability modern HLS tools becomes major obstacle them maintain high accuracy. To address this problem, we propose automated DSE framework— AutoDSE —that leverages bottleneck-guided coordinate optimizer systematically find better point. detects bottleneck each step focuses on high-impact parameters overcome it. The experimental results show able identify point achieves, geometric mean, 19.9× speedup over one CPU core MachSuite Rodinia benchmarks. Compared optimized vision kernels Xilinx Vitis libraries, can reduce their optimization pragmas 26.38× while achieving similar With less than pragma per average, making progress towards democratizing customizable computing enabling programmers accelerators.