arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2312.11279cs.AR
收藏

FPGA(也能获得一些)SATisfaction

FPGAs (Can Get Some) SATisfaction

Hariprasadh Godindasamy, Babak Esfandiari, Paulo Garcia

AI总结:

针对处理器/FPGA混合平台,本文提出通过运行时热插拔子句分配利用细粒度并行的开源硬件加速SAT求解器,公式规模仅受外部内存限制,并在Xilinx Zynq上取得最高6倍加速。

摘要:

我们提出了一种适用于处理器/现场可编程门阵列(FPGA)混合平台的硬件加速 SAT 求解器,此类平台已成为嵌入式领域的常态。我们的方案解决了 SAT 求解加速中的一个已知瓶颈:不同于先前通过限制可利用并行度来解决同一瓶颈的最先进方案,我们的求解器通过在运行时热插拔 FPGA 子句分配来利用细粒度并行化机会。它也是首个现代完全开源的 SAT 加速器,且公式大小仅受可用外部存储器容量限制,而非 FPGA 片上存储器。评估在 Xilinx Zynq 平台上进行:实验表明,取决于公式划分策略,硬件加速可在不同公式规模下缩短执行时间。在两个代表性基准测试上,我们分别比先前最先进方案快 1.7 倍和 1.1 倍,并且相比纯软件实现实现了最高 6 倍的性能提升。

英文摘要:

We present a hardware-accelerated SAT solver suitable for processor/Field Programmable Gate Arrays (FPGA) hybrid platforms, which have become the norm in the embedded domain. Our solution addresses a known bottleneck in SAT solving acceleration: unlike prior state-of-the-art solutions that have addressed the same bottleneck by limiting the amount of exploited parallelism, our solver takes advantage of fine-grained parallelization opportunities by hot-swapping FPGA clause assignments at runtime. It is also the first modern completely open-source SAT accelerator, and formula size is limited only by the amount of available external memory, not by on-chip FPGA memory. Evaluation is performed on a Xilinx Zynq platform: experiments support that hardware acceleration results in shorter execution time across varying formula sizes, subject to formula partitioning strategy. We outperform prior state-of-the-art by 1.7x and 1.1x, respectively, for 2 representative benchmarks, and boast up to 6x performance increase over software-only implementation.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑