High-Performance Genomic Analysis Heterogeneous System Using OpenCL.

Nowadays, genomic analysis significantly affects various fields, including vaccine development, viral evolution research, precision medicine, and crop breeding. Minimap2 [1] is widely used for long-read sequence alignment in third-generation sequencing. However, the chaining step in Minimap2 exposes a performance bottleneck, accounting for nearly 40% of the overall execution time. In this paper, we present a practical heterogeneous acceleration system for Minimap2 using OpenCL. We focus on three key optimizations: 1) Reordering the data flow in the scoring algorithm to maximize parallelism. 2) Introducing a coarse-grained batch-processing strategy to enhance data reuse and hide computation latency. 3) Implementing a hierarchical memory access strategy to optimize data transfer between software and hardware. Finally, we build a heterogeneous system with multi-threading and implement it on Intel Arria10 GX1150 FPGA. Our experimental results demonstrate a significant performance improvement compared to the software baseline. We achieve a 7.23 times speed-up for the chaining scoring, while the complete system achieves a 1.25 times speed-up without any loss in accuracy.