rapidsathkust / manymap

Accelerating Long Read Alignment on Three Processors

manymap

This repository contains the source code of the paper "Accelerating Long Read Alignment on Three Processors", by Zonghao Feng, Shuang Qiu, Lipeng Wang, and Qiong Luo.

Micro Benchmarks

Evaluation of the base-level alignment kernel. Run make to compile the following executables:

ksw_{sse2, sse41, avx2, avx512}: The original kernel in minimap2.

cpu_{sse2, sse41, avx2, avx512}: CPU version of the optimized kernel.

knl: KNL (Intel Xeon Phi processor) version.

gpu: GPU version. Variants: single block / multiple blocks with cooperative groups, global memory / shared memory (applicable to small data), score only / complete path

Macro Benchmarks

Evaluation of end-to-end runtime. Run make to compile. To use the MCDRAM on KNL, add numactl -m 1 before the run command.

Acknowledgment

This project is based on minimap2 (version 2.16).