SALoBa: Maximizing Data Locality and Workload Balance for Fast Sequence Alignment on GPUs

Abstract

Sequence alignment forms an important backbone in many sequencing applications. A commonly used strategy for sequence alignment is an approximate string matching with a two-dimensional dynamic programming approach. Although some prior work has been conducted on GPU acceleration of a sequence alignment, we identify several shortcomings that limit exploiting the full computational capability of modern GPUs. This paper presents SALoBa, a GPU-accelerated sequence alignment library focused on seed extension. Based on the analysis of previous work with real-world sequencing data, we propose techniques to exploit the data locality and improve workload balancing. The experimental results reveal that SALoBa significantly improves the seed extension kernel compared to state-of-the-art GPU-based methods.

AGAThA: Fast and Efficient GPU Acceleration of Guided Sequence Alignment for Long Read Mapping

“…In this section, we explain the techniques used in the stateof-the-art methods [1,5,42] for accelerating sequence alignment on GPUs with CUDA support. Input Packing.…”

“…Intra-query Parallelism. Intra-query parallelism [5,42] is an essential technique to exploit massive parallelism on the GPU. It assigns multiple threads to an alignment task as shown in Figure 2 (b).…”

“…Since threads calculate each block on the same anti-diagonal in sync, high-numbered threads start late and low-numbered threads end early. This can be mitigated by splitting a warp into smaller subwarps [17,42] and assigning a task to each subwarp as in Figure 2 (c). This reduces external fragmentation at the cost of warp divergence (internal fragmentation), but the benefits surpass the penalties.…”

“…• SALoBa [42] utilizes intra-query parallelism. We applied banding heuristic that gives further speedup, similar to GASAL2.…”

“…However, to the extent of our knowledge, existing attempts compromise the structure [12] or target different alignment algorithms [1,42,57]. This is problematic because read alignment is not a stand-alone application, but an early part of complicated bioinformatics pipelines.…”