No abstract

No abstract

Tiling is a key technique to reduce data movement in matrix computations. While tiling is well understood and widely used for dense matrix/tensor computations, effective tiling of sparse matrix computations remains a challenging problem. This paper proposes a novel method to efficiently summarize the impact of the sparsity structure of a matrix on achievable data reuse as a one-dimensional signature, which is then used to build an analytical cost model for tile size optimization for sparse matrix computations. The proposed model-driven approach to sparse tiling is evaluated on two key sparse matrix kernels: Sparse Matrix-Dense Matrix Multiplication (SpMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM). Experimental results demonstrate that model-based tiled SpMM and SDDMM achieve high performance relative to the current state-of-the-art.

No abstract

De novo genome assembly is a fundamental problem in computational molecular biology that aims to reconstruct an unknown genome sequence from a set of short DNA sequences (or reads) obtained from the genome. High throughput sequencers could generate several billions of such short reads in a single run. However, the relative ordering of the reads along the target genome is not known a priori. This lack of information is one of the main contributors to the increased complexity of the assembly process. Typically, state-of-the-art approaches produce an ordering of the reads toward the end of the assembly process, making it rather too late to benefit from the ordering information. In this paper, with the dual objective of improving assembly quality as well as exposing a high degree of parallelism for assemblers, we present a partitioning-based approach. Our framework—which we call BOA (for bucket-order-assemble)—uses a bucketing alongside graph- and hypergraph-based partitioning techniques to produce a partial ordering of the reads. This partial ordering enables us to divide the read set into disjoint blocks that can be independently assembled in parallel using any state-of-the-art serial assembler of choice. We tested the BOA framework on a variety of genomes. Experimental results show that the hypergraph variant of our approach, Hyper-BOA, consistently improves both the overall assembly quality and performance. For the inputs tested, the Hyper-BOA framework consistently improves the N50 values of the popular standalone MEGAHIT assembler by an average of 1.70× and up to 2.13×; while the largest alignment length improves 1.47× on average and up to 1.94×. The time to solution also consistently improves between 3-4× for the system sizes tested.