For years, SIMD/vector units have enhanced the capabilities of modern CPUs in High-Performance Computing (HPC) and mobile technology. Typical commercially-available SIMD units process up to 8 double-precision elements with one instruction. The optimal vector width and its impact on CPU throughput due to memory latency and bandwidth remain challenging research areas. This study examines the behavior of four computational kernels on a RISC-V core connected to a customizable vector unit, capable of operating up to 256 double precision elements per instruction. The four codes have been purposefully selected to represent non-dense workloads: SpMV, BFS, PageRank, FFT. The experimental setup allows us to measure their performance while varying the vector length, the memory latency, and bandwidth. Our results not only show that larger vector lengths allow for better tolerance of limitations in the memory subsystem but also offer hope to code developers beyond dense linear algebra.
2 November 2022 in Concurrency and Computation: Practice and Experience
Summary: Novel architectures leveraging long and variable vector lengths like the NEC SX‐Aurora or the vector extension of RISCV are appearing as promising solutions on the supercomputing market. These architectures often require re‐coding of scientific kernels. For example, traditional implementations of algorithms for computing the fast Fourier transform (FFT) cannot take full advantage of vector architectures. In this article, we present the implementation of FFT algorithms able to leverage these novel architectures. We evaluate these codes on NEC SX‐Aurora , comparing them with the optimized NEC libraries; and in a prototype of a RISC‐V core with a vector processing unit. We present the benefits and limitations of two approaches of RADIX‐2 FFT vector implementations. We show that our approach makes better use of the vector unit of the NEC SX‐Aurora , reaching higher or equal performance than the optimized NEC library. More generally, we prove the importance of maximizing the vector length usage of the algorithm, taking advantage of the FFT properties to reduce long‐latency vector operations, and reordering the instructions according to the specific hardware features to boost the performance of FFT‐like computational kernels.
The HPC market is witnessing the appearance of emerging architectures exploiting long and variable vector lengths, like the NEC SX-Aurora or the RISC-V vector extension. These new architectures often require rethinking scientific algorithms to leverage their potential, and sparse computations can be the most challenging codes to adapt. Specifically, we showcase the obstacles of vectorizing graph algorithms and propose a graph representation suited for long vector architectures. We evaluate our work implementing three graph algorithms, Breadth-First-Search (BFS), Connected Components (CC), and Page Rank (PR), for a prototype RISC-V core with a long Vector Unit, presenting optimizations that benefit long vector architectures. We also run our implementations on the NEC SX-Aurora and compare our results with other graph libraries for CPUs and vector acceleration. Our BFS and CC implementations outperform CPU approaches by a factor around 4× depending on the algorithm and graph, with PR having a speedup closer to 20×. We also match or outperform the single-core performance of other vector implementations in the NEC SX-Aurora by 2× on average. We exhibit the efficiency of the RISC-V prototype, obtaining more performance out of its available bandwidth than the other evaluated single-core architectures. More generally, we show that adapting data structures and algorithms unleashes the potential of long vector architectures for sparse computations, demonstrating that these emerging technologies strive against traditional accelerated computing.