针对广义稀疏矩阵乘法(SpGEMM)内存瓶颈问题,提出基于近行缓冲存储器(PIM)的HighP加速器,通过非冲突列并行处理机制提升内存带宽利用率,实验表明性能较现有加速器提升17.88倍,较PIM方案提升8.19倍。 摘要: 广义稀疏矩阵乘法(SpGEMM)是一种关键的计算原 ...
A new technical paper titled “Optimization of SpGEMM with Risc-V vector instructions” was published (preprint) by researchers at the Barcelona Supercomputing Center.