arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2305.02444cs.DCcs.PF

FT-GEMM:x86 CPU 上的一种容错高性能 GEMM 实现

FT-GEMM: A Fault Tolerant High Performance GEMM Implementation on x86 CPUs

Shixun Wu, Yujia Zhai, Jiajun Huang, Zizhe Jian, Zizhong Chen

AI总结:

FT-GEMM 是一种在 x86 CPU 上实现的高性能容错 GEMM,通过算法级融合内存操作实现软错误容忍,在 Intel Cascade Lake 上比 MKL、OpenBLAS 和 BLIS 快 3.50% 至 22.14%。

摘要:

通用矩阵乘法(GEMM)对于科学计算和机器学习至关重要。然而,计算平台规模的增大引发了对硬件和软件可靠性的担忧。在本海报中,我们提出了 FT-GEMM,一种能够实时容忍软错误的高性能 GEMM。我们通过将内存密集型操作融合到 GEMM 汇编内核中,在算法层面集成了容错功能。我们为并行 FT-GEMM 设计了一种缓存友好的方案。在 Intel Cascade Lake 上的实验结果表明,FT-GEMM 提供了高可靠性和高性能——在串行和并行 GEMM 中,即使每分钟注入数百个错误,其速度也比 Intel MKL、OpenBLAS 和 BLIS 快 3.50% 至 22.14%。

英文摘要:

General matrix/matrix multiplication (GEMM) is crucial for scientific computing and machine learning. However, the increased scale of the computing platforms raises concerns about hardware and software reliability. In this poster, we present FT-GEMM, a high-performance GEMM being capable of tolerating soft errors on-the-fly. We incorporate the fault tolerant functionality at algorithmic level by fusing the memory-intensive operations into the GEMM assembly kernels. We design a cache-friendly scheme for parallel FT-GEMM. Experimental results on Intel Cascade Lake demonstrate that FT-GEMM offers high reliability and performance -- faster than Intel MKL, OpenBLAS, and BLIS by 3.50\%$\sim$ 22.14\% for both serial and parallel GEMM, even under hundreds of errors injected per minute.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑