Ft-Blas

Abstract

Basic Linear Algebra Subprograms (BLAS) is a core library in scientific computing and machine learning. This paper presents FT-BLAS, a new implementation of BLAS routines that not only tolerates soft errors on the fly, but also provides comparable performance to modern state-of-the-art BLAS libraries on widely-used processors such as Intel Skylake and Cascade Lake. To accommodate the features of BLAS, which contains both memory-bound and computing-bound routines, we propose a hybrid strategy to incorporate fault tolerance into our brand-new BLAS implementation: duplicating computing instructions for memory-bound Level-1 and Level-2 BLAS routines and incorporating an Algorithm-Based Fault Tolerance mechanism for computing-bound Level-3 BLAS routines. Our high performance and low overhead are obtained from delicate assembly-level optimization and a kernel-fusion approach to the computing kernels. Experimental results demonstrate that FT-BLAS offers high reliability and high performance -faster than Intel MKL, OpenBLAS, and BLIS by up to 3.50%, 22.14% and 21.70%, respectively, for routines spanning all three levels of BLAS we benchmarked, even under hundreds of errors injected per minute.

Machine learning and high-performance computing in numerical simulation of quantum many-body systems

“…GEMM employs a series of architecture-aware optimization strategies, such as cacheand register-level data re-use, prefetching, and vectorization that improve the hardware utilization of a program from a marginal < 1% to a near-optimal efficacy (> 90%) [83,230,207]. To leverage the highly optimized GEMM subroutine, the order of data in memory for spin configurations S α r (t) strategy that fuses the memory footprint of the element-wise operation with the compute-bound GEMM operation to hide the memory latency, is a sound solution that benefits a series of GEMMbased scientific computing and machine-learning applications [263,264]. Therefore, we delve into the black box of GEMM kernels, enabling memory-bandwidth efficient computations for "Daxpy"…”