Engineering Papers⌕ Search

Engineering topics

Mullin, Lenore

Publications and source records attributed to Mullin, Lenore.

Threaded Multi-Core GEMM with MoA and Cache-Blocking: Preprint

A threaded multi-core implementation of the high performance dense linear algebra matrix-matrix multiply GEMM kernel is described. This kernel is widely implemented by vendors in the basic linear algebra subroutine BLAS library. The mathematics of arrays (MoA) paradigm due to Mullin (1988) results in contiguous memory accesses by employing outer-product forms. Our performance studies demonstrate that the MoA implementation of double precision DGEMM combined with optimal cache-blocking strategies results in at least a 25% performance gain on the Intel Xeon Skylake processor over the vendor supplied Intel MKL basic linear algebra libraries. Results are presented for the NREL Eagle supercomputer. The multi-core DGEMM achieves over 100 GigaFlops/sec with eight openMP threads.

cache-blocking↗

Improving the Performance of DGEMM with MoA and Cache-Blocking: Preprint

The goal of this paper is to demonstrate performance enhancements of the high performance dense linear algebra matrix-matrix multiply DGEMM kernel, widely implemented by vendors in the basic linear algebra subroutine BLAS library. The mathematics of arrays (MoA) paradigm due to Mullin (1988) results in contiguous memory accesses in combination with Church-Rosser complete language constructs optimized for target processor architectures [3]. Our performance studies demonstrate that the MoA implementation of DGEMM combined with optimal cache-blocking strategies results in at least a 25% performance gain on both Intel Xeon Skylake and IBM Power-9 processors over the vendor supplied Intel MKL and IBM ESSL basic linear algebra libraries. Results are presented for the NREL Eagle and ORNL Summit supercomputers.

cache-blocking↗

Exascale and Beyond: Configuring, Reasoning, Scaling. Report of the 2011 Workshop on Architectures II: Exascale and Beyond, held August 8-10, Albuquerque, N.M.

The push toward exascale computing (1) and beyond is torn between contradictory and possibly incompatible goals. On the one hand, one wants a system that will be available within a short span of time, on a relatively small development budget, and with the least changes possible to the current software stack. On the other hand, one wants a system that is orders of magnitude more efficient in its energy consumption. An exascale system could be as different from current systems as scalable parallel systems were from vector machines at the time of the last major transition in the field, but the amount of application software developed at DOE in the last decades makes it undesirable to take a “clean sheet of paper” approach, if at all possible. Furthermore, one wishes to explore and develop distinct exascale architectures to reduce risk while maintaining some unity so that application codes easily port among them. Finally, the design of an exascale system may involve a large number of vendors, some of which have limited experience in the field of High-Performance Computing (HPC). A proper integration of these multiple contributions requires a proper definition of many new interfaces, as well as consistent design principles for the handling of cross-cutting issues such as resilience, performance, and power management.

97 MATHEMATICS AND COMPUTING↗