The multicomputer toolbox approach to concurrent BLAS and LACs
We describe many of the issues involved in general-purpose Concurrent Basic Linear Algebra Subprograms (Concurrent BLAS or CBLAS), discuss data-distribution independence, while further generalizing data distributions. We comment on the utility of Linear Algebra Communication Subprograms (LACS). We describe an algorithm for dense matrix-matrix multiplication and also discuss matrix-vector multiplication issues. With regard to communication, we conclude the there is limited leverage in LACS per se as a stand-alone message-passing standard, and propose that needed capabilities instead be integrated in general, application-level message passing standard, focusing attention on CBLAS and large-scale application needs. Most of the proposed LACS features are similar to existing or needed general-purpose primitives anyway. All of the ideas discussed here have been implemented or, as noted, are under current development within the Multicomputer Toolbox open software system. 13 refs.
- Research Organization:
- Lawrence Livermore National Lab., CA (United States)
- Sponsoring Organization:
- DOE; USDOE, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 5671193
- Report Number(s):
- UCRL-JC-109775; CONF-9204106--1; ON: DE92010401
- Country of Publication:
- United States
- Language:
- English
Similar Records
The multicomputer toolbox: Scalable parallel libraries for large-scale concurrent applications
The multicomputer toolbox: Scalable parallel libraries for large-scale concurrent applications