The Real Benefits and Costs to Industry of the Ownership of HPF Codes

Tim Cooper
Parallel Applications Centre, 2 Venture Road, Chilworth, Southampton, SO16 7NP, UK.
tpc@pac.soton.ac.uk

Mike Delves
N.A. Software Ltd, Roscoe House, 62 Roscoe Street, Liverpool, L1 9DW, UK.
delves@nasoftware.co.uk

Abstract

Performance characteristics, in terms of numerical accuracy and speed up, are necessary conditions to be met if HPF is to be used by industry, but they do not satisfy all the criteria of industrial users, particularly with regards to the cost of ownership. In order to address this problem the EC has funded, under the HPCN TTN Network, the HEIDI project to assess the full cost of porting a legacy, in-house F77 code to HPF, and the subsequent cost of ongoing development of the code.

This paper presents the technical detail of that project, an assessment of the cost of the original port from F77 to HPF and a preliminary analysis of the ongoing development cost. The original numerical solver was implicitly serial and has been replaced by the preconditioned ORTHOMIN solver, which is more amenable to parallelisation. The ORTHOMIN solver requires different convergence criteria from the original solver, but the results obtained on the Jones combustor test case are identical to the original solver, both in serial and in parallel. The code was ported by N.A. Software, owners and developers of the HPF Plus compiler and PAC, a centre of excellence in parallel computing and advanced IT. An analysis of the cost of porting the code is presented. The development of the code by the owners (Alfa Romeo Avio) is underway and so only preliminary results on this stage of the project are presented.

Notes by Chuck Koelbel