Mercury Computer Systems, Inc. (NASDAQ: MRCY) released a binary distribution of its Parallel Acceleration System (PAS) middleware to High Performance Embedded Computing Software Initiative (HPEC-SI) members for use on a Georgia Tech Research Institute (GTRI) 64-node Linux cluster. This powerful development lab system is based on commercial-off-the-shelf (COTS) x86 computers. Members of the HPEC-SI working group are now evaluating PAS as middleware which allows straightforward migration of Parallel Vector Signal Image Processing Library++ (Parallel VSIPL++) implementations between benign environment workstations, parallel Linux clusters, and ruggedized, deployable multicomputer platforms.
“CodeSourcery leverages the GTRI cluster in the development of Sourcery VSIPL++(TM), our optimized implementation of the Parallel VSIPL++ API,” said Jules Bergmann, Software Architect of CodeSourcery. “The availability of PAS middleware in a standard Linux x86 cluster will greatly ease development by allowing embedded applications to be coded on lab systems and then migrated to rugged, deployed multicomputers.”
PAS is Mercury’s optimized middleware for data movement among compute nodes in scalable, high-performance embedded computing applications. The HPEC-SI distribution makes PAS available for non-Mercury hardware platforms for the first time. The release implements PAS over standard TCP/IP to facilitate the easy migration of the PAS application programming interface (API) to a wide variety of target environments. PAS is supported on all Mercury multicomputer system platforms and has been implemented to run on a variety of fabrics including RapidIO, InfiniBand, and Ethernet.
Parallel VSIPL++ builds upon VSIPL(TM) which is an open standard mathematical subroutine application programming interface (API). Parallel VSIPL++ extends VSIPL by adding an initial component of data movement functionality targeted at distributed, embedded signal and image processing applications. Application partitioning, which traditionally has been hardware-dependent, is expressed in Parallel VSIPL++ using a data object constructor. This allows algorithm developers to concentrate on mathematics without having to focus undue attention on the memory, processor, and switch fabric details of the target platform.
“Parallel VSIPL++ is a major step toward meeting the grand challenge of software development ‘write-once, run-anywhere.’ It is a breakthrough for C++ embedded developers, allowing them to code portable mathematical expressions,” said Eran Strod, Director of Product Marketing for the defense business at Mercury.
Mercury plans to make a click-through license of the PAS Linux-x86 TCP/IP reference distribution available on its website.