Libraries Optimized for High-End Symmetric Multiprocessing Systems
Silicon Graphics, Inc. announced the availability of a comprehensive collection of high-performance, industry-standard mathematical libraries. The CHALLENGEcomplib(tm) Scientific Math Library provides C and Fortran programmers with access to a set of easy-to-use mathematical and numerical algorithms for use in scientific and technical computing.
The library is available immediately and ships with a Fortran compiler, which is an optional product for Silicon Graphics' high-end systems. The CHALLENGEcomplib allows scientific and technical computing users to develop, optimize and convert application programs to Silicon Graphics' symmetric multiprocessing (SMP) CHALLENGE(tm) servers, POWER CHALLENGE(tm) supercomputing servers and Onyx(tm) graphics supercomputers. It consists of optimized, blocked and multi-threaded mathematical routines which extract optimal performance from cache-based hierarchical memory systems and allows users to effectively utilize the parallel processing capabilities of the SMP architecture.
"The optimized parallel routines improve the performance and efficiency of applications, provide increased application throughput and reduce the time to solution," said Paul Koontz, director of marketing of Silicon Graphics' Supercomputing Systems Division. "This comprehensive selection of mathematical and scientific subprograms conforms to de-facto industry standard functional interfaces. This facilitates the porting of applications from traditional supercomputers and other UNIX(r) platforms to the leading price/performance products of the Silicon Graphics' family of SMP RISC systems."The CHALLENGEcomplib Library contains an extensive collection of industry standard libraries such as Basic Linear Algebra Subprograms (BLAS), Extended BLAS (Level 2 and Level 3) and LAPACK. Internally developed libraries which contain Fast Fourier Transforms and Convolution routines are also included, as well as select routines from robust libraries such as SCILIB(tm) from Cray Research and SLATEC(tm) from the Energy, Science and Technology Software Center."The Fast Fourier transforms (FFT) routines in the Silicon Graphics math library are twice as fast as our portable Fortran FFT routines, which enabled us to significantly reduce the runtime of our 3D prestack depth migration program," said Dwight Sukup, senior research associate at Texaco, Inc. "Because the library routines themselves are tuned to make optimal use of the cache, we were able to achieve linear speed-up on our 3D migration benchmark."
