14.7. ScaLAPACK PDGETRF

LAPACK is designed to be highly efficient on vector processors, high-performance "super-scalar" workstations, and shared-memory multiprocessors. LAPACK can also be used sat-isfactorily on all types of scalar machines (PCs, workstations, and mainframes). However, LAPACK in its present form is less likely to give good performance on other types of parallel architectures—for example, massively parallel Single Instruction Multiple Data (SIMD) machines, or Multiple Instruction Multiple Data (MIMD) distributed-memory machines. The ScaLAPACK effort was intended to adapt LAPACK to these new architectures.

By creating the ScaLAPACK software library, we extended the LAPACK library to scalable MIMD, distributed-memory, concurrent computers. For such machines, the memory hierarchy includes the off-processor memory of other processors, in addition to the hierarchy of registers, cache, and local memory on each processor.