Parallel QR factorization for hybrid message passing/shared memory operation

The capability to blend the message passing and shared memory programming environments is a common feature of many high performance computer architectures. To probe the performance benefits of this feature on a 64-processor HP SPP-2000, a 32-processor IBM SP3, and a 128-processor SGI Origin 2000, we propose a new hybrid parallel fast Givens QR factorization algorithm that can interpolate between pure message passing and pure shared memory.