Numerical stability analysis of the class of communication hiding pipelined Conjugate Gradient methods
share
Krylov subspace methods are widely known as efficient algebraic methods for solving linear systems. However, on massively parallel hardware their performance is typically limited by communication latency rather than floating point performance. With HPC hardware advancing towards the exascale regime the gap between computation (i.e. flops) and communication (i.e. internode communication, as well as data movement within the memory hierarchy) keeps steadily increasing, imposing the need for scalable alternatives to traditional Krylov subspace methods. One such approach are pipelined Krylov subspace methods, which reduce the number of global synchronization points and overlap global communication latency with local arithmetic operations, thus `hiding' the global reduction phases behind useful computations. To obtain this overlap the algorithm is reformulated by introducing a number of auxiliary vector quantities, which are computed using additional recurrence relations. Although pipelined Krylov subspace methods are equivalent to traditional Krylov subspace methods in exact arithmetic, the behavior of local rounding errors induced by the multi-term recurrence relations in finite precision may in practice affect convergence significantly. This numerical stability study aims to characterize the effect of local rounding errors in various pipelined versions of the popular Conjugate Gradient method. We derive expressions for the gaps between the true and (recursively) computed variables that are used to update the search directions in the different CG variants. Furthermore, we show how these results can be used to analyze and correct the effect of local rounding error propagation on the maximal attainable accuracy of pipelined CG methods. The analysis in this work is supplemented by various numerical experiments that demonstrate the numerical stability of the pipelined CG methods.
READ FULL TEXT
