Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
ice_dyn_vp: add global_norm, global_dot_product functions
The VP solver uses a linear solver, FGMRES, as part of the non-linear iteration. The FGMRES algorithm involves computing the norm of a distributed vector field, thus performing global sums. These norms are computed by first summing the squared X and Y components of a vector field in subroutine 'calc_L2norm_squared', summing these over the local blocks, and then doing a global (MPI) sum using 'global_sum'. This approach does not lead to reproducible results when the MPI distribution, or the number of local blocks, is changed, for reasons explained in the "Reproducible sums" section of the Developer Guide (mostly, floating point addition is not associative). This was partly pointed out in [1] but I failed to realize it at the time. Introduce a new function, 'global_dot_product', to encapsulate the computation of the dot product of two grid vectors, each split into two arrays (for the X and Y components). Compute the reduction locally as is done in 'calc_L2norm_squared', but throw away the result and use the existing 'global_sum' function when 'bfbflag' is active, passing it the temporary array used to compute the element-by-element product. This approach avoids a performance regression from the added work done in 'global_sum', such that non-bfbflag runs are as fast as before. Note that since 'global_sum' loops on the whole array (and not just ice points as 'global_dot_product'), make sure to zero-initialize the 'prod' local array. Also add a 'global_norm' function implemented using 'global_dot_product'. Both functions will be used in subsequent commits to ensure bit-for-bit reproducibility.
- Loading branch information