GitLab

Combined the old and new contraction mapping mechanisms and select mapping with lowest time among the two sets.

made other corrections to get new mapping scheme working properly

Finished a seemingly working implementation of a more exhaustive mapping scheme, but performance seems like there are still some issues.

Added model output/input functionality and setup initial models based on running CCSD on a 12-core linux machine

Could use a bit more correctness verification.

Fixed correctness bug involving different processors folding differently due to differing performance models.

Fix is temporary, simply broadcasts processor 0s solution. Planned solution is to do flat-tree TSQR.

Still seeing a bug for sparse_mp3 on 24 processes.

Made all bcast and allreduce calls of nontrivial size use the wrapped MPI routine through CommData, where times get measured and contribute to time estimation.

undoing commit correcting flop count correction, in order to make merge with sparsity easier, since that code was moved elsewhere there, and the correction should have been applied in the right location

changes to performance models for sparse contractions. Also made redistribution time account for home buffer back-redistribution