Repository navigation
Two issues with mpi communicator configuration #22
Description
From #21:
Two related issues found while investigating, not addressed here:
- h_comm_size is writable on both config objects, but tpb::diag / gdb::diag
recompute it from mpi_size, so setting it has no effect.- gdb_diag's docstring says b_comm_size must be 1, which is enforced nowhere and
unexplained. If true it deserves a reason; if stale it blocks the one dimension that
would parallelise the determinant list.
Activity
Drop h_comm_size from the config bindings, and warn on a bad rank count
hfwen0502 commentedon Sep 14, 2026
Jim Garrison (@garrison) agree. h_comm_size will be removed from the wrapper layer.
The GDB case needs a lot of more work. There are a lot of changes in the sbd upstream and we would need to use the new APIs in order to lift the b_comm_size=1 constraint. Kevin J. Sung (@kevinsung): Will you have time to work on this? Or you would prefer me to look into it?
kevinsung commentedon Sep 14, 2026
I won't have time to look into this soon, so It would be great if you could.
Drop h_comm_size from the config bindings, and warn on a bad rank count
Separate SQD parameters from SBD parameters, and expose the ones that…
README: fix broken anchor, remove stale h_comm_size row (#30)
gdb_diag: take a sharded determinant list, and numpy instead of neste…
gdb_diag: take a sharded determinant list, and numpy instead of neste…
gdb_diag: take a sharded determinant list, and numpy instead of neste…
gdb_diag: take a sharded determinant list, and numpy instead of neste…
gdb_diag: take a sharded determinant list, and numpy instead of neste…
gdb_diag: take a sharded determinant list, and numpy instead of neste…
gdb_diag: take a sharded determinant list, and numpy instead of neste…
gdb_diag: take a sharded determinant list, and numpy instead of neste…
gdb_diag: take a sharded determinant list, and numpy instead of neste…
GDB: sharded determinant basis, numpy marshalling, drivers and first …