Reusable distribution of a rank-ordered host array over NEKO_COMM. Counts are vector elements or matrix columns, not total matrix elements. Matrices distribute their second dimension; all ranks have the same number of rows and write every row of their local columns. Initialise collectively once, then reuse while every rank's local size and NEKO_COMM remain unchanged. Reinitialise after redistribution.
More...
|
| procedure | init (this, local_count) |
| | Calculate the distribution collectively.
|
|
| integer | offset = 0 |
| | First entry owned by this rank, using zero-based indexing.
|
| integer | total_count = 0 |
| | Sum and maximum of local counts over NEKO_COMM.
|
| integer | max_count = 0 |
|
| integer | local_count = -1 |
| | Number of entries on this rank; -1 marks an uninitialised layout.
|
Definition at line 73 of file hdf5_file.F90.
◆ init()
| procedure hdf5_file::hdf5_layout_t::init |
( |
class(hdf5_layout_t), intent(inout) | this, |
|
|
integer, intent(in) | local_count ) |
- Parameters
-
| local_count | Scalar integer: local number of vector elements or matrix columns. For a matrix use matget_ncols(), not matsize(). Pass only this rank's count, not an array of counts or dimensions. |
- Note
- All ranks in NEKO_COMM must call this routine. A supplied layout removes size collectives from subsequent writes; callers must initialise it again on all ranks if any local count or the communicator changes.
Definition at line 84 of file hdf5_file.F90.
◆ local_count
| integer hdf5_file::hdf5_layout_t::local_count = -1 |
|
private |
◆ max_count
| integer hdf5_file::hdf5_layout_t::max_count = 0 |
◆ offset
| integer hdf5_file::hdf5_layout_t::offset = 0 |
◆ total_count
| integer hdf5_file::hdf5_layout_t::total_count = 0 |
The documentation for this type was generated from the following file: