| 21 | |---------------- |
| 22 | {{{#!td style="vertical-align:top;width: 50px" |
| 23 | [[CRev(0,4683)]] |
| 24 | }}} |
| 25 | {{{#!td style="vertical-align:top;width: 50px" |
| 26 | [[CRev(1,4683)]] |
| 27 | }}} |
| 28 | {{{#!td style="vertical-align:top;width: 75px" |
| 29 | [[CRev(2,4683)]] |
| 30 | }}} |
| 31 | {{{#!td style="vertical-align:top" |
| 32 | 6.0 |
| 33 | }}} |
| 34 | {{{#!td style="vertical-align:top" |
| 35 | N |
| 36 | }}} |
| 37 | {{{#!td style="vertical-align:top" |
| 38 | SVN commit message: [[CRev(3,4683)]] (radiation_model_mod.f90) |
| 39 | |
| 40 | Lowering the size of each MPI_Alltoall call could alleviate problems with certain MPI implementations/installations that may have problem with large exchanges. Setting {{{bufsize_alltoall=1048576}}} could be a reasonable starting point for debugging in such cases. |
| 41 | |
| 42 | The complete exchange (calculation of displacements, negotiation of number of exchanges and the individual exchanges of integer and real arrays has been moved to a new procedure {{{radiation_exchange_alltoall}}} which may be applied to similar alltoall calls in the future. |
| 43 | }}} |