Order of parallel communication #252
Description
While working on the turbo features Salvo and I noticed that there is currently a problem with the order of the communication (i.e. the order in which we loop through the SEND_RECEIVE markers.)
In the code we always have loops like this
for (iMarker = 0; iMarker < config->GetnMarker_All(); iMarker++) {
if ((config->GetMarker_All_KindBC(iMarker) == SEND_RECEIVE) &&
(config->GetMarker_All_SendRecv(iMarker) > 0)) {
...
This way the send/receive involving the periodic boundaries is always done before the send/receive involving the boundaries related to the parallel partitioning (this is because the periodic structure is created before the partitioning is done). Hence the wrong values are send to the periodic ghost cells.
An easy solution is to change the order of the loop, i.e.
for (iMarker = config->GetnMarker_All() - 1; iMarker >= 0; iMarker--) {
if ((config->GetMarker_All_KindBC(iMarker) == SEND_RECEIVE) &&
(config->GetMarker_All_SendRecv(iMarker) > 0)) {
...
Since I am not completely sure whether this is a valid solution (I dont know if this gives performance issues or will break something else) I opened this as an issue.
Related to this I would also suggest to maybe implement a general send/recv routine that every class can use, because all these SendReceive_* and Set_MPI_* routines do essentially the same at the moment (just with different data).