Hui Zhou
Hui Zhou
We need somehow get these tests into at least nightly Jenkins tests.
For the last passed mpich version 3.4.1, do we have the configure line? I am wondering if that was --with-device=ch3 or ch4.
The last message in https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1006788 says it was happy with mpich 4.0, right?
@edoapra Can I get confirmation that only MPICH+UCX have (and always have) issue, and MPICH+OFI has always been okay, regardless of MPICH versions? The ucx netmod have shared memory disabled...
The requests passed to `MPIX_Waitall_enqueue` must be from `MPIX_Isend_enqueue` or `MPIX_Irecv_enqueue`, and they have to be all enqueued to the same gpu stream.
Yes, `MPI_REQUEST_NULL` is accepted in `MPI_Waitall`. The use case is for applications that may use `MPI_Test` before and it is inconvenient to track which requests have already been completed and...
@AtlantaPepsi Do you have further questions regarding this issue?
Most functions in mpich return an error code even for functions that its semantics only return MPI_SUCCESS -- for uniformity reason or for future proof. I suspect MPIR_Setup_intercomm_localcomm is such...
Yes, but inside the if clause in the example.
@crubiog I just checked the code for `MPIR_Setup_intercomm_localcomm` and it does not always return `MPI_SUCCESS` and your static analysis does indicate a potential bug. I do suspect there could be...