bosilca comments

Results 318 comments of


                                            bosilca

Efficiency improvements for error code/class management

I'm torn about this. The error code/class management is not something where most users will get an intense add/remove, which means the proposed optimization impact will be minimal, even non-existent...

MPI_ERR_IN_STATUS for persistent requests

looking into the failure.

segfault due to thread-unsafe getenv() call

This particular issue aside, a quick scan of the OMPI code indicates that we are using setenv and opal_setenv in a lot of places that could technically be triggered simultaneously...

MPI_Waitall doesn't return MPI_ERR_IN_STATUS for persistent request

Thanks for the report. I reviewed the code and your analysis is correct, there are few cases where we fail to properly report the error code upstream. For your last...

ODR violation with libopen-palcommon_sm

I'm puzzled by the comment at the top of the opal/mca/common/Makefile.am is accurate, because the shared library case has too many pitfalls. I assumed the common is build into the...

v4.1.5 UCX_NET_DEVICES not selecting TCP devices correctly

All this looks good to me. Allowing UCX to pick the communication device gives you IB (and a latency of 4us for the all-to-all) while enforcing a specific device (mostly...

TCP transport hangs during cleanup

The OMPI proposed fix does not really fix the issue, it makes it go away by forcing a PMIX fence before destroying endpoints. As such, I don't think it is...

Is having more than one GPU initialized UB in standard cuda-aware MPI?

You are correct, most MPI libraries are optimized for one GPU per process. OMPI might work with multiple GPU, depending on what exactly you are doing, but it will certainly...