bosilca

1. Using a slightly older version of the Intel compiler **2021.10.0** (gcc version 11.4.1 compatibility), everything **works** (including the OpenMP support) without the patch. ```fundamental configure: C compiler: icc -O3...

I confirm the new patch works as expected.

Your code works just fine on my setup. My main issue arise from the fact that prte DVM decide to only use the 'lo' interface which forces me to start...

As a first step you should attach to the running processes with a debugger (`gdb -p`) and take a look at the stack trace to see where exactly the processes...

The decision to call another algorithm under certain conditions is out of scope in a base function, because technically it shall be done by a higher level decision function (such...

I was planning to review it after they figure out the issue with replacing the blocking ompi_coll_base_sendrecv with non-blocking communications.

No, I was wondering how would a multi-rail enabled collective component (which will indeed post multiple isend requests) interacts with a multi-rail enabled low-level communication library (which will split all...

Let me add some generic comment here, mostly as a reminder to self. The reason is that I don't think this is how we should use these op modules, especially...

Thanks for taking a look. I have to say your results are concerning, because the extra overhead is terribly high, so high that I wonder how one small memcpy (for...

From a code perspective this looks ok. But I still question the need for this, from a logical perspective. If I look at the results posted earlier here, the k=2...