bosilca

Correctly releasing the memory is an issue because calling free on your object does not clean all the inherited classes. If you think of a simple, one-level, inheritance of a...

From a user perspective this is mind blowingly complicated and cumbersome. And I'm trying to be polite.

It seems to me that all these issues arise from the fact that we persist at separating the MCA for the different layers. This also forces us to prepend all...

Make sense, that code needs to know the correct number of datatypes in order to work properly. As we have control over `ompi_datatype_get_args` we can make it return the correct...

I'm not sure the patch is enough. UCC uses UCX to move data, and UCX is also used by OMPI as the default PML. The collective framework being initialized after...

I went though the selection code and if my reading (and recollection) of that code is correct we have a problem. The loop reading the different rules expects 4 longs...

@burlen all looks good. Can you please add or update the copyright in the files you touched. Something like this: ``` * Copyright (c) 2024 NVIDIA CORPORATION. All rights reserved....

the cudaIPC does not allocate memory but instead maps the memory of another GPU onto the local process memory space (for GPU access). Thus, while the memory use seems to...

There is no magic in the 3, it is skipping the 3 predefined communicators in the world model, communicators statically registered under comm_id 0, 1 and 2.

I don't have a mingw machine to test it anymore. Also it appears you are compiling the single node version, definitively not the standard way to use parsec. Please try...