Ken Raffenetti

ch4/ucx: Support GPU buffers as window memory

3

## Pull Request Description Recent versions of UCX handle this case correctly. ## Author Checklist * [x] **Provide Description** Particularly focus on _why_, not _what_. Reference background, issues, test failures,...

Add `mpi_memory_alloc_kind` and `mpi_assert_memory_alloc_kind`. https://github.com/mpi-forum/mpi-issues/issues/580 https://github.com/mpi-forum/mpi-standard/pull/714

The wire protocol functions treat all values as strings and determine their length using `strlen(value)`. ```c void PMIU_msg_set_query_kvsput(struct PMIU_cmd *pmi_query, int version, bool is_static, const char *key, const char *value);...

Use of `PMIx_Spawn` was recently introduced into MPICH. We should support it in our internal PMIx client library and Hydra.

~release_gather-based reduce collectives assume uniform data layout among participating processes. Data in temporary shared memory buffers is copied in/out using the layout of the local datatype. The problem is that...

See https://github.com/open-mpi/ompi/issues/11677. Users are running into issues building MPICH/ROMIO/mpl in the wild due to an incompatible Level Zero library installed on their system. 1. Should Level Zero support default to...

[1.15.x] segfault in IOV GPU recv

2

### Describe the bug Segfault when receiving an IOV datatype using GPU device memory. ### Steps to Reproduce - Configure MPICH main branch (494c4cf50c4a5d95c76d9fcd454b462adeb4d15e or later) with UCX - ./configure...

feature: Support building CUDA backend kernels with `clang++`

2

A user reported being unable to build MPICH using `clang++` as the CUDA compiler used to build Yaksa pmodels/mpich#6954. Noting down the issues encountered trying to build like this: 1....

## Pull Request Description These flags were ignored when the user specified a compiler other than the nvcc included in the CUDA installation. Make sure to include them for consistency....