Leo Fang

Support for `std::complex<__half>`?

3

Hello, suggested by @allisonvacanti I am opening an issue for information gathering. I am wondering if libcu++ can support half complex natively or not. We are evaluating the possibility of...

[BLD] Better handling of MS-MPI version check at build time

9

See https://github.com/mpi4py/mpi4py/issues/19#issuecomment-826221898.

`memory.{__getitem__, __setitem__}` should be disabled for GPU buffers

3

`__getitem__` with a slice might work, but it's easier to just disable them altogether.

[DOC] Document more env vars?

3

We should make a distinction between env vars used at build time and those used at runtime. Runtime env vars include: - `MPI4PY_FUTURES_MAX_WORKERS` ([now documented!](https://mpi4py.readthedocs.io/en/latest/mpi4py.futures.html#envvar-MPI4PY_FUTURES_MAX_WORKERS)) - `MPI4PY_FUTURES_SERVICE` - `MPI4PY_FUTURES_SERVER_HOST` -...

[DOC] Document that certain objects require manual cleanup

12

For example, users need to call `Request.Free()` explicitly, see the discussion around https://github.com/mpi4py/mpi4py/pull/29#discussion_r593384270.

Installing 3.4.x fails with OS X + cmake + Fortran

5

Hello, not sure if this is a duplicate of #5004: We (Conda-Forge) experienced several failures when downstream packages try to build MPI support against MPICH 3.4.x (we tested 3.4.0 and...

Test the cuTensorNet backend in CI?

1

Follow-up of #6677. I wonder if we can just piggyback on, say, the `upy.linux.cuda-rapids` CI and run `CUPY_ACCELERATORS=cutensornet pytest test_einsum.py`?

Support new kwargs `initial` and `where` in reduction functions

1

### Description It seems - `numpy.min()` - `numpy.amin()` - `numpy.max()` - `numpy.amax()` - ...? have added additional kwargs `initial` and `where`, but CuPy's counterparts do not. Would be nice to...

Discussion: Restore (or document how to restore) the old `Device` context manager behavior (pre-v10)?

3

Follow-up of #5705, #6867. For downstream libraries depending on CuPy, having a context manager that respects any `cudaSetDevice`/`cudaGetDevice` calls inside _or outside_ CuPy is useful. However, this is not possible...