qimpy.mpi.Iallreduce_in_place
- class Iallreduce_in_place(comm, buf, op)
Bases:
object
Perform async Iallreduce in-place on buf. Provides async semantics and completes on wait() of the return value, even if the MPI implementation does not support Iallreduce. This is true of some MPI implementations on GPU buffers eg. OpenMPI.
- Parameters:
comm (Comm) –
buf (Tensor) –
op (Op) –
- __init__(comm, buf, op)
- Parameters:
comm (Comm) –
buf (Tensor) –
op (Op) –
- Return type:
None
Methods
wait