Pass NULL to the collective components for unused buffers. #11552
Conversation
Neither Allreduce or Alltoall provide the opportunity for the user to provide a buffer that shall not be used. That would be possible for the other flavors of |
bosilca
force-pushed
the
topic/null_as_unused_buffers_for_collectives
branch
4 times, most recently
from
7ae2b93 to
88ff1d2
Compare
Right, I was thinking in terms of |
We want the underlying algorithms to know when they can use a buffer as a temporary, such as when one module delegate a collective to another. Instead of allocating temporaries everywhere, we allow the upper level module to handle the temporary buffers, and pass them into all the others. Thus, we need to make sure that the user provided buffers are screened such that they don't make it as temporaries. Signed-off-by: George Bosilca <[email protected]>
bosilca
force-pushed
the
topic/null_as_unused_buffers_for_collectives
branch
from
88ff1d2 to
1edc1b5
Compare
With open-mpi#11552, rbuf is *either* valid & appropriately sized, or NULL, and collectives implementations downstream from HAN (e.g. XHC) rely on this convention. Signed-off-by: George Katevenis <[email protected]>
Improve handling of the rbuf pointer. Previously, it was possible for the pointer to become non-NULL, even though it does not point to valid memory. This becomes a problem for the assurances that open-mpi#11552 tries to introduce. Signed-off-by: George Katevenis <[email protected]>
|
Hi, were there any concerns or issues left to resolve with this? Is it good to be merged? |
We want the underlying algorithms to know when they can use a buffer as a temporary, such as when one module delegate a collective to another. Instead of allocating temporaries everywhere, we allow the upper level module to handle the temporary buffers, and pass them into all the others. Thus, we need to make sure that the user provided buffers are screened such that they don't make it as temporaries.
This could help #11418 to streamline the call to the xhc reduction.