Skip to content

WIP - Bug 4495: detect missing HMEM support#13

Draft
PHHargrove wants to merge 1 commit into
BerkeleyLab:developfrom
PHHargrove:bug4495-hmem-probe
Draft

WIP - Bug 4495: detect missing HMEM support#13
PHHargrove wants to merge 1 commit into
BerkeleyLab:developfrom
PHHargrove:bug4495-hmem-probe

Conversation

@PHHargrove
Copy link
Copy Markdown
Collaborator

This is an updated (rebased and ZE kinds support added) replacement for BitBucket PR#565

Quoting from the prose in that old PR:

Status
Ready for review, but WIP due to insufficient testing

So far, tested only on Dirac's CX5/Maxwell nodes with two builds of libfabric 1.16.1, one each with and without the necessary FI_HMEM_CUDA support.


ofi: fix bug 4495 by probing HMEM support

For libfabric 1.16.0 and newer, this commit resolves bug 4495 "ofi: detect when libfabric provider was not configure to support a specific memory kind" by attempting a small registration with the appropriate attr.iface value.

For libfabric 1.16.0 and newer, this commit resolves bug 4495 "ofi:
detect when libfabric provider was not configure to support a specific
memory kind" by attempting a small registration with the appropriate
`attr.iface` value.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant