Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

full cosmo on kesch #32

Closed
jgphpc opened this issue Jun 19, 2017 · 2 comments
Closed

full cosmo on kesch #32

jgphpc opened this issue Jun 19, 2017 · 2 comments

Comments

@jgphpc
Copy link
Contributor

jgphpc commented Jun 19, 2017

scorep/2.0 (Carlos)

  • /project/csstaff/piccinal/cosmo/cosmo-pompa.scorep_nonblockcomm
    eff

scorep/3.0 (Maxime)

  • /project/csstaff/piccinal/cosmo/cosmo-e_1m_2h/
    eff2
@jgphpc jgphpc mentioned this issue Jun 19, 2017
@jgphpc
Copy link
Contributor Author

jgphpc commented Jun 19, 2017

This is no longer an issue.

i can't get reports with mvapich2/2.2gdr, see: #4 (comment)

SCOREP_IpcGroup_GetSize: Assertion `SCOREP_Status_IsMppInitialized()' failed.

From: Ronny Tschueter <ronny.tschueter@tu-dresden.de>
Date: Thu, 13 Apr 2017 10:40:45 +0200

Dear jg,

The Score-P MPI adapter was designed for MPI execution on the host in mind.
We've never tested NVIDIA GPU Direct. IMHO, you are right and it's illegal
to call 'memcpy' for GPU Direct RDMA memory areas that bypass CPU host
memory.

Best regards, Ronny

@jgphpc
Copy link
Contributor Author

jgphpc commented Jun 19, 2017

For reference, trace from the comm_overlap benchmark:
eff

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

No branches or pull requests

1 participant