Project

General

Profile

Feature #1769

AMPI implementation of Buffered sends

Added by Sam White 6 months ago. Updated about 2 months ago.

Status:
New
Priority:
Normal
Assignee:
Category:
AMPI
Target version:
Start date:
12/20/2017
Due date:
% Done:

0%


Description

Now that AMPI supports in-place sends via RDMA, we should actually support buffered sends (Bsend, Ibsend, etc) and the related Buffer_{attach, detach} routines

History

#1 Updated by Sam White 6 months ago

Should also be able to optimize for ready sends (Rsend, Irsend, etc).

#2 Updated by Sam White about 2 months ago

One quick mpich-test that we can cross of is bsendalign. We currently return NULL for MPI_Buffer_detach, but we should return the same pointer that was given at MPI_Buffer_attach. The overhead for implementing this is one pointer per AMPI rank (stored in ampiParent).

Also available in: Atom PDF