Ticket #1559 (closed changeset move request: fixed)

Opened 8 years ago

Last modified 3 years ago

Move trunk r29542, and r29566 to 1.7 (Fix reduce_scatter algorithms to handle recvcounts of 0)

Reported by: rolfv Owned by: ompi-rm1.7
Priority: major Milestone: Open MPI 1.7.4
Version: trunk Keywords:
Cc:

Description (last modified by hjelmn) (diff)

See ticket #1550 for more details on this.

We have implemented a fix for recvcounts of 0 with r19702. However, rather then detecting this issue and just falling back to simpler algorithms, we should fix the various algorithms to handle this case. This is a placeholder so we remember to do that.

Change History

comment:1 Changed 8 years ago by rolfv

  • Description modified (diff)

comment:2 Changed 8 years ago by jsquyres

George asked me to move this to 1.3.2 at the earliest. Might slip to 1.4.

comment:3 Changed 8 years ago by bbenton

  • Milestone changed from Open MPI 1.3.1 to Open MPI 1.3.2

Finishing the move to 1.3.2

comment:4 Changed 8 years ago by bbenton

  • Milestone changed from Open MPI 1.3.2 to Open MPI 1.3.3

comment:5 Changed 7 years ago by bbenton

  • Milestone changed from Open MPI 1.3.3 to Open MPI 1.3.4

comment:6 Changed 7 years ago by bosilca

  • Milestone changed from Open MPI 1.3.4 to Open MPI 1.4

comment:7 Changed 7 years ago by bbenton

  • Milestone changed from Open MPI 1.4 to Open MPI 1.4.2

comment:8 Changed 7 years ago by bbenton

  • Milestone changed from Open MPI 1.4.2 to Open MPI 1.4.3

comment:9 Changed 5 years ago by bbenton

  • Milestone changed from Open MPI 1.4.5 to Open MPI 1.4.6

Milestone Open MPI 1.4.5 deleted

comment:10 Changed 4 years ago by bbenton

  • Milestone changed from Open MPI 1.4.6 to Open MPI 1.7.1

Kicking this can down the 1.7 road.

comment:11 Changed 3 years ago by hjelmn

(In [29542]) Fix the recursive halfing algorithms for reduce scatter in both basic and tuned to correctly handle 0 recvcounts.

Tested with the reproducer from #1550.

Refs #1559

comment:12 Changed 3 years ago by hjelmn

  • Owner changed from bosilca to rolfv
  • Status changed from new to assigned
  • Type changed from defect to changeset move request

There were a couple of errors in both the tuned and basic recursive halfing reduce scatter algorithms. Fixed in r29542.

Rolf, please review for 1.7.4.

comment:13 Changed 3 years ago by rolfv

  • Owner changed from rolfv to hjelmn

Impressive! Fixing a 5 year old bug. Things look OK. You need to remove the comment at the top of the coll_basic_reduce_scatter.c file like you did for the one of the other files. I also ran the IMB test like in the original bug report and it passed for me too. So, remove the comment and this is reviewed OK.

comment:14 Changed 3 years ago by hjelmn

(In [29566]) Remove outdated comments in coll_basic_reduce_scatter.c.

Refs #1559

comment:15 Changed 3 years ago by hjelmn

  • Owner changed from hjelmn to ompi-rm1.7
  • Description modified (diff)
  • Summary changed from Fix reduce_scatter algorithms to handle recvcounts of 0 to Move trunk r29542, and r29566 to 1.7 (Fix reduce_scatter algorithms to handle recvcounts of 0)

Yeah. Saw this one and figured it might be time to finally fix it.

comment:16 Changed 3 years ago by rhc

  • Status changed from assigned to closed
  • Resolution set to fixed

(In [29575]) Fixes #1559: Move trunk r29542, and r29566 to 1.7 (Fix reduce_scatter algorithms to handle recvcounts of 0)

---svn-pre-commit-ignore-below---

r29542
Fix the recursive halfing algorithms for reduce scatter in both basic and tuned to correctly handle 0 recvcounts.

Tested with the reproducer from #1550.

Refs #1559

r29566
Remove outdated comments in coll_basic_reduce_scatter.c.

Refs #1559

Note: See TracTickets for help on using tickets.