--cpu-list does not bind in master nor 4.0 branch

[awlauria]

Thank you for taking the time to submit an issue!

Background information

What version of Open MPI are you using? (e.g., v1.10.3, v2.1.0, git branch name and hash, etc.)

master/4.0

Describe how Open MPI was installed (e.g., from a source/distribution tarball, from a git clone, from an operating system distribution package, etc.)

Git clone.

Please describe the system on which you are running

• Operating system/version: RHEL 7.6
• Computer hardware: Power9
• Network type:

---

Details of the problem

This can be reproduced with the following command, it seems that the --cpu-list option isn't getting processed correctly/ignored when it comes to binding:

[awlauria@c685f10n17 master]$ mpirun --report-bindings --np 2 --cpu-list 3,4 --bind-to core hostname
c685f10n17
[c685f10n17:102892] MCW rank 0 is not bound (or bound to all available processors)
[c685f10n17:102892] MCW rank 1 is not bound (or bound to all available processors)
c685f10n17

Changing --bind-to core to hwthread or removing it completely yields the same results.

Another somewhat related issue is that --cpu-list cannot be used with --map-by. Is this intended behavior?

[awlauria@c685f10n17 master]$ mpirun -np 2 --map-by ppr:2:node --cpu-list 2,3 --bind-to core --report-bindings --tag-output hostname
--------------------------------------------------------------------------
Conflicting directives for mapping policy are causing the policy
to be redefined:

  New policy:   RANK_FILE
  Prior policy:  UNKNOWN

Please check that only one policy is defined.
--------------------------------------------------------------------------

Thanks.

[gpaulsen]

@markalle has a PR to address this, that he'll be posting.

[awlauria]

@markalle opened #6584 to address this.

[awlauria]

It looks like it's still an issue on v4.0.6:

[awlauria@f8n02 v4.0.x_5_21]$ ./exports/bin/mpirun --report-bindings --np 2 --cpu-list 3,4 --bind-to core hostname
f8n02
[f8n02:1693364] MCW rank 0 is not bound (or bound to all available processors)
[f8n02:1693364] MCW rank 1 is not bound (or bound to all available processors)
f8n02

[awlauria]

#6584 doesn't look like it was ever brought back to the v4 series by my eye - which is probably why I can still reproduce.

This seems to work with v5.0.x/master + prrte with a slightly different run line

**[awlauria@f8n02 v5.0.x]$ ./exports/bin/mpirun --report-bindings --np 2 --cpu-list 3,4 --bind-to core hostname
--------------------------------------------------------------------------
WARNING: A deprecated command line option was used.

  Deprecated option:   --report-bindings
  Corrected option:    --display bind

We have updated this for you and will proceed. However, this will be treated
as an error in a future release. Please update your command line.
--------------------------------------------------------------------------

--------------------------------------------------------------------------
WARNING: A deprecated command line option was used.

  Deprecated option:   --cpu-list
  Corrected option:    --map-by :PE-LIST=3,4

We have updated this for you and will proceed. However, this will be treated
as an error in a future release. Please update your command line.
--------------------------------------------------------------------------


******* Corrected cmd line: ./exports/bin/mpirun --display bind --np 2 --map-by :PE-LIST=3,4 --bind-to core hostname


[f8n02:1693833] MCW rank 1 bound to package[0][core:4]
[f8n02:1693833] MCW rank 0 bound to package[0][core:3]
f8n02
f8n02
**

[awlauria]

@markalle @gpaulsen fyi ^^

[awlauria]

Closing as fixed in v5.