task assignment report missing GPU update
The note on update-constraints being assigned to the GPU is in an entirely different place than the rest of the task assignment outputs and it is not present on the terminal output either; i.e. in a run that explicitly requests
On the log and the terminal we have:
1 GPU selected for this run. Mapping of GPU IDs to the 1 GPU task in the 1 rank on this node: PP:0 PP tasks will do (non-perturbed) short-ranged interactions on the GPU
And somewhere further down (only in the log):
Updating coordinates and applying constraints on the GPU.
Report if update is running on the GPU or on the CPU
This adds a line which states where update is offloaded next
to the rest of the GPU usage information.
#3 Updated by Szilárd Páll 3 months ago
- Status changed from Resolved to Feedback wanted
I think improvements to the documentation are still necessary. As noted on gerrit:
documentation only refers to "PP task" and "PME task" so it is now inconsistent in that it neither tells the user that there is a new "GPU task" that has to be mapped with the "-gputask" option nor does it document that this new task is part of the PP task and it is mapped with the respective identifier in the "-gputasks" specifier.
As we have not added a separate designator for the "update-constraints" tasks, we either need to add one (i.e. something like "PP:0,PME:0,UPDATE:0") or document that this is part of the PP task and print it consistently.
#4 Updated by Erik Lindahl about 1 month ago
Although Artem might already have a fix, remember that we also have a clear policy we should ONLY patch actual bugs in the release series, never cosmetics or features.
Each change we put in release-2020 is something that also has to be pulled into the master branch, and in practice that ends up adding to Paul's task list.
#5 Updated by Artem Zhmurov about 1 month ago
The fix was merged on Jan 13th (https://gerrit.gromacs.org/#/c/gromacs/+/15282/). I think the message was not exactly clear with respect to what we call tasks (see Szilard comment above). Shall we bump this to the next version or shall I update the message as Szilard suggested?