samuelkgutierrez closed issue hpc/quo-vadis#105.
Test double split operations
Create a program in tests to test the following scenarios. Let’s say we have a dual socket node with 2 GPUs per socket. We also have four MPI tasks.
…View Comment
wihobbs commented on issue flux-framework/rfc#529.
Oh, of the options so far I like sched.selection_type the most. @cmoussa1 @garlick your thoughts?…
trws commented on issue flux-framework/rfc#529.
I’m leaning toward something like your suggestion @grondo, maybe alloc_type or sched_selection_type? That and make the strings the same here and in the PR and I think we’re probably good….
wihobbs commented on issue flux-framework/flux-sched#1528.
I’ll look at #1424, as you point out with this going into the next release it might be good to tackle that as well….
trws commented on issue flux-framework/flux-sched#1528.
That’s a fair point. The advantage to having the reserved boolean tracked here is I think I would need to do the same thing to solve #1424, because we need to know that something was reserved (and thus sent a time estimate) to clear those estimates the next time around. I suppose we could separate them and set “reserved” along with the time estimate but not track that it was reserved at the end? …
vsoch open issue converged-computing/fleetq#4.
Descriptive Dispatch Experiments Planning
Putting some notes here before I head out:…View Comment
JustinPrivitera commented on issue visit-dav/visit#21036.
Hi @marcdurant. I apologize if you have already done this, but I asked codex to have a look at your crash report and comments here and it had a theory about what might be going wrong:…
alecbcs commented on issue spack/spack-packages#4437.
@bernhardkaindl went in and fixed the merge conflicts on the behalf of @dacolombo. …
v-dobrev commented on issue spack/spack-packages#143.
Update: I planning to run some tests with GPU builds and also see if more of the conflicts("^mfem build_system=cmake") are really necessary….
gardner48 commented on issue llnl/sundials#976.
This will depend on the index size selected at configuration time. By default this is a 64-bit signed integer type e.g., int64_t….
JustinPrivitera commented on issue llnl/conduit#1646.
@cyrush can you have a look at this? We discovered that the blueprint transforms can give wrong results if you call them with empty nodes. …
adrienbernede commented on issue llnl/Umpire#1076.
Can’t merge it myself……
artv3 commented on issue llnl/RAJA#2054.
> > Regarding old implementation of atomicGeneric in CUDA is giving incorrect results is that a CUDA related bug? or why is that happening?
…
adayton1 commented on issue llnl/CARE#379.
> If you can get access to a bank (other than guests) on matrix, the best is to define it as the default one. But otherwise, yes, we can easily set the CI to use it in the alloc options (.gitlab/custom-variables.yml). …
adrienbernede commented on issue llnl/CARE#379.
If you can get access to a bank (other than guests) on matrix, the best is to define it as the default one. …
vsoch commented on issue kubeflow/trainer#3712.
Primarily “here is a container or Dockerfile with out app and the command to run it at a small size.” The goal would be to reproduce what you want to do for testing….
tgamblin open issue hpsfoundation/tac#88.
[Project Proposal] mars-earth / pymars readiness inquiry
### 1. Name of Project…View Comment
milroy commented on issue flux-framework/flux-sched#1316.
> Final review from @trws then merge? …
jameshcorbett commented on issue flux-framework/flux-sched#1316.
Final review from @trws then merge?…
tpatki commented on issue flux-framework/flux-rest-server#3.
No worries, @garlick. The new PR is up….
garlick commented on issue flux-framework/flux-pmix#131.
It was kind of a baddie, and @sam-maloney ‘s excellent issue description made it easy to quickly narrow down. Thanks for the review! I’ll set MWP and ask Sam to confirm the fix when he has a chance….
sam-maloney commented on issue flux-framework/flux-pmix#130.
To add a couple of quick tests running flux pmi:…
garlick open issue flux-framework/flux-core#7720.
content: flux kvs put --sync can hang under certain error conditions
Problem: as noted in #7702 , the content cache can get stuck flushing out dirty blobs under certain conditions:…View Comment
chu11 commented on issue flux-framework/flux-core#5083.
I’ve had this on my “to cleanup” list for a long time, but as @grondo says above, what is the best/easiest solution. Using Claude Opus tried to get a gauge on how disruptive a change would be using a number of the suggested paths above:…
grondo commented on issue flux-framework/flux-core#7699.
In a meeting this was actually described as not a real requirement. Rather than try to guess at what solution is really desired here, let’s close this and later open a different issue with clear goals….
jeremylt commented on issue CEED/libCEED#1987.
Note - merging main into your branch instead of rebasing means I’d recommend we squash-merge this PR…
hppritcha open issue open-mpi/ompi#14145.
ABI: Code Duplication: c_header.py Re-implements Domain Knowledge from ompi_bindings
## Code Duplication: c_header.py Re-implements Domain Knowledge from ompi_bindings…View Comment
BradWhitlock closed issue llnl/axom#1689.
Centralize logic related to units and conversion
We have logic in Klee, in the Quest Open Cascade reader and in the shaping tutorial related to units and conversions between them….View Comment
hppritcha commented on issue hpsfoundation/tac#93.
@jwillenbring …
grondo commented on issue flux-framework/rfc#530.
Required by mergify:…
jameshcorbett commented on issue flux-framework/flux-multi-cluster-utilities#29.
While working on this, I ran into another issue, which was that I can’t seem to update the jobspec in the jobtap plugin to have attributes.system.exec.test.override=1 . I’ve been working around this by setting that attribute directly on the jobspec at submission time. @grondo suggested making this edit in flux-core to track down the issue:…
sam-maloney commented on issue flux-framework/flux-core#7677.
I would immediately think that target should only ever be 0 or 1, as otherwise the request would skip over intermediate instances, which feels like a violation of the hierarchical model. Certainly for a shrink/partial release, those resources would have to be removed from the resource set of each instance in the hierarchy in any case, and then you have the more philosophical question of what gives a lower level application/instance the right to determine what its parent/ancestor instances do with their resources. I suppose a user might reasonably want a way to indicate that an application is releasing resources which are no longer needed by its own local instance for further workflow steps, but that would be different from the target proposed here, as the shrink would still need to be dealt with by the local instance first, it would just be guaranteed to be immediately followed by an equivalent job_manager.dyn_alloc_request from the local instance to its immediate parent.
…
cmoussa1 commented on issue flux-framework/flux-accounting#896.
Thanks @jameshcorbett! That is a good question you brought up. This segregation between child banks and users is intentional and is not allowed behavior for a hierarchy in flux-accounting. So, no, banks and users cannot exist under the same parent bank, so I think we should be okay here….
jeremylt commented on issue 4C-multiphysics/4C#2012.
Ah, this was a missing piece …
becker33 open issue spack/spack-tutorial#572.
Add docker to container image
We now use docker in the binary caching section, but we never updated the image to include it….View Comment
imguoguo commented on issue sipeed/picoclaw#2251.
@loafoe Hi! Since there’s been no activity after the earlier reminder, I’m closing this Grafana Alertmanager channel PR for now to keep the queue tidy. The work looked solid, so if you’d like to pick it back up, feel free to reopen anytime and we’ll gladly continue the review….
tzanio commented on issue mfem/mfem#5397.
Merged in next for testing……
chu11 commented on issue flux-framework/rfc#528.
> One point that might need clarification: can MODEL_VERSION contain whitespace? …
wihobbs commented on issue flux-framework/flux-core#7716.
Thanks for calling my attention to annotations, which do seem like the right approach. The RFC doesn’t stipulate what happens to a job’s annotations after the job enters the CLEAN state, unless the job fails (they are discarded). Do they persist in the KVS? Or is the way to reference a past job’s annotations through the eventlog?…
sam-maloney commented on issue flux-framework/flux-accounting#885.
@cmoussa1 I haven’t actually run a build since, but once v0.59.0 releases I will rebuild my RPMs and report back here on the outcome!…
jedbrown commented on issue rsmpi/rsmpi#224.
Thank you! Sorry about the delay….
jedbrown commented on issue openjournals/joss-reviews#9333.
The MPI reference @mpi1994standard does not appear in the bibtex file. You might be intending to cite @mpi50. The Ringler 2013 paper above does have a DOI, but is not cited. If it’s not intended to be cited, then the check above can be considered a false positive….
jandrej commented on issue mfem/mfem#5350.
> This is an interesting PR but I think it requires some discussion with the MFEM team at large. One of the basic tenets of the finite element method is that there is one mesh which is subdivided into elements on which we represent our fields. This PR relaxes this requirement for the special case where one mesh is a sub-mesh of another. Another alternative would be to compute bilinear forms on the primary and sub-mesh separately and combine these separate operators into a block system using coupling operators to define the mapping between finite element spaces. …
balos1 commented on issue llnl/sundials#966.
@copilot apply the changes in the diff produced by the failing litgen_check…
markcmiller86 commented on issue llnl/Silo#535.
Working on it. Just fixing a bug in DBCp() when copying between drivers
…
cyrush commented on issue visit-dav/visit#21047.
```…
cjvogl commented on issue pnnl/SPAECIES#52.
@quantheory I’ve rebased this branch off main that has #49 merged in….
v-dobrev commented on issue mfem/mfem#5334.
The diffs in “Files Changed” still show some changes that were already merged in master through PR #5330 – I suspect merging master here will fix that….
bgunnar5 commented on issue llnl/mada-tools#13.
Closing this as #15 will include these files….
cyrush commented on issue llnl/conduit#1642.
This is not a good change, the ownership semantics have to do with how the schema tree is owned, and a simple swap will not work….
tzanio commented on issue mfem/mfem#5393.
This PR is now under review (see the table in the PR description). To help with the review process, please do not force push to the branch….
v-dobrev commented on issue mfem/mfem#5259.
Note: I used a rebase + squash to merge the last two commits into next (ignoring the last merge of master into this branch)….
adrienbernede commented on issue llnl/RAJA#2045.
@johnbowen42 OK fixed !…
garlick commented on issue flux-framework/flux-core#7681.
Just pushed a big update that I think addressed all the review comments: …
trws commented on issue GTNewHorizons/CropsNH#189.
Thanks for the reasoning. It’s unfortunate but makes sense, honestly I didn’t even remember that the old version could be used for sand. Great work on making the migration mostly seamless….
becker33 commented on issue spack/spack#52652.
@haampie yes, it is. The reason I want it as a separate flag is because it will need to be the primary avenue for uninstalling specs with circular dependencies, but it seemed separate enough to be a separate PR….
adrienbernede commented on issue llnl/radiuss-spack-configs#187.
@johnbowen42 I invited you to the radiuss group in GitLab so that you automatically get added to subprojects like radiuss-spack-configs and raja mirror repos….
grondo commented on issue flux-framework/flux-core#7701.
Ok, I’ve addressed @chu11’s comments and pushed the result. Will set MWP….
cmoussa1 commented on issue flux-framework/flux-accounting#871.
I’ve rebased to catch up after a number of approved PRs. I know this has already been approved, but I believe that was before we noticed a couple of issues fixed by #880. @grondo, I’m not sure if you would want to give this another a look before I officially merge this! …
garlick commented on issue chaos/nodediag#24.
Thanks!…
jeremylt commented on issue 4C-multiphysics/4C#2012.
The concern about build time makes sense. Though, I was under the impression that we do not currently have CUDA hardware that we can run CI on and OpenMP could be used as a surrogate to ensure shared memory parallelism is covered by CI once it is added in follow up development work?…
cyrush commented on issue visit-dav/visit#21045.
@andymwood thanks for the report….
alecbcs commented on issue spack/spack-packages#5428.
@spackbot run pipeline…
alecbcs commented on issue llnl/repligit#190.
I think I prefer to simplify down to only instantiating the variable for line2 since we only ever use line1 once….
alecbcs commented on issue llnl/radiuss-spack-configs#187.
Haha well this is unintuitive [at]lc-hubcast run pipeline reruns the pipeline on the last commit that was synced. But since this wasn’t synced recently it’s just effectively rerunning the pipeline on the old commit.
…
bgunnar5 commented on issue llnl/mada-tools#13.
@lc-hubcast run pipeline…
bgunnar5 commented on issue llnl/hubcast#355.
@cmelone Ok thanks for the heads up. Any idea how to resolve this? Does this have to do with how my GitLab CI is configured or my Hubcast YAML?…
cyrush open issue llnl/conduit#1641.
add example of conduit::relay::mpi::io::blueprint::save_mesh to docs
We have examples of the non mpi variant, but folks don’t realize the mpi variant exists. …View Comment
adayton1 commented on issue llnl/Umpire#1081.
This call now fails with the error below: blt_convert_to_system_includes(TARGET umpire)…
vsoch commented on issue kubeflow/trainer#3427.
@andreyvelich is the user application specifying the entrypoint with mpirun, and the issue is about the environment being discovered for it? If yes, the approach the flux operator takes is to take the application command, and wrap that. In the case of PYTORCH, the problem (I think, if I understand correctly) is that normally there is another layer (the workload manager) like Flux or Slurm that you use to run the job, and its the workload manager that ensures the environment is properly set. If you just expect a TrainJob to be executing a pytorch script, and then have all the envars set (and correct) for leader and launcher, then you are essentially re-implementing a workload manager. …
garlick commented on issue flux-framework/flux-sched#1489.
I’ll go ahead and set MWP since this was already approved. Thanks!…
chu11 open issue flux-framework/flux-core#7703.
libkvs/treewalk: support treewalk_strerror / errormsg
Between flux-fsck, flux-dump, and flux-gc, there’s a lot of handling of error codes -> log messages in the walk error handling. Might be worthwhile to reduce repeat code and deal with newer error codes in the future….View Comment
jedbrown commented on issue RDycore/RDycore#398.
Does temperature converge at second order if you shut off transport?…
emily-howell closed issue Alpine-DAV/ascent#1494.
support for actions_file[s] ?
cyrush commented on issue Alpine-DAV/ascent#1652.
@emily-howell …
jedbrown commented on issue openjournals/joss-reviews#6099.
The JOSS paper DOI loads now for me. Thanks, all!…
balos1 commented on issue llnl/sundials#973.
@lc-hubcast approve…
BradWhitlock closed issue llnl/axom#1750.
Centralize CMake logic about available policies for tests
@BradWhitlock pointed out in #1749 that we have a lot of repeated CMake logic in our tests related to which policies are available. We should centralize this into a utility function/macro….View Comment
vsoch commented on issue kubernetes/enhancements#4671.
Just wanted to say I’ve been developing with this during June using the kind image 1.36.1 (or GKE 1.36.0-gke.2684000) and am so thrilled to see it moving through. This has been a long time coming, from Coscheduling in the scheduler-plugins repository to many other plugins, and (looking at the dates of this issue from 2024) I am so pumped to have this naively! Bring on the gangs, and everything that supports! Awesome work to everyone here!
samuelkgutierrez open issue hpc/quo-vadis#104.
Update Device Split Semantics
Assume the following scenario: …View Comment
eleon commented on issue hpc/quo-vadis#104.
Thank you, @samuelkgutierrez . This example brings an interesting semantics consideration. …
jeremylt commented on issue CEED/libCEED#1985.
Ope, I need to update the LIBXSMM hash to use in the gitlab yml…
cjvogl commented on issue mfem/mfem#4244.
Documenting here that I did run the modified ex1p provided above, which runs successfully on this branch while causing an error on master.
…
artv3 commented on issue llnl/RAJA#2006.
> ```c++ …
artv3 closed issue llnl/RAJAPerf#689.
Incorrect problem config for MASSVEC3DPA
Describe the bug…View Comment
wihobbs open issue flux-framework/flux-core#7695.
`flux-job-wait-event(1)` has `--help` output, but isn't listed in `flux-job(1)`
LivChat taught me about this command this morning, is there a reason it’s left out of the manpage? Maybe just meant for the testsuite?…View Comment
cjekel open issue cjekel/piecewise_linear_fit_py#133.
PiecewiseLinFit.fit no longer working after Scipy 1.18 upgrade
scipy.linalg.lstsq slightly modified the residues argument in their lastest (1.18) release. Causing a crash in the fit routine…View Comment
hppritcha commented on issue open-mpi/ompi#12796.
@mdosanjh are you still working on this? …
grondo commented on issue flux-framework/flux-core#7621.
Great! Already approved so feel free to merge when ready!…
garlick commented on issue flux-framework/flux-core#7621.
OK, just pushed an update with the following changes: …
markcmiller86 commented on issue visit-dav/visit#21034.
Will do…will have to wait ‘til monday though….
v-dobrev commented on issue spack/spack-packages#143.
Hi @jacobmerson, …
tgamblin commented on issue spack/spack#52613.
@haampie: I’ve reworked this to use rsvg, which preserves fonts properly and renders all the figures as a browser would. CI is also updated….
tzanio commented on issue mfem/mfem#5383.
This PR is now under review (see the table in the PR description). To help with the review process, please do not force push to the branch….
v-dobrev commented on issue mfem/mfem#5383.
ping: @tomstitt …
artv3 commented on issue mfem/mfem#5382.
@helloworld922 , the RAJA API takes resource objects that can be used to specify the stream, we have examples here: …
adayton1 commented on issue mfem/mfem#5383.
Thank you!…
balos1 closed issue llnl/sundials#955.
[BUG] arkLsMassSolve ignores some return values from user callbacks.
One of the folks working on the deal.II SUNDIALS interfaces, @vovannikov, has spent what I think must have been a quality afternoon tracing why one of our tests fails. The details are in https://github.com/dealii/dealii/pull/19838 but come down to code that I (mis-)wrote: If a user callback fails in user-provided linear solvers, I either returned +1 or -1, depending on whether it’s a recoverable or non-recoverable failure (and zero if everything is fine). I think that’s what the documentation of SUNDIALS at some point said I should do, but it’s possible that that was about the nonlinear solver callbacks in KINSOL and not about the SUNDIALS linear solvers. Regardless, for the linear solvers, I need to return SUNLS_PSOLVE_FAIL_REC or SUNLS_PSOLVE_FAIL_UNREC, with values +805 and -808, respectively….View Comment
jameshcorbett commented on issue llnl/scr#614.
Sorry for the delay in getting back to you! Yes we would be interested in working with you on this. I’m not sure yet how complex it would be to support this in SCR but I’d be very happy to investigate with you!…
adrienbernede open issue llnl/radiuss-shared-ci#79.
Hmm, interesting... I read the same statement that it was "gpu allocated" associated with the suggestion to use "--gpu".
Hmm, interesting… I read the same statement that it was “gpu allocated” associated with the suggestion to use “–gpu”….View Comment