-
Notifications
You must be signed in to change notification settings - Fork 4.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Redesign all GPU workflows to detect if a GPU is present, and fall back to CPU otherwise #33428
Conversation
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-33428/22092
|
A new Pull Request was created by @fwyzard (Andrea Bocci) for master. It involves the following packages: Configuration/PyReleaseValidation @perrotta, @jordan-martins, @chayanit, @wajidalikhan, @kpedro88, @cmsbuild, @srimanob, @slava77, @jpata can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
7bed67a
to
b955402
Compare
code checks |
please test |
It's not actually completed, but I wanted to ask the bot to test it so far ... |
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-33428/22108
|
Pull request #33428 was updated. @perrotta, @jordan-martins, @chayanit, @wajidalikhan, @kpedro88, @srimanob, @slava77, @jpata can you please check and sign again. |
enable gpu |
@smuzaffar @silviodonato is there a way to ask the bot to run the 11634.502 on cpu ? |
+1 Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-a96ae4/14239/summary.html Comparison SummarySummary:
|
+Upgrade |
Maybe a question for @qliphy : do we know why this PR is not merged yet given the consensus about the changes it brings? |
@cms-sw/alca-l2 @cms-sw/pdmv-l2 Do you have any comment? |
Several signatures are missing. This could be done quickly, otherwise we can discuss this at tomorrow's ORP. |
Thanks! |
+1 |
+1 |
+operations |
This pull request is fully signed and it will be integrated in one of the next master IBs (tests are also fine). This pull request will now be reviewed by the release team before it's merged. @silviodonato, @dpiparo, @qliphy (and backports should be raised in the release meeting by the corresponding L2) |
+1 |
PR description:
Redesign the GPU workflows:
###.501
) and GPU (###.502
) workflows should now be as close as possible;SwitchProducerCUDA
mechanism to detect if a GPU is available and offload a module or task to the GPU; if not, they automatically fall back to the equivalent CPU modules and tasks;pixelNtupletFit
) instead of a customisation, in line with the other workflows;Some changes to the relevant EDProducers have made the definition of the workflows easier:
Some other general changes and code clean up:
clone()
a module used in aSwitchProducerCUDA
gpuVertexFinder
kernels from gpuVertexFinderImpl.h to gpuVertexFinder.ccThe update has been presented here: https://indico.cern.ch/event/1033022/#47-gpu-workflows .
PR validation:
The GPU workflows (e.g.
###.502
) now work also without a GPU: