-
Notifications
You must be signed in to change notification settings - Fork 184
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
Keep the PTX in CUDA binaries, so it can be JIT'ted for newer devices #6851
Keep the PTX in CUDA binaries, so it can be JIT'ted for newer devices #6851
Conversation
please test |
A new Pull Request was created by @fwyzard (Andrea Bocci) for branch IB/CMSSW_12_0_X/master. @smuzaffar, @mrodozov can you please review it and eventually sign? Thanks. |
enable gpu |
please test |
please test for slc7_aarch64_gcc9 |
please test for slc7_ppc64le_gcc9 |
-1 Failed Tests: UnitTests RelVals Unit TestsI found errors in the following unit tests: ---> test import-yaml had ERRORS RelVals
|
-1 Failed Tests: UnitTests Unit TestsI found errors in the following unit tests: ---> test import-yaml had ERRORS |
unrelated ? |
unrelated. for some weird reason on Arm and PPC this test is failing. |
+1 Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-0fc616/14657/summary.html GPU Comparison SummarySummary:
Comparison SummarySummary:
|
Out of curiosity, how much does the PTX add to the (shared) object file size? (even on qualitative order-of-magnitude scale) |
Here is a comparison of the libraries from a working area I had around. before (no PTX)
after (with PTX)
So, around 10% of the size for the CUDA-enabled plugins, on average ? |
Thanks. So noticeable but not large. Do you think the JIT'ting would be intended only for cases where new architecture appears after the build, or should we think of mainly JIT'ting the code? |
For the moment I would consider it as a fallback option for new architectures (e.g. Ampere), and to possibly improve the performance on less used architectures (e.g. JIT'ting for Somewhere in the NVIDIA documentation I've see a suggestion that we should build with I consider this PR a reasonable compromise. |
Sure, thanks. I was thinking more the future, e.g. should we add |
+externals |
This pull request is fully signed and it will be integrated in one of the next IB/CMSSW_12_0_X/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) |
Sure - well, maybe once we actually have some Ampere card to test :-) |
test parameters:
|
please test |
abort we need to wait for next IB with SCRAMV3 for external PR testing |
please test |
+1 Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-0fc616/14730/summary.html GPU Comparison SummarySummary:
Comparison SummarySummary:
|
No description provided.