You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
I am looking at this flaky test, had some queries and comments:
The test compares normal memcpy/memset against with using openmp, and expects normal memcpy to be always faster . Although its will be true for small data size ( coz of overhead of creating threads with Openmp), but Openmp will be faster for larger data size , as with multiple threads utilizes greater memory bandwidth.
Could you please let me know if I am missing something here. (reference to test code )
@spidydev I found this test to vary a lot depending on machine type as well. On my AMD machine it was consistently failing (openmp was always faster) and my intel machine it was consistently passing. I think this would be confusing for users so my recommendation would actually be that we remove this test.
Description
Test relies on timings, which may or may not explain why it sometimes fails in CI.
Things to try
We could do a 'near' comparison.
http://jenkins.mxnet-ci.amazon-ml.com/blue/rest/organizations/jenkins/pipelines/incubator-mxnet/branches/PR-9609/runs/22/nodes/427/log/?start=0
Environment info (Required)
CI Environment failure.
The text was updated successfully, but these errors were encountered: