Exclude a couple of tests which were found to be flaky.
For OpenCL, there is a set of tests which sometimes timeout when tests are run parallel. Sometimes these tests explode in time taken to complete - from 15s to 1500s, for example. If you set --repeat until-fail:10 it's almost guaranteed that some of them will timeout.
Most of these tests seem to involve managed memory.
hipEventRecord sometimes reports negative time for OpenCL
This needs to investigated, but for now let's just run these tests in a single thread to get the CI system to be more stable.
--repeat until-fail:10it's almost guaranteed that some of them will timeout.This needs to investigated, but for now let's just run these tests in a single thread to get the CI system to be more stable.