Move performance tests and benchmarks from "make test" to a new "make benchmark"
A non-trivial portion of "make test" jobs is spent on performance testing. These tests should be moved to a separate "make benchmark" so that automated unit and integration testing can complete more quickly.
#1 Updated by Sam White over 1 year ago
And at least some of the results of "make benchmark" should be plotted online somewhere every night based on some of the autobuilds whose performance we care about tracking. It would be useful for diagnosing performance regressions. We already do this for netlrts charm++ pingpong on the PPL wiki page.
I would want example/ampi/pingpong and tests/charm++/pingpong at a minimum. Potentially some reduction/broadcast tests, Converse/LRTS tests, and maybe some example programs to all have that.
This could be a good Charmworks intern task.