Use the code motion trace produced by TraceOptoPipelining (excluding traces of
stubs) to assert that two compilations with the same seed cause StressLCM and
StressGCM to take the same randomized decisions. Previously, the entire output
produced by PrintOptoStatistics was used instead, which has shown to be too
fragile. Also, disable inlining in both TestStressCM.java and the similar
TestStressIGVN.java to prevent flaky behavior, and run both tests for ten
different seeds to improve coverage.
Reviewed-by: kvn, thartmann
Remove test assertion checking that different random seeds lead to different
code motion decisions. This was the case for the specific pair of random seeds,
IR fed to code motion, and target platforms tested originally; but does not need
to hold in general. Remove similar test assertion in IGVN randomization test
case. Re-enable the test case.
Reviewed-by: shade, kvn
Use the compilation-local seed in 'StressLCM' and 'StressGCM' rather than the
global one. As a consequence, these options use by default a fresh seed in every
compilation, unless 'StressSeed=N' is specified, in which case they behave
deterministically. Annotate tests that use 'StressLCM' and 'StressGCM' with the
'stress' and 'randomness' keys to reflect this change in default behavior.
Reviewed-by: kvn, thartmann
Add 'StressIGVN' option to let C2 randomize IGVN worklist order. When enabled,
the worklist is shuffled before each main run of the IGVN loop. Also add
'StressSeed=N' option to specify the seed. If the seed is not specified, a
random one is generated. In either case, the seed is logged if 'LogCompilation'
is enabled. The new options are declared as production+diagnostic for
consistency with the existing 'StressLCM' and 'StressGCM' options.
Reviewed-by: kvn, chagedorn, thartmann
