[12] RFR(M): 8210215: C2 should optimize trichotomy calculations

Vladimir Kozlov vladimir.kozlov at oracle.com
Tue Oct 9 17:35:24 UTC 2018

Looks good.


On 10/9/18 10:29 AM, Tobias Hartmann wrote:
> Hi,
> I had an off thread discussion with John and he suggested the following changes in addition:
> - Use randomization for integer values in the test
> - Add some asserts and comments to improve readability
> - Don't apply optimization to floating point compares because they don't obey trichotomy laws (for
> example, NaN always compares to false)
> - Handle the case where the merge result is BoolTest::illegal (just for robustness, I haven't seen
> this ever during testing but it might show up when overflow tests mix with "normal" other tests)
> - While testing, I found that we also have to add a call to add_users_to_worklist(iff1) to make sure
> that the dead if is removed (otherwise we hit failures like JDK-8075922).
> Incremental webrev:
> http://cr.openjdk.java.net/~thartmann/8210215/webrev.04_inc/
> Full webrev:
> http://cr.openjdk.java.net/~thartmann/8210215/webrev.04/
> Best regards,
> Tobias

More information about the hotspot-compiler-dev mailing list