InlineCachingCallSite perf degradation due to JRUBY-2477 fix

Charles Oliver Nutter added a comment - 23/May/08 3:15 AM Fixed in 6759 with a benchmark added in 6757. I fixed it for now by adding a "misses" counter to InlineCachingCallSite. The counter is incremented for every time a new method must be cached. If it reaches 50 (arbitrarily chosen by me) it will no longer attempt to cache methods. For smaller, cheaper methods, the improvement is substantial: Before: control, monomorphic dispatch 0.899000 0.000000 0.899000 ( 0.898000) test, bimorphic dispatch 3.032000 0.000000 3.032000 ( 3.032000) After: control, monomorphic dispatch 1.065000 0.000000 1.065000 ( 1.065000) test, bimorphic dispatch 1.493000 0.000000 1.493000 ( 1.493000) I will investigate the slight degradation in monomorphic dispatch, but for now it appears to be only noise.

Charles Oliver Nutter added a comment - 23/May/08 3:55 AM The degradation does appear to be non-anomalous...I will continue investigating.

Charles Oliver Nutter added a comment - 23/May/08 4:13 AM I discovered the degradation was actually due to a flaw in the test: because the classes under test were singletons, and because they were recreated repeatedly during the course of the test, the monomorphic (control) test eventually flipped polymorphic, causing the call sites to deoptimize. This does potentially cause a small concern, since sites that are "mostly monomorphic" may still benefit from caching even if they are "occasionally polymorphic". We also have too few metrics being gathered on call sites to either know what's a good deoptimzation threshold or how well the call sites are working. I will file a separate bug for this. Ignoring the sometimes over-zealous deoptimization, the new repaired benchmark is significantly faster for both control and test: Before: control, monomorphic dispatch 1.295000 0.000000 1.295000 ( 1.295000) test, bimorphic dispatch 3.551000 0.000000 3.551000 ( 3.551000) control, monomorphic dispatch 1.772000 0.000000 1.772000 ( 1.773000) test, bimorphic dispatch 3.032000 0.000000 3.032000 ( 3.032000) control, monomorphic dispatch 0.896000 0.000000 0.896000 ( 0.896000) test, bimorphic dispatch 2.965000 0.000000 2.965000 ( 2.965000) control, monomorphic dispatch 0.899000 0.000000 0.899000 ( 0.898000) test, bimorphic dispatch 2.951000 0.000000 2.951000 ( 2.950000) control, monomorphic dispatch 0.889000 0.000000 0.889000 ( 0.889000) test, bimorphic dispatch 2.983000 0.000000 2.983000 ( 2.983000) After: control, monomorphic dispatch 1.171000 0.000000 1.171000 ( 1.171000) test, bimorphic dispatch 1.065000 0.000000 1.065000 ( 1.065000) control, monomorphic dispatch 0.750000 0.000000 0.750000 ( 0.750000) test, bimorphic dispatch 0.957000 0.000000 0.957000 ( 0.957000) control, monomorphic dispatch 0.773000 0.000000 0.773000 ( 0.773000) test, bimorphic dispatch 0.927000 0.000000 0.927000 ( 0.927000) control, monomorphic dispatch 0.779000 0.000000 0.779000 ( 0.779000) test, bimorphic dispatch 0.960000 0.000000 0.960000 ( 0.960000) control, monomorphic dispatch 0.773000 0.000000 0.773000 ( 0.773000) test, bimorphic dispatch 0.966000 0.000000 0.966000 ( 0.966000)