-
Enhancement
-
Resolution: Unresolved
-
P4
-
25
In JDK-8234863 we increased the default MaxInlineLevel based on user feedback and benchmark numbers.
The new level was relatively conservative based on observations that a higher inlining limit might have negative consequences, especially in C1. We've seen specific benchmarks benefit of a higher level though, and are aware that certain customers tune this more aggressively.
Some flags were later split so that C1 has its own set (and the C1 inlining increase rolled back) byJDK-8235673. This removes a few stability concerns and might make more aggressive (C2) defaults somewhat more appealing.
We are getting occasional requests to increase some limits, especially those which might make functional style programming (with large chains or trees of simple methods, e.g., stream pipelines) more performant.
I'm opening this RFE mostly to give a heads up. We need to collect fresh data on this and try to look at the big picture changes in C2, including changes in overheads over time.
The new level was relatively conservative based on observations that a higher inlining limit might have negative consequences, especially in C1. We've seen specific benchmarks benefit of a higher level though, and are aware that certain customers tune this more aggressively.
Some flags were later split so that C1 has its own set (and the C1 inlining increase rolled back) by
We are getting occasional requests to increase some limits, especially those which might make functional style programming (with large chains or trees of simple methods, e.g., stream pipelines) more performant.
I'm opening this RFE mostly to give a heads up. We need to collect fresh data on this and try to look at the big picture changes in C2, including changes in overheads over time.