Please review JDK-8061391

Marcus Lagergren marcus.lagergren at oracle.com
Wed Oct 22 15:08:50 UTC 2014


Hi! Please review this!

I added specialized builtins for concat, which are up to 10 times faster in microbenchmarks with just one element (or array) being concatenated.
While doing this I happened upon some bugs with preventExtensions for arrays, and removed the hateful unreliable SwitchPoint from ContinuousArray, replacing non extensible arrays with an ArrayFilter, which is the way I should have done it in the first place.
I also had to turn NoTypedArrayData into a structure that inherited ContinuousArray, or I threw away a lot of optimistic link sites to untouched arrays in ScriptObjects. This has other performance implications (good ones).

ant test runs cleanly
ant test262parallel runs cleanly
octane runs with no regressions, some benchmarks like raytrace and crypto are even a bit faster.

There are regressions in octane since the last time I checked, but not due to my fix. I will investigate and file bugs.

Bug: https://bugs.openjdk.java.net/browse/JDK-8061391
Webrev: http://cr.openjdk.java.net/~lagergren/8061391/

Regards
Marcus



More information about the nashorn-dev mailing list