A simple optimization proposal (original) (raw)

Martin Grajcar maaartinus at gmail.com
Thu Feb 13 20:40:23 PST 2014

Hi Kris,

some comments on http://cr.openjdk.java.net/~kmo/8003585/webrev.02/src/share/vm/opto/subnode.cpp.sdiff.html

I could imagine adding cmp1->in(1) == cmp2 as an alternative at line 1217. While hardly anybody writes a.length & index, some previous transformation could rearrange the expression.

The description at line 1221 mentions unsigned compare only while you're handling both. I'm unsure about the correctness in the signed case (but didn't check it). Some sort of proof would be nice, especially for extreme values.

I like the idea at line 1237, though I don't exactly understand how it works. Is there a good description of all the nodes and all normalizations done?

Regards, Martin. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20140214/539b99a8/attachment.html

More information about the hotspot-compiler-dev mailing list