> Interested in some Core i5 benchmarks? Sure, add them to the JIRA issue if you can, please.
> I just ran the benchmark locally on latest JDK 6 and it was slightly better > than the I7 results you posted which made me wonder.. Well, like I said -- they may depend on the architecture of the computer itself, not only on the CPU (memory reading throughput will be the key). The part you should be looking at is not how results compare between machines, but if you gain anything between the JRE version and the implementation used in Lucene's trunk. The intrinsics for popcnt have been introduced in Java 1.6u18 -- I thought they're in 1.7 only, but was wrong on this one: http://java.sun.com/javase/6/webnotes/6u18.html What Yonik mentioned would be equally interesting (a simple loop instead of the unrolled algorithm). I did try this in my previous tests (not on hardware popcnt) and it was slower. I'll try to see into it tomorrow if I find a spare minute. D. --------------------------------------------------------------------- To unsubscribe, e-mail: java-dev-unsubscr...@lucene.apache.org For additional commands, e-mail: java-dev-h...@lucene.apache.org
