Hi Chris, On 03/08/2016 12:55 AM, Chris Hegarty wrote: > Updated links: > http://cr.openjdk.java.net/~chegar/8151384/webrev.01/
*) Your previous patch had the explicit access to CharacterDataLatin1.instance.(toLowerCase|toUpperCase). Any reason not to use it in your new patch? Probably saves a shift and branch on a hot path -- j.l.String is one of those magical places where it actually matters. Otherwise looks good. > http://cr.openjdk.java.net/~chegar/8151384/bench.01/ *) This one worries me a little bit. ASCIICaseInsensitive has the anomalous 360 us/op result in one of the forks, and that "best" is better than average StringCaseInsensitive: http://cr.openjdk.java.net/~chegar/8151384/bench.01/AvailableCharsetsCompare_afterChanges.txt This might be the genuine run-to-run variance with compiling the hot loop; I'd recommend using Blackhole.consume, not the accumulator variable to try avoiding that. Otherwise looks good. Thanks, -Aleksey
