Possible compression performance optimization about compress HOT 6 CLOSED

Thanks, I will try (micro)benchmarking this. It is difficult to say for sure, given how much hotspot rearranges things, but I agree in that conditional expressions can be much more costly (due to branching etc).

from compress.

Hmmh. Replacement with byte masking (I hate java's signed bytes...) becomes something more like:

                || (((in[ref] & 0xFF) << 8) | (in[ref+1] & 0xFF)) != ((hash >> 8) & 0xffff)) {

which unfortunately is bit slower on my tests...

from compress.

You are right, in my code I also have "& 255" I wonder why I forgot it here.

I guess the reason why it is a bit faster on my machine is that I have different test data.

from compress.

Ok. Which test file(s) did you use? I have simplest of tests under src/test/java/perf/ManualPerfComparison.java which I used on just one of compression files.

from compress.

I used the data I need to compress in my case, which is database page data (from the H2 database).

I guess it would be better to optimize performance for a more common mix of data, for example the test data benchmarks: http://github.com/ning/jvm-compressor-benchmark/wiki

from compress.

Yeah I use test data from that test benchmark. Actually if you know of additional data sets I could also those to jvm-compressor-becnhmark. I do believe that there may be improvements to do for compress side, for what that's worth, just need to figure out good ordering for test etc.

But I also need to occasionally compare performance between versions; I did this to check how 0.7 and 0.8.5 compare (and determine there's 10-15% improvement for uncompress).

from compress.