these were assigned to me by accident

Could you assign this to me ... working on it.

Fixed:
https://github.com/clojure/test.benchmark/commit/f4c0802af58f5f3341816be40bf1388029bc0d3e
https://github.com/clojure/test.benchmark/commit/f4c0802af58f5f3341816be40bf1388029bc0d3e

https://github.com/clojure/test.benchmark/commit/f4c0802af58f5f3341816be40bf1388029bc0d3e

This is a first pass at a mandelbrot that takes advantage of clojure 1.3 performance features.

Approach:
Avoid idiomatic functional code.
Extensive iteration, in-place mutation, primitive arrays, manage our own threads etc. Not "pretty."
(currently .8 to .9 % of fastest java version).

Tests:
Correctness tests, verified against a set of known-good resource files.

Caveats:
This is the first pass and my first time writing optimized clojure 1.3. I'm sure there are things that can be improved, so any feedback welcome.

Future:
I've included a modified version of the fastest java program in the patch as I've found it useful for comparison
(# time script/run_java alioth.java.mandelbrot 16000)
One thing we don't have is repeatable performance metrics/tests.
One approach would be to have tests that fail whenever the clojure version falls below a certain percentage of the java version.
That's something I'm working on.

Per discussion on mailing list (https://groups.google.com/d/topic/clojure-dev/Kr5teQh69Pw/discussion) removed java files.
Files pushed.

Paul, it is fine for you to close tickets yourself on contrib projects (all the notes about "core team only" apply to Clojure itself, not the contribs).

This can be closed. Patch pushed.

TBENCH-4

Fixed, https://github.com/clojure/test.benchmark/commit/642279639c99ce222c10a4d1f1f69aa5565845c6

these were assigned to me by accident

done as far I can tell.

Done, https://github.com/clojure/test.benchmark/commit/b0772ea5858a3924b15d87003cb039af20c0cfa7

these were assigned to me by accident

Done, https://github.com/clojure/test.benchmark/commit/642279639c99ce222c10a4d1f1f69aa5565845c6

re-worked A. Fingerhut's implementation to use nio and a hand-rolled reduce loop for code replacement.
Result is a ~2x speedup and 92% as fast as the fastest Java version.

https://github.com/clojure/test.benchmark/blob/0977c41e430b85f452f0064aad24562e9be48fff/src/main/clojure/alioth/regexdna.clj

these were assigned to me by accident

these were assigned to me by accident

these were assigned to me by accident

Have added my latest version of the k-nucleotide benchmark in the following repository:

https://github.com/mikera/test.benchmark/

Please create a patch of your changes and add it to the ticket, thanks!

If you add a patch, I can take a look at this tonight.

Mike,
I did pull down your repo and ran some tests.
This version of k-nucleotide does run faster than the currently submitted version in alioth.
However, we've got a ways to go to catch the faster Java implementations.

I've added a few comments in your repo.
Of note, the (set! unchecked-math true) directive at the top of files allows us to use inc, dec, etc directly without explicitly specifying unchecked-inc, etc.

http://shootout.alioth.debian.org/u64q/program.php?test=knucleotide&lang=java&id=9 is the fastest java version, but is a bit gnarly.
http://shootout.alioth.debian.org/u64q/program.php?test=knucleotide&lang=java&id=2 is a little slower, but in the same order of magnitude for performance, and significantly more straightforward.

You might check those out for some inspiration to boost the performance of clojure's knucleotide.
Good luck and thanks for the help!

64bit Linux, Quad-core

Patch which adds an stm-namespace containing a collection of performance test functions.