Author of Acovea looking for new ideas in Next Version

[ScottRobertLadd]

I'm working on Acovea 5.0 (http://acovea.coyotegulch.com) for release early next week. I appreciate the interest and support of the Gentoo community, and wnted to let people know about the forthcoming update so they can make comments.

The major changes are still in flux, but Acovea 5 <i>will</i> be updated for GCC 4.0 and 4.1. The existing XML configs have minor tweaks.

I've added the ability to optimize for compile time to Acovea's existing "optimize for code speed".

The new release includes a GUI frontend. The Acovea engine is now a library file called by command-line and GUI drivers.

And, if time permits, I have a few other extensions under development, including the ability to compare optimizations across sets of programs as opposed to single programs.

Any and all comments are welcome.

[brantgurga]

The sample programs for using acovea to analyze compiler arguments are generally program-type specific, emphasizing integer stuff or floating point, or repitition, etc. To better model general programs since Gentoo doesn't have per-package CFLAGS yet, perhaps it could include a sample program that is essentially all the other programs ran in sequence or something along those lines.

[SoTired]

I know that makefile support has been talked about before, and that's a feature that I'd really like. Basically anything that makes it easier to run the algorithm on real programs, rather than the default ones would be very nice (perhaps the gui does this already?)

Aside from that the other things that I can think of off the top of my head would be: Testing for binary size. Or, maybe a composite test with flexible weights for binary size, compile time, and binary speed.

Mainly though just better/easier support for getting it to work with other programs.

EDIT:
I know this could be added manually, but perhaps default support for some of the param arguements in the configuration files. It would be interesting to see how close the defaults, for things like automatic inlining, are to their optimum values (and how much those optimum values range over program type).

[brantgurga]

I've not done alot with acovea, but with the latest suggestion, it sounds like an extensibility mechanism of a configuration like:

build() {
; user-defined function using $acovea_flags for building program
}

success() {
; user-defined function to determine success criteria
}

perhaps more functions would be useful, but that is one obvious way to make these things extensible

[teh-steve]

[ScottRobertLadd]

Makefile support seems to be a big request - and it just so happens that I've got it covered (I think... it's testing right now).

I have a version that accumulates statistics across multiple tests, using a database; however, this is a big change that probably will wait for the next version. The forthcoming 5.0 focuses on architectural changes: improved threading support, a genericized engine that can be wrapped in various drivers, and better statistical output. And yes, version 5.0 (which will be out in a day or two) includes makefile support.

Plans for 6.0 include the statistics database, (maybe) a configuration editor/generator, and a battery of new benchmarks, including several composite tests.