Re: Perl 5 Optimizing Compiler, Part 10: Kickstarter & Performance Benchmarks

[Laurent_R]

Sure enough, the benchmark results you are reporting look great. But you are not saying at all how they were obtained. And it seems from you own admission that some of the program versions were "hand-compiled", whatever this means. Sorry, too sketchy to be convincing. Your idea might be great, and I would like to believe in it, but so far, I wish you have convinced me.

[Will_the_Chill]

Hi Laurent_R,

Sorry if I'm not being clear. In the second paragraph of the Performance Benchmarks page you will see I have linked to the actual source code files that were run to generate the timings.

The first step toward reproducing the benchmarks yourself is to get RPerl installed and passing the basic tests, please see the use RPerl; page for more info. RPerl is still alpha software being developed, so don't hesitate to let me know if you need help or have any questions. I'm often in #perl11 on irc.perl.org if you use IRC.

We already have the basic scalar, array, and hash data types working in RPerl. The "hand-compiled" part of the performance benchmark refers to the fact that RPerl v1.0 is not complete yet and thus there is not yet an actual automatic compiling mechanism in place. Using the working RPerl data types I played the part of a compiler and manually compiled the sorting algorithm code from pure Perl 5 into both RPerl Perl-data mode and RPerl C-data mode, as explained in the Performance Benchmark page.

Did I answer your questions? My goal is to win your support and convince you to make a pledge to the RPerl Kickstarter campaign.

Thanks!

Perling,
~ Will

[vkon]

reini stated that his typed data in perl will be a huge speed win (preseented at Riga YAPC2011)

will his approach outperform yours and by what %-tage? where to invest?

[Will_the_Chill]

vkon,

Types are required in RPerl, and I believe they are optional in P2.

RPerl compiles low-magic Perl 5 source code into C/C++, there's just no way you can get faster than that, so we say RPerl is "very high performance". P2 is a JIT for high-magic Perl 5 and Perl 6 source code, it is based on the Potion VM and we say P2 is "high performance".

We will run full comparison benchmarks once RPerl v1.0 is complete, but the bottom line is that JIT dynamic compilation can never beat the speed of static compilation, all other things being equal.

Thank you for your generous pledge to support the future development of RPerl!

Perling,
~ Will the Chill