I’ve gotten smash++’s Rakudo benchmarks running again, and I’m trying to get a full set of benchmark graphs generated. While moritz and I are debugging the graphing program, here are the first results. (Assuming I can get them to display here!)
Okay, two days later, I’ve finally got inkscape running again to generate PNG versions of the graphs. And with any luck, here they are:
Update: I forgot people who weren’t familiar with the original benchmarks would be reading this. All graphs are labeled with the number of seconds on the left hand side. Times given are the average execution time (including Rakudo start-up and compiling the benchmark code) over ten runs, on my reasonably fast 64-bit Linux machine. The benchmark scripts are at https://github.com/perl6/bench-scripts, and are a fairly random bunch at the moment. Contributions of new scripts would be gladly accepted.
As for the meaning of the results, I haven’t had time to analyze them in any depth. latest-rakudo seems pretty consistently a bit slower than the latest Rakudo Star, which makes me wonder if there are some different flags set in the build process for each. (I used the default build process for both R* and latest-rakudo, but they are two different build methods and might have some different compiler flags.) It also looks like a couple of the benchmarks have backslid very badly in the last two months. Certainly a full investigation is warranted.