Hi all, Michael has been running microbenchmarks on the tuatara machine (a PowerPC OS/X) for quite a while now. In case you don't know them, they are at http://tuatara.cs.uni-duesseldorf.de/benchmark.html. Now I've plugged this information through a CGI script using gnuplot, and it gives this: http://wyvern.cs.uni-duesseldorf.de/~arigo/plots.html (lower is better).
In the last couple of days it was a useful tool to find out a few mistakes we made over the months - notaby, several innocent-looking checkins that disabled some useful optimization as a side-effect. As you can see on the right edge of the graphs, as of today - with these mistakes corrected - we are faster than ever on Richards and Pystone, and our custom GCs beat Boehm hand-down, too. A bientot, Armin _______________________________________________ [email protected] http://codespeak.net/mailman/listinfo/pypy-dev
