I see drastic speed up in raytracer test from ffconsultancy .
Running "low-level optimizations" test on SBCL-1.0.29 gives me:
6.094 seconds of real time
5.995090 seconds of total run time (5.933099 user, 0.061991 system)
[ Run times consist of 0.150 seconds GC time, and 5.846 seconds non-GC time. ]
13,372,554,118 processor cycles
460,231,520 bytes consed
And on latest 1.0.30:
4.090 seconds of real time
4.037387 seconds of total run time (3.990394 user, 0.046993 system)
[ Run times consist of 0.159 seconds GC time, and 3.879 seconds non-GC time. ]
8,973,724,122 processor cycles
462,330,704 bytes consed
1.0.30 crunches complex float-point arithmetic up to 33% better, and
it goes now almost on the same speed as in OCaml.
Thank you guys, who committed code to 1.0.30. Special thanks to Paul
and Nikodemus, you did a really great job!