>> Of course Ben's hint at unsafe-reprovide is something I'd like to >> investigate. > > My hint turned into a PR and merged a few years ago. > https://github.com/racket/typed-racket/pull/657 >
I read that one yesterday - it does not address the issue fully. I meant the hint at really implementing full (unsafe-reprovide ...) form. > Here's the output I see from `typed-performance.rkt` . I'm on a commit > from Jan 20. Does this look more like you'd expect? > > ``` > --- > typed: cpu time: 2990 real time: 3023 gc time: 65 > unsafe: cpu time: 405 real time: 408 gc time: 5 > typed, unsafe provided: cpu time: 692 real time: 697 gc time: 10 > plain: cpu time: 748 real time: 752 gc time: 3 > --- > typed: cpu time: 2817 real time: 2825 gc time: 11 > unsafe: cpu time: 414 real time: 417 gc time: 3 > typed, unsafe provided: cpu time: 671 real time: 675 gc time: 4 > plain: cpu time: 750 real time: 754 gc time: 5 > === > --- > typed: cpu time: 1678 real time: 1684 gc time: 0 > unsafe: cpu time: 161 real time: 165 gc time: 0 > typed, unsafe provided: cpu time: 175 real time: 178 gc time: 0 > plain: cpu time: 180 real time: 184 gc time: 0 > --- > typed: cpu time: 1744 real time: 1755 gc time: 0 > unsafe: cpu time: 165 real time: 168 gc time: 0 > typed, unsafe provided: cpu time: 164 real time: 167 gc time: 0 > plain: cpu time: 195 real time: 198 gc time: 0 > ``` > No, this is still "wrong". I experimented a bit with re-providing renamed / copied bindings as discussed in those two issues and the results are definitely much better. Almost perfect: ``` ===rgb-average --- typed: cpu time: 62 real time: 62 gc time: 10 typed-orig: cpu time: 1275 real time: 1275 gc time: 9 unsafe: cpu time: 50 real time: 50 gc time: 0 typed, unsafe provided: cpu time: 271 real time: 271 gc time: 0 plain: cpu time: 416 real time: 416 gc time: 0 --- typed: cpu time: 54 real time: 54 gc time: 1 typed-orig: cpu time: 1273 real time: 1273 gc time: 2 unsafe: cpu time: 49 real time: 49 gc time: 0 typed, unsafe provided: cpu time: 271 real time: 271 gc time: 0 plain: cpu time: 414 real time: 414 gc time: 0 ===rgb-distance^2 --- typed: cpu time: 672 real time: 672 gc time: 1 typed-orig: cpu time: 664 real time: 664 gc time: 1 unsafe: cpu time: 12 real time: 12 gc time: 0 typed, unsafe provided: cpu time: 13 real time: 13 gc time: 0 plain: cpu time: 13 real time: 13 gc time: 0 --- typed: cpu time: 676 real time: 676 gc time: 1 typed-orig: cpu time: 664 real time: 664 gc time: 1 unsafe: cpu time: 13 real time: 13 gc time: 0 typed, unsafe provided: cpu time: 13 real time: 13 gc time: 0 plain: cpu time: 13 real time: 13 gc time: 0 ``` Now the "typed" version of "rgb-average" is almost on par with usafe one. I re-provided only rgb-average, I left rgb-distance intact. Also I optimized it a bit (basically restructured the way the bindings are introduced in the loop) and that helped a lot. But now... Why is the re-provided through rename of copied binding (typed) 5 times faster than the one which is unsafe-provide'd (typed, unsafe provided) directly? I must be overlooking something. Cheers, Dominik -- You received this message because you are subscribed to the Google Groups "Racket Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/racket-users/427cc569-1929-d70c-3d0d-894de2b89176%40trustica.cz.
