At Fri, 18 Jun 2010 10:05:57 -0500, Robby Findler wrote: > Perhaps it suggests a bug in the interaction between that fancy > require operator and the optimizer?
Actually, the benchmark on the shootout webpage is named mandelbrot-unsafe.rkt in the repository. The version I had in mind is mandelbrot.rkt, which does use fl+ without renaming it. After running mandelbrot-unsafe.rkt 50 times, it does show some variance in the running times (it runs for between 7 and 10 seconds, with times following what looks like a normal distribution), but much less than mandelbrot.rkt (between 19 and 40 seconds, with times following a bimodal distribution, with peaks around 20 and 40). The small variation for mandelbrot-unsafe.rkt (combined with the fact that times are distributed normally) indicates that it's probably noise. On the other hand, the bimodal distribution observed on mandelbrot.rkt is probably a sign that something weird is going on. The slowdown observed on the shootout webpage, since it's based on mandelbrot-unsafe.rkt, could thus be caused by the upgrade to 5.0. The running time inconsistencies due to fl+ and co is likely a separate issue. Vincent _________________________________________________ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
