Hallo Fred, vous ecrit au Mon, 28 Nov 2022 23:29:41 +0000:
> >Yes, it has improved a whole lot compared to fpc 3,2,2, ESPECIALLY on > >floating point operations. > > In my case, because using extendly float calculation in uos sound > library, it is a big plus. My tests don't use really heavy calulations, a sI wrote, just a simple prime sifter, performed very often. I.e., the control operations probabely outweigh the calculations by a big factor, drowning out the floating point performance noticably. > > With integer operations, it has slowed down > > appreciably, by about 30% (i.e. it takes nearly 50% LONGER). > > And this is a few sad ( and difficult to undertand, maybe change all > integer calculation into float and then round it, should be faster? ) Yes, this IS quite sad, and it's hard to imagine a mechanism. It might even suggest a noticable degradation of control function performance as its cause, which would be very bad, because this would compromise much more applications than could benefit from the better floating point performance. (And, interestingly, the llvm-based fpc 3.3.1 isn't much better either.) > Very interesting benchmark. Not so much, just a q&d performance check aimed at the special topic of numeric operations. Perhaps one should also test the control flow operations; perhaps a loop producing all permutations of a given string might work? -- (Weitergabe von Adressdaten, Telefonnummern u.ä. ohne Zustimmung nicht gestattet, ebenso Zusendung von Werbung oder ähnlichem) ----------------------------------------------------------- Mit freundlichen Grüßen, S. Schicktanz ----------------------------------------------------------- _______________________________________________ mseide-msegui-talk mailing list mseide-msegui-talk@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mseide-msegui-talk