On Tue, Feb 18, 2014 at 9:37 AM, Sven Panne <[email protected]> wrote:
> On Tue, Feb 18, 2014 at 9:08 AM, Jakob Kummerow <[email protected]>wrote: > >> On Tue, Feb 18, 2014 at 8:28 AM, <[email protected]> wrote: >> >>> Math.imul, Math.clz32, ... What will be next? :) >> >> >> Math.f2xm1(), I think. >> > > :-P * * * > > >> >> AFAICT "count leading zeroes" has HW support for all >>> our platforms and boils down to a single CPU instruction, >> >> >> It would be interesting to see how fast that single instruction is. If >> it's implemented as a bit test loop in microcode, then rolling our own >> (based on binary search + lookup table or something?) might be faster :-) >> Then again, as long as there's no common use case where clz32 is the >> bottleneck, this is of purely academic interest. >> > > IIRC there was an article about using compiler intrinsics in one of the > 2013 issues of either "c't" or "Linux Magazin", don't remember, but clz > seems to be crucial for some real world programs and there was some > significant speedup for some CPU models (<= that much I can remember ;-). > That is motivation for clz - apparently Mozilla did some tests somewhere [citation needed] -- -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev --- You received this message because you are subscribed to the Google Groups "v8-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/groups/opt_out.
