Messages in this thread | | | From | "Alexander van Heukelum" <> | Subject | Re: Alternative implementation of the generic __ffs | Date | Sun, 20 Apr 2008 10:42:21 +0200 |
| |
On Sat, 19 Apr 2008 20:06:57 -0700, "Joe Perches" <joe@perches.com> said: > On Sun, 2008-04-20 at 01:29 +0300, Matti Aarnio wrote: > > I am curious, why not take the code already in glibc ffs() for ARM ? > > That is, if the ffs() is all that important detail in kernel ?
Hi,
The glibc version is based on a table-lookup. This makes it behave differently in hot and cold cache situations. That's fine if __ffs is used in tight loops, but in the kernel such use of __ffs is avoided because it might be slow. I added it to the benchmark, but it would need testing for the cold cache case too.
As for the importance of __ffs in the kernel: as far as I know the hot-spots in the kernel using __ffs are the schedular (sched_find_first_bit) and the cpu mask walking code (for_each_cpu_mask).
Greetings, Alexander
> Here's test results with the glibc ffs implementation. > (small const is still using slower add rather than or)
Added, thanks.
> $ gcc -Os -fomit-frame-pointer ffs.c > $ ./a.out > Original: 3155 tics, 8331 tics > New: 4211 tics, 8793 tics > Smallest: 4019 tics, 7754 tics > Small const: 3552 tics, 6308 tics > glibc: 2816 tics, 6911 tics > Empty loop: 1516 tics, 2244 tics > > $ gcc -O2 -fomit-frame-pointer ffs.c > $ ./a.out > Original: 3155 tics, 7828 tics > New: 4792 tics, 8825 tics > Smallest: 4401 tics, 7155 tics > Small const: 3539 tics, 5805 tics > glibc: 2720 tics, 7061 tics > Empty loop: 1516 tics, 2148 tics > > $ gcc -O3 -fomit-frame-pointer ffs.c > $ ./a.out > Original: 3080 tics, 7706 tics > New: 4721 tics, 8663 tics > Smallest: 4334 tics, 7116 tics > Small const: 3466 tics, 5672 tics > glibc: 2649 tics, 6939 tics > Empty loop: 1444 tics, 2012 tics > > -- Alexander van Heukelum heukelum@fastmail.fm
-- http://www.fastmail.fm - A no graphics, no pop-ups email service
| |