On Sun, Nov 04, 2012 at 10:27:48PM -0800, Ben Pfaff wrote:
> On Fri, Oct 26, 2012 at 11:36:06AM +0900, Simon Horman wrote:
> > Make use of __builtin_clz if available which should optimize
> > ofputil_version_bitmap_scanr() by replacing a loop with
> > a single CLZ instruction when available.
> >
> > I'm unsure if this approach is worth it or not.
> > But a similar approach could be taken to use ffs()
> > in bitmap_scan().
> >
> > Signed-off-by: Simon Horman <ho...@verge.net.au>
>
> I didn't notice this until now; I see now that I implemented something
> with the same purpose as:
>
> /* Returns the index of the rightmost 1-bit in 'x' (e.g. 01011000 => 6), or 32
> * if 'x' is 0.
> *
> * This function only works with 32-bit integers. */
> static inline uint32_t
> leftmost_1bit_idx(uint32_t x)
> {
> return x ? log_2_floor(x) : 32;
> }
Thanks, I will make use of that.
Do you have any thoughts on if it is worth implementing something
similar for bitmap_scan() using ffs()?
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
