On Fri, Oct 30, 2015 at 2:02 PM, Al Viro [off-list ref] wrote:

Your variant has 1:64 ratio; obviously better than now, but we can actually
do 1:bits-per-cacheline quite easily.

Ok, but in that case you end up needing a counter for each cacheline
too (to count how many bits, in order to know when to say "cacheline
is entirely full").

Because otherwise you'll end up having to check the whole cacheline
(rather than check just one word, or check the "bit counter") when you
set a bit.

So I suspect a "bitmap of full words" ends up being much simpler.  But
hey, if you can make your more complex thing work, go right ahead.

               Linus