On Thu, May 25, 2017 at 01:08:21PM -0700, Joe Stringer wrote:
> On 25 May 2017 at 10:36, Ben Pfaff <[email protected]> wrote:
> > On Tue, May 23, 2017 at 04:02:14PM -0700, Joe Stringer wrote:
> >> When running 256B hash check, we currently iterate from 0 up to and
> >> including bit 2048, which is beyond the range of bits that 256B holds.
> >> For bit 2048, set_bit128() doesn't set a bit due to the range check.
> >> Simplify the code by dropping the handling of bit 2048.
> >>
> >> Signed-off-by: Joe Stringer <[email protected]>
> >
> > Hmm, weird code.
> >
> > Looking at check_word_hash(), I think the goal here is to test that the
> > hash of all-bits-0 is different from the hash for any single bit being
> > set.  That does seem like a valuable check.  Do you think that there is
> > a better way to still accomplish that goal for the larger cases?
> 
> I think that the above is part of it, but it's also validating that
> for murmurhash operating on 64-bit chunks at a time, it doesn't make a
> difference whether the input data is 64-bit aligned or not.

Right; I just meant the reason why it originally went one-past-the-end.
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to