On Tue, Sep 30, 2014 at 11:30:38AM +1300, Joe Stringer wrote: > On 30 September 2014 09:30, Ben Pfaff <b...@nicira.com> wrote: > > > On Fri, Sep 26, 2014 at 09:28:13PM +1200, Joe Stringer wrote: > > > Add the 128-bit murmurhash by Austin Appleby, r150 from: > > > http://code.google.com/p/smhasher/source/browse/trunk/MurmurHash3.cpp > > > > > > Signed-off-by: Joe Stringer <joestrin...@nicira.com> > > > --- > > > The x64 version is approximately 2x faster than x86 in the tests I ran. > > > > What makes these x86 and x64 versions? Aren't they just 32-bit word > > and 64-bit word versions? > > Yes. Apparently I confused myself a bit with looking at different hash > functions, but this is how they are referred to in the original source as > well. I don't have any information on whether there is a performance > difference between them on a 32-bit machine, I only tested on a 64-bit > machine.
OK. I did run this on my 32-bit machine (I actually use a 32-bit distro because everyone else uses 64-bit and I want to make sure there's coverage) and the test case passed. > > The comment on getblock32() doesn't seem to match what it actually > > does; it's just an array access. > > > > My intention was to point out that I haven't tried to address this issue at > all. I could drop it or place "XXX" at the start of the comment. Not > familiar with which platforms this is necessary for. If the hash function is intended to hash misaligned data, then getblock32() should call get_unaligned_u32() from unaligned.h, and similarly for getblock64(). If the hash function isn't intended to hash misaligned data, then the current implementation is fine. > > Acked-by: Ben Pfaff <b...@nicira.com> > > Thanks for review, will hold on for review of final patches in series. Thanks. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
