I am trying to see if I can get an adaquate avalanche distribution using a hash of popcount(src),(popcount(dst),srcport,dstport, protocol, seed.
popcount is, regrettably, an sse4.2-only instruction, and this version of the assembly routine can actually popcount up to 8 ipv6 addresses in a row, but you typically just pass it 2. I had a great deal of fun writing this tho I haven't got around to actually seeing how good a resulting hash would be! The ipv4 version does both src and dst with a single popcnt. The startup cost to further hash this is pretty insane - 30 cycles in http://burtleburtle.net/bob/hash/spooky.html. Xor? CRC? 00000000004005a0 <popcount2asm>: 4005a0: 31 c0 xor %eax,%eax 4005a2: 89 f1 mov %esi,%ecx 00000000004005a4 <poploop7>: 4005a4: 48 c1 e0 08 shl $0x8,%rax 4005a8: f3 4c 0f b8 07 popcnt (%rdi),%r8 4005ad: 48 83 c7 10 add $0x10,%rdi 4005b1: 4c 09 c0 or %r8,%rax 4005b4: f3 4c 0f b8 47 f8 popcnt -0x8(%rdi),%r8 4005ba: 4c 01 c0 add %r8,%rax 4005bd: e2 e5 loop 4005a4 <poploop7> 4005bf: c3 retq -- Dave Täht Let's go make home routers and wifi faster! With better software! http://blog.cerowrt.org _______________________________________________ Cake mailing list [email protected] https://lists.bufferbloat.net/listinfo/cake
