mirror of
https://github.com/openvswitch/ovs
synced 2025-10-19 14:37:21 +00:00
c49d1dd168
murmurhash is faster than Jenkins and slightly higher quality, so switch to it for hashing words. The best timings I got for hashing for data lengths of the following numbers of 32-bit words, in seconds per 1,000,000,000 hashes, were: words murmurhash Jenkins hash ----- ---------- ------------ 1 8.4 10.4 2 10.3 10.3 3 11.2 10.7 4 12.6 18.0 5 13.9 18.3 6 15.2 18.7 In other words, murmurhash outperforms Jenkins for all input lengths other than exactly 3 32-bit words (12 bytes). (It's understandable that Jenkins would have a best case at 12 bytes, because Jenkins works in 12-byte chunks.) Even in the case where Jenkins is faster, it's only by 5%. On average within this data set, murmurhash is 15% faster, and for 4-word input it is 30% faster. We retain Jenkins for flow_hash_symmetric_l4() and flow_hash_fields(), which are cases where the hash value is exposed externally. This commit appears to improve "ovs-benchmark rate" results slightly by a few hundred connections per second (under 1%), when used with an NVP controller. Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Ethan Jackson <ethan@nicira.com>
126 lines
3.3 KiB
C
126 lines
3.3 KiB
C
/*
|
|
* Copyright (c) 2008, 2009, 2010, 2012 Nicira, Inc.
|
|
*
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at:
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
*/
|
|
|
|
#include <config.h>
|
|
#include "jhash.h"
|
|
#include <string.h>
|
|
#include "unaligned.h"
|
|
|
|
/* This is the public domain lookup3 hash by Bob Jenkins from
|
|
* http://burtleburtle.net/bob/c/lookup3.c, modified for style. */
|
|
|
|
static inline uint32_t
|
|
jhash_rot(uint32_t x, int k)
|
|
{
|
|
return (x << k) | (x >> (32 - k));
|
|
}
|
|
|
|
static inline void
|
|
jhash_mix(uint32_t *a, uint32_t *b, uint32_t *c)
|
|
{
|
|
*a -= *c; *a ^= jhash_rot(*c, 4); *c += *b;
|
|
*b -= *a; *b ^= jhash_rot(*a, 6); *a += *c;
|
|
*c -= *b; *c ^= jhash_rot(*b, 8); *b += *a;
|
|
*a -= *c; *a ^= jhash_rot(*c, 16); *c += *b;
|
|
*b -= *a; *b ^= jhash_rot(*a, 19); *a += *c;
|
|
*c -= *b; *c ^= jhash_rot(*b, 4); *b += *a;
|
|
}
|
|
|
|
static inline void
|
|
jhash_final(uint32_t *a, uint32_t *b, uint32_t *c)
|
|
{
|
|
*c ^= *b; *c -= jhash_rot(*b, 14);
|
|
*a ^= *c; *a -= jhash_rot(*c, 11);
|
|
*b ^= *a; *b -= jhash_rot(*a, 25);
|
|
*c ^= *b; *c -= jhash_rot(*b, 16);
|
|
*a ^= *c; *a -= jhash_rot(*c, 4);
|
|
*b ^= *a; *b -= jhash_rot(*a, 14);
|
|
*c ^= *b; *c -= jhash_rot(*b, 24);
|
|
}
|
|
|
|
/* Returns the Jenkins hash of the 'n' 32-bit words at 'p', starting from
|
|
* 'basis'. 'p' must be properly aligned.
|
|
*
|
|
* Use hash_words() instead, unless you're computing a hash function whose
|
|
* value is exposed "on the wire" so we don't want to change it. */
|
|
uint32_t
|
|
jhash_words(const uint32_t *p, size_t n, uint32_t basis)
|
|
{
|
|
uint32_t a, b, c;
|
|
|
|
a = b = c = 0xdeadbeef + (((uint32_t) n) << 2) + basis;
|
|
|
|
while (n > 3) {
|
|
a += p[0];
|
|
b += p[1];
|
|
c += p[2];
|
|
jhash_mix(&a, &b, &c);
|
|
n -= 3;
|
|
p += 3;
|
|
}
|
|
|
|
switch (n) {
|
|
case 3:
|
|
c += p[2];
|
|
/* fall through */
|
|
case 2:
|
|
b += p[1];
|
|
/* fall through */
|
|
case 1:
|
|
a += p[0];
|
|
jhash_final(&a, &b, &c);
|
|
/* fall through */
|
|
case 0:
|
|
break;
|
|
}
|
|
return c;
|
|
}
|
|
|
|
/* Returns the Jenkins hash of the 'n' bytes at 'p', starting from 'basis'.
|
|
*
|
|
* Use jhash_bytes() instead, unless you're computing a hash function whose
|
|
* value is exposed "on the wire" so we don't want to change it. */
|
|
uint32_t
|
|
jhash_bytes(const void *p_, size_t n, uint32_t basis)
|
|
{
|
|
const uint8_t *p = p_;
|
|
uint32_t a, b, c;
|
|
|
|
a = b = c = 0xdeadbeef + n + basis;
|
|
|
|
while (n >= 12) {
|
|
a += get_unaligned_u32((uint32_t *) p);
|
|
b += get_unaligned_u32((uint32_t *) (p + 4));
|
|
c += get_unaligned_u32((uint32_t *) (p + 8));
|
|
jhash_mix(&a, &b, &c);
|
|
n -= 12;
|
|
p += 12;
|
|
}
|
|
|
|
if (n) {
|
|
uint32_t tmp[3];
|
|
|
|
tmp[0] = tmp[1] = tmp[2] = 0;
|
|
memcpy(tmp, p, n);
|
|
a += tmp[0];
|
|
b += tmp[1];
|
|
c += tmp[2];
|
|
jhash_final(&a, &b, &c);
|
|
}
|
|
|
|
return c;
|
|
}
|