2009-07-08 13:19:16 -07:00
|
|
|
/*
|
2014-08-12 11:12:12 +12:00
|
|
|
* Copyright (c) 2008, 2009, 2010, 2012, 2013, 2014 Nicira, Inc.
|
2009-07-08 13:19:16 -07:00
|
|
|
*
|
2009-06-15 15:11:30 -07:00
|
|
|
* 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:
|
2009-07-08 13:19:16 -07:00
|
|
|
*
|
2009-06-15 15:11:30 -07:00
|
|
|
* 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.
|
2009-07-08 13:19:16 -07:00
|
|
|
*/
|
|
|
|
#include <config.h>
|
|
|
|
#include "hash.h"
|
|
|
|
#include <string.h>
|
2012-04-09 14:33:33 -07:00
|
|
|
#include "unaligned.h"
|
2009-07-08 13:19:16 -07:00
|
|
|
|
2010-12-16 13:53:29 -08:00
|
|
|
/* Returns the hash of 'a', 'b', and 'c'. */
|
2009-10-14 17:03:55 -07:00
|
|
|
uint32_t
|
2010-12-16 13:53:29 -08:00
|
|
|
hash_3words(uint32_t a, uint32_t b, uint32_t c)
|
2009-10-14 17:03:55 -07:00
|
|
|
{
|
2014-07-04 07:57:18 -07:00
|
|
|
return hash_finish(hash_add(hash_add(hash_add(a, 0), b), c), 12);
|
2010-12-16 13:53:29 -08:00
|
|
|
}
|
|
|
|
|
2009-07-08 13:19:16 -07:00
|
|
|
/* Returns the hash of the 'n' bytes at 'p', starting from 'basis'. */
|
|
|
|
uint32_t
|
|
|
|
hash_bytes(const void *p_, size_t n, uint32_t basis)
|
|
|
|
{
|
2024-05-03 07:44:13 +02:00
|
|
|
const uint8_t *p = p_;
|
hash: Replace primary hash functions by murmurhash.
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>
2013-01-16 16:14:42 -08:00
|
|
|
size_t orig_n = n;
|
|
|
|
uint32_t hash;
|
2009-07-08 13:19:16 -07:00
|
|
|
|
hash: Replace primary hash functions by murmurhash.
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>
2013-01-16 16:14:42 -08:00
|
|
|
hash = basis;
|
|
|
|
while (n >= 4) {
|
2024-05-03 07:44:13 +02:00
|
|
|
hash = hash_add(hash,
|
|
|
|
get_unaligned_u32(ALIGNED_CAST(const uint32_t *, p)));
|
hash: Replace primary hash functions by murmurhash.
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>
2013-01-16 16:14:42 -08:00
|
|
|
n -= 4;
|
2024-05-03 07:44:13 +02:00
|
|
|
p += 4;
|
2009-07-08 13:19:16 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
if (n) {
|
hash: Replace primary hash functions by murmurhash.
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>
2013-01-16 16:14:42 -08:00
|
|
|
uint32_t tmp = 0;
|
2012-04-09 14:33:33 -07:00
|
|
|
|
hash: Replace primary hash functions by murmurhash.
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>
2013-01-16 16:14:42 -08:00
|
|
|
memcpy(&tmp, p, n);
|
2014-07-04 07:57:18 -07:00
|
|
|
hash = hash_add(hash, tmp);
|
2009-07-08 13:19:16 -07:00
|
|
|
}
|
|
|
|
|
2014-07-04 07:57:18 -07:00
|
|
|
return hash_finish(hash, orig_n);
|
2009-07-08 13:19:16 -07:00
|
|
|
}
|
2012-08-21 14:26:23 -07:00
|
|
|
|
hash: Replace primary hash functions by murmurhash.
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>
2013-01-16 16:14:42 -08:00
|
|
|
uint32_t
|
|
|
|
hash_double(double x, uint32_t basis)
|
|
|
|
{
|
|
|
|
uint32_t value[2];
|
|
|
|
BUILD_ASSERT_DECL(sizeof x == sizeof value);
|
|
|
|
|
|
|
|
memcpy(value, &x, sizeof value);
|
|
|
|
return hash_3words(value[0], value[1], basis);
|
|
|
|
}
|
2014-07-11 05:57:11 -07:00
|
|
|
|
|
|
|
uint32_t
|
hash: Fix compilation error on Fedora 34 with GCC 11 and -O0.
With newest version of GCC, OVS fails to compile with -O0 due to
false-positive overread detection in hash-related functions. Those
function declares "const uint32_t p[]" as argument and it throws off
the compiler into thinking that it reads from memory region of size 0.
To fix that behavior, a change in argument declaration needs to be made:
instead of using "[]" notation for a pointer, simply use "*".
The reported error in question:
lib/conntrack.c: In function ‘conn_key_hash’:
lib/conntrack.c:2154:12: error: ‘hash_words’ reading 4 bytes \
from a region of size 0 [-Werror=stringop-overread]
2154| return hash_words((uint32_t *) (&key->dst + 1),
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2155| (uint32_t *) (key + 1) - (uint32_t *) (&key->dst + 1),
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2156| hash);
| ~~~~~
lib/conntrack.c:2154:12: note: referencing argument 1 \
of type ‘const uint32_t *’ {aka ‘const unsigned int *’}
In file included from lib/packets.h:31,
from lib/ct-dpif.h:21,
from lib/conntrack.h:23,
from lib/conntrack.c:26:
lib/hash.h:294:1: note: in a call to function ‘hash_words’
294 | hash_words(const uint32_t p[], size_t n_words, uint32_t basis)
| ^~~~~~~~~~
Signed-off-by: Guzowski Adrian <adrian.guzowski@exatel.pl>
Tested-by: Numan Siddique <numans@ovn.org>
Signed-off-by: Ilya Maximets <i.maximets@ovn.org>
2021-07-28 12:03:29 +00:00
|
|
|
hash_words__(const uint32_t *p, size_t n_words, uint32_t basis)
|
2014-07-11 05:57:11 -07:00
|
|
|
{
|
|
|
|
return hash_words_inline(p, n_words, basis);
|
|
|
|
}
|
|
|
|
|
|
|
|
uint32_t
|
hash: Fix compilation error on Fedora 34 with GCC 11 and -O0.
With newest version of GCC, OVS fails to compile with -O0 due to
false-positive overread detection in hash-related functions. Those
function declares "const uint32_t p[]" as argument and it throws off
the compiler into thinking that it reads from memory region of size 0.
To fix that behavior, a change in argument declaration needs to be made:
instead of using "[]" notation for a pointer, simply use "*".
The reported error in question:
lib/conntrack.c: In function ‘conn_key_hash’:
lib/conntrack.c:2154:12: error: ‘hash_words’ reading 4 bytes \
from a region of size 0 [-Werror=stringop-overread]
2154| return hash_words((uint32_t *) (&key->dst + 1),
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2155| (uint32_t *) (key + 1) - (uint32_t *) (&key->dst + 1),
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2156| hash);
| ~~~~~
lib/conntrack.c:2154:12: note: referencing argument 1 \
of type ‘const uint32_t *’ {aka ‘const unsigned int *’}
In file included from lib/packets.h:31,
from lib/ct-dpif.h:21,
from lib/conntrack.h:23,
from lib/conntrack.c:26:
lib/hash.h:294:1: note: in a call to function ‘hash_words’
294 | hash_words(const uint32_t p[], size_t n_words, uint32_t basis)
| ^~~~~~~~~~
Signed-off-by: Guzowski Adrian <adrian.guzowski@exatel.pl>
Tested-by: Numan Siddique <numans@ovn.org>
Signed-off-by: Ilya Maximets <i.maximets@ovn.org>
2021-07-28 12:03:29 +00:00
|
|
|
hash_words64__(const uint64_t *p, size_t n_words, uint32_t basis)
|
2014-07-11 05:57:11 -07:00
|
|
|
{
|
|
|
|
return hash_words64_inline(p, n_words, basis);
|
|
|
|
}
|
2014-08-12 11:12:12 +12:00
|
|
|
|
2019-02-27 17:44:06 +08:00
|
|
|
#if !(defined(__x86_64__)) && !(defined(__aarch64__))
|
2014-08-12 11:12:12 +12:00
|
|
|
void
|
|
|
|
hash_bytes128(const void *p_, size_t len, uint32_t basis, ovs_u128 *out)
|
|
|
|
{
|
|
|
|
const uint32_t c1 = 0x239b961b;
|
|
|
|
const uint32_t c2 = 0xab0e9789;
|
|
|
|
const uint32_t c3 = 0x38b34ae5;
|
|
|
|
const uint32_t c4 = 0xa1e38b93;
|
|
|
|
const uint8_t *tail, *data = (const uint8_t *)p_;
|
|
|
|
const uint32_t *blocks = (const uint32_t *)p_;
|
|
|
|
const int nblocks = len / 16;
|
|
|
|
uint32_t h1 = basis;
|
|
|
|
uint32_t h2 = basis;
|
|
|
|
uint32_t h3 = basis;
|
|
|
|
uint32_t h4 = basis;
|
|
|
|
|
|
|
|
/* Body */
|
|
|
|
for (int i = 0; i < nblocks; i++) {
|
|
|
|
uint32_t k1 = get_unaligned_u32(&blocks[i * 4 + 0]);
|
|
|
|
uint32_t k2 = get_unaligned_u32(&blocks[i * 4 + 1]);
|
|
|
|
uint32_t k3 = get_unaligned_u32(&blocks[i * 4 + 2]);
|
|
|
|
uint32_t k4 = get_unaligned_u32(&blocks[i * 4 + 3]);
|
|
|
|
|
|
|
|
k1 *= c1;
|
|
|
|
k1 = hash_rot(k1, 15);
|
|
|
|
k1 *= c2;
|
|
|
|
h1 ^= k1;
|
|
|
|
|
|
|
|
h1 = hash_rot(h1, 19);
|
|
|
|
h1 += h2;
|
|
|
|
h1 = h1 * 5 + 0x561ccd1b;
|
|
|
|
|
|
|
|
k2 *= c2;
|
|
|
|
k2 = hash_rot(k2, 16);
|
|
|
|
k2 *= c3;
|
|
|
|
h2 ^= k2;
|
|
|
|
|
|
|
|
h2 = hash_rot(h2, 17);
|
|
|
|
h2 += h3;
|
|
|
|
h2 = h2 * 5 + 0x0bcaa747;
|
|
|
|
|
|
|
|
k3 *= c3;
|
|
|
|
k3 = hash_rot(k3, 17);
|
|
|
|
k3 *= c4;
|
|
|
|
h3 ^= k3;
|
|
|
|
|
|
|
|
h3 = hash_rot(h3, 15);
|
|
|
|
h3 += h4;
|
|
|
|
h3 = h3 * 5 + 0x96cd1c35;
|
|
|
|
|
|
|
|
k4 *= c4;
|
|
|
|
k4 = hash_rot(k4, 18);
|
|
|
|
k4 *= c1;
|
|
|
|
h4 ^= k4;
|
|
|
|
|
|
|
|
h4 = hash_rot(h4, 13);
|
|
|
|
h4 += h1;
|
|
|
|
h4 = h4 * 5 + 0x32ac3b17;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Tail */
|
2017-08-02 15:03:06 -07:00
|
|
|
uint32_t k1, k2, k3, k4;
|
2014-08-12 11:12:12 +12:00
|
|
|
k1 = k2 = k3 = k4 = 0;
|
|
|
|
tail = data + nblocks * 16;
|
|
|
|
switch (len & 15) {
|
|
|
|
case 15:
|
|
|
|
k4 ^= tail[14] << 16;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 14:
|
|
|
|
k4 ^= tail[13] << 8;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 13:
|
|
|
|
k4 ^= tail[12] << 0;
|
|
|
|
k4 *= c4;
|
|
|
|
k4 = hash_rot(k4, 18);
|
|
|
|
k4 *= c1;
|
|
|
|
h4 ^= k4;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
|
|
|
|
case 12:
|
|
|
|
k3 ^= tail[11] << 24;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 11:
|
|
|
|
k3 ^= tail[10] << 16;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 10:
|
|
|
|
k3 ^= tail[9] << 8;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 9:
|
|
|
|
k3 ^= tail[8] << 0;
|
|
|
|
k3 *= c3;
|
|
|
|
k3 = hash_rot(k3, 17);
|
|
|
|
k3 *= c4;
|
|
|
|
h3 ^= k3;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
|
|
|
|
case 8:
|
|
|
|
k2 ^= tail[7] << 24;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 7:
|
|
|
|
k2 ^= tail[6] << 16;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 6:
|
|
|
|
k2 ^= tail[5] << 8;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 5:
|
|
|
|
k2 ^= tail[4] << 0;
|
|
|
|
k2 *= c2;
|
|
|
|
k2 = hash_rot(k2, 16);
|
|
|
|
k2 *= c3;
|
|
|
|
h2 ^= k2;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
|
|
|
|
case 4:
|
|
|
|
k1 ^= tail[3] << 24;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 3:
|
|
|
|
k1 ^= tail[2] << 16;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 2:
|
|
|
|
k1 ^= tail[1] << 8;
|
2017-07-13 10:15:42 -07:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 1:
|
|
|
|
k1 ^= tail[0] << 0;
|
|
|
|
k1 *= c1;
|
|
|
|
k1 = hash_rot(k1, 15);
|
|
|
|
k1 *= c2;
|
|
|
|
h1 ^= k1;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* Finalization */
|
|
|
|
h1 ^= len;
|
|
|
|
h2 ^= len;
|
|
|
|
h3 ^= len;
|
|
|
|
h4 ^= len;
|
|
|
|
|
|
|
|
h1 += h2;
|
|
|
|
h1 += h3;
|
|
|
|
h1 += h4;
|
|
|
|
h2 += h1;
|
|
|
|
h3 += h1;
|
|
|
|
h4 += h1;
|
|
|
|
|
|
|
|
h1 = mhash_finish(h1);
|
|
|
|
h2 = mhash_finish(h2);
|
|
|
|
h3 = mhash_finish(h3);
|
|
|
|
h4 = mhash_finish(h4);
|
|
|
|
|
|
|
|
h1 += h2;
|
|
|
|
h1 += h3;
|
|
|
|
h1 += h4;
|
|
|
|
h2 += h1;
|
|
|
|
h3 += h1;
|
|
|
|
h4 += h1;
|
|
|
|
|
|
|
|
out->u32[0] = h1;
|
|
|
|
out->u32[1] = h2;
|
|
|
|
out->u32[2] = h3;
|
|
|
|
out->u32[3] = h4;
|
|
|
|
}
|
|
|
|
|
2019-02-27 17:44:06 +08:00
|
|
|
#else /* __x86_64__ or __aarch64__*/
|
2014-08-12 11:12:12 +12:00
|
|
|
|
|
|
|
static inline uint64_t
|
|
|
|
hash_rot64(uint64_t x, int8_t r)
|
|
|
|
{
|
|
|
|
return (x << r) | (x >> (64 - r));
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline uint64_t
|
|
|
|
fmix64(uint64_t k)
|
|
|
|
{
|
|
|
|
k ^= k >> 33;
|
|
|
|
k *= 0xff51afd7ed558ccdULL;
|
|
|
|
k ^= k >> 33;
|
|
|
|
k *= 0xc4ceb9fe1a85ec53ULL;
|
|
|
|
k ^= k >> 33;
|
|
|
|
|
|
|
|
return k;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
hash_bytes128(const void *p_, size_t len, uint32_t basis, ovs_u128 *out)
|
|
|
|
{
|
|
|
|
const uint64_t c1 = 0x87c37b91114253d5ULL;
|
|
|
|
const uint64_t c2 = 0x4cf5ad432745937fULL;
|
|
|
|
const uint8_t *tail, *data = (const uint8_t *)p_;
|
|
|
|
const uint64_t *blocks = (const uint64_t *)p_;
|
|
|
|
const int nblocks = len / 16;
|
|
|
|
uint64_t h1 = basis;
|
|
|
|
uint64_t h2 = basis;
|
|
|
|
uint64_t k1, k2;
|
|
|
|
|
|
|
|
/* Body */
|
|
|
|
for (int i = 0; i < nblocks; i++) {
|
|
|
|
k1 = get_unaligned_u64(&blocks[i * 2 + 0]);
|
|
|
|
k2 = get_unaligned_u64(&blocks[i * 2 + 1]);
|
|
|
|
|
|
|
|
k1 *= c1;
|
|
|
|
k1 = hash_rot64(k1, 31);
|
|
|
|
k1 *= c2;
|
|
|
|
h1 ^= k1;
|
|
|
|
|
|
|
|
h1 = hash_rot64(h1, 27);
|
|
|
|
h1 += h2;
|
|
|
|
h1 = h1 * 5 + 0x52dce729;
|
|
|
|
|
|
|
|
k2 *= c2;
|
|
|
|
k2 = hash_rot64(k2, 33);
|
|
|
|
k2 *= c1;
|
|
|
|
h2 ^= k2;
|
|
|
|
|
|
|
|
h2 = hash_rot64(h2, 31);
|
|
|
|
h2 += h1;
|
|
|
|
h2 = h2 * 5 + 0x38495ab5;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Tail */
|
|
|
|
k1 = 0;
|
|
|
|
k2 = 0;
|
|
|
|
tail = data + nblocks * 16;
|
|
|
|
switch (len & 15) {
|
|
|
|
case 15:
|
|
|
|
k2 ^= ((uint64_t) tail[14]) << 48;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 14:
|
|
|
|
k2 ^= ((uint64_t) tail[13]) << 40;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 13:
|
|
|
|
k2 ^= ((uint64_t) tail[12]) << 32;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 12:
|
|
|
|
k2 ^= ((uint64_t) tail[11]) << 24;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 11:
|
|
|
|
k2 ^= ((uint64_t) tail[10]) << 16;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 10:
|
|
|
|
k2 ^= ((uint64_t) tail[9]) << 8;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 9:
|
|
|
|
k2 ^= ((uint64_t) tail[8]) << 0;
|
|
|
|
k2 *= c2;
|
|
|
|
k2 = hash_rot64(k2, 33);
|
|
|
|
k2 *= c1;
|
|
|
|
h2 ^= k2;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 8:
|
|
|
|
k1 ^= ((uint64_t) tail[7]) << 56;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 7:
|
|
|
|
k1 ^= ((uint64_t) tail[6]) << 48;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 6:
|
|
|
|
k1 ^= ((uint64_t) tail[5]) << 40;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 5:
|
|
|
|
k1 ^= ((uint64_t) tail[4]) << 32;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 4:
|
|
|
|
k1 ^= ((uint64_t) tail[3]) << 24;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 3:
|
|
|
|
k1 ^= ((uint64_t) tail[2]) << 16;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 2:
|
|
|
|
k1 ^= ((uint64_t) tail[1]) << 8;
|
2017-06-23 18:12:49 +02:00
|
|
|
/* fall through */
|
2014-08-12 11:12:12 +12:00
|
|
|
case 1:
|
|
|
|
k1 ^= ((uint64_t) tail[0]) << 0;
|
|
|
|
k1 *= c1;
|
|
|
|
k1 = hash_rot64(k1, 31);
|
|
|
|
k1 *= c2;
|
|
|
|
h1 ^= k1;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* Finalization */
|
|
|
|
h1 ^= len;
|
|
|
|
h2 ^= len;
|
|
|
|
h1 += h2;
|
|
|
|
h2 += h1;
|
|
|
|
h1 = fmix64(h1);
|
|
|
|
h2 = fmix64(h2);
|
|
|
|
h1 += h2;
|
|
|
|
h2 += h1;
|
|
|
|
|
|
|
|
out->u64.lo = h1;
|
|
|
|
out->u64.hi = h2;
|
|
|
|
}
|
2019-02-27 17:44:06 +08:00
|
|
|
#endif /* __x86_64__ or __aarch64__*/
|