1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
|
// SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0-only
/*
* Implement fast Fletcher4 with AVX2 instructions. (x86_64)
*
* Use the 256-bit AVX2 SIMD instructions and registers to compute
* Fletcher4 in four incremental 64-bit parallel accumulator streams,
* and then combine the streams to form the final four checksum words.
*
* Copyright (C) 2015 Intel Corporation.
*
* Authors:
* James Guilford <james.guilford@intel.com>
* Jinshan Xiong <jinshan.xiong@intel.com>
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
* General Public License (GPL) Version 2, available from the file
* COPYING in the main directory of this source tree, or the
* OpenIB.org BSD license below:
*
* Redistribution and use in source and binary forms, with or
* without modification, are permitted provided that the following
* conditions are met:
*
* - Redistributions of source code must retain the above
* copyright notice, this list of conditions and the following
* disclaimer.
*
* - Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials
* provided with the distribution.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#if defined(HAVE_AVX) && defined(HAVE_AVX2)
#include <sys/spa_checksum.h>
#include <sys/string.h>
#include <sys/simd.h>
#include <zfs_fletcher.h>
static void
fletcher_4_avx2_init(fletcher_4_ctx_t *ctx)
{
memset(ctx->avx, 0, 4 * sizeof (zfs_fletcher_avx_t));
}
static void
fletcher_4_avx2_fini(fletcher_4_ctx_t *ctx, zio_cksum_t *zcp)
{
uint64_t A, B, C, D;
A = ctx->avx[0].v[0] + ctx->avx[0].v[1] +
ctx->avx[0].v[2] + ctx->avx[0].v[3];
B = 0 - ctx->avx[0].v[1] - 2 * ctx->avx[0].v[2] - 3 * ctx->avx[0].v[3] +
4 * ctx->avx[1].v[0] + 4 * ctx->avx[1].v[1] + 4 * ctx->avx[1].v[2] +
4 * ctx->avx[1].v[3];
C = ctx->avx[0].v[2] + 3 * ctx->avx[0].v[3] - 6 * ctx->avx[1].v[0] -
10 * ctx->avx[1].v[1] - 14 * ctx->avx[1].v[2] -
18 * ctx->avx[1].v[3] + 16 * ctx->avx[2].v[0] +
16 * ctx->avx[2].v[1] + 16 * ctx->avx[2].v[2] +
16 * ctx->avx[2].v[3];
D = 0 - ctx->avx[0].v[3] + 4 * ctx->avx[1].v[0] +
10 * ctx->avx[1].v[1] + 20 * ctx->avx[1].v[2] +
34 * ctx->avx[1].v[3] - 48 * ctx->avx[2].v[0] -
64 * ctx->avx[2].v[1] - 80 * ctx->avx[2].v[2] -
96 * ctx->avx[2].v[3] + 64 * ctx->avx[3].v[0] +
64 * ctx->avx[3].v[1] + 64 * ctx->avx[3].v[2] +
64 * ctx->avx[3].v[3];
ZIO_SET_CHECKSUM(zcp, A, B, C, D);
}
#define FLETCHER_4_AVX2_RESTORE_CTX(ctx) \
{ \
asm volatile("vmovdqu %0, %%ymm0" :: "m" ((ctx)->avx[0])); \
asm volatile("vmovdqu %0, %%ymm1" :: "m" ((ctx)->avx[1])); \
asm volatile("vmovdqu %0, %%ymm2" :: "m" ((ctx)->avx[2])); \
asm volatile("vmovdqu %0, %%ymm3" :: "m" ((ctx)->avx[3])); \
}
#define FLETCHER_4_AVX2_SAVE_CTX(ctx) \
{ \
asm volatile("vmovdqu %%ymm0, %0" : "=m" ((ctx)->avx[0])); \
asm volatile("vmovdqu %%ymm1, %0" : "=m" ((ctx)->avx[1])); \
asm volatile("vmovdqu %%ymm2, %0" : "=m" ((ctx)->avx[2])); \
asm volatile("vmovdqu %%ymm3, %0" : "=m" ((ctx)->avx[3])); \
}
static void
fletcher_4_avx2_native(fletcher_4_ctx_t *ctx, const void *buf, uint64_t size)
{
const uint64_t *ip = buf;
const uint64_t *ipend = (uint64_t *)((uint8_t *)ip + size);
FLETCHER_4_AVX2_RESTORE_CTX(ctx);
do {
asm volatile("vpmovzxdq %0, %%ymm4"::"m" (*ip));
asm volatile("vpaddq %ymm4, %ymm0, %ymm0");
asm volatile("vpaddq %ymm0, %ymm1, %ymm1");
asm volatile("vpaddq %ymm1, %ymm2, %ymm2");
asm volatile("vpaddq %ymm2, %ymm3, %ymm3");
} while ((ip += 2) < ipend);
FLETCHER_4_AVX2_SAVE_CTX(ctx);
asm volatile("vzeroupper");
}
static void
fletcher_4_avx2_byteswap(fletcher_4_ctx_t *ctx, const void *buf, uint64_t size)
{
static const zfs_fletcher_avx_t mask = {
.v = { 0xFFFFFFFF00010203, 0xFFFFFFFF08090A0B,
0xFFFFFFFF00010203, 0xFFFFFFFF08090A0B }
};
const uint64_t *ip = buf;
const uint64_t *ipend = (uint64_t *)((uint8_t *)ip + size);
FLETCHER_4_AVX2_RESTORE_CTX(ctx);
asm volatile("vmovdqu %0, %%ymm5" :: "m" (mask));
do {
asm volatile("vpmovzxdq %0, %%ymm4"::"m" (*ip));
asm volatile("vpshufb %ymm5, %ymm4, %ymm4");
asm volatile("vpaddq %ymm4, %ymm0, %ymm0");
asm volatile("vpaddq %ymm0, %ymm1, %ymm1");
asm volatile("vpaddq %ymm1, %ymm2, %ymm2");
asm volatile("vpaddq %ymm2, %ymm3, %ymm3");
} while ((ip += 2) < ipend);
FLETCHER_4_AVX2_SAVE_CTX(ctx);
asm volatile("vzeroupper");
}
static boolean_t fletcher_4_avx2_valid(void)
{
return (kfpu_allowed() && zfs_avx_available() && zfs_avx2_available());
}
const fletcher_4_ops_t fletcher_4_avx2_ops = {
.init_native = fletcher_4_avx2_init,
.fini_native = fletcher_4_avx2_fini,
.compute_native = fletcher_4_avx2_native,
.init_byteswap = fletcher_4_avx2_init,
.fini_byteswap = fletcher_4_avx2_fini,
.compute_byteswap = fletcher_4_avx2_byteswap,
.valid = fletcher_4_avx2_valid,
.uses_fpu = B_TRUE,
.name = "avx2"
};
#endif /* defined(HAVE_AVX) && defined(HAVE_AVX2) */
|
