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
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
|
#include <my_global.h>
#include <string.h>
#if defined(__GNUC__) && defined(__linux__) && defined(HAVE_ARMV8_CRC)
#include <sys/auxv.h>
#include <asm/hwcap.h>
#ifndef HWCAP_CRC32
#define HWCAP_CRC32 (1 << 7)
#endif
unsigned int crc32c_aarch64_available(void)
{
unsigned long auxv = getauxval(AT_HWCAP);
return (auxv & HWCAP_CRC32) != 0;
}
#endif
#ifndef HAVE_ARMV8_CRC_CRYPTO_INTRINSICS
/* Request crc extension capabilities from the assembler */
asm(".arch_extension crc");
#ifdef HAVE_ARMV8_CRYPTO
/* crypto extension */
asm(".arch_extension crypto");
#endif
#define CRC32CX(crc, value) __asm__("crc32cx %w[c], %w[c], %x[v]":[c]"+r"(crc):[v]"r"(value))
#define CRC32CW(crc, value) __asm__("crc32cw %w[c], %w[c], %w[v]":[c]"+r"(crc):[v]"r"(value))
#define CRC32CH(crc, value) __asm__("crc32ch %w[c], %w[c], %w[v]":[c]"+r"(crc):[v]"r"(value))
#define CRC32CB(crc, value) __asm__("crc32cb %w[c], %w[c], %w[v]":[c]"+r"(crc):[v]"r"(value))
#define CRC32C3X8(buffer, ITR) \
__asm__("crc32cx %w[c1], %w[c1], %x[v]":[c1]"+r"(crc1):[v]"r"(*((const uint64_t *)buffer + 42*1 + (ITR))));\
__asm__("crc32cx %w[c2], %w[c2], %x[v]":[c2]"+r"(crc2):[v]"r"(*((const uint64_t *)buffer + 42*2 + (ITR))));\
__asm__("crc32cx %w[c0], %w[c0], %x[v]":[c0]"+r"(crc0):[v]"r"(*((const uint64_t *)buffer + 42*0 + (ITR))));
#define CRC32C3X8_ZERO \
__asm__("crc32cx %w[c0], %w[c0], xzr":[c0]"+r"(crc0));
#else /* HAVE_ARMV8_CRC_CRYPTO_INTRINSICS */
/* Intrinsics header*/
#include <arm_acle.h>
#include <arm_neon.h>
#define CRC32CX(crc, value) (crc) = __crc32cd((crc), (value))
#define CRC32CW(crc, value) (crc) = __crc32cw((crc), (value))
#define CRC32CH(crc, value) (crc) = __crc32ch((crc), (value))
#define CRC32CB(crc, value) (crc) = __crc32cb((crc), (value))
#define CRC32C3X8(buffer, ITR) \
crc1 = __crc32cd(crc1, *((const uint64_t *)buffer + 42*1 + (ITR)));\
crc2 = __crc32cd(crc2, *((const uint64_t *)buffer + 42*2 + (ITR)));\
crc0 = __crc32cd(crc0, *((const uint64_t *)buffer + 42*0 + (ITR)));
#define CRC32C3X8_ZERO \
crc0 = __crc32cd(crc0, (const uint64_t)0);
#endif /* HAVE_ARMV8_CRC_CRYPTO_INTRINSICS */
#define CRC32C7X3X8(buffer, ITR) do {\
CRC32C3X8(buffer, ((ITR) * 7 + 0)) \
CRC32C3X8(buffer, ((ITR) * 7 + 1)) \
CRC32C3X8(buffer, ((ITR) * 7 + 2)) \
CRC32C3X8(buffer, ((ITR) * 7 + 3)) \
CRC32C3X8(buffer, ((ITR) * 7 + 4)) \
CRC32C3X8(buffer, ((ITR) * 7 + 5)) \
CRC32C3X8(buffer, ((ITR) * 7 + 6)) \
} while(0)
#define CRC32C7X3X8_ZERO do {\
CRC32C3X8_ZERO \
CRC32C3X8_ZERO \
CRC32C3X8_ZERO \
CRC32C3X8_ZERO \
CRC32C3X8_ZERO \
CRC32C3X8_ZERO \
CRC32C3X8_ZERO \
} while(0)
#define PREF4X64L1(buffer, PREF_OFFSET, ITR) \
__asm__("PRFM PLDL1KEEP, [%x[v],%[c]]"::[v]"r"(buffer), [c]"I"((PREF_OFFSET) + ((ITR) + 0)*64));\
__asm__("PRFM PLDL1KEEP, [%x[v],%[c]]"::[v]"r"(buffer), [c]"I"((PREF_OFFSET) + ((ITR) + 1)*64));\
__asm__("PRFM PLDL1KEEP, [%x[v],%[c]]"::[v]"r"(buffer), [c]"I"((PREF_OFFSET) + ((ITR) + 2)*64));\
__asm__("PRFM PLDL1KEEP, [%x[v],%[c]]"::[v]"r"(buffer), [c]"I"((PREF_OFFSET) + ((ITR) + 3)*64));
#define PREF1KL1(buffer, PREF_OFFSET) \
PREF4X64L1(buffer,(PREF_OFFSET), 0) \
PREF4X64L1(buffer,(PREF_OFFSET), 4) \
PREF4X64L1(buffer,(PREF_OFFSET), 8) \
PREF4X64L1(buffer,(PREF_OFFSET), 12)
#define PREF4X64L2(buffer, PREF_OFFSET, ITR) \
__asm__("PRFM PLDL2KEEP, [%x[v],%[c]]"::[v]"r"(buffer), [c]"I"((PREF_OFFSET) + ((ITR) + 0)*64));\
__asm__("PRFM PLDL2KEEP, [%x[v],%[c]]"::[v]"r"(buffer), [c]"I"((PREF_OFFSET) + ((ITR) + 1)*64));\
__asm__("PRFM PLDL2KEEP, [%x[v],%[c]]"::[v]"r"(buffer), [c]"I"((PREF_OFFSET) + ((ITR) + 2)*64));\
__asm__("PRFM PLDL2KEEP, [%x[v],%[c]]"::[v]"r"(buffer), [c]"I"((PREF_OFFSET) + ((ITR) + 3)*64));
#define PREF1KL2(buffer, PREF_OFFSET) \
PREF4X64L2(buffer,(PREF_OFFSET), 0) \
PREF4X64L2(buffer,(PREF_OFFSET), 4) \
PREF4X64L2(buffer,(PREF_OFFSET), 8) \
PREF4X64L2(buffer,(PREF_OFFSET), 12)
uint32_t crc32c_aarch64(uint32_t crc, const unsigned char *buffer, uint64_t len)
{
uint32_t crc0, crc1, crc2;
int64_t length = (int64_t)len;
crc = 0xFFFFFFFFU;
if (buffer) {
/* Crypto extension Support
* Process 1024 Bytes (per block)
*/
#ifdef HAVE_ARMV8_CRYPTO
/* Intrinsics Support */
#ifdef HAVE_ARMV8_CRC_CRYPTO_INTRINSICS
const poly64_t k1 = 0xe417f38a, k2 = 0x8f158014;
uint64_t t0, t1;
/* Process per block size of 1024 Bytes
* A block size = 8 + 42*3*sizeof(uint64_t) + 8
*/
while ((length -= 1024) >= 0) {
/* Prefetch 3*1024 data for avoiding L2 cache miss */
PREF1KL2(buffer, 1024*3);
/* Do first 8 bytes here for better pipelining */
crc0 = __crc32cd(crc, *(const uint64_t *)buffer);
crc1 = 0;
crc2 = 0;
buffer += sizeof(uint64_t);
/* Process block inline
* Process crc0 last to avoid dependency with above
*/
CRC32C7X3X8(buffer, 0);
CRC32C7X3X8(buffer, 1);
CRC32C7X3X8(buffer, 2);
CRC32C7X3X8(buffer, 3);
CRC32C7X3X8(buffer, 4);
CRC32C7X3X8(buffer, 5);
buffer += 42*3*sizeof(uint64_t);
/* Prefetch data for following block to avoid L1 cache miss */
PREF1KL1(buffer, 1024);
/* Last 8 bytes
* Merge crc0 and crc1 into crc2
* crc1 multiply by K2
* crc0 multiply by K1
*/
t1 = (uint64_t)vmull_p64(crc1, k2);
t0 = (uint64_t)vmull_p64(crc0, k1);
crc = __crc32cd(crc2, *(const uint64_t *)buffer);
crc1 = __crc32cd(0, t1);
crc ^= crc1;
crc0 = __crc32cd(0, t0);
crc ^= crc0;
buffer += sizeof(uint64_t);
}
#else /* HAVE_ARMV8_CRC_CRYPTO_INTRINSICS */
/*No intrinsics*/
__asm__("mov x16, #0xf38a \n\t"
"movk x16, #0xe417, lsl 16 \n\t"
"mov v1.2d[0], x16 \n\t"
"mov x16, #0x8014 \n\t"
"movk x16, #0x8f15, lsl 16 \n\t"
"mov v0.2d[0], x16 \n\t"
:::"x16");
while ((length -= 1024) >= 0) {
PREF1KL2(buffer, 1024*3);
__asm__("crc32cx %w[c0], %w[c], %x[v]\n\t"
:[c0]"=r"(crc0):[c]"r"(crc), [v]"r"(*(const uint64_t *)buffer):);
crc1 = 0;
crc2 = 0;
buffer += sizeof(uint64_t);
CRC32C7X3X8(buffer, 0);
CRC32C7X3X8(buffer, 1);
CRC32C7X3X8(buffer, 2);
CRC32C7X3X8(buffer, 3);
CRC32C7X3X8(buffer, 4);
CRC32C7X3X8(buffer, 5);
buffer += 42*3*sizeof(uint64_t);
PREF1KL1(buffer, 1024);
__asm__("mov v2.2d[0], %x[c1] \n\t"
"pmull v2.1q, v2.1d, v0.1d \n\t"
"mov v3.2d[0], %x[c0] \n\t"
"pmull v3.1q, v3.1d, v1.1d \n\t"
"crc32cx %w[c], %w[c2], %x[v] \n\t"
"mov %x[c1], v2.2d[0] \n\t"
"crc32cx %w[c1], wzr, %x[c1] \n\t"
"eor %w[c], %w[c], %w[c1] \n\t"
"mov %x[c0], v3.2d[0] \n\t"
"crc32cx %w[c0], wzr, %x[c0] \n\t"
"eor %w[c], %w[c], %w[c0] \n\t"
:[c1]"+r"(crc1), [c0]"+r"(crc0), [c2]"+r"(crc2), [c]"+r"(crc)
:[v]"r"(*((const uint64_t *)buffer)));
buffer += sizeof(uint64_t);
}
#endif /* HAVE_ARMV8_CRC_CRYPTO_INTRINSICS */
/* Done if Input data size is aligned with 1024 */
if(!(length += 1024))
return (~crc);
#endif /* HAVE_ARMV8_CRYPTO */
while ((length -= sizeof(uint64_t)) >= 0) {
CRC32CX(crc, *(uint64_t *)buffer);
buffer += sizeof(uint64_t);
}
/* The following is more efficient than the straight loop */
if (length & sizeof(uint32_t)) {
CRC32CW(crc, *(uint32_t *)buffer);
buffer += sizeof(uint32_t);
}
if (length & sizeof(uint16_t)) {
CRC32CH(crc, *(uint16_t *)buffer);
buffer += sizeof(uint16_t);
}
if (length & sizeof(uint8_t))
CRC32CB(crc, *buffer);
} else {
#ifdef HAVE_ARMV8_CRYPTO
#ifdef HAVE_ARMV8_CRC_CRYPTO_INTRINSICS
const poly64_t k1 = 0xe417f38a;
uint64_t t0;
while ((length -= 1024) >= 0) {
crc0 = __crc32cd(crc, 0);
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
/* Merge crc0 into crc: crc0 multiply by K1 */
t0 = (uint64_t)vmull_p64(crc0, k1);
crc = __crc32cd(0, t0);
}
#else /* !HAVE_ARMV8_CRC_CRYPTO_INTRINSICS */
__asm__("mov x16, #0xf38a \n\t"
"movk x16, #0xe417, lsl 16 \n\t"
"mov v1.2d[0], x16 \n\t"
:::"x16");
while ((length -= 1024) >= 0) {
__asm__("crc32cx %w[c0], %w[c], xzr\n\t"
:[c0]"=r"(crc0):[c]"r"(crc));
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
CRC32C7X3X8_ZERO;
__asm__("mov v3.2d[0], %x[c0] \n\t"
"pmull v3.1q, v3.1d, v1.1d \n\t"
"mov %x[c0], v3.2d[0] \n\t"
"crc32cx %w[c], wzr, %x[c0] \n\t"
:[c]"=r"(crc)
:[c0]"r"(crc0));
}
#endif /* HAVE_ARMV8_CRC_CRYPTO_INTRINSICS */
if(!(length += 1024))
return (~crc);
#endif /* HAVE_ARMV8_CRYPTO */
while ((length -= sizeof(uint64_t)) >= 0)
CRC32CX(crc, 0);
/* The following is more efficient than the straight loop */
if (length & sizeof(uint32_t))
CRC32CW(crc, 0);
if (length & sizeof(uint16_t))
CRC32CH(crc, 0);
if (length & sizeof(uint8_t))
CRC32CB(crc, 0);
}
return (~crc);
}
|
