1 /*
2 * Based on node-yencode library by Anime Tosho:
3 * https://github.com/animetosho/node-yencode
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program. If not, see <http://www.gnu.org/licenses/>.
17 */
18
19 // taken from zlib-ng / Intel's zlib patch, modified to remove zlib dependencies
20 /*
21 * Compute the CRC32 using a parallelized folding approach with the PCLMULQDQ
22 * instruction.
23 *
24 * A white paper describing this algorithm can be found at:
25 * http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/fast-crc-computation-generic-polynomials-pclmulqdq-paper.pdf
26 *
27 * Copyright (C) 2013 Intel Corporation. All rights reserved.
28 * Authors:
29 * Wajdi Feghali <wajdi.k.feghali@intel.com>
30 * Jim Guilford <james.guilford@intel.com>
31 * Vinodh Gopal <vinodh.gopal@intel.com>
32 * Erdinc Ozturk <erdinc.ozturk@intel.com>
33 * Jim Kukunas <james.t.kukunas@linux.intel.com>
34 *
35 * For conditions of distribution and use, see copyright notice in zlib.h
36 */
37
38 #include "nzbget.h"
39
40 #include "YEncode.h"
41
42 #ifdef __PCLMUL__
43 #include <immintrin.h>
44 #endif
45
46 namespace YEncode
47 {
48 #ifdef __PCLMUL__
49
fold_1(__m128i * xmm_crc0,__m128i * xmm_crc1,__m128i * xmm_crc2,__m128i * xmm_crc3)50 void fold_1(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2, __m128i *xmm_crc3) {
51 const __m128i xmm_fold4 = _mm_set_epi32(
52 0x00000001, 0x54442bd4,
53 0x00000001, 0xc6e41596);
54
55 __m128i x_tmp3;
56 __m128 ps_crc0, ps_crc3, ps_res;
57
58 x_tmp3 = *xmm_crc3;
59
60 *xmm_crc3 = *xmm_crc0;
61 *xmm_crc0 = _mm_clmulepi64_si128(*xmm_crc0, xmm_fold4, 0x01);
62 *xmm_crc3 = _mm_clmulepi64_si128(*xmm_crc3, xmm_fold4, 0x10);
63 ps_crc0 = _mm_castsi128_ps(*xmm_crc0);
64 ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
65 ps_res = _mm_xor_ps(ps_crc0, ps_crc3);
66
67 *xmm_crc0 = *xmm_crc1;
68 *xmm_crc1 = *xmm_crc2;
69 *xmm_crc2 = x_tmp3;
70 *xmm_crc3 = _mm_castps_si128(ps_res);
71 }
72
fold_2(__m128i * xmm_crc0,__m128i * xmm_crc1,__m128i * xmm_crc2,__m128i * xmm_crc3)73 void fold_2(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2, __m128i *xmm_crc3) {
74 const __m128i xmm_fold4 = _mm_set_epi32(
75 0x00000001, 0x54442bd4,
76 0x00000001, 0xc6e41596);
77
78 __m128i x_tmp3, x_tmp2;
79 __m128 ps_crc0, ps_crc1, ps_crc2, ps_crc3, ps_res31, ps_res20;
80
81 x_tmp3 = *xmm_crc3;
82 x_tmp2 = *xmm_crc2;
83
84 *xmm_crc3 = *xmm_crc1;
85 *xmm_crc1 = _mm_clmulepi64_si128(*xmm_crc1, xmm_fold4, 0x01);
86 *xmm_crc3 = _mm_clmulepi64_si128(*xmm_crc3, xmm_fold4, 0x10);
87 ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
88 ps_crc1 = _mm_castsi128_ps(*xmm_crc1);
89 ps_res31 = _mm_xor_ps(ps_crc3, ps_crc1);
90
91 *xmm_crc2 = *xmm_crc0;
92 *xmm_crc0 = _mm_clmulepi64_si128(*xmm_crc0, xmm_fold4, 0x01);
93 *xmm_crc2 = _mm_clmulepi64_si128(*xmm_crc2, xmm_fold4, 0x10);
94 ps_crc0 = _mm_castsi128_ps(*xmm_crc0);
95 ps_crc2 = _mm_castsi128_ps(*xmm_crc2);
96 ps_res20 = _mm_xor_ps(ps_crc0, ps_crc2);
97
98 *xmm_crc0 = x_tmp2;
99 *xmm_crc1 = x_tmp3;
100 *xmm_crc2 = _mm_castps_si128(ps_res20);
101 *xmm_crc3 = _mm_castps_si128(ps_res31);
102 }
103
fold_3(__m128i * xmm_crc0,__m128i * xmm_crc1,__m128i * xmm_crc2,__m128i * xmm_crc3)104 void fold_3(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2, __m128i *xmm_crc3) {
105 const __m128i xmm_fold4 = _mm_set_epi32(
106 0x00000001, 0x54442bd4,
107 0x00000001, 0xc6e41596);
108
109 __m128i x_tmp3;
110 __m128 ps_crc0, ps_crc1, ps_crc2, ps_crc3, ps_res32, ps_res21, ps_res10;
111
112 x_tmp3 = *xmm_crc3;
113
114 *xmm_crc3 = *xmm_crc2;
115 *xmm_crc2 = _mm_clmulepi64_si128(*xmm_crc2, xmm_fold4, 0x01);
116 *xmm_crc3 = _mm_clmulepi64_si128(*xmm_crc3, xmm_fold4, 0x10);
117 ps_crc2 = _mm_castsi128_ps(*xmm_crc2);
118 ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
119 ps_res32 = _mm_xor_ps(ps_crc2, ps_crc3);
120
121 *xmm_crc2 = *xmm_crc1;
122 *xmm_crc1 = _mm_clmulepi64_si128(*xmm_crc1, xmm_fold4, 0x01);
123 *xmm_crc2 = _mm_clmulepi64_si128(*xmm_crc2, xmm_fold4, 0x10);
124 ps_crc1 = _mm_castsi128_ps(*xmm_crc1);
125 ps_crc2 = _mm_castsi128_ps(*xmm_crc2);
126 ps_res21 = _mm_xor_ps(ps_crc1, ps_crc2);
127
128 *xmm_crc1 = *xmm_crc0;
129 *xmm_crc0 = _mm_clmulepi64_si128(*xmm_crc0, xmm_fold4, 0x01);
130 *xmm_crc1 = _mm_clmulepi64_si128(*xmm_crc1, xmm_fold4, 0x10);
131 ps_crc0 = _mm_castsi128_ps(*xmm_crc0);
132 ps_crc1 = _mm_castsi128_ps(*xmm_crc1);
133 ps_res10 = _mm_xor_ps(ps_crc0, ps_crc1);
134
135 *xmm_crc0 = x_tmp3;
136 *xmm_crc1 = _mm_castps_si128(ps_res10);
137 *xmm_crc2 = _mm_castps_si128(ps_res21);
138 *xmm_crc3 = _mm_castps_si128(ps_res32);
139 }
140
fold_4(__m128i * xmm_crc0,__m128i * xmm_crc1,__m128i * xmm_crc2,__m128i * xmm_crc3)141 void fold_4(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2, __m128i *xmm_crc3) {
142 const __m128i xmm_fold4 = _mm_set_epi32(
143 0x00000001, 0x54442bd4,
144 0x00000001, 0xc6e41596);
145
146 __m128i x_tmp0, x_tmp1, x_tmp2, x_tmp3;
147 __m128 ps_crc0, ps_crc1, ps_crc2, ps_crc3;
148 __m128 ps_t0, ps_t1, ps_t2, ps_t3;
149 __m128 ps_res0, ps_res1, ps_res2, ps_res3;
150
151 x_tmp0 = *xmm_crc0;
152 x_tmp1 = *xmm_crc1;
153 x_tmp2 = *xmm_crc2;
154 x_tmp3 = *xmm_crc3;
155
156 *xmm_crc0 = _mm_clmulepi64_si128(*xmm_crc0, xmm_fold4, 0x01);
157 x_tmp0 = _mm_clmulepi64_si128(x_tmp0, xmm_fold4, 0x10);
158 ps_crc0 = _mm_castsi128_ps(*xmm_crc0);
159 ps_t0 = _mm_castsi128_ps(x_tmp0);
160 ps_res0 = _mm_xor_ps(ps_crc0, ps_t0);
161
162 *xmm_crc1 = _mm_clmulepi64_si128(*xmm_crc1, xmm_fold4, 0x01);
163 x_tmp1 = _mm_clmulepi64_si128(x_tmp1, xmm_fold4, 0x10);
164 ps_crc1 = _mm_castsi128_ps(*xmm_crc1);
165 ps_t1 = _mm_castsi128_ps(x_tmp1);
166 ps_res1 = _mm_xor_ps(ps_crc1, ps_t1);
167
168 *xmm_crc2 = _mm_clmulepi64_si128(*xmm_crc2, xmm_fold4, 0x01);
169 x_tmp2 = _mm_clmulepi64_si128(x_tmp2, xmm_fold4, 0x10);
170 ps_crc2 = _mm_castsi128_ps(*xmm_crc2);
171 ps_t2 = _mm_castsi128_ps(x_tmp2);
172 ps_res2 = _mm_xor_ps(ps_crc2, ps_t2);
173
174 *xmm_crc3 = _mm_clmulepi64_si128(*xmm_crc3, xmm_fold4, 0x01);
175 x_tmp3 = _mm_clmulepi64_si128(x_tmp3, xmm_fold4, 0x10);
176 ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
177 ps_t3 = _mm_castsi128_ps(x_tmp3);
178 ps_res3 = _mm_xor_ps(ps_crc3, ps_t3);
179
180 *xmm_crc0 = _mm_castps_si128(ps_res0);
181 *xmm_crc1 = _mm_castps_si128(ps_res1);
182 *xmm_crc2 = _mm_castps_si128(ps_res2);
183 *xmm_crc3 = _mm_castps_si128(ps_res3);
184 }
185
186 alignas(32) const unsigned pshufb_shf_table[60] = {
187 0x84838281, 0x88878685, 0x8c8b8a89, 0x008f8e8d, /* shl 15 (16 - 1)/shr1 */
188 0x85848382, 0x89888786, 0x8d8c8b8a, 0x01008f8e, /* shl 14 (16 - 3)/shr2 */
189 0x86858483, 0x8a898887, 0x8e8d8c8b, 0x0201008f, /* shl 13 (16 - 4)/shr3 */
190 0x87868584, 0x8b8a8988, 0x8f8e8d8c, 0x03020100, /* shl 12 (16 - 4)/shr4 */
191 0x88878685, 0x8c8b8a89, 0x008f8e8d, 0x04030201, /* shl 11 (16 - 5)/shr5 */
192 0x89888786, 0x8d8c8b8a, 0x01008f8e, 0x05040302, /* shl 10 (16 - 6)/shr6 */
193 0x8a898887, 0x8e8d8c8b, 0x0201008f, 0x06050403, /* shl 9 (16 - 7)/shr7 */
194 0x8b8a8988, 0x8f8e8d8c, 0x03020100, 0x07060504, /* shl 8 (16 - 8)/shr8 */
195 0x8c8b8a89, 0x008f8e8d, 0x04030201, 0x08070605, /* shl 7 (16 - 9)/shr9 */
196 0x8d8c8b8a, 0x01008f8e, 0x05040302, 0x09080706, /* shl 6 (16 -10)/shr10*/
197 0x8e8d8c8b, 0x0201008f, 0x06050403, 0x0a090807, /* shl 5 (16 -11)/shr11*/
198 0x8f8e8d8c, 0x03020100, 0x07060504, 0x0b0a0908, /* shl 4 (16 -12)/shr12*/
199 0x008f8e8d, 0x04030201, 0x08070605, 0x0c0b0a09, /* shl 3 (16 -13)/shr13*/
200 0x01008f8e, 0x05040302, 0x09080706, 0x0d0c0b0a, /* shl 2 (16 -14)/shr14*/
201 0x0201008f, 0x06050403, 0x0a090807, 0x0e0d0c0b /* shl 1 (16 -15)/shr15*/
202 };
203
partial_fold(const size_t len,__m128i * xmm_crc0,__m128i * xmm_crc1,__m128i * xmm_crc2,__m128i * xmm_crc3,__m128i * xmm_crc_part)204 void partial_fold(const size_t len, __m128i *xmm_crc0, __m128i *xmm_crc1,
205 __m128i *xmm_crc2, __m128i *xmm_crc3, __m128i *xmm_crc_part) {
206
207 const __m128i xmm_fold4 = _mm_set_epi32(
208 0x00000001, 0x54442bd4,
209 0x00000001, 0xc6e41596);
210 const __m128i xmm_mask3 = _mm_set1_epi32(0x80808080);
211
212 __m128i xmm_shl, xmm_shr, xmm_tmp1, xmm_tmp2, xmm_tmp3;
213 __m128i xmm_a0_0, xmm_a0_1;
214 __m128 ps_crc3, psa0_0, psa0_1, ps_res;
215
216 xmm_shl = _mm_load_si128((__m128i *)pshufb_shf_table + (len - 1));
217 xmm_shr = xmm_shl;
218 xmm_shr = _mm_xor_si128(xmm_shr, xmm_mask3);
219
220 xmm_a0_0 = _mm_shuffle_epi8(*xmm_crc0, xmm_shl);
221
222 *xmm_crc0 = _mm_shuffle_epi8(*xmm_crc0, xmm_shr);
223 xmm_tmp1 = _mm_shuffle_epi8(*xmm_crc1, xmm_shl);
224 *xmm_crc0 = _mm_or_si128(*xmm_crc0, xmm_tmp1);
225
226 *xmm_crc1 = _mm_shuffle_epi8(*xmm_crc1, xmm_shr);
227 xmm_tmp2 = _mm_shuffle_epi8(*xmm_crc2, xmm_shl);
228 *xmm_crc1 = _mm_or_si128(*xmm_crc1, xmm_tmp2);
229
230 *xmm_crc2 = _mm_shuffle_epi8(*xmm_crc2, xmm_shr);
231 xmm_tmp3 = _mm_shuffle_epi8(*xmm_crc3, xmm_shl);
232 *xmm_crc2 = _mm_or_si128(*xmm_crc2, xmm_tmp3);
233
234 *xmm_crc3 = _mm_shuffle_epi8(*xmm_crc3, xmm_shr);
235 *xmm_crc_part = _mm_shuffle_epi8(*xmm_crc_part, xmm_shl);
236 *xmm_crc3 = _mm_or_si128(*xmm_crc3, *xmm_crc_part);
237
238 xmm_a0_1 = _mm_clmulepi64_si128(xmm_a0_0, xmm_fold4, 0x10);
239 xmm_a0_0 = _mm_clmulepi64_si128(xmm_a0_0, xmm_fold4, 0x01);
240
241 ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
242 psa0_0 = _mm_castsi128_ps(xmm_a0_0);
243 psa0_1 = _mm_castsi128_ps(xmm_a0_1);
244
245 ps_res = _mm_xor_ps(ps_crc3, psa0_0);
246 ps_res = _mm_xor_ps(ps_res, psa0_1);
247
248 *xmm_crc3 = _mm_castps_si128(ps_res);
249 }
250
251 alignas(16) const unsigned crc_k[] = {
252 0xccaa009e, 0x00000000, /* rk1 */
253 0x751997d0, 0x00000001, /* rk2 */
254 0xccaa009e, 0x00000000, /* rk5 */
255 0x63cd6124, 0x00000001, /* rk6 */
256 0xf7011640, 0x00000001, /* rk7 */
257 0xdb710640, 0x00000001 /* rk8 */
258 };
259
260 alignas(16) const unsigned crc_mask[4] = {
261 0xFFFFFFFF, 0xFFFFFFFF, 0x00000000, 0x00000000
262 };
263
264 alignas(16) const unsigned crc_mask2[4] = {
265 0x00000000, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF
266 };
267
268 #define CRC_LOAD(s) \
269 do { \
270 __m128i xmm_crc0 = _mm_loadu_si128((__m128i *)s->crc0 + 0);\
271 __m128i xmm_crc1 = _mm_loadu_si128((__m128i *)s->crc0 + 1);\
272 __m128i xmm_crc2 = _mm_loadu_si128((__m128i *)s->crc0 + 2);\
273 __m128i xmm_crc3 = _mm_loadu_si128((__m128i *)s->crc0 + 3);\
274 __m128i xmm_crc_part = _mm_loadu_si128((__m128i *)s->crc0 + 4);
275
276 #define CRC_SAVE(s) \
277 _mm_storeu_si128((__m128i *)s->crc0 + 0, xmm_crc0);\
278 _mm_storeu_si128((__m128i *)s->crc0 + 1, xmm_crc1);\
279 _mm_storeu_si128((__m128i *)s->crc0 + 2, xmm_crc2);\
280 _mm_storeu_si128((__m128i *)s->crc0 + 3, xmm_crc3);\
281 _mm_storeu_si128((__m128i *)s->crc0 + 4, xmm_crc_part);\
282 } while (0);
283
crc_fold_init(crc_state * const s)284 void crc_fold_init(crc_state *const s) {
285 CRC_LOAD(s)
286
287 xmm_crc0 = _mm_cvtsi32_si128(0x9db42487);
288 xmm_crc1 = _mm_setzero_si128();
289 xmm_crc2 = _mm_setzero_si128();
290 xmm_crc3 = _mm_setzero_si128();
291
292 CRC_SAVE(s)
293 }
294
crc_fold(crc_state * const s,const unsigned char * src,long len)295 void crc_fold(crc_state *const s, const unsigned char *src, long len) {
296 unsigned long algn_diff;
297 __m128i xmm_t0, xmm_t1, xmm_t2, xmm_t3;
298
299 CRC_LOAD(s)
300
301 if (len < 16) {
302 if (len == 0)
303 return;
304 xmm_crc_part = _mm_loadu_si128((__m128i *)src);
305 goto partial;
306 }
307
308 algn_diff = (0 - (uintptr_t)src) & 0xF;
309 if (algn_diff) {
310 xmm_crc_part = _mm_loadu_si128((__m128i *)src);
311 src += algn_diff;
312 len -= algn_diff;
313
314 partial_fold(algn_diff, &xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3, &xmm_crc_part);
315 }
316
317 while ((len -= 64) >= 0) {
318 xmm_t0 = _mm_load_si128((__m128i *)src);
319 xmm_t1 = _mm_load_si128((__m128i *)src + 1);
320 xmm_t2 = _mm_load_si128((__m128i *)src + 2);
321 xmm_t3 = _mm_load_si128((__m128i *)src + 3);
322
323 fold_4(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
324
325 xmm_crc0 = _mm_xor_si128(xmm_crc0, xmm_t0);
326 xmm_crc1 = _mm_xor_si128(xmm_crc1, xmm_t1);
327 xmm_crc2 = _mm_xor_si128(xmm_crc2, xmm_t2);
328 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t3);
329
330 src += 64;
331 }
332
333 /*
334 * len = num bytes left - 64
335 */
336 if (len + 16 >= 0) {
337 len += 16;
338
339 xmm_t0 = _mm_load_si128((__m128i *)src);
340 xmm_t1 = _mm_load_si128((__m128i *)src + 1);
341 xmm_t2 = _mm_load_si128((__m128i *)src + 2);
342
343 fold_3(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
344
345 xmm_crc1 = _mm_xor_si128(xmm_crc1, xmm_t0);
346 xmm_crc2 = _mm_xor_si128(xmm_crc2, xmm_t1);
347 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t2);
348
349 if (len == 0)
350 goto done;
351
352 xmm_crc_part = _mm_load_si128((__m128i *)src + 3);
353 } else if (len + 32 >= 0) {
354 len += 32;
355
356 xmm_t0 = _mm_load_si128((__m128i *)src);
357 xmm_t1 = _mm_load_si128((__m128i *)src + 1);
358
359 fold_2(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
360
361 xmm_crc2 = _mm_xor_si128(xmm_crc2, xmm_t0);
362 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t1);
363
364 if (len == 0)
365 goto done;
366
367 xmm_crc_part = _mm_load_si128((__m128i *)src + 2);
368 } else if (len + 48 >= 0) {
369 len += 48;
370
371 xmm_t0 = _mm_load_si128((__m128i *)src);
372
373 fold_1(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
374
375 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t0);
376
377 if (len == 0)
378 goto done;
379
380 xmm_crc_part = _mm_load_si128((__m128i *)src + 1);
381 } else {
382 len += 64;
383 if (len == 0)
384 goto done;
385 xmm_crc_part = _mm_load_si128((__m128i *)src);
386 }
387
388 partial:
389 partial_fold(len, &xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3, &xmm_crc_part);
390 done:
391 CRC_SAVE(s)
392 }
393
crc_fold_512to32(crc_state * const s)394 uint32_t crc_fold_512to32(crc_state *const s) {
395 const __m128i xmm_mask = _mm_load_si128((__m128i *)crc_mask);
396 const __m128i xmm_mask2 = _mm_load_si128((__m128i *)crc_mask2);
397
398 uint32_t crc;
399 __m128i x_tmp0, x_tmp1, x_tmp2, crc_fold;
400
401 CRC_LOAD(s)
402
403 /*
404 * k1
405 */
406 crc_fold = _mm_load_si128((__m128i *)crc_k);
407
408 x_tmp0 = _mm_clmulepi64_si128(xmm_crc0, crc_fold, 0x10);
409 xmm_crc0 = _mm_clmulepi64_si128(xmm_crc0, crc_fold, 0x01);
410 xmm_crc1 = _mm_xor_si128(xmm_crc1, x_tmp0);
411 xmm_crc1 = _mm_xor_si128(xmm_crc1, xmm_crc0);
412
413 x_tmp1 = _mm_clmulepi64_si128(xmm_crc1, crc_fold, 0x10);
414 xmm_crc1 = _mm_clmulepi64_si128(xmm_crc1, crc_fold, 0x01);
415 xmm_crc2 = _mm_xor_si128(xmm_crc2, x_tmp1);
416 xmm_crc2 = _mm_xor_si128(xmm_crc2, xmm_crc1);
417
418 x_tmp2 = _mm_clmulepi64_si128(xmm_crc2, crc_fold, 0x10);
419 xmm_crc2 = _mm_clmulepi64_si128(xmm_crc2, crc_fold, 0x01);
420 xmm_crc3 = _mm_xor_si128(xmm_crc3, x_tmp2);
421 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc2);
422
423 /*
424 * k5
425 */
426 crc_fold = _mm_load_si128((__m128i *)crc_k + 1);
427
428 xmm_crc0 = xmm_crc3;
429 xmm_crc3 = _mm_clmulepi64_si128(xmm_crc3, crc_fold, 0);
430 xmm_crc0 = _mm_srli_si128(xmm_crc0, 8);
431 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc0);
432
433 xmm_crc0 = xmm_crc3;
434 xmm_crc3 = _mm_slli_si128(xmm_crc3, 4);
435 xmm_crc3 = _mm_clmulepi64_si128(xmm_crc3, crc_fold, 0x10);
436 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc0);
437 xmm_crc3 = _mm_and_si128(xmm_crc3, xmm_mask2);
438
439 /*
440 * k7
441 */
442 xmm_crc1 = xmm_crc3;
443 xmm_crc2 = xmm_crc3;
444 crc_fold = _mm_load_si128((__m128i *)crc_k + 2);
445
446 xmm_crc3 = _mm_clmulepi64_si128(xmm_crc3, crc_fold, 0);
447 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc2);
448 xmm_crc3 = _mm_and_si128(xmm_crc3, xmm_mask);
449
450 xmm_crc2 = xmm_crc3;
451 xmm_crc3 = _mm_clmulepi64_si128(xmm_crc3, crc_fold, 0x10);
452 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc2);
453 xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc1);
454
455 crc = _mm_extract_epi32(xmm_crc3, 2);
456 return ~crc;
457 CRC_SAVE(s)
458 }
459 #endif
460
init_crc_pclmul()461 void init_crc_pclmul()
462 {
463 #ifdef __PCLMUL__
464 crc_init = &crc_fold_init;
465 crc_incr = &crc_fold;
466 crc_finish = &crc_fold_512to32;
467 crc_simd = true;
468 #endif
469 }
470
471 }
472