9e052883 |
1 | /* crc32.c -- compute the CRC-32 of a data stream |
2 | * Copyright (C) 1995-2022 Mark Adler |
3 | * For conditions of distribution and use, see copyright notice in zlib.h |
4 | * |
5 | * This interleaved implementation of a CRC makes use of pipelined multiple |
6 | * arithmetic-logic units, commonly found in modern CPU cores. It is due to |
7 | * Kadatch and Jenkins (2010). See doc/crc-doc.1.0.pdf in this distribution. |
8 | */ |
9 | |
10 | /* @(#) $Id$ */ |
11 | |
12 | /* |
13 | Note on the use of DYNAMIC_CRC_TABLE: there is no mutex or semaphore |
14 | protection on the static variables used to control the first-use generation |
15 | of the crc tables. Therefore, if you #define DYNAMIC_CRC_TABLE, you should |
16 | first call get_crc_table() to initialize the tables before allowing more than |
17 | one thread to use crc32(). |
18 | |
19 | MAKECRCH can be #defined to write out crc32.h. A main() routine is also |
20 | produced, so that this one source file can be compiled to an executable. |
21 | */ |
22 | |
23 | #ifdef MAKECRCH |
24 | # include <stdio.h> |
25 | # ifndef DYNAMIC_CRC_TABLE |
26 | # define DYNAMIC_CRC_TABLE |
27 | # endif /* !DYNAMIC_CRC_TABLE */ |
28 | #endif /* MAKECRCH */ |
29 | |
30 | #include "zutil.h" /* for Z_U4, Z_U8, z_crc_t, and FAR definitions */ |
31 | |
32 | /* |
33 | A CRC of a message is computed on N braids of words in the message, where |
34 | each word consists of W bytes (4 or 8). If N is 3, for example, then three |
35 | running sparse CRCs are calculated respectively on each braid, at these |
36 | indices in the array of words: 0, 3, 6, ..., 1, 4, 7, ..., and 2, 5, 8, ... |
37 | This is done starting at a word boundary, and continues until as many blocks |
38 | of N * W bytes as are available have been processed. The results are combined |
39 | into a single CRC at the end. For this code, N must be in the range 1..6 and |
40 | W must be 4 or 8. The upper limit on N can be increased if desired by adding |
41 | more #if blocks, extending the patterns apparent in the code. In addition, |
42 | crc32.h would need to be regenerated, if the maximum N value is increased. |
43 | |
44 | N and W are chosen empirically by benchmarking the execution time on a given |
45 | processor. The choices for N and W below were based on testing on Intel Kaby |
46 | Lake i7, AMD Ryzen 7, ARM Cortex-A57, Sparc64-VII, PowerPC POWER9, and MIPS64 |
47 | Octeon II processors. The Intel, AMD, and ARM processors were all fastest |
48 | with N=5, W=8. The Sparc, PowerPC, and MIPS64 were all fastest at N=5, W=4. |
49 | They were all tested with either gcc or clang, all using the -O3 optimization |
50 | level. Your mileage may vary. |
51 | */ |
52 | |
53 | /* Define N */ |
54 | #ifdef Z_TESTN |
55 | # define N Z_TESTN |
56 | #else |
57 | # define N 5 |
58 | #endif |
59 | #if N < 1 || N > 6 |
60 | # error N must be in 1..6 |
61 | #endif |
62 | |
63 | /* |
64 | z_crc_t must be at least 32 bits. z_word_t must be at least as long as |
65 | z_crc_t. It is assumed here that z_word_t is either 32 bits or 64 bits, and |
66 | that bytes are eight bits. |
67 | */ |
68 | |
69 | /* |
70 | Define W and the associated z_word_t type. If W is not defined, then a |
71 | braided calculation is not used, and the associated tables and code are not |
72 | compiled. |
73 | */ |
74 | #ifdef Z_TESTW |
75 | # if Z_TESTW-1 != -1 |
76 | # define W Z_TESTW |
77 | # endif |
78 | #else |
79 | # ifdef MAKECRCH |
80 | # define W 8 /* required for MAKECRCH */ |
81 | # else |
82 | # if defined(__x86_64__) || defined(__aarch64__) |
83 | # define W 8 |
84 | # else |
85 | # define W 4 |
86 | # endif |
87 | # endif |
88 | #endif |
89 | #ifdef W |
90 | # if W == 8 && defined(Z_U8) |
91 | typedef Z_U8 z_word_t; |
92 | # elif defined(Z_U4) |
93 | # undef W |
94 | # define W 4 |
95 | typedef Z_U4 z_word_t; |
96 | # else |
97 | # undef W |
98 | # endif |
99 | #endif |
100 | |
101 | /* If available, use the ARM processor CRC32 instruction. */ |
102 | #if defined(__aarch64__) && defined(__ARM_FEATURE_CRC32) && W == 8 |
103 | # define ARMCRC32 |
104 | #endif |
105 | |
106 | /* Local functions. */ |
107 | local z_crc_t multmodp OF((z_crc_t a, z_crc_t b)); |
108 | local z_crc_t x2nmodp OF((z_off64_t n, unsigned k)); |
109 | |
110 | #if defined(W) && (!defined(ARMCRC32) || defined(DYNAMIC_CRC_TABLE)) |
111 | local z_word_t byte_swap OF((z_word_t word)); |
112 | #endif |
113 | |
114 | #if defined(W) && !defined(ARMCRC32) |
115 | local z_crc_t crc_word OF((z_word_t data)); |
116 | local z_word_t crc_word_big OF((z_word_t data)); |
117 | #endif |
118 | |
119 | #if defined(W) && (!defined(ARMCRC32) || defined(DYNAMIC_CRC_TABLE)) |
120 | /* |
121 | Swap the bytes in a z_word_t to convert between little and big endian. Any |
122 | self-respecting compiler will optimize this to a single machine byte-swap |
123 | instruction, if one is available. This assumes that word_t is either 32 bits |
124 | or 64 bits. |
125 | */ |
126 | local z_word_t byte_swap(word) |
127 | z_word_t word; |
128 | { |
129 | # if W == 8 |
130 | return |
131 | (word & 0xff00000000000000) >> 56 | |
132 | (word & 0xff000000000000) >> 40 | |
133 | (word & 0xff0000000000) >> 24 | |
134 | (word & 0xff00000000) >> 8 | |
135 | (word & 0xff000000) << 8 | |
136 | (word & 0xff0000) << 24 | |
137 | (word & 0xff00) << 40 | |
138 | (word & 0xff) << 56; |
139 | # else /* W == 4 */ |
140 | return |
141 | (word & 0xff000000) >> 24 | |
142 | (word & 0xff0000) >> 8 | |
143 | (word & 0xff00) << 8 | |
144 | (word & 0xff) << 24; |
145 | # endif |
146 | } |
147 | #endif |
148 | |
149 | /* CRC polynomial. */ |
150 | #define POLY 0xedb88320 /* p(x) reflected, with x^32 implied */ |
151 | |
152 | #ifdef DYNAMIC_CRC_TABLE |
153 | |
154 | local z_crc_t FAR crc_table[256]; |
155 | local z_crc_t FAR x2n_table[32]; |
156 | local void make_crc_table OF((void)); |
157 | #ifdef W |
158 | local z_word_t FAR crc_big_table[256]; |
159 | local z_crc_t FAR crc_braid_table[W][256]; |
160 | local z_word_t FAR crc_braid_big_table[W][256]; |
161 | local void braid OF((z_crc_t [][256], z_word_t [][256], int, int)); |
162 | #endif |
163 | #ifdef MAKECRCH |
164 | local void write_table OF((FILE *, const z_crc_t FAR *, int)); |
165 | local void write_table32hi OF((FILE *, const z_word_t FAR *, int)); |
166 | local void write_table64 OF((FILE *, const z_word_t FAR *, int)); |
167 | #endif /* MAKECRCH */ |
168 | |
169 | /* |
170 | Define a once() function depending on the availability of atomics. If this is |
171 | compiled with DYNAMIC_CRC_TABLE defined, and if CRCs will be computed in |
172 | multiple threads, and if atomics are not available, then get_crc_table() must |
173 | be called to initialize the tables and must return before any threads are |
174 | allowed to compute or combine CRCs. |
175 | */ |
176 | |
177 | /* Definition of once functionality. */ |
178 | typedef struct once_s once_t; |
179 | local void once OF((once_t *, void (*)(void))); |
180 | |
181 | /* Check for the availability of atomics. */ |
182 | #if defined(__STDC__) && __STDC_VERSION__ >= 201112L && \ |
183 | !defined(__STDC_NO_ATOMICS__) |
184 | |
185 | #include <stdatomic.h> |
186 | |
187 | /* Structure for once(), which must be initialized with ONCE_INIT. */ |
188 | struct once_s { |
189 | atomic_flag begun; |
190 | atomic_int done; |
191 | }; |
192 | #define ONCE_INIT {ATOMIC_FLAG_INIT, 0} |
193 | |
194 | /* |
195 | Run the provided init() function exactly once, even if multiple threads |
196 | invoke once() at the same time. The state must be a once_t initialized with |
197 | ONCE_INIT. |
198 | */ |
199 | local void once(state, init) |
200 | once_t *state; |
201 | void (*init)(void); |
202 | { |
203 | if (!atomic_load(&state->done)) { |
204 | if (atomic_flag_test_and_set(&state->begun)) |
205 | while (!atomic_load(&state->done)) |
206 | ; |
207 | else { |
208 | init(); |
209 | atomic_store(&state->done, 1); |
210 | } |
211 | } |
212 | } |
213 | |
214 | #else /* no atomics */ |
215 | |
216 | /* Structure for once(), which must be initialized with ONCE_INIT. */ |
217 | struct once_s { |
218 | volatile int begun; |
219 | volatile int done; |
220 | }; |
221 | #define ONCE_INIT {0, 0} |
222 | |
223 | /* Test and set. Alas, not atomic, but tries to minimize the period of |
224 | vulnerability. */ |
225 | local int test_and_set OF((int volatile *)); |
226 | local int test_and_set(flag) |
227 | int volatile *flag; |
228 | { |
229 | int was; |
230 | |
231 | was = *flag; |
232 | *flag = 1; |
233 | return was; |
234 | } |
235 | |
236 | /* Run the provided init() function once. This is not thread-safe. */ |
237 | local void once(state, init) |
238 | once_t *state; |
239 | void (*init)(void); |
240 | { |
241 | if (!state->done) { |
242 | if (test_and_set(&state->begun)) |
243 | while (!state->done) |
244 | ; |
245 | else { |
246 | init(); |
247 | state->done = 1; |
248 | } |
249 | } |
250 | } |
251 | |
252 | #endif |
253 | |
254 | /* State for once(). */ |
255 | local once_t made = ONCE_INIT; |
256 | |
257 | /* |
258 | Generate tables for a byte-wise 32-bit CRC calculation on the polynomial: |
259 | x^32+x^26+x^23+x^22+x^16+x^12+x^11+x^10+x^8+x^7+x^5+x^4+x^2+x+1. |
260 | |
261 | Polynomials over GF(2) are represented in binary, one bit per coefficient, |
262 | with the lowest powers in the most significant bit. Then adding polynomials |
263 | is just exclusive-or, and multiplying a polynomial by x is a right shift by |
264 | one. If we call the above polynomial p, and represent a byte as the |
265 | polynomial q, also with the lowest power in the most significant bit (so the |
266 | byte 0xb1 is the polynomial x^7+x^3+x^2+1), then the CRC is (q*x^32) mod p, |
267 | where a mod b means the remainder after dividing a by b. |
268 | |
269 | This calculation is done using the shift-register method of multiplying and |
270 | taking the remainder. The register is initialized to zero, and for each |
271 | incoming bit, x^32 is added mod p to the register if the bit is a one (where |
272 | x^32 mod p is p+x^32 = x^26+...+1), and the register is multiplied mod p by x |
273 | (which is shifting right by one and adding x^32 mod p if the bit shifted out |
274 | is a one). We start with the highest power (least significant bit) of q and |
275 | repeat for all eight bits of q. |
276 | |
277 | The table is simply the CRC of all possible eight bit values. This is all the |
278 | information needed to generate CRCs on data a byte at a time for all |
279 | combinations of CRC register values and incoming bytes. |
280 | */ |
281 | |
282 | local void make_crc_table() |
283 | { |
284 | unsigned i, j, n; |
285 | z_crc_t p; |
286 | |
287 | /* initialize the CRC of bytes tables */ |
288 | for (i = 0; i < 256; i++) { |
289 | p = i; |
290 | for (j = 0; j < 8; j++) |
291 | p = p & 1 ? (p >> 1) ^ POLY : p >> 1; |
292 | crc_table[i] = p; |
293 | #ifdef W |
294 | crc_big_table[i] = byte_swap(p); |
295 | #endif |
296 | } |
297 | |
298 | /* initialize the x^2^n mod p(x) table */ |
299 | p = (z_crc_t)1 << 30; /* x^1 */ |
300 | x2n_table[0] = p; |
301 | for (n = 1; n < 32; n++) |
302 | x2n_table[n] = p = multmodp(p, p); |
303 | |
304 | #ifdef W |
305 | /* initialize the braiding tables -- needs x2n_table[] */ |
306 | braid(crc_braid_table, crc_braid_big_table, N, W); |
307 | #endif |
308 | |
309 | #ifdef MAKECRCH |
310 | { |
311 | /* |
312 | The crc32.h header file contains tables for both 32-bit and 64-bit |
313 | z_word_t's, and so requires a 64-bit type be available. In that case, |
314 | z_word_t must be defined to be 64-bits. This code then also generates |
315 | and writes out the tables for the case that z_word_t is 32 bits. |
316 | */ |
317 | #if !defined(W) || W != 8 |
318 | # error Need a 64-bit integer type in order to generate crc32.h. |
319 | #endif |
320 | FILE *out; |
321 | int k, n; |
322 | z_crc_t ltl[8][256]; |
323 | z_word_t big[8][256]; |
324 | |
325 | out = fopen("crc32.h", "w"); |
326 | if (out == NULL) return; |
327 | |
328 | /* write out little-endian CRC table to crc32.h */ |
329 | fprintf(out, |
330 | "/* crc32.h -- tables for rapid CRC calculation\n" |
331 | " * Generated automatically by crc32.c\n */\n" |
332 | "\n" |
333 | "local const z_crc_t FAR crc_table[] = {\n" |
334 | " "); |
335 | write_table(out, crc_table, 256); |
336 | fprintf(out, |
337 | "};\n"); |
338 | |
339 | /* write out big-endian CRC table for 64-bit z_word_t to crc32.h */ |
340 | fprintf(out, |
341 | "\n" |
342 | "#ifdef W\n" |
343 | "\n" |
344 | "#if W == 8\n" |
345 | "\n" |
346 | "local const z_word_t FAR crc_big_table[] = {\n" |
347 | " "); |
348 | write_table64(out, crc_big_table, 256); |
349 | fprintf(out, |
350 | "};\n"); |
351 | |
352 | /* write out big-endian CRC table for 32-bit z_word_t to crc32.h */ |
353 | fprintf(out, |
354 | "\n" |
355 | "#else /* W == 4 */\n" |
356 | "\n" |
357 | "local const z_word_t FAR crc_big_table[] = {\n" |
358 | " "); |
359 | write_table32hi(out, crc_big_table, 256); |
360 | fprintf(out, |
361 | "};\n" |
362 | "\n" |
363 | "#endif\n"); |
364 | |
365 | /* write out braid tables for each value of N */ |
366 | for (n = 1; n <= 6; n++) { |
367 | fprintf(out, |
368 | "\n" |
369 | "#if N == %d\n", n); |
370 | |
371 | /* compute braid tables for this N and 64-bit word_t */ |
372 | braid(ltl, big, n, 8); |
373 | |
374 | /* write out braid tables for 64-bit z_word_t to crc32.h */ |
375 | fprintf(out, |
376 | "\n" |
377 | "#if W == 8\n" |
378 | "\n" |
379 | "local const z_crc_t FAR crc_braid_table[][256] = {\n"); |
380 | for (k = 0; k < 8; k++) { |
381 | fprintf(out, " {"); |
382 | write_table(out, ltl[k], 256); |
383 | fprintf(out, "}%s", k < 7 ? ",\n" : ""); |
384 | } |
385 | fprintf(out, |
386 | "};\n" |
387 | "\n" |
388 | "local const z_word_t FAR crc_braid_big_table[][256] = {\n"); |
389 | for (k = 0; k < 8; k++) { |
390 | fprintf(out, " {"); |
391 | write_table64(out, big[k], 256); |
392 | fprintf(out, "}%s", k < 7 ? ",\n" : ""); |
393 | } |
394 | fprintf(out, |
395 | "};\n"); |
396 | |
397 | /* compute braid tables for this N and 32-bit word_t */ |
398 | braid(ltl, big, n, 4); |
399 | |
400 | /* write out braid tables for 32-bit z_word_t to crc32.h */ |
401 | fprintf(out, |
402 | "\n" |
403 | "#else /* W == 4 */\n" |
404 | "\n" |
405 | "local const z_crc_t FAR crc_braid_table[][256] = {\n"); |
406 | for (k = 0; k < 4; k++) { |
407 | fprintf(out, " {"); |
408 | write_table(out, ltl[k], 256); |
409 | fprintf(out, "}%s", k < 3 ? ",\n" : ""); |
410 | } |
411 | fprintf(out, |
412 | "};\n" |
413 | "\n" |
414 | "local const z_word_t FAR crc_braid_big_table[][256] = {\n"); |
415 | for (k = 0; k < 4; k++) { |
416 | fprintf(out, " {"); |
417 | write_table32hi(out, big[k], 256); |
418 | fprintf(out, "}%s", k < 3 ? ",\n" : ""); |
419 | } |
420 | fprintf(out, |
421 | "};\n" |
422 | "\n" |
423 | "#endif\n" |
424 | "\n" |
425 | "#endif\n"); |
426 | } |
427 | fprintf(out, |
428 | "\n" |
429 | "#endif\n"); |
430 | |
431 | /* write out zeros operator table to crc32.h */ |
432 | fprintf(out, |
433 | "\n" |
434 | "local const z_crc_t FAR x2n_table[] = {\n" |
435 | " "); |
436 | write_table(out, x2n_table, 32); |
437 | fprintf(out, |
438 | "};\n"); |
439 | fclose(out); |
440 | } |
441 | #endif /* MAKECRCH */ |
442 | } |
443 | |
444 | #ifdef MAKECRCH |
445 | |
446 | /* |
447 | Write the 32-bit values in table[0..k-1] to out, five per line in |
448 | hexadecimal separated by commas. |
449 | */ |
450 | local void write_table(out, table, k) |
451 | FILE *out; |
452 | const z_crc_t FAR *table; |
453 | int k; |
454 | { |
455 | int n; |
456 | |
457 | for (n = 0; n < k; n++) |
458 | fprintf(out, "%s0x%08lx%s", n == 0 || n % 5 ? "" : " ", |
459 | (unsigned long)(table[n]), |
460 | n == k - 1 ? "" : (n % 5 == 4 ? ",\n" : ", ")); |
461 | } |
462 | |
463 | /* |
464 | Write the high 32-bits of each value in table[0..k-1] to out, five per line |
465 | in hexadecimal separated by commas. |
466 | */ |
467 | local void write_table32hi(out, table, k) |
468 | FILE *out; |
469 | const z_word_t FAR *table; |
470 | int k; |
471 | { |
472 | int n; |
473 | |
474 | for (n = 0; n < k; n++) |
475 | fprintf(out, "%s0x%08lx%s", n == 0 || n % 5 ? "" : " ", |
476 | (unsigned long)(table[n] >> 32), |
477 | n == k - 1 ? "" : (n % 5 == 4 ? ",\n" : ", ")); |
478 | } |
479 | |
480 | /* |
481 | Write the 64-bit values in table[0..k-1] to out, three per line in |
482 | hexadecimal separated by commas. This assumes that if there is a 64-bit |
483 | type, then there is also a long long integer type, and it is at least 64 |
484 | bits. If not, then the type cast and format string can be adjusted |
485 | accordingly. |
486 | */ |
487 | local void write_table64(out, table, k) |
488 | FILE *out; |
489 | const z_word_t FAR *table; |
490 | int k; |
491 | { |
492 | int n; |
493 | |
494 | for (n = 0; n < k; n++) |
495 | fprintf(out, "%s0x%016llx%s", n == 0 || n % 3 ? "" : " ", |
496 | (unsigned long long)(table[n]), |
497 | n == k - 1 ? "" : (n % 3 == 2 ? ",\n" : ", ")); |
498 | } |
499 | |
500 | /* Actually do the deed. */ |
501 | int main() |
502 | { |
503 | make_crc_table(); |
504 | return 0; |
505 | } |
506 | |
507 | #endif /* MAKECRCH */ |
508 | |
509 | #ifdef W |
510 | /* |
511 | Generate the little and big-endian braid tables for the given n and z_word_t |
512 | size w. Each array must have room for w blocks of 256 elements. |
513 | */ |
514 | local void braid(ltl, big, n, w) |
515 | z_crc_t ltl[][256]; |
516 | z_word_t big[][256]; |
517 | int n; |
518 | int w; |
519 | { |
520 | int k; |
521 | z_crc_t i, p, q; |
522 | for (k = 0; k < w; k++) { |
523 | p = x2nmodp((n * w + 3 - k) << 3, 0); |
524 | ltl[k][0] = 0; |
525 | big[w - 1 - k][0] = 0; |
526 | for (i = 1; i < 256; i++) { |
527 | ltl[k][i] = q = multmodp(i << 24, p); |
528 | big[w - 1 - k][i] = byte_swap(q); |
529 | } |
530 | } |
531 | } |
532 | #endif |
533 | |
534 | #else /* !DYNAMIC_CRC_TABLE */ |
535 | /* ======================================================================== |
536 | * Tables for byte-wise and braided CRC-32 calculations, and a table of powers |
537 | * of x for combining CRC-32s, all made by make_crc_table(). |
538 | */ |
539 | #include "crc32.h" |
540 | #endif /* DYNAMIC_CRC_TABLE */ |
541 | |
542 | /* ======================================================================== |
543 | * Routines used for CRC calculation. Some are also required for the table |
544 | * generation above. |
545 | */ |
546 | |
547 | /* |
548 | Return a(x) multiplied by b(x) modulo p(x), where p(x) is the CRC polynomial, |
549 | reflected. For speed, this requires that a not be zero. |
550 | */ |
551 | local z_crc_t multmodp(a, b) |
552 | z_crc_t a; |
553 | z_crc_t b; |
554 | { |
555 | z_crc_t m, p; |
556 | |
557 | m = (z_crc_t)1 << 31; |
558 | p = 0; |
559 | for (;;) { |
560 | if (a & m) { |
561 | p ^= b; |
562 | if ((a & (m - 1)) == 0) |
563 | break; |
564 | } |
565 | m >>= 1; |
566 | b = b & 1 ? (b >> 1) ^ POLY : b >> 1; |
567 | } |
568 | return p; |
569 | } |
570 | |
571 | /* |
572 | Return x^(n * 2^k) modulo p(x). Requires that x2n_table[] has been |
573 | initialized. |
574 | */ |
575 | local z_crc_t x2nmodp(n, k) |
576 | z_off64_t n; |
577 | unsigned k; |
578 | { |
579 | z_crc_t p; |
580 | |
581 | p = (z_crc_t)1 << 31; /* x^0 == 1 */ |
582 | while (n) { |
583 | if (n & 1) |
584 | p = multmodp(x2n_table[k & 31], p); |
585 | n >>= 1; |
586 | k++; |
587 | } |
588 | return p; |
589 | } |
590 | |
591 | /* ========================================================================= |
592 | * This function can be used by asm versions of crc32(), and to force the |
593 | * generation of the CRC tables in a threaded application. |
594 | */ |
595 | const z_crc_t FAR * ZEXPORT get_crc_table() |
596 | { |
597 | #ifdef DYNAMIC_CRC_TABLE |
598 | once(&made, make_crc_table); |
599 | #endif /* DYNAMIC_CRC_TABLE */ |
600 | return (const z_crc_t FAR *)crc_table; |
601 | } |
602 | |
603 | /* ========================================================================= |
604 | * Use ARM machine instructions if available. This will compute the CRC about |
605 | * ten times faster than the braided calculation. This code does not check for |
606 | * the presence of the CRC instruction at run time. __ARM_FEATURE_CRC32 will |
607 | * only be defined if the compilation specifies an ARM processor architecture |
608 | * that has the instructions. For example, compiling with -march=armv8.1-a or |
609 | * -march=armv8-a+crc, or -march=native if the compile machine has the crc32 |
610 | * instructions. |
611 | */ |
612 | #ifdef ARMCRC32 |
613 | |
614 | /* |
615 | Constants empirically determined to maximize speed. These values are from |
616 | measurements on a Cortex-A57. Your mileage may vary. |
617 | */ |
618 | #define Z_BATCH 3990 /* number of words in a batch */ |
619 | #define Z_BATCH_ZEROS 0xa10d3d0c /* computed from Z_BATCH = 3990 */ |
620 | #define Z_BATCH_MIN 800 /* fewest words in a final batch */ |
621 | |
622 | unsigned long ZEXPORT crc32_z(crc, buf, len) |
623 | unsigned long crc; |
624 | const unsigned char FAR *buf; |
625 | z_size_t len; |
626 | { |
627 | z_crc_t val; |
628 | z_word_t crc1, crc2; |
629 | const z_word_t *word; |
630 | z_word_t val0, val1, val2; |
631 | z_size_t last, last2, i; |
632 | z_size_t num; |
633 | |
634 | /* Return initial CRC, if requested. */ |
635 | if (buf == Z_NULL) return 0; |
636 | |
637 | #ifdef DYNAMIC_CRC_TABLE |
638 | once(&made, make_crc_table); |
639 | #endif /* DYNAMIC_CRC_TABLE */ |
640 | |
641 | /* Pre-condition the CRC */ |
642 | crc = (~crc) & 0xffffffff; |
643 | |
644 | /* Compute the CRC up to a word boundary. */ |
645 | while (len && ((z_size_t)buf & 7) != 0) { |
646 | len--; |
647 | val = *buf++; |
648 | __asm__ volatile("crc32b %w0, %w0, %w1" : "+r"(crc) : "r"(val)); |
649 | } |
650 | |
651 | /* Prepare to compute the CRC on full 64-bit words word[0..num-1]. */ |
652 | word = (z_word_t const *)buf; |
653 | num = len >> 3; |
654 | len &= 7; |
655 | |
656 | /* Do three interleaved CRCs to realize the throughput of one crc32x |
657 | instruction per cycle. Each CRC is calculated on Z_BATCH words. The |
658 | three CRCs are combined into a single CRC after each set of batches. */ |
659 | while (num >= 3 * Z_BATCH) { |
660 | crc1 = 0; |
661 | crc2 = 0; |
662 | for (i = 0; i < Z_BATCH; i++) { |
663 | val0 = word[i]; |
664 | val1 = word[i + Z_BATCH]; |
665 | val2 = word[i + 2 * Z_BATCH]; |
666 | __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0)); |
667 | __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc1) : "r"(val1)); |
668 | __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc2) : "r"(val2)); |
669 | } |
670 | word += 3 * Z_BATCH; |
671 | num -= 3 * Z_BATCH; |
672 | crc = multmodp(Z_BATCH_ZEROS, crc) ^ crc1; |
673 | crc = multmodp(Z_BATCH_ZEROS, crc) ^ crc2; |
674 | } |
675 | |
676 | /* Do one last smaller batch with the remaining words, if there are enough |
677 | to pay for the combination of CRCs. */ |
678 | last = num / 3; |
679 | if (last >= Z_BATCH_MIN) { |
680 | last2 = last << 1; |
681 | crc1 = 0; |
682 | crc2 = 0; |
683 | for (i = 0; i < last; i++) { |
684 | val0 = word[i]; |
685 | val1 = word[i + last]; |
686 | val2 = word[i + last2]; |
687 | __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0)); |
688 | __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc1) : "r"(val1)); |
689 | __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc2) : "r"(val2)); |
690 | } |
691 | word += 3 * last; |
692 | num -= 3 * last; |
693 | val = x2nmodp(last, 6); |
694 | crc = multmodp(val, crc) ^ crc1; |
695 | crc = multmodp(val, crc) ^ crc2; |
696 | } |
697 | |
698 | /* Compute the CRC on any remaining words. */ |
699 | for (i = 0; i < num; i++) { |
700 | val0 = word[i]; |
701 | __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0)); |
702 | } |
703 | word += num; |
704 | |
705 | /* Complete the CRC on any remaining bytes. */ |
706 | buf = (const unsigned char FAR *)word; |
707 | while (len) { |
708 | len--; |
709 | val = *buf++; |
710 | __asm__ volatile("crc32b %w0, %w0, %w1" : "+r"(crc) : "r"(val)); |
711 | } |
712 | |
713 | /* Return the CRC, post-conditioned. */ |
714 | return crc ^ 0xffffffff; |
715 | } |
716 | |
717 | #else |
718 | |
719 | #ifdef W |
720 | |
721 | /* |
722 | Return the CRC of the W bytes in the word_t data, taking the |
723 | least-significant byte of the word as the first byte of data, without any pre |
724 | or post conditioning. This is used to combine the CRCs of each braid. |
725 | */ |
726 | local z_crc_t crc_word(data) |
727 | z_word_t data; |
728 | { |
729 | int k; |
730 | for (k = 0; k < W; k++) |
731 | data = (data >> 8) ^ crc_table[data & 0xff]; |
732 | return (z_crc_t)data; |
733 | } |
734 | |
735 | local z_word_t crc_word_big(data) |
736 | z_word_t data; |
737 | { |
738 | int k; |
739 | for (k = 0; k < W; k++) |
740 | data = (data << 8) ^ |
741 | crc_big_table[(data >> ((W - 1) << 3)) & 0xff]; |
742 | return data; |
743 | } |
744 | |
745 | #endif |
746 | |
747 | /* ========================================================================= */ |
748 | unsigned long ZEXPORT crc32_z(crc, buf, len) |
749 | unsigned long crc; |
750 | const unsigned char FAR *buf; |
751 | z_size_t len; |
752 | { |
753 | /* Return initial CRC, if requested. */ |
754 | if (buf == Z_NULL) return 0; |
755 | |
756 | #ifdef DYNAMIC_CRC_TABLE |
757 | once(&made, make_crc_table); |
758 | #endif /* DYNAMIC_CRC_TABLE */ |
759 | |
760 | /* Pre-condition the CRC */ |
761 | crc = (~crc) & 0xffffffff; |
762 | |
763 | #ifdef W |
764 | |
765 | /* If provided enough bytes, do a braided CRC calculation. */ |
766 | if (len >= N * W + W - 1) { |
767 | z_size_t blks; |
768 | z_word_t const *words; |
769 | unsigned endian; |
770 | int k; |
771 | |
772 | /* Compute the CRC up to a z_word_t boundary. */ |
773 | while (len && ((z_size_t)buf & (W - 1)) != 0) { |
774 | len--; |
775 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
776 | } |
777 | |
778 | /* Compute the CRC on as many N z_word_t blocks as are available. */ |
779 | blks = len / (N * W); |
780 | len -= blks * N * W; |
781 | words = (z_word_t const *)buf; |
782 | |
783 | /* Do endian check at execution time instead of compile time, since ARM |
784 | processors can change the endianess at execution time. If the |
785 | compiler knows what the endianess will be, it can optimize out the |
786 | check and the unused branch. */ |
787 | endian = 1; |
788 | if (*(unsigned char *)&endian) { |
789 | /* Little endian. */ |
790 | |
791 | z_crc_t crc0; |
792 | z_word_t word0; |
793 | #if N > 1 |
794 | z_crc_t crc1; |
795 | z_word_t word1; |
796 | #if N > 2 |
797 | z_crc_t crc2; |
798 | z_word_t word2; |
799 | #if N > 3 |
800 | z_crc_t crc3; |
801 | z_word_t word3; |
802 | #if N > 4 |
803 | z_crc_t crc4; |
804 | z_word_t word4; |
805 | #if N > 5 |
806 | z_crc_t crc5; |
807 | z_word_t word5; |
808 | #endif |
809 | #endif |
810 | #endif |
811 | #endif |
812 | #endif |
813 | |
814 | /* Initialize the CRC for each braid. */ |
815 | crc0 = crc; |
816 | #if N > 1 |
817 | crc1 = 0; |
818 | #if N > 2 |
819 | crc2 = 0; |
820 | #if N > 3 |
821 | crc3 = 0; |
822 | #if N > 4 |
823 | crc4 = 0; |
824 | #if N > 5 |
825 | crc5 = 0; |
826 | #endif |
827 | #endif |
828 | #endif |
829 | #endif |
830 | #endif |
831 | |
832 | /* |
833 | Process the first blks-1 blocks, computing the CRCs on each braid |
834 | independently. |
835 | */ |
836 | while (--blks) { |
837 | /* Load the word for each braid into registers. */ |
838 | word0 = crc0 ^ words[0]; |
839 | #if N > 1 |
840 | word1 = crc1 ^ words[1]; |
841 | #if N > 2 |
842 | word2 = crc2 ^ words[2]; |
843 | #if N > 3 |
844 | word3 = crc3 ^ words[3]; |
845 | #if N > 4 |
846 | word4 = crc4 ^ words[4]; |
847 | #if N > 5 |
848 | word5 = crc5 ^ words[5]; |
849 | #endif |
850 | #endif |
851 | #endif |
852 | #endif |
853 | #endif |
854 | words += N; |
855 | |
856 | /* Compute and update the CRC for each word. The loop should |
857 | get unrolled. */ |
858 | crc0 = crc_braid_table[0][word0 & 0xff]; |
859 | #if N > 1 |
860 | crc1 = crc_braid_table[0][word1 & 0xff]; |
861 | #if N > 2 |
862 | crc2 = crc_braid_table[0][word2 & 0xff]; |
863 | #if N > 3 |
864 | crc3 = crc_braid_table[0][word3 & 0xff]; |
865 | #if N > 4 |
866 | crc4 = crc_braid_table[0][word4 & 0xff]; |
867 | #if N > 5 |
868 | crc5 = crc_braid_table[0][word5 & 0xff]; |
869 | #endif |
870 | #endif |
871 | #endif |
872 | #endif |
873 | #endif |
874 | for (k = 1; k < W; k++) { |
875 | crc0 ^= crc_braid_table[k][(word0 >> (k << 3)) & 0xff]; |
876 | #if N > 1 |
877 | crc1 ^= crc_braid_table[k][(word1 >> (k << 3)) & 0xff]; |
878 | #if N > 2 |
879 | crc2 ^= crc_braid_table[k][(word2 >> (k << 3)) & 0xff]; |
880 | #if N > 3 |
881 | crc3 ^= crc_braid_table[k][(word3 >> (k << 3)) & 0xff]; |
882 | #if N > 4 |
883 | crc4 ^= crc_braid_table[k][(word4 >> (k << 3)) & 0xff]; |
884 | #if N > 5 |
885 | crc5 ^= crc_braid_table[k][(word5 >> (k << 3)) & 0xff]; |
886 | #endif |
887 | #endif |
888 | #endif |
889 | #endif |
890 | #endif |
891 | } |
892 | } |
893 | |
894 | /* |
895 | Process the last block, combining the CRCs of the N braids at the |
896 | same time. |
897 | */ |
898 | crc = crc_word(crc0 ^ words[0]); |
899 | #if N > 1 |
900 | crc = crc_word(crc1 ^ words[1] ^ crc); |
901 | #if N > 2 |
902 | crc = crc_word(crc2 ^ words[2] ^ crc); |
903 | #if N > 3 |
904 | crc = crc_word(crc3 ^ words[3] ^ crc); |
905 | #if N > 4 |
906 | crc = crc_word(crc4 ^ words[4] ^ crc); |
907 | #if N > 5 |
908 | crc = crc_word(crc5 ^ words[5] ^ crc); |
909 | #endif |
910 | #endif |
911 | #endif |
912 | #endif |
913 | #endif |
914 | words += N; |
915 | } |
916 | else { |
917 | /* Big endian. */ |
918 | |
919 | z_word_t crc0, word0, comb; |
920 | #if N > 1 |
921 | z_word_t crc1, word1; |
922 | #if N > 2 |
923 | z_word_t crc2, word2; |
924 | #if N > 3 |
925 | z_word_t crc3, word3; |
926 | #if N > 4 |
927 | z_word_t crc4, word4; |
928 | #if N > 5 |
929 | z_word_t crc5, word5; |
930 | #endif |
931 | #endif |
932 | #endif |
933 | #endif |
934 | #endif |
935 | |
936 | /* Initialize the CRC for each braid. */ |
937 | crc0 = byte_swap(crc); |
938 | #if N > 1 |
939 | crc1 = 0; |
940 | #if N > 2 |
941 | crc2 = 0; |
942 | #if N > 3 |
943 | crc3 = 0; |
944 | #if N > 4 |
945 | crc4 = 0; |
946 | #if N > 5 |
947 | crc5 = 0; |
948 | #endif |
949 | #endif |
950 | #endif |
951 | #endif |
952 | #endif |
953 | |
954 | /* |
955 | Process the first blks-1 blocks, computing the CRCs on each braid |
956 | independently. |
957 | */ |
958 | while (--blks) { |
959 | /* Load the word for each braid into registers. */ |
960 | word0 = crc0 ^ words[0]; |
961 | #if N > 1 |
962 | word1 = crc1 ^ words[1]; |
963 | #if N > 2 |
964 | word2 = crc2 ^ words[2]; |
965 | #if N > 3 |
966 | word3 = crc3 ^ words[3]; |
967 | #if N > 4 |
968 | word4 = crc4 ^ words[4]; |
969 | #if N > 5 |
970 | word5 = crc5 ^ words[5]; |
971 | #endif |
972 | #endif |
973 | #endif |
974 | #endif |
975 | #endif |
976 | words += N; |
977 | |
978 | /* Compute and update the CRC for each word. The loop should |
979 | get unrolled. */ |
980 | crc0 = crc_braid_big_table[0][word0 & 0xff]; |
981 | #if N > 1 |
982 | crc1 = crc_braid_big_table[0][word1 & 0xff]; |
983 | #if N > 2 |
984 | crc2 = crc_braid_big_table[0][word2 & 0xff]; |
985 | #if N > 3 |
986 | crc3 = crc_braid_big_table[0][word3 & 0xff]; |
987 | #if N > 4 |
988 | crc4 = crc_braid_big_table[0][word4 & 0xff]; |
989 | #if N > 5 |
990 | crc5 = crc_braid_big_table[0][word5 & 0xff]; |
991 | #endif |
992 | #endif |
993 | #endif |
994 | #endif |
995 | #endif |
996 | for (k = 1; k < W; k++) { |
997 | crc0 ^= crc_braid_big_table[k][(word0 >> (k << 3)) & 0xff]; |
998 | #if N > 1 |
999 | crc1 ^= crc_braid_big_table[k][(word1 >> (k << 3)) & 0xff]; |
1000 | #if N > 2 |
1001 | crc2 ^= crc_braid_big_table[k][(word2 >> (k << 3)) & 0xff]; |
1002 | #if N > 3 |
1003 | crc3 ^= crc_braid_big_table[k][(word3 >> (k << 3)) & 0xff]; |
1004 | #if N > 4 |
1005 | crc4 ^= crc_braid_big_table[k][(word4 >> (k << 3)) & 0xff]; |
1006 | #if N > 5 |
1007 | crc5 ^= crc_braid_big_table[k][(word5 >> (k << 3)) & 0xff]; |
1008 | #endif |
1009 | #endif |
1010 | #endif |
1011 | #endif |
1012 | #endif |
1013 | } |
1014 | } |
1015 | |
1016 | /* |
1017 | Process the last block, combining the CRCs of the N braids at the |
1018 | same time. |
1019 | */ |
1020 | comb = crc_word_big(crc0 ^ words[0]); |
1021 | #if N > 1 |
1022 | comb = crc_word_big(crc1 ^ words[1] ^ comb); |
1023 | #if N > 2 |
1024 | comb = crc_word_big(crc2 ^ words[2] ^ comb); |
1025 | #if N > 3 |
1026 | comb = crc_word_big(crc3 ^ words[3] ^ comb); |
1027 | #if N > 4 |
1028 | comb = crc_word_big(crc4 ^ words[4] ^ comb); |
1029 | #if N > 5 |
1030 | comb = crc_word_big(crc5 ^ words[5] ^ comb); |
1031 | #endif |
1032 | #endif |
1033 | #endif |
1034 | #endif |
1035 | #endif |
1036 | words += N; |
1037 | crc = byte_swap(comb); |
1038 | } |
1039 | |
1040 | /* |
1041 | Update the pointer to the remaining bytes to process. |
1042 | */ |
1043 | buf = (unsigned char const *)words; |
1044 | } |
1045 | |
1046 | #endif /* W */ |
1047 | |
1048 | /* Complete the computation of the CRC on any remaining bytes. */ |
1049 | while (len >= 8) { |
1050 | len -= 8; |
1051 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1052 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1053 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1054 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1055 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1056 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1057 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1058 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1059 | } |
1060 | while (len) { |
1061 | len--; |
1062 | crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff]; |
1063 | } |
1064 | |
1065 | /* Return the CRC, post-conditioned. */ |
1066 | return crc ^ 0xffffffff; |
1067 | } |
1068 | |
1069 | #endif |
1070 | |
1071 | /* ========================================================================= */ |
1072 | unsigned long ZEXPORT crc32(crc, buf, len) |
1073 | unsigned long crc; |
1074 | const unsigned char FAR *buf; |
1075 | uInt len; |
1076 | { |
1077 | return crc32_z(crc, buf, len); |
1078 | } |
1079 | |
1080 | /* ========================================================================= */ |
1081 | uLong ZEXPORT crc32_combine64(crc1, crc2, len2) |
1082 | uLong crc1; |
1083 | uLong crc2; |
1084 | z_off64_t len2; |
1085 | { |
1086 | #ifdef DYNAMIC_CRC_TABLE |
1087 | once(&made, make_crc_table); |
1088 | #endif /* DYNAMIC_CRC_TABLE */ |
1089 | return multmodp(x2nmodp(len2, 3), crc1) ^ (crc2 & 0xffffffff); |
1090 | } |
1091 | |
1092 | /* ========================================================================= */ |
1093 | uLong ZEXPORT crc32_combine(crc1, crc2, len2) |
1094 | uLong crc1; |
1095 | uLong crc2; |
1096 | z_off_t len2; |
1097 | { |
1098 | return crc32_combine64(crc1, crc2, (z_off64_t)len2); |
1099 | } |
1100 | |
1101 | /* ========================================================================= */ |
1102 | uLong ZEXPORT crc32_combine_gen64(len2) |
1103 | z_off64_t len2; |
1104 | { |
1105 | #ifdef DYNAMIC_CRC_TABLE |
1106 | once(&made, make_crc_table); |
1107 | #endif /* DYNAMIC_CRC_TABLE */ |
1108 | return x2nmodp(len2, 3); |
1109 | } |
1110 | |
1111 | /* ========================================================================= */ |
1112 | uLong ZEXPORT crc32_combine_gen(len2) |
1113 | z_off_t len2; |
1114 | { |
1115 | return crc32_combine_gen64((z_off64_t)len2); |
1116 | } |
1117 | |
1118 | /* ========================================================================= */ |
1119 | uLong ZEXPORT crc32_combine_op(crc1, crc2, op) |
1120 | uLong crc1; |
1121 | uLong crc2; |
1122 | uLong op; |
1123 | { |
1124 | return multmodp(op, crc1) ^ (crc2 & 0xffffffff); |
1125 | } |