1 /* crc32.c -- compute the CRC-32 of a data stream
2 * Copyright (C) 1995-2006, 2010, 2011, 2012, 2016, 2018 Mark Adler
3 * For conditions of distribution and use, see copyright notice in zlib.h
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.
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().
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.
25 # ifndef DYNAMIC_CRC_TABLE
26 # define DYNAMIC_CRC_TABLE
27 # endif /* !DYNAMIC_CRC_TABLE */
30 #include "zutil.h" /* for Z_U4, Z_U8, z_crc_t, and FAR definitions */
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.
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.
60 # error N must be in 1..6
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.
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
80 # define W 8 /* required for MAKECRCH */
82 # if defined(__x86_64__) || defined(__aarch64__)
90 # if W == 8 && defined(Z_U8)
91 typedef Z_U8 z_word_t;
95 typedef Z_U4 z_word_t;
101 /* Local functions. */
102 local z_crc_t multmodp OF((z_crc_t a, z_crc_t b));
103 local z_crc_t x2nmodp OF((z_off64_t n, unsigned k));
105 local z_word_t byte_swap OF((z_word_t word));
106 local z_crc_t crc_word OF((z_word_t data));
107 local z_word_t crc_word_big OF((z_word_t data));
110 /* CRC polynomial. */
111 #define POLY 0xedb88320 /* p(x) reflected, with x^32 implied */
113 #ifdef DYNAMIC_CRC_TABLE
115 local volatile int crc_table_empty = 1;
116 local z_crc_t FAR crc_table[256];
117 local z_crc_t FAR x2n_table[32];
118 local void make_crc_table OF((void));
120 local z_word_t FAR crc_big_table[256];
121 local z_crc_t FAR crc_braid_table[W][256];
122 local z_word_t FAR crc_braid_big_table[W][256];
123 local void braid OF((z_crc_t [][256], z_word_t [][256], int, int));
126 local void write_table OF((FILE *, const z_crc_t FAR *, int));
127 local void write_table32hi OF((FILE *, const z_word_t FAR *, int));
128 local void write_table64 OF((FILE *, const z_word_t FAR *, int));
129 #endif /* MAKECRCH */
132 Generate tables for a byte-wise 32-bit CRC calculation on the polynomial:
133 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.
135 Polynomials over GF(2) are represented in binary, one bit per coefficient,
136 with the lowest powers in the most significant bit. Then adding polynomials
137 is just exclusive-or, and multiplying a polynomial by x is a right shift by
138 one. If we call the above polynomial p, and represent a byte as the
139 polynomial q, also with the lowest power in the most significant bit (so the
140 byte 0xb1 is the polynomial x^7+x^3+x+1), then the CRC is (q*x^32) mod p,
141 where a mod b means the remainder after dividing a by b.
143 This calculation is done using the shift-register method of multiplying and
144 taking the remainder. The register is initialized to zero, and for each
145 incoming bit, x^32 is added mod p to the register if the bit is a one (where
146 x^32 mod p is p+x^32 = x^26+...+1), and the register is multiplied mod p by x
147 (which is shifting right by one and adding x^32 mod p if the bit shifted out
148 is a one). We start with the highest power (least significant bit) of q and
149 repeat for all eight bits of q.
151 The table is simply the CRC of all possible eight bit values. This is all the
152 information needed to generate CRCs on data a byte at a time for all
153 combinations of CRC register values and incoming bytes.
156 local void make_crc_table()
159 static volatile int first = 1; /* flag to limit concurrent making */
161 /* See if another task is already doing this (not thread-safe, but better
162 than nothing -- significantly reduces duration of vulnerability in
163 case the advice about DYNAMIC_CRC_TABLE is ignored) */
168 /* initialize the CRC of bytes tables */
169 for (i = 0; i < 256; i++) {
171 for (j = 0; j < 8; j++)
172 p = p & 1 ? (p >> 1) ^ POLY : p >> 1;
175 crc_big_table[i] = byte_swap(p);
179 /* initialize the x^2^n mod p(x) table */
180 p = (z_crc_t)1 << 30; /* x^1 */
182 for (n = 1; n < 32; n++)
183 x2n_table[n] = p = multmodp(p, p);
185 /* initialize the braiding tables -- needs x2n_table[] */
186 braid(crc_braid_table, crc_braid_big_table, N, W);
189 /* mark tables as complete, in case someone else is waiting */
192 else { /* not first */
193 /* wait for the other guy to finish (not efficient, but rare) */
194 while (crc_table_empty)
200 The crc32.h header file contains tables for both 32-bit and 64-bit
201 z_word_t's, and so requires a 64-bit type be available. In that case,
202 z_word_t must be defined to be 64-bits. This code then also generates
203 and writes out the tables for the case that z_word_t is 32 bits.
205 #if !defined(W) || W != 8
206 # error Need a 64-bit integer type in order to generate crc32.h.
211 z_word_t big[8][256];
213 out = fopen("crc32.h", "w");
214 if (out == NULL) return;
216 /* write out little-endian CRC table to crc32.h */
218 "/* crc32.h -- tables for rapid CRC calculation\n"
219 " * Generated automatically by crc32.c\n */\n"
221 "local const z_crc_t FAR crc_table[] = {\n"
223 write_table(out, crc_table, 256);
227 /* write out big-endian CRC table for 64-bit z_word_t to crc32.h */
234 "local const z_word_t FAR crc_big_table[] = {\n"
236 write_table64(out, crc_big_table, 256);
240 /* write out big-endian CRC table for 32-bit z_word_t to crc32.h */
243 "#else /* W == 4 */\n"
245 "local const z_word_t FAR crc_big_table[] = {\n"
247 write_table32hi(out, crc_big_table, 256);
253 /* write out braid tables for each value of N */
254 for (n = 1; n <= 6; n++) {
259 /* compute braid tables for this N and 64-bit word_t */
260 braid(ltl, big, n, 8);
262 /* write out braid tables for 64-bit z_word_t to crc32.h */
267 "local const z_crc_t FAR crc_braid_table[][256] = {\n");
268 for (k = 0; k < 8; k++) {
270 write_table(out, ltl[k], 256);
271 fprintf(out, "}%s", k < 7 ? ",\n" : "");
276 "local const z_word_t FAR crc_braid_big_table[][256] = {\n");
277 for (k = 0; k < 8; k++) {
279 write_table64(out, big[k], 256);
280 fprintf(out, "}%s", k < 7 ? ",\n" : "");
285 /* compute braid tables for this N and 32-bit word_t */
286 braid(ltl, big, n, 4);
288 /* write out braid tables for 32-bit z_word_t to crc32.h */
291 "#else /* W == 4 */\n"
293 "local const z_crc_t FAR crc_braid_table[][256] = {\n");
294 for (k = 0; k < 4; k++) {
296 write_table(out, ltl[k], 256);
297 fprintf(out, "}%s", k < 3 ? ",\n" : "");
302 "local const z_word_t FAR crc_braid_big_table[][256] = {\n");
303 for (k = 0; k < 4; k++) {
305 write_table32hi(out, big[k], 256);
306 fprintf(out, "}%s", k < 3 ? ",\n" : "");
319 /* write out zeros operator table to crc32.h */
322 "local const z_crc_t FAR x2n_table[] = {\n"
324 write_table(out, x2n_table, 32);
329 #endif /* MAKECRCH */
335 Write the 32-bit values in table[0..k-1] to out, five per line in
336 hexadecimal separated by commas.
338 local void write_table(out, table, k)
340 const z_crc_t FAR *table;
345 for (n = 0; n < k; n++)
346 fprintf(out, "%s0x%08lx%s", n == 0 || n % 5 ? "" : " ",
347 (unsigned long)(table[n]),
348 n == k - 1 ? "" : (n % 5 == 4 ? ",\n" : ", "));
352 Write the high 32-bits of each value in table[0..k-1] to out, five per line
353 in hexadecimal separated by commas.
355 local void write_table32hi(out, table, k)
357 const z_word_t FAR *table;
362 for (n = 0; n < k; n++)
363 fprintf(out, "%s0x%08lx%s", n == 0 || n % 5 ? "" : " ",
364 (unsigned long)(table[n] >> 32),
365 n == k - 1 ? "" : (n % 5 == 4 ? ",\n" : ", "));
369 Write the 64-bit values in table[0..k-1] to out, three per line in
370 hexadecimal separated by commas. This assumes that if there is a 64-bit
371 type, then there is also a long long integer type, and it is at least 64
372 bits. If not, then the type cast and format string can be adjusted
375 local void write_table64(out, table, k)
377 const z_word_t FAR *table;
382 for (n = 0; n < k; n++)
383 fprintf(out, "%s0x%016llx%s", n == 0 || n % 3 ? "" : " ",
384 (unsigned long long)(table[n]),
385 n == k - 1 ? "" : (n % 3 == 2 ? ",\n" : ", "));
388 /* Actually do the deed. */
395 #endif /* MAKECRCH */
399 Generate the little and big-endian braid tables for the given n and z_word_t
400 size w. Each array must have room for w blocks of 256 elements.
402 local void braid(ltl, big, n, w)
410 for (k = 0; k < w; k++) {
411 p = x2nmodp((n * w + 3 - k) << 3, 0);
413 big[w - 1 - k][0] = 0;
414 for (i = 1; i < 256; i++) {
415 ltl[k][i] = q = multmodp(i << 24, p);
416 big[w - 1 - k][i] = byte_swap(q);
422 #else /* !DYNAMIC_CRC_TABLE */
423 /* ========================================================================
424 * Tables for byte-wise and braided CRC-32 calculations, and a table of powers
425 * of x for combining CRC-32s, all made by make_crc_table().
428 #endif /* DYNAMIC_CRC_TABLE */
430 /* ========================================================================
431 * Routines used for CRC calculation. Some are also required for the table
436 Return a(x) multiplied by b(x) modulo p(x), where p(x) is the CRC polynomial,
437 reflected. For speed, this requires that a not be zero.
439 local z_crc_t multmodp(a, b)
445 m = (z_crc_t)1 << 31;
450 if ((a & (m - 1)) == 0)
454 b = b & 1 ? (b >> 1) ^ POLY : b >> 1;
460 Return x^(n+k) modulo p(x). Requires that x2n_table[] has been initialized.
462 local z_crc_t x2nmodp(n, k)
468 p = (z_crc_t)1 << 31; /* x^0 == 1 */
471 p = multmodp(x2n_table[k & 31], p);
481 Swap the bytes in a z_word_t to convert between little and big endian. Any
482 self-respecting compiler will optimize this to a single machine byte-swap
483 instruction, if one is available. This assumes that word_t is either 32 bits
486 local z_word_t byte_swap(word)
491 (word & 0xff00000000000000) >> 56 |
492 (word & 0xff000000000000) >> 40 |
493 (word & 0xff0000000000) >> 24 |
494 (word & 0xff00000000) >> 8 |
495 (word & 0xff000000) << 8 |
496 (word & 0xff0000) << 24 |
497 (word & 0xff00) << 40 |
501 (word & 0xff000000) >> 24 |
502 (word & 0xff0000) >> 8 |
503 (word & 0xff00) << 8 |
509 Return the CRC of the W bytes in the word_t data, taking the
510 least-significant byte of the word as the first byte of data, without any pre
511 or post conditioning. This is used to combine the CRCs of each braid.
513 local z_crc_t crc_word(data)
517 for (k = 0; k < W; k++)
518 data = (data >> 8) ^ crc_table[data & 0xff];
519 return (z_crc_t)data;
522 local z_word_t crc_word_big(data)
526 for (k = 0; k < W; k++)
528 crc_big_table[(data >> ((W - 1) << 3)) & 0xff];
534 /* =========================================================================
535 * This function can be used by asm versions of crc32()
537 const z_crc_t FAR * ZEXPORT get_crc_table()
539 #ifdef DYNAMIC_CRC_TABLE
542 #endif /* DYNAMIC_CRC_TABLE */
543 return (const z_crc_t FAR *)crc_table;
546 /* ========================================================================= */
547 unsigned long ZEXPORT crc32_z(crc, buf, len)
549 const unsigned char FAR *buf;
552 /* Return initial CRC, if requested. */
553 if (buf == Z_NULL) return 0;
555 #ifdef DYNAMIC_CRC_TABLE
558 #endif /* DYNAMIC_CRC_TABLE */
560 /* Pre-condition the CRC */
565 /* If provided enough bytes, do a braided CRC calculation. */
566 if (len >= N * W + W - 1) {
568 z_word_t const *words;
572 /* Compute the CRC up to a z_word_t boundary. */
573 while (len && ((z_size_t)buf & (W - 1)) != 0) {
575 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
578 /* Compute the CRC on as many N z_word_t blocks as are available. */
579 blks = len / (N * W);
581 words = (z_word_t const *)buf;
583 /* Do endian check at execution time instead of compile time, since ARM
584 processors can change the endianess at execution time. If the
585 compiler knows what the endianess will be, it can optimize out the
586 check and the unused branch. */
588 if (*(unsigned char *)&endian) {
614 /* Initialize the CRC for each braid. */
633 Process the first blks-1 blocks, computing the CRCs on each braid
637 /* Load the word for each braid into registers. */
638 word0 = crc0 ^ words[0];
640 word1 = crc1 ^ words[1];
642 word2 = crc2 ^ words[2];
644 word3 = crc3 ^ words[3];
646 word4 = crc4 ^ words[4];
648 word5 = crc5 ^ words[5];
656 /* Compute and update the CRC for each word. The loop should
658 crc0 = crc_braid_table[0][word0 & 0xff];
660 crc1 = crc_braid_table[0][word1 & 0xff];
662 crc2 = crc_braid_table[0][word2 & 0xff];
664 crc3 = crc_braid_table[0][word3 & 0xff];
666 crc4 = crc_braid_table[0][word4 & 0xff];
668 crc5 = crc_braid_table[0][word5 & 0xff];
674 for (k = 1; k < W; k++) {
675 crc0 ^= crc_braid_table[k][(word0 >> (k << 3)) & 0xff];
677 crc1 ^= crc_braid_table[k][(word1 >> (k << 3)) & 0xff];
679 crc2 ^= crc_braid_table[k][(word2 >> (k << 3)) & 0xff];
681 crc3 ^= crc_braid_table[k][(word3 >> (k << 3)) & 0xff];
683 crc4 ^= crc_braid_table[k][(word4 >> (k << 3)) & 0xff];
685 crc5 ^= crc_braid_table[k][(word5 >> (k << 3)) & 0xff];
695 Process the last block, combining the CRCs of the N braids at the
698 crc = crc_word(crc0 ^ words[0]);
700 crc = crc_word(crc1 ^ words[1] ^ crc);
702 crc = crc_word(crc2 ^ words[2] ^ crc);
704 crc = crc_word(crc3 ^ words[3] ^ crc);
706 crc = crc_word(crc4 ^ words[4] ^ crc);
708 crc = crc_word(crc5 ^ words[5] ^ crc);
719 z_word_t crc0, word0, comb;
721 z_word_t crc1, word1;
723 z_word_t crc2, word2;
725 z_word_t crc3, word3;
727 z_word_t crc4, word4;
729 z_word_t crc5, word5;
736 /* Initialize the CRC for each braid. */
737 crc0 = byte_swap(crc);
755 Process the first blks-1 blocks, computing the CRCs on each braid
759 /* Load the word for each braid into registers. */
760 word0 = crc0 ^ words[0];
762 word1 = crc1 ^ words[1];
764 word2 = crc2 ^ words[2];
766 word3 = crc3 ^ words[3];
768 word4 = crc4 ^ words[4];
770 word5 = crc5 ^ words[5];
778 /* Compute and update the CRC for each word. The loop should
780 crc0 = crc_braid_big_table[0][word0 & 0xff];
782 crc1 = crc_braid_big_table[0][word1 & 0xff];
784 crc2 = crc_braid_big_table[0][word2 & 0xff];
786 crc3 = crc_braid_big_table[0][word3 & 0xff];
788 crc4 = crc_braid_big_table[0][word4 & 0xff];
790 crc5 = crc_braid_big_table[0][word5 & 0xff];
796 for (k = 1; k < W; k++) {
797 crc0 ^= crc_braid_big_table[k][(word0 >> (k << 3)) & 0xff];
799 crc1 ^= crc_braid_big_table[k][(word1 >> (k << 3)) & 0xff];
801 crc2 ^= crc_braid_big_table[k][(word2 >> (k << 3)) & 0xff];
803 crc3 ^= crc_braid_big_table[k][(word3 >> (k << 3)) & 0xff];
805 crc4 ^= crc_braid_big_table[k][(word4 >> (k << 3)) & 0xff];
807 crc5 ^= crc_braid_big_table[k][(word5 >> (k << 3)) & 0xff];
817 Process the last block, combining the CRCs of the N braids at the
820 comb = crc_word_big(crc0 ^ words[0]);
822 comb = crc_word_big(crc1 ^ words[1] ^ comb);
824 comb = crc_word_big(crc2 ^ words[2] ^ comb);
826 comb = crc_word_big(crc3 ^ words[3] ^ comb);
828 comb = crc_word_big(crc4 ^ words[4] ^ comb);
830 comb = crc_word_big(crc5 ^ words[5] ^ comb);
837 crc = byte_swap(comb);
841 Update the pointer to the remaining bytes to process.
843 buf = (unsigned char const *)words;
848 /* Complete the computation of the CRC on any remaining bytes. */
851 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
852 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
853 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
854 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
855 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
856 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
857 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
858 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
862 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
865 /* Return the CRC, post-conditioned. */
866 return crc ^ 0xffffffff;
869 /* ========================================================================= */
870 unsigned long ZEXPORT crc32(crc, buf, len)
872 const unsigned char FAR *buf;
875 return crc32_z(crc, buf, len);
878 /* ========================================================================= */
879 uLong ZEXPORT crc32_combine64(crc1, crc2, len2)
884 #ifdef DYNAMIC_CRC_TABLE
887 #endif /* DYNAMIC_CRC_TABLE */
888 return multmodp(x2nmodp(len2, 3), crc1) ^ crc2;
891 /* ========================================================================= */
892 uLong ZEXPORT crc32_combine(crc1, crc2, len2)
897 return crc32_combine64(crc1, crc2, len2);
900 /* ========================================================================= */
901 uLong ZEXPORT crc32_combine_gen64(len2)
904 #ifdef DYNAMIC_CRC_TABLE
907 #endif /* DYNAMIC_CRC_TABLE */
908 return x2nmodp(len2, 3);
911 /* ========================================================================= */
912 uLong ZEXPORT crc32_combine_gen(len2)
915 return crc32_combine_gen64(len2);
918 /* ========================================================================= */
919 uLong crc32_combine_op(crc1, crc2, op)
924 return multmodp(op, crc1) ^ crc2;