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 z_crc_t FAR crc_table[256];
116 local z_crc_t FAR x2n_table[32];
117 local void make_crc_table OF((void));
119 local z_word_t FAR crc_big_table[256];
120 local z_crc_t FAR crc_braid_table[W][256];
121 local z_word_t FAR crc_braid_big_table[W][256];
122 local void braid OF((z_crc_t [][256], z_word_t [][256], int, int));
125 local void write_table OF((FILE *, const z_crc_t FAR *, int));
126 local void write_table32hi OF((FILE *, const z_word_t FAR *, int));
127 local void write_table64 OF((FILE *, const z_word_t FAR *, int));
128 #endif /* MAKECRCH */
131 Define a once() function depending on the availability of atomics. If this is
132 compiled with DYNAMIC_CRC_TABLE defined, and if CRCs will be computed in
133 multiple threads, and if atomics are not available, then get_crc_table() must
134 be called to initialize the tables and must return before any threads are
135 allowed to compute or combine CRCs.
138 /* Definition of once functionality. */
139 typedef struct once_s once_t;
140 local void once OF((once_t *, void (*)(void)));
142 /* Check for the availability of atomics. */
143 #if defined(__STDC__) && __STDC_VERSION__ >= 201112L && \
144 !defined(__STDC_NO_ATOMICS__)
146 #include <stdatomic.h>
148 /* Structure for once(), which must be initialized with ONCE_INIT. */
153 #define ONCE_INIT {ATOMIC_FLAG_INIT, 0}
156 Run the provided init() function exactly once, even if multiple threads
157 invoke once() at the same time. The state must be a once_t initialized with
160 local void once(state, init)
164 if (!atomic_load(&state->done)) {
165 if (atomic_flag_test_and_set(&state->begun))
166 while (!atomic_load(&state->done))
170 atomic_store(&state->done, 1);
175 #else /* no atomics */
177 /* Structure for once(), which must be initialized with ONCE_INIT. */
182 #define ONCE_INIT {0, 0}
184 /* Test and set. Alas, not atomic, but tries to minimize the period of
186 local int test_and_set OF((int volatile *));
187 local int test_and_set(flag)
197 /* Run the provided init() function once. This is not thread-safe. */
198 local void once(state, init)
203 if (test_and_set(&state->begun))
215 /* State for once(). */
216 local once_t made = ONCE_INIT;
219 Generate tables for a byte-wise 32-bit CRC calculation on the polynomial:
220 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.
222 Polynomials over GF(2) are represented in binary, one bit per coefficient,
223 with the lowest powers in the most significant bit. Then adding polynomials
224 is just exclusive-or, and multiplying a polynomial by x is a right shift by
225 one. If we call the above polynomial p, and represent a byte as the
226 polynomial q, also with the lowest power in the most significant bit (so the
227 byte 0xb1 is the polynomial x^7+x^3+x^2+1), then the CRC is (q*x^32) mod p,
228 where a mod b means the remainder after dividing a by b.
230 This calculation is done using the shift-register method of multiplying and
231 taking the remainder. The register is initialized to zero, and for each
232 incoming bit, x^32 is added mod p to the register if the bit is a one (where
233 x^32 mod p is p+x^32 = x^26+...+1), and the register is multiplied mod p by x
234 (which is shifting right by one and adding x^32 mod p if the bit shifted out
235 is a one). We start with the highest power (least significant bit) of q and
236 repeat for all eight bits of q.
238 The table is simply the CRC of all possible eight bit values. This is all the
239 information needed to generate CRCs on data a byte at a time for all
240 combinations of CRC register values and incoming bytes.
243 local void make_crc_table()
248 /* initialize the CRC of bytes tables */
249 for (i = 0; i < 256; i++) {
251 for (j = 0; j < 8; j++)
252 p = p & 1 ? (p >> 1) ^ POLY : p >> 1;
255 crc_big_table[i] = byte_swap(p);
259 /* initialize the x^2^n mod p(x) table */
260 p = (z_crc_t)1 << 30; /* x^1 */
262 for (n = 1; n < 32; n++)
263 x2n_table[n] = p = multmodp(p, p);
266 /* initialize the braiding tables -- needs x2n_table[] */
267 braid(crc_braid_table, crc_braid_big_table, N, W);
273 The crc32.h header file contains tables for both 32-bit and 64-bit
274 z_word_t's, and so requires a 64-bit type be available. In that case,
275 z_word_t must be defined to be 64-bits. This code then also generates
276 and writes out the tables for the case that z_word_t is 32 bits.
278 #if !defined(W) || W != 8
279 # error Need a 64-bit integer type in order to generate crc32.h.
284 z_word_t big[8][256];
286 out = fopen("crc32.h", "w");
287 if (out == NULL) return;
289 /* write out little-endian CRC table to crc32.h */
291 "/* crc32.h -- tables for rapid CRC calculation\n"
292 " * Generated automatically by crc32.c\n */\n"
294 "local const z_crc_t FAR crc_table[] = {\n"
296 write_table(out, crc_table, 256);
300 /* write out big-endian CRC table for 64-bit z_word_t to crc32.h */
307 "local const z_word_t FAR crc_big_table[] = {\n"
309 write_table64(out, crc_big_table, 256);
313 /* write out big-endian CRC table for 32-bit z_word_t to crc32.h */
316 "#else /* W == 4 */\n"
318 "local const z_word_t FAR crc_big_table[] = {\n"
320 write_table32hi(out, crc_big_table, 256);
326 /* write out braid tables for each value of N */
327 for (n = 1; n <= 6; n++) {
332 /* compute braid tables for this N and 64-bit word_t */
333 braid(ltl, big, n, 8);
335 /* write out braid tables for 64-bit z_word_t to crc32.h */
340 "local const z_crc_t FAR crc_braid_table[][256] = {\n");
341 for (k = 0; k < 8; k++) {
343 write_table(out, ltl[k], 256);
344 fprintf(out, "}%s", k < 7 ? ",\n" : "");
349 "local const z_word_t FAR crc_braid_big_table[][256] = {\n");
350 for (k = 0; k < 8; k++) {
352 write_table64(out, big[k], 256);
353 fprintf(out, "}%s", k < 7 ? ",\n" : "");
358 /* compute braid tables for this N and 32-bit word_t */
359 braid(ltl, big, n, 4);
361 /* write out braid tables for 32-bit z_word_t to crc32.h */
364 "#else /* W == 4 */\n"
366 "local const z_crc_t FAR crc_braid_table[][256] = {\n");
367 for (k = 0; k < 4; k++) {
369 write_table(out, ltl[k], 256);
370 fprintf(out, "}%s", k < 3 ? ",\n" : "");
375 "local const z_word_t FAR crc_braid_big_table[][256] = {\n");
376 for (k = 0; k < 4; k++) {
378 write_table32hi(out, big[k], 256);
379 fprintf(out, "}%s", k < 3 ? ",\n" : "");
392 /* write out zeros operator table to crc32.h */
395 "local const z_crc_t FAR x2n_table[] = {\n"
397 write_table(out, x2n_table, 32);
402 #endif /* MAKECRCH */
408 Write the 32-bit values in table[0..k-1] to out, five per line in
409 hexadecimal separated by commas.
411 local void write_table(out, table, k)
413 const z_crc_t FAR *table;
418 for (n = 0; n < k; n++)
419 fprintf(out, "%s0x%08lx%s", n == 0 || n % 5 ? "" : " ",
420 (unsigned long)(table[n]),
421 n == k - 1 ? "" : (n % 5 == 4 ? ",\n" : ", "));
425 Write the high 32-bits of each value in table[0..k-1] to out, five per line
426 in hexadecimal separated by commas.
428 local void write_table32hi(out, table, k)
430 const z_word_t FAR *table;
435 for (n = 0; n < k; n++)
436 fprintf(out, "%s0x%08lx%s", n == 0 || n % 5 ? "" : " ",
437 (unsigned long)(table[n] >> 32),
438 n == k - 1 ? "" : (n % 5 == 4 ? ",\n" : ", "));
442 Write the 64-bit values in table[0..k-1] to out, three per line in
443 hexadecimal separated by commas. This assumes that if there is a 64-bit
444 type, then there is also a long long integer type, and it is at least 64
445 bits. If not, then the type cast and format string can be adjusted
448 local void write_table64(out, table, k)
450 const z_word_t FAR *table;
455 for (n = 0; n < k; n++)
456 fprintf(out, "%s0x%016llx%s", n == 0 || n % 3 ? "" : " ",
457 (unsigned long long)(table[n]),
458 n == k - 1 ? "" : (n % 3 == 2 ? ",\n" : ", "));
461 /* Actually do the deed. */
468 #endif /* MAKECRCH */
472 Generate the little and big-endian braid tables for the given n and z_word_t
473 size w. Each array must have room for w blocks of 256 elements.
475 local void braid(ltl, big, n, w)
483 for (k = 0; k < w; k++) {
484 p = x2nmodp((n * w + 3 - k) << 3, 0);
486 big[w - 1 - k][0] = 0;
487 for (i = 1; i < 256; i++) {
488 ltl[k][i] = q = multmodp(i << 24, p);
489 big[w - 1 - k][i] = byte_swap(q);
495 #else /* !DYNAMIC_CRC_TABLE */
496 /* ========================================================================
497 * Tables for byte-wise and braided CRC-32 calculations, and a table of powers
498 * of x for combining CRC-32s, all made by make_crc_table().
501 #endif /* DYNAMIC_CRC_TABLE */
503 /* ========================================================================
504 * Routines used for CRC calculation. Some are also required for the table
509 Return a(x) multiplied by b(x) modulo p(x), where p(x) is the CRC polynomial,
510 reflected. For speed, this requires that a not be zero.
512 local z_crc_t multmodp(a, b)
518 m = (z_crc_t)1 << 31;
523 if ((a & (m - 1)) == 0)
527 b = b & 1 ? (b >> 1) ^ POLY : b >> 1;
533 Return x^(n * 2^k) modulo p(x). Requires that x2n_table[] has been
536 local z_crc_t x2nmodp(n, k)
542 p = (z_crc_t)1 << 31; /* x^0 == 1 */
545 p = multmodp(x2n_table[k & 31], p);
555 Swap the bytes in a z_word_t to convert between little and big endian. Any
556 self-respecting compiler will optimize this to a single machine byte-swap
557 instruction, if one is available. This assumes that word_t is either 32 bits
560 local z_word_t byte_swap(word)
565 (word & 0xff00000000000000) >> 56 |
566 (word & 0xff000000000000) >> 40 |
567 (word & 0xff0000000000) >> 24 |
568 (word & 0xff00000000) >> 8 |
569 (word & 0xff000000) << 8 |
570 (word & 0xff0000) << 24 |
571 (word & 0xff00) << 40 |
575 (word & 0xff000000) >> 24 |
576 (word & 0xff0000) >> 8 |
577 (word & 0xff00) << 8 |
583 Return the CRC of the W bytes in the word_t data, taking the
584 least-significant byte of the word as the first byte of data, without any pre
585 or post conditioning. This is used to combine the CRCs of each braid.
587 local z_crc_t crc_word(data)
591 for (k = 0; k < W; k++)
592 data = (data >> 8) ^ crc_table[data & 0xff];
593 return (z_crc_t)data;
596 local z_word_t crc_word_big(data)
600 for (k = 0; k < W; k++)
602 crc_big_table[(data >> ((W - 1) << 3)) & 0xff];
608 /* =========================================================================
609 * This function can be used by asm versions of crc32(), and to force the
610 * generation of the CRC tables in a threaded application.
612 const z_crc_t FAR * ZEXPORT get_crc_table()
614 #ifdef DYNAMIC_CRC_TABLE
615 once(&made, make_crc_table);
616 #endif /* DYNAMIC_CRC_TABLE */
617 return (const z_crc_t FAR *)crc_table;
620 /* =========================================================================
621 * Use ARM machine instructions if available. This will compute the CRC about
622 * ten times faster than the braided calculation. This code does not check for
623 * the presence of the CRC instruction at run time. __ARM_FEATURE_CRC32 will
624 * only be defined if the compilation specifies an ARM processor architecture
625 * that has the instructions. For example, compiling with -march=armv8.1-a or
626 * -march=armv8-a+crc, or -march=native if the compile machine has the crc32
629 #if defined(__aarch64__) && defined(__ARM_FEATURE_CRC32) && W == 8
632 Constants empirically determined to maximize speed. These values are from
633 measurements on a Cortex-A57. Your mileage may vary.
635 #define Z_BATCH 3990 /* number of words in a batch */
636 #define Z_BATCH_ZEROS 0xa10d3d0c /* computed from Z_BATCH = 3990 */
637 #define Z_BATCH_MIN 800 /* fewest words in a final batch */
639 unsigned long ZEXPORT crc32_z(crc, buf, len)
641 const unsigned char FAR *buf;
646 const z_word_t *word;
647 z_word_t val0, val1, val2;
648 z_size_t last, last2, i;
651 /* Return initial CRC, if requested. */
652 if (buf == Z_NULL) return 0;
654 #ifdef DYNAMIC_CRC_TABLE
655 once(&made, make_crc_table);
656 #endif /* DYNAMIC_CRC_TABLE */
658 /* Pre-condition the CRC */
661 /* Compute the CRC up to a word boundary. */
662 while (len && ((z_size_t)buf & 7) != 0) {
665 __asm__ volatile("crc32b %w0, %w0, %w1" : "+r"(crc) : "r"(val));
668 /* Prepare to compute the CRC on full 64-bit words word[0..num-1]. */
669 word = (z_word_t const *)buf;
673 /* Do three interleaved CRCs to realize the throughput of one crc32x
674 instruction per cycle. Each CRC is calcuated on Z_BATCH words. The three
675 CRCs are combined into a single CRC after each set of batches. */
676 while (num >= 3 * Z_BATCH) {
679 for (i = 0; i < Z_BATCH; i++) {
681 val1 = word[i + Z_BATCH];
682 val2 = word[i + 2 * Z_BATCH];
683 __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0));
684 __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc1) : "r"(val1));
685 __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc2) : "r"(val2));
689 crc = multmodp(Z_BATCH_ZEROS, crc) ^ crc1;
690 crc = multmodp(Z_BATCH_ZEROS, crc) ^ crc2;
693 /* Do one last smaller batch with the remaining words, if there are enough
694 to pay for the combination of CRCs. */
696 if (last >= Z_BATCH_MIN) {
700 for (i = 0; i < last; i++) {
702 val1 = word[i + last];
703 val2 = word[i + last2];
704 __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0));
705 __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc1) : "r"(val1));
706 __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc2) : "r"(val2));
710 val = x2nmodp(last, 6);
711 crc = multmodp(val, crc) ^ crc1;
712 crc = multmodp(val, crc) ^ crc2;
715 /* Compute the CRC on any remaining words. */
716 for (i = 0; i < num; i++) {
718 __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0));
722 /* Complete the CRC on any remaining bytes. */
723 buf = (const unsigned char FAR *)word;
727 __asm__ volatile("crc32b %w0, %w0, %w1" : "+r"(crc) : "r"(val));
730 /* Return the CRC, post-conditioned. */
731 return crc ^ 0xffffffff;
736 /* ========================================================================= */
737 unsigned long ZEXPORT crc32_z(crc, buf, len)
739 const unsigned char FAR *buf;
742 /* Return initial CRC, if requested. */
743 if (buf == Z_NULL) return 0;
745 #ifdef DYNAMIC_CRC_TABLE
746 once(&made, make_crc_table);
747 #endif /* DYNAMIC_CRC_TABLE */
749 /* Pre-condition the CRC */
754 /* If provided enough bytes, do a braided CRC calculation. */
755 if (len >= N * W + W - 1) {
757 z_word_t const *words;
761 /* Compute the CRC up to a z_word_t boundary. */
762 while (len && ((z_size_t)buf & (W - 1)) != 0) {
764 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
767 /* Compute the CRC on as many N z_word_t blocks as are available. */
768 blks = len / (N * W);
770 words = (z_word_t const *)buf;
772 /* Do endian check at execution time instead of compile time, since ARM
773 processors can change the endianess at execution time. If the
774 compiler knows what the endianess will be, it can optimize out the
775 check and the unused branch. */
777 if (*(unsigned char *)&endian) {
803 /* Initialize the CRC for each braid. */
822 Process the first blks-1 blocks, computing the CRCs on each braid
826 /* Load the word for each braid into registers. */
827 word0 = crc0 ^ words[0];
829 word1 = crc1 ^ words[1];
831 word2 = crc2 ^ words[2];
833 word3 = crc3 ^ words[3];
835 word4 = crc4 ^ words[4];
837 word5 = crc5 ^ words[5];
845 /* Compute and update the CRC for each word. The loop should
847 crc0 = crc_braid_table[0][word0 & 0xff];
849 crc1 = crc_braid_table[0][word1 & 0xff];
851 crc2 = crc_braid_table[0][word2 & 0xff];
853 crc3 = crc_braid_table[0][word3 & 0xff];
855 crc4 = crc_braid_table[0][word4 & 0xff];
857 crc5 = crc_braid_table[0][word5 & 0xff];
863 for (k = 1; k < W; k++) {
864 crc0 ^= crc_braid_table[k][(word0 >> (k << 3)) & 0xff];
866 crc1 ^= crc_braid_table[k][(word1 >> (k << 3)) & 0xff];
868 crc2 ^= crc_braid_table[k][(word2 >> (k << 3)) & 0xff];
870 crc3 ^= crc_braid_table[k][(word3 >> (k << 3)) & 0xff];
872 crc4 ^= crc_braid_table[k][(word4 >> (k << 3)) & 0xff];
874 crc5 ^= crc_braid_table[k][(word5 >> (k << 3)) & 0xff];
884 Process the last block, combining the CRCs of the N braids at the
887 crc = crc_word(crc0 ^ words[0]);
889 crc = crc_word(crc1 ^ words[1] ^ crc);
891 crc = crc_word(crc2 ^ words[2] ^ crc);
893 crc = crc_word(crc3 ^ words[3] ^ crc);
895 crc = crc_word(crc4 ^ words[4] ^ crc);
897 crc = crc_word(crc5 ^ words[5] ^ crc);
908 z_word_t crc0, word0, comb;
910 z_word_t crc1, word1;
912 z_word_t crc2, word2;
914 z_word_t crc3, word3;
916 z_word_t crc4, word4;
918 z_word_t crc5, word5;
925 /* Initialize the CRC for each braid. */
926 crc0 = byte_swap(crc);
944 Process the first blks-1 blocks, computing the CRCs on each braid
948 /* Load the word for each braid into registers. */
949 word0 = crc0 ^ words[0];
951 word1 = crc1 ^ words[1];
953 word2 = crc2 ^ words[2];
955 word3 = crc3 ^ words[3];
957 word4 = crc4 ^ words[4];
959 word5 = crc5 ^ words[5];
967 /* Compute and update the CRC for each word. The loop should
969 crc0 = crc_braid_big_table[0][word0 & 0xff];
971 crc1 = crc_braid_big_table[0][word1 & 0xff];
973 crc2 = crc_braid_big_table[0][word2 & 0xff];
975 crc3 = crc_braid_big_table[0][word3 & 0xff];
977 crc4 = crc_braid_big_table[0][word4 & 0xff];
979 crc5 = crc_braid_big_table[0][word5 & 0xff];
985 for (k = 1; k < W; k++) {
986 crc0 ^= crc_braid_big_table[k][(word0 >> (k << 3)) & 0xff];
988 crc1 ^= crc_braid_big_table[k][(word1 >> (k << 3)) & 0xff];
990 crc2 ^= crc_braid_big_table[k][(word2 >> (k << 3)) & 0xff];
992 crc3 ^= crc_braid_big_table[k][(word3 >> (k << 3)) & 0xff];
994 crc4 ^= crc_braid_big_table[k][(word4 >> (k << 3)) & 0xff];
996 crc5 ^= crc_braid_big_table[k][(word5 >> (k << 3)) & 0xff];
1006 Process the last block, combining the CRCs of the N braids at the
1009 comb = crc_word_big(crc0 ^ words[0]);
1011 comb = crc_word_big(crc1 ^ words[1] ^ comb);
1013 comb = crc_word_big(crc2 ^ words[2] ^ comb);
1015 comb = crc_word_big(crc3 ^ words[3] ^ comb);
1017 comb = crc_word_big(crc4 ^ words[4] ^ comb);
1019 comb = crc_word_big(crc5 ^ words[5] ^ comb);
1026 crc = byte_swap(comb);
1030 Update the pointer to the remaining bytes to process.
1032 buf = (unsigned char const *)words;
1037 /* Complete the computation of the CRC on any remaining bytes. */
1040 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1041 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1042 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1043 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1044 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1045 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1046 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1047 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1051 crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1054 /* Return the CRC, post-conditioned. */
1055 return crc ^ 0xffffffff;
1060 /* ========================================================================= */
1061 unsigned long ZEXPORT crc32(crc, buf, len)
1063 const unsigned char FAR *buf;
1066 return crc32_z(crc, buf, len);
1069 /* ========================================================================= */
1070 uLong ZEXPORT crc32_combine64(crc1, crc2, len2)
1075 #ifdef DYNAMIC_CRC_TABLE
1076 once(&made, make_crc_table);
1077 #endif /* DYNAMIC_CRC_TABLE */
1078 return multmodp(x2nmodp(len2, 3), crc1) ^ crc2;
1081 /* ========================================================================= */
1082 uLong ZEXPORT crc32_combine(crc1, crc2, len2)
1087 return crc32_combine64(crc1, crc2, len2);
1090 /* ========================================================================= */
1091 uLong ZEXPORT crc32_combine_gen64(len2)
1094 #ifdef DYNAMIC_CRC_TABLE
1095 once(&made, make_crc_table);
1096 #endif /* DYNAMIC_CRC_TABLE */
1097 return x2nmodp(len2, 3);
1100 /* ========================================================================= */
1101 uLong ZEXPORT crc32_combine_gen(len2)
1104 return crc32_combine_gen64(len2);
1107 /* ========================================================================= */
1108 uLong crc32_combine_op(crc1, crc2, op)
1113 return multmodp(op, crc1) ^ crc2;