xref: /freebsd/sys/contrib/zlib/crc32.c (revision 069ac184)
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 #if defined(W) && (!defined(ARMCRC32) || defined(DYNAMIC_CRC_TABLE))
107 /*
108   Swap the bytes in a z_word_t to convert between little and big endian. Any
109   self-respecting compiler will optimize this to a single machine byte-swap
110   instruction, if one is available. This assumes that word_t is either 32 bits
111   or 64 bits.
112  */
113 local z_word_t byte_swap(z_word_t word) {
114 #  if W == 8
115     return
116         (word & 0xff00000000000000) >> 56 |
117         (word & 0xff000000000000) >> 40 |
118         (word & 0xff0000000000) >> 24 |
119         (word & 0xff00000000) >> 8 |
120         (word & 0xff000000) << 8 |
121         (word & 0xff0000) << 24 |
122         (word & 0xff00) << 40 |
123         (word & 0xff) << 56;
124 #  else   /* W == 4 */
125     return
126         (word & 0xff000000) >> 24 |
127         (word & 0xff0000) >> 8 |
128         (word & 0xff00) << 8 |
129         (word & 0xff) << 24;
130 #  endif
131 }
132 #endif
133 
134 #ifdef DYNAMIC_CRC_TABLE
135 /* =========================================================================
136  * Table of powers of x for combining CRC-32s, filled in by make_crc_table()
137  * below.
138  */
139    local z_crc_t FAR x2n_table[32];
140 #else
141 /* =========================================================================
142  * Tables for byte-wise and braided CRC-32 calculations, and a table of powers
143  * of x for combining CRC-32s, all made by make_crc_table().
144  */
145 #  include "crc32.h"
146 #endif
147 
148 /* CRC polynomial. */
149 #define POLY 0xedb88320         /* p(x) reflected, with x^32 implied */
150 
151 /*
152   Return a(x) multiplied by b(x) modulo p(x), where p(x) is the CRC polynomial,
153   reflected. For speed, this requires that a not be zero.
154  */
155 local z_crc_t multmodp(z_crc_t a, z_crc_t b) {
156     z_crc_t m, p;
157 
158     m = (z_crc_t)1 << 31;
159     p = 0;
160     for (;;) {
161         if (a & m) {
162             p ^= b;
163             if ((a & (m - 1)) == 0)
164                 break;
165         }
166         m >>= 1;
167         b = b & 1 ? (b >> 1) ^ POLY : b >> 1;
168     }
169     return p;
170 }
171 
172 /*
173   Return x^(n * 2^k) modulo p(x). Requires that x2n_table[] has been
174   initialized.
175  */
176 local z_crc_t x2nmodp(z_off64_t n, unsigned k) {
177     z_crc_t p;
178 
179     p = (z_crc_t)1 << 31;           /* x^0 == 1 */
180     while (n) {
181         if (n & 1)
182             p = multmodp(x2n_table[k & 31], p);
183         n >>= 1;
184         k++;
185     }
186     return p;
187 }
188 
189 #ifdef DYNAMIC_CRC_TABLE
190 /* =========================================================================
191  * Build the tables for byte-wise and braided CRC-32 calculations, and a table
192  * of powers of x for combining CRC-32s.
193  */
194 local z_crc_t FAR crc_table[256];
195 #ifdef W
196    local z_word_t FAR crc_big_table[256];
197    local z_crc_t FAR crc_braid_table[W][256];
198    local z_word_t FAR crc_braid_big_table[W][256];
199    local void braid(z_crc_t [][256], z_word_t [][256], int, int);
200 #endif
201 #ifdef MAKECRCH
202    local void write_table(FILE *, const z_crc_t FAR *, int);
203    local void write_table32hi(FILE *, const z_word_t FAR *, int);
204    local void write_table64(FILE *, const z_word_t FAR *, int);
205 #endif /* MAKECRCH */
206 
207 /*
208   Define a once() function depending on the availability of atomics. If this is
209   compiled with DYNAMIC_CRC_TABLE defined, and if CRCs will be computed in
210   multiple threads, and if atomics are not available, then get_crc_table() must
211   be called to initialize the tables and must return before any threads are
212   allowed to compute or combine CRCs.
213  */
214 
215 /* Definition of once functionality. */
216 typedef struct once_s once_t;
217 
218 /* Check for the availability of atomics. */
219 #if defined(__STDC__) && __STDC_VERSION__ >= 201112L && \
220     !defined(__STDC_NO_ATOMICS__)
221 
222 #include <stdatomic.h>
223 
224 /* Structure for once(), which must be initialized with ONCE_INIT. */
225 struct once_s {
226     atomic_flag begun;
227     atomic_int done;
228 };
229 #define ONCE_INIT {ATOMIC_FLAG_INIT, 0}
230 
231 /*
232   Run the provided init() function exactly once, even if multiple threads
233   invoke once() at the same time. The state must be a once_t initialized with
234   ONCE_INIT.
235  */
236 local void once(once_t *state, void (*init)(void)) {
237     if (!atomic_load(&state->done)) {
238         if (atomic_flag_test_and_set(&state->begun))
239             while (!atomic_load(&state->done))
240                 ;
241         else {
242             init();
243             atomic_store(&state->done, 1);
244         }
245     }
246 }
247 
248 #else   /* no atomics */
249 
250 /* Structure for once(), which must be initialized with ONCE_INIT. */
251 struct once_s {
252     volatile int begun;
253     volatile int done;
254 };
255 #define ONCE_INIT {0, 0}
256 
257 /* Test and set. Alas, not atomic, but tries to minimize the period of
258    vulnerability. */
259 local int test_and_set(int volatile *flag) {
260     int was;
261 
262     was = *flag;
263     *flag = 1;
264     return was;
265 }
266 
267 /* Run the provided init() function once. This is not thread-safe. */
268 local void once(once_t *state, void (*init)(void)) {
269     if (!state->done) {
270         if (test_and_set(&state->begun))
271             while (!state->done)
272                 ;
273         else {
274             init();
275             state->done = 1;
276         }
277     }
278 }
279 
280 #endif
281 
282 /* State for once(). */
283 local once_t made = ONCE_INIT;
284 
285 /*
286   Generate tables for a byte-wise 32-bit CRC calculation on the polynomial:
287   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.
288 
289   Polynomials over GF(2) are represented in binary, one bit per coefficient,
290   with the lowest powers in the most significant bit. Then adding polynomials
291   is just exclusive-or, and multiplying a polynomial by x is a right shift by
292   one. If we call the above polynomial p, and represent a byte as the
293   polynomial q, also with the lowest power in the most significant bit (so the
294   byte 0xb1 is the polynomial x^7+x^3+x^2+1), then the CRC is (q*x^32) mod p,
295   where a mod b means the remainder after dividing a by b.
296 
297   This calculation is done using the shift-register method of multiplying and
298   taking the remainder. The register is initialized to zero, and for each
299   incoming bit, x^32 is added mod p to the register if the bit is a one (where
300   x^32 mod p is p+x^32 = x^26+...+1), and the register is multiplied mod p by x
301   (which is shifting right by one and adding x^32 mod p if the bit shifted out
302   is a one). We start with the highest power (least significant bit) of q and
303   repeat for all eight bits of q.
304 
305   The table is simply the CRC of all possible eight bit values. This is all the
306   information needed to generate CRCs on data a byte at a time for all
307   combinations of CRC register values and incoming bytes.
308  */
309 
310 local void make_crc_table(void) {
311     unsigned i, j, n;
312     z_crc_t p;
313 
314     /* initialize the CRC of bytes tables */
315     for (i = 0; i < 256; i++) {
316         p = i;
317         for (j = 0; j < 8; j++)
318             p = p & 1 ? (p >> 1) ^ POLY : p >> 1;
319         crc_table[i] = p;
320 #ifdef W
321         crc_big_table[i] = byte_swap(p);
322 #endif
323     }
324 
325     /* initialize the x^2^n mod p(x) table */
326     p = (z_crc_t)1 << 30;         /* x^1 */
327     x2n_table[0] = p;
328     for (n = 1; n < 32; n++)
329         x2n_table[n] = p = multmodp(p, p);
330 
331 #ifdef W
332     /* initialize the braiding tables -- needs x2n_table[] */
333     braid(crc_braid_table, crc_braid_big_table, N, W);
334 #endif
335 
336 #ifdef MAKECRCH
337     {
338         /*
339           The crc32.h header file contains tables for both 32-bit and 64-bit
340           z_word_t's, and so requires a 64-bit type be available. In that case,
341           z_word_t must be defined to be 64-bits. This code then also generates
342           and writes out the tables for the case that z_word_t is 32 bits.
343          */
344 #if !defined(W) || W != 8
345 #  error Need a 64-bit integer type in order to generate crc32.h.
346 #endif
347         FILE *out;
348         int k, n;
349         z_crc_t ltl[8][256];
350         z_word_t big[8][256];
351 
352         out = fopen("crc32.h", "w");
353         if (out == NULL) return;
354 
355         /* write out little-endian CRC table to crc32.h */
356         fprintf(out,
357             "/* crc32.h -- tables for rapid CRC calculation\n"
358             " * Generated automatically by crc32.c\n */\n"
359             "\n"
360             "local const z_crc_t FAR crc_table[] = {\n"
361             "    ");
362         write_table(out, crc_table, 256);
363         fprintf(out,
364             "};\n");
365 
366         /* write out big-endian CRC table for 64-bit z_word_t to crc32.h */
367         fprintf(out,
368             "\n"
369             "#ifdef W\n"
370             "\n"
371             "#if W == 8\n"
372             "\n"
373             "local const z_word_t FAR crc_big_table[] = {\n"
374             "    ");
375         write_table64(out, crc_big_table, 256);
376         fprintf(out,
377             "};\n");
378 
379         /* write out big-endian CRC table for 32-bit z_word_t to crc32.h */
380         fprintf(out,
381             "\n"
382             "#else /* W == 4 */\n"
383             "\n"
384             "local const z_word_t FAR crc_big_table[] = {\n"
385             "    ");
386         write_table32hi(out, crc_big_table, 256);
387         fprintf(out,
388             "};\n"
389             "\n"
390             "#endif\n");
391 
392         /* write out braid tables for each value of N */
393         for (n = 1; n <= 6; n++) {
394             fprintf(out,
395             "\n"
396             "#if N == %d\n", n);
397 
398             /* compute braid tables for this N and 64-bit word_t */
399             braid(ltl, big, n, 8);
400 
401             /* write out braid tables for 64-bit z_word_t to crc32.h */
402             fprintf(out,
403             "\n"
404             "#if W == 8\n"
405             "\n"
406             "local const z_crc_t FAR crc_braid_table[][256] = {\n");
407             for (k = 0; k < 8; k++) {
408                 fprintf(out, "   {");
409                 write_table(out, ltl[k], 256);
410                 fprintf(out, "}%s", k < 7 ? ",\n" : "");
411             }
412             fprintf(out,
413             "};\n"
414             "\n"
415             "local const z_word_t FAR crc_braid_big_table[][256] = {\n");
416             for (k = 0; k < 8; k++) {
417                 fprintf(out, "   {");
418                 write_table64(out, big[k], 256);
419                 fprintf(out, "}%s", k < 7 ? ",\n" : "");
420             }
421             fprintf(out,
422             "};\n");
423 
424             /* compute braid tables for this N and 32-bit word_t */
425             braid(ltl, big, n, 4);
426 
427             /* write out braid tables for 32-bit z_word_t to crc32.h */
428             fprintf(out,
429             "\n"
430             "#else /* W == 4 */\n"
431             "\n"
432             "local const z_crc_t FAR crc_braid_table[][256] = {\n");
433             for (k = 0; k < 4; k++) {
434                 fprintf(out, "   {");
435                 write_table(out, ltl[k], 256);
436                 fprintf(out, "}%s", k < 3 ? ",\n" : "");
437             }
438             fprintf(out,
439             "};\n"
440             "\n"
441             "local const z_word_t FAR crc_braid_big_table[][256] = {\n");
442             for (k = 0; k < 4; k++) {
443                 fprintf(out, "   {");
444                 write_table32hi(out, big[k], 256);
445                 fprintf(out, "}%s", k < 3 ? ",\n" : "");
446             }
447             fprintf(out,
448             "};\n"
449             "\n"
450             "#endif\n"
451             "\n"
452             "#endif\n");
453         }
454         fprintf(out,
455             "\n"
456             "#endif\n");
457 
458         /* write out zeros operator table to crc32.h */
459         fprintf(out,
460             "\n"
461             "local const z_crc_t FAR x2n_table[] = {\n"
462             "    ");
463         write_table(out, x2n_table, 32);
464         fprintf(out,
465             "};\n");
466         fclose(out);
467     }
468 #endif /* MAKECRCH */
469 }
470 
471 #ifdef MAKECRCH
472 
473 /*
474    Write the 32-bit values in table[0..k-1] to out, five per line in
475    hexadecimal separated by commas.
476  */
477 local void write_table(FILE *out, const z_crc_t FAR *table, int k) {
478     int n;
479 
480     for (n = 0; n < k; n++)
481         fprintf(out, "%s0x%08lx%s", n == 0 || n % 5 ? "" : "    ",
482                 (unsigned long)(table[n]),
483                 n == k - 1 ? "" : (n % 5 == 4 ? ",\n" : ", "));
484 }
485 
486 /*
487    Write the high 32-bits of each value in table[0..k-1] to out, five per line
488    in hexadecimal separated by commas.
489  */
490 local void write_table32hi(FILE *out, const z_word_t FAR *table, int k) {
491     int n;
492 
493     for (n = 0; n < k; n++)
494         fprintf(out, "%s0x%08lx%s", n == 0 || n % 5 ? "" : "    ",
495                 (unsigned long)(table[n] >> 32),
496                 n == k - 1 ? "" : (n % 5 == 4 ? ",\n" : ", "));
497 }
498 
499 /*
500   Write the 64-bit values in table[0..k-1] to out, three per line in
501   hexadecimal separated by commas. This assumes that if there is a 64-bit
502   type, then there is also a long long integer type, and it is at least 64
503   bits. If not, then the type cast and format string can be adjusted
504   accordingly.
505  */
506 local void write_table64(FILE *out, const z_word_t FAR *table, int k) {
507     int n;
508 
509     for (n = 0; n < k; n++)
510         fprintf(out, "%s0x%016llx%s", n == 0 || n % 3 ? "" : "    ",
511                 (unsigned long long)(table[n]),
512                 n == k - 1 ? "" : (n % 3 == 2 ? ",\n" : ", "));
513 }
514 
515 /* Actually do the deed. */
516 int main(void) {
517     make_crc_table();
518     return 0;
519 }
520 
521 #endif /* MAKECRCH */
522 
523 #ifdef W
524 /*
525   Generate the little and big-endian braid tables for the given n and z_word_t
526   size w. Each array must have room for w blocks of 256 elements.
527  */
528 local void braid(z_crc_t ltl[][256], z_word_t big[][256], int n, int w) {
529     int k;
530     z_crc_t i, p, q;
531     for (k = 0; k < w; k++) {
532         p = x2nmodp((n * w + 3 - k) << 3, 0);
533         ltl[k][0] = 0;
534         big[w - 1 - k][0] = 0;
535         for (i = 1; i < 256; i++) {
536             ltl[k][i] = q = multmodp(i << 24, p);
537             big[w - 1 - k][i] = byte_swap(q);
538         }
539     }
540 }
541 #endif
542 
543 #endif /* DYNAMIC_CRC_TABLE */
544 
545 /* =========================================================================
546  * This function can be used by asm versions of crc32(), and to force the
547  * generation of the CRC tables in a threaded application.
548  */
549 const z_crc_t FAR * ZEXPORT get_crc_table(void) {
550 #ifdef DYNAMIC_CRC_TABLE
551     once(&made, make_crc_table);
552 #endif /* DYNAMIC_CRC_TABLE */
553     return (const z_crc_t FAR *)crc_table;
554 }
555 
556 /* =========================================================================
557  * Use ARM machine instructions if available. This will compute the CRC about
558  * ten times faster than the braided calculation. This code does not check for
559  * the presence of the CRC instruction at run time. __ARM_FEATURE_CRC32 will
560  * only be defined if the compilation specifies an ARM processor architecture
561  * that has the instructions. For example, compiling with -march=armv8.1-a or
562  * -march=armv8-a+crc, or -march=native if the compile machine has the crc32
563  * instructions.
564  */
565 #ifdef ARMCRC32
566 
567 /*
568    Constants empirically determined to maximize speed. These values are from
569    measurements on a Cortex-A57. Your mileage may vary.
570  */
571 #define Z_BATCH 3990                /* number of words in a batch */
572 #define Z_BATCH_ZEROS 0xa10d3d0c    /* computed from Z_BATCH = 3990 */
573 #define Z_BATCH_MIN 800             /* fewest words in a final batch */
574 
575 unsigned long ZEXPORT crc32_z(unsigned long crc, const unsigned char FAR *buf,
576                               z_size_t len) {
577     z_crc_t val;
578     z_word_t crc1, crc2;
579     const z_word_t *word;
580     z_word_t val0, val1, val2;
581     z_size_t last, last2, i;
582     z_size_t num;
583 
584     /* Return initial CRC, if requested. */
585     if (buf == Z_NULL) return 0;
586 
587 #ifdef DYNAMIC_CRC_TABLE
588     once(&made, make_crc_table);
589 #endif /* DYNAMIC_CRC_TABLE */
590 
591     /* Pre-condition the CRC */
592     crc = (~crc) & 0xffffffff;
593 
594     /* Compute the CRC up to a word boundary. */
595     while (len && ((z_size_t)buf & 7) != 0) {
596         len--;
597         val = *buf++;
598         __asm__ volatile("crc32b %w0, %w0, %w1" : "+r"(crc) : "r"(val));
599     }
600 
601     /* Prepare to compute the CRC on full 64-bit words word[0..num-1]. */
602     word = (z_word_t const *)buf;
603     num = len >> 3;
604     len &= 7;
605 
606     /* Do three interleaved CRCs to realize the throughput of one crc32x
607        instruction per cycle. Each CRC is calculated on Z_BATCH words. The
608        three CRCs are combined into a single CRC after each set of batches. */
609     while (num >= 3 * Z_BATCH) {
610         crc1 = 0;
611         crc2 = 0;
612         for (i = 0; i < Z_BATCH; i++) {
613             val0 = word[i];
614             val1 = word[i + Z_BATCH];
615             val2 = word[i + 2 * Z_BATCH];
616             __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0));
617             __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc1) : "r"(val1));
618             __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc2) : "r"(val2));
619         }
620         word += 3 * Z_BATCH;
621         num -= 3 * Z_BATCH;
622         crc = multmodp(Z_BATCH_ZEROS, crc) ^ crc1;
623         crc = multmodp(Z_BATCH_ZEROS, crc) ^ crc2;
624     }
625 
626     /* Do one last smaller batch with the remaining words, if there are enough
627        to pay for the combination of CRCs. */
628     last = num / 3;
629     if (last >= Z_BATCH_MIN) {
630         last2 = last << 1;
631         crc1 = 0;
632         crc2 = 0;
633         for (i = 0; i < last; i++) {
634             val0 = word[i];
635             val1 = word[i + last];
636             val2 = word[i + last2];
637             __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0));
638             __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc1) : "r"(val1));
639             __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc2) : "r"(val2));
640         }
641         word += 3 * last;
642         num -= 3 * last;
643         val = x2nmodp(last, 6);
644         crc = multmodp(val, crc) ^ crc1;
645         crc = multmodp(val, crc) ^ crc2;
646     }
647 
648     /* Compute the CRC on any remaining words. */
649     for (i = 0; i < num; i++) {
650         val0 = word[i];
651         __asm__ volatile("crc32x %w0, %w0, %x1" : "+r"(crc) : "r"(val0));
652     }
653     word += num;
654 
655     /* Complete the CRC on any remaining bytes. */
656     buf = (const unsigned char FAR *)word;
657     while (len) {
658         len--;
659         val = *buf++;
660         __asm__ volatile("crc32b %w0, %w0, %w1" : "+r"(crc) : "r"(val));
661     }
662 
663     /* Return the CRC, post-conditioned. */
664     return crc ^ 0xffffffff;
665 }
666 
667 #else
668 
669 #ifdef W
670 
671 /*
672   Return the CRC of the W bytes in the word_t data, taking the
673   least-significant byte of the word as the first byte of data, without any pre
674   or post conditioning. This is used to combine the CRCs of each braid.
675  */
676 local z_crc_t crc_word(z_word_t data) {
677     int k;
678     for (k = 0; k < W; k++)
679         data = (data >> 8) ^ crc_table[data & 0xff];
680     return (z_crc_t)data;
681 }
682 
683 local z_word_t crc_word_big(z_word_t data) {
684     int k;
685     for (k = 0; k < W; k++)
686         data = (data << 8) ^
687             crc_big_table[(data >> ((W - 1) << 3)) & 0xff];
688     return data;
689 }
690 
691 #endif
692 
693 /* ========================================================================= */
694 unsigned long ZEXPORT crc32_z(unsigned long crc, const unsigned char FAR *buf,
695                               z_size_t len) {
696     /* Return initial CRC, if requested. */
697     if (buf == Z_NULL) return 0;
698 
699 #ifdef DYNAMIC_CRC_TABLE
700     once(&made, make_crc_table);
701 #endif /* DYNAMIC_CRC_TABLE */
702 
703     /* Pre-condition the CRC */
704     crc = (~crc) & 0xffffffff;
705 
706 #ifdef W
707 
708     /* If provided enough bytes, do a braided CRC calculation. */
709     if (len >= N * W + W - 1) {
710         z_size_t blks;
711         z_word_t const *words;
712         unsigned endian;
713         int k;
714 
715         /* Compute the CRC up to a z_word_t boundary. */
716         while (len && ((z_size_t)buf & (W - 1)) != 0) {
717             len--;
718             crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
719         }
720 
721         /* Compute the CRC on as many N z_word_t blocks as are available. */
722         blks = len / (N * W);
723         len -= blks * N * W;
724         words = (z_word_t const *)buf;
725 
726         /* Do endian check at execution time instead of compile time, since ARM
727            processors can change the endianness at execution time. If the
728            compiler knows what the endianness will be, it can optimize out the
729            check and the unused branch. */
730         endian = 1;
731         if (*(unsigned char *)&endian) {
732             /* Little endian. */
733 
734             z_crc_t crc0;
735             z_word_t word0;
736 #if N > 1
737             z_crc_t crc1;
738             z_word_t word1;
739 #if N > 2
740             z_crc_t crc2;
741             z_word_t word2;
742 #if N > 3
743             z_crc_t crc3;
744             z_word_t word3;
745 #if N > 4
746             z_crc_t crc4;
747             z_word_t word4;
748 #if N > 5
749             z_crc_t crc5;
750             z_word_t word5;
751 #endif
752 #endif
753 #endif
754 #endif
755 #endif
756 
757             /* Initialize the CRC for each braid. */
758             crc0 = crc;
759 #if N > 1
760             crc1 = 0;
761 #if N > 2
762             crc2 = 0;
763 #if N > 3
764             crc3 = 0;
765 #if N > 4
766             crc4 = 0;
767 #if N > 5
768             crc5 = 0;
769 #endif
770 #endif
771 #endif
772 #endif
773 #endif
774 
775             /*
776               Process the first blks-1 blocks, computing the CRCs on each braid
777               independently.
778              */
779             while (--blks) {
780                 /* Load the word for each braid into registers. */
781                 word0 = crc0 ^ words[0];
782 #if N > 1
783                 word1 = crc1 ^ words[1];
784 #if N > 2
785                 word2 = crc2 ^ words[2];
786 #if N > 3
787                 word3 = crc3 ^ words[3];
788 #if N > 4
789                 word4 = crc4 ^ words[4];
790 #if N > 5
791                 word5 = crc5 ^ words[5];
792 #endif
793 #endif
794 #endif
795 #endif
796 #endif
797                 words += N;
798 
799                 /* Compute and update the CRC for each word. The loop should
800                    get unrolled. */
801                 crc0 = crc_braid_table[0][word0 & 0xff];
802 #if N > 1
803                 crc1 = crc_braid_table[0][word1 & 0xff];
804 #if N > 2
805                 crc2 = crc_braid_table[0][word2 & 0xff];
806 #if N > 3
807                 crc3 = crc_braid_table[0][word3 & 0xff];
808 #if N > 4
809                 crc4 = crc_braid_table[0][word4 & 0xff];
810 #if N > 5
811                 crc5 = crc_braid_table[0][word5 & 0xff];
812 #endif
813 #endif
814 #endif
815 #endif
816 #endif
817                 for (k = 1; k < W; k++) {
818                     crc0 ^= crc_braid_table[k][(word0 >> (k << 3)) & 0xff];
819 #if N > 1
820                     crc1 ^= crc_braid_table[k][(word1 >> (k << 3)) & 0xff];
821 #if N > 2
822                     crc2 ^= crc_braid_table[k][(word2 >> (k << 3)) & 0xff];
823 #if N > 3
824                     crc3 ^= crc_braid_table[k][(word3 >> (k << 3)) & 0xff];
825 #if N > 4
826                     crc4 ^= crc_braid_table[k][(word4 >> (k << 3)) & 0xff];
827 #if N > 5
828                     crc5 ^= crc_braid_table[k][(word5 >> (k << 3)) & 0xff];
829 #endif
830 #endif
831 #endif
832 #endif
833 #endif
834                 }
835             }
836 
837             /*
838               Process the last block, combining the CRCs of the N braids at the
839               same time.
840              */
841             crc = crc_word(crc0 ^ words[0]);
842 #if N > 1
843             crc = crc_word(crc1 ^ words[1] ^ crc);
844 #if N > 2
845             crc = crc_word(crc2 ^ words[2] ^ crc);
846 #if N > 3
847             crc = crc_word(crc3 ^ words[3] ^ crc);
848 #if N > 4
849             crc = crc_word(crc4 ^ words[4] ^ crc);
850 #if N > 5
851             crc = crc_word(crc5 ^ words[5] ^ crc);
852 #endif
853 #endif
854 #endif
855 #endif
856 #endif
857             words += N;
858         }
859         else {
860             /* Big endian. */
861 
862             z_word_t crc0, word0, comb;
863 #if N > 1
864             z_word_t crc1, word1;
865 #if N > 2
866             z_word_t crc2, word2;
867 #if N > 3
868             z_word_t crc3, word3;
869 #if N > 4
870             z_word_t crc4, word4;
871 #if N > 5
872             z_word_t crc5, word5;
873 #endif
874 #endif
875 #endif
876 #endif
877 #endif
878 
879             /* Initialize the CRC for each braid. */
880             crc0 = byte_swap(crc);
881 #if N > 1
882             crc1 = 0;
883 #if N > 2
884             crc2 = 0;
885 #if N > 3
886             crc3 = 0;
887 #if N > 4
888             crc4 = 0;
889 #if N > 5
890             crc5 = 0;
891 #endif
892 #endif
893 #endif
894 #endif
895 #endif
896 
897             /*
898               Process the first blks-1 blocks, computing the CRCs on each braid
899               independently.
900              */
901             while (--blks) {
902                 /* Load the word for each braid into registers. */
903                 word0 = crc0 ^ words[0];
904 #if N > 1
905                 word1 = crc1 ^ words[1];
906 #if N > 2
907                 word2 = crc2 ^ words[2];
908 #if N > 3
909                 word3 = crc3 ^ words[3];
910 #if N > 4
911                 word4 = crc4 ^ words[4];
912 #if N > 5
913                 word5 = crc5 ^ words[5];
914 #endif
915 #endif
916 #endif
917 #endif
918 #endif
919                 words += N;
920 
921                 /* Compute and update the CRC for each word. The loop should
922                    get unrolled. */
923                 crc0 = crc_braid_big_table[0][word0 & 0xff];
924 #if N > 1
925                 crc1 = crc_braid_big_table[0][word1 & 0xff];
926 #if N > 2
927                 crc2 = crc_braid_big_table[0][word2 & 0xff];
928 #if N > 3
929                 crc3 = crc_braid_big_table[0][word3 & 0xff];
930 #if N > 4
931                 crc4 = crc_braid_big_table[0][word4 & 0xff];
932 #if N > 5
933                 crc5 = crc_braid_big_table[0][word5 & 0xff];
934 #endif
935 #endif
936 #endif
937 #endif
938 #endif
939                 for (k = 1; k < W; k++) {
940                     crc0 ^= crc_braid_big_table[k][(word0 >> (k << 3)) & 0xff];
941 #if N > 1
942                     crc1 ^= crc_braid_big_table[k][(word1 >> (k << 3)) & 0xff];
943 #if N > 2
944                     crc2 ^= crc_braid_big_table[k][(word2 >> (k << 3)) & 0xff];
945 #if N > 3
946                     crc3 ^= crc_braid_big_table[k][(word3 >> (k << 3)) & 0xff];
947 #if N > 4
948                     crc4 ^= crc_braid_big_table[k][(word4 >> (k << 3)) & 0xff];
949 #if N > 5
950                     crc5 ^= crc_braid_big_table[k][(word5 >> (k << 3)) & 0xff];
951 #endif
952 #endif
953 #endif
954 #endif
955 #endif
956                 }
957             }
958 
959             /*
960               Process the last block, combining the CRCs of the N braids at the
961               same time.
962              */
963             comb = crc_word_big(crc0 ^ words[0]);
964 #if N > 1
965             comb = crc_word_big(crc1 ^ words[1] ^ comb);
966 #if N > 2
967             comb = crc_word_big(crc2 ^ words[2] ^ comb);
968 #if N > 3
969             comb = crc_word_big(crc3 ^ words[3] ^ comb);
970 #if N > 4
971             comb = crc_word_big(crc4 ^ words[4] ^ comb);
972 #if N > 5
973             comb = crc_word_big(crc5 ^ words[5] ^ comb);
974 #endif
975 #endif
976 #endif
977 #endif
978 #endif
979             words += N;
980             crc = byte_swap(comb);
981         }
982 
983         /*
984           Update the pointer to the remaining bytes to process.
985          */
986         buf = (unsigned char const *)words;
987     }
988 
989 #endif /* W */
990 
991     /* Complete the computation of the CRC on any remaining bytes. */
992     while (len >= 8) {
993         len -= 8;
994         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
995         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
996         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
997         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
998         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
999         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1000         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1001         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1002     }
1003     while (len) {
1004         len--;
1005         crc = (crc >> 8) ^ crc_table[(crc ^ *buf++) & 0xff];
1006     }
1007 
1008     /* Return the CRC, post-conditioned. */
1009     return crc ^ 0xffffffff;
1010 }
1011 
1012 #endif
1013 
1014 /* ========================================================================= */
1015 unsigned long ZEXPORT crc32(unsigned long crc, const unsigned char FAR *buf,
1016                             uInt len) {
1017     return crc32_z(crc, buf, len);
1018 }
1019 
1020 /* ========================================================================= */
1021 uLong ZEXPORT crc32_combine64(uLong crc1, uLong crc2, z_off64_t len2) {
1022 #ifdef DYNAMIC_CRC_TABLE
1023     once(&made, make_crc_table);
1024 #endif /* DYNAMIC_CRC_TABLE */
1025     return multmodp(x2nmodp(len2, 3), crc1) ^ (crc2 & 0xffffffff);
1026 }
1027 
1028 /* ========================================================================= */
1029 uLong ZEXPORT crc32_combine(uLong crc1, uLong crc2, z_off_t len2) {
1030     return crc32_combine64(crc1, crc2, (z_off64_t)len2);
1031 }
1032 
1033 /* ========================================================================= */
1034 uLong ZEXPORT crc32_combine_gen64(z_off64_t len2) {
1035 #ifdef DYNAMIC_CRC_TABLE
1036     once(&made, make_crc_table);
1037 #endif /* DYNAMIC_CRC_TABLE */
1038     return x2nmodp(len2, 3);
1039 }
1040 
1041 /* ========================================================================= */
1042 uLong ZEXPORT crc32_combine_gen(z_off_t len2) {
1043     return crc32_combine_gen64((z_off64_t)len2);
1044 }
1045 
1046 /* ========================================================================= */
1047 uLong ZEXPORT crc32_combine_op(uLong crc1, uLong crc2, uLong op) {
1048     return multmodp(op, crc1) ^ (crc2 & 0xffffffff);
1049 }
1050