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
4 *
5 * Thanks to Rodney Brown <rbrown64@csc.com.au> for his contribution of faster
6 * CRC methods: exclusive-oring 32 bits of data at a time, and pre-computing
7 * tables for updating the shift register in one step with three exclusive-ors
8 * instead of four steps with four exclusive-ors. This results in about a
9 * factor of two increase in speed on a Power PC G4 (PPC7455) using gcc -O3.
10 */
11
12/* @(#) $Id$ */
13
14#include "zbuild.h"
15#include "zendian.h"
16#include <inttypes.h>
17#include "deflate.h"
18#include "functable.h"
19#include "crc32_p.h"
20#include "crc32.h"
21
22
23/* Local functions for crc concatenation */
24static uint32_t crc32_combine_(uint32_t crc1, uint32_t crc2, z_off64_t len2);
25static void crc32_combine_gen_(uint32_t *op, z_off64_t len2);
26
27/* =========================================================================
28 * This function can be used by asm versions of crc32()
29 */
30const uint32_t * ZEXPORT PREFIX(get_crc_table)(void) {
31 return (const uint32_t *)crc_table;
32}
33
34uint32_t ZEXPORT PREFIX(crc32_z)(uint32_t crc, const unsigned char *buf, size_t len) {
35 if (buf == NULL) return 0;
36
37 return functable.crc32(crc, buf, len);
38}
39/* ========================================================================= */
40#define DO1 crc = crc_table[0][((int)crc ^ (*buf++)) & 0xff] ^ (crc >> 8)
41#define DO8 DO1; DO1; DO1; DO1; DO1; DO1; DO1; DO1
42#define DO4 DO1; DO1; DO1; DO1
43
44/* ========================================================================= */
45ZLIB_INTERNAL uint32_t crc32_generic(uint32_t crc, const unsigned char *buf, uint64_t len)
46{
47 crc = crc ^ 0xffffffff;
48
49#ifdef UNROLL_MORE
50 while (len >= 8) {
51 DO8;
52 len -= 8;
53 }
54#else
55 while (len >= 4) {
56 DO4;
57 len -= 4;
58 }
59#endif
60
61 if (len) do {
62 DO1;
63 } while (--len);
64 return crc ^ 0xffffffff;
65}
66
67#ifdef ZLIB_COMPAT
68unsigned long ZEXPORT PREFIX(crc32)(unsigned long crc, const unsigned char *buf, unsigned int len) {
69 return (unsigned long) PREFIX(crc32_z)((uint32_t) crc, buf, len);
70}
71#else
72uint32_t ZEXPORT PREFIX(crc32)(uint32_t crc, const unsigned char *buf, uint32_t len) {
73 return PREFIX(crc32_z)(crc, buf, len);
74}
75#endif
76
77/*
78 This BYFOUR code accesses the passed unsigned char * buffer with a 32-bit
79 integer pointer type. This violates the strict aliasing rule, where a
80 compiler can assume, for optimization purposes, that two pointers to
81 fundamentally different types won't ever point to the same memory. This can
82 manifest as a problem only if one of the pointers is written to. This code
83 only reads from those pointers. So long as this code remains isolated in
84 this compilation unit, there won't be a problem. For this reason, this code
85 should not be copied and pasted into a compilation unit in which other code
86 writes to the buffer that is passed to these routines.
87 */
88
89/* ========================================================================= */
90#if BYTE_ORDER == LITTLE_ENDIAN
91#define DOLIT4 c ^= *buf4++; \
92 c = crc_table[3][c & 0xff] ^ crc_table[2][(c >> 8) & 0xff] ^ \
93 crc_table[1][(c >> 16) & 0xff] ^ crc_table[0][c >> 24]
94#define DOLIT32 DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4
95
96/* ========================================================================= */
97ZLIB_INTERNAL uint32_t crc32_little(uint32_t crc, const unsigned char *buf, uint64_t len) {
98 register uint32_t c;
99 register const uint32_t *buf4;
100
101 c = crc;
102 c = ~c;
103 while (len && ((ptrdiff_t)buf & 3)) {
104 c = crc_table[0][(c ^ *buf++) & 0xff] ^ (c >> 8);
105 len--;
106 }
107
108 buf4 = (const uint32_t *)(const void *)buf;
109
110#ifdef UNROLL_MORE
111 while (len >= 32) {
112 DOLIT32;
113 len -= 32;
114 }
115#endif
116
117 while (len >= 4) {
118 DOLIT4;
119 len -= 4;
120 }
121 buf = (const unsigned char *)buf4;
122
123 if (len) do {
124 c = crc_table[0][(c ^ *buf++) & 0xff] ^ (c >> 8);
125 } while (--len);
126 c = ~c;
127 return c;
128}
129#endif /* BYTE_ORDER == LITTLE_ENDIAN */
130
131/* ========================================================================= */
132#if BYTE_ORDER == BIG_ENDIAN
133#define DOBIG4 c ^= *buf4++; \
134 c = crc_table[4][c & 0xff] ^ crc_table[5][(c >> 8) & 0xff] ^ \
135 crc_table[6][(c >> 16) & 0xff] ^ crc_table[7][c >> 24]
136#define DOBIG32 DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4
137
138/* ========================================================================= */
139ZLIB_INTERNAL uint32_t crc32_big(uint32_t crc, const unsigned char *buf, uint64_t len) {
140 register uint32_t c;
141 register const uint32_t *buf4;
142
143 c = ZSWAP32(crc);
144 c = ~c;
145 while (len && ((ptrdiff_t)buf & 3)) {
146 c = crc_table[4][(c >> 24) ^ *buf++] ^ (c << 8);
147 len--;
148 }
149
150 buf4 = (const uint32_t *)(const void *)buf;
151
152#ifdef UNROLL_MORE
153 while (len >= 32) {
154 DOBIG32;
155 len -= 32;
156 }
157#endif
158
159 while (len >= 4) {
160 DOBIG4;
161 len -= 4;
162 }
163 buf = (const unsigned char *)buf4;
164
165 if (len) do {
166 c = crc_table[4][(c >> 24) ^ *buf++] ^ (c << 8);
167 } while (--len);
168 c = ~c;
169 return ZSWAP32(c);
170}
171#endif /* BYTE_ORDER == BIG_ENDIAN */
172
173
174/* ========================================================================= */
175static uint32_t crc32_combine_(uint32_t crc1, uint32_t crc2, z_off64_t len2) {
176 int n;
177
178 if (len2 > 0)
179 /* operator for 2^n zeros repeats every GF2_DIM n values */
180 for (n = 0; len2; n = (n + 1) % GF2_DIM, len2 >>= 1)
181 if (len2 & 1)
182 crc1 = gf2_matrix_times(crc_comb[n], crc1);
183 return crc1 ^ crc2;
184}
185
186/* ========================================================================= */
187uint32_t ZEXPORT PREFIX(crc32_combine)(uint32_t crc1, uint32_t crc2, z_off_t len2) {
188 return crc32_combine_(crc1, crc2, len2);
189}
190
191uint32_t ZEXPORT PREFIX(crc32_combine64)(uint32_t crc1, uint32_t crc2, z_off64_t len2) {
192 return crc32_combine_(crc1, crc2, len2);
193}
194
195#ifdef X86_PCLMULQDQ_CRC
196#include "arch/x86/x86.h"
197#include "arch/x86/crc_folding.h"
198
199ZLIB_INTERNAL void crc_finalize(deflate_state *const s) {
200 if (x86_cpu_has_pclmulqdq)
201 s->strm->adler = crc_fold_512to32(s);
202}
203#endif
204
205ZLIB_INTERNAL void crc_reset(deflate_state *const s) {
206#ifdef X86_PCLMULQDQ_CRC
207 if (x86_cpu_has_pclmulqdq) {
208 crc_fold_init(s);
209 return;
210 }
211#endif
212 s->strm->adler = PREFIX(crc32)(0L, NULL, 0);
213}
214
215ZLIB_INTERNAL void copy_with_crc(PREFIX3(stream) *strm, unsigned char *dst, unsigned long size) {
216#ifdef X86_PCLMULQDQ_CRC
217 if (x86_cpu_has_pclmulqdq) {
218 crc_fold_copy(strm->state, dst, strm->next_in, size);
219 return;
220 }
221#endif
222 memcpy(dst, strm->next_in, size);
223 strm->adler = PREFIX(crc32)(strm->adler, dst, size);
224}
225
226/* ========================================================================= */
227static void crc32_combine_gen_(uint32_t *op, z_off64_t len2)
228{
229 uint32_t row;
230 int j;
231 unsigned i;
232
233 /* if len2 is zero or negative, return the identity matrix */
234 if (len2 <= 0) {
235 row = 1;
236 for (j = 0; j < GF2_DIM; j++) {
237 op[j] = row;
238 row <<= 1;
239 }
240 return;
241 }
242
243 /* at least one bit in len2 is set -- find it, and copy the operator
244 corresponding to that position into op */
245 i = 0;
246 for (;;) {
247 if (len2 & 1) {
248 for (j = 0; j < GF2_DIM; j++)
249 op[j] = crc_comb[i][j];
250 break;
251 }
252 len2 >>= 1;
253 i = (i + 1) % GF2_DIM;
254 }
255
256 /* for each remaining bit set in len2 (if any), multiply op by the operator
257 corresponding to that position */
258 for (;;) {
259 len2 >>= 1;
260 i = (i + 1) % GF2_DIM;
261 if (len2 == 0)
262 break;
263 if (len2 & 1)
264 for (j = 0; j < GF2_DIM; j++)
265 op[j] = gf2_matrix_times(crc_comb[i], op[j]);
266 }
267}
268
269/* ========================================================================= */
270void ZEXPORT PREFIX(crc32_combine_gen)(uint32_t *op, z_off_t len2)
271{
272 crc32_combine_gen_(op, len2);
273}
274
275void ZEXPORT PREFIX(crc32_combine_gen64)(uint32_t *op, z_off64_t len2)
276{
277 crc32_combine_gen_(op, len2);
278}
279
280/* ========================================================================= */
281uint32_t ZEXPORT PREFIX(crc32_combine_op)(uint32_t crc1, uint32_t crc2, const uint32_t *op)
282{
283 return gf2_matrix_times(op, crc1) ^ crc2;
284}
285