1 | /***************************************************************************** |
2 | |
3 | Copyright (c) 2009, 2010 Facebook, Inc. All Rights Reserved. |
4 | Copyright (c) 2011, 2015, Oracle and/or its affiliates. All Rights Reserved. |
5 | Copyright (c) 2016, MariaDB Corporation. |
6 | |
7 | This program is free software; you can redistribute it and/or modify it under |
8 | the terms of the GNU General Public License as published by the Free Software |
9 | Foundation; version 2 of the License. |
10 | |
11 | This program is distributed in the hope that it will be useful, but WITHOUT |
12 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS |
13 | FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
14 | |
15 | You should have received a copy of the GNU General Public License along with |
16 | this program; if not, write to the Free Software Foundation, Inc., |
17 | 51 Franklin Street, Suite 500, Boston, MA 02110-1335 USA |
18 | |
19 | *****************************************************************************/ |
20 | |
21 | /***************************************************************//** |
22 | @file ut/ut0crc32.cc |
23 | CRC32 implementation from Facebook, based on the zlib implementation. |
24 | |
25 | Created Aug 8, 2011, Vasil Dimov, based on mysys/my_crc32.c and |
26 | mysys/my_perf.c, contributed by Facebook under the following license. |
27 | ********************************************************************/ |
28 | |
29 | /* Copyright (C) 2009-2010 Facebook, Inc. All Rights Reserved. |
30 | |
31 | Dual licensed under BSD license and GPLv2. |
32 | |
33 | Redistribution and use in source and binary forms, with or without |
34 | modification, are permitted provided that the following conditions are met: |
35 | 1. Redistributions of source code must retain the above copyright notice, |
36 | this list of conditions and the following disclaimer. |
37 | 2. Redistributions in binary form must reproduce the above copyright notice, |
38 | this list of conditions and the following disclaimer in the documentation |
39 | and/or other materials provided with the distribution. |
40 | |
41 | THIS SOFTWARE IS PROVIDED BY FACEBOOK, INC. ``AS IS'' AND ANY EXPRESS OR |
42 | IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF |
43 | MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
44 | EVENT SHALL FACEBOOK, INC. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
45 | SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
46 | PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
47 | OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
48 | WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR |
49 | OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF |
50 | ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
51 | |
52 | This program is free software; you can redistribute it and/or modify it |
53 | under the terms of the GNU General Public License as published by the Free |
54 | Software Foundation; version 2 of the License. |
55 | |
56 | This program is distributed in the hope that it will be useful, but WITHOUT |
57 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
58 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
59 | more details. |
60 | |
61 | You should have received a copy of the GNU General Public License along with |
62 | this program; if not, write to the Free Software Foundation, Inc., |
63 | 51 Franklin Street, Suite 500, Boston, MA 02110-1335 USA */ |
64 | |
65 | /* The below CRC32 implementation is based on the implementation included with |
66 | * zlib with modifications to process 8 bytes at a time and using SSE 4.2 |
67 | * extensions when available. The polynomial constant has been changed to |
68 | * match the one used by SSE 4.2 and does not return the same value as the |
69 | * version used by zlib. The original zlib copyright notice follows. */ |
70 | |
71 | /* crc32.c -- compute the CRC-32 of a buf stream |
72 | * Copyright (C) 1995-2005 Mark Adler |
73 | * For conditions of distribution and use, see copyright notice in zlib.h |
74 | * |
75 | * Thanks to Rodney Brown <rbrown64@csc.com.au> for his contribution of faster |
76 | * CRC methods: exclusive-oring 32 bits of buf at a time, and pre-computing |
77 | * tables for updating the shift register in one step with three exclusive-ors |
78 | * instead of four steps with four exclusive-ors. This results in about a |
79 | * factor of two increase in speed on a Power PC G4 (PPC7455) using gcc -O3. |
80 | */ |
81 | |
82 | // First include (the generated) my_config.h, to get correct platform defines. |
83 | #include "my_config.h" |
84 | #include <string.h> |
85 | |
86 | #include "univ.i" |
87 | #include "ut0crc32.h" |
88 | |
89 | #ifdef _MSC_VER |
90 | #include <intrin.h> |
91 | #endif |
92 | |
93 | /** Swap the byte order of an 8 byte integer. |
94 | @param[in] i 8-byte integer |
95 | @return 8-byte integer */ |
96 | inline |
97 | uint64_t |
98 | ut_crc32_swap_byteorder( |
99 | uint64_t i) |
100 | { |
101 | return(i << 56 |
102 | | (i & 0x000000000000FF00ULL) << 40 |
103 | | (i & 0x0000000000FF0000ULL) << 24 |
104 | | (i & 0x00000000FF000000ULL) << 8 |
105 | | (i & 0x000000FF00000000ULL) >> 8 |
106 | | (i & 0x0000FF0000000000ULL) >> 24 |
107 | | (i & 0x00FF000000000000ULL) >> 40 |
108 | | i >> 56); |
109 | } |
110 | |
111 | /* CRC32 hardware implementation. */ |
112 | |
113 | #ifdef HAVE_CRC32_VPMSUM |
114 | extern "C" { |
115 | unsigned int crc32c_vpmsum(unsigned int crc, const unsigned char *p, unsigned long len); |
116 | }; |
117 | UNIV_INLINE |
118 | ib_uint32_t |
119 | ut_crc32_power8( |
120 | /*===========*/ |
121 | const byte* buf, /*!< in: data over which to calculate CRC32 */ |
122 | ulint len) /*!< in: data length */ |
123 | { |
124 | return crc32c_vpmsum(0, buf, len); |
125 | } |
126 | |
127 | ut_crc32_func_t ut_crc32 = ut_crc32_power8; |
128 | const char* ut_crc32_implementation = "Using POWER8 crc32 instructions" ; |
129 | #else |
130 | uint32_t ut_crc32_sw(const byte* buf, ulint len); |
131 | ut_crc32_func_t ut_crc32 = ut_crc32_sw; |
132 | const char* ut_crc32_implementation = "Using generic crc32 instructions" ; |
133 | #endif |
134 | |
135 | #if (defined(__GNUC__) && defined(__x86_64__)) || defined(_MSC_VER) |
136 | /********************************************************************//** |
137 | Fetches CPU info */ |
138 | static |
139 | void |
140 | ut_cpuid( |
141 | /*=====*/ |
142 | uint32_t vend[3], /*!< out: CPU vendor */ |
143 | uint32_t* model, /*!< out: CPU model */ |
144 | uint32_t* family, /*!< out: CPU family */ |
145 | uint32_t* stepping, /*!< out: CPU stepping */ |
146 | uint32_t* features_ecx, /*!< out: CPU features ecx */ |
147 | uint32_t* features_edx) /*!< out: CPU features edx */ |
148 | { |
149 | uint32_t sig; |
150 | #ifdef _MSC_VER |
151 | int data[4]; |
152 | __cpuid(data, 0); |
153 | /* ebx */ |
154 | vend[0] = data[1]; |
155 | /* edx */ |
156 | vend[1] = data[3]; |
157 | /* ecx */ |
158 | vend[2] = data[2]; |
159 | |
160 | __cpuid(data, 1); |
161 | /* eax */ |
162 | sig = data[0]; |
163 | /* ecx */ |
164 | *features_ecx = data[2]; |
165 | /* edx */ |
166 | *features_edx = data[3]; |
167 | #else |
168 | asm("cpuid" : "=b" (vend[0]), "=c" (vend[2]), "=d" (vend[1]) : "a" (0)); |
169 | asm("cpuid" : "=a" (sig), "=c" (*features_ecx), "=d" (*features_edx) |
170 | : "a" (1) |
171 | : "ebx" ); |
172 | #endif |
173 | |
174 | *model = ((sig >> 4) & 0xF); |
175 | *family = ((sig >> 8) & 0xF); |
176 | *stepping = (sig & 0xF); |
177 | |
178 | if (memcmp(vend, "GenuineIntel" , 12) == 0 |
179 | || (memcmp(vend, "AuthenticAMD" , 12) == 0 && *family == 0xF)) { |
180 | |
181 | *model += (((sig >> 16) & 0xF) << 4); |
182 | *family += ((sig >> 20) & 0xFF); |
183 | } |
184 | } |
185 | |
186 | /** Calculate CRC32 over 8-bit data using a hardware/CPU instruction. |
187 | @param[in,out] crc crc32 checksum so far when this function is called, |
188 | when the function ends it will contain the new checksum |
189 | @param[in,out] data data to be checksummed, the pointer will be advanced |
190 | with 1 byte |
191 | @param[in,out] len remaining bytes, it will be decremented with 1 */ |
192 | inline |
193 | void |
194 | ut_crc32_8_hw( |
195 | uint32_t* crc, |
196 | const byte** data, |
197 | ulint* len) |
198 | { |
199 | #ifdef _MSC_VER |
200 | *crc = _mm_crc32_u8(*crc, (*data)[0]); |
201 | #else |
202 | asm("crc32b %1, %0" |
203 | /* output operands */ |
204 | : "+r" (*crc) |
205 | /* input operands */ |
206 | : "rm" ((*data)[0])); |
207 | #endif |
208 | |
209 | (*data)++; |
210 | (*len)--; |
211 | } |
212 | |
213 | /** Calculate CRC32 over a 64-bit integer using a hardware/CPU instruction. |
214 | @param[in] crc crc32 checksum so far |
215 | @param[in] data data to be checksummed |
216 | @return resulting checksum of crc + crc(data) */ |
217 | inline |
218 | uint32_t |
219 | ut_crc32_64_low_hw( |
220 | uint32_t crc, |
221 | uint64_t data) |
222 | { |
223 | uint64_t crc_64bit = crc; |
224 | #ifdef _MSC_VER |
225 | #ifdef _M_X64 |
226 | crc_64bit = _mm_crc32_u64(crc_64bit, data); |
227 | #elif defined(_M_IX86) |
228 | crc = _mm_crc32_u32(crc, static_cast<uint32_t>(data)); |
229 | crc_64bit = _mm_crc32_u32(crc, static_cast<uint32_t>(data >> 32)); |
230 | #else |
231 | #error Not Supported processors type. |
232 | #endif |
233 | #else |
234 | asm("crc32q %1, %0" |
235 | /* output operands */ |
236 | : "+r" (crc_64bit) |
237 | /* input operands */ |
238 | : "rm" (data)); |
239 | #endif |
240 | |
241 | return(static_cast<uint32_t>(crc_64bit)); |
242 | } |
243 | |
244 | /** Calculate CRC32 over 64-bit byte string using a hardware/CPU instruction. |
245 | @param[in,out] crc crc32 checksum so far when this function is called, |
246 | when the function ends it will contain the new checksum |
247 | @param[in,out] data data to be checksummed, the pointer will be advanced |
248 | with 8 bytes |
249 | @param[in,out] len remaining bytes, it will be decremented with 8 */ |
250 | inline |
251 | void |
252 | ut_crc32_64_hw( |
253 | uint32_t* crc, |
254 | const byte** data, |
255 | ulint* len) |
256 | { |
257 | uint64_t data_int = *reinterpret_cast<const uint64_t*>(*data); |
258 | |
259 | #ifdef WORDS_BIGENDIAN |
260 | /* Currently we only support x86_64 (little endian) CPUs. In case |
261 | some big endian CPU supports a CRC32 instruction, then maybe we will |
262 | need a byte order swap here. */ |
263 | #error Dont know how to handle big endian CPUs |
264 | /* |
265 | data_int = ut_crc32_swap_byteorder(data_int); |
266 | */ |
267 | #endif /* WORDS_BIGENDIAN */ |
268 | |
269 | *crc = ut_crc32_64_low_hw(*crc, data_int); |
270 | |
271 | *data += 8; |
272 | *len -= 8; |
273 | } |
274 | |
275 | /** Calculates CRC32 using hardware/CPU instructions. |
276 | @param[in] buf data over which to calculate CRC32 |
277 | @param[in] len data length |
278 | @return CRC-32C (polynomial 0x11EDC6F41) */ |
279 | uint32_t |
280 | ut_crc32_hw( |
281 | const byte* buf, |
282 | ulint len) |
283 | { |
284 | uint32_t crc = 0xFFFFFFFFU; |
285 | |
286 | /* Calculate byte-by-byte up to an 8-byte aligned address. After |
287 | this consume the input 8-bytes at a time. */ |
288 | while (len > 0 && (reinterpret_cast<uintptr_t>(buf) & 7) != 0) { |
289 | ut_crc32_8_hw(&crc, &buf, &len); |
290 | } |
291 | |
292 | /* Perf testing |
293 | ./unittest/gunit/innodb/merge_innodb_tests-t --gtest_filter=ut0crc32.perf |
294 | on CPU "Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz" |
295 | with different N in "while (len >= N) {" shows: |
296 | N=16 |
297 | 2.867254 sec |
298 | 2.866860 sec |
299 | 2.867973 sec |
300 | |
301 | N=32 |
302 | 2.715725 sec |
303 | 2.713008 sec |
304 | 2.712520 sec |
305 | (5.36% speedup over N=16) |
306 | |
307 | N=64 |
308 | 2.634140 sec |
309 | 2.636558 sec |
310 | 2.636488 sec |
311 | (2.88% speedup over N=32) |
312 | |
313 | N=128 |
314 | 2.599534 sec |
315 | 2.599919 sec |
316 | 2.598035 sec |
317 | (1.39% speedup over N=64) |
318 | |
319 | N=256 |
320 | 2.576993 sec |
321 | 2.576748 sec |
322 | 2.575700 sec |
323 | (0.87% speedup over N=128) |
324 | |
325 | N=512 |
326 | 2.693928 sec |
327 | 2.691663 sec |
328 | 2.692142 sec |
329 | (4.51% slowdown over N=256) |
330 | */ |
331 | while (len >= 128) { |
332 | /* This call is repeated 16 times. 16 * 8 = 128. */ |
333 | ut_crc32_64_hw(&crc, &buf, &len); |
334 | ut_crc32_64_hw(&crc, &buf, &len); |
335 | ut_crc32_64_hw(&crc, &buf, &len); |
336 | ut_crc32_64_hw(&crc, &buf, &len); |
337 | ut_crc32_64_hw(&crc, &buf, &len); |
338 | ut_crc32_64_hw(&crc, &buf, &len); |
339 | ut_crc32_64_hw(&crc, &buf, &len); |
340 | ut_crc32_64_hw(&crc, &buf, &len); |
341 | ut_crc32_64_hw(&crc, &buf, &len); |
342 | ut_crc32_64_hw(&crc, &buf, &len); |
343 | ut_crc32_64_hw(&crc, &buf, &len); |
344 | ut_crc32_64_hw(&crc, &buf, &len); |
345 | ut_crc32_64_hw(&crc, &buf, &len); |
346 | ut_crc32_64_hw(&crc, &buf, &len); |
347 | ut_crc32_64_hw(&crc, &buf, &len); |
348 | ut_crc32_64_hw(&crc, &buf, &len); |
349 | } |
350 | |
351 | while (len >= 8) { |
352 | ut_crc32_64_hw(&crc, &buf, &len); |
353 | } |
354 | |
355 | while (len > 0) { |
356 | ut_crc32_8_hw(&crc, &buf, &len); |
357 | } |
358 | |
359 | return(~crc); |
360 | } |
361 | #endif /* defined(__GNUC__) && defined(__x86_64__) || (_WIN64) */ |
362 | |
363 | /* CRC32 software implementation. */ |
364 | |
365 | /* Precalculated table used to generate the CRC32 if the CPU does not |
366 | have support for it */ |
367 | static uint32_t ut_crc32_slice8_table[8][256]; |
368 | static bool ut_crc32_slice8_table_initialized = false; |
369 | |
370 | /********************************************************************//** |
371 | Initializes the table that is used to generate the CRC32 if the CPU does |
372 | not have support for it. */ |
373 | static |
374 | void |
375 | ut_crc32_slice8_table_init() |
376 | /*========================*/ |
377 | { |
378 | /* bit-reversed poly 0x1EDC6F41 (from SSE42 crc32 instruction) */ |
379 | static const uint32_t poly = 0x82f63b78; |
380 | uint32_t n; |
381 | uint32_t k; |
382 | uint32_t c; |
383 | |
384 | for (n = 0; n < 256; n++) { |
385 | c = n; |
386 | for (k = 0; k < 8; k++) { |
387 | c = (c & 1) ? (poly ^ (c >> 1)) : (c >> 1); |
388 | } |
389 | ut_crc32_slice8_table[0][n] = c; |
390 | } |
391 | |
392 | for (n = 0; n < 256; n++) { |
393 | c = ut_crc32_slice8_table[0][n]; |
394 | for (k = 1; k < 8; k++) { |
395 | c = ut_crc32_slice8_table[0][c & 0xFF] ^ (c >> 8); |
396 | ut_crc32_slice8_table[k][n] = c; |
397 | } |
398 | } |
399 | |
400 | ut_crc32_slice8_table_initialized = true; |
401 | } |
402 | |
403 | /** Calculate CRC32 over 8-bit data using a software implementation. |
404 | @param[in,out] crc crc32 checksum so far when this function is called, |
405 | when the function ends it will contain the new checksum |
406 | @param[in,out] data data to be checksummed, the pointer will be advanced |
407 | with 1 byte |
408 | @param[in,out] len remaining bytes, it will be decremented with 1 */ |
409 | inline |
410 | void |
411 | ut_crc32_8_sw( |
412 | uint32_t* crc, |
413 | const byte** data, |
414 | ulint* len) |
415 | { |
416 | const uint8_t i = (*crc ^ (*data)[0]) & 0xFF; |
417 | |
418 | *crc = (*crc >> 8) ^ ut_crc32_slice8_table[0][i]; |
419 | |
420 | (*data)++; |
421 | (*len)--; |
422 | } |
423 | |
424 | /** Calculate CRC32 over a 64-bit integer using a software implementation. |
425 | @param[in] crc crc32 checksum so far |
426 | @param[in] data data to be checksummed |
427 | @return resulting checksum of crc + crc(data) */ |
428 | inline |
429 | uint32_t |
430 | ut_crc32_64_low_sw( |
431 | uint32_t crc, |
432 | uint64_t data) |
433 | { |
434 | const uint64_t i = crc ^ data; |
435 | |
436 | return( |
437 | ut_crc32_slice8_table[7][(i ) & 0xFF] ^ |
438 | ut_crc32_slice8_table[6][(i >> 8) & 0xFF] ^ |
439 | ut_crc32_slice8_table[5][(i >> 16) & 0xFF] ^ |
440 | ut_crc32_slice8_table[4][(i >> 24) & 0xFF] ^ |
441 | ut_crc32_slice8_table[3][(i >> 32) & 0xFF] ^ |
442 | ut_crc32_slice8_table[2][(i >> 40) & 0xFF] ^ |
443 | ut_crc32_slice8_table[1][(i >> 48) & 0xFF] ^ |
444 | ut_crc32_slice8_table[0][(i >> 56)] |
445 | ); |
446 | } |
447 | |
448 | /** Calculate CRC32 over 64-bit byte string using a software implementation. |
449 | @param[in,out] crc crc32 checksum so far when this function is called, |
450 | when the function ends it will contain the new checksum |
451 | @param[in,out] data data to be checksummed, the pointer will be advanced |
452 | with 8 bytes |
453 | @param[in,out] len remaining bytes, it will be decremented with 8 */ |
454 | inline |
455 | void |
456 | ut_crc32_64_sw( |
457 | uint32_t* crc, |
458 | const byte** data, |
459 | ulint* len) |
460 | { |
461 | uint64_t data_int = *reinterpret_cast<const uint64_t*>(*data); |
462 | |
463 | #ifdef WORDS_BIGENDIAN |
464 | data_int = ut_crc32_swap_byteorder(data_int); |
465 | #endif /* WORDS_BIGENDIAN */ |
466 | |
467 | *crc = ut_crc32_64_low_sw(*crc, data_int); |
468 | |
469 | *data += 8; |
470 | *len -= 8; |
471 | } |
472 | |
473 | /** Calculate CRC32 over 64-bit byte string using a software implementation. |
474 | The byte string is converted to a 64-bit integer using big endian byte order. |
475 | @param[in,out] crc crc32 checksum so far when this function is called, |
476 | when the function ends it will contain the new checksum |
477 | @param[in,out] data data to be checksummed, the pointer will be advanced |
478 | with 8 bytes |
479 | @param[in,out] len remaining bytes, it will be decremented with 8 */ |
480 | inline |
481 | void |
482 | ut_crc32_64_legacy_big_endian_sw( |
483 | uint32_t* crc, |
484 | const byte** data, |
485 | ulint* len) |
486 | { |
487 | uint64_t data_int = *reinterpret_cast<const uint64_t*>(*data); |
488 | |
489 | #ifndef WORDS_BIGENDIAN |
490 | data_int = ut_crc32_swap_byteorder(data_int); |
491 | #endif /* WORDS_BIGENDIAN */ |
492 | |
493 | *crc = ut_crc32_64_low_sw(*crc, data_int); |
494 | |
495 | *data += 8; |
496 | *len -= 8; |
497 | } |
498 | |
499 | /** Calculates CRC32 in software, without using CPU instructions. |
500 | @param[in] buf data over which to calculate CRC32 |
501 | @param[in] len data length |
502 | @return CRC-32C (polynomial 0x11EDC6F41) */ |
503 | uint32_t |
504 | ut_crc32_sw( |
505 | const byte* buf, |
506 | ulint len) |
507 | { |
508 | uint32_t crc = 0xFFFFFFFFU; |
509 | |
510 | ut_a(ut_crc32_slice8_table_initialized); |
511 | |
512 | /* Calculate byte-by-byte up to an 8-byte aligned address. After |
513 | this consume the input 8-bytes at a time. */ |
514 | while (len > 0 && (reinterpret_cast<uintptr_t>(buf) & 7) != 0) { |
515 | ut_crc32_8_sw(&crc, &buf, &len); |
516 | } |
517 | |
518 | while (len >= 128) { |
519 | /* This call is repeated 16 times. 16 * 8 = 128. */ |
520 | ut_crc32_64_sw(&crc, &buf, &len); |
521 | ut_crc32_64_sw(&crc, &buf, &len); |
522 | ut_crc32_64_sw(&crc, &buf, &len); |
523 | ut_crc32_64_sw(&crc, &buf, &len); |
524 | ut_crc32_64_sw(&crc, &buf, &len); |
525 | ut_crc32_64_sw(&crc, &buf, &len); |
526 | ut_crc32_64_sw(&crc, &buf, &len); |
527 | ut_crc32_64_sw(&crc, &buf, &len); |
528 | ut_crc32_64_sw(&crc, &buf, &len); |
529 | ut_crc32_64_sw(&crc, &buf, &len); |
530 | ut_crc32_64_sw(&crc, &buf, &len); |
531 | ut_crc32_64_sw(&crc, &buf, &len); |
532 | ut_crc32_64_sw(&crc, &buf, &len); |
533 | ut_crc32_64_sw(&crc, &buf, &len); |
534 | ut_crc32_64_sw(&crc, &buf, &len); |
535 | ut_crc32_64_sw(&crc, &buf, &len); |
536 | } |
537 | |
538 | while (len >= 8) { |
539 | ut_crc32_64_sw(&crc, &buf, &len); |
540 | } |
541 | |
542 | while (len > 0) { |
543 | ut_crc32_8_sw(&crc, &buf, &len); |
544 | } |
545 | |
546 | return(~crc); |
547 | } |
548 | |
549 | /** Calculates CRC32 in software, without using CPU instructions. |
550 | This function uses big endian byte ordering when converting byte sequence to |
551 | integers. |
552 | @param[in] buf data over which to calculate CRC32 |
553 | @param[in] len data length |
554 | @return CRC-32C (polynomial 0x11EDC6F41) */ |
555 | uint32_t |
556 | ut_crc32_legacy_big_endian( |
557 | const byte* buf, |
558 | ulint len) |
559 | { |
560 | uint32_t crc = 0xFFFFFFFFU; |
561 | |
562 | ut_a(ut_crc32_slice8_table_initialized); |
563 | |
564 | /* Calculate byte-by-byte up to an 8-byte aligned address. After |
565 | this consume the input 8-bytes at a time. */ |
566 | while (len > 0 && (reinterpret_cast<uintptr_t>(buf) & 7) != 0) { |
567 | ut_crc32_8_sw(&crc, &buf, &len); |
568 | } |
569 | |
570 | while (len >= 128) { |
571 | /* This call is repeated 16 times. 16 * 8 = 128. */ |
572 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
573 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
574 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
575 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
576 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
577 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
578 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
579 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
580 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
581 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
582 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
583 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
584 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
585 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
586 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
587 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
588 | } |
589 | |
590 | while (len >= 8) { |
591 | ut_crc32_64_legacy_big_endian_sw(&crc, &buf, &len); |
592 | } |
593 | |
594 | while (len > 0) { |
595 | ut_crc32_8_sw(&crc, &buf, &len); |
596 | } |
597 | |
598 | return(~crc); |
599 | } |
600 | |
601 | /********************************************************************//** |
602 | Initializes the data structures used by ut_crc32*(). Does not do any |
603 | allocations, would not hurt if called twice, but would be pointless. */ |
604 | void |
605 | ut_crc32_init() |
606 | /*===========*/ |
607 | { |
608 | ut_crc32_slice8_table_init(); |
609 | |
610 | #if (defined(__GNUC__) && defined(__x86_64__)) || defined(_MSC_VER) |
611 | uint32_t vend[3]; |
612 | uint32_t model; |
613 | uint32_t family; |
614 | uint32_t stepping; |
615 | uint32_t features_ecx; |
616 | uint32_t features_edx; |
617 | |
618 | ut_cpuid(vend, &model, &family, &stepping, |
619 | &features_ecx, &features_edx); |
620 | |
621 | /* Valgrind does not understand the CRC32 instructions: |
622 | |
623 | vex amd64->IR: unhandled instruction bytes: 0xF2 0x48 0xF 0x38 0xF0 0xA |
624 | valgrind: Unrecognised instruction at address 0xad3db5. |
625 | Your program just tried to execute an instruction that Valgrind |
626 | did not recognise. There are two possible reasons for this. |
627 | 1. Your program has a bug and erroneously jumped to a non-code |
628 | location. If you are running Memcheck and you just saw a |
629 | warning about a bad jump, it's probably your program's fault. |
630 | 2. The instruction is legitimate but Valgrind doesn't handle it, |
631 | i.e. it's Valgrind's fault. If you think this is the case or |
632 | you are not sure, please let us know and we'll try to fix it. |
633 | Either way, Valgrind will now raise a SIGILL signal which will |
634 | probably kill your program. |
635 | |
636 | */ |
637 | |
638 | if (features_ecx & 1 << 20) { |
639 | ut_crc32 = ut_crc32_hw; |
640 | ut_crc32_implementation = "Using SSE2 crc32 instructions" ; |
641 | } |
642 | #endif |
643 | } |
644 | |