2 * Copyright (C) 2012, 2013
6 * Permission is hereby granted, free of charge, to any person obtaining a copy of
7 * this software and associated documentation files (the "Software"), to deal in
8 * the Software without restriction, including without limitation the rights to
9 * use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
10 * of the Software, and to permit persons to whom the Software is furnished to do
11 * so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 void util_debug(const char *area, const char *ms, ...) {
31 if (!OPTS_OPTION_BOOL(OPTION_DEBUG))
34 if (!strcmp(area, "MEM") && !OPTS_OPTION_BOOL(OPTION_MEMCHK))
38 con_out ("[%s] ", area);
44 * only required if big endian .. otherwise no need to swap
47 #if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_BIG
48 static GMQCC_INLINE void util_swap16(uint16_t *d, size_t l) {
50 d[l] = (d[l] << 8) | (d[l] >> 8);
54 static GMQCC_INLINE void util_swap32(uint32_t *d, size_t l) {
57 v = ((d[l] << 8) & 0xFF00FF00) | ((d[l] >> 8) & 0x00FF00FF);
58 d[l] = (v << 16) | (v >> 16);
62 /* Some strange system doesn't like constants that big, AND doesn't recognize an ULL suffix
63 * so let's go the safe way
65 static GMQCC_INLINE void util_swap64(uint32_t *d, size_t l) {
69 v = ((d[l] << 8) & 0xFF00FF00FF00FF00) | ((d[l] >> 8) & 0x00FF00FF00FF00FF);
70 v = ((v << 16) & 0xFFFF0000FFFF0000) | ((v >> 16) & 0x0000FFFF0000FFFF);
71 d[l] = (v << 32) | (v >> 32);
75 for (i = 0; i < l; i += 2) {
84 void util_endianswap(void *_data, size_t length, unsigned int typesize) {
85 # if PLATFORM_BYTE_ORDER == -1 /* runtime check */
86 if (*((char*)&typesize))
89 /* prevent unused warnings */
94 # if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_LITTLE
100 util_swap16((uint16_t*)_data, length>>1);
103 util_swap32((uint32_t*)_data, length>>2);
106 util_swap64((uint32_t*)_data, length>>3);
109 default: exit(EXIT_FAILURE); /* please blow the fuck up! */
116 * CRC algorithms vary in the width of the polynomial, the value of said polynomial,
117 * the initial value used for the register, weather the bits of each byte are reflected
118 * before being processed, weather the algorithm itself feeds input bytes through the
119 * register or XORs them with a byte from one end and then straight into the table, as
120 * well as (but not limited to the idea of reflected versions) where the final register
121 * value becomes reversed, and finally weather the value itself is used to XOR the final
122 * register value. AS such you can already imagine how painfully annoying CRCs are,
123 * of course we stand to target Quake, which expects it's certian set of rules for proper
124 * calculation of a CRC.
126 * In most traditional CRC algorithms on uses a reflected table driven method where a value
127 * or register is reflected if it's bits are swapped around it's center. For example:
128 * take the bits 0101 is the 4-bit reflection of 1010, and respectfully 0011 would be the
129 * reflection of 1100. Quake however expects a NON-Reflected CRC on the output, but still
130 * requires a final XOR on the values (0xFFFF and 0x0000) this is a standard CCITT CRC-16
131 * which I respectfully as a programmer don't agree with.
133 * So now you know what we target, and why we target it, despite how unsettling it may seem
134 * but those are what Quake seems to request.
137 static const uint16_t util_crc16_table[] = {
138 0x0000, 0x1021, 0x2042, 0x3063, 0x4084, 0x50A5,
139 0x60C6, 0x70E7, 0x8108, 0x9129, 0xA14A, 0xB16B,
140 0xC18C, 0xD1AD, 0xE1CE, 0xF1EF, 0x1231, 0x0210,
141 0x3273, 0x2252, 0x52B5, 0x4294, 0x72F7, 0x62D6,
142 0x9339, 0x8318, 0xB37B, 0xA35A, 0xD3BD, 0xC39C,
143 0xF3FF, 0xE3DE, 0x2462, 0x3443, 0x0420, 0x1401,
144 0x64E6, 0x74C7, 0x44A4, 0x5485, 0xA56A, 0xB54B,
145 0x8528, 0x9509, 0xE5EE, 0xF5CF, 0xC5AC, 0xD58D,
146 0x3653, 0x2672, 0x1611, 0x0630, 0x76D7, 0x66F6,
147 0x5695, 0x46B4, 0xB75B, 0xA77A, 0x9719, 0x8738,
148 0xF7DF, 0xE7FE, 0xD79D, 0xC7BC, 0x48C4, 0x58E5,
149 0x6886, 0x78A7, 0x0840, 0x1861, 0x2802, 0x3823,
150 0xC9CC, 0xD9ED, 0xE98E, 0xF9AF, 0x8948, 0x9969,
151 0xA90A, 0xB92B, 0x5AF5, 0x4AD4, 0x7AB7, 0x6A96,
152 0x1A71, 0x0A50, 0x3A33, 0x2A12, 0xDBFD, 0xCBDC,
153 0xFBBF, 0xEB9E, 0x9B79, 0x8B58, 0xBB3B, 0xAB1A,
154 0x6CA6, 0x7C87, 0x4CE4, 0x5CC5, 0x2C22, 0x3C03,
155 0x0C60, 0x1C41, 0xEDAE, 0xFD8F, 0xCDEC, 0xDDCD,
156 0xAD2A, 0xBD0B, 0x8D68, 0x9D49, 0x7E97, 0x6EB6,
157 0x5ED5, 0x4EF4, 0x3E13, 0x2E32, 0x1E51, 0x0E70,
158 0xFF9F, 0xEFBE, 0xDFDD, 0xCFFC, 0xBF1B, 0xAF3A,
159 0x9F59, 0x8F78, 0x9188, 0x81A9, 0xB1CA, 0xA1EB,
160 0xD10C, 0xC12D, 0xF14E, 0xE16F, 0x1080, 0x00A1,
161 0x30C2, 0x20E3, 0x5004, 0x4025, 0x7046, 0x6067,
162 0x83B9, 0x9398, 0xA3FB, 0xB3DA, 0xC33D, 0xD31C,
163 0xE37F, 0xF35E, 0x02B1, 0x1290, 0x22F3, 0x32D2,
164 0x4235, 0x5214, 0x6277, 0x7256, 0xB5EA, 0xA5CB,
165 0x95A8, 0x8589, 0xF56E, 0xE54F, 0xD52C, 0xC50D,
166 0x34E2, 0x24C3, 0x14A0, 0x0481, 0x7466, 0x6447,
167 0x5424, 0x4405, 0xA7DB, 0xB7FA, 0x8799, 0x97B8,
168 0xE75F, 0xF77E, 0xC71D, 0xD73C, 0x26D3, 0x36F2,
169 0x0691, 0x16B0, 0x6657, 0x7676, 0x4615, 0x5634,
170 0xD94C, 0xC96D, 0xF90E, 0xE92F, 0x99C8, 0x89E9,
171 0xB98A, 0xA9AB, 0x5844, 0x4865, 0x7806, 0x6827,
172 0x18C0, 0x08E1, 0x3882, 0x28A3, 0xCB7D, 0xDB5C,
173 0xEB3F, 0xFB1E, 0x8BF9, 0x9BD8, 0xABBB, 0xBB9A,
174 0x4A75, 0x5A54, 0x6A37, 0x7A16, 0x0AF1, 0x1AD0,
175 0x2AB3, 0x3A92, 0xFD2E, 0xED0F, 0xDD6C, 0xCD4D,
176 0xBDAA, 0xAD8B, 0x9DE8, 0x8DC9, 0x7C26, 0x6C07,
177 0x5C64, 0x4C45, 0x3CA2, 0x2C83, 0x1CE0, 0x0CC1,
178 0xEF1F, 0xFF3E, 0xCF5D, 0xDF7C, 0xAF9B, 0xBFBA,
179 0x8FD9, 0x9FF8, 0x6E17, 0x7E36, 0x4E55, 0x5E74,
180 0x2E93, 0x3EB2, 0x0ED1, 0x1EF0
183 /* Non - Reflected */
184 uint16_t util_crc16(uint16_t current, const char *k, size_t len) {
185 register uint16_t h = current;
186 for (; len; --len, ++k)
187 h = util_crc16_table[(h>>8)^((unsigned char)*k)]^(h<<8);
190 /* Reflective Varation (for reference) */
192 uint16_t util_crc16(const char *k, int len, const short clamp) {
193 register uint16_t h= (uint16_t)0xFFFFFFFF;
194 for (; len; --len, ++k)
195 h = util_crc16_table[(h^((unsigned char)*k))&0xFF]^(h>>8);
200 size_t util_strtocmd(const char *in, char *out, size_t outsz) {
202 for (; *in && sz < outsz; ++in, ++out, ++sz)
203 *out = (*in == '-') ? '_' : (util_isalpha(*in) && !util_isupper(*in)) ? *in + 'A' - 'a': *in;
208 size_t util_strtononcmd(const char *in, char *out, size_t outsz) {
210 for (; *in && sz < outsz; ++in, ++out, ++sz)
211 *out = (*in == '_') ? '-' : (util_isalpha(*in) && util_isupper(*in)) ? *in + 'a' - 'A' : *in;
217 * Portable implementation of vasprintf/asprintf. Assumes vsnprintf
218 * exists, otherwise compiler error.
220 * TODO: fix for MSVC ....
222 int util_vasprintf(char **dat, const char *fmt, va_list args) {
228 * For visuals tido _vsnprintf doesn't tell you the length of a
229 * formatted string if it overflows. However there is a MSVC
230 * intrinsic (which is documented wrong) called _vcsprintf which
231 * will return the required amount to allocate.
234 if ((len = _vscprintf(fmt, args)) < 0) {
239 tmp = (char*)mem_a(len + 1);
240 if ((ret = _vsnprintf_s(tmp, len+1, len+1, fmt, args)) != len) {
249 * For everything else we have a decent conformint vsnprintf that
250 * returns the number of bytes needed. We give it a try though on
251 * a short buffer, since efficently speaking, it could be nice to
252 * above a second vsnprintf call.
257 len = vsnprintf(buf, sizeof(buf), fmt, cpy);
260 if (len < (int)sizeof(buf)) {
261 *dat = util_strdup(buf);
265 /* not large enough ... */
266 tmp = (char*)mem_a(len + 1);
267 if ((ret = vsnprintf(tmp, len + 1, fmt, args)) != len) {
277 int util_asprintf(char **ret, const char *fmt, ...) {
281 read = util_vasprintf(ret, fmt, args);
288 * These are various re-implementations (wrapping the real ones) of
289 * string functions that MSVC consideres unsafe. We wrap these up and
290 * use the safe varations on MSVC.
293 static char **util_strerror_allocated() {
294 static char **data = NULL;
298 static void util_strerror_cleanup(void) {
300 char **data = util_strerror_allocated();
301 for (i = 0; i < vec_size(data); i++)
306 const char *util_strerror(int num) {
307 char *allocated = NULL;
308 static bool install = false;
309 static size_t tries = 0;
310 char **vector = util_strerror_allocated();
312 /* try installing cleanup handler */
317 install = !atexit(&util_strerror_cleanup);
321 allocated = (char*)mem_a(4096); /* A page must be enough */
322 strerror_s(allocated, 4096, num);
324 vec_push(vector, allocated);
325 return (const char *)allocated;
328 int util_snprintf(char *src, size_t bytes, const char *format, ...) {
331 va_start(va, format);
333 rt = vsprintf_s(src, bytes, format, va);
339 char *util_strcat(char *dest, const char *src) {
340 strcat_s(dest, strlen(src), src);
344 char *util_strncpy(char *dest, const char *src, size_t num) {
345 strncpy_s(dest, num, src, num);
349 const char *util_strerror(int num) {
350 return strerror(num);
353 int util_snprintf(char *src, size_t bytes, const char *format, ...) {
356 va_start(va, format);
357 rt = vsnprintf(src, bytes, format, va);
363 char *util_strcat(char *dest, const char *src) {
364 return strcat(dest, src);
367 char *util_strncpy(char *dest, const char *src, size_t num) {
368 return strncpy(dest, src, num);
371 #endif /*! _MSC_VER */
374 * Implementation of the Mersenne twister PRNG (pseudo random numer
375 * generator). Implementation of MT19937. Has a period of 2^19937-1
376 * which is a Mersenne Prime (hence the name).
378 * Implemented from specification and original paper:
379 * http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/ARTICLES/mt.pdf
381 * This code is placed in the public domain by me personally
382 * (Dale Weiler, a.k.a graphitemaster).
386 #define MT_PERIOD 397
387 #define MT_SPACE (MT_SIZE - MT_PERIOD)
389 static uint32_t mt_state[MT_SIZE];
390 static size_t mt_index = 0;
392 static GMQCC_INLINE void mt_generate(void) {
394 * The loop has been unrolled here: the original paper and implemenation
395 * Called for the following code:
396 * for (register unsigned i = 0; i < MT_SIZE; ++i) {
397 * register uint32_t load;
398 * load = (0x80000000 & mt_state[i]) // most significant 32nd bit
399 * load |= (0x7FFFFFFF & mt_state[(i + 1) % MT_SIZE]) // least significant 31nd bit
401 * mt_state[i] = mt_state[(i + MT_PERIOD) % MT_SIZE] ^ (load >> 1);
403 * if (load & 1) mt_state[i] ^= 0x9908B0DF;
406 * This essentially is a waste: we have two modulus operations, and
407 * a branch that is executed every iteration from [0, MT_SIZE).
409 * Please see: http://www.quadibloc.com/crypto/co4814.htm for more
410 * information on how this clever trick works.
412 static const uint32_t matrix[2] = {
417 * This register gives up a little more speed by instructing the compiler
418 * to force these into CPU registers (they're counters for indexing mt_state
419 * which we can force the compiler to generate prefetch instructions for)
425 * Said loop has been unrolled for MT_SPACE (226 iterations), opposed
426 * to [0, MT_SIZE) (634 iterations).
428 for (i = 0; i < MT_SPACE; ++i) {
429 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
430 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
432 i ++; /* loop unroll */
434 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
435 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
439 * collapsing the walls unrolled (evenly dividing 396 [632-227 = 396
443 while (i < MT_SIZE - 1) {
445 * We expand this 11 times .. manually, no macros are required
446 * here. This all fits in the CPU cache.
448 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
449 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
451 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
452 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
454 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
455 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
457 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
458 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
460 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
461 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
463 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
464 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
466 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
467 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
469 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
470 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
472 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
473 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
475 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
476 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
478 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
479 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
483 /* i = mt_state[623] */
484 y = (0x80000000 & mt_state[MT_SIZE - 1]) | (0x7FFFFFFF & mt_state[MT_SIZE - 1]);
485 mt_state[MT_SIZE - 1] = mt_state[MT_PERIOD - 1] ^ (y >> 1) ^ matrix[y & 1];
488 void util_seed(uint32_t value) {
490 * We seed the mt_state with a LCG (linear congruential generator)
491 * We're operating exactly on exactly m=32, so there is no need to
494 * The multipler of choice is 0x6C07865, also knows as the Borosh-
495 * Niederreiter multipler used for modulus 2^32. More can be read
496 * about this in Knuth's TAOCP Volume 2, page 106.
498 * If you don't own TAOCP something is wrong with you :-) .. so I
499 * also provided a link to the original paper by Borosh and
500 * Niederreiter. It's called "Optional Multipliers for PRNG by The
501 * Linear Congruential Method" (1983).
502 * http://en.wikipedia.org/wiki/Linear_congruential_generator
504 * From said page, it says the following:
505 * "A common Mersenne twister implementation, interestingly enough
506 * used an LCG to generate seed data."
509 * The data we're operating on is 32-bits for the mt_state array, so
510 * there is no masking required with 0xFFFFFFFF
515 for (i = 1; i < MT_SIZE; ++i)
516 mt_state[i] = 0x6C078965 * (mt_state[i - 1] ^ mt_state[i - 1] >> 30) + i;
519 uint32_t util_rand() {
523 * This is inlined with any sane compiler (I checked)
524 * for some reason though, SubC seems to be generating invalid
525 * code when it inlines this.
530 y = mt_state[mt_index];
532 /* Standard tempering */
533 y ^= y >> 11; /* +7 */
534 y ^= y << 7 & 0x9D2C5680; /* +4 */
535 y ^= y << 15 & 0xEFC60000; /* -4 */
536 y ^= y >> 18; /* -7 */
538 if(++mt_index == MT_SIZE)