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
30 void util_debug(const char *area, const char *ms, ...) {
32 if (!OPTS_OPTION_BOOL(OPTION_DEBUG))
35 if (!strcmp(area, "MEM") && !OPTS_OPTION_BOOL(OPTION_MEMCHK))
39 con_out ("[%s] ", area);
45 * only required if big endian .. otherwise no need to swap
48 #if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_BIG
49 static GMQCC_INLINE void util_swap16(uint16_t *d, size_t l) {
51 d[l] = (d[l] << 8) | (d[l] >> 8);
55 static GMQCC_INLINE void util_swap32(uint32_t *d, size_t l) {
58 v = ((d[l] << 8) & 0xFF00FF00) | ((d[l] >> 8) & 0x00FF00FF);
59 d[l] = (v << 16) | (v >> 16);
63 /* Some strange system doesn't like constants that big, AND doesn't recognize an ULL suffix
64 * so let's go the safe way
66 static GMQCC_INLINE void util_swap64(uint32_t *d, size_t l) {
70 v = ((d[l] << 8) & 0xFF00FF00FF00FF00) | ((d[l] >> 8) & 0x00FF00FF00FF00FF);
71 v = ((v << 16) & 0xFFFF0000FFFF0000) | ((v >> 16) & 0x0000FFFF0000FFFF);
72 d[l] = (v << 32) | (v >> 32);
76 for (i = 0; i < l; i += 2) {
85 void util_endianswap(void *_data, size_t length, unsigned int typesize) {
86 # if PLATFORM_BYTE_ORDER == -1 /* runtime check */
87 if (*((char*)&typesize))
90 /* prevent unused warnings */
95 # if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_LITTLE
101 util_swap16((uint16_t*)_data, length>>1);
104 util_swap32((uint32_t*)_data, length>>2);
107 util_swap64((uint32_t*)_data, length>>3);
110 default: exit(EXIT_FAILURE); /* please blow the fuck up! */
117 * CRC algorithms vary in the width of the polynomial, the value of said polynomial,
118 * the initial value used for the register, weather the bits of each byte are reflected
119 * before being processed, weather the algorithm itself feeds input bytes through the
120 * register or XORs them with a byte from one end and then straight into the table, as
121 * well as (but not limited to the idea of reflected versions) where the final register
122 * value becomes reversed, and finally weather the value itself is used to XOR the final
123 * register value. AS such you can already imagine how painfully annoying CRCs are,
124 * of course we stand to target Quake, which expects it's certian set of rules for proper
125 * calculation of a CRC.
127 * In most traditional CRC algorithms on uses a reflected table driven method where a value
128 * or register is reflected if it's bits are swapped around it's center. For example:
129 * take the bits 0101 is the 4-bit reflection of 1010, and respectfully 0011 would be the
130 * reflection of 1100. Quake however expects a NON-Reflected CRC on the output, but still
131 * requires a final XOR on the values (0xFFFF and 0x0000) this is a standard CCITT CRC-16
132 * which I respectfully as a programmer don't agree with.
134 * So now you know what we target, and why we target it, despite how unsettling it may seem
135 * but those are what Quake seems to request.
138 static const uint16_t util_crc16_table[] = {
139 0x0000, 0x1021, 0x2042, 0x3063, 0x4084, 0x50A5,
140 0x60C6, 0x70E7, 0x8108, 0x9129, 0xA14A, 0xB16B,
141 0xC18C, 0xD1AD, 0xE1CE, 0xF1EF, 0x1231, 0x0210,
142 0x3273, 0x2252, 0x52B5, 0x4294, 0x72F7, 0x62D6,
143 0x9339, 0x8318, 0xB37B, 0xA35A, 0xD3BD, 0xC39C,
144 0xF3FF, 0xE3DE, 0x2462, 0x3443, 0x0420, 0x1401,
145 0x64E6, 0x74C7, 0x44A4, 0x5485, 0xA56A, 0xB54B,
146 0x8528, 0x9509, 0xE5EE, 0xF5CF, 0xC5AC, 0xD58D,
147 0x3653, 0x2672, 0x1611, 0x0630, 0x76D7, 0x66F6,
148 0x5695, 0x46B4, 0xB75B, 0xA77A, 0x9719, 0x8738,
149 0xF7DF, 0xE7FE, 0xD79D, 0xC7BC, 0x48C4, 0x58E5,
150 0x6886, 0x78A7, 0x0840, 0x1861, 0x2802, 0x3823,
151 0xC9CC, 0xD9ED, 0xE98E, 0xF9AF, 0x8948, 0x9969,
152 0xA90A, 0xB92B, 0x5AF5, 0x4AD4, 0x7AB7, 0x6A96,
153 0x1A71, 0x0A50, 0x3A33, 0x2A12, 0xDBFD, 0xCBDC,
154 0xFBBF, 0xEB9E, 0x9B79, 0x8B58, 0xBB3B, 0xAB1A,
155 0x6CA6, 0x7C87, 0x4CE4, 0x5CC5, 0x2C22, 0x3C03,
156 0x0C60, 0x1C41, 0xEDAE, 0xFD8F, 0xCDEC, 0xDDCD,
157 0xAD2A, 0xBD0B, 0x8D68, 0x9D49, 0x7E97, 0x6EB6,
158 0x5ED5, 0x4EF4, 0x3E13, 0x2E32, 0x1E51, 0x0E70,
159 0xFF9F, 0xEFBE, 0xDFDD, 0xCFFC, 0xBF1B, 0xAF3A,
160 0x9F59, 0x8F78, 0x9188, 0x81A9, 0xB1CA, 0xA1EB,
161 0xD10C, 0xC12D, 0xF14E, 0xE16F, 0x1080, 0x00A1,
162 0x30C2, 0x20E3, 0x5004, 0x4025, 0x7046, 0x6067,
163 0x83B9, 0x9398, 0xA3FB, 0xB3DA, 0xC33D, 0xD31C,
164 0xE37F, 0xF35E, 0x02B1, 0x1290, 0x22F3, 0x32D2,
165 0x4235, 0x5214, 0x6277, 0x7256, 0xB5EA, 0xA5CB,
166 0x95A8, 0x8589, 0xF56E, 0xE54F, 0xD52C, 0xC50D,
167 0x34E2, 0x24C3, 0x14A0, 0x0481, 0x7466, 0x6447,
168 0x5424, 0x4405, 0xA7DB, 0xB7FA, 0x8799, 0x97B8,
169 0xE75F, 0xF77E, 0xC71D, 0xD73C, 0x26D3, 0x36F2,
170 0x0691, 0x16B0, 0x6657, 0x7676, 0x4615, 0x5634,
171 0xD94C, 0xC96D, 0xF90E, 0xE92F, 0x99C8, 0x89E9,
172 0xB98A, 0xA9AB, 0x5844, 0x4865, 0x7806, 0x6827,
173 0x18C0, 0x08E1, 0x3882, 0x28A3, 0xCB7D, 0xDB5C,
174 0xEB3F, 0xFB1E, 0x8BF9, 0x9BD8, 0xABBB, 0xBB9A,
175 0x4A75, 0x5A54, 0x6A37, 0x7A16, 0x0AF1, 0x1AD0,
176 0x2AB3, 0x3A92, 0xFD2E, 0xED0F, 0xDD6C, 0xCD4D,
177 0xBDAA, 0xAD8B, 0x9DE8, 0x8DC9, 0x7C26, 0x6C07,
178 0x5C64, 0x4C45, 0x3CA2, 0x2C83, 0x1CE0, 0x0CC1,
179 0xEF1F, 0xFF3E, 0xCF5D, 0xDF7C, 0xAF9B, 0xBFBA,
180 0x8FD9, 0x9FF8, 0x6E17, 0x7E36, 0x4E55, 0x5E74,
181 0x2E93, 0x3EB2, 0x0ED1, 0x1EF0
184 /* Non - Reflected */
185 uint16_t util_crc16(uint16_t current, const char *k, size_t len) {
186 register uint16_t h = current;
187 for (; len; --len, ++k)
188 h = util_crc16_table[(h>>8)^((unsigned char)*k)]^(h<<8);
191 /* Reflective Varation (for reference) */
193 uint16_t util_crc16(const char *k, int len, const short clamp) {
194 register uint16_t h= (uint16_t)0xFFFFFFFF;
195 for (; len; --len, ++k)
196 h = util_crc16_table[(h^((unsigned char)*k))&0xFF]^(h>>8);
201 size_t util_strtocmd(const char *in, char *out, size_t outsz) {
203 for (; *in && sz < outsz; ++in, ++out, ++sz)
204 *out = (*in == '-') ? '_' : (isalpha(*in) && !isupper(*in)) ? *in + 'A' - 'a': *in;
209 size_t util_strtononcmd(const char *in, char *out, size_t outsz) {
211 for (; *in && sz < outsz; ++in, ++out, ++sz)
212 *out = (*in == '_') ? '-' : (isalpha(*in) && isupper(*in)) ? *in + 'a' - 'A' : *in;
218 * Portable implementation of vasprintf/asprintf. Assumes vsnprintf
219 * exists, otherwise compiler error.
221 * TODO: fix for MSVC ....
223 int util_vasprintf(char **dat, const char *fmt, va_list args) {
229 * For visuals tido _vsnprintf doesn't tell you the length of a
230 * formatted string if it overflows. However there is a MSVC
231 * intrinsic (which is documented wrong) called _vcsprintf which
232 * will return the required amount to allocate.
235 if ((len = _vscprintf(fmt, args)) < 0) {
240 tmp = (char*)mem_a(len + 1);
241 if ((ret = _vsnprintf_s(tmp, len+1, len+1, fmt, args)) != len) {
250 * For everything else we have a decent conformint vsnprintf that
251 * returns the number of bytes needed. We give it a try though on
252 * a short buffer, since efficently speaking, it could be nice to
253 * above a second vsnprintf call.
258 len = vsnprintf(buf, sizeof(buf), fmt, cpy);
261 if (len < (int)sizeof(buf)) {
262 *dat = util_strdup(buf);
266 /* not large enough ... */
267 tmp = (char*)mem_a(len + 1);
268 if ((ret = vsnprintf(tmp, len + 1, fmt, args)) != len) {
278 int util_asprintf(char **ret, const char *fmt, ...) {
282 read = util_vasprintf(ret, fmt, args);
289 * These are various re-implementations (wrapping the real ones) of
290 * string functions that MSVC consideres unsafe. We wrap these up and
291 * use the safe varations on MSVC.
294 static char **util_strerror_allocated() {
295 static char **data = NULL;
299 static void util_strerror_cleanup(void) {
301 char **data = util_strerror_allocated();
302 for (i = 0; i < vec_size(data); i++)
307 const char *util_strerror(int num) {
308 char *allocated = NULL;
309 static bool install = false;
310 static size_t tries = 0;
311 char **vector = util_strerror_allocated();
313 /* try installing cleanup handler */
318 install = !atexit(&util_strerror_cleanup);
322 allocated = (char*)mem_a(4096); /* A page must be enough */
323 strerror_s(allocated, 4096, num);
325 vec_push(vector, allocated);
326 return (const char *)allocated;
329 int util_snprintf(char *src, size_t bytes, const char *format, ...) {
332 va_start(va, format);
334 rt = vsprintf_s(src, bytes, format, va);
340 char *util_strcat(char *dest, const char *src) {
341 strcat_s(dest, strlen(src), src);
345 char *util_strncpy(char *dest, const char *src, size_t num) {
346 strncpy_s(dest, num, src, num);
350 const char *util_strerror(int num) {
351 return strerror(num);
354 int util_snprintf(char *src, size_t bytes, const char *format, ...) {
357 va_start(va, format);
358 rt = vsnprintf(src, bytes, format, va);
364 char *util_strcat(char *dest, const char *src) {
365 return strcat(dest, src);
368 char *util_strncpy(char *dest, const char *src, size_t num) {
369 return strncpy(dest, src, num);
372 #endif /*! _MSC_VER */
375 * Implementation of the Mersenne twister PRNG (pseudo random numer
376 * generator). Implementation of MT19937. Has a period of 2^19937-1
377 * which is a Mersenne Prime (hence the name).
379 * Implemented from specification and original paper:
380 * http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/ARTICLES/mt.pdf
382 * This code is placed in the public domain by me personally
383 * (Dale Weiler, a.k.a graphitemaster).
387 #define MT_PERIOD 397
388 #define MT_SPACE (MT_SIZE - MT_PERIOD)
390 static uint32_t mt_state[MT_SIZE];
391 static size_t mt_index = 0;
393 static GMQCC_INLINE void mt_generate(void) {
395 * The loop has been unrolled here: the original paper and implemenation
396 * Called for the following code:
397 * for (register unsigned i = 0; i < MT_SIZE; ++i) {
398 * register uint32_t load;
399 * load = (0x80000000 & mt_state[i]) // most significant 32nd bit
400 * load |= (0x7FFFFFFF & mt_state[(i + 1) % MT_SIZE]) // least significant 31nd bit
402 * mt_state[i] = mt_state[(i + MT_PERIOD) % MT_SIZE] ^ (load >> 1);
404 * if (load & 1) mt_state[i] ^= 0x9908B0DF;
407 * This essentially is a waste: we have two modulus operations, and
408 * a branch that is executed every iteration from [0, MT_SIZE).
410 * Please see: http://www.quadibloc.com/crypto/co4814.htm for more
411 * information on how this clever trick works.
413 static const uint32_t matrix[2] = {
418 * This register gives up a little more speed by instructing the compiler
419 * to force these into CPU registers (they're counters for indexing mt_state
420 * which we can force the compiler to generate prefetch instructions for)
426 * Said loop has been unrolled for MT_SPACE (226 iterations), opposed
427 * to [0, MT_SIZE) (634 iterations).
429 for (i = 0; i < MT_SPACE; ++i) {
430 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
431 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
433 i ++; /* loop unroll */
435 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
436 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
440 * collapsing the walls unrolled (evenly dividing 396 [632-227 = 396
444 while (i < MT_SIZE - 1) {
446 * We expand this 11 times .. manually, no macros are required
447 * here. This all fits in the CPU cache.
449 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
450 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
452 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
453 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
455 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
456 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
458 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
459 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
461 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
462 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
464 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
465 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
467 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
468 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
470 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
471 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
473 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
474 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
476 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
477 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
479 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
480 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
484 /* i = mt_state[623] */
485 y = (0x80000000 & mt_state[MT_SIZE - 1]) | (0x7FFFFFFF & mt_state[MT_SIZE - 1]);
486 mt_state[MT_SIZE - 1] = mt_state[MT_PERIOD - 1] ^ (y >> 1) ^ matrix[y & 1];
489 void util_seed(uint32_t value) {
491 * We seed the mt_state with a LCG (linear congruential generator)
492 * We're operating exactly on exactly m=32, so there is no need to
495 * The multipler of choice is 0x6C07865, also knows as the Borosh-
496 * Niederreiter multipler used for modulus 2^32. More can be read
497 * about this in Knuth's TAOCP Volume 2, page 106.
499 * If you don't own TAOCP something is wrong with you :-) .. so I
500 * also provided a link to the original paper by Borosh and
501 * Niederreiter. It's called "Optional Multipliers for PRNG by The
502 * Linear Congruential Method" (1983).
503 * http://en.wikipedia.org/wiki/Linear_congruential_generator
505 * From said page, it says the following:
506 * "A common Mersenne twister implementation, interestingly enough
507 * used an LCG to generate seed data."
510 * The data we're operating on is 32-bits for the mt_state array, so
511 * there is no masking required with 0xFFFFFFFF
516 for (i = 1; i < MT_SIZE; ++i)
517 mt_state[i] = 0x6C078965 * (mt_state[i - 1] ^ mt_state[i - 1] >> 30) + i;
520 uint32_t util_rand() {
524 * This is inlined with any sane compiler (I checked)
525 * for some reason though, SubC seems to be generating invalid
526 * code when it inlines this.
531 y = mt_state[mt_index];
533 /* Standard tempering */
534 y ^= y >> 11; /* +7 */
535 y ^= y << 7 & 0x9D2C5680; /* +4 */
536 y ^= y << 15 & 0xEFC60000; /* -4 */
537 y ^= y >> 18; /* -7 */
539 if(++mt_index == MT_SIZE)