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
28 /* TODO: remove globals ... */
40 struct memblock_t *next;
41 struct memblock_t *prev;
46 if (mem_hw > mem_pk) \
50 static struct memblock_t *mem_start = NULL;
52 void *util_memory_a(size_t byte, unsigned int line, const char *file) {
53 struct memblock_t *info = (struct memblock_t*)malloc(sizeof(struct memblock_t) + byte);
54 void *data = (void*)(info+1);
55 if (!info) return NULL;
60 info->next = mem_start;
62 mem_start->prev = info;
74 void util_memory_d(void *ptrn) {
75 struct memblock_t *info = NULL;
78 info = ((struct memblock_t*)ptrn - 1);
85 info->prev->next = info->next;
87 info->next->prev = info->prev;
88 if (info == mem_start)
89 mem_start = info->next;
94 void *util_memory_r(void *ptrn, size_t byte, unsigned int line, const char *file) {
95 struct memblock_t *oldinfo = NULL;
97 struct memblock_t *newinfo;
100 return util_memory_a(byte, line, file);
106 oldinfo = ((struct memblock_t*)ptrn - 1);
107 newinfo = ((struct memblock_t*)malloc(sizeof(struct memblock_t) + byte));
111 util_memory_d(oldinfo+1);
116 memcpy(newinfo+1, oldinfo+1, oldinfo->byte);
120 oldinfo->prev->next = oldinfo->next;
122 oldinfo->next->prev = oldinfo->prev;
123 if (oldinfo == mem_start)
124 mem_start = oldinfo->next;
127 newinfo->line = line;
128 newinfo->byte = byte;
129 newinfo->file = file;
130 newinfo->prev = NULL;
131 newinfo->next = mem_start;
133 mem_start->prev = newinfo;
136 mem_ab -= oldinfo->byte;
137 mem_hw -= oldinfo->byte;
138 mem_ab += newinfo->byte;
139 mem_hw += newinfo->byte;
148 static void util_dumpmem(struct memblock_t *memory, uint16_t cols) {
150 for (i = 0; i < memory->byte + ((memory->byte % cols) ? (cols - memory->byte % cols) : 0); i++) {
151 if (i % cols == 0) con_out(" 0x%06X: ", i);
152 if (i < memory->byte) con_out("%02X " , 0xFF & ((char*)(memory + 1))[i]);
155 if ((uint16_t)(i % cols) == (cols - 1)) {
156 for (j = i - (cols - 1); j <= i; j++) {
160 : (isprint(((char*)(memory + 1))[j]))
161 ? 0xFF & ((char*)(memory + 1)) [j]
170 void util_meminfo() {
171 struct memblock_t *info;
174 if (OPTS_OPTION_BOOL(OPTION_DEBUG)) {
175 for (info = mem_start; info; info = info->next) {
176 con_out("lost: %u (bytes) at %s:%u\n",
181 util_dumpmem(info, OPTS_OPTION_U16(OPTION_MEMDUMPCOLS));
185 if (OPTS_OPTION_BOOL(OPTION_DEBUG) ||
186 OPTS_OPTION_BOOL(OPTION_MEMCHK)) {
187 con_out("Memory information:\n\
188 Total allocations: %llu\n\
189 Total deallocations: %llu\n\
190 Total allocated: %f (MB)\n\
191 Total deallocated: %f (MB)\n\
192 Total peak memory: %f (MB)\n\
193 Total leaked memory: %f (MB) in %llu allocations\n",
196 (float)(mem_ab) / 1048576.0f,
197 (float)(mem_db) / 1048576.0f,
198 (float)(mem_pk) / 1048576.0f,
199 (float)(mem_ab - mem_db) / 1048576.0f,
201 /* could be more clever */
208 * Some string utility functions, because strdup uses malloc, and we want
209 * to track all memory (without replacing malloc).
211 char *_util_Estrdup(const char *s, const char *file, size_t line) {
215 /* in case of -DNOTRACK */
222 if ((len = strlen(s)) && (ptr = (char*)mem_af(len+1, line, file))) {
229 char *_util_Estrdup_empty(const char *s, const char *file, size_t line) {
233 /* in case of -DNOTRACK */
241 if ((ptr = (char*)mem_af(len+1, line, file))) {
248 void util_debug(const char *area, const char *ms, ...) {
250 if (!OPTS_OPTION_BOOL(OPTION_DEBUG))
253 if (!strcmp(area, "MEM") && !OPTS_OPTION_BOOL(OPTION_MEMCHK))
257 con_out ("[%s] ", area);
263 * only required if big endian .. otherwise no need to swap
266 #if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_BIG
267 static GMQCC_INLINE void util_swap16(uint16_t *d, size_t l) {
269 d[l] = (d[l] << 8) | (d[l] >> 8);
273 static GMQCC_INLINE void util_swap32(uint32_t *d, size_t l) {
276 v = ((d[l] << 8) & 0xFF00FF00) | ((d[l] >> 8) & 0x00FF00FF);
277 d[l] = (v << 16) | (v >> 16);
281 /* Some strange system doesn't like constants that big, AND doesn't recognize an ULL suffix
282 * so let's go the safe way
284 static GMQCC_INLINE void util_swap64(uint32_t *d, size_t l) {
288 v = ((d[l] << 8) & 0xFF00FF00FF00FF00) | ((d[l] >> 8) & 0x00FF00FF00FF00FF);
289 v = ((v << 16) & 0xFFFF0000FFFF0000) | ((v >> 16) & 0x0000FFFF0000FFFF);
290 d[l] = (v << 32) | (v >> 32);
294 for (i = 0; i < l; i += 2) {
303 void util_endianswap(void *_data, size_t length, unsigned int typesize) {
304 # if PLATFORM_BYTE_ORDER == -1 /* runtime check */
305 if (*((char*)&typesize))
308 /* prevent unused warnings */
313 # if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_LITTLE
319 util_swap16((uint16_t*)_data, length>>1);
322 util_swap32((uint32_t*)_data, length>>2);
325 util_swap64((uint32_t*)_data, length>>3);
328 default: exit(EXIT_FAILURE); /* please blow the fuck up! */
335 * CRC algorithms vary in the width of the polynomial, the value of said polynomial,
336 * the initial value used for the register, weather the bits of each byte are reflected
337 * before being processed, weather the algorithm itself feeds input bytes through the
338 * register or XORs them with a byte from one end and then straight into the table, as
339 * well as (but not limited to the idea of reflected versions) where the final register
340 * value becomes reversed, and finally weather the value itself is used to XOR the final
341 * register value. AS such you can already imagine how painfully annoying CRCs are,
342 * of course we stand to target Quake, which expects it's certian set of rules for proper
343 * calculation of a CRC.
345 * In most traditional CRC algorithms on uses a reflected table driven method where a value
346 * or register is reflected if it's bits are swapped around it's center. For example:
347 * take the bits 0101 is the 4-bit reflection of 1010, and respectfully 0011 would be the
348 * reflection of 1100. Quake however expects a NON-Reflected CRC on the output, but still
349 * requires a final XOR on the values (0xFFFF and 0x0000) this is a standard CCITT CRC-16
350 * which I respectfully as a programmer don't agree with.
352 * So now you know what we target, and why we target it, despite how unsettling it may seem
353 * but those are what Quake seems to request.
356 static const uint16_t util_crc16_table[] = {
357 0x0000, 0x1021, 0x2042, 0x3063, 0x4084, 0x50A5,
358 0x60C6, 0x70E7, 0x8108, 0x9129, 0xA14A, 0xB16B,
359 0xC18C, 0xD1AD, 0xE1CE, 0xF1EF, 0x1231, 0x0210,
360 0x3273, 0x2252, 0x52B5, 0x4294, 0x72F7, 0x62D6,
361 0x9339, 0x8318, 0xB37B, 0xA35A, 0xD3BD, 0xC39C,
362 0xF3FF, 0xE3DE, 0x2462, 0x3443, 0x0420, 0x1401,
363 0x64E6, 0x74C7, 0x44A4, 0x5485, 0xA56A, 0xB54B,
364 0x8528, 0x9509, 0xE5EE, 0xF5CF, 0xC5AC, 0xD58D,
365 0x3653, 0x2672, 0x1611, 0x0630, 0x76D7, 0x66F6,
366 0x5695, 0x46B4, 0xB75B, 0xA77A, 0x9719, 0x8738,
367 0xF7DF, 0xE7FE, 0xD79D, 0xC7BC, 0x48C4, 0x58E5,
368 0x6886, 0x78A7, 0x0840, 0x1861, 0x2802, 0x3823,
369 0xC9CC, 0xD9ED, 0xE98E, 0xF9AF, 0x8948, 0x9969,
370 0xA90A, 0xB92B, 0x5AF5, 0x4AD4, 0x7AB7, 0x6A96,
371 0x1A71, 0x0A50, 0x3A33, 0x2A12, 0xDBFD, 0xCBDC,
372 0xFBBF, 0xEB9E, 0x9B79, 0x8B58, 0xBB3B, 0xAB1A,
373 0x6CA6, 0x7C87, 0x4CE4, 0x5CC5, 0x2C22, 0x3C03,
374 0x0C60, 0x1C41, 0xEDAE, 0xFD8F, 0xCDEC, 0xDDCD,
375 0xAD2A, 0xBD0B, 0x8D68, 0x9D49, 0x7E97, 0x6EB6,
376 0x5ED5, 0x4EF4, 0x3E13, 0x2E32, 0x1E51, 0x0E70,
377 0xFF9F, 0xEFBE, 0xDFDD, 0xCFFC, 0xBF1B, 0xAF3A,
378 0x9F59, 0x8F78, 0x9188, 0x81A9, 0xB1CA, 0xA1EB,
379 0xD10C, 0xC12D, 0xF14E, 0xE16F, 0x1080, 0x00A1,
380 0x30C2, 0x20E3, 0x5004, 0x4025, 0x7046, 0x6067,
381 0x83B9, 0x9398, 0xA3FB, 0xB3DA, 0xC33D, 0xD31C,
382 0xE37F, 0xF35E, 0x02B1, 0x1290, 0x22F3, 0x32D2,
383 0x4235, 0x5214, 0x6277, 0x7256, 0xB5EA, 0xA5CB,
384 0x95A8, 0x8589, 0xF56E, 0xE54F, 0xD52C, 0xC50D,
385 0x34E2, 0x24C3, 0x14A0, 0x0481, 0x7466, 0x6447,
386 0x5424, 0x4405, 0xA7DB, 0xB7FA, 0x8799, 0x97B8,
387 0xE75F, 0xF77E, 0xC71D, 0xD73C, 0x26D3, 0x36F2,
388 0x0691, 0x16B0, 0x6657, 0x7676, 0x4615, 0x5634,
389 0xD94C, 0xC96D, 0xF90E, 0xE92F, 0x99C8, 0x89E9,
390 0xB98A, 0xA9AB, 0x5844, 0x4865, 0x7806, 0x6827,
391 0x18C0, 0x08E1, 0x3882, 0x28A3, 0xCB7D, 0xDB5C,
392 0xEB3F, 0xFB1E, 0x8BF9, 0x9BD8, 0xABBB, 0xBB9A,
393 0x4A75, 0x5A54, 0x6A37, 0x7A16, 0x0AF1, 0x1AD0,
394 0x2AB3, 0x3A92, 0xFD2E, 0xED0F, 0xDD6C, 0xCD4D,
395 0xBDAA, 0xAD8B, 0x9DE8, 0x8DC9, 0x7C26, 0x6C07,
396 0x5C64, 0x4C45, 0x3CA2, 0x2C83, 0x1CE0, 0x0CC1,
397 0xEF1F, 0xFF3E, 0xCF5D, 0xDF7C, 0xAF9B, 0xBFBA,
398 0x8FD9, 0x9FF8, 0x6E17, 0x7E36, 0x4E55, 0x5E74,
399 0x2E93, 0x3EB2, 0x0ED1, 0x1EF0
402 /* Non - Reflected */
403 uint16_t util_crc16(uint16_t current, const char *k, size_t len) {
404 register uint16_t h = current;
405 for (; len; --len, ++k)
406 h = util_crc16_table[(h>>8)^((unsigned char)*k)]^(h<<8);
409 /* Reflective Varation (for reference) */
411 uint16_t util_crc16(const char *k, int len, const short clamp) {
412 register uint16_t h= (uint16_t)0xFFFFFFFF;
413 for (; len; --len, ++k)
414 h = util_crc16_table[(h^((unsigned char)*k))&0xFF]^(h>>8);
419 size_t util_strtocmd(const char *in, char *out, size_t outsz) {
421 for (; *in && sz < outsz; ++in, ++out, ++sz)
422 *out = (*in == '-') ? '_' : (isalpha(*in) && !isupper(*in)) ? *in + 'A' - 'a': *in;
427 size_t util_strtononcmd(const char *in, char *out, size_t outsz) {
429 for (; *in && sz < outsz; ++in, ++out, ++sz)
430 *out = (*in == '_') ? '-' : (isalpha(*in) && isupper(*in)) ? *in + 'a' - 'A' : *in;
435 /* TODO: rewrite ... when I redo the ve cleanup */
436 void _util_vec_grow(void **a, size_t i, size_t s) {
437 vector_t *d = vec_meta(*a);
438 size_t m = *a ? 2 * d->allocated +i : i+1;
439 void *p = mem_r((*a ? d : NULL), s * m + sizeof(vector_t));
442 ((vector_t*)p)->used = 0;
443 *a = (vector_t*)p + 1;
445 vec_meta(*a)->allocated = m;
449 * Hash table for generic data, based on dynamic memory allocations
450 * all around. This is the internal interface, please look for
451 * EXPOSED INTERFACE comment below
453 typedef struct hash_node_t {
454 char *key; /* the key for this node in table */
455 void *value; /* pointer to the data as void* */
456 struct hash_node_t *next; /* next node (linked list) */
459 GMQCC_INLINE size_t util_hthash(hash_table_t *ht, const char *key) {
460 const uint32_t mix = 0x5BD1E995;
461 const uint32_t rot = 24;
462 size_t size = strlen(key);
463 uint32_t hash = 0x1EF0 /* LICRC TAB */ ^ size;
465 const unsigned char *data = (const unsigned char*)key;
468 alias = *(uint32_t*)data;
471 alias ^= alias >> rot;
482 case 3: hash ^= data[2] << 16;
483 case 2: hash ^= data[1] << 8;
484 case 1: hash ^= data[0];
492 return (size_t) (hash % ht->size);
495 hash_node_t *_util_htnewpair(const char *key, void *value) {
497 if (!(node = (hash_node_t*)mem_a(sizeof(hash_node_t))))
500 if (!(node->key = util_strdupe(key))) {
512 * EXPOSED INTERFACE for the hashtable implementation
513 * util_htnew(size) -- to make a new hashtable
514 * util_htset(table, key, value, sizeof(value)) -- to set something in the table
515 * util_htget(table, key) -- to get something from the table
516 * util_htdel(table) -- to delete the table
518 hash_table_t *util_htnew(size_t size) {
519 hash_table_t *hashtable = NULL;
523 if (!(hashtable = (hash_table_t*)mem_a(sizeof(hash_table_t))))
526 if (!(hashtable->table = (hash_node_t**)mem_a(sizeof(hash_node_t*) * size))) {
531 hashtable->size = size;
532 memset(hashtable->table, 0, sizeof(hash_node_t*) * size);
537 void util_htseth(hash_table_t *ht, const char *key, size_t bin, void *value) {
538 hash_node_t *newnode = NULL;
539 hash_node_t *next = NULL;
540 hash_node_t *last = NULL;
542 next = ht->table[bin];
544 while (next && next->key && strcmp(key, next->key) > 0)
545 last = next, next = next->next;
547 /* already in table, do a replace */
548 if (next && next->key && strcmp(key, next->key) == 0) {
551 /* not found, grow a pair man :P */
552 newnode = _util_htnewpair(key, value);
553 if (next == ht->table[bin]) {
554 newnode->next = next;
555 ht->table[bin] = newnode;
557 last->next = newnode;
559 newnode->next = next;
560 last->next = newnode;
565 void util_htset(hash_table_t *ht, const char *key, void *value) {
566 util_htseth(ht, key, util_hthash(ht, key), value);
569 void *util_htgeth(hash_table_t *ht, const char *key, size_t bin) {
570 hash_node_t *pair = ht->table[bin];
572 while (pair && pair->key && strcmp(key, pair->key) > 0)
575 if (!pair || !pair->key || strcmp(key, pair->key) != 0)
581 void *util_htget(hash_table_t *ht, const char *key) {
582 return util_htgeth(ht, key, util_hthash(ht, key));
585 void *code_util_str_htgeth(hash_table_t *ht, const char *key, size_t bin) {
590 keylen = strlen(key);
592 pair = ht->table[bin];
593 while (pair && pair->key) {
594 len = strlen(pair->key);
600 cmp = strcmp(key, pair->key);
608 cmp = strcmp(key, pair->key + len - keylen);
610 uintptr_t up = (uintptr_t)pair->value;
620 * Free all allocated data in a hashtable, this is quite the amount
623 void util_htrem(hash_table_t *ht, void (*callback)(void *data)) {
625 for (; i < ht->size; i++) {
626 hash_node_t *n = ht->table[i];
646 void util_htrmh(hash_table_t *ht, const char *key, size_t bin, void (*cb)(void*)) {
647 hash_node_t **pair = &ht->table[bin];
650 while (*pair && (*pair)->key && strcmp(key, (*pair)->key) > 0)
651 pair = &(*pair)->next;
654 if (!tmp || !tmp->key || strcmp(key, tmp->key) != 0)
665 void util_htrm(hash_table_t *ht, const char *key, void (*cb)(void*)) {
666 util_htrmh(ht, key, util_hthash(ht, key), cb);
669 void util_htdel(hash_table_t *ht) {
670 util_htrem(ht, NULL);
674 * Portable implementation of vasprintf/asprintf. Assumes vsnprintf
675 * exists, otherwise compiler error.
677 * TODO: fix for MSVC ....
679 int util_vasprintf(char **dat, const char *fmt, va_list args) {
685 * For visuals tido _vsnprintf doesn't tell you the length of a
686 * formatted string if it overflows. However there is a MSVC
687 * intrinsic (which is documented wrong) called _vcsprintf which
688 * will return the required amount to allocate.
692 if ((len = _vscprintf(fmt, args)) < 0) {
697 tmp = mem_a(len + 1);
698 if ((ret = _vsnprintf(tmp, len+1, fmt, args)) != len) {
707 * For everything else we have a decent conformint vsnprintf that
708 * returns the number of bytes needed. We give it a try though on
709 * a short buffer, since efficently speaking, it could be nice to
710 * above a second vsnprintf call.
715 len = vsnprintf(buf, sizeof(buf), fmt, cpy);
718 if (len < (int)sizeof(buf)) {
719 *dat = util_strdup(buf);
723 /* not large enough ... */
724 tmp = (char*)mem_a(len + 1);
725 if ((ret = vsnprintf(tmp, len + 1, fmt, args)) != len) {
735 int util_asprintf(char **ret, const char *fmt, ...) {
739 read = util_vasprintf(ret, fmt, args);
746 * Implementation of the Mersenne twister PRNG (pseudo random numer
747 * generator). Implementation of MT19937. Has a period of 2^19937-1
748 * which is a Mersenne Prime (hence the name).
750 * Implemented from specification and original paper:
751 * http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/ARTICLES/mt.pdf
753 * This code is placed in the public domain by me personally
754 * (Dale Weiler, a.k.a graphitemaster).
758 #define MT_PERIOD 397
759 #define MT_SPACE (MT_SIZE - MT_PERIOD)
761 static uint32_t mt_state[MT_SIZE];
762 static size_t mt_index = 0;
764 static GMQCC_INLINE void mt_generate() {
766 * The loop has been unrolled here: the original paper and implemenation
767 * Called for the following code:
768 * for (register unsigned i = 0; i < MT_SIZE; ++i) {
769 * register uint32_t load;
770 * load = (0x80000000 & mt_state[i]) // most significant 32nd bit
771 * load |= (0x7FFFFFFF & mt_state[(i + 1) % MT_SIZE]) // least significant 31nd bit
773 * mt_state[i] = mt_state[(i + MT_PERIOD) % MT_SIZE] ^ (load >> 1);
775 * if (load & 1) mt_state[i] ^= 0x9908B0DF;
778 * This essentially is a waste: we have two modulus operations, and
779 * a branch that is executed every iteration from [0, MT_SIZE).
781 * Please see: http://www.quadibloc.com/crypto/co4814.htm for more
782 * information on how this clever trick works.
784 static const uint32_t matrix[2] = {
789 * This register gives up a little more speed by instructing the compiler
790 * to force these into CPU registers (they're counters for indexing mt_state
791 * which we can force the compiler to generate prefetch instructions for)
797 * Said loop has been unrolled for MT_SPACE (226 iterations), opposed
798 * to [0, MT_SIZE) (634 iterations).
800 for (i = 0; i < MT_SPACE; ++i) {
801 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
802 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
804 i ++; /* loop unroll */
806 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
807 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
811 * collapsing the walls unrolled (evenly dividing 396 [632-227 = 396
815 while (i < MT_SIZE - 1) {
817 * We expand this 11 times .. manually, no macros are required
818 * here. This all fits in the CPU cache.
820 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
821 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
823 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
824 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
826 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
827 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
829 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
830 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
832 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
833 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
835 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
836 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
838 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
839 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
841 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
842 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
844 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
845 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
847 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
848 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
850 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
851 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
855 /* i = mt_state[623] */
856 y = (0x80000000 & mt_state[MT_SIZE - 1]) | (0x7FFFFFFF & mt_state[MT_SIZE - 1]);
857 mt_state[MT_SIZE - 1] = mt_state[MT_PERIOD - 1] ^ (y >> 1) ^ matrix[y & 1];
860 void util_seed(uint32_t value) {
862 * We seed the mt_state with a LCG (linear congruential generator)
863 * We're operating exactly on exactly m=32, so there is no need to
866 * The multipler of choice is 0x6C07865, also knows as the Borosh-
867 * Niederreiter multipler used for modulus 2^32. More can be read
868 * about this in Knuth's TAOCP Volume 2, page 106.
870 * If you don't own TAOCP something is wrong with you :-) .. so I
871 * also provided a link to the original paper by Borosh and
872 * Niederreiter. It's called "Optional Multipliers for PRNG by The
873 * Linear Congruential Method" (1983).
874 * http://en.wikipedia.org/wiki/Linear_congruential_generator
876 * From said page, it says the following:
877 * "A common Mersenne twister implementation, interestingly enough
878 * used an LCG to generate seed data."
881 * The data we're operating on is 32-bits for the mt_state array, so
882 * there is no masking required with 0xFFFFFFFF
887 for (i = 1; i < MT_SIZE; ++i)
888 mt_state[i] = 0x6C078965 * (mt_state[i - 1] ^ mt_state[i - 1] >> 30) + i;
891 uint32_t util_rand() {
895 * This is inlined with any sane compiler (I checked)
896 * for some reason though, SubC seems to be generating invalid
897 * code when it inlines this.
902 y = mt_state[mt_index];
904 /* Standard tempering */
905 y ^= y >> 11; /* +7 */
906 y ^= y << 7 & 0x9D2C5680; /* +4 */
907 y ^= y << 15 & 0xEFC60000; /* -4 */
908 y ^= y >> 18; /* -7 */
910 if(++mt_index == MT_SIZE)