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 void util_debug(const char *area, const char *ms, ...) {
231 if (!OPTS_OPTION_BOOL(OPTION_DEBUG))
234 if (!strcmp(area, "MEM") && !OPTS_OPTION_BOOL(OPTION_MEMCHK))
238 con_out ("[%s] ", area);
244 * only required if big endian .. otherwise no need to swap
247 #if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_BIG
248 static GMQCC_INLINE void util_swap16(uint16_t *d, size_t l) {
250 d[l] = (d[l] << 8) | (d[l] >> 8);
254 static GMQCC_INLINE void util_swap32(uint32_t *d, size_t l) {
257 v = ((d[l] << 8) & 0xFF00FF00) | ((d[l] >> 8) & 0x00FF00FF);
258 d[l] = (v << 16) | (v >> 16);
262 /* Some strange system doesn't like constants that big, AND doesn't recognize an ULL suffix
263 * so let's go the safe way
265 static GMQCC_INLINE void util_swap64(uint32_t *d, size_t l) {
269 v = ((d[l] << 8) & 0xFF00FF00FF00FF00) | ((d[l] >> 8) & 0x00FF00FF00FF00FF);
270 v = ((v << 16) & 0xFFFF0000FFFF0000) | ((v >> 16) & 0x0000FFFF0000FFFF);
271 d[l] = (v << 32) | (v >> 32);
275 for (i = 0; i < l; i += 2) {
284 void util_endianswap(void *_data, size_t length, unsigned int typesize) {
285 # if PLATFORM_BYTE_ORDER == -1 /* runtime check */
286 if (*((char*)&typesize))
289 /* prevent unused warnings */
294 # if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_LITTLE
300 util_swap16((uint16_t*)_data, length>>1);
303 util_swap32((uint32_t*)_data, length>>2);
306 util_swap64((uint32_t*)_data, length>>3);
309 default: exit(EXIT_FAILURE); /* please blow the fuck up! */
316 * CRC algorithms vary in the width of the polynomial, the value of said polynomial,
317 * the initial value used for the register, weather the bits of each byte are reflected
318 * before being processed, weather the algorithm itself feeds input bytes through the
319 * register or XORs them with a byte from one end and then straight into the table, as
320 * well as (but not limited to the idea of reflected versions) where the final register
321 * value becomes reversed, and finally weather the value itself is used to XOR the final
322 * register value. AS such you can already imagine how painfully annoying CRCs are,
323 * of course we stand to target Quake, which expects it's certian set of rules for proper
324 * calculation of a CRC.
326 * In most traditional CRC algorithms on uses a reflected table driven method where a value
327 * or register is reflected if it's bits are swapped around it's center. For example:
328 * take the bits 0101 is the 4-bit reflection of 1010, and respectfully 0011 would be the
329 * reflection of 1100. Quake however expects a NON-Reflected CRC on the output, but still
330 * requires a final XOR on the values (0xFFFF and 0x0000) this is a standard CCITT CRC-16
331 * which I respectfully as a programmer don't agree with.
333 * So now you know what we target, and why we target it, despite how unsettling it may seem
334 * but those are what Quake seems to request.
337 static const uint16_t util_crc16_table[] = {
338 0x0000, 0x1021, 0x2042, 0x3063, 0x4084, 0x50A5,
339 0x60C6, 0x70E7, 0x8108, 0x9129, 0xA14A, 0xB16B,
340 0xC18C, 0xD1AD, 0xE1CE, 0xF1EF, 0x1231, 0x0210,
341 0x3273, 0x2252, 0x52B5, 0x4294, 0x72F7, 0x62D6,
342 0x9339, 0x8318, 0xB37B, 0xA35A, 0xD3BD, 0xC39C,
343 0xF3FF, 0xE3DE, 0x2462, 0x3443, 0x0420, 0x1401,
344 0x64E6, 0x74C7, 0x44A4, 0x5485, 0xA56A, 0xB54B,
345 0x8528, 0x9509, 0xE5EE, 0xF5CF, 0xC5AC, 0xD58D,
346 0x3653, 0x2672, 0x1611, 0x0630, 0x76D7, 0x66F6,
347 0x5695, 0x46B4, 0xB75B, 0xA77A, 0x9719, 0x8738,
348 0xF7DF, 0xE7FE, 0xD79D, 0xC7BC, 0x48C4, 0x58E5,
349 0x6886, 0x78A7, 0x0840, 0x1861, 0x2802, 0x3823,
350 0xC9CC, 0xD9ED, 0xE98E, 0xF9AF, 0x8948, 0x9969,
351 0xA90A, 0xB92B, 0x5AF5, 0x4AD4, 0x7AB7, 0x6A96,
352 0x1A71, 0x0A50, 0x3A33, 0x2A12, 0xDBFD, 0xCBDC,
353 0xFBBF, 0xEB9E, 0x9B79, 0x8B58, 0xBB3B, 0xAB1A,
354 0x6CA6, 0x7C87, 0x4CE4, 0x5CC5, 0x2C22, 0x3C03,
355 0x0C60, 0x1C41, 0xEDAE, 0xFD8F, 0xCDEC, 0xDDCD,
356 0xAD2A, 0xBD0B, 0x8D68, 0x9D49, 0x7E97, 0x6EB6,
357 0x5ED5, 0x4EF4, 0x3E13, 0x2E32, 0x1E51, 0x0E70,
358 0xFF9F, 0xEFBE, 0xDFDD, 0xCFFC, 0xBF1B, 0xAF3A,
359 0x9F59, 0x8F78, 0x9188, 0x81A9, 0xB1CA, 0xA1EB,
360 0xD10C, 0xC12D, 0xF14E, 0xE16F, 0x1080, 0x00A1,
361 0x30C2, 0x20E3, 0x5004, 0x4025, 0x7046, 0x6067,
362 0x83B9, 0x9398, 0xA3FB, 0xB3DA, 0xC33D, 0xD31C,
363 0xE37F, 0xF35E, 0x02B1, 0x1290, 0x22F3, 0x32D2,
364 0x4235, 0x5214, 0x6277, 0x7256, 0xB5EA, 0xA5CB,
365 0x95A8, 0x8589, 0xF56E, 0xE54F, 0xD52C, 0xC50D,
366 0x34E2, 0x24C3, 0x14A0, 0x0481, 0x7466, 0x6447,
367 0x5424, 0x4405, 0xA7DB, 0xB7FA, 0x8799, 0x97B8,
368 0xE75F, 0xF77E, 0xC71D, 0xD73C, 0x26D3, 0x36F2,
369 0x0691, 0x16B0, 0x6657, 0x7676, 0x4615, 0x5634,
370 0xD94C, 0xC96D, 0xF90E, 0xE92F, 0x99C8, 0x89E9,
371 0xB98A, 0xA9AB, 0x5844, 0x4865, 0x7806, 0x6827,
372 0x18C0, 0x08E1, 0x3882, 0x28A3, 0xCB7D, 0xDB5C,
373 0xEB3F, 0xFB1E, 0x8BF9, 0x9BD8, 0xABBB, 0xBB9A,
374 0x4A75, 0x5A54, 0x6A37, 0x7A16, 0x0AF1, 0x1AD0,
375 0x2AB3, 0x3A92, 0xFD2E, 0xED0F, 0xDD6C, 0xCD4D,
376 0xBDAA, 0xAD8B, 0x9DE8, 0x8DC9, 0x7C26, 0x6C07,
377 0x5C64, 0x4C45, 0x3CA2, 0x2C83, 0x1CE0, 0x0CC1,
378 0xEF1F, 0xFF3E, 0xCF5D, 0xDF7C, 0xAF9B, 0xBFBA,
379 0x8FD9, 0x9FF8, 0x6E17, 0x7E36, 0x4E55, 0x5E74,
380 0x2E93, 0x3EB2, 0x0ED1, 0x1EF0
383 /* Non - Reflected */
384 uint16_t util_crc16(uint16_t current, const char *k, size_t len) {
385 register uint16_t h = current;
386 for (; len; --len, ++k)
387 h = util_crc16_table[(h>>8)^((unsigned char)*k)]^(h<<8);
390 /* Reflective Varation (for reference) */
392 uint16_t util_crc16(const char *k, int len, const short clamp) {
393 register uint16_t h= (uint16_t)0xFFFFFFFF;
394 for (; len; --len, ++k)
395 h = util_crc16_table[(h^((unsigned char)*k))&0xFF]^(h>>8);
400 size_t util_strtocmd(const char *in, char *out, size_t outsz) {
402 for (; *in && sz < outsz; ++in, ++out, ++sz)
403 *out = (*in == '-') ? '_' : (isalpha(*in) && !isupper(*in)) ? *in + 'A' - 'a': *in;
408 size_t util_strtononcmd(const char *in, char *out, size_t outsz) {
410 for (; *in && sz < outsz; ++in, ++out, ++sz)
411 *out = (*in == '_') ? '-' : (isalpha(*in) && isupper(*in)) ? *in + 'a' - 'A' : *in;
416 /* TODO: rewrite ... when I redo the ve cleanup */
417 void _util_vec_grow(void **a, size_t i, size_t s) {
418 vector_t *d = vec_meta(*a);
419 size_t m = *a ? 2 * d->allocated +i : i+1;
420 void *p = mem_r((*a ? d : NULL), s * m + sizeof(vector_t));
423 ((vector_t*)p)->used = 0;
424 *a = (vector_t*)p + 1;
426 vec_meta(*a)->allocated = m;
430 * Hash table for generic data, based on dynamic memory allocations
431 * all around. This is the internal interface, please look for
432 * EXPOSED INTERFACE comment below
434 typedef struct hash_node_t {
435 char *key; /* the key for this node in table */
436 void *value; /* pointer to the data as void* */
437 struct hash_node_t *next; /* next node (linked list) */
440 GMQCC_INLINE size_t util_hthash(hash_table_t *ht, const char *key) {
441 const uint32_t mix = 0x5BD1E995;
442 const uint32_t rot = 24;
443 size_t size = strlen(key);
444 uint32_t hash = 0x1EF0 /* LICRC TAB */ ^ size;
446 const unsigned char *data = (const unsigned char*)key;
449 alias = *(uint32_t*)data;
452 alias ^= alias >> rot;
463 case 3: hash ^= data[2] << 16;
464 case 2: hash ^= data[1] << 8;
465 case 1: hash ^= data[0];
473 return (size_t) (hash % ht->size);
476 hash_node_t *_util_htnewpair(const char *key, void *value) {
478 if (!(node = (hash_node_t*)mem_a(sizeof(hash_node_t))))
481 if (!(node->key = util_strdup(key))) {
493 * EXPOSED INTERFACE for the hashtable implementation
494 * util_htnew(size) -- to make a new hashtable
495 * util_htset(table, key, value, sizeof(value)) -- to set something in the table
496 * util_htget(table, key) -- to get something from the table
497 * util_htdel(table) -- to delete the table
499 hash_table_t *util_htnew(size_t size) {
500 hash_table_t *hashtable = NULL;
504 if (!(hashtable = (hash_table_t*)mem_a(sizeof(hash_table_t))))
507 if (!(hashtable->table = (hash_node_t**)mem_a(sizeof(hash_node_t*) * size))) {
512 hashtable->size = size;
513 memset(hashtable->table, 0, sizeof(hash_node_t*) * size);
518 void util_htseth(hash_table_t *ht, const char *key, size_t bin, void *value) {
519 hash_node_t *newnode = NULL;
520 hash_node_t *next = NULL;
521 hash_node_t *last = NULL;
523 next = ht->table[bin];
525 while (next && next->key && strcmp(key, next->key) > 0)
526 last = next, next = next->next;
528 /* already in table, do a replace */
529 if (next && next->key && strcmp(key, next->key) == 0) {
532 /* not found, grow a pair man :P */
533 newnode = _util_htnewpair(key, value);
534 if (next == ht->table[bin]) {
535 newnode->next = next;
536 ht->table[bin] = newnode;
538 last->next = newnode;
540 newnode->next = next;
541 last->next = newnode;
546 void util_htset(hash_table_t *ht, const char *key, void *value) {
547 util_htseth(ht, key, util_hthash(ht, key), value);
550 void *util_htgeth(hash_table_t *ht, const char *key, size_t bin) {
551 hash_node_t *pair = ht->table[bin];
553 while (pair && pair->key && strcmp(key, pair->key) > 0)
556 if (!pair || !pair->key || strcmp(key, pair->key) != 0)
562 void *util_htget(hash_table_t *ht, const char *key) {
563 return util_htgeth(ht, key, util_hthash(ht, key));
566 void *code_util_str_htgeth(hash_table_t *ht, const char *key, size_t bin) {
571 keylen = strlen(key);
573 pair = ht->table[bin];
574 while (pair && pair->key) {
575 len = strlen(pair->key);
581 cmp = strcmp(key, pair->key);
589 cmp = strcmp(key, pair->key + len - keylen);
591 uintptr_t up = (uintptr_t)pair->value;
601 * Free all allocated data in a hashtable, this is quite the amount
604 void util_htdel(hash_table_t *ht) {
606 for (; i < ht->size; i++) {
607 hash_node_t *n = ht->table[i];
626 * A basic implementation of a hash-set. Unlike a hashtable, a hash
627 * set doesn't maintain key-value pairs. It simply maintains a key
628 * that can be set, removed, and checked for.
630 * See EXPOSED interface comment below
632 #define GMQCC_HASHSET_PRIME0 0x0049
633 #define GMQCC_HASHSET_PRIME1 0x1391
635 static int util_hsput(hash_set_t *set, void *item) {
636 size_t hash = (size_t)item; /* shouldn't drop the bits */
639 /* a == 0 || a == 1 */
643 iter = set->mask & (GMQCC_HASHSET_PRIME0 * hash);
645 /* while (set->items[iter] != 0 && set->items[iter] != 1) */
646 while (!(set->items[iter] >> 1)) {
647 if (set->items[iter] == hash)
650 iter = set->mask & (iter + GMQCC_HASHSET_PRIME1);
654 set->items[iter] = hash;
659 static void util_hsupdate(hash_set_t *set) {
664 /* time to rehash? */
665 if ((float)set->total >= (size_t)((double)set->capacity * 0.85)) {
670 set->capacity = (size_t)(1 << set->bits);
671 set->mask = set->capacity - 1;
672 set->items = (size_t*)mem_a(set->capacity * sizeof(size_t));
675 /*assert(set->items);*/
678 * this shouldn't be slow? if so unroll it a little perhaps
679 * (shouldn't be though)
681 for (itr = 0; itr < end; itr++)
682 util_hsput(set, (void*)old[itr]);
689 * EXPOSED interface: all of these functions are exposed to the outside
690 * for use. The stuff above is static because it's the "internal" mechanics
691 * for syncronizing the set for updating, and putting data into the set.
693 int util_hsadd(hash_set_t *set, void *item) {
694 int run = util_hsput(set, item); /* inlined */
700 /* remove item in set */
701 int util_hsrem(hash_set_t *set, void *item) {
702 size_t hash = (size_t)item;
703 size_t iter = set->mask & (GMQCC_HASHSET_PRIME0 * hash);
705 while (set->items[iter]) {
706 if (set->items[iter] == hash) {
707 set->items[iter] = 1;
712 iter = set->mask & (iter + GMQCC_HASHSET_PRIME1);
718 /* check if item is set */
719 int util_hshas(hash_set_t *set, void *item) {
720 size_t hash = (size_t)item;
721 size_t iter = set->mask & (GMQCC_HASHSET_PRIME0 * hash);
723 while (set->items[iter]) {
724 if (set->items[iter] == hash)
727 iter = set->mask & (iter + GMQCC_HASHSET_PRIME1);
733 hash_set_t *util_hsnew(void) {
736 if (!(set = (hash_set_t*)mem_a(sizeof(hash_set_t))))
741 set->capacity = (size_t)(1 << set->bits);
742 set->mask = set->capacity - 1;
743 set->items = (size_t*)mem_a(set->capacity * sizeof(size_t));
753 void util_hsdel(hash_set_t *set) {
761 #undef GMQCC_HASHSET_PRIME0
762 #undef GMQCC_HASHSET_PRIME1
766 * Portable implementation of vasprintf/asprintf. Assumes vsnprintf
767 * exists, otherwise compiler error.
769 * TODO: fix for MSVC ....
771 int util_vasprintf(char **dat, const char *fmt, va_list args) {
777 * For visuals tido _vsnprintf doesn't tell you the length of a
778 * formatted string if it overflows. However there is a MSVC
779 * intrinsic (which is documented wrong) called _vcsprintf which
780 * will return the required amount to allocate.
784 if ((len = _vscprintf(fmt, args)) < 0) {
789 tmp = mem_a(len + 1);
790 if ((ret = _vsnprintf(tmp, len+1, fmt, args)) != len) {
799 * For everything else we have a decent conformint vsnprintf that
800 * returns the number of bytes needed. We give it a try though on
801 * a short buffer, since efficently speaking, it could be nice to
802 * above a second vsnprintf call.
807 len = vsnprintf(buf, sizeof(buf), fmt, cpy);
810 if (len < (int)sizeof(buf)) {
811 *dat = util_strdup(buf);
815 /* not large enough ... */
816 tmp = (char*)mem_a(len + 1);
817 if ((ret = vsnprintf(tmp, len + 1, fmt, args)) != len) {
827 int util_asprintf(char **ret, const char *fmt, ...) {
831 read = util_vasprintf(ret, fmt, args);
838 * Implementation of the Mersenne twister PRNG (pseudo random numer
839 * generator). Implementation of MT19937. Has a period of 2^19937-1
840 * which is a Mersenne Prime (hence the name).
842 * Implemented from specification and original paper:
843 * http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/ARTICLES/mt.pdf
845 * This code is placed in the public domain by me personally
846 * (Dale Weiler, a.k.a graphitemaster).
850 #define MT_PERIOD 397
851 #define MT_SPACE (MT_SIZE - MT_PERIOD)
853 static uint32_t mt_state[MT_SIZE];
854 static size_t mt_index = 0;
856 static GMQCC_INLINE void mt_generate() {
858 * The loop has been unrolled here: the original paper and implemenation
859 * Called for the following code:
860 * for (register unsigned i = 0; i < MT_SIZE; ++i) {
861 * register uint32_t load;
862 * load = (0x80000000 & mt_state[i]) // most significant 32nd bit
863 * load |= (0x7FFFFFFF & mt_state[(i + 1) % MT_SIZE]) // least significant 31nd bit
865 * mt_state[i] = mt_state[(i + MT_PERIOD) % MT_SIZE] ^ (load >> 1);
867 * if (load & 1) mt_state[i] ^= 0x9908B0DF;
870 * This essentially is a waste: we have two modulus operations, and
871 * a branch that is executed every iteration from [0, MT_SIZE).
873 * Please see: http://www.quadibloc.com/crypto/co4814.htm for more
874 * information on how this clever trick works.
876 static const uint32_t matrix[2] = {
881 * This register gives up a little more speed by instructing the compiler
882 * to force these into CPU registers (they're counters for indexing mt_state
883 * which we can force the compiler to generate prefetch instructions for)
889 * Said loop has been unrolled for MT_SPACE (226 iterations), opposed
890 * to [0, MT_SIZE) (634 iterations).
892 for (i = 0; i < MT_SPACE; ++i) {
893 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
894 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
896 i ++; /* loop unroll */
898 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
899 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
903 * collapsing the walls unrolled (evenly dividing 396 [632-227 = 396
907 while (i < MT_SIZE - 1) {
909 * We expand this 11 times .. manually, no macros are required
910 * here. This all fits in the CPU cache.
912 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
913 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
915 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
916 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
918 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
919 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
921 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
922 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
924 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
925 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
927 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
928 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
930 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
931 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
933 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
934 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
936 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
937 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
939 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
940 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
942 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
943 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
947 /* i = mt_state[623] */
948 y = (0x80000000 & mt_state[MT_SIZE - 1]) | (0x7FFFFFFF & mt_state[MT_SIZE - 1]);
949 mt_state[MT_SIZE - 1] = mt_state[MT_PERIOD - 1] ^ (y >> 1) ^ matrix[y & 1];
952 void util_seed(uint32_t value) {
954 * We seed the mt_state with a LCG (linear congruential generator)
955 * We're operating exactly on exactly m=32, so there is no need to
958 * The multipler of choice is 0x6C07865, also knows as the Borosh-
959 * Niederreiter multipler used for modulus 2^32. More can be read
960 * about this in Knuth's TAOCP Volume 2, page 106.
962 * If you don't own TAOCP something is wrong with you :-) .. so I
963 * also provided a link to the original paper by Borosh and
964 * Niederreiter. It's called "Optional Multipliers for PRNG by The
965 * Linear Congruential Method" (1983).
966 * http://en.wikipedia.org/wiki/Linear_congruential_generator
968 * From said page, it says the following:
969 * "A common Mersenne twister implementation, interestingly enough
970 * used an LCG to generate seed data."
973 * The data we're operating on is 32-bits for the mt_state array, so
974 * there is no masking required with 0xFFFFFFFF
979 for (i = 1; i < MT_SIZE; ++i)
980 mt_state[i] = 0x6C078965 * (mt_state[i - 1] ^ mt_state[i - 1] >> 30) + i;
983 uint32_t util_rand() {
987 * This is inlined with any sane compiler (I checked)
988 * for some reason though, SubC seems to be generating invalid
989 * code when it inlines this.
994 y = mt_state[mt_index];
996 /* Standard tempering */
997 y ^= y >> 11; /* +7 */
998 y ^= y << 7 & 0x9D2C5680; /* +4 */
999 y ^= y << 15 & 0xEFC60000; /* -4 */
1000 y ^= y >> 18; /* -7 */
1002 if(++mt_index == MT_SIZE)