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 ... */
29 static uint64_t mem_ab = 0;
30 static uint64_t mem_db = 0;
31 static uint64_t mem_at = 0;
32 static uint64_t mem_dt = 0;
33 static uint64_t mem_pk = 0;
34 static uint64_t mem_hw = 0;
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]
171 * The following is a VERY tight, efficent, hashtable for integer
172 * values and keys, and for nothing more. We could make our existing
173 * hashtable support type-genericness through a void * pointer but,
174 * ideally that would make things more complicated. We also don't need
175 * that much of a bloat for something as basic as this.
183 typedef size_entry_t **size_table_t;
185 static size_table_t util_st_new() {
186 return (size_table_t)memset(
187 mem_a(sizeof(size_entry_t*) * ST_SIZE),
188 0, ST_SIZE * sizeof(size_entry_t*)
191 static void util_st_del(size_table_t table) {
193 for (; i < ST_SIZE; i++) if(table[i]) mem_d(table[i]);
196 static size_entry_t *util_st_get(size_table_t table, size_t key) {
197 size_t hash = (key % ST_SIZE);
198 while (table[hash] && table[hash]->key != key)
199 hash = (hash + 1) % ST_SIZE;
202 static void util_st_put(size_table_t table, size_t key, size_t value) {
203 size_t hash = (key % ST_SIZE);
204 while (table[hash] && table[hash]->key != key)
205 hash = (hash + 1) % ST_SIZE;
206 table[hash] = (size_entry_t*)mem_a(sizeof(size_entry_t));
207 table[hash]->key = key;
208 table[hash]->value = value;
211 static uint64_t strdups = 0;
212 static uint64_t vectors = 0;
213 static uint64_t vector_sizes = 0;
214 static uint64_t hashtables = 0;
215 static uint64_t hashtable_sizes = 0;
216 static size_table_t vector_usage = NULL;
217 static size_table_t hashtable_usage = NULL;
219 void util_meminfo() {
220 struct memblock_t *info;
222 if (OPTS_OPTION_BOOL(OPTION_DEBUG)) {
223 for (info = mem_start; info; info = info->next) {
224 con_out("lost: %u (bytes) at %s:%u\n",
229 util_dumpmem(info, OPTS_OPTION_U16(OPTION_MEMDUMPCOLS));
233 if (OPTS_OPTION_BOOL(OPTION_DEBUG) ||
234 OPTS_OPTION_BOOL(OPTION_MEMCHK)) {
235 con_out("Memory information:\n\
236 Total allocations: %llu\n\
237 Total deallocations: %llu\n\
238 Total allocated: %f (MB)\n\
239 Total deallocated: %f (MB)\n\
240 Total peak memory: %f (MB)\n\
241 Total leaked memory: %f (MB) in %llu allocations\n",
244 (float)(mem_ab) / 1048576.0f,
245 (float)(mem_db) / 1048576.0f,
246 (float)(mem_pk) / 1048576.0f,
247 (float)(mem_ab - mem_db) / 1048576.0f,
249 /* could be more clever */
254 if (OPTS_OPTION_BOOL(OPTION_STATISTICS) ||
255 OPTS_OPTION_BOOL(OPTION_MEMCHK)) {
258 uint64_t vectormem = 0;
260 con_out("\nAdditional Statistics:\n\
261 Total vectors allocated: %llu\n\
262 Total string duplicates: %llu\n\
263 Total hashtables allocated: %llu\n\
264 Total unique vector sizes: %llu\n",
271 for (; i < ST_SIZE; i++) {
274 if (!(entry = vector_usage[i]))
277 con_out(" %2u| # of %4u byte vectors: %u\n",
279 (unsigned)entry->key,
280 (unsigned)entry->value
284 vectormem += entry->key * entry->value;
288 Total unique hashtable sizes: %llu\n",
292 for (i = 0, e = 1; i < ST_SIZE; i++) {
295 if (!(entry = hashtable_usage[i]))
298 con_out(" %2u| # of %4u element hashtables: %u\n",
300 (unsigned)entry->key,
301 (unsigned)entry->value
306 con_out(" Total vector memory: %f (MB)\n",
307 (float)(vectormem) / 1048576.0f
312 util_st_del(vector_usage);
314 util_st_del(hashtable_usage);
318 * Some string utility functions, because strdup uses malloc, and we want
319 * to track all memory (without replacing malloc).
321 char *_util_Estrdup(const char *s, const char *file, size_t line) {
325 /* in case of -DNOTRACK */
332 if ((len = strlen(s)) && (ptr = (char*)mem_af(len+1, line, file))) {
340 char *_util_Estrdup_empty(const char *s, const char *file, size_t line) {
344 /* in case of -DNOTRACK */
352 if ((ptr = (char*)mem_af(len+1, line, file))) {
360 void util_debug(const char *area, const char *ms, ...) {
362 if (!OPTS_OPTION_BOOL(OPTION_DEBUG))
365 if (!strcmp(area, "MEM") && !OPTS_OPTION_BOOL(OPTION_MEMCHK))
369 con_out ("[%s] ", area);
375 * only required if big endian .. otherwise no need to swap
378 #if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_BIG
379 static GMQCC_INLINE void util_swap16(uint16_t *d, size_t l) {
381 d[l] = (d[l] << 8) | (d[l] >> 8);
385 static GMQCC_INLINE void util_swap32(uint32_t *d, size_t l) {
388 v = ((d[l] << 8) & 0xFF00FF00) | ((d[l] >> 8) & 0x00FF00FF);
389 d[l] = (v << 16) | (v >> 16);
393 /* Some strange system doesn't like constants that big, AND doesn't recognize an ULL suffix
394 * so let's go the safe way
396 static GMQCC_INLINE void util_swap64(uint32_t *d, size_t l) {
400 v = ((d[l] << 8) & 0xFF00FF00FF00FF00) | ((d[l] >> 8) & 0x00FF00FF00FF00FF);
401 v = ((v << 16) & 0xFFFF0000FFFF0000) | ((v >> 16) & 0x0000FFFF0000FFFF);
402 d[l] = (v << 32) | (v >> 32);
406 for (i = 0; i < l; i += 2) {
415 void util_endianswap(void *_data, size_t length, unsigned int typesize) {
416 # if PLATFORM_BYTE_ORDER == -1 /* runtime check */
417 if (*((char*)&typesize))
420 /* prevent unused warnings */
425 # if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_LITTLE
431 util_swap16((uint16_t*)_data, length>>1);
434 util_swap32((uint32_t*)_data, length>>2);
437 util_swap64((uint32_t*)_data, length>>3);
440 default: exit(EXIT_FAILURE); /* please blow the fuck up! */
447 * CRC algorithms vary in the width of the polynomial, the value of said polynomial,
448 * the initial value used for the register, weather the bits of each byte are reflected
449 * before being processed, weather the algorithm itself feeds input bytes through the
450 * register or XORs them with a byte from one end and then straight into the table, as
451 * well as (but not limited to the idea of reflected versions) where the final register
452 * value becomes reversed, and finally weather the value itself is used to XOR the final
453 * register value. AS such you can already imagine how painfully annoying CRCs are,
454 * of course we stand to target Quake, which expects it's certian set of rules for proper
455 * calculation of a CRC.
457 * In most traditional CRC algorithms on uses a reflected table driven method where a value
458 * or register is reflected if it's bits are swapped around it's center. For example:
459 * take the bits 0101 is the 4-bit reflection of 1010, and respectfully 0011 would be the
460 * reflection of 1100. Quake however expects a NON-Reflected CRC on the output, but still
461 * requires a final XOR on the values (0xFFFF and 0x0000) this is a standard CCITT CRC-16
462 * which I respectfully as a programmer don't agree with.
464 * So now you know what we target, and why we target it, despite how unsettling it may seem
465 * but those are what Quake seems to request.
468 static const uint16_t util_crc16_table[] = {
469 0x0000, 0x1021, 0x2042, 0x3063, 0x4084, 0x50A5,
470 0x60C6, 0x70E7, 0x8108, 0x9129, 0xA14A, 0xB16B,
471 0xC18C, 0xD1AD, 0xE1CE, 0xF1EF, 0x1231, 0x0210,
472 0x3273, 0x2252, 0x52B5, 0x4294, 0x72F7, 0x62D6,
473 0x9339, 0x8318, 0xB37B, 0xA35A, 0xD3BD, 0xC39C,
474 0xF3FF, 0xE3DE, 0x2462, 0x3443, 0x0420, 0x1401,
475 0x64E6, 0x74C7, 0x44A4, 0x5485, 0xA56A, 0xB54B,
476 0x8528, 0x9509, 0xE5EE, 0xF5CF, 0xC5AC, 0xD58D,
477 0x3653, 0x2672, 0x1611, 0x0630, 0x76D7, 0x66F6,
478 0x5695, 0x46B4, 0xB75B, 0xA77A, 0x9719, 0x8738,
479 0xF7DF, 0xE7FE, 0xD79D, 0xC7BC, 0x48C4, 0x58E5,
480 0x6886, 0x78A7, 0x0840, 0x1861, 0x2802, 0x3823,
481 0xC9CC, 0xD9ED, 0xE98E, 0xF9AF, 0x8948, 0x9969,
482 0xA90A, 0xB92B, 0x5AF5, 0x4AD4, 0x7AB7, 0x6A96,
483 0x1A71, 0x0A50, 0x3A33, 0x2A12, 0xDBFD, 0xCBDC,
484 0xFBBF, 0xEB9E, 0x9B79, 0x8B58, 0xBB3B, 0xAB1A,
485 0x6CA6, 0x7C87, 0x4CE4, 0x5CC5, 0x2C22, 0x3C03,
486 0x0C60, 0x1C41, 0xEDAE, 0xFD8F, 0xCDEC, 0xDDCD,
487 0xAD2A, 0xBD0B, 0x8D68, 0x9D49, 0x7E97, 0x6EB6,
488 0x5ED5, 0x4EF4, 0x3E13, 0x2E32, 0x1E51, 0x0E70,
489 0xFF9F, 0xEFBE, 0xDFDD, 0xCFFC, 0xBF1B, 0xAF3A,
490 0x9F59, 0x8F78, 0x9188, 0x81A9, 0xB1CA, 0xA1EB,
491 0xD10C, 0xC12D, 0xF14E, 0xE16F, 0x1080, 0x00A1,
492 0x30C2, 0x20E3, 0x5004, 0x4025, 0x7046, 0x6067,
493 0x83B9, 0x9398, 0xA3FB, 0xB3DA, 0xC33D, 0xD31C,
494 0xE37F, 0xF35E, 0x02B1, 0x1290, 0x22F3, 0x32D2,
495 0x4235, 0x5214, 0x6277, 0x7256, 0xB5EA, 0xA5CB,
496 0x95A8, 0x8589, 0xF56E, 0xE54F, 0xD52C, 0xC50D,
497 0x34E2, 0x24C3, 0x14A0, 0x0481, 0x7466, 0x6447,
498 0x5424, 0x4405, 0xA7DB, 0xB7FA, 0x8799, 0x97B8,
499 0xE75F, 0xF77E, 0xC71D, 0xD73C, 0x26D3, 0x36F2,
500 0x0691, 0x16B0, 0x6657, 0x7676, 0x4615, 0x5634,
501 0xD94C, 0xC96D, 0xF90E, 0xE92F, 0x99C8, 0x89E9,
502 0xB98A, 0xA9AB, 0x5844, 0x4865, 0x7806, 0x6827,
503 0x18C0, 0x08E1, 0x3882, 0x28A3, 0xCB7D, 0xDB5C,
504 0xEB3F, 0xFB1E, 0x8BF9, 0x9BD8, 0xABBB, 0xBB9A,
505 0x4A75, 0x5A54, 0x6A37, 0x7A16, 0x0AF1, 0x1AD0,
506 0x2AB3, 0x3A92, 0xFD2E, 0xED0F, 0xDD6C, 0xCD4D,
507 0xBDAA, 0xAD8B, 0x9DE8, 0x8DC9, 0x7C26, 0x6C07,
508 0x5C64, 0x4C45, 0x3CA2, 0x2C83, 0x1CE0, 0x0CC1,
509 0xEF1F, 0xFF3E, 0xCF5D, 0xDF7C, 0xAF9B, 0xBFBA,
510 0x8FD9, 0x9FF8, 0x6E17, 0x7E36, 0x4E55, 0x5E74,
511 0x2E93, 0x3EB2, 0x0ED1, 0x1EF0
514 /* Non - Reflected */
515 uint16_t util_crc16(uint16_t current, const char *k, size_t len) {
516 register uint16_t h = current;
517 for (; len; --len, ++k)
518 h = util_crc16_table[(h>>8)^((unsigned char)*k)]^(h<<8);
521 /* Reflective Varation (for reference) */
523 uint16_t util_crc16(const char *k, int len, const short clamp) {
524 register uint16_t h= (uint16_t)0xFFFFFFFF;
525 for (; len; --len, ++k)
526 h = util_crc16_table[(h^((unsigned char)*k))&0xFF]^(h>>8);
531 size_t util_strtocmd(const char *in, char *out, size_t outsz) {
533 for (; *in && sz < outsz; ++in, ++out, ++sz)
534 *out = (*in == '-') ? '_' : (isalpha(*in) && !isupper(*in)) ? *in + 'A' - 'a': *in;
539 size_t util_strtononcmd(const char *in, char *out, size_t outsz) {
541 for (; *in && sz < outsz; ++in, ++out, ++sz)
542 *out = (*in == '_') ? '-' : (isalpha(*in) && isupper(*in)) ? *in + 'a' - 'A' : *in;
547 /* TODO: rewrite ... when I redo the ve cleanup */
548 void _util_vec_grow(void **a, size_t i, size_t s) {
549 vector_t *d = vec_meta(*a);
551 size_entry_t *e = NULL;
555 m = 2 * d->allocated + i;
556 p = mem_r(d, s * m + sizeof(vector_t));
559 p = mem_a(s * m + sizeof(vector_t));
560 ((vector_t*)p)->used = 0;
565 vector_usage = util_st_new();
567 if ((e = util_st_get(vector_usage, s))) {
570 util_st_put(vector_usage, s, 1); /* start off with 1 */
574 *a = (vector_t*)p + 1;
575 vec_meta(*a)->allocated = m;
579 * Hash table for generic data, based on dynamic memory allocations
580 * all around. This is the internal interface, please look for
581 * EXPOSED INTERFACE comment below
583 typedef struct hash_node_t {
584 char *key; /* the key for this node in table */
585 void *value; /* pointer to the data as void* */
586 struct hash_node_t *next; /* next node (linked list) */
589 GMQCC_INLINE size_t util_hthash(hash_table_t *ht, const char *key) {
590 const uint32_t mix = 0x5BD1E995;
591 const uint32_t rot = 24;
592 size_t size = strlen(key);
593 uint32_t hash = 0x1EF0 /* LICRC TAB */ ^ size;
595 const unsigned char *data = (const unsigned char*)key;
598 alias = (data[0] | (data[1] << 8) | (data[2] << 16) | (data[3] << 24));
600 alias ^= alias >> rot;
611 case 3: hash ^= data[2] << 16;
612 case 2: hash ^= data[1] << 8;
613 case 1: hash ^= data[0];
621 return (size_t) (hash % ht->size);
624 static hash_node_t *_util_htnewpair(const char *key, void *value) {
626 if (!(node = (hash_node_t*)mem_a(sizeof(hash_node_t))))
629 if (!(node->key = util_strdupe(key))) {
641 * EXPOSED INTERFACE for the hashtable implementation
642 * util_htnew(size) -- to make a new hashtable
643 * util_htset(table, key, value, sizeof(value)) -- to set something in the table
644 * util_htget(table, key) -- to get something from the table
645 * util_htdel(table) -- to delete the table
647 hash_table_t *util_htnew(size_t size) {
648 hash_table_t *hashtable = NULL;
654 if (!hashtable_usage)
655 hashtable_usage = util_st_new();
657 if (!(hashtable = (hash_table_t*)mem_a(sizeof(hash_table_t))))
660 if (!(hashtable->table = (hash_node_t**)mem_a(sizeof(hash_node_t*) * size))) {
665 if ((find = util_st_get(hashtable_usage, size)))
669 util_st_put(hashtable_usage, size, 1);
672 hashtable->size = size;
673 memset(hashtable->table, 0, sizeof(hash_node_t*) * size);
679 void util_htseth(hash_table_t *ht, const char *key, size_t bin, void *value) {
680 hash_node_t *newnode = NULL;
681 hash_node_t *next = NULL;
682 hash_node_t *last = NULL;
684 next = ht->table[bin];
686 while (next && next->key && strcmp(key, next->key) > 0)
687 last = next, next = next->next;
689 /* already in table, do a replace */
690 if (next && next->key && strcmp(key, next->key) == 0) {
693 /* not found, grow a pair man :P */
694 newnode = _util_htnewpair(key, value);
695 if (next == ht->table[bin]) {
696 newnode->next = next;
697 ht->table[bin] = newnode;
699 last->next = newnode;
701 newnode->next = next;
702 last->next = newnode;
707 void util_htset(hash_table_t *ht, const char *key, void *value) {
708 util_htseth(ht, key, util_hthash(ht, key), value);
711 void *util_htgeth(hash_table_t *ht, const char *key, size_t bin) {
712 hash_node_t *pair = ht->table[bin];
714 while (pair && pair->key && strcmp(key, pair->key) > 0)
717 if (!pair || !pair->key || strcmp(key, pair->key) != 0)
723 void *util_htget(hash_table_t *ht, const char *key) {
724 return util_htgeth(ht, key, util_hthash(ht, key));
727 void *code_util_str_htgeth(hash_table_t *ht, const char *key, size_t bin) {
732 keylen = strlen(key);
734 pair = ht->table[bin];
735 while (pair && pair->key) {
736 len = strlen(pair->key);
742 cmp = strcmp(key, pair->key);
750 cmp = strcmp(key, pair->key + len - keylen);
752 uintptr_t up = (uintptr_t)pair->value;
762 * Free all allocated data in a hashtable, this is quite the amount
765 void util_htrem(hash_table_t *ht, void (*callback)(void *data)) {
767 for (; i < ht->size; i++) {
768 hash_node_t *n = ht->table[i];
788 void util_htrmh(hash_table_t *ht, const char *key, size_t bin, void (*cb)(void*)) {
789 hash_node_t **pair = &ht->table[bin];
792 while (*pair && (*pair)->key && strcmp(key, (*pair)->key) > 0)
793 pair = &(*pair)->next;
796 if (!tmp || !tmp->key || strcmp(key, tmp->key) != 0)
807 void util_htrm(hash_table_t *ht, const char *key, void (*cb)(void*)) {
808 util_htrmh(ht, key, util_hthash(ht, key), cb);
811 void util_htdel(hash_table_t *ht) {
812 util_htrem(ht, NULL);
816 * Portable implementation of vasprintf/asprintf. Assumes vsnprintf
817 * exists, otherwise compiler error.
819 * TODO: fix for MSVC ....
821 int util_vasprintf(char **dat, const char *fmt, va_list args) {
827 * For visuals tido _vsnprintf doesn't tell you the length of a
828 * formatted string if it overflows. However there is a MSVC
829 * intrinsic (which is documented wrong) called _vcsprintf which
830 * will return the required amount to allocate.
833 if ((len = _vscprintf(fmt, args)) < 0) {
838 tmp = (char*)mem_a(len + 1);
839 if ((ret = _vsnprintf_s(tmp, len+1, len+1, fmt, args)) != len) {
848 * For everything else we have a decent conformint vsnprintf that
849 * returns the number of bytes needed. We give it a try though on
850 * a short buffer, since efficently speaking, it could be nice to
851 * above a second vsnprintf call.
856 len = vsnprintf(buf, sizeof(buf), fmt, cpy);
859 if (len < (int)sizeof(buf)) {
860 *dat = util_strdup(buf);
864 /* not large enough ... */
865 tmp = (char*)mem_a(len + 1);
866 if ((ret = vsnprintf(tmp, len + 1, fmt, args)) != len) {
876 int util_asprintf(char **ret, const char *fmt, ...) {
880 read = util_vasprintf(ret, fmt, args);
887 * These are various re-implementations (wrapping the real ones) of
888 * string functions that MSVC consideres unsafe. We wrap these up and
889 * use the safe varations on MSVC.
892 static char **util_strerror_allocated() {
893 static char **data = NULL;
897 static void util_strerror_cleanup(void) {
899 char **data = util_strerror_allocated();
900 for (i = 0; i < vec_size(data); i++)
905 const char *util_strerror(int num) {
906 char *allocated = NULL;
907 static bool install = false;
908 static size_t tries = 0;
909 char **vector = util_strerror_allocated();
911 /* try installing cleanup handler */
916 install = !atexit(&util_strerror_cleanup);
920 allocated = (char*)mem_a(4096); /* A page must be enough */
921 strerror_s(allocated, 4096, num);
923 vec_push(vector, allocated);
924 return (const char *)allocated;
927 int util_snprintf(char *src, size_t bytes, const char *format, ...) {
930 va_start(va, format);
932 rt = vsprintf_s(src, bytes, format, va);
938 char *util_strcat(char *dest, const char *src) {
939 strcat_s(dest, strlen(src), src);
943 char *util_strncpy(char *dest, const char *src, size_t num) {
944 strncpy_s(dest, num, src, num);
948 const char *util_strerror(int num) {
949 return strerror(num);
952 int util_snprintf(char *src, size_t bytes, const char *format, ...) {
955 va_start(va, format);
956 rt = vsnprintf(src, bytes, format, va);
962 char *util_strcat(char *dest, const char *src) {
963 return strcat(dest, src);
966 char *util_strncpy(char *dest, const char *src, size_t num) {
967 return strncpy(dest, src, num);
970 #endif /*! _MSC_VER */
973 * Implementation of the Mersenne twister PRNG (pseudo random numer
974 * generator). Implementation of MT19937. Has a period of 2^19937-1
975 * which is a Mersenne Prime (hence the name).
977 * Implemented from specification and original paper:
978 * http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/ARTICLES/mt.pdf
980 * This code is placed in the public domain by me personally
981 * (Dale Weiler, a.k.a graphitemaster).
985 #define MT_PERIOD 397
986 #define MT_SPACE (MT_SIZE - MT_PERIOD)
988 static uint32_t mt_state[MT_SIZE];
989 static size_t mt_index = 0;
991 static GMQCC_INLINE void mt_generate() {
993 * The loop has been unrolled here: the original paper and implemenation
994 * Called for the following code:
995 * for (register unsigned i = 0; i < MT_SIZE; ++i) {
996 * register uint32_t load;
997 * load = (0x80000000 & mt_state[i]) // most significant 32nd bit
998 * load |= (0x7FFFFFFF & mt_state[(i + 1) % MT_SIZE]) // least significant 31nd bit
1000 * mt_state[i] = mt_state[(i + MT_PERIOD) % MT_SIZE] ^ (load >> 1);
1002 * if (load & 1) mt_state[i] ^= 0x9908B0DF;
1005 * This essentially is a waste: we have two modulus operations, and
1006 * a branch that is executed every iteration from [0, MT_SIZE).
1008 * Please see: http://www.quadibloc.com/crypto/co4814.htm for more
1009 * information on how this clever trick works.
1011 static const uint32_t matrix[2] = {
1016 * This register gives up a little more speed by instructing the compiler
1017 * to force these into CPU registers (they're counters for indexing mt_state
1018 * which we can force the compiler to generate prefetch instructions for)
1020 register uint32_t y;
1021 register uint32_t i;
1024 * Said loop has been unrolled for MT_SPACE (226 iterations), opposed
1025 * to [0, MT_SIZE) (634 iterations).
1027 for (i = 0; i < MT_SPACE; ++i) {
1028 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
1029 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
1031 i ++; /* loop unroll */
1033 y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
1034 mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
1038 * collapsing the walls unrolled (evenly dividing 396 [632-227 = 396
1042 while (i < MT_SIZE - 1) {
1044 * We expand this 11 times .. manually, no macros are required
1045 * here. This all fits in the CPU cache.
1047 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1048 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1050 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1051 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1053 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1054 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1056 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1057 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1059 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1060 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1062 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1063 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1065 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1066 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1068 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1069 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1071 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1072 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1074 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1075 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1077 y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
1078 mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
1082 /* i = mt_state[623] */
1083 y = (0x80000000 & mt_state[MT_SIZE - 1]) | (0x7FFFFFFF & mt_state[MT_SIZE - 1]);
1084 mt_state[MT_SIZE - 1] = mt_state[MT_PERIOD - 1] ^ (y >> 1) ^ matrix[y & 1];
1087 void util_seed(uint32_t value) {
1089 * We seed the mt_state with a LCG (linear congruential generator)
1090 * We're operating exactly on exactly m=32, so there is no need to
1093 * The multipler of choice is 0x6C07865, also knows as the Borosh-
1094 * Niederreiter multipler used for modulus 2^32. More can be read
1095 * about this in Knuth's TAOCP Volume 2, page 106.
1097 * If you don't own TAOCP something is wrong with you :-) .. so I
1098 * also provided a link to the original paper by Borosh and
1099 * Niederreiter. It's called "Optional Multipliers for PRNG by The
1100 * Linear Congruential Method" (1983).
1101 * http://en.wikipedia.org/wiki/Linear_congruential_generator
1103 * From said page, it says the following:
1104 * "A common Mersenne twister implementation, interestingly enough
1105 * used an LCG to generate seed data."
1108 * The data we're operating on is 32-bits for the mt_state array, so
1109 * there is no masking required with 0xFFFFFFFF
1113 mt_state[0] = value;
1114 for (i = 1; i < MT_SIZE; ++i)
1115 mt_state[i] = 0x6C078965 * (mt_state[i - 1] ^ mt_state[i - 1] >> 30) + i;
1118 uint32_t util_rand() {
1119 register uint32_t y;
1122 * This is inlined with any sane compiler (I checked)
1123 * for some reason though, SubC seems to be generating invalid
1124 * code when it inlines this.
1129 y = mt_state[mt_index];
1131 /* Standard tempering */
1132 y ^= y >> 11; /* +7 */
1133 y ^= y << 7 & 0x9D2C5680; /* +4 */
1134 y ^= y << 15 & 0xEFC60000; /* -4 */
1135 y ^= y >> 18; /* -7 */
1137 if(++mt_index == MT_SIZE)