static struct memblock_t *mem_start = NULL;
void *util_memory_a(size_t byte, unsigned int line, const char *file) {
- struct memblock_t *info = malloc(sizeof(struct memblock_t) + byte);
+ struct memblock_t *info = (struct memblock_t*)malloc(sizeof(struct memblock_t) + byte);
void *data = (void*)(info+1);
if (!info) return NULL;
info->line = line;
if (!s)
return NULL;
- if ((len = strlen(s)) && (ptr = mem_a(len+1))) {
+ if ((len = strlen(s)) && (ptr = (char*)mem_a(len+1))) {
memcpy(ptr, s, len);
ptr[len] = '\0';
}
* data.
*/
#if PLATFORM_BYTE_ORDER == GMQCC_BYTE_ORDER_BIG
- static void util_swap16(uint16_t *d, size_t l) {
+ static GMQCC_INLINE void util_swap16(uint16_t *d, size_t l) {
while (l--) {
d[l] = (d[l] << 8) | (d[l] >> 8);
}
}
- static void util_swap32(uint32_t *d, size_t l) {
+ static GMQCC_INLINE void util_swap32(uint32_t *d, size_t l) {
while (l--) {
uint32_t v;
v = ((d[l] << 8) & 0xFF00FF00) | ((d[l] >> 8) & 0x00FF00FF);
/* Some strange system doesn't like constants that big, AND doesn't recognize an ULL suffix
* so let's go the safe way
*/
- static void util_swap64(uint32_t *d, size_t l) {
+ static GMQCC_INLINE void util_swap64(uint32_t *d, size_t l) {
/*
while (l--) {
uint64_t v;
/* TODO: rewrite ... when I redo the ve cleanup */
void _util_vec_grow(void **a, size_t i, size_t s) {
- size_t m = *a ? 2*_vec_beg(*a)+i : i+1;
- void *p = mem_r((*a ? _vec_raw(*a) : NULL), s * m + sizeof(size_t)*2);
+ vector_t *d = vec_meta(*a);
+ size_t m = *a ? 2 * d->allocated +i : i+1;
+ void *p = mem_r((*a ? d : NULL), s * m + sizeof(vector_t));
+
if (!*a)
- ((size_t*)p)[1] = 0;
- *a = (void*)((size_t*)p + 2);
- _vec_beg(*a) = m;
+ ((vector_t*)p)->used = 0;
+ *a = (vector_t*)p + 1;
+
+ vec_meta(*a)->allocated = m;
}
/*
hash_node_t *_util_htnewpair(const char *key, void *value) {
hash_node_t *node;
- if (!(node = mem_a(sizeof(hash_node_t))))
+ if (!(node = (hash_node_t*)mem_a(sizeof(hash_node_t))))
return NULL;
if (!(node->key = util_strdup(key))) {
if (size < 1)
return NULL;
- if (!(hashtable = mem_a(sizeof(hash_table_t))))
+ if (!(hashtable = (hash_table_t*)mem_a(sizeof(hash_table_t))))
return NULL;
- if (!(hashtable->table = mem_a(sizeof(hash_node_t*) * size))) {
+ if (!(hashtable->table = (hash_node_t**)mem_a(sizeof(hash_node_t*) * size))) {
mem_d(hashtable);
return NULL;
}
return util_htgeth(ht, key, util_hthash(ht, key));
}
+void *code_util_str_htgeth(hash_table_t *ht, const char *key, size_t bin) {
+ hash_node_t *pair;
+ size_t len, keylen;
+ int cmp;
+
+ keylen = strlen(key);
+
+ pair = ht->table[bin];
+ while (pair && pair->key) {
+ len = strlen(pair->key);
+ if (len < keylen) {
+ pair = pair->next;
+ continue;
+ }
+ if (keylen == len) {
+ cmp = strcmp(key, pair->key);
+ if (cmp == 0)
+ return pair->value;
+ if (cmp < 0)
+ return NULL;
+ pair = pair->next;
+ continue;
+ }
+ cmp = strcmp(key, pair->key + len - keylen);
+ if (cmp == 0) {
+ uintptr_t up = (uintptr_t)pair->value;
+ up += len - keylen;
+ return (void*)up;
+ }
+ pair = pair->next;
+ }
+ return NULL;
+}
+
/*
* Free all allocated data in a hashtable, this is quite the amount
* of work.
mem_d(ht->table);
mem_d(ht);
}
+
+/*
+ * Portable implementation of vasprintf/asprintf. Assumes vsnprintf
+ * exists, otherwise compiler error.
+ */
+int util_vasprintf(char **ret, const char *fmt, va_list args) {
+ int read;
+ va_list copy;
+ va_copy(copy, args);
+
+ *ret = 0;
+ if ((read = vsnprintf(NULL, 0, fmt, args)) >= 0) {
+ char *buffer;
+ if ((buffer = (char*)mem_a(read + 1))) {
+ if ((read = vsnprintf(buffer, read + 1, fmt, copy)) < 0)
+ mem_d(buffer);
+ else
+ *ret = buffer;
+ }
+ }
+ va_end(copy);
+ return read;
+}
+int util_asprintf(char **ret, const char *fmt, ...) {
+ va_list args;
+ int read;
+ va_start(args, fmt);
+ read = util_vasprintf(ret, fmt, args);
+ va_end (args);
+
+ return read;
+}
+
+/*
+ * Implementation of the Mersenne twister PRNG (pseudo random numer
+ * generator). Implementation of MT19937. Has a period of 2^19937-1
+ * which is a Mersenne Prime (hence the name).
+ *
+ * Implemented from specification and original paper:
+ * http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/ARTICLES/mt.pdf
+ *
+ * This code is placed in the public domain by me personally
+ * (Dale Weiler, a.k.a graphitemaster).
+ */
+
+#define MT_SIZE 624
+#define MT_PERIOD 397
+#define MT_SPACE (MT_SIZE - MT_PERIOD)
+
+static uint32_t mt_state[MT_SIZE];
+static size_t mt_index = 0;
+
+static GMQCC_INLINE void mt_generate() {
+ /*
+ * The loop has been unrolled here: the original paper and implemenation
+ * Called for the following code:
+ * for (register unsigned i = 0; i < MT_SIZE; ++i) {
+ * register uint32_t load;
+ * load = (0x80000000 & mt_state[i]) // most significant 32nd bit
+ * load |= (0x7FFFFFFF & mt_state[(i + 1) % MT_SIZE]) // least significant 31nd bit
+ *
+ * mt_state[i] = mt_state[(i + MT_PERIOD) % MT_SIZE] ^ (load >> 1);
+ *
+ * if (load & 1) mt_state[i] ^= 0x9908B0DF;
+ * }
+ *
+ * This essentially is a waste: we have two modulus operations, and
+ * a branch that is executed every iteration from [0, MT_SIZE).
+ *
+ * Please see: http://www.quadibloc.com/crypto/co4814.htm for more
+ * information on how this clever trick works.
+ */
+ static const uint32_t matrix[2] = {
+ 0x00000000,
+ 0x9908B0Df
+ };
+ /*
+ * This register gives up a little more speed by instructing the compiler
+ * to force these into CPU registers (they're counters for indexing mt_state
+ * which we can force the compiler to generate prefetch instructions for)
+ */
+ register uint32_t y;
+ register uint32_t i;
+
+ /*
+ * Said loop has been unrolled for MT_SPACE (226 iterations), opposed
+ * to [0, MT_SIZE) (634 iterations).
+ */
+ for (i = 0; i < MT_SPACE; ++i) {
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
+
+ i ++; /* loop unroll */
+
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i + MT_PERIOD] ^ (y >> 1) ^ matrix[y & 1];
+ }
+
+ /*
+ * collapsing the walls unrolled (evenly dividing 396 [632-227 = 396
+ * = 2*2*3*3*11])
+ */
+ i = MT_SPACE;
+ while (i < MT_SIZE - 1) {
+ /*
+ * We expand this 11 times .. manually, no macros are required
+ * here. This all fits in the CPU cache.
+ */
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ y = (0x80000000 & mt_state[i]) | (0x7FFFFFFF & mt_state[i + 1]);
+ mt_state[i] = mt_state[i - MT_SPACE] ^ (y >> 1) ^ matrix[y & 1];
+ ++i;
+ }
+
+ /* i = mt_state[623] */
+ y = (0x80000000 & mt_state[MT_SIZE - 1]) | (0x7FFFFFFF & mt_state[MT_SIZE - 1]);
+ mt_state[MT_SIZE - 1] = mt_state[MT_PERIOD - 1] ^ (y >> 1) ^ matrix[y & 1];
+}
+
+void util_seed(uint32_t value) {
+ /*
+ * We seed the mt_state with a LCG (linear congruential generator)
+ * We're operating exactly on exactly m=32, so there is no need to
+ * use modulus.
+ *
+ * The multipler of choice is 0x6C07865, also knows as the Borosh-
+ * Niederreiter multipler used for modulus 2^32. More can be read
+ * about this in Knuth's TAOCP Volume 2, page 106.
+ *
+ * If you don't own TAOCP something is wrong with you :-) .. so I
+ * also provided a link to the original paper by Borosh and
+ * Niederreiter. It's called "Optional Multipliers for PRNG by The
+ * Linear Congruential Method" (1983).
+ * http://en.wikipedia.org/wiki/Linear_congruential_generator
+ *
+ * From said page, it says the following:
+ * "A common Mersenne twister implementation, interestingly enough
+ * used an LCG to generate seed data."
+ *
+ * Remarks:
+ * The data we're operating on is 32-bits for the mt_state array, so
+ * there is no masking required with 0xFFFFFFFF
+ */
+ register size_t i;
+
+ mt_state[0] = value;
+ for (i = 1; i < MT_SIZE; ++i)
+ mt_state[i] = 0x6C078965 * (mt_state[i - 1] ^ mt_state[i - 1] >> 30) + i;
+}
+
+uint32_t util_rand() {
+ register uint32_t y;
+
+ /*
+ * This is inlined with any sane compiler (I checked)
+ * for some reason though, SubC seems to be generating invalid
+ * code when it inlines this.
+ */
+ if (!mt_index)
+ mt_generate();
+
+ y = mt_state[mt_index];
+
+ /* Standard tempering */
+ y ^= y >> 11; /* +7 */
+ y ^= y << 7 & 0x9D2C5680; /* +4 */
+ y ^= y << 15 & 0xEFC60000; /* -4 */
+ y ^= y >> 18; /* -7 */
+
+ if(++mt_index == MT_SIZE)
+ mt_index = 0;
+
+ return y;
+}