struct stem_cache *stem_cache_new(void (*stemmer) (void *opaque, char *term),
void *opaque, unsigned int entries) {
struct stem_cache *cache;
if ((cache = malloc(sizeof(*cache)))
&& (cache->arr = malloc(sizeof(*cache->arr) * entries))
&& (cache->lookup = chash_ptr_new(bit_log2(entries) + 1, 2.0,
(unsigned int (*)(const void *)) str_hash,
(int (*)(const void *, const void *)) str_cmp))) {
cache->capacity = entries;
cache->size = 0;
cache->stem = stemmer;
cache->opaque = opaque;
cache->clock_pos = 0;
cache->stemmed = cache->cached = 0;
} else {
if (cache) {
if (cache->arr) {
free(cache->arr);
}
free(cache);
cache = NULL;
}
}
return cache;
}
uint64_t random_uint64()
{
uint64_t v = 0;
const int rand_shift = bit_log2(RAND_MAX) + 1;
int shift = 0;
while (shift < 64) {
v |= (uint64_t)rand() << shift;
shift += rand_shift;
}
return v;
}
struct postings* postings_new(unsigned int tablesize,
void (*stem)(void *opaque, char *term), void *opaque, struct stop *list) {
struct postings *p = malloc(sizeof(*p));
unsigned int bits = bit_log2(tablesize);
tablesize = bit_pow2(bits); /* tablesize is now guaranteed to be a power
* of two */
if (p && (p->hash = malloc(sizeof(struct postings_node *) * tablesize))
&& (p->string_mem = poolalloc_new(!!DEAR_DEBUG, MEMPOOL_SIZE, NULL))
&& (p->node_mem = objalloc_new(sizeof(struct postings_node), 0,
!!DEAR_DEBUG, MEMPOOL_SIZE, NULL))) {
p->stop = list;
p->stem = stem;
p->stem_opaque = opaque;
p->tblbits = bits;
p->tblsize = tablesize;
assert(p->tblbits && p->tblsize);
p->size = p->dterms = p->terms = 0;
p->err = 0;
p->docno = 0;
p->docs = 0;
p->update = NULL;
p->update_required = 0;
BIT_ARRAY_NULL(p->hash, tablesize);
} else if (p) {
if (p->hash) {
free(p->hash);
if (p->string_mem) {
poolalloc_delete(p->string_mem);
if (p->node_mem) {
objalloc_delete(p->node_mem);
}
}
}
p = NULL;
}
return p;
}
sval
insert_ea_map(struct cpu_thread *thr, uval vsid, uval ea, union ptel entry)
{
int j = 0;
uval lp = LOG_PGSIZE; /* FIXME: get large page size */
uval vaddr = (vsid << LOG_SEGMENT_SIZE) | (ea & (SEGMENT_SIZE - 1));
uval log_ht_size = bit_log2(thr->cpu->os->htab.num_ptes)+ LOG_PTE_SIZE;
uval pteg = NUM_PTES_IN_PTEG * get_pteg(vaddr, vsid, lp, log_ht_size);
union pte *ht = (union pte *)GET_HTAB(thr->cpu->os);
union pte pte;
pte.words.vsidWord = 0;
pte.words.rpnWord = entry.word;
pte.bits.avpn = (vsid << 5) | VADDR_TO_API(vaddr);
pte.bits.v = 1;
union pte *target = &ht[pteg];
int empty = NUM_PTES_IN_PTEG;
uval remove = 0;
redo_search:
for (; j < NUM_PTES_IN_PTEG; ++j, ++target) {
if (target->bits.avpn == pte.bits.avpn) {
empty = j;
remove = 1;
break;
}
if (j < empty && target->bits.v == 0) {
empty = j;
}
}
assert(empty != NUM_PTES_IN_PTEG, "Full htab\n");
target = &ht[pteg + empty];
pte_lock(target);
/* Things may have changed since we got the lock */
if (target->bits.v && target->bits.avpn != pte.bits.avpn)
goto redo_search;
if (remove) {
pte_invalidate(&thr->cpu->os->htab, target);
ptesync();
do_tlbie(target, pteg + empty);
}
// hprintf("insert: ea: 0x%lx vsid: 0x%lx -> idx: 0x%lx\n",
// ea, vsid, j + pteg);
if (pte.bits.v) {
pte.bits.lock = 1; /* Make sure new entry is still locked */
pte_insert(&thr->cpu->os->htab, empty + pteg,
pte.words.vsidWord, pte.words.rpnWord, ea);
}
pte_unlock(target);
return H_Success;
}
int main(int argc, char** argv)
{
cl_platform_id platform_id;
cl_device_id device_id;
cl_context context;
cl_command_queue command_queue;
cl_mem a_mem;
cl_program program;
cl_kernel kernel;
cl_int result;
cl_uint count;
cl_int* a = malloc(ARRAY_LENGTH * sizeof(cl_int));
cl_int* a0 = malloc(ARRAY_LENGTH * sizeof(cl_int));
cl_uint i;
cl_uint stage, pass;
(void)argc;
(void)argv;
if (!a || !a0)
return 1;
srand(time(NULL));
for (i = 0; i < ARRAY_LENGTH; i++)
a[i] = a0[i] = rand();
qsort(a0, ARRAY_LENGTH, sizeof(int), compare);
result = clGetPlatformIDs(1, &platform_id, &count);
if (result || count != 1)
return 1;
result = clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_DEFAULT, 1, &device_id,
&count);
if (result || count != 1)
return 1;
context = clCreateContext(NULL, 1, &device_id, NULL, NULL, &result);
if (result)
return 1;
command_queue = clCreateCommandQueue(context, device_id, 0, &result);
if (result)
return 1;
a_mem = clCreateBuffer(context, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR,
ARRAY_LENGTH * sizeof(cl_int), a, &result);
if (result)
return 1;
program = clCreateProgramWithSource(context, 1, &program_source,
&program_size, &result);
if (result)
return 1;
result = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);
if (result)
return 1;
kernel = clCreateKernel(program, "kernel_bitonic", &result);
if (result)
return 1;
result = clSetKernelArg(kernel, 0, sizeof(cl_mem), &a_mem);
if (result)
return 1;
for (stage = 0; stage < bit_log2(ARRAY_LENGTH); stage++) {
result = clSetKernelArg(kernel, 1, sizeof(cl_uint), &stage);
if (result)
return 1;
for (pass = 0; pass <= stage; pass++) {
result = clSetKernelArg(kernel, 2, sizeof(cl_uint), &pass);
if (result)
return 1;
result = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL,
&global_item_size, &local_item_size, 0, NULL, NULL);
if (result)
return 1;
}
}
result = clFinish(command_queue);
if (result)
return 1;
result = clEnqueueReadBuffer(command_queue, a_mem, CL_TRUE, 0,
ARRAY_LENGTH * sizeof(cl_int), a, 0, NULL, NULL);
if (result)
return 1;
for (i = 0; i < ARRAY_LENGTH; i++) {
if (a[i] != a0[i])
return 1;
}
result = clReleaseKernel(kernel);
if (result)
return 1;
result = clReleaseProgram(program);
if (result)
return 1;
result = clReleaseMemObject(a_mem);
if (result)
return 1;
result = clReleaseCommandQueue(command_queue);
if (result)
return 1;
result = clReleaseContext(context);
if (result)
return 1;
free(a0);
free(a);
return 0;
}
