int openblas_get_num_threads(void) {
#ifndef SMP
return 1;
#else
// init blas_cpu_number if needed
blas_get_cpu_number();
return blas_cpu_number;
#endif
}
void CONSTRUCTOR gotoblas_init(void) {
if (gotoblas_initialized) return;
#ifdef SMP
openblas_fork_handler();
#endif
#ifdef PROFILE
moncontrol (0);
#endif
#ifdef DYNAMIC_ARCH
gotoblas_dynamic_init();
#endif
#if defined(SMP) && defined(OS_LINUX) && !defined(NO_AFFINITY)
gotoblas_affinity_init();
#endif
#if defined(OS_LINUX) && !defined(NO_WARMUP)
gotoblas_memory_init();
#endif
#if defined(OS_LINUX)
struct rlimit curlimit;
if ( getrlimit(RLIMIT_STACK, &curlimit ) == 0 )
{
if ( curlimit.rlim_cur != curlimit.rlim_max )
{
curlimit.rlim_cur = curlimit.rlim_max;
setrlimit(RLIMIT_STACK, &curlimit);
}
}
#endif
#ifdef SMP
if (blas_cpu_number == 0) blas_get_cpu_number();
#ifdef SMP_SERVER
if (blas_server_avail == 0) blas_thread_init();
#endif
#endif
#ifdef FUNCTION_PROFILE
gotoblas_profile_init();
#endif
gotoblas_initialized = 1;
#ifdef PROFILE
moncontrol (1);
#endif
}
void CONSTRUCTOR gotoblas_init(void) {
if (gotoblas_initialized) return;
#ifdef PROFILE
moncontrol (0);
#endif
#ifdef DYNAMIC_ARCH
gotoblas_dynamic_init();
#endif
#if defined(SMP) && defined(OS_LINUX) && !defined(NO_AFFINITY)
gotoblas_affinity_init();
#endif
#if defined(OS_LINUX) && !defined(NO_WARMUP)
gotoblas_memory_init();
#endif
#ifdef SMP
if (blas_cpu_number == 0) blas_get_cpu_number();
#ifdef SMP_SERVER
if (blas_server_avail == 0) blas_thread_init();
#endif
#endif
#ifdef FUNCTION_PROFILE
gotoblas_profile_init();
#endif
gotoblas_initialized = 1;
#ifdef PROFILE
moncontrol (1);
#endif
}
static void gotoblas_memory_init(void) {
void *buffer;
hot_alloc = 1;
buffer = (void *)blas_memory_alloc(0);
#ifdef SMP
if (blas_cpu_number == 0) blas_get_cpu_number();
#ifdef SMP_SERVER
if (blas_server_avail == 0) blas_thread_init();
#endif
_init_thread_memory((void *)((BLASULONG)buffer + GEMM_OFFSET_A));
#else
_touch_memory(NULL, NULL, NULL, (void *)((BLASULONG)buffer + GEMM_OFFSET_A), NULL, 0);
#endif
blas_memory_free(buffer);
}
void *blas_memory_alloc(int procpos){
int position;
#if defined(WHEREAMI) && !defined(USE_OPENMP)
int mypos;
#endif
void *map_address;
void *(*memoryalloc[])(void *address) = {
#ifdef ALLOC_DEVICEDRIVER
alloc_devicedirver,
#endif
#if defined OS_LINUX || defined OS_AIX || defined __sun__ || defined OS_WINDOWS
alloc_hugetlb,
#endif
#ifdef ALLOC_SHM
alloc_shm,
#endif
#ifdef ALLOC_MMAP
alloc_mmap,
#endif
#ifdef ALLOC_QALLOC
alloc_qalloc,
#endif
#ifdef ALLOC_WINDOWS
alloc_windows,
#endif
#ifdef ALLOC_MALLOC
alloc_malloc,
#endif
NULL,
};
void *(**func)(void *address);
if (!memory_initialized) {
LOCK_COMMAND(&alloc_lock);
if (!memory_initialized) {
#if defined(WHEREAMI) && !defined(USE_OPENMP)
for (position = 0; position < NUM_BUFFERS; position ++){
memory[position].addr = (void *)0;
memory[position].pos = -1;
memory[position].used = 0;
memory[position].lock = 0;
}
#endif
#ifdef DYNAMIC_ARCH
gotoblas_dynamic_init();
#endif
#if defined(SMP) && defined(OS_LINUX) && !defined(NO_AFFINITY)
gotoblas_affinity_init();
#endif
#ifdef SMP
if (!blas_num_threads) blas_cpu_number = blas_get_cpu_number();
#endif
#if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64)
#ifndef DYNAMIC_ARCH
blas_set_parameter();
#endif
#endif
memory_initialized = 1;
}
UNLOCK_COMMAND(&alloc_lock);
}
#ifdef DEBUG
printf("Alloc Start ...\n");
#endif
#if defined(WHEREAMI) && !defined(USE_OPENMP)
mypos = WhereAmI();
position = mypos;
while (position > NUM_BUFFERS) position >>= 1;
do {
if (!memory[position].used && (memory[position].pos == mypos)) {
blas_lock(&memory[position].lock);
if (!memory[position].used) goto allocation;
blas_unlock(&memory[position].lock);
}
position ++;
} while (position < NUM_BUFFERS);
#endif
position = 0;
do {
if (!memory[position].used) {
blas_lock(&memory[position].lock);
if (!memory[position].used) goto allocation;
blas_unlock(&memory[position].lock);
}
position ++;
} while (position < NUM_BUFFERS);
goto error;
allocation :
#ifdef DEBUG
printf(" Position -> %d\n", position);
#endif
memory[position].used = 1;
blas_unlock(&memory[position].lock);
if (!memory[position].addr) {
do {
#ifdef DEBUG
printf("Allocation Start : %lx\n", base_address);
#endif
map_address = (void *)-1;
func = &memoryalloc[0];
while ((func != NULL) && (map_address == (void *) -1)) {
map_address = (*func)((void *)base_address);
#ifdef ALLOC_DEVICEDRIVER
if ((*func == alloc_devicedirver) && (map_address == (void *)-1)) {
fprintf(stderr, "OpenBLAS Warning ... Physically contigous allocation was failed.\n");
}
#endif
#ifdef ALLOC_HUGETLBFILE
if ((*func == alloc_hugetlbfile) && (map_address == (void *)-1)) {
#ifndef OS_WINDOWS
fprintf(stderr, "OpenBLAS Warning ... HugeTLB(File) allocation was failed.\n");
#endif
}
#endif
#if defined OS_LINUX || defined OS_AIX || defined __sun__ || defined OS_WINDOWS
if ((*func == alloc_hugetlb) && (map_address != (void *)-1)) hugetlb_allocated = 1;
#endif
func ++;
}
#ifdef DEBUG
printf(" Success -> %08lx\n", map_address);
#endif
if (((BLASLONG) map_address) == -1) base_address = 0UL;
if (base_address) base_address += BUFFER_SIZE + FIXED_PAGESIZE;
} while ((BLASLONG)map_address == -1);
memory[position].addr = map_address;
#ifdef DEBUG
printf(" Mapping Succeeded. %p(%d)\n", (void *)memory[position].addr, position);
#endif
}
#if defined(WHEREAMI) && !defined(USE_OPENMP)
if (memory[position].pos == -1) memory[position].pos = mypos;
#endif
#ifdef DYNAMIC_ARCH
if (memory_initialized == 1) {
LOCK_COMMAND(&alloc_lock);
if (memory_initialized == 1) {
if (!gotoblas) gotoblas_dynamic_init();
memory_initialized = 2;
}
UNLOCK_COMMAND(&alloc_lock);
}
#endif
#ifdef DEBUG
printf("Mapped : %p %3d\n\n",
(void *)memory[position].addr, position);
#endif
return (void *)memory[position].addr;
error:
printf("BLAS : Program is Terminated. Because you tried to allocate too many memory regions.\n");
return NULL;
}