void CPUID::Go()
{
CoreAssert(this != NULL);
#ifdef TARGET_OS_WINDOWS
DWORD dThread = NULL;
SYSTEM_INFO siSystem;
int iCount = 0;
struct GoThreadProc_Params params;
params.cpuid_class = this;
GetSystemInfo(&siSystem);
iCount = siSystem.dwNumberOfProcessors;
if (iCount > MAX_PROCESSORS) {
iCount = MAX_PROCESSORS;
}
for (params.processor = 0; params.processor < iCount;
params.processor++) {
HANDLE hThread =
CreateThread(NULL, 0, ( LPTHREAD_START_ROUTINE )s_GoThreadProc,
¶ms, CREATE_SUSPENDED, &dThread);
SetThreadAffinityMask(hThread,
( DWORD )pow(( double )2,
( double )params.
processor));
Sleep(0); /* Wait for affinity switch. */
SetThreadPriority(hThread, THREAD_PRIORITY_ABOVE_NORMAL);
ResumeThread(hThread);
WaitForSingleObject(hThread, INFINITE);
}
#elif defined (TARGET_OS_LINUX)
int NUM_PROCS = sysconf(_SC_NPROCESSORS_CONF), i;
cpu_set_t mask;
cpu_set_t originalmask;
sched_getaffinity(0, sizeof(originalmask), &originalmask);
for (i = 0; i < NUM_PROCS; i++) {
CPU_ZERO(&mask);
CPU_SET(( int )pow(2, i), &mask);
sched_setaffinity(0, sizeof(mask), &mask);
GoThread(i);
}
sched_setaffinity(0, sizeof(originalmask), &originalmask);
#elif defined (TARGET_OS_MACOSX)
int NUM_PROCS = chudProcessorCount();
for (int i = 0; i < NUM_PROCS; i++) {
utilUnbindThreadFromCPU();
utilBindThreadToCPU(i);
GoThread(i);
}
utilUnbindThreadFromCPU();
#endif
}
int thread_bind_native(__unused_variable struct cpuid_state_t *state, __unused_variable uint32_t id)
{
#ifdef TARGET_OS_WINDOWS
BOOL ret;
HANDLE hThread = GetCurrentThread();
#if _WIN32_WINNT >= 0x0601
GROUP_AFFINITY affinity;
ZeroMemory(&affinity, sizeof(GROUP_AFFINITY));
affinity.Group = id / 64;
affinity.Mask = 1 << (id % 64);
ret = SetThreadGroupAffinity(hThread, &affinity, NULL);
#else
DWORD mask;
if (id > 32)
return 1;
mask = (1 << id);
ret = SetThreadAffinityMask(hThread, mask);
#endif
return (ret != FALSE) ? 0 : 1;
#elif defined(TARGET_OS_LINUX) || defined(TARGET_OS_FREEBSD)
int ret;
#ifdef CPU_SET_S
size_t setsize = CPU_ALLOC_SIZE(MAX_CPUS);
CPUSET_T *set = CPU_ALLOC(MAX_CPUS);
pthread_t pth;
pth = pthread_self();
CPU_ZERO_S(setsize, set);
CPU_SET_S(id, setsize, set);
ret = pthread_setaffinity_np(pth, setsize, set);
CPU_FREE(set);
#else
size_t bits_per_set = sizeof(CPUSET_T) * 8;
size_t bits_per_subset = sizeof(CPUSET_MASK_T) * 8;
size_t setsize = sizeof(CPUSET_T) * (MAX_CPUS / bits_per_set);
size_t set_id, subset_id;
unsigned long long mask;
CPUSET_T *set = malloc(setsize);
pthread_t pth;
pth = pthread_self();
for (set_id = 0; set_id < (MAX_CPUS / bits_per_set); set_id++)
CPU_ZERO(&set[set_id]);
set_id = id / bits_per_set;
id %= bits_per_set;
subset_id = id / bits_per_subset;
id %= bits_per_subset;
mask = 1ULL << (unsigned long long)id;
((unsigned long *)set[set_id].__bits)[subset_id] |= mask;
ret = pthread_setaffinity_np(pth, setsize, set);
free(set);
#endif
return (ret == 0) ? 0 : 1;
#elif defined(TARGET_OS_MACOSX)
#ifdef USE_CHUD
return (utilBindThreadToCPU(id) == 0) ? 0 : 1;
#else
return 1;
#endif
#else
#error "thread_bind_native() not defined for this platform"
#endif
}
int thread_bind_native(__unused_variable struct cpuid_state_t *state, uint32_t id)
{
#ifdef TARGET_OS_WINDOWS
BOOL ret = FALSE;
HANDLE hThread = GetCurrentThread();
#if _WIN32_WINNT < 0x0601
if (is_windows7_or_greater()) {
#endif
DWORD threadsInGroup = 0;
WORD groupId, groupCount;
GROUP_AFFINITY affinity;
ZeroMemory(&affinity, sizeof(GROUP_AFFINITY));
groupCount = GetActiveProcessorGroupCount();
for (groupId = 0; groupId < groupCount; groupId++) {
threadsInGroup = GetActiveProcessorCount(groupId);
if (id < threadsInGroup)
break;
id -= threadsInGroup;
}
if (groupId < groupCount && id < threadsInGroup) {
affinity.Group = groupId;
affinity.Mask = 1ULL << id;
ret = SetThreadGroupAffinity(hThread, &affinity, NULL);
}
#if _WIN32_WINNT < 0x0601
} else {
DWORD mask;
if (id > 32)
return 1;
mask = (1 << id);
ret = SetThreadAffinityMask(hThread, mask);
}
#endif
if (state && ret != FALSE)
state->cpu_bound_index = id;
return (ret != FALSE) ? 0 : 1;
#elif defined(TARGET_OS_LINUX) || defined(TARGET_OS_FREEBSD)
int ret;
#ifdef CPU_SET_S
size_t setsize = CPU_ALLOC_SIZE(MAX_CPUS);
CPUSET_T *set = CPU_ALLOC(MAX_CPUS);
pthread_t pth;
pth = pthread_self();
CPU_ZERO_S(setsize, set);
CPU_SET_S(id, setsize, set);
ret = pthread_setaffinity_np(pth, setsize, set);
CPU_FREE(set);
#else
size_t bits_per_set = sizeof(CPUSET_T) * 8;
size_t bits_per_subset = sizeof(CPUSET_MASK_T) * 8;
size_t setsize = sizeof(CPUSET_T) * (MAX_CPUS / bits_per_set);
size_t set_id, subset_id;
unsigned long long mask;
CPUSET_T *set = malloc(setsize);
pthread_t pth;
pth = pthread_self();
for (set_id = 0; set_id < (MAX_CPUS / bits_per_set); set_id++)
CPU_ZERO(&set[set_id]);
set_id = id / bits_per_set;
id %= bits_per_set;
subset_id = id / bits_per_subset;
id %= bits_per_subset;
mask = 1ULL << (unsigned long long)id;
((unsigned long *)set[set_id].__bits)[subset_id] |= mask;
ret = pthread_setaffinity_np(pth, setsize, set);
free(set);
#endif
if (state && ret == 0)
state->cpu_bound_index = id;
return (ret == 0) ? 0 : 1;
#elif defined(TARGET_OS_SOLARIS)
/*
* This requires permissions, so can easily fail.
*/
if (processor_bind(P_LWPID, P_MYID, id, NULL) != 0) {
fprintf(stderr, "warning: failed to bind to CPU%u: %s\n",
id, strerror(errno));
return 1;
}
if (state)
state->cpu_bound_index = id;
return 0;
#elif defined(TARGET_OS_MACOSX)
int ret = 1;
#ifdef USE_CHUD
ret = (utilBindThreadToCPU(id) == 0) ? 0 : 1;
#endif
if (state && ret == 0)
state->cpu_bound_index = id;
return ret == 0 ? 0 : 1;
#else
#error "thread_bind_native() not defined for this platform"
#endif
}