static int
hwloc_look_windows(struct hwloc_backend *backend)
{
struct hwloc_topology *topology = backend->topology;
hwloc_bitmap_t groups_pu_set = NULL;
SYSTEM_INFO SystemInfo;
DWORD length;
if (topology->levels[0][0]->cpuset)
/* somebody discovered things */
return -1;
hwloc_alloc_obj_cpusets(topology->levels[0][0]);
GetSystemInfo(&SystemInfo);
if (!GetLogicalProcessorInformationExProc && GetLogicalProcessorInformationProc) {
PSYSTEM_LOGICAL_PROCESSOR_INFORMATION procInfo, tmpprocInfo;
unsigned id;
unsigned i;
struct hwloc_obj *obj;
hwloc_obj_type_t type;
length = 0;
procInfo = NULL;
while (1) {
if (GetLogicalProcessorInformationProc(procInfo, &length))
break;
if (GetLastError() != ERROR_INSUFFICIENT_BUFFER)
return -1;
tmpprocInfo = realloc(procInfo, length);
if (!tmpprocInfo) {
free(procInfo);
goto out;
}
procInfo = tmpprocInfo;
}
assert(!length || procInfo);
for (i = 0; i < length / sizeof(*procInfo); i++) {
/* Ignore unknown caches */
if (procInfo->Relationship == RelationCache
&& procInfo->Cache.Type != CacheUnified
&& procInfo->Cache.Type != CacheData
&& procInfo->Cache.Type != CacheInstruction)
continue;
id = -1;
switch (procInfo[i].Relationship) {
case RelationNumaNode:
type = HWLOC_OBJ_NUMANODE;
id = procInfo[i].NumaNode.NodeNumber;
break;
case RelationProcessorPackage:
type = HWLOC_OBJ_PACKAGE;
break;
case RelationCache:
type = (procInfo[i].Cache.Type == CacheInstruction ? HWLOC_OBJ_L1ICACHE : HWLOC_OBJ_L1CACHE) + procInfo[i].Cache.Level - 1;
break;
case RelationProcessorCore:
type = HWLOC_OBJ_CORE;
break;
case RelationGroup:
default:
type = HWLOC_OBJ_GROUP;
break;
}
if (!hwloc_filter_check_keep_object_type(topology, type))
continue;
obj = hwloc_alloc_setup_object(topology, type, id);
obj->cpuset = hwloc_bitmap_alloc();
hwloc_debug("%s#%u mask %lx\n", hwloc_type_name(type), id, procInfo[i].ProcessorMask);
/* ProcessorMask is a ULONG_PTR */
hwloc_bitmap_set_ith_ULONG_PTR(obj->cpuset, 0, procInfo[i].ProcessorMask);
hwloc_debug_2args_bitmap("%s#%u bitmap %s\n", hwloc_type_name(type), id, obj->cpuset);
switch (type) {
case HWLOC_OBJ_NUMANODE:
{
ULONGLONG avail;
obj->nodeset = hwloc_bitmap_alloc();
hwloc_bitmap_set(obj->nodeset, id);
if ((GetNumaAvailableMemoryNodeExProc && GetNumaAvailableMemoryNodeExProc(id, &avail))
|| (GetNumaAvailableMemoryNodeProc && GetNumaAvailableMemoryNodeProc(id, &avail)))
obj->memory.local_memory = avail;
obj->memory.page_types_len = 2;
obj->memory.page_types = malloc(2 * sizeof(*obj->memory.page_types));
memset(obj->memory.page_types, 0, 2 * sizeof(*obj->memory.page_types));
obj->memory.page_types_len = 1;
obj->memory.page_types[0].size = SystemInfo.dwPageSize;
#if HAVE_DECL__SC_LARGE_PAGESIZE
obj->memory.page_types_len++;
obj->memory.page_types[1].size = sysconf(_SC_LARGE_PAGESIZE);
#endif
break;
}
case HWLOC_OBJ_L1CACHE:
case HWLOC_OBJ_L2CACHE:
case HWLOC_OBJ_L3CACHE:
case HWLOC_OBJ_L4CACHE:
case HWLOC_OBJ_L5CACHE:
case HWLOC_OBJ_L1ICACHE:
case HWLOC_OBJ_L2ICACHE:
case HWLOC_OBJ_L3ICACHE:
obj->attr->cache.size = procInfo[i].Cache.Size;
obj->attr->cache.associativity = procInfo[i].Cache.Associativity == CACHE_FULLY_ASSOCIATIVE ? -1 : procInfo[i].Cache.Associativity ;
obj->attr->cache.linesize = procInfo[i].Cache.LineSize;
obj->attr->cache.depth = procInfo[i].Cache.Level;
switch (procInfo->Cache.Type) {
case CacheUnified:
obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
break;
case CacheData:
obj->attr->cache.type = HWLOC_OBJ_CACHE_DATA;
break;
case CacheInstruction:
obj->attr->cache.type = HWLOC_OBJ_CACHE_INSTRUCTION;
break;
default:
hwloc_free_unlinked_object(obj);
continue;
}
break;
case HWLOC_OBJ_GROUP:
obj->attr->group.kind = procInfo[i].Relationship == RelationGroup ? HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP : HWLOC_GROUP_KIND_WINDOWS_RELATIONSHIP_UNKNOWN;
break;
default:
break;
}
hwloc_insert_object_by_cpuset(topology, obj);
}
free(procInfo);
}
if (GetLogicalProcessorInformationExProc) {
PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX procInfoTotal, tmpprocInfoTotal, procInfo;
unsigned id;
struct hwloc_obj *obj;
hwloc_obj_type_t type;
length = 0;
procInfoTotal = NULL;
while (1) {
if (GetLogicalProcessorInformationExProc(RelationAll, procInfoTotal, &length))
break;
if (GetLastError() != ERROR_INSUFFICIENT_BUFFER)
return -1;
tmpprocInfoTotal = realloc(procInfoTotal, length);
if (!tmpprocInfoTotal) {
free(procInfoTotal);
goto out;
}
procInfoTotal = tmpprocInfoTotal;
}
for (procInfo = procInfoTotal;
(void*) procInfo < (void*) ((uintptr_t) procInfoTotal + length);
procInfo = (void*) ((uintptr_t) procInfo + procInfo->Size)) {
unsigned num, i;
GROUP_AFFINITY *GroupMask;
/* Ignore unknown caches */
if (procInfo->Relationship == RelationCache
&& procInfo->Cache.Type != CacheUnified
&& procInfo->Cache.Type != CacheData
&& procInfo->Cache.Type != CacheInstruction)
continue;
id = -1;
switch (procInfo->Relationship) {
case RelationNumaNode:
type = HWLOC_OBJ_NUMANODE;
num = 1;
GroupMask = &procInfo->NumaNode.GroupMask;
id = procInfo->NumaNode.NodeNumber;
break;
case RelationProcessorPackage:
type = HWLOC_OBJ_PACKAGE;
num = procInfo->Processor.GroupCount;
GroupMask = procInfo->Processor.GroupMask;
break;
case RelationCache:
type = (procInfo->Cache.Type == CacheInstruction ? HWLOC_OBJ_L1ICACHE : HWLOC_OBJ_L1CACHE) + procInfo->Cache.Level - 1;
num = 1;
GroupMask = &procInfo->Cache.GroupMask;
break;
case RelationProcessorCore:
type = HWLOC_OBJ_CORE;
num = procInfo->Processor.GroupCount;
GroupMask = procInfo->Processor.GroupMask;
break;
case RelationGroup:
/* So strange an interface... */
for (id = 0; id < procInfo->Group.ActiveGroupCount; id++) {
KAFFINITY mask;
hwloc_bitmap_t set;
set = hwloc_bitmap_alloc();
mask = procInfo->Group.GroupInfo[id].ActiveProcessorMask;
hwloc_debug("group %u %d cpus mask %lx\n", id,
procInfo->Group.GroupInfo[id].ActiveProcessorCount, mask);
/* KAFFINITY is ULONG_PTR */
hwloc_bitmap_set_ith_ULONG_PTR(set, id, mask);
/* FIXME: what if running 32bits on a 64bits windows with 64-processor groups?
* ULONG_PTR is 32bits, so half the group is invisible?
* maybe scale id to id*8/sizeof(ULONG_PTR) so that groups are 64-PU aligned?
*/
hwloc_debug_2args_bitmap("group %u %d bitmap %s\n", id, procInfo->Group.GroupInfo[id].ActiveProcessorCount, set);
/* save the set of PUs so that we can create them at the end */
if (!groups_pu_set)
groups_pu_set = hwloc_bitmap_alloc();
hwloc_bitmap_or(groups_pu_set, groups_pu_set, set);
if (hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_GROUP)) {
obj = hwloc_alloc_setup_object(topology, HWLOC_OBJ_GROUP, id);
obj->cpuset = set;
obj->attr->group.kind = HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP;
hwloc_insert_object_by_cpuset(topology, obj);
} else
hwloc_bitmap_free(set);
}
continue;
default:
/* Don't know how to get the mask. */
hwloc_debug("unknown relation %d\n", procInfo->Relationship);
continue;
}
if (!hwloc_filter_check_keep_object_type(topology, type))
continue;
obj = hwloc_alloc_setup_object(topology, type, id);
obj->cpuset = hwloc_bitmap_alloc();
for (i = 0; i < num; i++) {
hwloc_debug("%s#%u %d: mask %d:%lx\n", hwloc_type_name(type), id, i, GroupMask[i].Group, GroupMask[i].Mask);
/* GROUP_AFFINITY.Mask is KAFFINITY, which is ULONG_PTR */
hwloc_bitmap_set_ith_ULONG_PTR(obj->cpuset, GroupMask[i].Group, GroupMask[i].Mask);
/* FIXME: scale id to id*8/sizeof(ULONG_PTR) as above? */
}
hwloc_debug_2args_bitmap("%s#%u bitmap %s\n", hwloc_type_name(type), id, obj->cpuset);
switch (type) {
case HWLOC_OBJ_NUMANODE:
{
ULONGLONG avail;
obj->nodeset = hwloc_bitmap_alloc();
hwloc_bitmap_set(obj->nodeset, id);
if ((GetNumaAvailableMemoryNodeExProc && GetNumaAvailableMemoryNodeExProc(id, &avail))
|| (GetNumaAvailableMemoryNodeProc && GetNumaAvailableMemoryNodeProc(id, &avail)))
obj->memory.local_memory = avail;
obj->memory.page_types = malloc(2 * sizeof(*obj->memory.page_types));
memset(obj->memory.page_types, 0, 2 * sizeof(*obj->memory.page_types));
obj->memory.page_types_len = 1;
obj->memory.page_types[0].size = SystemInfo.dwPageSize;
#if HAVE_DECL__SC_LARGE_PAGESIZE
obj->memory.page_types_len++;
obj->memory.page_types[1].size = sysconf(_SC_LARGE_PAGESIZE);
#endif
break;
}
case HWLOC_OBJ_L1CACHE:
case HWLOC_OBJ_L2CACHE:
case HWLOC_OBJ_L3CACHE:
case HWLOC_OBJ_L4CACHE:
case HWLOC_OBJ_L5CACHE:
case HWLOC_OBJ_L1ICACHE:
case HWLOC_OBJ_L2ICACHE:
case HWLOC_OBJ_L3ICACHE:
obj->attr->cache.size = procInfo->Cache.CacheSize;
obj->attr->cache.associativity = procInfo->Cache.Associativity == CACHE_FULLY_ASSOCIATIVE ? -1 : procInfo->Cache.Associativity ;
obj->attr->cache.linesize = procInfo->Cache.LineSize;
obj->attr->cache.depth = procInfo->Cache.Level;
switch (procInfo->Cache.Type) {
case CacheUnified:
obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
break;
case CacheData:
obj->attr->cache.type = HWLOC_OBJ_CACHE_DATA;
break;
case CacheInstruction:
obj->attr->cache.type = HWLOC_OBJ_CACHE_INSTRUCTION;
break;
default:
hwloc_free_unlinked_object(obj);
continue;
}
break;
default:
break;
}
hwloc_insert_object_by_cpuset(topology, obj);
}
free(procInfoTotal);
}
if (groups_pu_set) {
/* the system supports multiple Groups.
* PU indexes may be discontiguous, especially if Groups contain less than 64 procs.
*/
hwloc_obj_t obj;
unsigned idx;
hwloc_bitmap_foreach_begin(idx, groups_pu_set) {
obj = hwloc_alloc_setup_object(topology, HWLOC_OBJ_PU, idx);
obj->cpuset = hwloc_bitmap_alloc();
hwloc_bitmap_only(obj->cpuset, idx);
hwloc_debug_1arg_bitmap("cpu %u has cpuset %s\n",
idx, obj->cpuset);
hwloc_insert_object_by_cpuset(topology, obj);
} hwloc_bitmap_foreach_end();
hwloc_bitmap_free(groups_pu_set);
} else {
static int
hwloc_look_darwin(struct hwloc_backend *backend)
{
struct hwloc_topology *topology = backend->topology;
int64_t _nprocs;
unsigned nprocs;
int64_t _npackages;
unsigned i, j, cpu;
struct hwloc_obj *obj;
size_t size;
int64_t l1dcachesize, l1icachesize;
int64_t cacheways[2];
int64_t l2cachesize;
int64_t cachelinesize;
int64_t memsize;
char cpumodel[64];
if (topology->levels[0][0]->cpuset)
/* somebody discovered things */
return -1;
hwloc_alloc_obj_cpusets(topology->levels[0][0]);
if (hwloc_get_sysctlbyname("hw.ncpu", &_nprocs) || _nprocs <= 0)
return -1;
nprocs = _nprocs;
topology->support.discovery->pu = 1;
hwloc_debug("%u procs\n", nprocs);
size = sizeof(cpumodel);
if (sysctlbyname("machdep.cpu.brand_string", cpumodel, &size, NULL, 0))
cpumodel[0] = '\0';
if (!hwloc_get_sysctlbyname("hw.packages", &_npackages) && _npackages > 0) {
unsigned npackages = _npackages;
int64_t _cores_per_package;
int64_t _logical_per_package;
unsigned logical_per_package;
hwloc_debug("%u packages\n", npackages);
if (!hwloc_get_sysctlbyname("machdep.cpu.logical_per_package", &_logical_per_package) && _logical_per_package > 0)
logical_per_package = _logical_per_package;
else
/* Assume the trivia. */
logical_per_package = nprocs / npackages;
hwloc_debug("%u threads per package\n", logical_per_package);
if (nprocs == npackages * logical_per_package
&& hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_PACKAGE))
for (i = 0; i < npackages; i++) {
obj = hwloc_alloc_setup_object(HWLOC_OBJ_PACKAGE, i);
obj->cpuset = hwloc_bitmap_alloc();
for (cpu = i*logical_per_package; cpu < (i+1)*logical_per_package; cpu++)
hwloc_bitmap_set(obj->cpuset, cpu);
hwloc_debug_1arg_bitmap("package %u has cpuset %s\n",
i, obj->cpuset);
if (cpumodel[0] != '\0')
hwloc_obj_add_info(obj, "CPUModel", cpumodel);
hwloc_insert_object_by_cpuset(topology, obj);
}
else
if (cpumodel[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUModel", cpumodel);
if (!hwloc_get_sysctlbyname("machdep.cpu.cores_per_package", &_cores_per_package) && _cores_per_package > 0
&& hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_CORE)) {
unsigned cores_per_package = _cores_per_package;
hwloc_debug("%u cores per package\n", cores_per_package);
if (!(logical_per_package % cores_per_package))
for (i = 0; i < npackages * cores_per_package; i++) {
obj = hwloc_alloc_setup_object(HWLOC_OBJ_CORE, i);
obj->cpuset = hwloc_bitmap_alloc();
for (cpu = i*(logical_per_package/cores_per_package);
cpu < (i+1)*(logical_per_package/cores_per_package);
cpu++)
hwloc_bitmap_set(obj->cpuset, cpu);
hwloc_debug_1arg_bitmap("core %u has cpuset %s\n",
i, obj->cpuset);
hwloc_insert_object_by_cpuset(topology, obj);
}
}
} else
if (cpumodel[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUModel", cpumodel);
if (hwloc_get_sysctlbyname("hw.l1dcachesize", &l1dcachesize))
l1dcachesize = 0;
if (hwloc_get_sysctlbyname("hw.l1icachesize", &l1icachesize))
l1icachesize = 0;
if (hwloc_get_sysctlbyname("hw.l2cachesize", &l2cachesize))
l2cachesize = 0;
if (hwloc_get_sysctlbyname("machdep.cpu.cache.L1_associativity", &cacheways[0]))
cacheways[0] = 0;
else if (cacheways[0] == 0xff)
cacheways[0] = -1;
if (hwloc_get_sysctlbyname("machdep.cpu.cache.L2_associativity", &cacheways[1]))
cacheways[1] = 0;
else if (cacheways[1] == 0xff)
cacheways[1] = -1;
if (hwloc_get_sysctlbyname("hw.cachelinesize", &cachelinesize))
cachelinesize = 0;
if (hwloc_get_sysctlbyname("hw.memsize", &memsize))
memsize = 0;
if (!sysctlbyname("hw.cacheconfig", NULL, &size, NULL, 0)) {
unsigned n = size / sizeof(uint32_t);
uint64_t *cacheconfig = NULL;
uint64_t *cachesize = NULL;
uint32_t *cacheconfig32 = NULL;
cacheconfig = malloc(sizeof(uint64_t) * n);
if (NULL == cacheconfig) {
goto out;
}
cachesize = malloc(sizeof(uint64_t) * n);
if (NULL == cachesize) {
goto out;
}
cacheconfig32 = malloc(sizeof(uint32_t) * n);
if (NULL == cacheconfig32) {
goto out;
}
if ((!sysctlbyname("hw.cacheconfig", cacheconfig, &size, NULL, 0))) {
/* Yeech. Darwin seemingly has changed from 32bit to 64bit integers for
* cacheconfig, with apparently no way for detection. Assume the machine
* won't have more than 4 billion cpus */
if (cacheconfig[0] > 0xFFFFFFFFUL) {
memcpy(cacheconfig32, cacheconfig, size);
for (i = 0 ; i < size / sizeof(uint32_t); i++)
cacheconfig[i] = cacheconfig32[i];
}
memset(cachesize, 0, sizeof(uint64_t) * n);
size = sizeof(uint64_t) * n;
if (sysctlbyname("hw.cachesize", cachesize, &size, NULL, 0)) {
if (n > 0)
cachesize[0] = memsize;
if (n > 1)
cachesize[1] = l1dcachesize;
if (n > 2)
cachesize[2] = l2cachesize;
}
hwloc_debug("%s", "caches");
for (i = 0; i < n && cacheconfig[i]; i++)
hwloc_debug(" %"PRIu64"(%"PRIu64"kB)", cacheconfig[i], cachesize[i] / 1024);
/* Now we know how many caches there are */
n = i;
hwloc_debug("\n%u cache levels\n", n - 1);
/* For each cache level (0 is memory) */
for (i = 0; i < n; i++) {
/* cacheconfig tells us how many cpus share it, let's iterate on each cache */
for (j = 0; j < (nprocs / cacheconfig[i]); j++) {
if (!i) {
obj = hwloc_alloc_setup_object(HWLOC_OBJ_NUMANODE, j);
obj->nodeset = hwloc_bitmap_alloc();
hwloc_bitmap_set(obj->nodeset, j);
} else {
obj = hwloc_alloc_setup_object(HWLOC_OBJ_L1CACHE+i-1, -1);
}
obj->cpuset = hwloc_bitmap_alloc();
for (cpu = j*cacheconfig[i];
cpu < ((j+1)*cacheconfig[i]);
cpu++)
hwloc_bitmap_set(obj->cpuset, cpu);
if (i == 1 && l1icachesize
&& hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_L1ICACHE)) {
/* FIXME assuming that L1i and L1d are shared the same way. Darwin
* does not yet provide a way to know. */
hwloc_obj_t l1i = hwloc_alloc_setup_object(HWLOC_OBJ_L1ICACHE, -1);
l1i->cpuset = hwloc_bitmap_dup(obj->cpuset);
hwloc_debug_1arg_bitmap("L1icache %u has cpuset %s\n",
j, l1i->cpuset);
l1i->attr->cache.depth = i;
l1i->attr->cache.size = l1icachesize;
l1i->attr->cache.linesize = cachelinesize;
l1i->attr->cache.associativity = 0;
l1i->attr->cache.type = HWLOC_OBJ_CACHE_INSTRUCTION;
hwloc_insert_object_by_cpuset(topology, l1i);
}
if (i) {
hwloc_debug_2args_bitmap("L%ucache %u has cpuset %s\n",
i, j, obj->cpuset);
obj->attr->cache.depth = i;
obj->attr->cache.size = cachesize[i];
obj->attr->cache.linesize = cachelinesize;
if (i <= sizeof(cacheways) / sizeof(cacheways[0]))
obj->attr->cache.associativity = cacheways[i-1];
else
obj->attr->cache.associativity = 0;
if (i == 1 && l1icachesize)
obj->attr->cache.type = HWLOC_OBJ_CACHE_DATA;
else
obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
} else {
hwloc_debug_1arg_bitmap("node %u has cpuset %s\n",
j, obj->cpuset);
obj->memory.local_memory = cachesize[i];
obj->memory.page_types_len = 2;
obj->memory.page_types = malloc(2*sizeof(*obj->memory.page_types));
memset(obj->memory.page_types, 0, 2*sizeof(*obj->memory.page_types));
obj->memory.page_types[0].size = hwloc_getpagesize();
#ifdef HAVE__SC_LARGE_PAGESIZE
obj->memory.page_types[1].size = sysconf(_SC_LARGE_PAGESIZE);
#endif
}
if (hwloc_filter_check_keep_object_type(topology, obj->type))
hwloc_insert_object_by_cpuset(topology, obj);
else
hwloc_free_unlinked_object(obj); /* FIXME: don't built at all, just build the cpuset in case l1i needs it */
}
}
}
out:
free(cacheconfig);
free(cachesize);
free(cacheconfig32);
}
/* add PU objects */
hwloc_setup_pu_level(topology, nprocs);
hwloc_obj_add_info(topology->levels[0][0], "Backend", "Darwin");
hwloc_add_uname_info(topology, NULL);
return 0;
}
static int
hwloc_look_darwin(struct hwloc_backend *backend)
{
struct hwloc_topology *topology = backend->topology;
int64_t _nprocs;
unsigned nprocs;
int64_t _npackages;
unsigned i, j, cpu;
struct hwloc_obj *obj;
size_t size;
int64_t l1dcachesize, l1icachesize;
int64_t cacheways[2];
int64_t l2cachesize;
int64_t l3cachesize;
int64_t cachelinesize;
int64_t memsize;
int64_t _tmp;
char cpumodel[64];
char cpuvendor[64];
char cpufamilynumber[20], cpumodelnumber[20], cpustepping[20];
int gotnuma = 0;
int gotnumamemory = 0;
if (topology->levels[0][0]->cpuset)
/* somebody discovered things */
return -1;
hwloc_alloc_root_sets(topology->levels[0][0]);
/* Don't use hwloc_fallback_nbprocessors() because it would return online cpus only,
* while we need all cpus when computing logical_per_package, etc below.
* We don't know which CPUs are offline, but Darwin doesn't support binding anyway.
*
* TODO: try hw.logicalcpu_max
*/
if (hwloc_get_sysctlbyname("hw.logicalcpu", &_nprocs) || _nprocs <= 0)
/* fallback to deprecated way */
if (hwloc_get_sysctlbyname("hw.ncpu", &_nprocs) || _nprocs <= 0)
return -1;
nprocs = _nprocs;
topology->support.discovery->pu = 1;
hwloc_debug("%u procs\n", nprocs);
size = sizeof(cpuvendor);
if (sysctlbyname("machdep.cpu.vendor", cpuvendor, &size, NULL, 0))
cpuvendor[0] = '\0';
size = sizeof(cpumodel);
if (sysctlbyname("machdep.cpu.brand_string", cpumodel, &size, NULL, 0))
cpumodel[0] = '\0';
if (hwloc_get_sysctlbyname("machdep.cpu.family", &_tmp))
cpufamilynumber[0] = '\0';
else
snprintf(cpufamilynumber, sizeof(cpufamilynumber), "%lld", (long long) _tmp);
if (hwloc_get_sysctlbyname("machdep.cpu.model", &_tmp))
cpumodelnumber[0] = '\0';
else
snprintf(cpumodelnumber, sizeof(cpumodelnumber), "%lld", (long long) _tmp);
/* .extfamily and .extmodel are already added to .family and .model */
if (hwloc_get_sysctlbyname("machdep.cpu.stepping", &_tmp))
cpustepping[0] = '\0';
else
snprintf(cpustepping, sizeof(cpustepping), "%lld", (long long) _tmp);
if (!hwloc_get_sysctlbyname("hw.packages", &_npackages) && _npackages > 0) {
unsigned npackages = _npackages;
int64_t _cores_per_package;
unsigned cores_per_package;
int64_t _logical_per_package;
unsigned logical_per_package;
hwloc_debug("%u packages\n", npackages);
if (!hwloc_get_sysctlbyname("machdep.cpu.thread_count", &_logical_per_package) && _logical_per_package > 0)
/* official/modern way */
logical_per_package = _logical_per_package;
else if (!hwloc_get_sysctlbyname("machdep.cpu.logical_per_package", &_logical_per_package) && _logical_per_package > 0)
/* old way, gives the max supported by this "kind" of processor,
* can be larger than the actual number for this model.
*/
logical_per_package = _logical_per_package;
else
/* Assume the trivia. */
logical_per_package = nprocs / npackages;
hwloc_debug("%u threads per package\n", logical_per_package);
if (nprocs == npackages * logical_per_package
&& hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_PACKAGE))
for (i = 0; i < npackages; i++) {
obj = hwloc_alloc_setup_object(topology, HWLOC_OBJ_PACKAGE, i);
obj->cpuset = hwloc_bitmap_alloc();
for (cpu = i*logical_per_package; cpu < (i+1)*logical_per_package; cpu++)
hwloc_bitmap_set(obj->cpuset, cpu);
hwloc_debug_1arg_bitmap("package %u has cpuset %s\n",
i, obj->cpuset);
if (cpuvendor[0] != '\0')
hwloc_obj_add_info(obj, "CPUVendor", cpuvendor);
if (cpumodel[0] != '\0')
hwloc_obj_add_info(obj, "CPUModel", cpumodel);
if (cpufamilynumber[0] != '\0')
hwloc_obj_add_info(obj, "CPUFamilyNumber", cpufamilynumber);
if (cpumodelnumber[0] != '\0')
hwloc_obj_add_info(obj, "CPUModelNumber", cpumodelnumber);
if (cpustepping[0] != '\0')
hwloc_obj_add_info(obj, "CPUStepping", cpustepping);
hwloc_insert_object_by_cpuset(topology, obj);
}
else {
if (cpuvendor[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUVendor", cpuvendor);
if (cpumodel[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUModel", cpumodel);
if (cpufamilynumber[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUFamilyNumber", cpufamilynumber);
if (cpumodelnumber[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUModelNumber", cpumodelnumber);
if (cpustepping[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUStepping", cpustepping);
}
if (!hwloc_get_sysctlbyname("machdep.cpu.core_count", &_cores_per_package) && _cores_per_package > 0)
/* official/modern way */
cores_per_package = _cores_per_package;
else if (!hwloc_get_sysctlbyname("machdep.cpu.cores_per_package", &_cores_per_package) && _cores_per_package > 0)
/* old way, gives the max supported by this "kind" of processor,
* can be larger than the actual number for this model.
*/
cores_per_package = _cores_per_package;
else
/* no idea */
cores_per_package = 0;
if (cores_per_package > 0
&& hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_CORE)) {
hwloc_debug("%u cores per package\n", cores_per_package);
if (!(logical_per_package % cores_per_package))
for (i = 0; i < npackages * cores_per_package; i++) {
obj = hwloc_alloc_setup_object(topology, HWLOC_OBJ_CORE, i);
obj->cpuset = hwloc_bitmap_alloc();
for (cpu = i*(logical_per_package/cores_per_package);
cpu < (i+1)*(logical_per_package/cores_per_package);
cpu++)
hwloc_bitmap_set(obj->cpuset, cpu);
hwloc_debug_1arg_bitmap("core %u has cpuset %s\n",
i, obj->cpuset);
hwloc_insert_object_by_cpuset(topology, obj);
}
}
} else {
if (cpuvendor[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUVendor", cpuvendor);
if (cpumodel[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUModel", cpumodel);
if (cpufamilynumber[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUFamilyNumber", cpufamilynumber);
if (cpumodelnumber[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUModelNumber", cpumodelnumber);
if (cpustepping[0] != '\0')
hwloc_obj_add_info(topology->levels[0][0], "CPUStepping", cpustepping);
}
if (hwloc_get_sysctlbyname("hw.l1dcachesize", &l1dcachesize))
l1dcachesize = 0;
if (hwloc_get_sysctlbyname("hw.l1icachesize", &l1icachesize))
l1icachesize = 0;
if (hwloc_get_sysctlbyname("hw.l2cachesize", &l2cachesize))
l2cachesize = 0;
if (hwloc_get_sysctlbyname("hw.l3cachesize", &l3cachesize))
l3cachesize = 0;
if (hwloc_get_sysctlbyname("machdep.cpu.cache.L1_associativity", &cacheways[0]))
cacheways[0] = 0;
else if (cacheways[0] == 0xff)
cacheways[0] = -1;
if (hwloc_get_sysctlbyname("machdep.cpu.cache.L2_associativity", &cacheways[1]))
cacheways[1] = 0;
else if (cacheways[1] == 0xff)
cacheways[1] = -1;
if (hwloc_get_sysctlbyname("hw.cachelinesize", &cachelinesize))
cachelinesize = 0;
if (hwloc_get_sysctlbyname("hw.memsize", &memsize))
memsize = 0;
if (!sysctlbyname("hw.cacheconfig", NULL, &size, NULL, 0)) {
unsigned n = size / sizeof(uint32_t);
uint64_t cacheconfig[n];
uint64_t cachesize[n];
uint32_t cacheconfig32[n];
if ((!sysctlbyname("hw.cacheconfig", cacheconfig, &size, NULL, 0))) {
/* Yeech. Darwin seemingly has changed from 32bit to 64bit integers for
* cacheconfig, with apparently no way for detection. Assume the machine
* won't have more than 4 billion cpus */
if (cacheconfig[0] > 0xFFFFFFFFUL) {
memcpy(cacheconfig32, cacheconfig, size);
for (i = 0 ; i < size / sizeof(uint32_t); i++)
cacheconfig[i] = cacheconfig32[i];
}
memset(cachesize, 0, sizeof(uint64_t) * n);
size = sizeof(uint64_t) * n;
if (sysctlbyname("hw.cachesize", cachesize, &size, NULL, 0)) {
if (n > 0)
cachesize[0] = memsize;
if (n > 1)
cachesize[1] = l1dcachesize;
if (n > 2)
cachesize[2] = l2cachesize;
if (n > 3)
cachesize[3] = l3cachesize;
}
hwloc_debug("%s", "caches");
for (i = 0; i < n && cacheconfig[i]; i++)
hwloc_debug(" %"PRIu64"(%"PRIu64"kB)", cacheconfig[i], cachesize[i] / 1024);
/* Now we know how many caches there are */
n = i;
hwloc_debug("\n%u cache levels\n", n - 1);
/* For each cache level (0 is memory) */
for (i = 0; i < n; i++) {
/* cacheconfig tells us how many cpus share it, let's iterate on each cache */
for (j = 0; j < (nprocs / cacheconfig[i]); j++) {
if (!i) {
obj = hwloc_alloc_setup_object(topology, HWLOC_OBJ_NUMANODE, j);
obj->nodeset = hwloc_bitmap_alloc();
hwloc_bitmap_set(obj->nodeset, j);
gotnuma++;
} else {
obj = hwloc_alloc_setup_object(topology, HWLOC_OBJ_L1CACHE+i-1, HWLOC_UNKNOWN_INDEX);
}
obj->cpuset = hwloc_bitmap_alloc();
for (cpu = j*cacheconfig[i];
cpu < ((j+1)*cacheconfig[i]);
cpu++)
hwloc_bitmap_set(obj->cpuset, cpu);
if (i == 1 && l1icachesize
&& hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_L1ICACHE)) {
/* FIXME assuming that L1i and L1d are shared the same way. Darwin
* does not yet provide a way to know. */
hwloc_obj_t l1i = hwloc_alloc_setup_object(topology, HWLOC_OBJ_L1ICACHE, HWLOC_UNKNOWN_INDEX);
l1i->cpuset = hwloc_bitmap_dup(obj->cpuset);
hwloc_debug_1arg_bitmap("L1icache %u has cpuset %s\n",
j, l1i->cpuset);
l1i->attr->cache.depth = i;
l1i->attr->cache.size = l1icachesize;
l1i->attr->cache.linesize = cachelinesize;
l1i->attr->cache.associativity = 0;
l1i->attr->cache.type = HWLOC_OBJ_CACHE_INSTRUCTION;
hwloc_insert_object_by_cpuset(topology, l1i);
}
if (i) {
hwloc_debug_2args_bitmap("L%ucache %u has cpuset %s\n",
i, j, obj->cpuset);
obj->attr->cache.depth = i;
obj->attr->cache.size = cachesize[i];
obj->attr->cache.linesize = cachelinesize;
if (i <= sizeof(cacheways) / sizeof(cacheways[0]))
obj->attr->cache.associativity = cacheways[i-1];
else
obj->attr->cache.associativity = 0;
if (i == 1 && l1icachesize)
obj->attr->cache.type = HWLOC_OBJ_CACHE_DATA;
else
obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
} else {
hwloc_debug_1arg_bitmap("node %u has cpuset %s\n",
j, obj->cpuset);
if (cachesize[i]) {
obj->attr->numanode.local_memory = cachesize[i];
gotnumamemory++;
}
obj->attr->numanode.page_types_len = 2;
obj->attr->numanode.page_types = malloc(2*sizeof(*obj->attr->numanode.page_types));
memset(obj->attr->numanode.page_types, 0, 2*sizeof(*obj->attr->numanode.page_types));
obj->attr->numanode.page_types[0].size = hwloc_getpagesize();
#if HAVE_DECL__SC_LARGE_PAGESIZE
obj->attr->numanode.page_types[1].size = sysconf(_SC_LARGE_PAGESIZE);
#endif
}
if (hwloc_filter_check_keep_object_type(topology, obj->type))
hwloc_insert_object_by_cpuset(topology, obj);
else
hwloc_free_unlinked_object(obj); /* FIXME: don't built at all, just build the cpuset in case l1i needs it */
}
}
}
}
if (gotnuma)
topology->support.discovery->numa = 1;
if (gotnumamemory)
topology->support.discovery->numa = 1;
/* add PU objects */
hwloc_setup_pu_level(topology, nprocs);
hwloc_obj_add_info(topology->levels[0][0], "Backend", "Darwin");
hwloc_add_uname_info(topology, NULL);
return 0;
}
