topology

#include <compatibility/hpx/util/cache_aligned_data.hpp>

#include <compatibility/hpx/runtime/threads/topology.hpp>

#include <compatibility/hpx/runtime/threads/cpu_mask.hpp>

#include <hpx/topology/topology.hpp>

namespace hpx
namespace threads

Typedefs

using hpx::threads::hwloc_bitmap_ptr = typedef std::shared_ptr<hpx_hwloc_bitmap_wrapper>

Enums

enum hpx_hwloc_membind_policy

Please see hwloc documentation for the corresponding enums HWLOC_MEMBIND_XXX.

Values:

membind_default = HWLOC_MEMBIND_DEFAULT
membind_firsttouch = HWLOC_MEMBIND_FIRSTTOUCH
membind_bind = HWLOC_MEMBIND_BIND
membind_interleave = HWLOC_MEMBIND_INTERLEAVE
membind_replicate = HWLOC_MEMBIND_REPLICATE
membind_nexttouch = HWLOC_MEMBIND_NEXTTOUCH
membind_mixed = HWLOC_MEMBIND_MIXED
membind_user = HWLOC_MEMBIND_MIXED + 256

Functions

topology &create_topology()
std::size_t hardware_concurrency()
std::size_t get_memory_page_size()
struct hpx_hwloc_bitmap_wrapper

Public Functions

HPX_NON_COPYABLE(hpx_hwloc_bitmap_wrapper)
hpx_hwloc_bitmap_wrapper()
hpx_hwloc_bitmap_wrapper(void *bmp)
~hpx_hwloc_bitmap_wrapper()
void reset(hwloc_bitmap_t bmp)
operator bool() const
hwloc_bitmap_t get_bmp() const

Private Members

hwloc_bitmap_t bmp_

Friends

std::ostream &operator<<(std::ostream &os, hpx_hwloc_bitmap_wrapper const *bmp)
struct topology

Public Functions

topology()
~topology()
std::size_t get_socket_number(std::size_t num_thread, error_code& = throws) const

Return the Socket number of the processing unit the given thread is running on.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

std::size_t get_numa_node_number(std::size_t num_thread, error_code& = throws) const

Return the NUMA node number of the processing unit the given thread is running on.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

mask_cref_type get_machine_affinity_mask(error_code &ec = throws) const

Return a bit mask where each set bit corresponds to a processing unit available to the application.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

mask_type get_service_affinity_mask(mask_cref_type used_processing_units, error_code &ec = throws) const

Return a bit mask where each set bit corresponds to a processing unit available to the service threads in the application.

Parameters
  • used_processing_units: [in] This is the mask of processing units which are not available for service threads.
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

mask_cref_type get_socket_affinity_mask(std::size_t num_thread, error_code &ec = throws) const

Return a bit mask where each set bit corresponds to a processing unit available to the given thread inside the socket it is running on.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

mask_cref_type get_numa_node_affinity_mask(std::size_t num_thread, error_code &ec = throws) const

Return a bit mask where each set bit corresponds to a processing unit available to the given thread inside the NUMA domain it is running on.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

mask_type get_numa_node_affinity_mask_from_numa_node(std::size_t num_node) const

Return a bit mask where each set bit corresponds to a processing unit associated with the given NUMA node.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

mask_cref_type get_core_affinity_mask(std::size_t num_thread, error_code &ec = throws) const

Return a bit mask where each set bit corresponds to a processing unit available to the given thread inside the core it is running on.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

mask_cref_type get_thread_affinity_mask(std::size_t num_thread, error_code &ec = throws) const

Return a bit mask where each set bit corresponds to a processing unit available to the given thread.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

void set_thread_affinity_mask(mask_cref_type mask, error_code &ec = throws) const

Use the given bit mask to set the affinity of the given thread. Each set bit corresponds to a processing unit the thread will be allowed to run on.

Note
Use this function on systems where the affinity must be set from inside the thread itself.
Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

mask_type get_thread_affinity_mask_from_lva(void const *lva, error_code &ec = throws) const

Return a bit mask where each set bit corresponds to a processing unit co-located with the memory the given address is currently allocated on.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

void print_affinity_mask(std::ostream &os, std::size_t num_thread, mask_cref_type m, const std::string &pool_name) const

Prints the.

Parameters
  • m: to os in a human readable form

bool reduce_thread_priority(error_code &ec = throws) const

Reduce thread priority of the current thread.

Parameters
  • ec: [in,out] this represents the error status on exit, if this is pre-initialized to hpx::throws the function will throw on error instead.

std::size_t get_number_of_sockets() const

Return the number of available NUMA domains.

std::size_t get_number_of_numa_nodes() const

Return the number of available NUMA domains.

std::size_t get_number_of_cores() const

Return the number of available cores.

std::size_t get_number_of_pus() const

Return the number of available hardware processing units.

std::size_t get_number_of_numa_node_cores(std::size_t numa) const

Return number of cores in given numa domain.

std::size_t get_number_of_numa_node_pus(std::size_t numa) const

Return number of processing units in a given numa domain.

std::size_t get_number_of_socket_pus(std::size_t socket) const

Return number of processing units in a given socket.

std::size_t get_number_of_core_pus(std::size_t core) const

Return number of processing units in given core.

std::size_t get_number_of_socket_cores(std::size_t socket) const

Return number of cores units in given socket.

std::size_t get_core_number(std::size_t num_thread, error_code& = throws) const
std::size_t get_pu_number(std::size_t num_core, std::size_t num_pu, error_code &ec = throws) const
mask_type get_cpubind_mask(error_code &ec = throws) const
mask_type get_cpubind_mask(std::thread &handle, error_code &ec = throws) const
hwloc_bitmap_ptr cpuset_to_nodeset(mask_cref_type cpuset) const

convert a cpu mask into a numa node mask in hwloc bitmap form

void write_to_log() const
void *allocate(std::size_t len) const

This is equivalent to malloc(), except that it tries to allocate page-aligned memory from the OS.

void *allocate_membind(std::size_t len, hwloc_bitmap_ptr bitmap, hpx_hwloc_membind_policy policy, int flags) const

allocate memory with binding to a numa node set as specified by the policy and flags (see hwloc docs)

threads::mask_type get_area_membind_nodeset(const void *addr, std::size_t len) const
bool set_area_membind_nodeset(const void *addr, std::size_t len, void *nodeset) const
int get_numa_domain(const void *addr) const
void deallocate(void *addr, std::size_t len) const

Free memory that was previously allocated by allocate.

void print_vector(std::ostream &os, std::vector<std::size_t> const &v) const
void print_mask_vector(std::ostream &os, std::vector<mask_type> const &v) const
void print_hwloc(std::ostream&) const
mask_type init_socket_affinity_mask_from_socket(std::size_t num_socket) const
mask_type init_numa_node_affinity_mask_from_numa_node(std::size_t num_numa_node) const
mask_type init_core_affinity_mask_from_core(std::size_t num_core, mask_cref_type default_mask = empty_mask) const
mask_type init_thread_affinity_mask(std::size_t num_thread) const
mask_type init_thread_affinity_mask(std::size_t num_core, std::size_t num_pu) const
hwloc_bitmap_t mask_to_bitmap(mask_cref_type mask, hwloc_obj_type_t htype) const
mask_type bitmap_to_mask(hwloc_bitmap_t bitmap, hwloc_obj_type_t htype) const

Private Types

using mutex_type = hpx::util::spinlock

Private Functions

std::size_t init_node_number(std::size_t num_thread, hwloc_obj_type_t type)
std::size_t init_socket_number(std::size_t num_thread)
std::size_t init_numa_node_number(std::size_t num_thread)
std::size_t init_core_number(std::size_t num_thread)
void extract_node_mask(hwloc_obj_t parent, mask_type &mask) const
std::size_t extract_node_count(hwloc_obj_t parent, hwloc_obj_type_t type, std::size_t count) const
mask_type init_machine_affinity_mask() const
mask_type init_socket_affinity_mask(std::size_t num_thread) const
mask_type init_numa_node_affinity_mask(std::size_t num_thread) const
mask_type init_core_affinity_mask(std::size_t num_thread) const
void init_num_of_pus()

Private Members

hwloc_topology_t topo
std::size_t num_of_pus_
mutex_type topo_mtx
std::vector<std::size_t> socket_numbers_
std::vector<std::size_t> numa_node_numbers_
std::vector<std::size_t> core_numbers_
mask_type machine_affinity_mask_
std::vector<mask_type> socket_affinity_masks_
std::vector<mask_type> numa_node_affinity_masks_
std::vector<mask_type> core_affinity_masks_
std::vector<mask_type> thread_affinity_masks_

Private Static Attributes

mask_type empty_mask
std::size_t memory_page_size_
const std::size_t pu_offset = 0
const std::size_t core_offset = 0

Friends

std::size_t get_memory_page_size()

#include <hpx/topology/cpu_mask.hpp>