tbb_bind.cpp

Go to the documentation of this file.

/*
    Copyright (c) 2005-2020 Intel Corporation

    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at

        http://www.apache.org/licenses/LICENSE-2.0

    Unless required by applicable law or agreed to in writing, software
    distributed under the License is distributed on an "AS IS" BASIS,
    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    See the License for the specific language governing permissions and
    limitations under the License.
*/

#include "../tbb/tbb_assert_impl.h" // Out-of-line TBB assertion handling routines are instantiated here.
#include "tbb/tbb_stddef.h"

#if _MSC_VER && !__INTEL_COMPILER
#pragma warning( push )
#pragma warning( disable : 4100 )
#endif
#include <hwloc.h>
#if _MSC_VER && !__INTEL_COMPILER
#pragma warning( pop )
#endif

#include <vector>

// Most of hwloc calls returns negative exit code on error.
// This macro tracks error codes that are returned from the hwloc interfaces.
#define assertion_hwloc_wrapper(command, ...) \
        __TBB_ASSERT_EX( (command(__VA_ARGS__)) >= 0, "Error occurred during call to hwloc API.");

namespace tbb {
namespace internal {

//------------------------------------------------------------------------
// Information about the machine's hardware TBB is happen to work on
//------------------------------------------------------------------------
class platform_topology {
    friend class numa_affinity_handler;

    // TODO: add the `my_` prefix to the members
    hwloc_topology_t topology;
    hwloc_cpuset_t   process_cpu_affinity_mask;
    hwloc_nodeset_t  process_node_affinity_mask;
    std::vector<hwloc_cpuset_t>  affinity_masks_list;

    std::vector<int> default_concurrency_list;
    std::vector<int> numa_indexes_list;
    int  numa_nodes_count;

    enum init_stages { uninitialized,
                       started,
                       topology_allocated,
                       topology_loaded,
                       topology_parsed } initialization_state;

    // Binding threads to NUMA nodes that locates in another Windows Processor groups
    // is allowed only if machine topology contains several Windows Processors groups
    // and process affinity mask wasn`t limited manually (affinity mask cannot violates
    // processors group boundaries).
    bool intergroup_binding_allowed(size_t groups_num) { return groups_num > 1; }

    platform_topology() : topology(NULL),
                          process_cpu_affinity_mask(NULL),
                          process_node_affinity_mask(NULL),
                          numa_nodes_count(0),
                          initialization_state(uninitialized) {}

public:
    typedef hwloc_cpuset_t             affinity_mask;
    typedef hwloc_const_cpuset_t const_affinity_mask;

    static platform_topology& instance() {
        static platform_topology topology;
        return topology;
    }

    bool is_topology_parsed() { return initialization_state == topology_parsed; }

    void initialize( size_t groups_num ) {
        if ( initialization_state != uninitialized )
            return;
        initialization_state = started;

        // Parse topology
        if ( hwloc_topology_init( &topology ) == 0 ) {
            initialization_state = topology_allocated;
            if ( hwloc_topology_load( topology ) == 0 ) {
                initialization_state = topology_loaded;
            }
        }

        // Fill parameters with stubs if topology parsing is broken.
        if ( initialization_state != topology_loaded ) {
            if ( initialization_state == topology_allocated ) {
                hwloc_topology_destroy(topology);
            }
            numa_nodes_count = 1;
            numa_indexes_list.push_back(-1);
            default_concurrency_list.push_back(-1);
            return;
        }

        // Getting process affinity mask
        if ( intergroup_binding_allowed(groups_num) ) {
            process_cpu_affinity_mask  = hwloc_bitmap_dup(hwloc_topology_get_complete_cpuset (topology));
            process_node_affinity_mask = hwloc_bitmap_dup(hwloc_topology_get_complete_nodeset(topology));
        } else {
            process_cpu_affinity_mask  = hwloc_bitmap_alloc();
            process_node_affinity_mask = hwloc_bitmap_alloc();

            assertion_hwloc_wrapper(hwloc_get_cpubind, topology, process_cpu_affinity_mask, 0);
            hwloc_cpuset_to_nodeset(topology, process_cpu_affinity_mask, process_node_affinity_mask);
        }

        // If system contains no NUMA nodes, HWLOC 1.11 returns an infinitely filled bitmap.
        // hwloc_bitmap_weight() returns negative value for such bitmaps, so we use this check
        // to change way of topology initialization.
        if (hwloc_bitmap_weight(process_node_affinity_mask) < 0) {
            numa_nodes_count = 1;
            numa_indexes_list.push_back(0);
            default_concurrency_list.push_back(hwloc_bitmap_weight(process_cpu_affinity_mask));

            affinity_masks_list.push_back(hwloc_bitmap_dup(process_cpu_affinity_mask));
            initialization_state = topology_parsed;
            return;
        }

        // Get number of available NUMA nodes
        numa_nodes_count = hwloc_bitmap_weight(process_node_affinity_mask);
        __TBB_ASSERT(numa_nodes_count > 0, "Any system must contain one or more NUMA nodes");

        // Get NUMA logical indexes list
        unsigned counter = 0;
        int i = 0;
        int max_numa_index = -1;
        numa_indexes_list.resize(numa_nodes_count);
        hwloc_obj_t node_buffer;
        hwloc_bitmap_foreach_begin(i, process_node_affinity_mask) {
            node_buffer = hwloc_get_obj_by_type(topology, HWLOC_OBJ_NUMANODE, i);
            numa_indexes_list[counter] = static_cast<int>(node_buffer->logical_index);

            if ( numa_indexes_list[counter] > max_numa_index ) {
                max_numa_index = numa_indexes_list[counter];
            }

            counter++;
        } hwloc_bitmap_foreach_end();
        __TBB_ASSERT(max_numa_index >= 0, "Maximal NUMA index must not be negative");

        // Fill concurrency and affinity masks lists
        default_concurrency_list.resize(max_numa_index + 1);
        affinity_masks_list.resize(max_numa_index + 1);

        int index = 0;
        hwloc_bitmap_foreach_begin(i, process_node_affinity_mask) {
            node_buffer = hwloc_get_obj_by_type(topology, HWLOC_OBJ_NUMANODE, i);
            index = static_cast<int>(node_buffer->logical_index);

            hwloc_cpuset_t& current_mask = affinity_masks_list[index];
            current_mask = hwloc_bitmap_dup(node_buffer->cpuset);

            hwloc_bitmap_and(current_mask, current_mask, process_cpu_affinity_mask);
            __TBB_ASSERT(!hwloc_bitmap_iszero(current_mask), "hwloc detected unavailable NUMA node");
            default_concurrency_list[index] = hwloc_bitmap_weight(current_mask);
        } hwloc_bitmap_foreach_end();
        initialization_state = topology_parsed;
    }

    ~platform_topology() {
        if ( is_topology_parsed() ) {
            for (int i = 0; i < numa_nodes_count; i++) {
                hwloc_bitmap_free(affinity_masks_list[numa_indexes_list[i]]);
            }
            hwloc_bitmap_free(process_node_affinity_mask);
            hwloc_bitmap_free(process_cpu_affinity_mask);
        }

        if ( initialization_state >= topology_allocated ) {
            hwloc_topology_destroy(topology);
        }

        initialization_state = uninitialized;
    }

    void fill(int& nodes_count, int*& indexes_list, int*& concurrency_list ) {
        __TBB_ASSERT(is_topology_parsed(), "Trying to get access to uninitialized platform_topology");
        nodes_count = numa_nodes_count;
        indexes_list = &numa_indexes_list.front();
        concurrency_list = &default_concurrency_list.front();
    }

    affinity_mask allocate_process_affinity_mask() {
        __TBB_ASSERT(is_topology_parsed(), "Trying to get access to uninitialized platform_topology");
        return hwloc_bitmap_dup(process_cpu_affinity_mask);
    }

    void free_affinity_mask( affinity_mask mask_to_free ) {
        hwloc_bitmap_free(mask_to_free); // If bitmap is NULL, no operation is performed.
    }

    void store_current_affinity_mask( affinity_mask current_mask ) {
        assertion_hwloc_wrapper(hwloc_get_cpubind, topology, current_mask, HWLOC_CPUBIND_THREAD);

        hwloc_bitmap_and(current_mask, current_mask, process_cpu_affinity_mask);
        __TBB_ASSERT(!hwloc_bitmap_iszero(current_mask),
            "Current affinity mask must intersects with process affinity mask");
    }

    void set_new_affinity_mask( const_affinity_mask new_mask ) {
        assertion_hwloc_wrapper(hwloc_set_cpubind, topology, new_mask, HWLOC_CPUBIND_THREAD);
    }

    const_affinity_mask get_node_affinity_mask( int node_index ) {
        __TBB_ASSERT((int)affinity_masks_list.size() > node_index,
            "Trying to get affinity mask for uninitialized NUMA node");
        return affinity_masks_list[node_index];
    }
};

class binding_handler {
    // Following vector saves thread affinity mask on scheduler entry to return it to this thread 
    // on scheduler exit.
    typedef std::vector<platform_topology::affinity_mask> affinity_masks_container;
    affinity_masks_container affinity_backup;

public:
    binding_handler( size_t size ) : affinity_backup(size) {
        for (affinity_masks_container::iterator it = affinity_backup.begin();
             it != affinity_backup.end(); it++) {
            *it = platform_topology::instance().allocate_process_affinity_mask();
        }
    }

    ~binding_handler() {
        for (affinity_masks_container::iterator it = affinity_backup.begin();
             it != affinity_backup.end(); it++) {
            platform_topology::instance().free_affinity_mask(*it);
        }
    }

    void bind_thread_to_node( unsigned slot_num, unsigned numa_node_id ) {
        __TBB_ASSERT(slot_num < affinity_backup.size(),
            "The slot number is greater than the number of slots in the arena");
        __TBB_ASSERT(platform_topology::instance().is_topology_parsed(),
            "Trying to get access to uninitialized platform_topology");
        platform_topology::instance().store_current_affinity_mask(affinity_backup[slot_num]);

        platform_topology::instance().set_new_affinity_mask(
            platform_topology::instance().get_node_affinity_mask(numa_node_id));
    }

    void restore_previous_affinity_mask( unsigned slot_num ) {
        __TBB_ASSERT(platform_topology::instance().is_topology_parsed(),
            "Trying to get access to uninitialized platform_topology");
        platform_topology::instance().set_new_affinity_mask(affinity_backup[slot_num]);
    };

};

extern "C" { // exported to TBB interfaces

void initialize_numa_topology( size_t groups_num,
                               int& nodes_count, int*& indexes_list, int*& concurrency_list ) {
    platform_topology::instance().initialize(groups_num);
    platform_topology::instance().fill(nodes_count, indexes_list, concurrency_list);
}

binding_handler* allocate_binding_handler(int slot_num) {
    __TBB_ASSERT(slot_num > 0, "Trying to create numa handler for 0 threads.");
    return new binding_handler(slot_num);
}

void deallocate_binding_handler(binding_handler* handler_ptr) {
    __TBB_ASSERT(handler_ptr != NULL, "Trying to deallocate NULL pointer.");
    delete handler_ptr;
}

void bind_to_node(binding_handler* handler_ptr, int slot_num, int numa_id) {
    __TBB_ASSERT(handler_ptr != NULL, "Trying to get access to uninitialized metadata.");
    __TBB_ASSERT(platform_topology::instance().is_topology_parsed(),
        "Trying to get access to uninitialized platform_topology.");
    handler_ptr->bind_thread_to_node(slot_num, numa_id);
}

void restore_affinity(binding_handler* handler_ptr, int slot_num) {
    __TBB_ASSERT(handler_ptr != NULL, "Trying to get access to uninitialized metadata.");
    __TBB_ASSERT(platform_topology::instance().is_topology_parsed(),
        "Trying to get access to uninitialized platform_topology.");
    handler_ptr->restore_previous_affinity_mask(slot_num);
}

} // extern "C"

} // namespace internal
} // namespace tbb

#undef assertion_hwloc_wrapper