task_arena.h

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.
*/

#ifndef __TBB_task_arena_H
#define __TBB_task_arena_H

#define __TBB_task_arena_H_include_area
#include "internal/_warning_suppress_enable_notice.h"

#include "task.h"
#include "tbb_exception.h"
#include "internal/_template_helpers.h"
#if __TBB_NUMA_SUPPORT
#include "info.h"
#endif /*__TBB_NUMA_SUPPORT*/
#if TBB_USE_THREADING_TOOLS
#include "atomic.h" // for as_atomic
#endif
#include "aligned_space.h"

namespace tbb {

namespace this_task_arena {
    int max_concurrency();
} // namespace this_task_arena

namespace internal {

    class arena;
    class task_scheduler_observer_v3;
} // namespace internal

namespace interface7 {
class task_arena;

namespace internal {
using namespace tbb::internal; //e.g. function_task from task.h

class delegate_base : no_assign {
public:
    virtual void operator()() const = 0;
    virtual ~delegate_base() {}
};

// If decltype is available, the helper detects the return type of functor of specified type,
// otherwise it defines the void type.
template <typename F>
struct return_type_or_void {
#if __TBB_CPP11_DECLTYPE_PRESENT && !__TBB_CPP11_DECLTYPE_OF_FUNCTION_RETURN_TYPE_BROKEN
    typedef decltype(declval<F>()()) type;
#else
    typedef void type;
#endif
};

template<typename F, typename R>
class delegated_function : public delegate_base {
    F &my_func;
    tbb::aligned_space<R> my_return_storage;
    // The function should be called only once.
    void operator()() const __TBB_override {
        new (my_return_storage.begin()) R(my_func());
    }
public:
    delegated_function(F& f) : my_func(f) {}
    // The function can be called only after operator() and only once.
    R consume_result() const {
        return tbb::internal::move(*(my_return_storage.begin()));
    }
    ~delegated_function() {
        my_return_storage.begin()->~R();
    }
};

template<typename F>
class delegated_function<F,void> : public delegate_base {
    F &my_func;
    void operator()() const __TBB_override {
        my_func();
    }
public:
    delegated_function(F& f) : my_func(f) {}
    void consume_result() const {}

    friend class task_arena_base;
};

class task_arena_base {
#if __TBB_NUMA_SUPPORT
public:
    // TODO: consider version approach to resolve backward compatibility potential issues.
    struct constraints {
        constraints(numa_node_id id = automatic, int maximal_concurrency = automatic)
            : numa_id(id)
            , max_concurrency(maximal_concurrency)
        {}
        numa_node_id numa_id;
        int max_concurrency;
    };
#endif /*__TBB_NUMA_SUPPORT*/
protected:
    internal::arena* my_arena;

#if __TBB_TASK_GROUP_CONTEXT
    task_group_context *my_context;
#endif

    int my_max_concurrency;

    unsigned my_master_slots;

    intptr_t my_version_and_traits;

    bool my_initialized;

#if __TBB_NUMA_SUPPORT
    numa_node_id my_numa_id;

    // Do not access my_numa_id without the following runtime check.
    // Despite my_numa_id is accesible, it does not exist in task_arena_base on user side
    // if TBB_PREVIEW_NUMA_SUPPORT macro is not defined by the user. To be sure that
    // my_numa_id exists in task_arena_base layout we check the traits.
    // TODO: Consider increasing interface version for task_arena_base instead of this runtime check.
    numa_node_id numa_id() {
        return (my_version_and_traits & numa_support_flag) == numa_support_flag ? my_numa_id : automatic;
    }
#endif

    enum {
        default_flags = 0
#if __TBB_TASK_GROUP_CONTEXT
        | (task_group_context::default_traits & task_group_context::exact_exception)  // 0 or 1 << 16
        , exact_exception_flag = task_group_context::exact_exception // used to specify flag for context directly
#endif
#if __TBB_NUMA_SUPPORT
        , numa_support_flag = 1
#endif
    };

    task_arena_base(int max_concurrency, unsigned reserved_for_masters)
        : my_arena(0)
#if __TBB_TASK_GROUP_CONTEXT
        , my_context(0)
#endif
        , my_max_concurrency(max_concurrency)
        , my_master_slots(reserved_for_masters)
#if __TBB_NUMA_SUPPORT
        , my_version_and_traits(default_flags | numa_support_flag)
#else
        , my_version_and_traits(default_flags)
#endif
        , my_initialized(false)
#if __TBB_NUMA_SUPPORT
        , my_numa_id(automatic)
#endif
        {}

#if __TBB_NUMA_SUPPORT
    task_arena_base(const constraints& constraints_, unsigned reserved_for_masters)
        : my_arena(0)
#if __TBB_TASK_GROUP_CONTEXT
        , my_context(0)
#endif
        , my_max_concurrency(constraints_.max_concurrency)
        , my_master_slots(reserved_for_masters)
        , my_version_and_traits(default_flags | numa_support_flag)
        , my_initialized(false)
        , my_numa_id(constraints_.numa_id )
        {}
#endif /*__TBB_NUMA_SUPPORT*/

    void __TBB_EXPORTED_METHOD internal_initialize();
    void __TBB_EXPORTED_METHOD internal_terminate();
    void __TBB_EXPORTED_METHOD internal_attach();
    void __TBB_EXPORTED_METHOD internal_enqueue( task&, intptr_t ) const;
    void __TBB_EXPORTED_METHOD internal_execute( delegate_base& ) const;
    void __TBB_EXPORTED_METHOD internal_wait() const;
    static int __TBB_EXPORTED_FUNC internal_current_slot();
    static int __TBB_EXPORTED_FUNC internal_max_concurrency( const task_arena * );
public:
    static const int automatic = -1;
    static const int not_initialized = -2;

};

#if __TBB_TASK_ISOLATION
void __TBB_EXPORTED_FUNC isolate_within_arena( delegate_base& d, intptr_t isolation = 0 );

template<typename R, typename F>
R isolate_impl(F& f) {
    delegated_function<F, R> d(f);
    isolate_within_arena(d);
    return d.consume_result();
}
#endif /* __TBB_TASK_ISOLATION */
} // namespace internal

class task_arena : public internal::task_arena_base {
    friend class tbb::internal::task_scheduler_observer_v3;
    friend void task::enqueue(task&, task_arena&
#if __TBB_TASK_PRIORITY
        , priority_t
#endif
    );
    friend int tbb::this_task_arena::max_concurrency();
    void mark_initialized() {
        __TBB_ASSERT( my_arena, "task_arena initialization is incomplete" );
#if __TBB_TASK_GROUP_CONTEXT
        __TBB_ASSERT( my_context, "task_arena initialization is incomplete" );
#endif
#if TBB_USE_THREADING_TOOLS
        // Actual synchronization happens in internal_initialize & internal_attach.
        // The race on setting my_initialized is benign, but should be hidden from Intel(R) Inspector
        internal::as_atomic(my_initialized).fetch_and_store<release>(true);
#else
        my_initialized = true;
#endif
    }

    template<typename F>
    void enqueue_impl( __TBB_FORWARDING_REF(F) f
#if __TBB_TASK_PRIORITY
        , priority_t p = priority_t(0)
#endif
    ) {
#if !__TBB_TASK_PRIORITY
        intptr_t p = 0;
#endif
        initialize();
#if __TBB_TASK_GROUP_CONTEXT
        internal_enqueue(*new(task::allocate_root(*my_context)) internal::function_task< typename internal::strip<F>::type >(internal::forward<F>(f)), p);
#else
        internal_enqueue(*new(task::allocate_root()) internal::function_task< typename internal::strip<F>::type >(internal::forward<F>(f)), p);
#endif /* __TBB_TASK_GROUP_CONTEXT */
    }

    template<typename R, typename F>
    R execute_impl(F& f) {
        initialize();
        internal::delegated_function<F, R> d(f);
        internal_execute(d);
        return d.consume_result();
    }

public:

    task_arena(int max_concurrency_ = automatic, unsigned reserved_for_masters = 1)
        : task_arena_base(max_concurrency_, reserved_for_masters)
    {}

#if __TBB_NUMA_SUPPORT
    task_arena(const constraints& constraints_, unsigned reserved_for_masters = 1)
        : task_arena_base(constraints_, reserved_for_masters)
    {}

    task_arena(const task_arena &s) // copy settings but not the reference or instance
        : task_arena_base(constraints(s.my_numa_id, s.my_max_concurrency), s.my_master_slots)
    {}
#else
    task_arena(const task_arena &s) // copy settings but not the reference or instance
        : task_arena_base(s.my_max_concurrency, s.my_master_slots)
    {}
#endif /*__TBB_NUMA_SUPPORT*/

    struct attach {};

    explicit task_arena( attach )
        : task_arena_base(automatic, 1) // use default settings if attach fails
    {
        internal_attach();
        if( my_arena ) my_initialized = true;
    }

    inline void initialize() {
        if( !my_initialized ) {
            internal_initialize();
            mark_initialized();
        }
    }

    inline void initialize(int max_concurrency_, unsigned reserved_for_masters = 1) {
        // TODO: decide if this call must be thread-safe
        __TBB_ASSERT(!my_arena, "Impossible to modify settings of an already initialized task_arena");
        if( !my_initialized ) {
            my_max_concurrency = max_concurrency_;
            my_master_slots = reserved_for_masters;
            initialize();
        }
    }

#if __TBB_NUMA_SUPPORT
    inline void initialize(constraints constraints_, unsigned reserved_for_masters = 1) {
        // TODO: decide if this call must be thread-safe
        __TBB_ASSERT(!my_arena, "Impossible to modify settings of an already initialized task_arena");
        if( !my_initialized ) {
            my_numa_id = constraints_.numa_id;
            my_max_concurrency = constraints_.max_concurrency;
            my_master_slots = reserved_for_masters;
            initialize();
        }
    }
#endif /*__TBB_NUMA_SUPPORT*/

    inline void initialize(attach) {
        // TODO: decide if this call must be thread-safe
        __TBB_ASSERT(!my_arena, "Impossible to modify settings of an already initialized task_arena");
        if( !my_initialized ) {
            internal_attach();
            if ( !my_arena ) internal_initialize();
            mark_initialized();
        }
    }

    inline void terminate() {
        if( my_initialized ) {
            internal_terminate();
            my_initialized = false;
        }
    }

    ~task_arena() {
        terminate();
    }

    bool is_active() const { return my_initialized; }


#if __TBB_CPP11_RVALUE_REF_PRESENT
    template<typename F>
    void enqueue( F&& f ) {
        enqueue_impl(std::forward<F>(f));
    }
#else
    template<typename F>
    void enqueue( const F& f ) {
        enqueue_impl(f);
    }
#endif

#if __TBB_TASK_PRIORITY
    template<typename F>
#if __TBB_CPP11_RVALUE_REF_PRESENT
    __TBB_DEPRECATED void enqueue( F&& f, priority_t p ) {
#if __TBB_PREVIEW_CRITICAL_TASKS
        __TBB_ASSERT(p == priority_low || p == priority_normal || p == priority_high
                     || p == internal::priority_critical, "Invalid priority level value");
#else
        __TBB_ASSERT(p == priority_low || p == priority_normal || p == priority_high, "Invalid priority level value");
#endif
        enqueue_impl(std::forward<F>(f), p);
    }
#else
    __TBB_DEPRECATED void enqueue( const F& f, priority_t p ) {
#if __TBB_PREVIEW_CRITICAL_TASKS
        __TBB_ASSERT(p == priority_low || p == priority_normal || p == priority_high
                     || p == internal::priority_critical, "Invalid priority level value");
#else
        __TBB_ASSERT(p == priority_low || p == priority_normal || p == priority_high, "Invalid priority level value");
#endif
        enqueue_impl(f,p);
    }
#endif
#endif// __TBB_TASK_PRIORITY

    template<typename F>
    typename internal::return_type_or_void<F>::type execute(F& f) {
        return execute_impl<typename internal::return_type_or_void<F>::type>(f);
    }

    template<typename F>
    typename internal::return_type_or_void<F>::type execute(const F& f) {
        return execute_impl<typename internal::return_type_or_void<F>::type>(f);
    }

#if __TBB_EXTRA_DEBUG
    void debug_wait_until_empty() {
        initialize();
        internal_wait();
    }
#endif //__TBB_EXTRA_DEBUG

    inline static int current_thread_index() {
        return internal_current_slot();
    }

    inline int max_concurrency() const {
        // Handle special cases inside the library
        return (my_max_concurrency>1) ? my_max_concurrency : internal_max_concurrency(this);
    }
};

namespace this_task_arena {
#if __TBB_TASK_ISOLATION
    template<typename F>
    typename internal::return_type_or_void<F>::type isolate(F& f) {
        return internal::isolate_impl<typename internal::return_type_or_void<F>::type>(f);
    }

    template<typename F>
    typename internal::return_type_or_void<F>::type isolate(const F& f) {
        return internal::isolate_impl<typename internal::return_type_or_void<F>::type>(f);
    }
#endif /* __TBB_TASK_ISOLATION */
} // namespace this_task_arena
} // namespace interfaceX

using interface7::task_arena;

namespace this_task_arena {
    using namespace interface7::this_task_arena;

    inline int current_thread_index() {
        int idx = tbb::task_arena::current_thread_index();
        return idx == -1 ? tbb::task_arena::not_initialized : idx;
    }

    inline int max_concurrency() {
        return tbb::task_arena::internal_max_concurrency(NULL);
    }
} // namespace this_task_arena

#if __TBB_TASK_PRIORITY
void task::enqueue( task& t, task_arena& arena, priority_t p ) {
#else
void task::enqueue( task& t, task_arena& arena ) {
    intptr_t p = 0;
#endif
    arena.initialize();
    arena.internal_enqueue(t, p);
}
} // namespace tbb

#include "internal/_warning_suppress_disable_notice.h"
#undef __TBB_task_arena_H_include_area

#endif /* __TBB_task_arena_H */