concurrent_vector.h

/*
    Copyright 2005-2009 Intel Corporation.  All Rights Reserved.

    The source code contained or described herein and all documents related
    to the source code ("Material") are owned by Intel Corporation or its
    suppliers or licensors.  Title to the Material remains with Intel
    Corporation or its suppliers and licensors.  The Material is protected
    by worldwide copyright laws and treaty provisions.  No part of the
    Material may be used, copied, reproduced, modified, published, uploaded,
    posted, transmitted, distributed, or disclosed in any way without
    Intel's prior express written permission.

    No license under any patent, copyright, trade secret or other
    intellectual property right is granted to or conferred upon you by
    disclosure or delivery of the Materials, either expressly, by
    implication, inducement, estoppel or otherwise.  Any license under such
    intellectual property rights must be express and approved by Intel in
    writing.
*/

#ifndef __TBB_concurrent_vector_H
#define __TBB_concurrent_vector_H

#include "tbb_stddef.h"
#include <algorithm>
#include <iterator>
#include <new>
#include <cstring>
#include "atomic.h"
#include "cache_aligned_allocator.h"
#include "blocked_range.h"

#include "tbb_machine.h"

#if _MSC_VER==1500 && !__INTEL_COMPILER
    // VS2008/VC9 seems to have an issue; limits pull in math.h
    #pragma warning( push )
    #pragma warning( disable: 4985 )
#endif
#include <limits> /* std::numeric_limits */
#if _MSC_VER==1500 && !__INTEL_COMPILER
    #pragma warning( pop )
#endif

#if defined(_MSC_VER) && !defined(__INTEL_COMPILER) && defined(_Wp64)
    // Workaround for overzealous compiler warnings in /Wp64 mode
    #pragma warning (push)
    #pragma warning (disable: 4267)
#endif

namespace tbb {

template<typename T, class A = cache_aligned_allocator<T> >
class concurrent_vector;


namespace internal {

    static void *const vector_allocation_error_flag = reinterpret_cast<void*>(size_t(63));
    void* __TBB_EXPORTED_FUNC itt_load_pointer_v3( const void* src );


    class concurrent_vector_base_v3 {
    protected:

        // Basic types declarations
        typedef size_t segment_index_t;
        typedef size_t size_type;

        // Using enumerations due to Mac linking problems of static const variables
        enum {
            // Size constants
            default_initial_segments = 1, // 2 initial items
            pointers_per_short_table = 3, // to fit into 8 words of entire structure
            pointers_per_long_table = sizeof(segment_index_t) * 8 // one segment per bit
        };

        // Segment pointer. Can be zero-initialized
        struct segment_t {
            void* array;
#if TBB_USE_ASSERT
            ~segment_t() {
                __TBB_ASSERT( array <= internal::vector_allocation_error_flag, "should have been freed by clear" );
            }
#endif /* TBB_USE_ASSERT */
        };
 
        // Data fields

        void* (*vector_allocator_ptr)(concurrent_vector_base_v3 &, size_t);

        atomic<size_type> my_first_block;

        atomic<size_type> my_early_size;

        atomic<segment_t*> my_segment;

        segment_t my_storage[pointers_per_short_table];

        // Methods

        concurrent_vector_base_v3() {
            my_early_size = 0;
            my_first_block = 0; // here is not default_initial_segments
            for( segment_index_t i = 0; i < pointers_per_short_table; i++)
                my_storage[i].array = NULL;
            my_segment = my_storage;
        }
        __TBB_EXPORTED_METHOD ~concurrent_vector_base_v3();

        static segment_index_t segment_index_of( size_type index ) {
            return segment_index_t( __TBB_Log2( index|1 ) );
        }

        static segment_index_t segment_base( segment_index_t k ) {
            return (segment_index_t(1)<<k & ~segment_index_t(1));
        }

        static inline segment_index_t segment_base_index_of( segment_index_t &index ) {
            segment_index_t k = segment_index_of( index );
            index -= segment_base(k);
            return k;
        }

        static size_type segment_size( segment_index_t k ) {
            return segment_index_t(1)<<k; // fake value for k==0
        }

        typedef void (__TBB_EXPORTED_FUNC *internal_array_op1)(void* begin, size_type n );

        typedef void (__TBB_EXPORTED_FUNC *internal_array_op2)(void* dst, const void* src, size_type n );

        struct internal_segments_table {
            segment_index_t first_block;
            void* table[pointers_per_long_table];
        };

        void __TBB_EXPORTED_METHOD internal_reserve( size_type n, size_type element_size, size_type max_size );
        size_type __TBB_EXPORTED_METHOD internal_capacity() const;
        void internal_grow( size_type start, size_type finish, size_type element_size, internal_array_op2 init, const void *src );
        size_type __TBB_EXPORTED_METHOD internal_grow_by( size_type delta, size_type element_size, internal_array_op2 init, const void *src );
        void* __TBB_EXPORTED_METHOD internal_push_back( size_type element_size, size_type& index );
        segment_index_t __TBB_EXPORTED_METHOD internal_clear( internal_array_op1 destroy );
        void* __TBB_EXPORTED_METHOD internal_compact( size_type element_size, void *table, internal_array_op1 destroy, internal_array_op2 copy );
        void __TBB_EXPORTED_METHOD internal_copy( const concurrent_vector_base_v3& src, size_type element_size, internal_array_op2 copy );
        void __TBB_EXPORTED_METHOD internal_assign( const concurrent_vector_base_v3& src, size_type element_size,
                              internal_array_op1 destroy, internal_array_op2 assign, internal_array_op2 copy );
        void __TBB_EXPORTED_METHOD internal_throw_exception(size_type) const;
        void __TBB_EXPORTED_METHOD internal_swap(concurrent_vector_base_v3& v);

        void __TBB_EXPORTED_METHOD internal_resize( size_type n, size_type element_size, size_type max_size, const void *src,
                                                    internal_array_op1 destroy, internal_array_op2 init );
        size_type __TBB_EXPORTED_METHOD internal_grow_to_at_least_with_result( size_type new_size, size_type element_size, internal_array_op2 init, const void *src );

        void __TBB_EXPORTED_METHOD internal_grow_to_at_least( size_type new_size, size_type element_size, internal_array_op2 init, const void *src );
private:
        class helper;
        friend class helper;
    };
    
    typedef concurrent_vector_base_v3 concurrent_vector_base;


    template<typename Container, typename Value>
    class vector_iterator 
    {
        Container* my_vector;

        size_t my_index;


        mutable Value* my_item;

        template<typename C, typename T>
        friend vector_iterator<C,T> operator+( ptrdiff_t offset, const vector_iterator<C,T>& v );

        template<typename C, typename T, typename U>
        friend bool operator==( const vector_iterator<C,T>& i, const vector_iterator<C,U>& j );

        template<typename C, typename T, typename U>
        friend bool operator<( const vector_iterator<C,T>& i, const vector_iterator<C,U>& j );

        template<typename C, typename T, typename U>
        friend ptrdiff_t operator-( const vector_iterator<C,T>& i, const vector_iterator<C,U>& j );
    
        template<typename C, typename U>
        friend class internal::vector_iterator;

#if !defined(_MSC_VER) || defined(__INTEL_COMPILER)
        template<typename T, class A>
        friend class tbb::concurrent_vector;
#else
public: // workaround for MSVC
#endif 

        vector_iterator( const Container& vector, size_t index, void *ptr = 0 ) : 
            my_vector(const_cast<Container*>(&vector)), 
            my_index(index), 
            my_item(static_cast<Value*>(ptr))
        {}

    public:
        vector_iterator() : my_vector(NULL), my_index(~size_t(0)), my_item(NULL) {}

        vector_iterator( const vector_iterator<Container,typename Container::value_type>& other ) :
            my_vector(other.my_vector),
            my_index(other.my_index),
            my_item(other.my_item)
        {}

        vector_iterator operator+( ptrdiff_t offset ) const {
            return vector_iterator( *my_vector, my_index+offset );
        }
        vector_iterator &operator+=( ptrdiff_t offset ) {
            my_index+=offset;
            my_item = NULL;
            return *this;
        }
        vector_iterator operator-( ptrdiff_t offset ) const {
            return vector_iterator( *my_vector, my_index-offset );
        }
        vector_iterator &operator-=( ptrdiff_t offset ) {
            my_index-=offset;
            my_item = NULL;
            return *this;
        }
        Value& operator*() const {
            Value* item = my_item;
            if( !item ) {
                item = my_item = &my_vector->internal_subscript(my_index);
            }
            __TBB_ASSERT( item==&my_vector->internal_subscript(my_index), "corrupt cache" );
            return *item;
        }
        Value& operator[]( ptrdiff_t k ) const {
            return my_vector->internal_subscript(my_index+k);
        }
        Value* operator->() const {return &operator*();}

        vector_iterator& operator++() {
            size_t k = ++my_index;
            if( my_item ) {
                // Following test uses 2's-complement wizardry
                if( (k& (k-2))==0 ) {
                    // k is a power of two that is at least k-2
                    my_item= NULL;
                } else {
                    ++my_item;
                }
            }
            return *this;
        }

        vector_iterator& operator--() {
            __TBB_ASSERT( my_index>0, "operator--() applied to iterator already at beginning of concurrent_vector" ); 
            size_t k = my_index--;
            if( my_item ) {
                // Following test uses 2's-complement wizardry
                if( (k& (k-2))==0 ) {
                    // k is a power of two that is at least k-2  
                    my_item= NULL;
                } else {
                    --my_item;
                }
            }
            return *this;
        }

        vector_iterator operator++(int) {
            vector_iterator result = *this;
            operator++();
            return result;
        }

        vector_iterator operator--(int) {
            vector_iterator result = *this;
            operator--();
            return result;
        }

        // STL support

        typedef ptrdiff_t difference_type;
        typedef Value value_type;
        typedef Value* pointer;
        typedef Value& reference;
        typedef std::random_access_iterator_tag iterator_category;
    };

    template<typename Container, typename T>
    vector_iterator<Container,T> operator+( ptrdiff_t offset, const vector_iterator<Container,T>& v ) {
        return vector_iterator<Container,T>( *v.my_vector, v.my_index+offset );
    }

    template<typename Container, typename T, typename U>
    bool operator==( const vector_iterator<Container,T>& i, const vector_iterator<Container,U>& j ) {
        return i.my_index==j.my_index && i.my_vector == j.my_vector;
    }

    template<typename Container, typename T, typename U>
    bool operator!=( const vector_iterator<Container,T>& i, const vector_iterator<Container,U>& j ) {
        return !(i==j);
    }

    template<typename Container, typename T, typename U>
    bool operator<( const vector_iterator<Container,T>& i, const vector_iterator<Container,U>& j ) {
        return i.my_index<j.my_index;
    }

    template<typename Container, typename T, typename U>
    bool operator>( const vector_iterator<Container,T>& i, const vector_iterator<Container,U>& j ) {
        return j<i;
    }

    template<typename Container, typename T, typename U>
    bool operator>=( const vector_iterator<Container,T>& i, const vector_iterator<Container,U>& j ) {
        return !(i<j);
    }

    template<typename Container, typename T, typename U>
    bool operator<=( const vector_iterator<Container,T>& i, const vector_iterator<Container,U>& j ) {
        return !(j<i);
    }

    template<typename Container, typename T, typename U>
    ptrdiff_t operator-( const vector_iterator<Container,T>& i, const vector_iterator<Container,U>& j ) {
        return ptrdiff_t(i.my_index)-ptrdiff_t(j.my_index);
    }

    template<typename T, class A>
    class allocator_base {
    public:
        typedef typename A::template
            rebind<T>::other allocator_type;
        allocator_type my_allocator;

        allocator_base(const allocator_type &a = allocator_type() ) : my_allocator(a) {}
    };

} // namespace internal


template<typename T, class A>
class concurrent_vector: protected internal::allocator_base<T, A>,
                         private internal::concurrent_vector_base {
private:
    template<typename I>
    class generic_range_type: public blocked_range<I> {
    public:
        typedef T value_type;
        typedef T& reference;
        typedef const T& const_reference;
        typedef I iterator;
        typedef ptrdiff_t difference_type;
        generic_range_type( I begin_, I end_, size_t grainsize = 1) : blocked_range<I>(begin_,end_,grainsize) {} 
        template<typename U>
        generic_range_type( const generic_range_type<U>& r) : blocked_range<I>(r.begin(),r.end(),r.grainsize()) {} 
        generic_range_type( generic_range_type& r, split ) : blocked_range<I>(r,split()) {}
    };

    template<typename C, typename U>
    friend class internal::vector_iterator;
public:
    //------------------------------------------------------------------------
    // STL compatible types
    //------------------------------------------------------------------------
    typedef internal::concurrent_vector_base_v3::size_type size_type;
    typedef typename internal::allocator_base<T, A>::allocator_type allocator_type;

    typedef T value_type;
    typedef ptrdiff_t difference_type;
    typedef T& reference;
    typedef const T& const_reference;
    typedef T *pointer;
    typedef const T *const_pointer;

    typedef internal::vector_iterator<concurrent_vector,T> iterator;
    typedef internal::vector_iterator<concurrent_vector,const T> const_iterator;

#if !defined(_MSC_VER) || _CPPLIB_VER>=300 
    // Assume ISO standard definition of std::reverse_iterator
    typedef std::reverse_iterator<iterator> reverse_iterator;
    typedef std::reverse_iterator<const_iterator> const_reverse_iterator;
#else
    // Use non-standard std::reverse_iterator
    typedef std::reverse_iterator<iterator,T,T&,T*> reverse_iterator;
    typedef std::reverse_iterator<const_iterator,T,const T&,const T*> const_reverse_iterator;
#endif /* defined(_MSC_VER) && (_MSC_VER<1300) */

    //------------------------------------------------------------------------
    // Parallel algorithm support
    //------------------------------------------------------------------------
    typedef generic_range_type<iterator> range_type;
    typedef generic_range_type<const_iterator> const_range_type;

    //------------------------------------------------------------------------
    // STL compatible constructors & destructors
    //------------------------------------------------------------------------

    explicit concurrent_vector(const allocator_type &a = allocator_type())
        : internal::allocator_base<T, A>(a)
    {
        vector_allocator_ptr = &internal_allocator;
    }

    concurrent_vector( const concurrent_vector& vector, const allocator_type& a = allocator_type() )
        : internal::allocator_base<T, A>(a)
    {
        vector_allocator_ptr = &internal_allocator;
        try {
            internal_copy(vector, sizeof(T), &copy_array);
        } catch(...) {
            segment_t *table = my_segment;
            internal_free_segments( reinterpret_cast<void**>(table), internal_clear(&destroy_array), my_first_block );
            throw;
        }
    }

    template<class M>
    concurrent_vector( const concurrent_vector<T, M>& vector, const allocator_type& a = allocator_type() )
        : internal::allocator_base<T, A>(a)
    {
        vector_allocator_ptr = &internal_allocator;
        try {
            internal_copy(vector.internal_vector_base(), sizeof(T), &copy_array);
        } catch(...) {
            segment_t *table = my_segment;
            internal_free_segments( reinterpret_cast<void**>(table), internal_clear(&destroy_array), my_first_block );
            throw;
        }
    }

    explicit concurrent_vector(size_type n)
    {
        vector_allocator_ptr = &internal_allocator;
        try {
            internal_resize( n, sizeof(T), max_size(), NULL, &destroy_array, &initialize_array );
        } catch(...) {
            segment_t *table = my_segment;
            internal_free_segments( reinterpret_cast<void**>(table), internal_clear(&destroy_array), my_first_block );
            throw;
        }
    }

    concurrent_vector(size_type n, const_reference t, const allocator_type& a = allocator_type())
        : internal::allocator_base<T, A>(a)
    {
        vector_allocator_ptr = &internal_allocator;
        try {
            internal_resize( n, sizeof(T), max_size(), static_cast<const void*>(&t), &destroy_array, &initialize_array_by );
        } catch(...) {
            segment_t *table = my_segment;
            internal_free_segments( reinterpret_cast<void**>(table), internal_clear(&destroy_array), my_first_block );
            throw;
        }
    }

    template<class I>
    concurrent_vector(I first, I last, const allocator_type &a = allocator_type())
        : internal::allocator_base<T, A>(a)
    {
        vector_allocator_ptr = &internal_allocator;
        try {
            internal_assign_range(first, last, static_cast<is_integer_tag<std::numeric_limits<I>::is_integer> *>(0) );
        } catch(...) {
            segment_t *table = my_segment;
            internal_free_segments( reinterpret_cast<void**>(table), internal_clear(&destroy_array), my_first_block );
            throw;
        }
    }

    concurrent_vector& operator=( const concurrent_vector& vector ) {
        if( this != &vector )
            internal_assign(vector, sizeof(T), &destroy_array, &assign_array, &copy_array);
        return *this;
    }

    template<class M>
    concurrent_vector& operator=( const concurrent_vector<T, M>& vector ) {
        if( static_cast<void*>( this ) != static_cast<const void*>( &vector ) )
            internal_assign(vector.internal_vector_base(),
                sizeof(T), &destroy_array, &assign_array, &copy_array);
        return *this;
    }

    //------------------------------------------------------------------------
    // Concurrent operations
    //------------------------------------------------------------------------
#if TBB_DEPRECATED

    size_type grow_by( size_type delta ) {
        return delta ? internal_grow_by( delta, sizeof(T), &initialize_array, NULL ) : my_early_size;
    }
#else

    iterator grow_by( size_type delta ) {
        return iterator(*this, delta ? internal_grow_by( delta, sizeof(T), &initialize_array, NULL ) : my_early_size);
    }
#endif

#if TBB_DEPRECATED

    size_type grow_by( size_type delta, const_reference t ) {
        return delta ? internal_grow_by( delta, sizeof(T), &initialize_array_by, static_cast<const void*>(&t) ) : my_early_size;
    }
#else

    iterator grow_by( size_type delta, const_reference t ) {
        return iterator(*this, delta ? internal_grow_by( delta, sizeof(T), &initialize_array_by, static_cast<const void*>(&t) ) : my_early_size);
    }
#endif

#if TBB_DEPRECATED

    void grow_to_at_least( size_type n ) {
        if( n ) internal_grow_to_at_least_with_result( n, sizeof(T), &initialize_array, NULL );
    };
#else

    iterator grow_to_at_least( size_type n ) {
        size_type m=0;
        if( n ) {
            m = internal_grow_to_at_least_with_result( n, sizeof(T), &initialize_array, NULL );
            if( m>n ) m=n;
        }
        return iterator(*this, m);
    };
#endif

#if TBB_DEPRECATED
    size_type push_back( const_reference item )
#else

    iterator push_back( const_reference item )
#endif
    {
        size_type k;
        void *ptr = internal_push_back(sizeof(T),k);
        internal_loop_guide loop(1, ptr);
        loop.init(&item);
#if TBB_DEPRECATED
        return k;
#else
        return iterator(*this, k, ptr);
#endif
    }


    reference operator[]( size_type index ) {
        return internal_subscript(index);
    }

    const_reference operator[]( size_type index ) const {
        return internal_subscript(index);
    }

    reference at( size_type index ) {
        return internal_subscript_with_exceptions(index);
    }

    const_reference at( size_type index ) const {
        return internal_subscript_with_exceptions(index);
    }

    range_type range( size_t grainsize = 1) {
        return range_type( begin(), end(), grainsize );
    }

    const_range_type range( size_t grainsize = 1 ) const {
        return const_range_type( begin(), end(), grainsize );
    }
    //------------------------------------------------------------------------
    // Capacity
    //------------------------------------------------------------------------
    size_type size() const {
        size_type sz = my_early_size, cp = internal_capacity();
        return cp < sz ? cp : sz;
    }

    bool empty() const {return !my_early_size;}

    size_type capacity() const {return internal_capacity();}


    void reserve( size_type n ) {
        if( n )
            internal_reserve(n, sizeof(T), max_size());
    }

    void resize( size_type n ) {
        internal_resize( n, sizeof(T), max_size(), NULL, &destroy_array, &initialize_array );
    }
    
    void resize( size_type n, const_reference t ) {
        internal_resize( n, sizeof(T), max_size(), static_cast<const void*>(&t), &destroy_array, &initialize_array_by );
    }
   
#if TBB_DEPRECATED 
    void compact() {shrink_to_fit();}
#endif /* TBB_DEPRECATED */

    void shrink_to_fit();

    size_type max_size() const {return (~size_type(0))/sizeof(T);}

    //------------------------------------------------------------------------
    // STL support
    //------------------------------------------------------------------------

    iterator begin() {return iterator(*this,0);}
    iterator end() {return iterator(*this,size());}
    const_iterator begin() const {return const_iterator(*this,0);}
    const_iterator end() const {return const_iterator(*this,size());}
    const_iterator cbegin() const {return const_iterator(*this,0);}
    const_iterator cend() const {return const_iterator(*this,size());}
    reverse_iterator rbegin() {return reverse_iterator(end());}
    reverse_iterator rend() {return reverse_iterator(begin());}
    const_reverse_iterator rbegin() const {return const_reverse_iterator(end());}
    const_reverse_iterator rend() const {return const_reverse_iterator(begin());}
    const_reverse_iterator crbegin() const {return const_reverse_iterator(end());}
    const_reverse_iterator crend() const {return const_reverse_iterator(begin());}
    reference front() {
        __TBB_ASSERT( size()>0, NULL);
        return static_cast<T*>(my_segment[0].array)[0];
    }
    const_reference front() const {
        __TBB_ASSERT( size()>0, NULL);
        return static_cast<const T*>(my_segment[0].array)[0];
    }
    reference back() {
        __TBB_ASSERT( size()>0, NULL);
        return internal_subscript( size()-1 );
    }
    const_reference back() const {
        __TBB_ASSERT( size()>0, NULL);
        return internal_subscript( size()-1 );
    }
    allocator_type get_allocator() const { return this->my_allocator; }

    void assign(size_type n, const_reference t) {
        clear();
        internal_resize( n, sizeof(T), max_size(), static_cast<const void*>(&t), &destroy_array, &initialize_array_by );
    }

    template<class I>
    void assign(I first, I last) {
        clear(); internal_assign_range( first, last, static_cast<is_integer_tag<std::numeric_limits<I>::is_integer> *>(0) );
    }

    void swap(concurrent_vector &vector) {
        if( this != &vector ) {
            concurrent_vector_base_v3::internal_swap(static_cast<concurrent_vector_base_v3&>(vector));
            std::swap(this->my_allocator, vector.my_allocator);
        }
    }


    void clear() {
        internal_clear(&destroy_array);
    }

    ~concurrent_vector() {
        segment_t *table = my_segment;
        internal_free_segments( reinterpret_cast<void**>(table), internal_clear(&destroy_array), my_first_block );
        // base class destructor call should be then
    }

    const internal::concurrent_vector_base_v3 &internal_vector_base() const { return *this; }
private:
    static void *internal_allocator(internal::concurrent_vector_base_v3 &vb, size_t k) {
        return static_cast<concurrent_vector<T, A>&>(vb).my_allocator.allocate(k);
    }
    void internal_free_segments(void *table[], segment_index_t k, segment_index_t first_block);

    T& internal_subscript( size_type index ) const;

    T& internal_subscript_with_exceptions( size_type index ) const;

    void internal_assign_n(size_type n, const_pointer p) {
        internal_resize( n, sizeof(T), max_size(), static_cast<const void*>(p), &destroy_array, p? &initialize_array_by : &initialize_array );
    }

    template<bool B> class is_integer_tag;

    template<class I>
    void internal_assign_range(I first, I last, is_integer_tag<true> *) {
        internal_assign_n(static_cast<size_type>(first), &static_cast<T&>(last));
    }
    template<class I>
    void internal_assign_range(I first, I last, is_integer_tag<false> *) {
        internal_assign_iterators(first, last);
    }
    template<class I>
    void internal_assign_iterators(I first, I last);

    static void __TBB_EXPORTED_FUNC initialize_array( void* begin, const void*, size_type n );

    static void __TBB_EXPORTED_FUNC initialize_array_by( void* begin, const void* src, size_type n );

    static void __TBB_EXPORTED_FUNC copy_array( void* dst, const void* src, size_type n );

    static void __TBB_EXPORTED_FUNC assign_array( void* dst, const void* src, size_type n );

    static void __TBB_EXPORTED_FUNC destroy_array( void* begin, size_type n );

    class internal_loop_guide : internal::no_copy {
    public:
        const pointer array;
        const size_type n;
        size_type i;
        internal_loop_guide(size_type ntrials, void *ptr)
            : array(static_cast<pointer>(ptr)), n(ntrials), i(0) {}
        void init() {   for(; i < n; ++i) new( &array[i] ) T(); }
        void init(const void *src) { for(; i < n; ++i) new( &array[i] ) T(*static_cast<const T*>(src)); }
        void copy(const void *src) { for(; i < n; ++i) new( &array[i] ) T(static_cast<const T*>(src)[i]); }
        void assign(const void *src) { for(; i < n; ++i) array[i] = static_cast<const T*>(src)[i]; }
        template<class I> void iterate(I &src) { for(; i < n; ++i, ++src) new( &array[i] ) T( *src ); }
        ~internal_loop_guide() {
            if(i < n) // if exception raised, do zerroing on the rest of items
                std::memset(array+i, 0, (n-i)*sizeof(value_type));
        }
    };
};

template<typename T, class A>
void concurrent_vector<T, A>::shrink_to_fit() {
    internal_segments_table old;
    try {
        if( internal_compact( sizeof(T), &old, &destroy_array, &copy_array ) )
            internal_free_segments( old.table, pointers_per_long_table, old.first_block ); // free joined and unnecessary segments
    } catch(...) {
        if( old.first_block ) // free segment allocated for compacting. Only for support of exceptions in ctor of user T[ype]
            internal_free_segments( old.table, 1, old.first_block );
        throw;
    }
}

template<typename T, class A>
void concurrent_vector<T, A>::internal_free_segments(void *table[], segment_index_t k, segment_index_t first_block) {
    // Free the arrays
    while( k > first_block ) {
        --k;
        T* array = static_cast<T*>(table[k]);
        table[k] = NULL;
        if( array > internal::vector_allocation_error_flag ) // check for correct segment pointer
            this->my_allocator.deallocate( array, segment_size(k) );
    }
    T* array = static_cast<T*>(table[0]);
    if( array > internal::vector_allocation_error_flag ) {
        __TBB_ASSERT( first_block > 0, NULL );
        while(k > 0) table[--k] = NULL;
        this->my_allocator.deallocate( array, segment_size(first_block) );
    }
}

template<typename T, class A>
T& concurrent_vector<T, A>::internal_subscript( size_type index ) const {
    __TBB_ASSERT( index < my_early_size, "index out of bounds" );
    size_type j = index;
    segment_index_t k = segment_base_index_of( j );
    __TBB_ASSERT( my_segment != (segment_t*)my_storage || k < pointers_per_short_table, "index is being allocated" );
    // no need in __TBB_load_with_acquire since thread works in own space or gets 
#if TBB_USE_THREADING_TOOLS
    T* array = static_cast<T*>( tbb::internal::itt_load_pointer_v3(&my_segment[k].array));
#else
    T* array = static_cast<T*>(my_segment[k].array);
#endif /* TBB_USE_THREADING_TOOLS */
    __TBB_ASSERT( array != internal::vector_allocation_error_flag, "the instance is broken by bad allocation. Use at() instead" );
    __TBB_ASSERT( array, "index is being allocated" );
    return array[j];
}

template<typename T, class A>
T& concurrent_vector<T, A>::internal_subscript_with_exceptions( size_type index ) const {
    if( index >= my_early_size )
        internal_throw_exception(0); // throw std::out_of_range
    size_type j = index;
    segment_index_t k = segment_base_index_of( j );
    if( my_segment == (segment_t*)my_storage && k >= pointers_per_short_table )
        internal_throw_exception(1); // throw std::range_error
    void *array = my_segment[k].array; // no need in __TBB_load_with_acquire
    if( array <= internal::vector_allocation_error_flag ) // check for correct segment pointer
        internal_throw_exception(2); // throw std::range_error
    return static_cast<T*>(array)[j];
}

template<typename T, class A> template<class I>
void concurrent_vector<T, A>::internal_assign_iterators(I first, I last) {
    __TBB_ASSERT(my_early_size == 0, NULL);
    size_type n = std::distance(first, last);
    if( !n ) return;
    internal_reserve(n, sizeof(T), max_size());
    my_early_size = n;
    segment_index_t k = 0;
    size_type sz = segment_size( my_first_block );
    while( sz < n ) {
        internal_loop_guide loop(sz, my_segment[k].array);
        loop.iterate(first);
        n -= sz;
        if( !k ) k = my_first_block;
        else { ++k; sz <<= 1; }
    }
    internal_loop_guide loop(n, my_segment[k].array);
    loop.iterate(first);
}

template<typename T, class A>
void concurrent_vector<T, A>::initialize_array( void* begin, const void *, size_type n ) {
    internal_loop_guide loop(n, begin); loop.init();
}

template<typename T, class A>
void concurrent_vector<T, A>::initialize_array_by( void* begin, const void *src, size_type n ) {
    internal_loop_guide loop(n, begin); loop.init(src);
}

template<typename T, class A>
void concurrent_vector<T, A>::copy_array( void* dst, const void* src, size_type n ) {
    internal_loop_guide loop(n, dst); loop.copy(src);
}

template<typename T, class A>
void concurrent_vector<T, A>::assign_array( void* dst, const void* src, size_type n ) {
    internal_loop_guide loop(n, dst); loop.assign(src);
}

#if defined(_MSC_VER) && !defined(__INTEL_COMPILER) 
    // Workaround for overzealous compiler warning
    #pragma warning (push)
    #pragma warning (disable: 4189)
#endif
template<typename T, class A>
void concurrent_vector<T, A>::destroy_array( void* begin, size_type n ) {
    T* array = static_cast<T*>(begin);
    for( size_type j=n; j>0; --j )
        array[j-1].~T(); // destructors are supposed to not throw any exceptions
}
#if defined(_MSC_VER) && !defined(__INTEL_COMPILER) 
    #pragma warning (pop)
#endif // warning 4189 is back 

// concurrent_vector's template functions
template<typename T, class A1, class A2>
inline bool operator==(const concurrent_vector<T, A1> &a, const concurrent_vector<T, A2> &b) {
    // Simply:    return a.size() == b.size() && std::equal(a.begin(), a.end(), b.begin());
    if(a.size() != b.size()) return false;
    typename concurrent_vector<T, A1>::const_iterator i(a.begin());
    typename concurrent_vector<T, A2>::const_iterator j(b.begin());
    for(; i != a.end(); ++i, ++j)
        if( !(*i == *j) ) return false;
    return true;
}

template<typename T, class A1, class A2>
inline bool operator!=(const concurrent_vector<T, A1> &a, const concurrent_vector<T, A2> &b)
{    return !(a == b); }

template<typename T, class A1, class A2>
inline bool operator<(const concurrent_vector<T, A1> &a, const concurrent_vector<T, A2> &b)
{    return (std::lexicographical_compare(a.begin(), a.end(), b.begin(), b.end())); }

template<typename T, class A1, class A2>
inline bool operator>(const concurrent_vector<T, A1> &a, const concurrent_vector<T, A2> &b)
{    return b < a; }

template<typename T, class A1, class A2>
inline bool operator<=(const concurrent_vector<T, A1> &a, const concurrent_vector<T, A2> &b)
{    return !(b < a); }

template<typename T, class A1, class A2>
inline bool operator>=(const concurrent_vector<T, A1> &a, const concurrent_vector<T, A2> &b)
{    return !(a < b); }

template<typename T, class A>
inline void swap(concurrent_vector<T, A> &a, concurrent_vector<T, A> &b)
{    a.swap( b ); }

} // namespace tbb

#if defined(_MSC_VER) && !defined(__INTEL_COMPILER) && defined(_Wp64)
    #pragma warning (pop)
#endif // warning 4267 is back

#endif /* __TBB_concurrent_vector_H */

---

Copyright © 2005-2009 Intel Corporation. All Rights Reserved.

Intel, Pentium, Intel Xeon, Itanium, Intel XScale and VTune are registered trademarks or trademarks of Intel Corporation or its subsidiaries in the United States and other countries.

* Other names and brands may be claimed as the property of others.