boost/container/detail/adaptive_node_pool_impl.hpp
//////////////////////////////////////////////////////////////////////////////
//
// (C) Copyright Ion Gaztanaga 2005-2013. Distributed under the Boost
// Software License, Version 1.0. (See accompanying file
// LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
//
// See http://www.boost.org/libs/container for documentation.
//
//////////////////////////////////////////////////////////////////////////////
#ifndef BOOST_CONTAINER_DETAIL_ADAPTIVE_NODE_POOL_IMPL_HPP
#define BOOST_CONTAINER_DETAIL_ADAPTIVE_NODE_POOL_IMPL_HPP
#ifndef BOOST_CONFIG_HPP
# include <boost/config.hpp>
#endif
#if defined(BOOST_HAS_PRAGMA_ONCE)
# pragma once
#endif
#include <boost/container/detail/config_begin.hpp>
#include <boost/container/detail/workaround.hpp>
// container
#include <boost/container/container_fwd.hpp>
#include <boost/container/throw_exception.hpp>
// container/detail
#include <boost/container/detail/pool_common.hpp>
#include <boost/container/detail/iterator.hpp>
#include <boost/move/detail/iterator_to_raw_pointer.hpp>
#include <boost/container/detail/math_functions.hpp>
#include <boost/container/detail/placement_new.hpp>
#include <boost/container/detail/mpl.hpp>
#include <boost/move/detail/to_raw_pointer.hpp>
#include <boost/move/detail/force_ptr.hpp>
#include <boost/container/detail/type_traits.hpp>
// intrusive
#include <boost/intrusive/pointer_traits.hpp>
#include <boost/intrusive/set.hpp>
#include <boost/intrusive/list.hpp>
#include <boost/intrusive/slist.hpp>
// other
#include <boost/assert.hpp>
#include <cstddef>
namespace boost {
namespace container {
namespace adaptive_pool_flag {
static const unsigned int none = 0u;
static const unsigned int align_only = 1u << 0u;
static const unsigned int size_ordered = 1u << 1u;
static const unsigned int address_ordered = 1u << 2u;
} //namespace adaptive_pool_flag{
namespace dtl {
template<class size_type>
struct hdr_offset_holder_t
{
hdr_offset_holder_t(size_type offset = 0)
: hdr_offset(offset)
{}
size_type hdr_offset;
};
template<class SizeType, unsigned int Flags>
struct less_func;
template<class SizeType>
struct less_func<SizeType, adaptive_pool_flag::none>
{
static bool less(SizeType, SizeType, const void *, const void *)
{ return true; }
};
template<class SizeType>
struct less_func<SizeType, adaptive_pool_flag::size_ordered>
{
static bool less(SizeType ls, SizeType rs, const void *, const void *)
{ return ls < rs; }
};
template<class SizeType>
struct less_func<SizeType, adaptive_pool_flag::address_ordered>
{
static bool less(SizeType, SizeType, const void *la, const void *ra)
{ return la < ra; }
};
template<class SizeType>
struct less_func<SizeType, adaptive_pool_flag::size_ordered | adaptive_pool_flag::address_ordered>
{
static bool less(SizeType ls, SizeType rs, const void *la, const void *ra)
{ return (ls < rs) || ((ls == rs) && (la < ra)); }
};
template<class VoidPointer, class SizeType, unsigned OrderFlags>
struct block_container_traits
{
typedef typename bi::make_set_base_hook
< bi::void_pointer<VoidPointer>
, bi::optimize_size<true>
, bi::link_mode<bi::normal_link> >::type hook_t;
template<class T>
struct container
{
typedef typename bi::make_multiset
<T, bi::base_hook<hook_t>, bi::size_type<SizeType> >::type type;
};
template<class Container>
static void reinsert_was_used(Container &container, typename Container::reference v, bool)
{
typedef typename Container::const_iterator const_block_iterator;
typedef typename Container::iterator block_iterator;
typedef typename Container::value_compare value_compare;
const block_iterator this_block(Container::s_iterator_to(v));
const const_block_iterator cendit(container.cend());
block_iterator next_block(this_block);
if(++next_block != cendit && value_compare()(*next_block, v)){
const_block_iterator next2_block(next_block);
//Test if v should be swapped with next (optimization)
if(++next2_block == cendit || !value_compare()(*next2_block, v)){
v.swap_nodes(*next_block);
BOOST_ASSERT(++next_block == this_block);
}
else{
container.erase(this_block);
container.insert(v);
}
}
}
template<class Container>
static void insert_was_empty(Container &container, typename Container::value_type &v, bool)
{
container.insert(v);
}
template<class Container>
static void erase_first(Container &container)
{
container.erase(container.cbegin());
}
template<class Container>
static void erase_last(Container &container)
{
container.erase(--container.cend());
}
};
template<class VoidPointer, class SizeType>
struct block_container_traits<VoidPointer, SizeType, 0u>
{
typedef typename bi::make_list_base_hook
< bi::void_pointer<VoidPointer>
, bi::link_mode<bi::normal_link> >::type hook_t;
template<class T>
struct container
{
typedef typename bi::make_list
<T, bi::base_hook<hook_t>, bi::size_type<SizeType>, bi::constant_time_size<false> >::type type;
};
template<class Container>
static void reinsert_was_used(Container &container, typename Container::value_type &v, bool is_full)
{
if(is_full){
container.erase(Container::s_iterator_to(v));
container.push_back(v);
}
}
template<class Container>
static void insert_was_empty(Container &container, typename Container::value_type &v, bool is_full)
{
if(is_full){
container.push_back(v);
}
else{
container.push_front(v);
}
}
template<class Container>
static void erase_first(Container &container)
{
container.pop_front();
}
template<class Container>
static void erase_last(Container &container)
{
container.pop_back();
}
};
/////////////////////////////
//
// adaptive_pool_types
//
/////////////////////////////
template<class MultiallocationChain, class VoidPointer, class SizeType, unsigned int Flags>
struct adaptive_pool_types
{
typedef VoidPointer void_pointer;
static const unsigned ordered = (Flags & (adaptive_pool_flag::size_ordered | adaptive_pool_flag::address_ordered));
typedef block_container_traits<VoidPointer, SizeType, ordered> block_container_traits_t;
typedef typename block_container_traits_t::hook_t hook_t;
typedef hdr_offset_holder_t<SizeType> hdr_offset_holder;
static const unsigned int order_flags = Flags & (adaptive_pool_flag::size_ordered | adaptive_pool_flag::address_ordered);
typedef MultiallocationChain free_nodes_t;
struct block_info_t
: public hdr_offset_holder,
public hook_t
{
//An intrusive list of free node from this block
free_nodes_t free_nodes;
friend bool operator <(const block_info_t &l, const block_info_t &r)
{
return less_func<SizeType, order_flags>::
less(l.free_nodes.size(), r.free_nodes.size(), &l , &r);
}
friend bool operator ==(const block_info_t &l, const block_info_t &r)
{ return &l == &r; }
};
typedef typename block_container_traits_t:: template container<block_info_t>::type block_container_t;
};
/////////////////////////////////////////////
//
// candidate_power_of_2_ct
//
/////////////////////////////////////////////
template< std::size_t alignment
, std::size_t real_node_size
, std::size_t payload_per_allocation
, std::size_t min_elements_per_block
, std::size_t hdr_size
, std::size_t hdr_offset_size
, std::size_t overhead_percent>
struct candidate_power_of_2_ct_helper
{
static const std::size_t hdr_subblock_elements_alone = (alignment - hdr_size - payload_per_allocation)/real_node_size;
static const std::size_t hdr_subblock_elements_first = (alignment - hdr_size - payload_per_allocation)/real_node_size;
static const std::size_t elements_per_b_subblock_mid = (alignment - hdr_offset_size)/real_node_size;
static const std::size_t elements_per_b_subblock_end = (alignment - hdr_offset_size - payload_per_allocation)/real_node_size;
static const std::size_t num_b_subblock =
hdr_subblock_elements_alone >= min_elements_per_block
? 0
: ( ((hdr_subblock_elements_first + elements_per_b_subblock_end) >= min_elements_per_block)
? 1
: 2 + (min_elements_per_block - hdr_subblock_elements_first - elements_per_b_subblock_end - 1)/elements_per_b_subblock_mid
)
;
static const std::size_t num_b_subblock_mid = (num_b_subblock > 1) ? (num_b_subblock - 1) : 0;
static const std::size_t total_nodes = (num_b_subblock == 0)
? hdr_subblock_elements_alone
: ( (num_b_subblock == 1)
? (hdr_subblock_elements_first + elements_per_b_subblock_end)
: (hdr_subblock_elements_first + num_b_subblock_mid*elements_per_b_subblock_mid + elements_per_b_subblock_end)
)
;
static const std::size_t total_data = total_nodes*real_node_size;
static const std::size_t total_size = alignment*(num_b_subblock+1);
static const bool overhead_satisfied = (total_size - total_data)*100/total_size < overhead_percent;
};
template< std::size_t initial_alignment
, std::size_t real_node_size
, std::size_t payload_per_allocation
, std::size_t min_elements_per_block
, std::size_t hdr_size
, std::size_t hdr_offset_size
, std::size_t overhead_percent
, bool Loop = true>
struct candidate_power_of_2_ct
{
typedef candidate_power_of_2_ct_helper
< initial_alignment
, real_node_size
, payload_per_allocation
, min_elements_per_block
, hdr_size
, hdr_offset_size
, overhead_percent> helper_t;
static const std::size_t candidate_power_of_2 = initial_alignment << std::size_t(!helper_t::overhead_satisfied);
typedef typename candidate_power_of_2_ct
< candidate_power_of_2
, real_node_size
, payload_per_allocation
, min_elements_per_block
, hdr_size
, hdr_offset_size
, overhead_percent
, !helper_t::overhead_satisfied
>::type type;
static const std::size_t alignment = type::alignment;
static const std::size_t num_subblocks = type::num_subblocks;
static const std::size_t real_num_node = type::real_num_node;
};
template< std::size_t initial_alignment
, std::size_t real_node_size
, std::size_t payload_per_allocation
, std::size_t min_elements_per_block
, std::size_t hdr_size
, std::size_t hdr_offset_size
, std::size_t overhead_percent
>
struct candidate_power_of_2_ct
< initial_alignment
, real_node_size
, payload_per_allocation
, min_elements_per_block
, hdr_size
, hdr_offset_size
, overhead_percent
, false>
{
typedef candidate_power_of_2_ct
< initial_alignment
, real_node_size
, payload_per_allocation
, min_elements_per_block
, hdr_size
, hdr_offset_size
, overhead_percent
, false> type;
typedef candidate_power_of_2_ct_helper
< initial_alignment
, real_node_size
, payload_per_allocation
, min_elements_per_block
, hdr_size
, hdr_offset_size
, overhead_percent> helper_t;
static const std::size_t alignment = initial_alignment;
static const std::size_t num_subblocks = helper_t::num_b_subblock+1;
static const std::size_t real_num_node = helper_t::total_nodes;
};
/////////////////////////////////////////////
//
// candidate_power_of_2_rt
//
/////////////////////////////////////////////
inline void candidate_power_of_2_rt ( std::size_t initial_alignment
, std::size_t real_node_size
, std::size_t payload_per_allocation
, std::size_t min_elements_per_block
, std::size_t hdr_size
, std::size_t hdr_offset_size
, std::size_t overhead_percent
, std::size_t &alignment
, std::size_t &num_subblocks
, std::size_t &real_num_node)
{
bool overhead_satisfied = false;
std::size_t num_b_subblock = 0;
std::size_t total_nodes = 0;
while(!overhead_satisfied)
{
std::size_t hdr_subblock_elements_alone = (initial_alignment - hdr_size - payload_per_allocation)/real_node_size;
std::size_t hdr_subblock_elements_first = (initial_alignment - hdr_size - payload_per_allocation)/real_node_size;
std::size_t elements_per_b_subblock_mid = (initial_alignment - hdr_offset_size)/real_node_size;
std::size_t elements_per_b_subblock_end = (initial_alignment - hdr_offset_size - payload_per_allocation)/real_node_size;
num_b_subblock =
hdr_subblock_elements_alone >= min_elements_per_block
? 0
: ( ((hdr_subblock_elements_first + elements_per_b_subblock_end) >= min_elements_per_block)
? 1
: 2 + (min_elements_per_block - hdr_subblock_elements_first - elements_per_b_subblock_end - 1)/elements_per_b_subblock_mid
)
;
std::size_t num_b_subblock_mid = (num_b_subblock > 1) ? (num_b_subblock - 1) : 0;
total_nodes = (num_b_subblock == 0)
? hdr_subblock_elements_alone
: ( (num_b_subblock == 1)
? (hdr_subblock_elements_first + elements_per_b_subblock_end)
: (hdr_subblock_elements_first + num_b_subblock_mid*elements_per_b_subblock_mid + elements_per_b_subblock_end)
)
;
std::size_t total_data = total_nodes*real_node_size;
std::size_t total_size = initial_alignment*(num_b_subblock+1);
overhead_satisfied = (total_size - total_data)*100/total_size < overhead_percent;
initial_alignment = initial_alignment << std::size_t(!overhead_satisfied);
}
alignment = initial_alignment;
num_subblocks = num_b_subblock+1;
real_num_node = total_nodes;
}
/////////////////////////////////////////////
//
// private_adaptive_node_pool_impl_common
//
/////////////////////////////////////////////
template< class SegmentManagerBase, unsigned int Flags>
class private_adaptive_node_pool_impl_common
{
public:
//!Segment manager typedef
typedef SegmentManagerBase segment_manager_base_type;
typedef typename SegmentManagerBase::multiallocation_chain multiallocation_chain;
typedef typename SegmentManagerBase::size_type size_type;
//Flags
//align_only
static const bool AlignOnly = (Flags & adaptive_pool_flag::align_only) != 0;
typedef bool_<AlignOnly> IsAlignOnly;
typedef true_ AlignOnlyTrue;
typedef false_ AlignOnlyFalse;
typedef typename SegmentManagerBase::void_pointer void_pointer;
static const typename SegmentManagerBase::
size_type PayloadPerAllocation = SegmentManagerBase::PayloadPerAllocation;
typedef typename boost::intrusive::pointer_traits
<void_pointer>::template rebind_pointer<segment_manager_base_type>::type segment_mngr_base_ptr_t;
protected:
typedef adaptive_pool_types
<multiallocation_chain, void_pointer, size_type, Flags> adaptive_pool_types_t;
typedef typename adaptive_pool_types_t::free_nodes_t free_nodes_t;
typedef typename adaptive_pool_types_t::block_info_t block_info_t;
typedef typename adaptive_pool_types_t::block_container_t block_container_t;
typedef typename adaptive_pool_types_t::block_container_traits_t block_container_traits_t;
typedef typename block_container_t::iterator block_iterator;
typedef typename block_container_t::const_iterator const_block_iterator;
typedef typename adaptive_pool_types_t::hdr_offset_holder hdr_offset_holder;
typedef private_adaptive_node_pool_impl_common this_type;
static const size_type MaxAlign = alignment_of<void_pointer>::value;
static const size_type HdrSize = ((sizeof(block_info_t)-1)/MaxAlign+1)*MaxAlign;
static const size_type HdrOffsetSize = ((sizeof(hdr_offset_holder)-1)/MaxAlign+1)*MaxAlign;
segment_mngr_base_ptr_t mp_segment_mngr_base; //Segment manager
block_container_t m_block_container; //Intrusive block list
size_type m_totally_free_blocks; //Free blocks
class block_destroyer;
friend class block_destroyer;
class block_destroyer
{
public:
block_destroyer(const this_type *impl, multiallocation_chain &chain, const size_type num_subblocks, const size_type real_block_alignment, const size_type real_num_node)
: mp_impl(impl), m_chain(chain), m_num_subblocks(num_subblocks), m_real_block_alignment(real_block_alignment), m_real_num_node(real_num_node)
{}
void operator()(typename block_container_t::pointer to_deallocate)
{ return this->do_destroy(to_deallocate, IsAlignOnly()); }
private:
void do_destroy(typename block_container_t::pointer to_deallocate, AlignOnlyTrue)
{
BOOST_ASSERT(to_deallocate->free_nodes.size() == m_real_num_node);
m_chain.push_back(to_deallocate);
}
void do_destroy(typename block_container_t::pointer to_deallocate, AlignOnlyFalse)
{
BOOST_ASSERT(to_deallocate->free_nodes.size() == m_real_num_node);
BOOST_ASSERT(0 == to_deallocate->hdr_offset);
hdr_offset_holder *hdr_off_holder =
mp_impl->priv_first_subblock_from_block(boost::movelib::to_raw_pointer(to_deallocate), m_num_subblocks, m_real_block_alignment);
m_chain.push_back(hdr_off_holder);
}
const this_type *mp_impl;
multiallocation_chain &m_chain;
const size_type m_num_subblocks;
const size_type m_real_block_alignment;
const size_type m_real_num_node;
};
//This macro will activate invariant checking. Slow, but helpful for debugging the code.
//#define BOOST_CONTAINER_ADAPTIVE_NODE_POOL_CHECK_INVARIANTS
void priv_invariants(const size_type real_num_node, const size_type num_subblocks, const size_type real_block_alignment) const
{
(void)real_num_node; (void)num_subblocks; (void)real_block_alignment;
#ifdef BOOST_CONTAINER_ADAPTIVE_NODE_POOL_CHECK_INVARIANTS
//Check that the total totally free blocks are correct
BOOST_ASSERT(m_block_container.size() >= m_totally_free_blocks);
const const_block_iterator itend(m_block_container.cend());
const const_block_iterator itbeg(m_block_container.cbegin());
{ //Try to do checks in a single iteration
const_block_iterator it(itbeg);
size_type total_free_nodes = 0;
size_type total_free_blocks = 0u;
for(; it != itend; ++it){
if(it != itbeg){
//Check order invariant
const_block_iterator prev(it);
--prev;
BOOST_ASSERT(!(m_block_container.key_comp()(*it, *prev)));
(void)prev; (void)it;
}
//free_nodes invariant
const size_type free_nodes = it->free_nodes.size();
BOOST_ASSERT(free_nodes <= real_num_node);
BOOST_ASSERT(free_nodes != 0);
//Acummulate total_free_nodes and total_free_blocks
total_free_nodes += free_nodes;
total_free_blocks += it->free_nodes.size() == real_num_node;
if (!AlignOnly) {
//Check that header offsets are correct
hdr_offset_holder *hdr_off_holder = this->priv_first_subblock_from_block(const_cast<block_info_t *>(&*it), num_subblocks, real_block_alignment);
for (size_type i = 0, max = num_subblocks; i < max; ++i) {
const size_type offset = size_type(reinterpret_cast<char*>(const_cast<block_info_t *>(&*it)) - reinterpret_cast<char*>(hdr_off_holder));
(void)offset;
BOOST_ASSERT(hdr_off_holder->hdr_offset == offset);
BOOST_ASSERT(0 == (reinterpret_cast<std::size_t>(hdr_off_holder) & (real_block_alignment - 1)));
BOOST_ASSERT(0 == (hdr_off_holder->hdr_offset & (real_block_alignment - 1)));
hdr_off_holder = move_detail::force_ptr<hdr_offset_holder *>(reinterpret_cast<char*>(hdr_off_holder) + real_block_alignment);
}
}
}
BOOST_ASSERT(total_free_blocks == m_totally_free_blocks);
BOOST_ASSERT(total_free_nodes >= m_totally_free_blocks*real_num_node);
}
#endif
}
void priv_deallocate_free_blocks( const size_type max_free_blocks, const size_type real_num_node
, const size_type num_subblocks, const size_type real_block_alignment)
{ //Trampoline function to ease inlining
if(m_totally_free_blocks > max_free_blocks){
this->priv_deallocate_free_blocks_impl(max_free_blocks, real_num_node, num_subblocks, real_block_alignment);
}
}
hdr_offset_holder *priv_first_subblock_from_block(block_info_t *block, const size_type num_subblocks, const size_type real_block_alignment) const
{ return this->priv_first_subblock_from_block(block, num_subblocks, real_block_alignment, IsAlignOnly()); }
hdr_offset_holder *priv_first_subblock_from_block(block_info_t *block, const size_type num_subblocks, const size_type real_block_alignment, AlignOnlyFalse) const
{
hdr_offset_holder *const hdr_off_holder = move_detail::force_ptr<hdr_offset_holder*>
(reinterpret_cast<char*>(block) - (num_subblocks-1)*real_block_alignment);
BOOST_ASSERT(hdr_off_holder->hdr_offset == size_type(reinterpret_cast<char*>(block) - reinterpret_cast<char*>(hdr_off_holder)));
BOOST_ASSERT(0 == ((std::size_t)hdr_off_holder & (real_block_alignment - 1)));
BOOST_ASSERT(0 == (hdr_off_holder->hdr_offset & (real_block_alignment - 1)));
return hdr_off_holder;
}
hdr_offset_holder *priv_first_subblock_from_block(block_info_t *block, const size_type num_subblocks, const size_type real_block_alignment, AlignOnlyTrue) const
{
(void)num_subblocks; (void)real_block_alignment;
return move_detail::force_ptr<hdr_offset_holder*>(block);
}
void priv_deallocate_free_blocks_impl( const size_type max_free_blocks, const size_type real_num_node
, const size_type num_subblocks, const size_type real_block_alignment)
{
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
//Now check if we've reached the free nodes limit
//and check if we have free blocks. If so, deallocate as much
//as we can to stay below the limit
multiallocation_chain chain;
{
if(Flags & adaptive_pool_flag::size_ordered){
const_block_iterator it = m_block_container.cend();
--it;
size_type totally_free_blocks = m_totally_free_blocks;
for( ; totally_free_blocks > max_free_blocks; --totally_free_blocks){
BOOST_ASSERT(it->free_nodes.size() == real_num_node);
void *addr = priv_first_subblock_from_block(const_cast<block_info_t*>(&*it), num_subblocks, real_block_alignment);
--it;
block_container_traits_t::erase_last(m_block_container);
chain.push_front(void_pointer(addr));
}
}
else{
const_block_iterator it = m_block_container.cend();
size_type totally_free_blocks = m_totally_free_blocks;
while(totally_free_blocks > max_free_blocks){
--it;
if(it->free_nodes.size() == real_num_node){
void *addr = priv_first_subblock_from_block(const_cast<block_info_t*>(&*it), num_subblocks, real_block_alignment);
it = m_block_container.erase(it);
chain.push_front(void_pointer(addr));
--totally_free_blocks;
}
}
}
BOOST_ASSERT((m_totally_free_blocks - max_free_blocks) == chain.size());
m_totally_free_blocks = max_free_blocks;
}
this->mp_segment_mngr_base->deallocate_many(chain);
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
}
void priv_fill_chain_remaining_to_block
( multiallocation_chain &chain, size_type target_elem_in_chain, block_info_t &c_info
, char *mem_address, size_type max_node_in_mem
, const size_type real_node_size)
{
BOOST_ASSERT(chain.size() <= target_elem_in_chain);
//First add all possible nodes to the chain
const size_type left = target_elem_in_chain - chain.size();
const size_type add_to_chain = (max_node_in_mem < left) ? max_node_in_mem : left;
char *free_mem_address = static_cast<char *>(boost::movelib::to_raw_pointer
(chain.incorporate_after(chain.last(), void_pointer(mem_address), real_node_size, add_to_chain)));
//Now store remaining nodes in the free list
if(const size_type free = max_node_in_mem - add_to_chain){
free_nodes_t & free_nodes = c_info.free_nodes;
free_nodes.incorporate_after(free_nodes.last(), void_pointer(free_mem_address), real_node_size, free);
}
}
//!Allocates a several blocks of nodes. Can throw
void priv_append_from_new_blocks( size_type min_elements, multiallocation_chain &chain
, const size_type max_free_blocks
, const size_type real_block_alignment, const size_type real_node_size
, const size_type real_num_node, const size_type num_subblocks
, AlignOnlyTrue)
{
(void)num_subblocks;
BOOST_ASSERT(m_block_container.empty());
BOOST_ASSERT(min_elements > 0);
const size_type n = (min_elements - 1)/real_num_node + 1;
const size_type real_block_size = real_block_alignment - PayloadPerAllocation;
const size_type target_elem_in_chain = chain.size() + min_elements;
for(size_type i = 0; i != n; ++i){
//We allocate a new NodeBlock and put it the last
//element of the tree
char *mem_address = static_cast<char*>
(mp_segment_mngr_base->allocate_aligned(real_block_size, real_block_alignment));
if(!mem_address){
//In case of error, free memory deallocating all nodes (the new ones allocated
//in this function plus previously stored nodes in chain).
this->priv_deallocate_nodes(chain, max_free_blocks, real_num_node, num_subblocks, real_block_alignment);
throw_bad_alloc();
}
block_info_t &c_info = *new(mem_address, boost_container_new_t())block_info_t();
mem_address += HdrSize;
this->priv_fill_chain_remaining_to_block(chain, target_elem_in_chain, c_info, mem_address, real_num_node, real_node_size);
const size_type free_nodes = c_info.free_nodes.size();
if(free_nodes){
const bool is_full = free_nodes == real_num_node;
BOOST_ASSERT(free_nodes < real_num_node);
m_totally_free_blocks += static_cast<size_type>(is_full);
block_container_traits_t::insert_was_empty(m_block_container, c_info, is_full);
}
}
}
void priv_append_from_new_blocks( size_type min_elements, multiallocation_chain &chain
, const size_type max_free_blocks
, const size_type real_block_alignment, const size_type real_node_size
, const size_type real_num_node, const size_type num_subblocks
, AlignOnlyFalse)
{
BOOST_ASSERT(m_block_container.empty());
BOOST_ASSERT(min_elements > 0);
const size_type n = (min_elements - 1)/real_num_node + 1;
const size_type real_block_size = real_block_alignment*num_subblocks - PayloadPerAllocation;
const size_type elements_per_subblock_mid = (real_block_alignment - HdrOffsetSize)/real_node_size;
const size_type elements_per_subblock_end = (real_block_alignment - HdrOffsetSize - PayloadPerAllocation) / real_node_size;
const size_type hdr_subblock_elements = (real_block_alignment - HdrSize - PayloadPerAllocation)/real_node_size;
const size_type target_elem_in_chain = chain.size() + min_elements;
for(size_type i = 0; i != n; ++i){
//We allocate a new NodeBlock and put it the last
//element of the tree
char *mem_address = static_cast<char*>
(mp_segment_mngr_base->allocate_aligned(real_block_size, real_block_alignment));
if(!mem_address){
//In case of error, free memory deallocating all nodes (the new ones allocated
//in this function plus previously stored nodes in chain).
this->priv_deallocate_nodes(chain, max_free_blocks, real_num_node, num_subblocks, real_block_alignment);
throw_bad_alloc();
}
//First initialize header information on the last subblock
char *hdr_addr = mem_address + real_block_alignment*(num_subblocks-1);
block_info_t &c_info = *new(hdr_addr, boost_container_new_t())block_info_t();
//Some structural checks
BOOST_ASSERT(static_cast<void*>(&static_cast<hdr_offset_holder&>(c_info).hdr_offset) ==
static_cast<void*>(&c_info)); (void)c_info;
for( size_type subblock = 0, maxsubblock = num_subblocks - 1
; subblock < maxsubblock
; ++subblock, mem_address += real_block_alignment){
//Initialize header offset mark
new(mem_address, boost_container_new_t()) hdr_offset_holder(size_type(hdr_addr - mem_address));
const size_type elements_per_subblock = (subblock != (maxsubblock - 1)) ? elements_per_subblock_mid : elements_per_subblock_end;
this->priv_fill_chain_remaining_to_block
(chain, target_elem_in_chain, c_info, mem_address + HdrOffsetSize, elements_per_subblock, real_node_size);
}
this->priv_fill_chain_remaining_to_block
(chain, target_elem_in_chain, c_info, hdr_addr + HdrSize, hdr_subblock_elements, real_node_size);
m_totally_free_blocks += static_cast<size_type>(c_info.free_nodes.size() == real_num_node);
if (c_info.free_nodes.size())
m_block_container.push_front(c_info);
}
}
//!Allocates array of count elements. Can throw
void *priv_allocate_node( const size_type max_free_blocks, const size_type real_block_alignment, const size_type real_node_size
, const size_type real_num_node, const size_type num_subblocks)
{
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
//If there are no free nodes we allocate a new block
if(!m_block_container.empty()){
//We take the first free node the multiset can't be empty
free_nodes_t &free_nodes = m_block_container.begin()->free_nodes;
BOOST_ASSERT(!free_nodes.empty());
const size_type free_nodes_count = free_nodes.size();
void *first_node = boost::movelib::to_raw_pointer(free_nodes.pop_front());
if(free_nodes.empty()){
block_container_traits_t::erase_first(m_block_container);
}
m_totally_free_blocks -= static_cast<size_type>(free_nodes_count == real_num_node);
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
return first_node;
}
else{
multiallocation_chain chain;
this->priv_append_from_new_blocks
(1, chain, max_free_blocks, real_block_alignment, real_node_size, real_num_node, num_subblocks, IsAlignOnly());
void *node = boost::movelib::to_raw_pointer(chain.pop_front());
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
return node;
}
}
void priv_allocate_nodes( const size_type n, multiallocation_chain &chain
, const size_type max_free_blocks, const size_type real_block_alignment, const size_type real_node_size
, const size_type real_num_node, const size_type num_subblocks)
{
size_type i = 0;
BOOST_CONTAINER_TRY{
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
while(i != n){
//If there are no free nodes we allocate all needed blocks
if (m_block_container.empty()){
this->priv_append_from_new_blocks
(n - i, chain, max_free_blocks, real_block_alignment, real_node_size, real_num_node, num_subblocks, IsAlignOnly());
BOOST_ASSERT(m_block_container.empty() || (++m_block_container.cbegin() == m_block_container.cend()));
BOOST_ASSERT(chain.size() == n);
break;
}
free_nodes_t &free_nodes = m_block_container.begin()->free_nodes;
const size_type free_nodes_count_before = free_nodes.size();
m_totally_free_blocks -= static_cast<size_type>(free_nodes_count_before == real_num_node);
const size_type num_left = n-i;
const size_type num_elems = (num_left < free_nodes_count_before) ? num_left : free_nodes_count_before;
typedef typename free_nodes_t::iterator free_nodes_iterator;
if(num_left < free_nodes_count_before){
const free_nodes_iterator it_bbeg(free_nodes.before_begin());
free_nodes_iterator it_bend(it_bbeg);
for(size_type j = 0; j != num_elems; ++j){
++it_bend;
}
free_nodes_iterator it_end = it_bend; ++it_end;
free_nodes_iterator it_beg = it_bbeg; ++it_beg;
free_nodes.erase_after(it_bbeg, it_end, num_elems);
chain.incorporate_after(chain.last(), &*it_beg, &*it_bend, num_elems);
//chain.splice_after(chain.last(), free_nodes, it_bbeg, it_bend, num_elems);
BOOST_ASSERT(!free_nodes.empty());
}
else{
const free_nodes_iterator it_beg(free_nodes.begin()), it_bend(free_nodes.last());
free_nodes.clear();
chain.incorporate_after(chain.last(), &*it_beg, &*it_bend, num_elems);
block_container_traits_t::erase_first(m_block_container);
}
i += num_elems;
}
}
BOOST_CONTAINER_CATCH(...){
this->priv_deallocate_nodes(chain, max_free_blocks, real_num_node, num_subblocks, real_block_alignment);
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
BOOST_CONTAINER_RETHROW
}
BOOST_CONTAINER_CATCH_END
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
}
//!Deallocates an array pointed by ptr. Never throws
void priv_deallocate_node( void *pElem
, const size_type max_free_blocks, const size_type real_num_node
, const size_type num_subblocks, const size_type real_block_alignment)
{
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
block_info_t &block_info = *this->priv_block_from_node(pElem, real_block_alignment);
const size_type prev_free_nodes = block_info.free_nodes.size();
BOOST_ASSERT(block_info.free_nodes.size() < real_num_node);
//We put the node at the beginning of the free node list
block_info.free_nodes.push_back(void_pointer(pElem));
//The loop reinserts all blocks except the last one
this->priv_reinsert_block(block_info, prev_free_nodes == 0, real_num_node);
this->priv_deallocate_free_blocks(max_free_blocks, real_num_node, num_subblocks, real_block_alignment);
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
}
void priv_deallocate_nodes( multiallocation_chain &nodes
, const size_type max_free_blocks, const size_type real_num_node
, const size_type num_subblocks, const size_type real_block_alignment)
{
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
//To take advantage of node locality, wait until two
//nodes belong to different blocks. Only then reinsert
//the block of the first node in the block tree.
//Cache of the previous block
block_info_t *prev_block_info = 0;
//If block was empty before this call, it's not already
//inserted in the block tree.
bool prev_block_was_empty = false;
typedef typename free_nodes_t::iterator free_nodes_iterator;
{
const free_nodes_iterator itbb(nodes.before_begin()), ite(nodes.end());
free_nodes_iterator itf(nodes.begin()), itbf(itbb);
size_type splice_node_count = size_type(-1);
while(itf != ite){
void *pElem = boost::movelib::to_raw_pointer(boost::movelib::iterator_to_raw_pointer(itf));
block_info_t &block_info = *this->priv_block_from_node(pElem, real_block_alignment);
BOOST_ASSERT(block_info.free_nodes.size() < real_num_node);
++splice_node_count;
//If block change is detected calculate the cached block position in the tree
if(&block_info != prev_block_info){
if(prev_block_info){ //Make sure we skip the initial "dummy" cache
free_nodes_iterator it(itbb); ++it;
nodes.erase_after(itbb, itf, splice_node_count);
prev_block_info->free_nodes.incorporate_after(prev_block_info->free_nodes.last(), &*it, &*itbf, splice_node_count);
this->priv_reinsert_block(*prev_block_info, prev_block_was_empty, real_num_node);
splice_node_count = 0;
}
//Update cache with new data
prev_block_was_empty = block_info.free_nodes.empty();
prev_block_info = &block_info;
}
itbf = itf;
++itf;
}
}
if(prev_block_info){
//The loop reinserts all blocks except the last one
const free_nodes_iterator itfirst(nodes.begin()), itlast(nodes.last());
const size_type splice_node_count = nodes.size();
nodes.clear();
prev_block_info->free_nodes.incorporate_after(prev_block_info->free_nodes.last(), &*itfirst, &*itlast, splice_node_count);
this->priv_reinsert_block(*prev_block_info, prev_block_was_empty, real_num_node);
this->priv_deallocate_free_blocks(max_free_blocks, real_num_node, num_subblocks, real_block_alignment);
}
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
}
void priv_reinsert_block(block_info_t &prev_block_info, const bool prev_block_was_empty, const size_type real_num_node)
{
//Cache the free nodes from the block
const size_type this_block_free_nodes = prev_block_info.free_nodes.size();
const bool is_full = this_block_free_nodes == real_num_node;
//Update free block count
m_totally_free_blocks += static_cast<size_type>(is_full);
if(prev_block_was_empty){
block_container_traits_t::insert_was_empty(m_block_container, prev_block_info, is_full);
}
else{
block_container_traits_t::reinsert_was_used(m_block_container, prev_block_info, is_full);
}
}
block_info_t *priv_block_from_node(void *node, const size_type real_block_alignment, AlignOnlyFalse) const
{
hdr_offset_holder *hdr_off_holder =
reinterpret_cast<hdr_offset_holder*>((std::size_t)node & size_type(~(real_block_alignment - 1)));
BOOST_ASSERT(0 == ((std::size_t)hdr_off_holder & (real_block_alignment - 1)));
BOOST_ASSERT(0 == (hdr_off_holder->hdr_offset & (real_block_alignment - 1)));
block_info_t *block = move_detail::force_ptr<block_info_t *>
(reinterpret_cast<char*>(hdr_off_holder) + hdr_off_holder->hdr_offset);
BOOST_ASSERT(block->hdr_offset == 0);
return block;
}
block_info_t *priv_block_from_node(void *node, const size_type real_block_alignment, AlignOnlyTrue) const
{
return (block_info_t *)((std::size_t)node & std::size_t(~(real_block_alignment - 1)));
}
block_info_t *priv_block_from_node(void *node, const size_type real_block_alignment) const
{ return this->priv_block_from_node(node, real_block_alignment, IsAlignOnly()); }
//!Deallocates all used memory. Never throws
void priv_clear(const size_type num_subblocks, const size_type real_block_alignment, const size_type real_num_node)
{
#ifndef NDEBUG
block_iterator it = m_block_container.begin();
block_iterator itend = m_block_container.end();
size_type n_free_nodes = 0;
for(; it != itend; ++it){
//Check for memory leak
BOOST_ASSERT(it->free_nodes.size() == real_num_node);
++n_free_nodes;
}
BOOST_ASSERT(n_free_nodes == m_totally_free_blocks);
#endif
//Check for memory leaks
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
multiallocation_chain chain;
m_block_container.clear_and_dispose(block_destroyer(this, chain, num_subblocks, real_block_alignment, real_num_node));
this->mp_segment_mngr_base->deallocate_many(chain);
m_totally_free_blocks = 0;
this->priv_invariants(real_num_node, num_subblocks, real_block_alignment);
}
public:
private_adaptive_node_pool_impl_common(segment_manager_base_type *segment_mngr_base)
//General purpose allocator
: mp_segment_mngr_base(segment_mngr_base)
, m_block_container()
, m_totally_free_blocks(0)
{}
size_type num_free_nodes()
{
typedef typename block_container_t::const_iterator citerator;
size_type count = 0;
citerator it (m_block_container.begin()), itend(m_block_container.end());
for(; it != itend; ++it){
count += it->free_nodes.size();
}
return count;
}
void swap(private_adaptive_node_pool_impl_common &other)
{
std::swap(mp_segment_mngr_base, other.mp_segment_mngr_base);
std::swap(m_totally_free_blocks, other.m_totally_free_blocks);
m_block_container.swap(other.m_block_container);
}
//!Returns the segment manager. Never throws
segment_manager_base_type* get_segment_manager_base()const
{ return boost::movelib::to_raw_pointer(mp_segment_mngr_base); }
};
template< class SizeType
, std::size_t HdrSize
, std::size_t PayloadPerAllocation
, std::size_t RealNodeSize
, std::size_t NodesPerBlock
, std::size_t HdrOffsetSize
, std::size_t OverheadPercent
, bool AlignOnly>
struct calculate_alignment_ct
{
static const std::size_t alignment = upper_power_of_2_ct<SizeType, HdrSize + RealNodeSize*NodesPerBlock>::value;
static const std::size_t num_subblocks = 0;
static const std::size_t real_num_node = (alignment - PayloadPerAllocation - HdrSize)/RealNodeSize;
};
template< class SizeType
, std::size_t HdrSize
, std::size_t PayloadPerAllocation
, std::size_t RealNodeSize
, std::size_t NodesPerBlock
, std::size_t HdrOffsetSize
, std::size_t OverheadPercent>
struct calculate_alignment_ct
< SizeType
, HdrSize
, PayloadPerAllocation
, RealNodeSize
, NodesPerBlock
, HdrOffsetSize
, OverheadPercent
, false>
{
typedef typename candidate_power_of_2_ct
< upper_power_of_2_ct<SizeType, HdrSize + PayloadPerAllocation + RealNodeSize>::value
, RealNodeSize
, PayloadPerAllocation
, NodesPerBlock
, HdrSize
, HdrOffsetSize
, OverheadPercent
>::type type;
static const std::size_t alignment = type::alignment;
static const std::size_t num_subblocks = type::num_subblocks;
static const std::size_t real_num_node = type::real_num_node;
};
/////////////////////////////////////////////
//
// private_adaptive_node_pool_impl_ct
//
/////////////////////////////////////////////
template< class SegmentManagerBase
, std::size_t MaxFreeBlocks
, std::size_t NodeSize
, std::size_t NodesPerBlock
, std::size_t OverheadPercent
, unsigned int Flags>
class private_adaptive_node_pool_impl_ct
: public private_adaptive_node_pool_impl_common<SegmentManagerBase, Flags>
{
typedef private_adaptive_node_pool_impl_common<SegmentManagerBase, Flags> base_t;
//Non-copyable
private_adaptive_node_pool_impl_ct();
private_adaptive_node_pool_impl_ct(const private_adaptive_node_pool_impl_ct &);
private_adaptive_node_pool_impl_ct &operator=(const private_adaptive_node_pool_impl_ct &);
public:
typedef typename base_t::void_pointer void_pointer;
typedef typename base_t::size_type size_type;
typedef typename base_t::multiallocation_chain multiallocation_chain;
typedef typename base_t::segment_manager_base_type segment_manager_base_type;
static const typename base_t::size_type PayloadPerAllocation = base_t::PayloadPerAllocation;
//align_only
static const bool AlignOnly = base_t::AlignOnly;
private:
static const size_type MaxAlign = base_t::MaxAlign;
static const size_type HdrSize = base_t::HdrSize;
static const size_type HdrOffsetSize = base_t::HdrOffsetSize;
static const size_type RealNodeSize = lcm_ct<NodeSize, alignment_of<void_pointer>::value>::value;
typedef calculate_alignment_ct
< size_type, HdrSize, PayloadPerAllocation
, RealNodeSize, NodesPerBlock, HdrOffsetSize, OverheadPercent, AlignOnly> data_t;
//Round the size to a power of two value.
//This is the total memory size (including payload) that we want to
//allocate from the general-purpose allocator
static const size_type NumSubBlocks = data_t::num_subblocks;
static const size_type RealNumNode = data_t::real_num_node;
static const size_type RealBlockAlignment = data_t::alignment;
public:
//!Constructor from a segment manager. Never throws
private_adaptive_node_pool_impl_ct(typename base_t::segment_manager_base_type *segment_mngr_base)
//General purpose allocator
: base_t(segment_mngr_base)
{}
//!Destructor. Deallocates all allocated blocks. Never throws
~private_adaptive_node_pool_impl_ct()
{ this->priv_clear(NumSubBlocks, data_t::alignment, RealNumNode); }
size_type get_real_num_node() const
{ return RealNumNode; }
//!Allocates array of count elements. Can throw
void *allocate_node()
{
return this->priv_allocate_node
(MaxFreeBlocks, data_t::alignment, RealNodeSize, RealNumNode, NumSubBlocks);
}
//!Allocates n nodes.
//!Can throw
void allocate_nodes(const size_type n, multiallocation_chain &chain)
{
this->priv_allocate_nodes
(n, chain, MaxFreeBlocks, data_t::alignment, RealNodeSize, RealNumNode, NumSubBlocks);
}
//!Deallocates an array pointed by ptr. Never throws
void deallocate_node(void *pElem)
{
this->priv_deallocate_node(pElem, MaxFreeBlocks, RealNumNode, NumSubBlocks, RealBlockAlignment);
}
//!Deallocates a linked list of nodes. Never throws
void deallocate_nodes(multiallocation_chain &nodes)
{
this->priv_deallocate_nodes(nodes, MaxFreeBlocks, RealNumNode, NumSubBlocks, data_t::alignment);
}
void deallocate_free_blocks()
{ this->priv_deallocate_free_blocks(0, RealNumNode, NumSubBlocks, data_t::alignment); }
//Deprecated, use deallocate_free_blocks
void deallocate_free_chunks()
{ this->priv_deallocate_free_blocks(0, RealNumNode, NumSubBlocks, data_t::alignment); }
};
/////////////////////////////////////////////
//
// private_adaptive_node_pool_impl_rt
//
/////////////////////////////////////////////
template<class SizeType>
struct private_adaptive_node_pool_impl_rt_data
{
typedef SizeType size_type;
private_adaptive_node_pool_impl_rt_data(size_type max_free_blocks, size_type real_node_size)
: m_max_free_blocks(max_free_blocks), m_real_node_size(real_node_size)
, m_real_block_alignment(), m_num_subblocks(), m_real_num_node()
{}
const size_type m_max_free_blocks;
const size_type m_real_node_size;
//Round the size to a power of two value.
//This is the total memory size (including payload) that we want to
//allocate from the general-purpose allocator
size_type m_real_block_alignment;
size_type m_num_subblocks;
//This is the real number of nodes per block
size_type m_real_num_node;
};
template<class SegmentManagerBase, unsigned int Flags>
class private_adaptive_node_pool_impl_rt
: private private_adaptive_node_pool_impl_rt_data<typename SegmentManagerBase::size_type>
, public private_adaptive_node_pool_impl_common<SegmentManagerBase, Flags>
{
typedef private_adaptive_node_pool_impl_common<SegmentManagerBase, Flags> impl_t;
typedef private_adaptive_node_pool_impl_rt_data<typename SegmentManagerBase::size_type> data_t;
//Non-copyable
private_adaptive_node_pool_impl_rt();
private_adaptive_node_pool_impl_rt(const private_adaptive_node_pool_impl_rt &);
private_adaptive_node_pool_impl_rt &operator=(const private_adaptive_node_pool_impl_rt &);
protected:
typedef typename impl_t::void_pointer void_pointer;
typedef typename impl_t::size_type size_type;
typedef typename impl_t::multiallocation_chain multiallocation_chain;
static const typename impl_t::size_type PayloadPerAllocation = impl_t::PayloadPerAllocation;
//Flags
//align_only
static const bool AlignOnly = impl_t::AlignOnly;
static const size_type HdrSize = impl_t::HdrSize;
static const size_type HdrOffsetSize = impl_t::HdrOffsetSize;
public:
//!Segment manager typedef
typedef SegmentManagerBase segment_manager_base_type;
//!Constructor from a segment manager. Never throws
private_adaptive_node_pool_impl_rt
( segment_manager_base_type *segment_mngr_base
, size_type node_size
, size_type nodes_per_block
, size_type max_free_blocks
, unsigned char overhead_percent
)
: data_t(max_free_blocks, lcm(node_size, size_type(alignment_of<void_pointer>::value)))
, impl_t(segment_mngr_base)
{
if(AlignOnly){
this->m_real_block_alignment = upper_power_of_2(HdrSize + this->m_real_node_size*nodes_per_block);
this->m_real_num_node = (this->m_real_block_alignment - PayloadPerAllocation - HdrSize)/this->m_real_node_size;
}
else{
candidate_power_of_2_rt ( upper_power_of_2(HdrSize + PayloadPerAllocation + this->m_real_node_size)
, this->m_real_node_size
, PayloadPerAllocation
, nodes_per_block
, HdrSize
, HdrOffsetSize
, overhead_percent
, this->m_real_block_alignment
, this->m_num_subblocks
, this->m_real_num_node);
}
}
//!Destructor. Deallocates all allocated blocks. Never throws
~private_adaptive_node_pool_impl_rt()
{ this->priv_clear(this->m_num_subblocks, this->m_real_block_alignment, this->m_real_num_node); }
size_type get_real_num_node() const
{ return this->m_real_num_node; }
//!Allocates array of count elements. Can throw
void *allocate_node()
{
return this->priv_allocate_node
(this->m_max_free_blocks, this->m_real_block_alignment, this->m_real_node_size, this->m_real_num_node, this->m_num_subblocks);
}
//!Allocates n nodes.
//!Can throw
void allocate_nodes(const size_type n, multiallocation_chain &chain)
{
this->priv_allocate_nodes
(n, chain, this->m_max_free_blocks, this->m_real_block_alignment, this->m_real_node_size, this->m_real_num_node, this->m_num_subblocks);
}
//!Deallocates an array pointed by ptr. Never throws
void deallocate_node(void *pElem)
{
this->priv_deallocate_node(pElem, this->m_max_free_blocks, this->m_real_num_node, this->m_num_subblocks, this->m_real_block_alignment);
}
//!Deallocates a linked list of nodes. Never throws
void deallocate_nodes(multiallocation_chain &nodes)
{
this->priv_deallocate_nodes(nodes, this->m_max_free_blocks, this->m_real_num_node, this->m_num_subblocks, this->m_real_block_alignment);
}
void deallocate_free_blocks()
{ this->priv_deallocate_free_blocks(0, this->m_real_num_node, this->m_num_subblocks, this->m_real_block_alignment); }
//Deprecated, use deallocate_free_blocks
void deallocate_free_chunks()
{ this->priv_deallocate_free_blocks(0, this->m_real_num_node, this->m_num_subblocks, this->m_real_block_alignment); }
};
} //namespace dtl {
} //namespace container {
} //namespace boost {
#include <boost/container/detail/config_end.hpp>
#endif //#ifndef BOOST_CONTAINER_DETAIL_ADAPTIVE_NODE_POOL_IMPL_HPP