mt_allocator.h

mingw32.rar

// MT-optimized allocator -*- C++ -*-

// Copyright (C) 2003, 2004 Free Software Foundation, Inc.
//
// This file is part of the GNU ISO C++ Library.  This library is free
// software; you can redistribute it and/or modify it under the
// terms of the GNU General Public License as published by the
// Free Software Foundation; either version 2, or (at your option)
// any later version.

// This library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU General Public License for more details.

// You should have received a copy of the GNU General Public License along
// with this library; see the file COPYING.  If not, write to the Free
// Software Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307,
// USA.

// As a special exception, you may use this file as part of a free software
// library without restriction.  Specifically, if other files instantiate
// templates or use macros or inline functions from this file, or you compile
// this file and link it with other files to produce an executable, this
// file does not by itself cause the resulting executable to be covered by
// the GNU General Public License.  This exception does not however
// invalidate any other reasons why the executable file might be covered by
// the GNU General Public License.

/** @file ext/mt_allocator.h
 *  This file is a GNU extension to the Standard C++ Library.
 *  You should only include this header if you are using GCC 3 or later.
 */

#ifndef _MT_ALLOCATOR_H
#define _MT_ALLOCATOR_H 1

#include <new>
#include <cstdlib>
#include <bits/functexcept.h>
#include <bits/gthr.h>
#include <bits/atomicity.h>

namespace __gnu_cxx
{
  /**
   *  This is a fixed size (power of 2) allocator which - when
   *  compiled with thread support - will maintain one freelist per
   *  size per thread plus a "global" one. Steps are taken to limit
   *  the per thread freelist sizes (by returning excess back to
   *  "global").
   *
   *  Further details:
   *  http://gcc.gnu.org/onlinedocs/libstdc++/ext/mt_allocator.html
   */
  template<typename _Tp>
    class __mt_alloc
    {
    public:
      typedef size_t                    size_type;
      typedef ptrdiff_t                 difference_type;
      typedef _Tp*                      pointer;
      typedef const _Tp*                const_pointer;
      typedef _Tp&                      reference;
      typedef const _Tp&                const_reference;
      typedef _Tp                       value_type;

      template<typename _Tp1>
        struct rebind
        { typedef __mt_alloc<_Tp1> other; };

      __mt_alloc() throw() 
      {
	// XXX
      }

      __mt_alloc(const __mt_alloc&) throw() 
      {
	// XXX
      }

      template<typename _Tp1>
        __mt_alloc(const __mt_alloc<_Tp1>& obj) throw()  
        {
	  // XXX
	}

      ~__mt_alloc() throw() { }

      pointer
      address(reference __x) const
      { return &__x; }

      const_pointer
      address(const_reference __x) const
      { return &__x; }

      size_type
      max_size() const throw() 
      { return size_t(-1) / sizeof(_Tp); }

      // _GLIBCXX_RESOLVE_LIB_DEFECTS
      // 402. wrong new expression in [some_] allocator::construct
      void 
      construct(pointer __p, const _Tp& __val) 
      { ::new(__p) _Tp(__val); }

      void 
      destroy(pointer __p) { __p->~_Tp(); }

      pointer
      allocate(size_type __n, const void* = 0);

      void
      deallocate(pointer __p, size_type __n);

      // Variables used to configure the behavior of the allocator,
      // assigned and explained in detail below.
      struct _Tune
      {
	// Alignment needed.
	// NB: In any case must be >= sizeof(_Block_record), that
	// is 4 on 32 bit machines and 8 on 64 bit machines.
	size_t  _M_align;

	// Allocation requests (after round-up to power of 2) below
	// this value will be handled by the allocator. A raw new/
	// call will be used for requests larger than this value.
	size_t	_M_max_bytes; 

	// Size in bytes of the smallest bin.
	// NB: Must be a power of 2 and >= _M_align.
	size_t  _M_min_bin;

	// In order to avoid fragmenting and minimize the number of
	// new() calls we always request new memory using this
	// value. Based on previous discussions on the libstdc++
	// mailing list we have choosen the value below.
	// See http://gcc.gnu.org/ml/libstdc++/2001-07/msg00077.html
	size_t 	_M_chunk_size;

	// The maximum number of supported threads. Our Linux 2.4.18
	// reports 4070 in /proc/sys/kernel/threads-max
	size_t 	_M_max_threads;

	// Each time a deallocation occurs in a threaded application
	// we make sure that there are no more than
	// _M_freelist_headroom % of used memory on the freelist. If
	// the number of additional records is more than
	// _M_freelist_headroom % of the freelist, we move these
	// records back to the global pool.
	size_t 	_M_freelist_headroom;

	// Set to true forces all allocations to use new().
	bool 	_M_force_new; 
     
	explicit
	_Tune()
	: _M_align(8), _M_max_bytes(128), _M_min_bin(8),
	  _M_chunk_size(4096 - 4 * sizeof(void*)), 
	  _M_max_threads(4096), _M_freelist_headroom(10), 
	  _M_force_new(getenv("GLIBCXX_FORCE_NEW") ? true : false)
	{ }

	explicit
	_Tune(size_t __align, size_t __maxb, size_t __minbin,
	      size_t __chunk, size_t __maxthreads, size_t __headroom,
	      bool __force) 
	: _M_align(__align), _M_max_bytes(__maxb), _M_min_bin(__minbin),
	  _M_chunk_size(__chunk), _M_max_threads(__maxthreads),
	  _M_freelist_headroom(__headroom), _M_force_new(__force)
	{ }
      };

    private:
      // We need to create the initial lists and set up some variables
      // before we can answer to the first request for memory.
#ifdef __GTHREADS
      static __gthread_once_t 		_S_once;
#endif
      static bool 			_S_init;

      static void
      _S_initialize();

      // Configuration options.
      static _Tune 	       		_S_options;

      static const _Tune
      _S_get_options()
      { return _S_options; }

      static void
      _S_set_options(_Tune __t)
      { 
	if (!_S_init)
	  _S_options = __t;
      }

      // Using short int as type for the binmap implies we are never
      // caching blocks larger than 65535 with this allocator
      typedef unsigned short int        _Binmap_type;
      static _Binmap_type* 		_S_binmap;

      // Each requesting thread is assigned an id ranging from 1 to
      // _S_max_threads. Thread id 0 is used as a global memory pool.
      // In order to get constant performance on the thread assignment
      // routine, we keep a list of free ids. When a thread first
      // requests memory we remove the first record in this list and
      // stores the address in a __gthread_key. When initializing the
      // __gthread_key we specify a destructor. When this destructor
      // (i.e. the thread dies) is called, we return the thread id to
      // the front of this list.
#ifdef __GTHREADS
      struct _Thread_record
      {
        // Points to next free thread id record. NULL if last record in list.
        _Thread_record* volatile        _M_next;

	// Thread id ranging from 1 to _S_max_threads.
        size_t                          _M_id;
      };

      static _Thread_record* volatile 	_S_thread_freelist_first;
      static __gthread_mutex_t 		_S_thread_freelist_mutex;
      static __gthread_key_t 		_S_thread_key;

      static void 
      _S_destroy_thread_key(void* __freelist_pos);
#endif

      static size_t 
      _S_get_thread_id();

      union _Block_record
      {
	// Points to the block_record of the next free block.
        _Block_record* volatile         _M_next;

#ifdef __GTHREADS
	// The thread id of the thread which has requested this block.
        size_t                          _M_thread_id;
#endif
      };

      struct _Bin_record
      {
	// An "array" of pointers to the first free block for each
	// thread id. Memory to this "array" is allocated in _S_initialize()
	// for _S_max_threads + global pool 0.
        _Block_record** volatile        _M_first;

#ifdef __GTHREADS
	// An "array" of counters used to keep track of the amount of
	// blocks that are on the freelist/used for each thread id.
	// Memory to these "arrays" is allocated in _S_initialize() for
	// _S_max_threads + global pool 0.
        size_t* volatile                _M_free;
        size_t* volatile                _M_used;

	// Each bin has its own mutex which is used to ensure data
	// integrity while changing "ownership" on a block.  The mutex
	// is initialized in _S_initialize().
        __gthread_mutex_t*              _M_mutex;
#endif
      };

      // An "array" of bin_records each of which represents a specific
      // power of 2 size. Memory to this "array" is allocated in
      // _S_initialize().
      static _Bin_record* volatile     	_S_bin;

      // Actual value calculated in _S_initialize().
      static size_t 	       	     	_S_bin_size; 
    };

  template<typename _Tp>
    typename __mt_alloc<_Tp>::pointer
    __mt_alloc<_Tp>::
    allocate(size_type __n, const void*)
    {
      // Although the test in __gthread_once() would suffice, we wrap
      // test of the once condition in our own unlocked check. This
      // saves one function call to pthread_once() (which itself only
      // tests for the once value unlocked anyway and immediately
      // returns if set)
      if (!_S_init)
	{
#ifdef __GTHREADS
	  if (__gthread_active_p())
	    __gthread_once(&_S_once, _S_initialize);
#endif
	  if (!_S_init)
	    _S_initialize();
	}
      
      // Requests larger than _M_max_bytes are handled by new/delete
      // directly.
      const size_t __bytes = __n * sizeof(_Tp);
      if (__bytes > _S_options._M_max_bytes || _S_options._M_force_new)
	{
	  void* __ret = ::operator new(__bytes);
	  return static_cast<_Tp*>(__ret);
	}

      // Round up to power of 2 and figure out which bin to use.
      const size_t __which = _S_binmap[__bytes];      
      const size_t __thread_id = _S_get_thread_id();
      
      // Find out if we have blocks on our freelist.  If so, go ahead
      // and use them directly without having to lock anything.
      const _Bin_record& __bin = _S_bin[__which];
      _Block_record* __block = NULL;
      if (__bin._M_first[__thread_id] == NULL)
	{
	  // NB: For alignment reasons, we can't use the first _M_align
	  // bytes, even when sizeof(_Block_record) < _M_align.
	  const size_t __bin_size = ((_S_options._M_min_bin << __which)
				     + _S_options._M_align);
	  size_t __block_count = _S_options._M_chunk_size / __bin_size;	  

	  // Are we using threads?
	  // - Yes, check if there are free blocks on the global
	  //   list. If so, grab up to __block_count blocks in one
	  //   lock and change ownership. If the global list is 
	  //   empty, we allocate a new chunk and add those blocks 
	  //   directly to our own freelist (with us as owner).
	  // - No, all operations are made directly to global pool 0
	  //   no need to lock or change ownership but check for free
	  //   blocks on global list (and if not add new ones) and
	  //   get the first one.
#ifdef __GTHREADS
	  if (__gthread_active_p())
	    {
	      __gthread_mutex_lock(__bin._M_mutex);
	      if (__bin._M_first[0] == NULL)
		{
		  // No need to hold the lock when we are adding a
		  // whole chunk to our own list.
		  __gthread_mutex_unlock(__bin._M_mutex);
		  
		  void* __v = ::operator new(_S_options._M_chunk_size);
		  __bin._M_first[__thread_id] = static_cast<_Block_record*>(__v);
		  __bin._M_free[__thread_id] = __block_count;

		  --__block_count;
		  __block = __bin._M_first[__thread_id];
		  while (__block_count-- > 0)
		    {
		      char* __c = reinterpret_cast<char*>(__block) + __bin_size;
		      __block->_M_next = reinterpret_cast<_Block_record*>(__c);
		      __block = __block->_M_next;
		    }
		  __block->_M_next = NULL;
		}
	      else
		{
		  // Is the number of required blocks greater than or
		  // equal to the number that can be provided by the
		  // global free list?