/* -*- Mode:C++; c-file-style:"gnu"; indent-tabs-mode:nil; -*- */

/*

 * Copyright (c) 2005,2006,2007 INRIA

 *

 * This program is free software; you can redistribute it and/or modify

 * it under the terms of the GNU General Public License version 2 as

 * published by the Free Software Foundation;

 *

 * This program 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 program; if not, write to the Free Software

 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

 *

 * Author: Mathieu Lacage <mathieu.lacage@sophia.inria.fr>

 */

#include "buffer.h"

#include "ns3/assert.h"

#include "ns3/log.h"

#include "ns3/fatal-error.h"

#include "ns3/test.h"

#include "ns3/random-variable.h"

#include <iomanip>

#include <iostream>

NS_LOG_COMPONENT_DEFINE ("Buffer");

#define LOG_INTERNAL_STATE(y)                                                                    \

NS_LOG_LOGIC (y << "start="<<m_start<<", end="<<m_end<<", zero start="<<m_zeroAreaStart<<              \

          ", zero end="<<m_zeroAreaEnd<<", count="<<m_data->m_count<<", size="<<m_data->m_size<<   \

          ", dirty start="<<m_data->m_dirtyStart<<", dirty end="<<m_data->m_dirtyEnd)

#ifdef BUFFER_HEURISTICS

#define HEURISTICS(x) x

#else

#define HEURISTICS(x)

#endif

//#define PRINT_STATS 1

namespace ns3 {

/**

 * This data structure is variable-sized through its last member whose size

 * is determined at allocation time and stored in the m_size field.

 *

 * The so-called "dirty area" describes the area in the buffer which

 * has been reserved and used by a user. Multiple Buffer instances

 * may reference the same BufferData object instance and may

 * reference different parts of the underlying byte buffer. The

 * "dirty area" is union of all the areas referenced by the Buffer

 * instances which reference the same BufferData instance.

 * New user data can be safely written only outside of the "dirty

 * area" if the reference count is higher than 1 (that is, if

 * more than one Buffer instance references the same BufferData).

 */

struct BufferData {

  /* The reference count of an instance of this data structure.

   * Each buffer which references an instance holds a count.

   */

  uint32_t m_count;

  /* the size of the m_data field below.

   */

  uint32_t m_size;

  /* offset from the start of the m_data field below to the

   * start of the area in which user bytes were written.

   */

  uint32_t m_dirtyStart;

  /* offset from the start of the m_data field below to the

   * end of the area in which user bytes were written.

   */

  uint32_t m_dirtyEnd;

  /* The real data buffer holds _at least_ one byte.

   * Its real size is stored in the m_size field.

   */

  uint8_t m_data[1];

};

typedef std::vector<struct BufferData*> BufferDataList;

static struct BufferData *BufferAllocate (uint32_t reqSize);

static void BufferDeallocate (struct BufferData *data);

} // namespace ns3

namespace ns3 {

#ifdef BUFFER_HEURISTICS

/* The following macros are pretty evil but they are needed to allow us to

 * keep track of 3 possible states for the g_freeList variable:

 *  - uninitialized means that no one has created a buffer yet

 *    so no one has created the associated free list (it is created

 *    on-demand when the first buffer is created)

 *  - initialized means that the free list exists and is valid

 *  - destroyed means that the static destructors of this compilation unit

 *    have run so, the free list has been cleared from its content

 * The key is that in destroyed state, we are careful not re-create it

 * which is a typical weakness of lazy evaluation schemes which use 

 * '0' as a special value to indicate both un-initialized and destroyed.

 * Note that it is important to use '0' as the marker for un-initialized state

 * because the variable holding this state information is initialized to zero

 * which the compiler assigns to zero-memory which is initialized to _zero_

 * before the constructors run so this ensures perfect handling of crazy 

 * constructor orderings.

 */

#define MAGIC_DESTROYED (~(long) 0)

#define IS_UNINITIALIZED(x) (x == (BufferDataList*)0)

#define IS_DESTROYED(x) (x == (BufferDataList*)MAGIC_DESTROYED)

#define IS_INITIALIZED(x) (!IS_UNINITIALIZED(x) && !IS_DESTROYED(x))

#define DESTROYED ((BufferDataList*)MAGIC_DESTROYED)

#define UNINITIALIZED ((BufferDataList*)0)

static uint32_t g_recommendedStart = 0;

static uint64_t g_nAddNoRealloc = 0;

static uint64_t g_nAddRealloc = 0;

static BufferDataList *g_freeList = 0;

static uint32_t g_maxSize = 0;

static uint64_t g_nAllocs = 0;

static uint64_t g_nCreates = 0;

#endif /* BUFFER_HEURISTICS */

static struct LocalStaticDestructor {

  LocalStaticDestructor(void)

  {

#ifdef PRINT_STATS

#ifdef BUFFER_HEURISTICS

    double efficiency;

    efficiency = g_nAllocs;

    efficiency /= g_nCreates;

    std::cout <<"buffer free list efficiency="<<efficiency<<" (lower is better)" << std::endl;

    std::cout <<"buffer free list max size="<<g_maxSize<<std::endl;

    std::cout <<"buffer free list recommended start="<<g_recommendedStart<<std::endl;

    double addEfficiency;

    addEfficiency = g_nAddRealloc;

    addEfficiency /= g_nAddNoRealloc;

    std::cout <<"buffer add efficiency=" << addEfficiency << " (lower is better)"<<std::endl;

    //std::cout <<"n add reallocs="<< g_nAddRealloc << std::endl;

    //std::cout <<"n add no reallocs="<< g_nAddNoRealloc << std::endl;

#endif /* BUFFER_HEURISTICS */

#endif /* PRINT_STATS */

    if (IS_INITIALIZED(g_freeList))

      {

        for (BufferDataList::iterator i = g_freeList->begin ();

             i != g_freeList->end (); i++)

          {

            BufferDeallocate (*i);

          }

        delete g_freeList;

        g_freeList = DESTROYED;

      }

  }

} g_localStaticDestructor;

struct BufferData *

BufferAllocate (uint32_t reqSize)

{

  if (reqSize == 0) 

    {

      reqSize = 1;

    }

  NS_ASSERT (reqSize >= 1);

  uint32_t size = reqSize - 1 + sizeof (struct BufferData);

  uint8_t *b = new uint8_t [size];

  struct BufferData *data = reinterpret_cast<struct BufferData*>(b);

  data->m_size = reqSize;

  data->m_count = 1;

  return data;

}

void

BufferDeallocate (struct BufferData *data)

{

  NS_ASSERT (data->m_count == 0);

  uint8_t *buf = reinterpret_cast<uint8_t *> (data);

  delete [] buf;

}

#ifdef BUFFER_HEURISTICS

void

Buffer::Recycle (struct BufferData *data)

{

  NS_ASSERT (data->m_count == 0);

  NS_ASSERT (!IS_UNINITIALIZED(g_freeList));

  g_maxSize = std::max (g_maxSize, data->m_size);

  /* feed into free list */

  if (data->m_size < g_maxSize ||

      IS_DESTROYED(g_freeList) ||

      g_freeList->size () > 1000)

    {

      BufferDeallocate (data);

    }

  else

    {

      NS_ASSERT (IS_INITIALIZED(g_freeList));

      g_freeList->push_back (data);

    }

}

BufferData *

Buffer::Create (uint32_t dataSize)

{

  /* try to find a buffer correctly sized. */

  g_nCreates++;

  if (IS_UNINITIALIZED(g_freeList))

    {

      g_freeList = new BufferDataList ();

    }

  else if (IS_INITIALIZED(g_freeList))

    {

      while (!g_freeList->empty ()) 

        {

          struct BufferData *data = g_freeList->back ();

          g_freeList->pop_back ();

          if (data->m_size >= dataSize) 

            {

              data->m_count = 1;

              return data;

            }

          BufferDeallocate (data);

        }

    }

  g_nAllocs++;

  struct BufferData *data = BufferAllocate (dataSize);

  NS_ASSERT (data->m_count == 1);

  return data;

}

#else

void

Buffer::Recycle (struct BufferData *data)

{

  NS_ASSERT (data->m_count == 0);

  BufferDeallocate (data);

}

BufferData *

Buffer::Create (uint32_t size)

{

  return BufferAllocate (size);

}

#endif

Buffer::Buffer ()

{

  NS_LOG_FUNCTION (this);

  Initialize (0);

}

Buffer::Buffer (uint32_t dataSize)

{

  NS_LOG_FUNCTION (this << dataSize);

  Initialize (dataSize);

}

bool

Buffer::CheckInternalState (void) const

{

  bool offsetsOk = 

    m_start <= m_zeroAreaStart &&

    m_zeroAreaStart <= m_zeroAreaEnd &&

    m_zeroAreaEnd <= m_end;

  bool dirtyOk =

    m_start >= m_data->m_dirtyStart &&

    m_end <= m_data->m_dirtyEnd;

  bool internalSizeOk = m_end - (m_zeroAreaEnd - m_zeroAreaStart) <= m_data->m_size &&

    m_start <= m_data->m_size &&

    m_zeroAreaStart <= m_data->m_size;

  bool ok = m_data->m_count > 0 && offsetsOk && dirtyOk && internalSizeOk;

  if (!ok)

    {

      LOG_INTERNAL_STATE ("check " << this << 

                          ", " << (offsetsOk?"true":"false") << 

                          ", " << (dirtyOk?"true":"false") << 

                          ", " << (internalSizeOk?"true":"false") << " ");

    }

  return ok;

}

void

Buffer::Initialize (uint32_t zeroSize)

{

  NS_LOG_FUNCTION (this << zeroSize);

  m_data = Buffer::Create (0);

#ifdef BUFFER_HEURISTICS

  m_start = std::min (m_data->m_size, g_recommendedStart);

  m_maxZeroAreaStart = m_start;

#else

  m_start = 0;

#endif /* BUFFER_HEURISTICS */

  m_zeroAreaStart = m_start;

  m_zeroAreaEnd = m_zeroAreaStart + zeroSize;

  m_end = m_zeroAreaEnd;

  m_data->m_dirtyStart = m_start;

  m_data->m_dirtyEnd = m_end;

  NS_ASSERT (CheckInternalState ());

}

Buffer::Buffer (Buffer const&o)

  : m_data (o.m_data),

#ifdef BUFFER_HEURISTICS

    m_maxZeroAreaStart (o.m_zeroAreaStart),

#endif

    m_zeroAreaStart (o.m_zeroAreaStart),

    m_zeroAreaEnd (o.m_zeroAreaEnd),

    m_start (o.m_start),

    m_end (o.m_end)

{

  NS_LOG_FUNCTION (this << &o);

  m_data->m_count++;

  NS_ASSERT (CheckInternalState ());

}

Buffer &

Buffer::operator = (Buffer const&o)

{

  NS_LOG_FUNCTION (this << &o);

  NS_ASSERT (CheckInternalState ());

  if (m_data != o.m_data) 

    {

      // not assignment to self.

      m_data->m_count--;

      if (m_data->m_count == 0) 

        {

          Recycle (m_data);

        }

      m_data = o.m_data;

      m_data->m_count++;

    }

  HEURISTICS (

  g_recommendedStart = std::max (g_recommendedStart, m_maxZeroAreaStart);

  m_maxZeroAreaStart = o.m_maxZeroAreaStart;

  );

  m_zeroAreaStart = o.m_zeroAreaStart;

  m_zeroAreaEnd = o.m_zeroAreaEnd;

  m_start = o.m_start;

  m_end = o.m_end;

  NS_ASSERT (CheckInternalState ());

  return *this;

}

Buffer::~Buffer ()

{

  NS_LOG_FUNCTION (this);

  NS_ASSERT (CheckInternalState ());

  HEURISTICS (g_recommendedStart = std::max (g_recommendedStart, m_maxZeroAreaStart));

  m_data->m_count--;

  if (m_data->m_count == 0) 

    {

      Recycle (m_data);

    }

}

uint32_t 

Buffer::GetSize (void) const

{

  NS_ASSERT (CheckInternalState ());

  return m_end - m_start;

}

Buffer::Iterator 

Buffer::Begin (void) const

{

  NS_ASSERT (CheckInternalState ());

  return Buffer::Iterator (this);

}

Buffer::Iterator 

Buffer::End (void) const

{

  NS_ASSERT (CheckInternalState ());

  return Buffer::Iterator (this, false);

}

uint32_t

Buffer::GetInternalSize (void) const

{

  return m_zeroAreaStart - m_start + m_end - m_zeroAreaEnd;

}

uint32_t

Buffer::GetInternalEnd (void) const

{

  return m_end - (m_zeroAreaEnd - m_zeroAreaStart);

}

bool

Buffer::AddAtStart (uint32_t start)

{

  NS_LOG_FUNCTION (this << start);

  bool dirty;

  NS_ASSERT (CheckInternalState ());

  bool isDirty = m_data->m_count > 1 && m_start > m_data->m_dirtyStart;

  if (m_start >= start && !isDirty)

    {

      /* enough space in the buffer and not dirty. 

       * To add: |..|

       * Before: |*****---------***|

       * After:  |***..---------***|

       */

      NS_ASSERT (m_data->m_count == 1 || m_start == m_data->m_dirtyStart);

      m_start -= start;

      dirty = m_start > m_data->m_dirtyStart;

      // update dirty area

      m_data->m_dirtyStart = m_start;

      HEURISTICS (g_nAddNoRealloc++);

    } 

  else

    {

      uint32_t newSize = GetInternalSize () + start;

      struct BufferData *newData = Buffer::Create (newSize);

      memcpy (newData->m_data + start, m_data->m_data + m_start, GetInternalSize ());

      m_data->m_count--;

      if (m_data->m_count == 0)

        {

          Buffer::Recycle (m_data);

        }

      m_data = newData;

      int32_t delta = start - m_start;

      m_start += delta;

      m_zeroAreaStart += delta;

      m_zeroAreaEnd += delta;

      m_end += delta;

      m_start -= start;

      // update dirty area

      m_data->m_dirtyStart = m_start;

      m_data->m_dirtyEnd = m_end;

      dirty = true;

      HEURISTICS (g_nAddRealloc++);

    }

  HEURISTICS (m_maxZeroAreaStart = std::max (m_maxZeroAreaStart, m_zeroAreaStart));

  LOG_INTERNAL_STATE ("add start=" << start << ", ");

  NS_ASSERT (CheckInternalState ());

  return dirty;

}

bool

Buffer::AddAtEnd (uint32_t end)

{

  NS_LOG_FUNCTION (this << end);

  bool dirty;

  NS_ASSERT (CheckInternalState ());

  bool isDirty = m_data->m_count > 1 && m_end < m_data->m_dirtyEnd;

  if (GetInternalEnd () + end <= m_data->m_size && !isDirty)

    {

      /* enough space in buffer and not dirty

       * Add:    |...|

       * Before: |**----*****|

       * After:  |**----...**|

       */

      NS_ASSERT (m_data->m_count == 1 || m_end == m_data->m_dirtyEnd);

      m_end += end;

      // update dirty area.

      m_data->m_dirtyEnd = m_end;

      dirty = m_end < m_data->m_dirtyEnd;

      HEURISTICS (g_nAddNoRealloc++);

    } 

  else

    {

      uint32_t newSize = GetInternalSize () + end;

      struct BufferData *newData = Buffer::Create (newSize);

      memcpy (newData->m_data, m_data->m_data + m_start, GetInternalSize ());

      m_data->m_count--;

      if (m_data->m_count == 0) 

        {

          Buffer::Recycle (m_data);

        }