//#define Trace
// ParallelDeflateOutputStream.cs
// ------------------------------------------------------------------
//
// A DeflateStream that does compression only, it uses a
// divide-and-conquer approach with multiple threads to exploit multiple
// CPUs for the DEFLATE computation.
//
// last saved: <2011-July-31 14:49:40>
//
// ------------------------------------------------------------------
//
// Copyright (c) 2009-2011 by Dino Chiesa
// All rights reserved!
//
// This code module is part of DotNetZip, a zipfile class library.
//
// ------------------------------------------------------------------
//
// This code is licensed under the Microsoft Public License.
// See the file License.txt for the license details.
// More info on: http://dotnetzip.codeplex.com
//
// ------------------------------------------------------------------
using System;
using System.Collections.Generic;
using System.Threading;
using System.IO;
namespace OfficeOpenXml.Packaging.Ionic.Zlib
{
internal class WorkItem
{
public byte[] buffer;
public byte[] compressed;
public int crc;
public int index;
public int ordinal;
public int inputBytesAvailable;
public int compressedBytesAvailable;
public ZlibCodec compressor;
public WorkItem(int size,
Ionic.Zlib.CompressionLevel compressLevel,
CompressionStrategy strategy,
int ix)
{
this.buffer= new byte[size];
// alloc 5 bytes overhead for every block (margin of safety= 2)
int n = size + ((size / 32768)+1) * 5 * 2;
this.compressed = new byte[n];
this.compressor = new ZlibCodec();
this.compressor.InitializeDeflate(compressLevel, false);
this.compressor.OutputBuffer = this.compressed;
this.compressor.InputBuffer = this.buffer;
this.index = ix;
}
}
///
/// A class for compressing streams using the
/// Deflate algorithm with multiple threads.
///
///
///
///
/// This class performs DEFLATE compression through writing. For
/// more information on the Deflate algorithm, see IETF RFC 1951,
/// "DEFLATE Compressed Data Format Specification version 1.3."
///
///
///
/// This class is similar to , except
/// that this class is for compression only, and this implementation uses an
/// approach that employs multiple worker threads to perform the DEFLATE. On
/// a multi-cpu or multi-core computer, the performance of this class can be
/// significantly higher than the single-threaded DeflateStream, particularly
/// for larger streams. How large? Anything over 10mb is a good candidate
/// for parallel compression.
///
///
///
/// The tradeoff is that this class uses more memory and more CPU than the
/// vanilla DeflateStream, and also is less efficient as a compressor. For
/// large files the size of the compressed data stream can be less than 1%
/// larger than the size of a compressed data stream from the vanialla
/// DeflateStream. For smaller files the difference can be larger. The
/// difference will also be larger if you set the BufferSize to be lower than
/// the default value. Your mileage may vary. Finally, for small files, the
/// ParallelDeflateOutputStream can be much slower than the vanilla
/// DeflateStream, because of the overhead associated to using the thread
/// pool.
///
///
///
///
public class ParallelDeflateOutputStream : System.IO.Stream
{
private static readonly int IO_BUFFER_SIZE_DEFAULT = 64 * 1024; // 128k
private static readonly int BufferPairsPerCore = 4;
private System.Collections.Generic.List _pool;
private bool _leaveOpen;
private bool emitting;
private System.IO.Stream _outStream;
private int _maxBufferPairs;
private int _bufferSize = IO_BUFFER_SIZE_DEFAULT;
private AutoResetEvent _newlyCompressedBlob;
//private ManualResetEvent _writingDone;
//private ManualResetEvent _sessionReset;
private object _outputLock = new object();
private bool _isClosed;
private bool _firstWriteDone;
private int _currentlyFilling;
private int _lastFilled;
private int _lastWritten;
private int _latestCompressed;
private int _Crc32;
private Ionic.Crc.CRC32 _runningCrc;
private object _latestLock = new object();
private System.Collections.Generic.Queue _toWrite;
private System.Collections.Generic.Queue _toFill;
private Int64 _totalBytesProcessed;
private Ionic.Zlib.CompressionLevel _compressLevel;
private volatile Exception _pendingException;
private bool _handlingException;
private object _eLock = new Object(); // protects _pendingException
// This bitfield is used only when Trace is defined.
//private TraceBits _DesiredTrace = TraceBits.Write | TraceBits.WriteBegin |
//TraceBits.WriteDone | TraceBits.Lifecycle | TraceBits.Fill | TraceBits.Flush |
//TraceBits.Session;
//private TraceBits _DesiredTrace = TraceBits.WriteBegin | TraceBits.WriteDone | TraceBits.Synch | TraceBits.Lifecycle | TraceBits.Session ;
private TraceBits _DesiredTrace =
TraceBits.Session |
TraceBits.Compress |
TraceBits.WriteTake |
TraceBits.WriteEnter |
TraceBits.EmitEnter |
TraceBits.EmitDone |
TraceBits.EmitLock |
TraceBits.EmitSkip |
TraceBits.EmitBegin;
///
/// Create a ParallelDeflateOutputStream.
///
///
///
///
/// This stream compresses data written into it via the DEFLATE
/// algorithm (see RFC 1951), and writes out the compressed byte stream.
///
///
///
/// The instance will use the default compression level, the default
/// buffer sizes and the default number of threads and buffers per
/// thread.
///
///
///
/// This class is similar to ,
/// except that this implementation uses an approach that employs
/// multiple worker threads to perform the DEFLATE. On a multi-cpu or
/// multi-core computer, the performance of this class can be
/// significantly higher than the single-threaded DeflateStream,
/// particularly for larger streams. How large? Anything over 10mb is
/// a good candidate for parallel compression.
///
///
///
///
///
///
/// This example shows how to use a ParallelDeflateOutputStream to compress
/// data. It reads a file, compresses it, and writes the compressed data to
/// a second, output file.
///
///
/// byte[] buffer = new byte[WORKING_BUFFER_SIZE];
/// int n= -1;
/// String outputFile = fileToCompress + ".compressed";
/// using (System.IO.Stream input = System.IO.File.OpenRead(fileToCompress))
/// {
/// using (var raw = System.IO.File.Create(outputFile))
/// {
/// using (Stream compressor = new ParallelDeflateOutputStream(raw))
/// {
/// while ((n= input.Read(buffer, 0, buffer.Length)) != 0)
/// {
/// compressor.Write(buffer, 0, n);
/// }
/// }
/// }
/// }
///
///
/// Dim buffer As Byte() = New Byte(4096) {}
/// Dim n As Integer = -1
/// Dim outputFile As String = (fileToCompress & ".compressed")
/// Using input As Stream = File.OpenRead(fileToCompress)
/// Using raw As FileStream = File.Create(outputFile)
/// Using compressor As Stream = New ParallelDeflateOutputStream(raw)
/// Do While (n <> 0)
/// If (n > 0) Then
/// compressor.Write(buffer, 0, n)
/// End If
/// n = input.Read(buffer, 0, buffer.Length)
/// Loop
/// End Using
/// End Using
/// End Using
///
///
/// The stream to which compressed data will be written.
public ParallelDeflateOutputStream(System.IO.Stream stream)
: this(stream, CompressionLevel.Default, CompressionStrategy.Default, false)
{
}
///
/// Create a ParallelDeflateOutputStream using the specified CompressionLevel.
///
///
/// See the
/// constructor for example code.
///
/// The stream to which compressed data will be written.
/// A tuning knob to trade speed for effectiveness.
public ParallelDeflateOutputStream(System.IO.Stream stream, CompressionLevel level)
: this(stream, level, CompressionStrategy.Default, false)
{
}
///
/// Create a ParallelDeflateOutputStream and specify whether to leave the captive stream open
/// when the ParallelDeflateOutputStream is closed.
///
///
/// See the
/// constructor for example code.
///
/// The stream to which compressed data will be written.
///
/// true if the application would like the stream to remain open after inflation/deflation.
///
public ParallelDeflateOutputStream(System.IO.Stream stream, bool leaveOpen)
: this(stream, CompressionLevel.Default, CompressionStrategy.Default, leaveOpen)
{
}
///
/// Create a ParallelDeflateOutputStream and specify whether to leave the captive stream open
/// when the ParallelDeflateOutputStream is closed.
///
///
/// See the
/// constructor for example code.
///
/// The stream to which compressed data will be written.
/// A tuning knob to trade speed for effectiveness.
///
/// true if the application would like the stream to remain open after inflation/deflation.
///
public ParallelDeflateOutputStream(System.IO.Stream stream, CompressionLevel level, bool leaveOpen)
: this(stream, CompressionLevel.Default, CompressionStrategy.Default, leaveOpen)
{
}
///
/// Create a ParallelDeflateOutputStream using the specified
/// CompressionLevel and CompressionStrategy, and specifying whether to
/// leave the captive stream open when the ParallelDeflateOutputStream is
/// closed.
///
///
/// See the
/// constructor for example code.
///
/// The stream to which compressed data will be written.
/// A tuning knob to trade speed for effectiveness.
///
/// By tweaking this parameter, you may be able to optimize the compression for
/// data with particular characteristics.
///
///
/// true if the application would like the stream to remain open after inflation/deflation.
///
public ParallelDeflateOutputStream(System.IO.Stream stream,
CompressionLevel level,
CompressionStrategy strategy,
bool leaveOpen)
{
TraceOutput(TraceBits.Lifecycle | TraceBits.Session, "-------------------------------------------------------");
TraceOutput(TraceBits.Lifecycle | TraceBits.Session, "Create {0:X8}", this.GetHashCode());
_outStream = stream;
_compressLevel= level;
Strategy = strategy;
_leaveOpen = leaveOpen;
this.MaxBufferPairs = 16; // default
}
///
/// The ZLIB strategy to be used during compression.
///
///
public CompressionStrategy Strategy
{
get;
private set;
}
///
/// The maximum number of buffer pairs to use.
///
///
///
///
/// This property sets an upper limit on the number of memory buffer
/// pairs to create. The implementation of this stream allocates
/// multiple buffers to facilitate parallel compression. As each buffer
/// fills up, this stream uses
/// ThreadPool.QueueUserWorkItem()
/// to compress those buffers in a background threadpool thread. After a
/// buffer is compressed, it is re-ordered and written to the output
/// stream.
///
///
///
/// A higher number of buffer pairs enables a higher degree of
/// parallelism, which tends to increase the speed of compression on
/// multi-cpu computers. On the other hand, a higher number of buffer
/// pairs also implies a larger memory consumption, more active worker
/// threads, and a higher cpu utilization for any compression. This
/// property enables the application to limit its memory consumption and
/// CPU utilization behavior depending on requirements.
///
///
///
/// For each compression "task" that occurs in parallel, there are 2
/// buffers allocated: one for input and one for output. This property
/// sets a limit for the number of pairs. The total amount of storage
/// space allocated for buffering will then be (N*S*2), where N is the
/// number of buffer pairs, S is the size of each buffer (). By default, DotNetZip allocates 4 buffer
/// pairs per CPU core, so if your machine has 4 cores, and you retain
/// the default buffer size of 128k, then the
/// ParallelDeflateOutputStream will use 4 * 4 * 2 * 128kb of buffer
/// memory in total, or 4mb, in blocks of 128kb. If you then set this
/// property to 8, then the number will be 8 * 2 * 128kb of buffer
/// memory, or 2mb.
///
///
///
/// CPU utilization will also go up with additional buffers, because a
/// larger number of buffer pairs allows a larger number of background
/// threads to compress in parallel. If you find that parallel
/// compression is consuming too much memory or CPU, you can adjust this
/// value downward.
///
///
///
/// The default value is 16. Different values may deliver better or
/// worse results, depending on your priorities and the dynamic
/// performance characteristics of your storage and compute resources.
///
///
///
/// This property is not the number of buffer pairs to use; it is an
/// upper limit. An illustration: Suppose you have an application that
/// uses the default value of this property (which is 16), and it runs
/// on a machine with 2 CPU cores. In that case, DotNetZip will allocate
/// 4 buffer pairs per CPU core, for a total of 8 pairs. The upper
/// limit specified by this property has no effect.
///
///
///
/// The application can set this value at any time, but it is effective
/// only before the first call to Write(), which is when the buffers are
/// allocated.
///
///
public int MaxBufferPairs
{
get
{
return _maxBufferPairs;
}
set
{
if (value < 4)
throw new ArgumentException("MaxBufferPairs",
"Value must be 4 or greater.");
_maxBufferPairs = value;
}
}
///
/// The size of the buffers used by the compressor threads.
///
///
///
///
/// The default buffer size is 128k. The application can set this value
/// at any time, but it is effective only before the first Write().
///
///
///
/// Larger buffer sizes implies larger memory consumption but allows
/// more efficient compression. Using smaller buffer sizes consumes less
/// memory but may result in less effective compression. For example,
/// using the default buffer size of 128k, the compression delivered is
/// within 1% of the compression delivered by the single-threaded . On the other hand, using a
/// BufferSize of 8k can result in a compressed data stream that is 5%
/// larger than that delivered by the single-threaded
/// DeflateStream. Excessively small buffer sizes can also cause
/// the speed of the ParallelDeflateOutputStream to drop, because of
/// larger thread scheduling overhead dealing with many many small
/// buffers.
///
///
///
/// The total amount of storage space allocated for buffering will be
/// (N*S*2), where N is the number of buffer pairs, and S is the size of
/// each buffer (this property). There are 2 buffers used by the
/// compressor, one for input and one for output. By default, DotNetZip
/// allocates 4 buffer pairs per CPU core, so if your machine has 4
/// cores, then the number of buffer pairs used will be 16. If you
/// accept the default value of this property, 128k, then the
/// ParallelDeflateOutputStream will use 16 * 2 * 128kb of buffer memory
/// in total, or 4mb, in blocks of 128kb. If you set this property to
/// 64kb, then the number will be 16 * 2 * 64kb of buffer memory, or
/// 2mb.
///
///
///
public int BufferSize
{
get { return _bufferSize;}
set
{
if (value < 1024)
throw new ArgumentOutOfRangeException("BufferSize",
"BufferSize must be greater than 1024 bytes");
_bufferSize = value;
}
}
///
/// The CRC32 for the data that was written out, prior to compression.
///
///
/// This value is meaningful only after a call to Close().
///
public int Crc32 { get { return _Crc32; } }
///
/// The total number of uncompressed bytes processed by the ParallelDeflateOutputStream.
///
///
/// This value is meaningful only after a call to Close().
///
public Int64 BytesProcessed { get { return _totalBytesProcessed; } }
private void _InitializePoolOfWorkItems()
{
_toWrite = new Queue();
_toFill = new Queue();
_pool = new System.Collections.Generic.List();
int nTasks = BufferPairsPerCore * Environment.ProcessorCount;
nTasks = Math.Min(nTasks, _maxBufferPairs);
for(int i=0; i < nTasks; i++)
{
_pool.Add(new WorkItem(_bufferSize, _compressLevel, Strategy, i));
_toFill.Enqueue(i);
}
_newlyCompressedBlob = new AutoResetEvent(false);
_runningCrc = new Ionic.Crc.CRC32();
_currentlyFilling = -1;
_lastFilled = -1;
_lastWritten = -1;
_latestCompressed = -1;
}
///
/// Write data to the stream.
///
///
///
///
///
/// To use the ParallelDeflateOutputStream to compress data, create a
/// ParallelDeflateOutputStream with CompressionMode.Compress, passing a
/// writable output stream. Then call Write() on that
/// ParallelDeflateOutputStream, providing uncompressed data as input. The
/// data sent to the output stream will be the compressed form of the data
/// written.
///
///
///
/// To decompress data, use the class.
///
///
///
/// The buffer holding data to write to the stream.
/// the offset within that data array to find the first byte to write.
/// the number of bytes to write.
public override void Write(byte[] buffer, int offset, int count)
{
bool mustWait = false;
// This method does this:
// 0. handles any pending exceptions
// 1. write any buffers that are ready to be written,
// 2. fills a work buffer; when full, flip state to 'Filled',
// 3. if more data to be written, goto step 1
if (_isClosed)
throw new InvalidOperationException();
// dispense any exceptions that occurred on the BG threads
if (_pendingException != null)
{
_handlingException = true;
var pe = _pendingException;
_pendingException = null;
throw pe;
}
if (count == 0) return;
if (!_firstWriteDone)
{
// Want to do this on first Write, first session, and not in the
// constructor. We want to allow MaxBufferPairs to
// change after construction, but before first Write.
_InitializePoolOfWorkItems();
_firstWriteDone = true;
}
do
{
// may need to make buffers available
EmitPendingBuffers(false, mustWait);
mustWait = false;
// use current buffer, or get a new buffer to fill
int ix = -1;
if (_currentlyFilling >= 0)
{
ix = _currentlyFilling;
TraceOutput(TraceBits.WriteTake,
"Write notake wi({0}) lf({1})",
ix,
_lastFilled);
}
else
{
TraceOutput(TraceBits.WriteTake, "Write take?");
if (_toFill.Count == 0)
{
// no available buffers, so... need to emit
// compressed buffers.
mustWait = true;
continue;
}
ix = _toFill.Dequeue();
TraceOutput(TraceBits.WriteTake,
"Write take wi({0}) lf({1})",
ix,
_lastFilled);
++_lastFilled; // TODO: consider rollover?
}
WorkItem workitem = _pool[ix];
int limit = ((workitem.buffer.Length - workitem.inputBytesAvailable) > count)
? count
: (workitem.buffer.Length - workitem.inputBytesAvailable);
workitem.ordinal = _lastFilled;
TraceOutput(TraceBits.Write,
"Write lock wi({0}) ord({1}) iba({2})",
workitem.index,
workitem.ordinal,
workitem.inputBytesAvailable
);
// copy from the provided buffer to our workitem, starting at
// the tail end of whatever data we might have in there currently.
Buffer.BlockCopy(buffer,
offset,
workitem.buffer,
workitem.inputBytesAvailable,
limit);
count -= limit;
offset += limit;
workitem.inputBytesAvailable += limit;
if (workitem.inputBytesAvailable == workitem.buffer.Length)
{
// No need for interlocked.increment: the Write()
// method is documented as not multi-thread safe, so
// we can assume Write() calls come in from only one
// thread.
TraceOutput(TraceBits.Write,
"Write QUWI wi({0}) ord({1}) iba({2}) nf({3})",
workitem.index,
workitem.ordinal,
workitem.inputBytesAvailable );
if (!ThreadPool.QueueUserWorkItem( _DeflateOne, workitem ))
throw new Exception("Cannot enqueue workitem");
_currentlyFilling = -1; // will get a new buffer next time
}
else
_currentlyFilling = ix;
if (count > 0)
TraceOutput(TraceBits.WriteEnter, "Write more");
}
while (count > 0); // until no more to write
TraceOutput(TraceBits.WriteEnter, "Write exit");
return;
}
private void _FlushFinish()
{
// After writing a series of compressed buffers, each one closed
// with Flush.Sync, we now write the final one as Flush.Finish,
// and then stop.
byte[] buffer = new byte[128];
var compressor = new ZlibCodec();
int rc = compressor.InitializeDeflate(_compressLevel, false);
compressor.InputBuffer = null;
compressor.NextIn = 0;
compressor.AvailableBytesIn = 0;
compressor.OutputBuffer = buffer;
compressor.NextOut = 0;
compressor.AvailableBytesOut = buffer.Length;
rc = compressor.Deflate(FlushType.Finish);
if (rc != ZlibConstants.Z_STREAM_END && rc != ZlibConstants.Z_OK)
throw new Exception("deflating: " + compressor.Message);
if (buffer.Length - compressor.AvailableBytesOut > 0)
{
TraceOutput(TraceBits.EmitBegin,
"Emit begin flush bytes({0})",
buffer.Length - compressor.AvailableBytesOut);
_outStream.Write(buffer, 0, buffer.Length - compressor.AvailableBytesOut);
TraceOutput(TraceBits.EmitDone,
"Emit done flush");
}
compressor.EndDeflate();
_Crc32 = _runningCrc.Crc32Result;
}
private void _Flush(bool lastInput)
{
if (_isClosed)
throw new InvalidOperationException();
if (emitting) return;
// compress any partial buffer
if (_currentlyFilling >= 0)
{
WorkItem workitem = _pool[_currentlyFilling];
_DeflateOne(workitem);
_currentlyFilling = -1; // get a new buffer next Write()
}
if (lastInput)
{
EmitPendingBuffers(true, false);
_FlushFinish();
}
else
{
EmitPendingBuffers(false, false);
}
}
///
/// Flush the stream.
///
public override void Flush()
{
if (_pendingException != null)
{
_handlingException = true;
var pe = _pendingException;
_pendingException = null;
throw pe;
}
if (_handlingException)
return;
_Flush(false);
}
///
/// Close the stream.
///
///
/// You must call Close on the stream to guarantee that all of the data written in has
/// been compressed, and the compressed data has been written out.
///
public override void Close()
{
TraceOutput(TraceBits.Session, "Close {0:X8}", this.GetHashCode());
if (_pendingException != null)
{
_handlingException = true;
var pe = _pendingException;
_pendingException = null;
throw pe;
}
if (_handlingException)
return;
if (_isClosed) return;
_Flush(true);
if (!_leaveOpen)
_outStream.Close();
_isClosed= true;
}
// workitem 10030 - implement a new Dispose method
/// Dispose the object
///
///
/// Because ParallelDeflateOutputStream is IDisposable, the
/// application must call this method when finished using the instance.
///
///
/// This method is generally called implicitly upon exit from
/// a using scope in C# (Using in VB).
///
///
new public void Dispose()
{
TraceOutput(TraceBits.Lifecycle, "Dispose {0:X8}", this.GetHashCode());
Close();
_pool = null;
Dispose(true);
}
/// The Dispose method
///
/// indicates whether the Dispose method was invoked by user code.
///
protected override void Dispose(bool disposing)
{
base.Dispose(disposing);
}
///
/// Resets the stream for use with another stream.
///
///
/// Because the ParallelDeflateOutputStream is expensive to create, it
/// has been designed so that it can be recycled and re-used. You have
/// to call Close() on the stream first, then you can call Reset() on
/// it, to use it again on another stream.
///
///
///
/// The new output stream for this era.
///
///
///
///
/// ParallelDeflateOutputStream deflater = null;
/// foreach (var inputFile in listOfFiles)
/// {
/// string outputFile = inputFile + ".compressed";
/// using (System.IO.Stream input = System.IO.File.OpenRead(inputFile))
/// {
/// using (var outStream = System.IO.File.Create(outputFile))
/// {
/// if (deflater == null)
/// deflater = new ParallelDeflateOutputStream(outStream,
/// CompressionLevel.Best,
/// CompressionStrategy.Default,
/// true);
/// deflater.Reset(outStream);
///
/// while ((n= input.Read(buffer, 0, buffer.Length)) != 0)
/// {
/// deflater.Write(buffer, 0, n);
/// }
/// }
/// }
/// }
///
///
public void Reset(Stream stream)
{
TraceOutput(TraceBits.Session, "-------------------------------------------------------");
TraceOutput(TraceBits.Session, "Reset {0:X8} firstDone({1})", this.GetHashCode(), _firstWriteDone);
if (!_firstWriteDone) return;
// reset all status
_toWrite.Clear();
_toFill.Clear();
foreach (var workitem in _pool)
{
_toFill.Enqueue(workitem.index);
workitem.ordinal = -1;
}
_firstWriteDone = false;
_totalBytesProcessed = 0L;
_runningCrc = new Ionic.Crc.CRC32();
_isClosed= false;
_currentlyFilling = -1;
_lastFilled = -1;
_lastWritten = -1;
_latestCompressed = -1;
_outStream = stream;
}
private void EmitPendingBuffers(bool doAll, bool mustWait)
{
// When combining parallel deflation with a ZipSegmentedStream, it's
// possible for the ZSS to throw from within this method. In that
// case, Close/Dispose will be called on this stream, if this stream
// is employed within a using or try/finally pair as required. But
// this stream is unaware of the pending exception, so the Close()
// method invokes this method AGAIN. This can lead to a deadlock.
// Therefore, failfast if re-entering.
if (emitting) return;
emitting = true;
if (doAll || mustWait)
_newlyCompressedBlob.WaitOne();
do
{
int firstSkip = -1;
int millisecondsToWait = doAll ? 200 : (mustWait ? -1 : 0);
int nextToWrite = -1;
do
{
if (Monitor.TryEnter(_toWrite, millisecondsToWait))
{
nextToWrite = -1;
try
{
if (_toWrite.Count > 0)
nextToWrite = _toWrite.Dequeue();
}
finally
{
Monitor.Exit(_toWrite);
}
if (nextToWrite >= 0)
{
WorkItem workitem = _pool[nextToWrite];
if (workitem.ordinal != _lastWritten + 1)
{
// out of order. requeue and try again.
TraceOutput(TraceBits.EmitSkip,
"Emit skip wi({0}) ord({1}) lw({2}) fs({3})",
workitem.index,
workitem.ordinal,
_lastWritten,
firstSkip);
lock(_toWrite)
{
_toWrite.Enqueue(nextToWrite);
}
if (firstSkip == nextToWrite)
{
// We went around the list once.
// None of the items in the list is the one we want.
// Now wait for a compressor to signal again.
_newlyCompressedBlob.WaitOne();
firstSkip = -1;
}
else if (firstSkip == -1)
firstSkip = nextToWrite;
continue;
}
firstSkip = -1;
TraceOutput(TraceBits.EmitBegin,
"Emit begin wi({0}) ord({1}) cba({2})",
workitem.index,
workitem.ordinal,
workitem.compressedBytesAvailable);
_outStream.Write(workitem.compressed, 0, workitem.compressedBytesAvailable);
_runningCrc.Combine(workitem.crc, workitem.inputBytesAvailable);
_totalBytesProcessed += workitem.inputBytesAvailable;
workitem.inputBytesAvailable = 0;
TraceOutput(TraceBits.EmitDone,
"Emit done wi({0}) ord({1}) cba({2}) mtw({3})",
workitem.index,
workitem.ordinal,
workitem.compressedBytesAvailable,
millisecondsToWait);
_lastWritten = workitem.ordinal;
_toFill.Enqueue(workitem.index);
// don't wait next time through
if (millisecondsToWait == -1) millisecondsToWait = 0;
}
}
else
nextToWrite = -1;
} while (nextToWrite >= 0);
//} while (doAll && (_lastWritten != _latestCompressed));
} while (doAll && (_lastWritten != _latestCompressed || _lastWritten != _lastFilled));
emitting = false;
}
#if OLD
private void _PerpetualWriterMethod(object state)
{
TraceOutput(TraceBits.WriterThread, "_PerpetualWriterMethod START");
try
{
do
{
// wait for the next session
TraceOutput(TraceBits.Synch | TraceBits.WriterThread, "Synch _sessionReset.WaitOne(begin) PWM");
_sessionReset.WaitOne();
TraceOutput(TraceBits.Synch | TraceBits.WriterThread, "Synch _sessionReset.WaitOne(done) PWM");
if (_isDisposed) break;
TraceOutput(TraceBits.Synch | TraceBits.WriterThread, "Synch _sessionReset.Reset() PWM");
_sessionReset.Reset();
// repeatedly write buffers as they become ready
WorkItem workitem = null;
Ionic.Zlib.CRC32 c= new Ionic.Zlib.CRC32();
do
{
workitem = _pool[_nextToWrite % _pc];
lock(workitem)
{
if (_noMoreInputForThisSegment)
TraceOutput(TraceBits.Write,
"Write drain wi({0}) stat({1}) canuse({2}) cba({3})",
workitem.index,
workitem.status,
(workitem.status == (int)WorkItem.Status.Compressed),
workitem.compressedBytesAvailable);
do
{
if (workitem.status == (int)WorkItem.Status.Compressed)
{
TraceOutput(TraceBits.WriteBegin,
"Write begin wi({0}) stat({1}) cba({2})",
workitem.index,
workitem.status,
workitem.compressedBytesAvailable);
workitem.status = (int)WorkItem.Status.Writing;
_outStream.Write(workitem.compressed, 0, workitem.compressedBytesAvailable);
c.Combine(workitem.crc, workitem.inputBytesAvailable);
_totalBytesProcessed += workitem.inputBytesAvailable;
_nextToWrite++;
workitem.inputBytesAvailable= 0;
workitem.status = (int)WorkItem.Status.Done;
TraceOutput(TraceBits.WriteDone,
"Write done wi({0}) stat({1}) cba({2})",
workitem.index,
workitem.status,
workitem.compressedBytesAvailable);
Monitor.Pulse(workitem);
break;
}
else
{
int wcycles = 0;
// I've locked a workitem I cannot use.
// Therefore, wake someone else up, and then release the lock.
while (workitem.status != (int)WorkItem.Status.Compressed)
{
TraceOutput(TraceBits.WriteWait,
"Write waiting wi({0}) stat({1}) nw({2}) nf({3}) nomore({4})",
workitem.index,
workitem.status,
_nextToWrite, _nextToFill,
_noMoreInputForThisSegment );
if (_noMoreInputForThisSegment && _nextToWrite == _nextToFill)
break;
wcycles++;
// wake up someone else
Monitor.Pulse(workitem);
// release and wait
Monitor.Wait(workitem);
if (workitem.status == (int)WorkItem.Status.Compressed)
TraceOutput(TraceBits.WriteWait,
"Write A-OK wi({0}) stat({1}) iba({2}) cba({3}) cyc({4})",
workitem.index,
workitem.status,
workitem.inputBytesAvailable,
workitem.compressedBytesAvailable,
wcycles);
}
if (_noMoreInputForThisSegment && _nextToWrite == _nextToFill)
break;
}
}
while (true);
}
if (_noMoreInputForThisSegment)
TraceOutput(TraceBits.Write,
"Write nomore nw({0}) nf({1}) break({2})",
_nextToWrite, _nextToFill, (_nextToWrite == _nextToFill));
if (_noMoreInputForThisSegment && _nextToWrite == _nextToFill)
break;
} while (true);
// Finish:
// After writing a series of buffers, closing each one with
// Flush.Sync, we now write the final one as Flush.Finish, and
// then stop.
byte[] buffer = new byte[128];
ZlibCodec compressor = new ZlibCodec();
int rc = compressor.InitializeDeflate(_compressLevel, false);
compressor.InputBuffer = null;
compressor.NextIn = 0;
compressor.AvailableBytesIn = 0;
compressor.OutputBuffer = buffer;
compressor.NextOut = 0;
compressor.AvailableBytesOut = buffer.Length;
rc = compressor.Deflate(FlushType.Finish);
if (rc != ZlibConstants.Z_STREAM_END && rc != ZlibConstants.Z_OK)
throw new Exception("deflating: " + compressor.Message);
if (buffer.Length - compressor.AvailableBytesOut > 0)
{
TraceOutput(TraceBits.WriteBegin,
"Write begin flush bytes({0})",
buffer.Length - compressor.AvailableBytesOut);
_outStream.Write(buffer, 0, buffer.Length - compressor.AvailableBytesOut);
TraceOutput(TraceBits.WriteBegin,
"Write done flush");
}
compressor.EndDeflate();
_Crc32 = c.Crc32Result;
// signal that writing is complete:
TraceOutput(TraceBits.Synch, "Synch _writingDone.Set() PWM");
_writingDone.Set();
}
while (true);
}
catch (System.Exception exc1)
{
lock(_eLock)
{
// expose the exception to the main thread
if (_pendingException!=null)
_pendingException = exc1;
}
}
TraceOutput(TraceBits.WriterThread, "_PerpetualWriterMethod FINIS");
}
#endif
private void _DeflateOne(Object wi)
{
// compress one buffer
WorkItem workitem = (WorkItem) wi;
try
{
int myItem = workitem.index;
Ionic.Crc.CRC32 crc = new Ionic.Crc.CRC32();
// calc CRC on the buffer
crc.SlurpBlock(workitem.buffer, 0, workitem.inputBytesAvailable);
// deflate it
DeflateOneSegment(workitem);
// update status
workitem.crc = crc.Crc32Result;
TraceOutput(TraceBits.Compress,
"Compress wi({0}) ord({1}) len({2})",
workitem.index,
workitem.ordinal,
workitem.compressedBytesAvailable
);
lock(_latestLock)
{
if (workitem.ordinal > _latestCompressed)
_latestCompressed = workitem.ordinal;
}
lock (_toWrite)
{
_toWrite.Enqueue(workitem.index);
}
_newlyCompressedBlob.Set();
}
catch (System.Exception exc1)
{
lock(_eLock)
{
// expose the exception to the main thread
if (_pendingException!=null)
_pendingException = exc1;
}
}
}
private bool DeflateOneSegment(WorkItem workitem)
{
ZlibCodec compressor = workitem.compressor;
int rc= 0;
compressor.ResetDeflate();
compressor.NextIn = 0;
compressor.AvailableBytesIn = workitem.inputBytesAvailable;
// step 1: deflate the buffer
compressor.NextOut = 0;
compressor.AvailableBytesOut = workitem.compressed.Length;
do
{
compressor.Deflate(FlushType.None);
}
while (compressor.AvailableBytesIn > 0 || compressor.AvailableBytesOut == 0);
// step 2: flush (sync)
rc = compressor.Deflate(FlushType.Sync);
workitem.compressedBytesAvailable= (int) compressor.TotalBytesOut;
return true;
}
[System.Diagnostics.ConditionalAttribute("Trace")]
private void TraceOutput(TraceBits bits, string format, params object[] varParams)
{
if ((bits & _DesiredTrace) != 0)
{
lock(_outputLock)
{
int tid = Thread.CurrentThread.GetHashCode();
#if !SILVERLIGHT
Console.ForegroundColor = (ConsoleColor) (tid % 8 + 8);
#endif
Console.Write("{0:000} PDOS ", tid);
Console.WriteLine(format, varParams);
#if !SILVERLIGHT
Console.ResetColor();
#endif
}
}
}
// used only when Trace is defined
[Flags]
enum TraceBits : uint
{
None = 0,
NotUsed1 = 1,
EmitLock = 2,
EmitEnter = 4, // enter _EmitPending
EmitBegin = 8, // begin to write out
EmitDone = 16, // done writing out
EmitSkip = 32, // writer skipping a workitem
EmitAll = 58, // All Emit flags
Flush = 64,
Lifecycle = 128, // constructor/disposer
Session = 256, // Close/Reset
Synch = 512, // thread synchronization
Instance = 1024, // instance settings
Compress = 2048, // compress task
Write = 4096, // filling buffers, when caller invokes Write()
WriteEnter = 8192, // upon entry to Write()
WriteTake = 16384, // on _toFill.Take()
All = 0xffffffff,
}
///
/// Indicates whether the stream supports Seek operations.
///
///
/// Always returns false.
///
public override bool CanSeek
{
get { return false; }
}
///
/// Indicates whether the stream supports Read operations.
///
///
/// Always returns false.
///
public override bool CanRead
{
get {return false;}
}
///
/// Indicates whether the stream supports Write operations.
///
///
/// Returns true if the provided stream is writable.
///
public override bool CanWrite
{
get { return _outStream.CanWrite; }
}
///
/// Reading this property always throws a NotSupportedException.
///
public override long Length
{
get { throw new NotSupportedException(); }
}
///
/// Returns the current position of the output stream.
///
///
///
/// Because the output gets written by a background thread,
/// the value may change asynchronously. Setting this
/// property always throws a NotSupportedException.
///
///
public override long Position
{
get { return _outStream.Position; }
set { throw new NotSupportedException(); }
}
///
/// This method always throws a NotSupportedException.
///
///
/// The buffer into which data would be read, IF THIS METHOD
/// ACTUALLY DID ANYTHING.
///
///
/// The offset within that data array at which to insert the
/// data that is read, IF THIS METHOD ACTUALLY DID
/// ANYTHING.
///
///
/// The number of bytes to write, IF THIS METHOD ACTUALLY DID
/// ANYTHING.
///
/// nothing.
public override int Read(byte[] buffer, int offset, int count)
{
throw new NotSupportedException();
}
///
/// This method always throws a NotSupportedException.
///
///
/// The offset to seek to....
/// IF THIS METHOD ACTUALLY DID ANYTHING.
///
///
/// The reference specifying how to apply the offset.... IF
/// THIS METHOD ACTUALLY DID ANYTHING.
///
/// nothing. It always throws.
public override long Seek(long offset, System.IO.SeekOrigin origin)
{
throw new NotSupportedException();
}
///
/// This method always throws a NotSupportedException.
///
///
/// The new value for the stream length.... IF
/// THIS METHOD ACTUALLY DID ANYTHING.
///
public override void SetLength(long value)
{
throw new NotSupportedException();
}
}
}