/*
Copyright (c) 2013, Michael Kazhdan
All rights reserved.

Redistribution and use in source and binary forms, with or without modification,
are permitted provided that the following conditions are met:

Redistributions of source code must retain the above copyright notice, this list of
conditions and the following disclaimer. Redistributions in binary form must reproduce
the above copyright notice, this list of conditions and the following disclaimer
in the documentation and/or other materials provided with the distribution. 

Neither the name of the Johns Hopkins University nor the names of its contributors
may be used to endorse or promote products derived from this software without specific
prior written permission. 

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY
EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO THE IMPLIED WARRANTIES 
OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT
SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
TO, PROCUREMENT OF SUBSTITUTE  GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
DAMAGE.
*/

#include "PreProcessor.h"

#undef USE_DOUBLE
#define DIMENSION 2
#define ROW_BLOCK_SIZE 16
#define DEFAULT_FEM_DEGREE 1

#ifndef USE_DEEP_TREE_NODES
#define USE_DEEP_TREE_NODES
#endif // USE_DEEP_TREE_NODES

#include <stdio.h>
#include <stdlib.h>
#include <float.h>
#include <algorithm>
#include "Image.h"
#include "MyMiscellany.h"
#include "Array.h"
#include "CmdLineParser.h"
#include "Geometry.h"
#include "FEMTree.h"

cmdLineParameterArray< char* , 2 >
	In( "in" );
cmdLineParameter< char* >
	Out( "out" );
cmdLineParameter< int >
#ifdef FAST_COMPILE
#else // !FAST_COMPILE
	Degree( "degree" , DEFAULT_FEM_DEGREE ) ,
#endif // FAST_COMPILE
	ParallelType( "parallel" , (int)ThreadPool::OPEN_MP ) ,
	ScheduleType( "schedule" , (int)ThreadPool::DefaultSchedule ) ,
	ThreadChunkSize( "chunkSize" , (int)ThreadPool::DefaultChunkSize ) ,
	Threads( "threads" , (int)std::thread::hardware_concurrency() ) ,
	MaxMemoryGB( "maxMemory" , 0 ) ,
	GSIterations( "iters" , 8 ) ,
	FullDepth( "fullDepth" , 6 ) ,
	BaseDepth( "baseDepth" , 6 ) ,
	BaseVCycles( "baseVCycles" , 4 );
cmdLineReadable
	Verbose( "verbose" ) ,
	ShowResidual( "residual" ) ,
	Performance( "performance" );
cmdLineParameter< float >
	WeightScale   ( "wScl", 0.125f ) ,
	WeightExponent( "wExp" , 6.f );

cmdLineReadable* params[] =
{
	&In , &Out , &Threads , &Verbose , &ShowResidual , &GSIterations , &FullDepth ,
	&BaseDepth , &BaseVCycles ,
	&WeightScale , &WeightExponent ,
	&Performance ,
	&MaxMemoryGB ,
#if !defined( FAST_COMPILE )
	&Degree , 
#endif // !FAST_COMPILE
	&ParallelType ,
	&ScheduleType ,
	&ThreadChunkSize ,
	NULL
};

void ShowUsage( char* ex )
{
	printf( "Usage: %s\n" , ex );
	printf( "\t --%s <input color / labels>\n" , In.name );
	printf( "\t[--%s <ouput stitched image>]\n" , Out.name );
#if !defined( FAST_COMPILE )
	printf( "\t[--%s <b-spline degree>=%d]\n" , Degree.name , Degree.value );
#endif // !FAST_COMPILE
	printf( "\t[--%s <GS iterations>=%d]\n" , GSIterations.name , GSIterations.value );
	printf( "\t[--%s <full depth>=%d]\n" , FullDepth.name , FullDepth.value );
	printf( "\t[--%s <num threads>=%d]\n" , Threads.name , Threads.value );
	printf( "\t[--%s <parallel type>=%d]\n" , ParallelType.name , ParallelType.value );
	for( size_t i=0 ; i<ThreadPool::ParallelNames.size() ; i++ ) printf( "\t\t%d] %s\n" , (int)i , ThreadPool::ParallelNames[i].c_str() );
	printf( "\t[--%s <schedue type>=%d]\n" , ScheduleType.name , ScheduleType.value );
	for( size_t i=0 ; i<ThreadPool::ScheduleNames.size() ; i++ ) printf( "\t\t%d] %s\n" , (int)i , ThreadPool::ScheduleNames[i].c_str() );
	printf( "\t[--%s <thread chunk size>=%d]\n" , ThreadChunkSize.name , ThreadChunkSize.value );
	printf( "\t[--%s <successive under-relaxation scale>=%f]\n", WeightScale.name , WeightScale.value );
	printf( "\t[--%s <successive under-relaxation exponent>=%f]\n", WeightExponent.name , WeightExponent.value );
	printf( "\t[--%s <maximum memory (in GB)>=%d]\n" , MaxMemoryGB.name , MaxMemoryGB.value );
	printf( "\t[--%s]\n" , Performance.name );
	printf( "\t[--%s <coarse MG solver depth>=%d]\n" , BaseDepth.name , BaseDepth.value );
	printf( "\t[--%s <coarse MG solver v-cycles>=%d]\n" , BaseVCycles.name , BaseVCycles.value );
	printf( "\t[--%s]\n" , ShowResidual.name );
	printf( "\t[--%s]\n" , Verbose.name );
}

struct RGBPixel
{
	unsigned char rgb[3];
	unsigned char& operator[]( int idx ){ return rgb[idx]; }
	const unsigned char& operator[]( int idx ) const { return rgb[idx]; }
	int mask( void ) const
	{
		if( rgb[0]==255 && rgb[1]==255 && rgb[2]==255 ) return -1;
		else return ( (int)rgb[0] )<<16 | ( (int)rgb[1] )<<8 | ( (int)rgb[2] );
	}

	RGBPixel( void ){ rgb[0] = rgb[1] = rgb[2] = 0; }
	RGBPixel( double r , double g , double b )
	{
		rgb[0] = (unsigned char)( std::max< int >( 0 , std::min< int >( 255 , (int)( r*255 ) ) ) );
		rgb[1] = (unsigned char)( std::max< int >( 0 , std::min< int >( 255 , (int)( g*255 ) ) ) );
		rgb[2] = (unsigned char)( std::max< int >( 0 , std::min< int >( 255 , (int)( b*255 ) ) ) );
	}
	RGBPixel( float r , float g , float b )
	{
		rgb[0] = (unsigned char)( std::max< int >( 0 , std::min< int >( 255 , (int)( r*255 ) ) ) );
		rgb[1] = (unsigned char)( std::max< int >( 0 , std::min< int >( 255 , (int)( g*255 ) ) ) );
		rgb[2] = (unsigned char)( std::max< int >( 0 , std::min< int >( 255 , (int)( b*255 ) ) ) );
	}

	template< class Real >
	static Point< Real , 3 > ToPoint( RGBPixel rgb )
	{
		Point< Real , 3 > p;
		for( int c=0 ; c<3 ; c++ ) p[c] = (Real)( ( (double)rgb[c] ) / 255. );
		return p;
	}
};

void WriteImage( char* fileName , RGBPixel* pixels , int w , int h )
{
	unsigned int _w = w , _h = h , _c = 3;
	ImageWriter::Write( fileName , (const unsigned char*)pixels , _w , _h , _c );
}

struct Profiler
{
	double t;
	Profiler( void ){ t = Time(); }
	void print( bool newLine=false ) const
	{
		printf( "%.2f (s) ; %d (MB)" , Time()-t , MemoryInfo::PeakMemoryUsageMB() );
		if( newLine ) printf( "\n" );
	}
};

template< unsigned int Colors >
void ReadAndWrite( ImageReader* pixels , ImageReader* labels , ImageWriter* output )
{
	RGBPixel* pixelRow = new RGBPixel[ pixels->width() ];
	RGBPixel* labelRow = new RGBPixel[ labels->width() ];
	for( unsigned int r=0 ; r<pixels->height() ; r++ )
	{
		if( Verbose.set ) printf( "%d / %d       \r" , r ,pixels->height() );
		pixels->nextRow( (unsigned char*)pixelRow );
		labels->nextRow( (unsigned char*)labelRow );
		output->nextRow( (unsigned char*)pixelRow );
	}
	if( Verbose.set ) printf( "\n" );
	delete[] pixelRow;
	delete[] labelRow;
}

template< class Real , unsigned int Colors >
struct BufferedImageDerivativeStream : public FEMTreeInitializer< DIMENSION , Real >::template DerivativeStream< Point< Real , Colors > >
{
	BufferedImageDerivativeStream( const unsigned int resolution[] , ImageReader* pixels , ImageReader* labels ) : _pixels( pixels ) , _labels( labels )
	{
		memcpy( _resolution , resolution , sizeof( unsigned int ) * DIMENSION );
		for( int i=0 ; i<3 ; i++ )
		{
			_pixelRows[i] = new RGBPixel[ _resolution[0] ];
			_labelRows[i] = new RGBPixel[ _resolution[0] ];
			_maskRows [i] = new      int[ _resolution[0] ];
		}
		if( pixels->channels()!=3 && pixels->channels()!=1 ) ERROR_OUT( "Pixel input must have 1 or 3 channels: " , pixels->channels() );
		if( labels->channels()!=3 && labels->channels()!=1 ) ERROR_OUT( "Label input must have 1 or 3 channels: " , labels->channels() );
		__pixelRow = pixels->channels()==3 ? NULL : new unsigned char[ _resolution[0] ];
		__labelRow = labels->channels()==3 ? NULL : new unsigned char[ _resolution[0] ];
		_r = -2 ; prefetch();
		_r = -1 ; prefetch();
		_c = _r = _dir = 0;
	}
	~BufferedImageDerivativeStream( void )
	{
		for( int i=0 ; i<3 ; i++ ) delete[] _pixelRows[i] , delete[] _labelRows[i] , delete[] _maskRows[i];
		if( __pixelRow ) delete[] __pixelRow;
		if( __labelRow ) delete[] __labelRow;
	}
	void resolution( unsigned int res[] ) const { memcpy( res , _resolution , sizeof(_resolution) ); }

	void advance( void ){ _c = _dir = 0 , _r++; }
	void prefetch( void )
	{
		if( _r+2<(int)_resolution[1] )
		{
			int j = (_r+2)%3;
			RGBPixel *pixelRow = _pixelRows[j] , *labelRow = _labelRows[j];
			int *maskRow = _maskRows[j];
			if( _pixels->channels()==3 ) _pixels->nextRow( (unsigned char*)pixelRow );
			else
			{
				_pixels->nextRow( __pixelRow );
				for( int i=0 ; i<(int)_resolution[0] ; i++ ) pixelRow[i][0] = pixelRow[i][1] = pixelRow[i][2] = __pixelRow[i];
			}
			if( _labels->channels()==3 ) _labels->nextRow( (unsigned char*)labelRow );
			else
			{
				_labels->nextRow( __labelRow );
				for( int i=0 ; i<(int)_resolution[0] ; i++ ) labelRow[i][0] = labelRow[i][1] = labelRow[i][2] = __labelRow[i];
			}
			ThreadPool::Parallel_for( 0 , _resolution[0] , [&]( unsigned int , size_t i ){ maskRow[i] = labelRow[i].mask(); } );
		}
	}

	bool nextDerivative( unsigned int idx[] , unsigned int& dir , Point< Real , Colors >& dValue )
	{
		const RGBPixel *pixelRow1 = _pixelRows[_r%3] , *pixelRow2 = _pixelRows[(_r+1)%3];
		const int *maskRow1 = _maskRows[_r%3] , *maskRow2 = _maskRows[(_r+1)%3];
		if( _dir==0 )
		{
			for( ; _c<(int)_resolution[0]-1 ; _c++ )
			{
				if( maskRow1[_c]!=maskRow1[_c+1] && maskRow1[_c]>=0 && maskRow1[_c+1]>=0 )
				{
					idx[0] = _c , idx[1] = _r , dir = _dir;
					dValue = RGBPixel::ToPoint< Real >( pixelRow1[_c+1] ) - RGBPixel::ToPoint< Real >( pixelRow1[_c] );
					_c++;
					return true;
				}
			}
			_dir = 1 , _c = 0;
		}
		if( _dir==1 )
		{
			if( _r+1<(int)_resolution[1] )
			{
				for( ; _c<(int)_resolution[0] ; _c++ )
				{
					if( maskRow1[_c]!=maskRow2[_c] && maskRow1[_c]>=0 && maskRow2[_c]>=0 )
					{
						idx[0] = _c , idx[1] = _r , dir = _dir;
						dValue = RGBPixel::ToPoint< Real >( pixelRow2[_c] ) - RGBPixel::ToPoint< Real >( pixelRow1[_c] );
						_c++;
						return true;
					}
				}
			}
		}
		return false;
	}
protected:
	int _r , _c , _dir;
	unsigned int _resolution[DIMENSION];
	ImageReader *_pixels , *_labels;
	RGBPixel *_pixelRows[3] , *_labelRows[3];
	unsigned char *__pixelRow , *__labelRow;
	int* _maskRows[3];
};

template< typename Real , unsigned int Degree >
void _Execute( void )
{
	ThreadPool::Init( (ThreadPool::ParallelType)ParallelType.value , Threads.value );
	int w , h;
	{
		unsigned int _w , _h , _c;
		ImageReader::GetInfo( In.values[0] , _w , _h , _c );
		w = _w , h = _h;
		ImageReader::GetInfo( In.values[1] , _w , _h , _c );
		if( w!=_w || h!=_h ) ERROR_OUT( "Pixel and label dimensions don't match: " , _w , " x " , _h , " != " , w , " x " , h );
	}
	if( Verbose.set ) printf( "Resolution: %d x %d\n" , w , h );

	static const unsigned int Dim = DIMENSION;
	static const unsigned int Colors = 3;
	static const unsigned int FEMSig = FEMDegreeAndBType< Degree , BOUNDARY_NEUMANN >::Signature;

	FEMTree< Dim , Real > tree( MEMORY_ALLOCATOR_BLOCK_SIZE );
	std::vector< NodeSample< Dim , Point< Real , Colors > > > derivatives[Dim];
	int maxDepth;
	DenseNodeData< Point< Real , Colors > , IsotropicUIntPack< Dim , FEMSig > > constraints;
	DenseNodeData< Point< Real , Colors > , IsotropicUIntPack< Dim , FEMSig > > solution;
	{
		Profiler p;
		ImageReader* pixels = ImageReader::Get( In.values[0] );
		ImageReader* labels = ImageReader::Get( In.values[1] );
		unsigned int resolution[] = { (unsigned int )w , (unsigned int )h };
		BufferedImageDerivativeStream< Real , Colors > dStream( resolution , pixels , labels );
		for( int j=0 ; j<h ; j++ )
		{
			ThreadPool::ParallelSections
				(
					[&]( void ){ dStream.prefetch(); } ,
					[&]( void ){ maxDepth = FEMTreeInitializer< Dim , Real >::template Initialize< (Degree&1)==0 , Point< Real , Colors > >( tree.spaceRoot() , dStream , derivatives , tree.nodeAllocators.size() ? tree.nodeAllocators[0] : NULL , tree.initializer() ); }
				);
			dStream.advance();
		}
		delete pixels;
		delete labels;
		{
			std::vector< typename FEMTree< Dim , Real >::FEMTreeNode* > nodes;
			nodes.reserve( derivatives[0].size() + derivatives[1].size() );
			for( int i=0 ; i<derivatives[0].size() ; i++ ) nodes.push_back( derivatives[0][i].node );
			for( int i=0 ; i<derivatives[1].size() ; i++ ) nodes.push_back( derivatives[1][i].node );
			tree.template thicken< 1 , 0 >( &nodes[0] , (int)nodes.size() );
		}
		tree.template finalizeForMultigrid< Degree >( FullDepth.value , []( const RegularTreeNode< Dim , FEMTreeNodeData , depth_and_offset_type >* ){ return true; } );
		if( Verbose.set )
		{
			printf( "Valid FEM Nodes / Edges: %llu %llu\n" , (unsigned long long)tree.validFEMNodes( IsotropicUIntPack< Dim , FEMSig >() ) , (unsigned long long)( derivatives[0].size() + derivatives[1].size() ) );
			printf( "Set tree [%d]: " , maxDepth ) , p.print( true );
		}
	}

	{
		Profiler p;
		constraints = tree.template initDenseNodeData< Point< Real , Colors > >( IsotropicUIntPack< Dim , FEMSig >() );
		static const unsigned int DFEMSig = FEMSignature< FEMSig >::DSignature();
		// Generate the partial-x constraints
		{
			typedef UIntPack< DFEMSig , FEMSig > CSignature;
			typedef IsotropicUIntPack< 2 , 0 > CDerivative;
			typedef UIntPack< FEMSig , FEMSig > FEMSignature;
			typedef UIntPack< 1 , 0 > FEMDerivative;
			SparseNodeData< Point< Real , Colors > , CSignature > partialX;
			for( int i=0 ; i<derivatives[0].size() ; i++ ) partialX[ derivatives[0][i].node ] = -derivatives[0][i].data * (1<<maxDepth);

			unsigned int derivatives1[] = { 1 , 0 } , derivatives2[] = { 0 , 0 };
			typename FEMIntegrator::template Constraint< FEMSignature , FEMDerivative , CSignature , CDerivative , 1 > F;
			F.weights[0][ TensorDerivatives< FEMDerivative >::Index( derivatives1 ) ][ TensorDerivatives< CDerivative >::Index( derivatives2 ) ] = 1;
			tree.addFEMConstraints( F , partialX , constraints , maxDepth );
		}
		// Generate the partial-y constraints
		{
			typedef UIntPack< FEMSig , DFEMSig > CSignature;
			typedef IsotropicUIntPack< 2 , 0 > CDerivative;
			typedef UIntPack< FEMSig , FEMSig > FEMSignature;
			typedef UIntPack< 0 , 1 > FEMDerivative;
			SparseNodeData< Point< Real , Colors > , CSignature > partialY;
			for( int i=0 ; i<derivatives[1].size() ; i++ ) partialY[ derivatives[1][i].node ] = -derivatives[1][i].data * (1<<maxDepth);

			unsigned int derivatives1[] = { 0 , 1 } , derivatives2[] = { 0 , 0 };
			typename FEMIntegrator::template Constraint< FEMSignature , FEMDerivative , CSignature , CDerivative , 1 > F;
			F.weights[0][ TensorDerivatives< FEMDerivative >::Index( derivatives1 ) ][ TensorDerivatives< CDerivative >::Index( derivatives2 ) ] = 1;
			tree.addFEMConstraints( F , partialY , constraints , maxDepth );
		}
		if( Verbose.set ) printf( "Set constraints: " ) , p.print( true );
	}
	// Solve the system
	{
		Profiler p;
		double t = Time();
		solution = tree.template initDenseNodeData< Point< Real , Colors > >( IsotropicUIntPack< Dim , FEMSig >() );
		typename FEMTree< Dim , Real >::SolverInfo sInfo;
		sInfo.cgDepth = 0 , sInfo.cascadic = false , sInfo.vCycles = 1 , sInfo.cgAccuracy = 0 , sInfo.verbose = Verbose.set , sInfo.showResidual = ShowResidual.set , sInfo.showGlobalResidual = false , sInfo.sliceBlockSize = ROW_BLOCK_SIZE;
		sInfo.baseDepth = BaseDepth.value , sInfo.baseVCycles = BaseVCycles.value;
		sInfo.iters = GSIterations.value;

		sInfo.useSupportWeights = true;
		sInfo.sorRestrictionFunction = [&] ( Real w , Real ){ return (Real)( WeightScale.value * pow( w , WeightExponent.value ) ); };
		sInfo.wCycle = false;
		typename FEMIntegrator::template System< IsotropicUIntPack< Dim , FEMSig > , IsotropicUIntPack< Dim , 1 > > F( { 0. , 1. } );
		DenseNodeData< Point< Real , Colors > , IsotropicUIntPack< Dim , FEMSig > > _constraints = tree.template initDenseNodeData< Point< Real , Colors > >( IsotropicUIntPack< Dim , FEMSig >() );
		tree.solveSystem( IsotropicUIntPack< Dim , FEMSig >() , F , constraints , solution , Point< Real , Colors >::Dot , maxDepth , sInfo );
		if( Verbose.set ) printf( "Solved system: " ) , p.print( true );
	}

	Point< Real , Colors > average;
	{
		Profiler p;
		Real begin[] = { 0 , 0 } , end[] = { (Real)w/(1<<maxDepth) , (Real)h/(1<<maxDepth) };
		average = tree.average( solution , begin , end );
		if( Verbose.set ) printf( "Got average: " ) , p.print( true );
	}
	// Stitch the image
	if( Out.set )
	{
		Profiler p;
		int begin[2] , end[2];
		ImageReader* in = ImageReader::Get( In.values[0] );
		ImageWriter* out = ImageWriter::Get( Out.value , w , h , 3 );

		RGBPixel *inRows[2] , *outRows[2];
		unsigned char* inRow = NULL;
		for( int i=0 ; i<2 ; i++ ) inRows[i] = new RGBPixel[w*ROW_BLOCK_SIZE] , outRows[i] = new RGBPixel[w*ROW_BLOCK_SIZE];
		if( in->channels()==1 ) inRow = new unsigned char[w];

		auto FetchInput = [&]( unsigned int block )
		{
			int rStart = block*ROW_BLOCK_SIZE;
			int rEnd = rStart + ROW_BLOCK_SIZE < h ? rStart + ROW_BLOCK_SIZE : h;
			for( int r=rStart , rr=0 ; r<rEnd ; r++ , rr++ )
			{
				if( in->channels()==3 ) in->nextRow( (unsigned char*)( inRows[block&1] + rr*w ) );
				else
				{
					in->nextRow( inRow );
					RGBPixel *_inRow = inRows[block&1] + rr*w;
					ThreadPool::Parallel_for( 0 , w , [&]( unsigned int , size_t i ){ _inRow[i][0] = _inRow[i][1] = _inRow[i][2] = inRow[i]; } );
				}
			}
		};
		auto SetOutput = [&]( unsigned int block )
		{
			int rStart = block*ROW_BLOCK_SIZE;
			int rEnd = rStart + ROW_BLOCK_SIZE < h ? rStart + ROW_BLOCK_SIZE : h;
			out->nextRows( (unsigned char*)outRows[block&1] , rEnd-rStart );
		};
		int blockNum = ( h + ROW_BLOCK_SIZE - 1 ) / ROW_BLOCK_SIZE;

		// Prefetch the first block
		FetchInput( 0 );
		for( int rStart=0 , block=0 ; rStart<h ; rStart+=ROW_BLOCK_SIZE , block++ )
		{
			ThreadPool::ParallelSections
			(
				[&]( void ){ if( block<blockNum ) FetchInput( block+1 ); } ,
				[&]( void ){ if( block>0 ) SetOutput( block-1 ); } ,
				[&]( void )
				{
					RGBPixel *_inRows = inRows[block&1] , *_outRows = outRows[block&1];
					int rEnd = rStart + ROW_BLOCK_SIZE < h ? rStart + ROW_BLOCK_SIZE : h;

					// Expand the next block of values
					begin[0] = 0 , begin[1] = rStart , end[0] = w , end[1] = rEnd;
					Pointer( Point< Real , Colors > ) outBlock = tree.template regularGridUpSample< true >( solution , begin , end );
					int size = (rEnd-rStart)*w;
					ThreadPool::Parallel_for( 0 , size , [&]( unsigned int , size_t ii )
					{
						Point< Real , Colors > c = Point< Real , Colors >( _inRows[ii][0] , _inRows[ii][1] , _inRows[ii][2] ) / 255;
						c += outBlock[ii] - average;
						_outRows[ii] = RGBPixel( c[0] , c[1] , c[2] );
					}
					);
					DeletePointer( outBlock );
				}
			);
		}
		// Write out the last block
		SetOutput( blockNum-1 );
		if( Verbose.set ) printf( "Wrote output: " ) , p.print( true );
		delete[] inRows[0];
		delete[] outRows[0];
		delete[] inRows[1];
		delete[] outRows[1];
		if( inRow ) delete[] inRow;
		delete in;
		delete out;
	}
}

#ifdef FAST_COMPILE
#else // !FAST_COMPILE
template< typename Real >
void _Execute( void )
{
	switch( Degree.value )
	{
	case 1: _Execute< Real , 1 >() ; break;
	case 2: _Execute< Real , 2 >() ; break;
//	case 3: _Execute< Real , 3 >() ; break;
//	case 4: _Execute< Real , 4 >() ; break;
	default: ERROR_OUT( "Only B-Splines of degree 1 - 2 are supported" );
	}
}
#endif // FAST_COMPILE

int main( int argc , char* argv[] )
{
	Timer timer;
#ifdef USE_SEG_FAULT_HANDLER
	WARN( "using seg-fault handler" );
	StackTracer::exec = argv[0];
	signal( SIGSEGV , SignalHandler );
#endif // USE_SEG_FAULT_HANDLER
#ifdef ARRAY_DEBUG
	WARN( "Array debugging enabled" );
#endif // ARRAY_DEBUG
	cmdLineParse( argc-1 , &argv[1] , params );
	if( MaxMemoryGB.value>0 ) SetPeakMemoryMB( MaxMemoryGB.value<<10 );
	ThreadPool::DefaultChunkSize = ThreadChunkSize.value;
	ThreadPool::DefaultSchedule = (ThreadPool::ScheduleType)ScheduleType.value;
	if( Verbose.set )
	{
		printf( "*********************************************\n" );
		printf( "*********************************************\n" );
		printf( "** Running Image Stitching (Version %s) **\n" , VERSION );
		printf( "*********************************************\n" );
		printf( "*********************************************\n" );
		if( !Threads.set ) printf( "Running with %d threads\n" , Threads.value );
	}

	if( !In.set )
	{
		ShowUsage( argv[0] );
		return EXIT_FAILURE;
	}
	if( BaseDepth.value>FullDepth.value )
	{
		if( BaseDepth.set ) WARN( "Base depth must be smaller than full depth: " , BaseDepth.value , " <= " , FullDepth.value );
		BaseDepth.value = FullDepth.value;
	}

#ifdef USE_DOUBLE
	typedef double Real;
#else // !USE_DOUBLE
	typedef float  Real;
#endif // USE_DOUBLE

#ifdef FAST_COMPILE
	static const int Degree = DEFAULT_FEM_DEGREE;
	WARN( "Compiled for degree-" , Degree , ", " , sizeof(Real)==4 ? "single" : "double" , "-precision _only_" );
	_Execute< Real , Degree >();
#else // !FAST_COMPILE
	_Execute< Real >();
#endif // FAST_COMPILE

	if( Performance.set )
	{
		printf( "Time (Wall/CPU): %.2f / %.2f\n" , timer.wallTime() , timer.cpuTime() );
		printf( "Peak Memory (MB): %d\n" , MemoryInfo::PeakMemoryUsageMB() );
	}
	return EXIT_SUCCESS;
}