diff --git a/src/Benchmarks/SpMV/tnlCusparseCSRMatrix.h b/src/Benchmarks/SpMV/tnlCusparseCSRMatrix.h
index cfb271d7d7c4090cb6fcfcb67766d0115757d1d9..8f6d376fe27ebed3cd67307bf8f24ea2c5d630d4 100644
--- a/src/Benchmarks/SpMV/tnlCusparseCSRMatrix.h
+++ b/src/Benchmarks/SpMV/tnlCusparseCSRMatrix.h
@@ -60,7 +60,7 @@ class CusparseCSRBase
void vectorProduct( const InVector& inVector,
OutVector& outVector ) const
{
- TNL_ASSERT( matrix, );
+ TNL_ASSERT_TRUE( matrix, "matrix was not initialized" );
#ifdef HAVE_CUDA
cusparseDcsrmv( *( this->cusparseHandle ),
CUSPARSE_OPERATION_NON_TRANSPOSE,
@@ -103,7 +103,7 @@ class CusparseCSR< double > : public CusparseCSRBase< double >
void vectorProduct( const InVector& inVector,
OutVector& outVector ) const
{
- TNL_ASSERT( matrix, "" );
+ TNL_ASSERT_TRUE( matrix, "matrix was not initialized" );
#ifdef HAVE_CUDA
double d = 1.0;
double* alpha = &d;
@@ -134,7 +134,7 @@ class CusparseCSR< float > : public CusparseCSRBase< float >
void vectorProduct( const InVector& inVector,
OutVector& outVector ) const
{
- TNL_ASSERT( matrix, "" );
+ TNL_ASSERT_TRUE( matrix, "matrix was not initialized" );
#ifdef HAVE_CUDA
float d = 1.0;
float* alpha = &d;
diff --git a/src/TNL/Matrices/AdEllpack_impl.h b/src/TNL/Matrices/AdEllpack_impl.h
index b70087b8d956de7d53603e2d8b72d4db064cb46b..daab3b8cdc351459120436186cb3060aeee54d8f 100644
--- a/src/TNL/Matrices/AdEllpack_impl.h
+++ b/src/TNL/Matrices/AdEllpack_impl.h
@@ -318,6 +318,8 @@ Index AdEllpack< Real, Device, Index >::getWarp( const IndexType row ) const
( ( this->rowOffset.getElement( searchedWarp ) < row ) && ( this->rowOffset.getElement( searchedWarp + 1 ) >= row ) ) )
return searchedWarp;
}
+ // FIXME: non-void function always has to return something sensible
+ throw "bug - row was not found";
}
template< typename Real,
@@ -474,7 +476,6 @@ bool AdEllpack< Real, Device, Index >::setRow( const IndexType row,
warp++;
bool found = false;
- IndexType length = 0;
IndexType elementPtr;
IndexType elPtr = 0;
while( ( !found ) && ( elPtr < elements ) )
diff --git a/src/TNL/Matrices/BiEllpackSymmetric_impl.h b/src/TNL/Matrices/BiEllpackSymmetric_impl.h
index 4a8ad24701334e2a0609190874e1871096c784cd..5b6f94b57b571963f5ba18c3c4c7e8fb7700fa99 100644
--- a/src/TNL/Matrices/BiEllpackSymmetric_impl.h
+++ b/src/TNL/Matrices/BiEllpackSymmetric_impl.h
@@ -149,7 +149,7 @@ Index BiEllpackSymmetric< Real, Device, Index, StripSize >::getNumberOfGroups( c
<< " this->getName() = " << std::endl );
IndexType strip = row / this->warpSize;
- IndexType rowStripPermutation = this->rowPermArray.getElement( row ) - this->warpSize * strip;
+ IndexType rowStripPermutation = this->rowPermArray[ row ] - this->warpSize * strip;
IndexType numberOfGroups = this->logWarpSize + 1;
IndexType bisection = 1;
for( IndexType i = 0; i < this->logWarpSize + 1; i++ )
@@ -158,6 +158,12 @@ Index BiEllpackSymmetric< Real, Device, Index, StripSize >::getNumberOfGroups( c
return ( numberOfGroups - i );
bisection *= 2;
}
+ // FIXME: non-void function always has to return something sensible
+#ifndef __CUDA_ARCH__
+ throw "bug - row was not found";
+#else
+ TNL_ASSERT_TRUE( false, "bug - row was not found" );
+#endif
}
template< typename Real,
@@ -390,6 +396,7 @@ bool BiEllpackSymmetric< Real, Device, Index, StripSize >::setRow( const IndexTy
}
if( thisElementPtr == numberOfElements )
return true;
+ return false;
}
template< typename Real,
@@ -576,8 +583,8 @@ __cuda_callable__
Index BiEllpackSymmetric< Real, Device, Index, StripSize >::getGroupLength( const Index strip,
const Index group ) const
{
- return this->groupPointers.getElement( strip * ( this->logWarpSize + 1 ) + group + 1 )
- - this->groupPointers.getElement( strip * ( this->logWarpSize + 1 ) + group );
+ return this->groupPointers[ strip * ( this->logWarpSize + 1 ) + group + 1 ]
+ - this->groupPointers[ strip * ( this->logWarpSize + 1 ) + group ];
}
template< typename Real,
diff --git a/src/TNL/Matrices/BiEllpack_impl.h b/src/TNL/Matrices/BiEllpack_impl.h
index b8533d62077285549564585569949edd662371d7..ea5e1efb9463915ec724b28b38abbfe64ac596b0 100644
--- a/src/TNL/Matrices/BiEllpack_impl.h
+++ b/src/TNL/Matrices/BiEllpack_impl.h
@@ -154,7 +154,7 @@ Index BiEllpack< Real, Device, Index, StripSize >::getNumberOfGroups( const Inde
<< " this->getName() = " << std::endl; );
IndexType strip = row / this->warpSize;
- IndexType rowStripPermutation = this->rowPermArray.getElement( row ) - this->warpSize * strip;
+ IndexType rowStripPermutation = this->rowPermArray[ row ] - this->warpSize * strip;
IndexType numberOfGroups = this->logWarpSize + 1;
IndexType bisection = 1;
for( IndexType i = 0; i < this->logWarpSize + 1; i++ )
@@ -163,6 +163,12 @@ Index BiEllpack< Real, Device, Index, StripSize >::getNumberOfGroups( const Inde
return ( numberOfGroups - i );
bisection *= 2;
}
+ // FIXME: non-void function always has to return something sensible
+#ifndef __CUDA_ARCH__
+ throw "bug - row was not found";
+#else
+ TNL_ASSERT_TRUE( false, "bug - row was not found" );
+#endif
}
template< typename Real,
@@ -397,6 +403,7 @@ setRow( const IndexType row,
}
if( thisElementPtr == numberOfElements )
return true;
+ return false;
}
template< typename Real,
@@ -581,8 +588,8 @@ __cuda_callable__
Index BiEllpack< Real, Device, Index, StripSize >::getGroupLength( const Index strip,
const Index group ) const
{
- return this->groupPointers.getElement( strip * ( this->logWarpSize + 1 ) + group + 1 )
- - this->groupPointers.getElement( strip * ( this->logWarpSize + 1 ) + group );
+ return this->groupPointers[ strip * ( this->logWarpSize + 1 ) + group + 1 ]
+ - this->groupPointers[ strip * ( this->logWarpSize + 1 ) + group ];
}
template< typename Real,
diff --git a/src/TNL/Matrices/EllpackSymmetricGraph.h b/src/TNL/Matrices/EllpackSymmetricGraph.h
index 26a5e18368182f9fda86cba2e00fe0341c2a6e07..3a282c796be439209184023fb76aa692ff1e4294 100644
--- a/src/TNL/Matrices/EllpackSymmetricGraph.h
+++ b/src/TNL/Matrices/EllpackSymmetricGraph.h
@@ -153,10 +153,8 @@ class EllpackSymmetricGraph : public Sparse< Real, Device, Index >
const int color ) const;
#endif
- __cuda_callable__
void computePermutationArray();
- __cuda_callable__
bool rearrangeMatrix( bool verbose );
bool save( File& file ) const;
@@ -182,21 +180,20 @@ class EllpackSymmetricGraph : public Sparse< Real, Device, Index >
__cuda_callable__
Index getRowsOfColor( IndexType color ) const;
- __cuda_callable__
void copyFromHostToCuda( EllpackSymmetricGraph< Real, Devices::Host, Index >& matrix );
__cuda_callable__
- Containers::Vector< Index, Device, Index > getPermutationArray();
+ Containers::Vector< Index, Device, Index >& getPermutationArray();
__cuda_callable__
- Containers::Vector< Index, Device, Index > getInversePermutation();
+ Containers::Vector< Index, Device, Index >& getInversePermutation();
__cuda_callable__
- Containers::Vector< Index, Device, Index > getColorPointers();
+ Containers::Vector< Index, Device, Index >& getColorPointers();
protected:
- bool allocateElements();
+ void allocateElements();
IndexType rowLengths, alignedRows;
diff --git a/src/TNL/Matrices/EllpackSymmetricGraph_impl.h b/src/TNL/Matrices/EllpackSymmetricGraph_impl.h
index 9d04b6b2255fd9c9293c9d81358ac2b214824b33..799d07281e62ae441c39162ee2fc5271997effac 100644
--- a/src/TNL/Matrices/EllpackSymmetricGraph_impl.h
+++ b/src/TNL/Matrices/EllpackSymmetricGraph_impl.h
@@ -102,7 +102,7 @@ template< typename Real,
__cuda_callable__
Index EllpackSymmetricGraph< Real, Device, Index >::getRowsOfColor( IndexType color ) const
{
- return this->colorPointers.getElement( color + 1 ) - this->colorPointers.getElement( color );
+ return this->colorPointers[ color + 1 ] - this->colorPointers[ color ];
}
/*
@@ -174,7 +174,6 @@ void EllpackSymmetricGraph< Real, Device, Index >::computeColorsVector( Containe
template< typename Real,
typename Device,
typename Index >
-__cuda_callable__
void EllpackSymmetricGraph< Real, Device, Index >::computePermutationArray()
{
// init vector of colors and permutation array
@@ -238,7 +237,6 @@ void EllpackSymmetricGraph< Real, Device, Index >::verifyPermutationArray()
template< typename Real,
typename Device,
typename Index >
-__cuda_callable__
bool EllpackSymmetricGraph< Real, Device, Index >::rearrangeMatrix( bool verbose )
{
// first we need to know permutation
@@ -296,7 +294,8 @@ template< typename Real,
typename Device,
typename Index >
__cuda_callable__
-Containers::Vector< Index, Device, Index > EllpackSymmetricGraph< Real, Device, Index >::getPermutationArray()
+Containers::Vector< Index, Device, Index >&
+EllpackSymmetricGraph< Real, Device, Index >::getPermutationArray()
{
return this->permutationArray;
}
@@ -305,7 +304,8 @@ template< typename Real,
typename Device,
typename Index >
__cuda_callable__
-Containers::Vector< Index, Device, Index > EllpackSymmetricGraph< Real, Device, Index >::getInversePermutation()
+Containers::Vector< Index, Device, Index >&
+EllpackSymmetricGraph< Real, Device, Index >::getInversePermutation()
{
return this->inversePermutationArray;
}
@@ -314,7 +314,8 @@ template< typename Real,
typename Device,
typename Index >
__cuda_callable__
-Containers::Vector< Index, Device, Index > EllpackSymmetricGraph< Real, Device, Index >::getColorPointers()
+Containers::Vector< Index, Device, Index >&
+EllpackSymmetricGraph< Real, Device, Index >::getColorPointers()
{
return this->colorPointers;
}
@@ -322,7 +323,6 @@ Containers::Vector< Index, Device, Index > EllpackSymmetricGraph< Real, Device,
template< typename Real,
typename Device,
typename Index >
-__cuda_callable__
void EllpackSymmetricGraph< Real, Device, Index >::copyFromHostToCuda( EllpackSymmetricGraph< Real, Devices::Host, Index >& matrix )
{
// TODO: fix
@@ -331,17 +331,17 @@ void EllpackSymmetricGraph< Real, Device, Index >::copyFromHostToCuda( EllpackSy
this->rearranged = true;
this->rowLengths = matrix.getRowLengthsInt();
this->alignedRows = matrix.getAlignedRows();
- Containers::Vector< Index, Devices::Host, Index > colorPointers = matrix.getColorPointers();
+ Containers::Vector< Index, Devices::Host, Index >& colorPointers = matrix.getColorPointers();
this->colorPointers.setSize( colorPointers.getSize() );
for( IndexType i = 0; i < colorPointers.getSize(); i++ )
this->colorPointers.setElement( i, colorPointers[ i ] );
- Containers::Vector< Index,Devices::Host, Index > permutationArray = matrix.getPermutationArray();
+ Containers::Vector< Index,Devices::Host, Index >& permutationArray = matrix.getPermutationArray();
this->permutationArray.setSize( permutationArray.getSize() );
for( IndexType i = 0; i < permutationArray.getSize(); i++ )
this->permutationArray.setElement( i, permutationArray[ i ] );
- Containers::Vector< Index, Devices::Host, Index > inversePermutation = matrix.getInversePermutation();
+ Containers::Vector< Index, Devices::Host, Index >& inversePermutation = matrix.getInversePermutation();
this->inversePermutationArray.setSize( inversePermutation.getSize() );
for( IndexType i = 0; i < inversePermutation.getSize(); i++ )
this->inversePermutationArray.setElement( i, inversePermutation[ i ] );
@@ -363,7 +363,7 @@ bool EllpackSymmetricGraph< Real, Device, Index >::setConstantRowLengths( const
TNL_ASSERT( rowLengths > 0, std::cerr << " rowLengths = " << rowLengths );
this->rowLengths = rowLengths;
if( this->rows > 0 )
- return allocateElements();
+ allocateElements();
return true;
}
@@ -445,37 +445,37 @@ bool EllpackSymmetricGraph< Real, Device, Index > :: addElementFast( const Index
const RealType& thisElementMultiplicator )
{
typedef EllpackSymmetricGraphDeviceDependentCode< DeviceType > DDCType;
- IndexType i = DDCType::getRowBegin( *this, this->permutationArray.getElement( row ) );
- const IndexType rowEnd = DDCType::getRowEnd( *this, this->permutationArray.getElement( row ) );
+ IndexType i = DDCType::getRowBegin( *this, this->permutationArray[ row ] );
+ const IndexType rowEnd = DDCType::getRowEnd( *this, this->permutationArray[ row ] );
const IndexType step = DDCType::getElementStep( *this );
while( i < rowEnd &&
- this->columnIndexes.getElement( i ) < column &&
- this->columnIndexes.getElement( i ) != this->getPaddingIndex() ) i += step;
+ this->columnIndexes[ i ] < column &&
+ this->columnIndexes[ i ] != this->getPaddingIndex() ) i += step;
if( i == rowEnd )
return false;
- if( this->columnIndexes.getElement( i ) == column )
+ if( this->columnIndexes[ i ] == column )
{
- this->values.setElement( i, thisElementMultiplicator * this->values.getElement( i ) + value);
+ this->values[ i ] = thisElementMultiplicator * this->values[ i ] + value;
return true;
}
else
- if( this->columnIndexes.getElement( i ) == this->getPaddingIndex() ) // artificial zero
+ if( this->columnIndexes[ i ] == this->getPaddingIndex() ) // artificial zero
{
- this->columnIndexes.setElement( i, column);
- this->values.setElement( i, value);
+ this->columnIndexes[ i ] = column;
+ this->values[ i ] = value;
}
else
{
Index j = rowEnd - step;
while( j > i )
{
- this->columnIndexes.setElement( j, this->columnIndexes.getElement( j - step ) );
- this->values.setElement( j, this->values.getElement( j - step ) );
+ this->columnIndexes[ j ] = this->columnIndexes[ j - step ];
+ this->values[ j ] = this->values[ j - step ];
j -= step;
}
- this->columnIndexes.setElement( i, column );
- this->values.setElement( i, value );
+ this->columnIndexes[ i ] = column;
+ this->values[ i ] = value;
}
return true;
}
@@ -783,6 +783,7 @@ bool EllpackSymmetricGraph< Real, Device, Index >::help( bool verbose )
{
if( !this->rearranged )
return this->rearrangeMatrix( verbose );
+ return true;
}
template< typename Real,
@@ -810,7 +811,7 @@ void EllpackSymmetricGraph< Real, Device, Index >::print( std::ostream& str ) co
template< typename Real,
typename Device,
typename Index >
-bool EllpackSymmetricGraph< Real, Device, Index >::allocateElements()
+void EllpackSymmetricGraph< Real, Device, Index >::allocateElements()
{
Sparse< Real, Device, Index >::allocateMatrixElements( this->alignedRows * this->rowLengths );
}
diff --git a/src/TNL/Matrices/EllpackSymmetric_impl.h b/src/TNL/Matrices/EllpackSymmetric_impl.h
index af3a645200027d08c429986849ee594fb0de2f33..42202a883c0c887317aaf68904dfa0bddf27a646 100644
--- a/src/TNL/Matrices/EllpackSymmetric_impl.h
+++ b/src/TNL/Matrices/EllpackSymmetric_impl.h
@@ -85,7 +85,7 @@ bool EllpackSymmetric< Real, Device, Index >::setConstantRowLengths( const Index
std::cerr << " rowLengths = " << rowLengths );
this->rowLengths = rowLengths;
if( this->rows > 0 )
- return allocateElements();
+ allocateElements();
return true;
}
diff --git a/src/TNL/Matrices/SlicedEllpackSymmetricGraph.h b/src/TNL/Matrices/SlicedEllpackSymmetricGraph.h
index 4cc04436b125004544032013639e1b919f6bec9d..a2ab000957227a88a9655888cb417cea6498f56f 100644
--- a/src/TNL/Matrices/SlicedEllpackSymmetricGraph.h
+++ b/src/TNL/Matrices/SlicedEllpackSymmetricGraph.h
@@ -195,10 +195,8 @@ class SlicedEllpackSymmetricGraph : public Sparse< Real, Device, Index >
void copyFromHostToCuda( SlicedEllpackSymmetricGraph< Real, Devices::Host, Index, SliceSize >& matrix );
- __cuda_callable__
bool rearrangeMatrix( bool verbose = false );
- __cuda_callable__
void computePermutationArray();
Containers::Vector< Index, Device, Index > getSlicePointers();
diff --git a/src/TNL/Matrices/SlicedEllpackSymmetricGraph_impl.h b/src/TNL/Matrices/SlicedEllpackSymmetricGraph_impl.h
index d5ad42534869944b4bd11dadca31c1271a839241..9f09a21c506bd08110bc8360e17d3bf1f3a4097c 100644
--- a/src/TNL/Matrices/SlicedEllpackSymmetricGraph_impl.h
+++ b/src/TNL/Matrices/SlicedEllpackSymmetricGraph_impl.h
@@ -701,7 +701,6 @@ Index SlicedEllpackSymmetricGraph< Real, Device, Index, SliceSize >::getRealRowL
Index rowBegin = slicePointer + rowLength * ( row - sliceIdx * SliceSize );
Index rowEnd = rowBegin + rowLength;
- Index step = 1;
Index length = 0;
for( Index i = rowBegin; i < rowEnd; i++ )
if( this->columnIndexes.getElement( i ) != this->getPaddingIndex() )
@@ -764,7 +763,6 @@ bool SlicedEllpackSymmetricGraph< Real, Device, Index, SliceSize >::rearrangeMat
for( IndexType row = slice * SliceSize; row < (slice + 1) * SliceSize && row < this->getRows(); row++ )
{
IndexType rowBegin = slicePointerOrig + rowLengthOrig * ( row - slice * SliceSize );
- IndexType rowEnd = rowBegin + rowLengthOrig;
IndexType elementPointer = rowBegin;
IndexType sliceNew = this->permutationArray.getElement( row ) / SliceSize;
@@ -1040,7 +1038,7 @@ class SlicedEllpackSymmetricGraphDeviceDependentCode< Devices::Host >
template< typename Real,
typename Index,
int SliceSize >
- static bool computeMaximalRowLengthInSlices( SlicedEllpackSymmetricGraph< Real, Device, Index, SliceSize >& matrix,
+ static void computeMaximalRowLengthInSlices( SlicedEllpackSymmetricGraph< Real, Device, Index, SliceSize >& matrix,
typename SlicedEllpackSymmetricGraph< Real, Device, Index >::ConstCompressedRowLengthsVectorView rowLengths,
Containers::Vector< Index, Device, Index >& sliceRowLengths,
Containers::Vector< Index, Device, Index >& slicePointers )
@@ -1233,7 +1231,7 @@ class SlicedEllpackSymmetricGraphDeviceDependentCode< Devices::Cuda >
template< typename Real,
typename Index,
int SliceSize >
- static bool computeMaximalRowLengthInSlices( SlicedEllpackSymmetricGraph< Real, Device, Index, SliceSize >& matrix,
+ static void computeMaximalRowLengthInSlices( SlicedEllpackSymmetricGraph< Real, Device, Index, SliceSize >& matrix,
typename SlicedEllpackSymmetricGraph< Real, Device, Index >::ConstCompressedRowLengthsVectorView rowLengths,
Containers::Vector< Index, Device, Index >& sliceRowLengths,
Containers::Vector< Index, Device, Index >& slicePointers )
diff --git a/src/TNL/Matrices/SlicedEllpackSymmetric_impl.h b/src/TNL/Matrices/SlicedEllpackSymmetric_impl.h
index 905c42080c814b21db156b79341478787cf289df..402ac5a6c8128ab67bbf2393528fb3a2b58b9077 100644
--- a/src/TNL/Matrices/SlicedEllpackSymmetric_impl.h
+++ b/src/TNL/Matrices/SlicedEllpackSymmetric_impl.h
@@ -75,7 +75,7 @@ void SlicedEllpackSymmetric< Real, Device, Index, SliceSize >::setCompressedRowL
// TODO: Uncomment the next line and fix the compilation
//DeviceDependentCode::computeMaximalRowLengthInSlices( *this, rowLengths );
- TNL_ASSERT( false, "code fix required" );
+ throw std::runtime_error("code fix required");
this->maxRowLength = rowLengths.max();
@@ -743,6 +743,7 @@ class SlicedEllpackSymmetricDeviceDependentCode< Devices::Host >
template< typename Real,
typename Index,
int SliceSize >
+ __cuda_callable__
static void initRowTraverseFast( const SlicedEllpackSymmetric< Real, Device, Index, SliceSize >& matrix,
const Index row,
Index& rowBegin,
@@ -762,7 +763,7 @@ class SlicedEllpackSymmetricDeviceDependentCode< Devices::Host >
template< typename Real,
typename Index,
int SliceSize >
- static bool computeMaximalRowLengthInSlices( SlicedEllpackSymmetric< Real, Device, Index, SliceSize >& matrix,
+ static void computeMaximalRowLengthInSlices( SlicedEllpackSymmetric< Real, Device, Index, SliceSize >& matrix,
typename SlicedEllpackSymmetric< Real, Device, Index >::ConstCompressedRowLengthsVectorView rowLengths )
{
Index row( 0 ), slice( 0 ), sliceRowLength( 0 );
@@ -862,7 +863,7 @@ class SlicedEllpackSymmetricDeviceDependentCode< Devices::Cuda >
template< typename Real,
typename Index,
int SliceSize >
- static bool computeMaximalRowLengthInSlices( SlicedEllpackSymmetric< Real, Device, Index, SliceSize >& matrix,
+ static void computeMaximalRowLengthInSlices( SlicedEllpackSymmetric< Real, Device, Index, SliceSize >& matrix,
typename SlicedEllpackSymmetric< Real, Device, Index >::ConstCompressedRowLengthsVectorView rowLengths )
{
#ifdef HAVE_CUDA