From 64c2d435eeacb92fb39066e0273d7d66082f26cd Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1=C5=A1=20Oberhuber?= <oberhuber.tomas@gmail.com>
Date: Thu, 23 Jan 2020 21:32:56 +0100
Subject: [PATCH] Fixed sparse matrix assignment.
---
src/TNL/Containers/Segments/CSR.hpp | 1 +
src/TNL/Containers/Segments/CSRView.h | 2 ++
src/TNL/Containers/Segments/CSRView.hpp | 6 ++--
src/TNL/Containers/Segments/Ellpack.hpp | 3 +-
src/TNL/Containers/Segments/EllpackView.h | 6 ++--
src/TNL/Containers/Segments/EllpackView.hpp | 9 +++--
src/TNL/Containers/Segments/SlicedEllpack.hpp | 1 +
.../Containers/Segments/SlicedEllpackView.h | 2 ++
.../Containers/Segments/SlicedEllpackView.hpp | 3 ++
src/TNL/Matrices/Dense.hpp | 1 -
src/TNL/Matrices/DenseMatrixView.hpp | 1 +
src/TNL/Matrices/MatrixView.hpp | 1 +
src/TNL/Matrices/Multidiagonal.h | 2 ++
src/TNL/Matrices/Multidiagonal.hpp | 3 +-
src/TNL/Matrices/MultidiagonalMatrixView.h | 2 ++
src/TNL/Matrices/MultidiagonalMatrixView.hpp | 1 +
src/TNL/Matrices/SparseMatrix.h | 2 +-
src/TNL/Matrices/SparseMatrix.hpp | 34 ++++++++++---------
src/TNL/Matrices/SparseMatrixView.hpp | 3 +-
src/TNL/Matrices/Tridiagonal.h | 2 ++
src/TNL/Matrices/Tridiagonal.hpp | 1 +
.../Containers/Segments/SegmentsTest.hpp | 5 ---
src/UnitTests/Matrices/SparseMatrixCopyTest.h | 5 +++
23 files changed, 62 insertions(+), 34 deletions(-)
diff --git a/src/TNL/Containers/Segments/CSR.hpp b/src/TNL/Containers/Segments/CSR.hpp
index 9a05d84f75..55dcba74c9 100644
--- a/src/TNL/Containers/Segments/CSR.hpp
+++ b/src/TNL/Containers/Segments/CSR.hpp
@@ -253,6 +253,7 @@ CSR< Device, Index, IndexAllocator >::
operator=( const CSR< Device_, Index_, IndexAllocator_ >& source )
{
this->offsets = source.offsets;
+ return *this;
}
template< typename Device,
diff --git a/src/TNL/Containers/Segments/CSRView.h b/src/TNL/Containers/Segments/CSRView.h
index 2ad849f976..759fe8ff7f 100644
--- a/src/TNL/Containers/Segments/CSRView.h
+++ b/src/TNL/Containers/Segments/CSRView.h
@@ -52,8 +52,10 @@ class CSRView
static String getSerializationType();
+ __cuda_callable__
ViewType getView();
+ __cuda_callable__
ConstViewType getConstView() const;
/**
diff --git a/src/TNL/Containers/Segments/CSRView.hpp b/src/TNL/Containers/Segments/CSRView.hpp
index d6ec55b6af..043e06e04e 100644
--- a/src/TNL/Containers/Segments/CSRView.hpp
+++ b/src/TNL/Containers/Segments/CSRView.hpp
@@ -66,6 +66,7 @@ getSerializationType()
template< typename Device,
typename Index >
+__cuda_callable__
typename CSRView< Device, Index >::ViewType
CSRView< Device, Index >::
getView()
@@ -75,6 +76,7 @@ getView()
template< typename Device,
typename Index >
+__cuda_callable__
typename CSRView< Device, Index >::ConstViewType
CSRView< Device, Index >::
getConstView() const
@@ -156,7 +158,6 @@ auto
CSRView< Device, Index >::
getSegmentView( const IndexType segmentIdx ) const -> SegmentViewType
{
- printf( "----> segmentIdx %d offset %d size %d ptr %p \n", segmentIdx, offsets[ segmentIdx ], offsets.getSize(), offsets.getData() );
return SegmentViewType( offsets[ segmentIdx ], offsets[ segmentIdx + 1 ] - offsets[ segmentIdx ], 1 );
}
@@ -167,7 +168,7 @@ void
CSRView< Device, Index >::
forSegments( IndexType first, IndexType last, Function& f, Args... args ) const
{
- const auto offsetsView = this->offsets.getConstView();
+ const auto offsetsView = this->offsets;
auto l = [=] __cuda_callable__ ( const IndexType segmentIdx, Args... args ) mutable {
const IndexType begin = offsetsView[ segmentIdx ];
const IndexType end = offsetsView[ segmentIdx + 1 ];
@@ -228,6 +229,7 @@ CSRView< Device, Index >::
operator=( const CSRView& view )
{
this->offsets.copy( view.offsets );
+ return *this;
}
template< typename Device,
diff --git a/src/TNL/Containers/Segments/Ellpack.hpp b/src/TNL/Containers/Segments/Ellpack.hpp
index 8763c2e5dc..663a65bc80 100644
--- a/src/TNL/Containers/Segments/Ellpack.hpp
+++ b/src/TNL/Containers/Segments/Ellpack.hpp
@@ -293,7 +293,7 @@ void
Ellpack< Device, Index, IndexAllocator, RowMajorOrder, Alignment >::
forAll( Function& f, Args... args ) const
{
- this->forSegments( 0, this->getSize(), f, args... );
+ this->forSegments( 0, this->getSegmentsCount(), f, args... );
}
template< typename Device,
@@ -364,6 +364,7 @@ operator=( const Ellpack< Device_, Index_, IndexAllocator_, RowMajorOrder_, Alig
this->segmentSize = source.segmentSize;
this->size = source.size;
this->alignedSize = roundUpDivision( size, this->getAlignment() ) * this->getAlignment();
+ return *this;
}
template< typename Device,
diff --git a/src/TNL/Containers/Segments/EllpackView.h b/src/TNL/Containers/Segments/EllpackView.h
index f64b04068e..3870f08028 100644
--- a/src/TNL/Containers/Segments/EllpackView.h
+++ b/src/TNL/Containers/Segments/EllpackView.h
@@ -37,7 +37,7 @@ class EllpackView
template< typename Device_, typename Index_ >
using ViewTemplate = EllpackView< Device_, Index_ >;
using ViewType = EllpackView;
- //using ConstViewType = EllpackView< Device, std::add_const_t< Index > >;
+ using ConstViewType = EllpackView< Device, std::add_const_t< Index > >;
using SegmentViewType = SegmentView< IndexType, RowMajorOrder >;
__cuda_callable__
@@ -54,9 +54,11 @@ class EllpackView
static String getSerializationType();
+ __cuda_callable__
ViewType getView();
- //ConstViewType getConstView() const;
+ __cuda_callable__
+ ConstViewType getConstView() const;
/**
* \brief Number segments.
diff --git a/src/TNL/Containers/Segments/EllpackView.hpp b/src/TNL/Containers/Segments/EllpackView.hpp
index c0d0b37215..ea2dc0d216 100644
--- a/src/TNL/Containers/Segments/EllpackView.hpp
+++ b/src/TNL/Containers/Segments/EllpackView.hpp
@@ -78,6 +78,7 @@ template< typename Device,
typename Index,
bool RowMajorOrder,
int Alignment >
+__cuda_callable__
typename EllpackView< Device, Index, RowMajorOrder, Alignment >::ViewType
EllpackView< Device, Index, RowMajorOrder, Alignment >::
getView()
@@ -85,16 +86,17 @@ getView()
return ViewType( segmentSize, size, alignedSize );
}
-/*template< typename Device,
+template< typename Device,
typename Index,
bool RowMajorOrder,
int Alignment >
+__cuda_callable__
typename EllpackView< Device, Index, RowMajorOrder, Alignment >::ConstViewType
EllpackView< Device, Index, RowMajorOrder, Alignment >::
getConstView() const
{
return ConstViewType( segmentSize, size, alignedSize );
-}*/
+}
template< typename Device,
typename Index,
@@ -233,7 +235,7 @@ void
EllpackView< Device, Index, RowMajorOrder, Alignment >::
forAll( Function& f, Args... args ) const
{
- this->forSegments( 0, this->getSize(), f, args... );
+ this->forSegments( 0, this->getSegmentsCount(), f, args... );
}
template< typename Device,
@@ -302,6 +304,7 @@ operator=( const EllpackView< Device, Index, RowMajorOrder, Alignment >& view )
this->segmentSize = view.segmentSize;
this->size = view.size;
this->alignedSize = view.alignedSize;
+ return *this;
}
template< typename Device,
diff --git a/src/TNL/Containers/Segments/SlicedEllpack.hpp b/src/TNL/Containers/Segments/SlicedEllpack.hpp
index 62e2ca7d58..3d3a6d8c3d 100644
--- a/src/TNL/Containers/Segments/SlicedEllpack.hpp
+++ b/src/TNL/Containers/Segments/SlicedEllpack.hpp
@@ -408,6 +408,7 @@ operator=( const SlicedEllpack< Device_, Index_, IndexAllocator_, RowMajorOrder_
this->segmentsCount = source.segmentsCount;
this->sliceOffsets = source.sliceOffsets;
this->sliceSegmentSizes = source.sliceSegmentSizes;
+ return *this;
}
template< typename Device,
diff --git a/src/TNL/Containers/Segments/SlicedEllpackView.h b/src/TNL/Containers/Segments/SlicedEllpackView.h
index c8c73c3f26..2b310a805a 100644
--- a/src/TNL/Containers/Segments/SlicedEllpackView.h
+++ b/src/TNL/Containers/Segments/SlicedEllpackView.h
@@ -56,8 +56,10 @@ class SlicedEllpackView
static String getSerializationType();
+ __cuda_callable__
ViewType getView();
+ __cuda_callable__
ConstViewType getConstView() const;
__cuda_callable__
diff --git a/src/TNL/Containers/Segments/SlicedEllpackView.hpp b/src/TNL/Containers/Segments/SlicedEllpackView.hpp
index 98a3d9b814..3e3c8c09c2 100644
--- a/src/TNL/Containers/Segments/SlicedEllpackView.hpp
+++ b/src/TNL/Containers/Segments/SlicedEllpackView.hpp
@@ -89,6 +89,7 @@ template< typename Device,
typename Index,
bool RowMajorOrder,
int SliceSize >
+__cuda_callable__
typename SlicedEllpackView< Device, Index, RowMajorOrder, SliceSize >::ViewType
SlicedEllpackView< Device, Index, RowMajorOrder, SliceSize >::
getView()
@@ -100,6 +101,7 @@ template< typename Device,
typename Index,
bool RowMajorOrder,
int SliceSize >
+__cuda_callable__
typename SlicedEllpackView< Device, Index, RowMajorOrder, SliceSize >::ConstViewType
SlicedEllpackView< Device, Index, RowMajorOrder, SliceSize >::
getConstView() const
@@ -357,6 +359,7 @@ operator=( const SlicedEllpackView< Device, Index, RowMajorOrder, SliceSize >& v
this->segmentsCount = view.segmentsCount;
this->sliceOffsets.copy( view.sliceOffsets );
this->sliceSegmentSizes.copy( view.sliceSegmentSizes );
+ return *this;
}
template< typename Device,
diff --git a/src/TNL/Matrices/Dense.hpp b/src/TNL/Matrices/Dense.hpp
index e1acfee67e..91a98e7f95 100644
--- a/src/TNL/Matrices/Dense.hpp
+++ b/src/TNL/Matrices/Dense.hpp
@@ -925,7 +925,6 @@ operator=( const Dense< RHSReal, RHSDevice, RHSIndex, RHSRowMajorOrder, RHSRealA
auto this_view = this->view;
if( std::is_same< DeviceType, RHSDeviceType >::value )
{
- const auto segments_view = this->segments.getView();
auto f = [=] __cuda_callable__ ( RHSIndexType rowIdx, RHSIndexType localIdx, RHSIndexType columnIdx, const RHSRealType& value, bool& compute ) mutable {
this_view( rowIdx, columnIdx ) = value;
};
diff --git a/src/TNL/Matrices/DenseMatrixView.hpp b/src/TNL/Matrices/DenseMatrixView.hpp
index 890606436c..a11ff263cd 100644
--- a/src/TNL/Matrices/DenseMatrixView.hpp
+++ b/src/TNL/Matrices/DenseMatrixView.hpp
@@ -901,6 +901,7 @@ operator=( const DenseMatrixView& matrix )
{
MatrixView< Real, Device, Index >::operator=( matrix );
this->segments = matrix.segments;
+ return *this;
}
template< typename Real,
diff --git a/src/TNL/Matrices/MatrixView.hpp b/src/TNL/Matrices/MatrixView.hpp
index 275a228704..360478d055 100644
--- a/src/TNL/Matrices/MatrixView.hpp
+++ b/src/TNL/Matrices/MatrixView.hpp
@@ -131,6 +131,7 @@ operator=( const MatrixView& view )
rows = view.rows;
columns = view.columns;
values.copy( view.values );
+ return *this;
}
template< typename Real,
diff --git a/src/TNL/Matrices/Multidiagonal.h b/src/TNL/Matrices/Multidiagonal.h
index 927e524491..749ddfae7f 100644
--- a/src/TNL/Matrices/Multidiagonal.h
+++ b/src/TNL/Matrices/Multidiagonal.h
@@ -113,8 +113,10 @@ class Multidiagonal : public Matrix< Real, Device, Index, RealAllocator >
template< typename Real_, typename Device_, typename Index_, bool RowMajorOrder_, typename RealAllocator_ >
bool operator != ( const Multidiagonal< Real_, Device_, Index_, RowMajorOrder_, RealAllocator_ >& matrix ) const;
+ __cuda_callable__
RowView getRow( const IndexType& rowIdx );
+ __cuda_callable__
const RowView getRow( const IndexType& rowIdx ) const;
void setValue( const RealType& v );
diff --git a/src/TNL/Matrices/Multidiagonal.hpp b/src/TNL/Matrices/Multidiagonal.hpp
index 5d83004f29..659d6d4ebe 100644
--- a/src/TNL/Matrices/Multidiagonal.hpp
+++ b/src/TNL/Matrices/Multidiagonal.hpp
@@ -683,9 +683,7 @@ operator=( const Multidiagonal< Real_, Device_, Index_, RowMajorOrder_, RealAllo
Containers::Vector< RealType, DeviceType, IndexType, RealAllocatorType > thisValuesBuffer( bufferSize );
Containers::Vector< IndexType, DeviceType, IndexType, IndexAllocatorType > thisColumnsBuffer( bufferSize );
auto matrixValuesBuffer_view = matrixValuesBuffer.getView();
- auto matrixColumnsBuffer_view = matrixColumnsBuffer.getView();
auto thisValuesBuffer_view = thisValuesBuffer.getView();
- auto thisColumnsBuffer_view = thisColumnsBuffer.getView();
IndexType baseRow( 0 );
const IndexType rowsCount = this->getRows();
@@ -716,6 +714,7 @@ operator=( const Multidiagonal< Real_, Device_, Index_, RowMajorOrder_, RealAllo
}
}
}
+ return *this;
}
template< typename Real,
diff --git a/src/TNL/Matrices/MultidiagonalMatrixView.h b/src/TNL/Matrices/MultidiagonalMatrixView.h
index f623a3ca66..97ff94f85c 100644
--- a/src/TNL/Matrices/MultidiagonalMatrixView.h
+++ b/src/TNL/Matrices/MultidiagonalMatrixView.h
@@ -87,8 +87,10 @@ class MultidiagonalMatrixView : public MatrixView< Real, Device, Index >
template< typename Real_, typename Device_, typename Index_, bool RowMajorOrder_ >
bool operator != ( const MultidiagonalMatrixView< Real_, Device_, Index_, RowMajorOrder_ >& matrix ) const;
+ __cuda_callable__
RowView getRow( const IndexType& rowIdx );
+ __cuda_callable__
const RowView getRow( const IndexType& rowIdx ) const;
void setValue( const RealType& v );
diff --git a/src/TNL/Matrices/MultidiagonalMatrixView.hpp b/src/TNL/Matrices/MultidiagonalMatrixView.hpp
index f35c6d713f..33010cebc8 100644
--- a/src/TNL/Matrices/MultidiagonalMatrixView.hpp
+++ b/src/TNL/Matrices/MultidiagonalMatrixView.hpp
@@ -338,6 +338,7 @@ operator=( const MultidiagonalMatrixView& view )
this->diagonalsShifts.copy( view.diagonalsShifts );
this->hostDiagonalsShifts.copy( view.hostDiagonalsShifts );
this->indexer = view.indexer;
+ return *this;
}
template< typename Real,
diff --git a/src/TNL/Matrices/SparseMatrix.h b/src/TNL/Matrices/SparseMatrix.h
index 26d5d2d840..9f91ee7d18 100644
--- a/src/TNL/Matrices/SparseMatrix.h
+++ b/src/TNL/Matrices/SparseMatrix.h
@@ -93,7 +93,7 @@ class SparseMatrix : public Matrix< Real, Device, Index, RealAllocator >
void getCompressedRowLengths( Vector& rowLengths ) const;
[[deprecated]]
- virtual IndexType getRowLength( const IndexType row ) const {};
+ virtual IndexType getRowLength( const IndexType row ) const { return 0;};
template< typename Matrix >
void setLike( const Matrix& matrix );
diff --git a/src/TNL/Matrices/SparseMatrix.hpp b/src/TNL/Matrices/SparseMatrix.hpp
index 447d8d2509..cf4472922b 100644
--- a/src/TNL/Matrices/SparseMatrix.hpp
+++ b/src/TNL/Matrices/SparseMatrix.hpp
@@ -210,7 +210,7 @@ Index
SparseMatrix< Real, Device, Index, MatrixType, Segments, RealAllocator, IndexAllocator >::
getNumberOfNonzeroMatrixElements() const
{
- this->view.getNumberOfNonzeroMatrixElements();
+ return this->view.getNumberOfNonzeroMatrixElements();
}
template< typename Real,
@@ -602,7 +602,6 @@ operator=( const Dense< Real_, Device_, Index_, RowMajorOrder, RealAllocator_ >&
Containers::Vector< IndexType, DeviceType, IndexType, IndexAllocatorType > thisColumnsBuffer( bufferSize );
auto matrixValuesBuffer_view = matrixValuesBuffer.getView();
auto thisValuesBuffer_view = thisValuesBuffer.getView();
- auto thisColumnsBuffer_view = thisColumnsBuffer.getView();
IndexType baseRow( 0 );
const IndexType rowsCount = this->getRows();
@@ -689,11 +688,10 @@ operator=( const RHSMatrix& matrix )
auto rowLocalIndexes_view = rowLocalIndexes.getView();
columns_view = paddingIndex;
- if( std::is_same< DeviceType, RHSDeviceType >::value )
+ /*if( std::is_same< DeviceType, RHSDeviceType >::value )
{
const auto segments_view = this->segments.getView();
auto f = [=] __cuda_callable__ ( RHSIndexType rowIdx, RHSIndexType localIdx_, RHSIndexType columnIndex, const RHSRealType& value, bool& compute ) mutable {
- RealType inValue( 0.0 );
IndexType localIdx( rowLocalIndexes_view[ rowIdx ] );
if( value != 0.0 && columnIndex != paddingIndex )
{
@@ -705,7 +703,7 @@ operator=( const RHSMatrix& matrix )
};
matrix.forAllRows( f );
}
- else
+ else*/
{
const IndexType maxRowLength = max( rowLengths );
const IndexType bufferRowsCount( 128 );
@@ -714,10 +712,13 @@ operator=( const RHSMatrix& matrix )
Containers::Vector< RHSIndexType, RHSDeviceType, RHSIndexType > matrixColumnsBuffer( bufferSize );
Containers::Vector< RealType, DeviceType, IndexType, RealAllocatorType > thisValuesBuffer( bufferSize );
Containers::Vector< IndexType, DeviceType, IndexType > thisColumnsBuffer( bufferSize );
+ Containers::Vector< IndexType, DeviceType, IndexType > thisRowLengths;
+ thisRowLengths = rowLengths;
auto matrixValuesBuffer_view = matrixValuesBuffer.getView();
auto matrixColumnsBuffer_view = matrixColumnsBuffer.getView();
auto thisValuesBuffer_view = thisValuesBuffer.getView();
auto thisColumnsBuffer_view = thisColumnsBuffer.getView();
+ matrixValuesBuffer_view = 0.0;
IndexType baseRow( 0 );
const IndexType rowsCount = this->getRows();
@@ -735,6 +736,7 @@ operator=( const RHSMatrix& matrix )
const IndexType bufferIdx = ( rowIdx - baseRow ) * maxRowLength + localIdx;
matrixColumnsBuffer_view[ bufferIdx ] = columnIndex;
matrixValuesBuffer_view[ bufferIdx ] = value;
+ //std::cerr << " <<<<< rowIdx = " << rowIdx << " localIdx = " << localIdx << " value = " << value << " bufferIdx = " << bufferIdx << std::endl;
}
};
matrix.forRows( baseRow, lastRow, f1 );
@@ -748,20 +750,20 @@ operator=( const RHSMatrix& matrix )
// Copy matrix elements from the buffer to the matrix and ignoring
// zero matrix elements
const IndexType matrix_columns = this->getColumns();
- auto matrix_view = matrix.getView();
- auto f2 = [=] __cuda_callable__ ( IndexType rowIdx, IndexType localIdx_, IndexType& columnIndex, RealType& value, bool& compute ) mutable {
+ const auto thisRowLengths_view = thisRowLengths.getConstView();
+ auto f2 = [=] __cuda_callable__ ( IndexType rowIdx, IndexType localIdx, IndexType& columnIndex, RealType& value, bool& compute ) mutable {
RealType inValue( 0.0 );
- IndexType bufferIdx, localIdx( rowLocalIndexes_view[ rowIdx ] );
- auto matrixRow = matrix_view.getRow( rowIdx );
- IndexType s = matrixRow.getSize();
- //printf( " row %d size %d \n", rowIdx, s );
- while( inValue == 0.0 && localIdx < 0 )
+ size_t bufferIdx;
+ IndexType bufferLocalIdx( rowLocalIndexes_view[ rowIdx ] );
+ while( inValue == 0.0 && localIdx < thisRowLengths_view[ rowIdx ] )
{
- bufferIdx = ( rowIdx - baseRow ) * maxRowLength + localIdx++;
+ bufferIdx = ( rowIdx - baseRow ) * maxRowLength + bufferLocalIdx++;
TNL_ASSERT_LT( bufferIdx, bufferSize, "" );
- //inValue = thisValuesBuffer_view[ bufferIdx ];
+ inValue = thisValuesBuffer_view[ bufferIdx ];
}
- /*rowLocalIndexes_view[ rowIdx ] = localIdx;
+ //std::cerr << "rowIdx = " << rowIdx << " localIdx = " << localIdx << " bufferLocalIdx = " << bufferLocalIdx
+ // << " inValue = " << inValue << " bufferIdx = " << bufferIdx << std::endl;
+ rowLocalIndexes_view[ rowIdx ] = bufferLocalIdx;
if( inValue == 0.0 )
{
columnIndex = paddingIndex;
@@ -771,7 +773,7 @@ operator=( const RHSMatrix& matrix )
{
columnIndex = thisColumnsBuffer_view[ bufferIdx ];//column - 1;
value = inValue;
- }*/
+ }
};
this->forRows( baseRow, lastRow, f2 );
baseRow += bufferRowsCount;
diff --git a/src/TNL/Matrices/SparseMatrixView.hpp b/src/TNL/Matrices/SparseMatrixView.hpp
index 965a51b8b6..4ac0a29b80 100644
--- a/src/TNL/Matrices/SparseMatrixView.hpp
+++ b/src/TNL/Matrices/SparseMatrixView.hpp
@@ -139,7 +139,7 @@ Index
SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView >::
getRowLength( const IndexType row ) const
{
-
+ return 0;
}
template< typename Real,
@@ -525,6 +525,7 @@ operator=( const SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView
MatrixView< Real, Device, Index >::operator=( matrix );
this->columnIndexes.copy( matrix.columnIndexes );
this->segments = matrix.segments;
+ return *this;
}
template< typename Real,
diff --git a/src/TNL/Matrices/Tridiagonal.h b/src/TNL/Matrices/Tridiagonal.h
index 3f89023106..6f0c6a548e 100644
--- a/src/TNL/Matrices/Tridiagonal.h
+++ b/src/TNL/Matrices/Tridiagonal.h
@@ -91,8 +91,10 @@ class Tridiagonal : public Matrix< Real, Device, Index, RealAllocator >
template< typename Real_, typename Device_, typename Index_, bool RowMajorOrder_, typename RealAllocator_ >
bool operator != ( const Tridiagonal< Real_, Device_, Index_, RowMajorOrder_, RealAllocator_ >& matrix ) const;
+ __cuda_callable__
RowView getRow( const IndexType& rowIdx );
+ __cuda_callable__
const RowView getRow( const IndexType& rowIdx ) const;
void setValue( const RealType& v );
diff --git a/src/TNL/Matrices/Tridiagonal.hpp b/src/TNL/Matrices/Tridiagonal.hpp
index d99715a47c..2ccdc4838f 100644
--- a/src/TNL/Matrices/Tridiagonal.hpp
+++ b/src/TNL/Matrices/Tridiagonal.hpp
@@ -592,6 +592,7 @@ operator=( const Tridiagonal< Real_, Device_, Index_, RowMajorOrder_, RealAlloca
this->forAllRows( f );
}
}
+ return *this;
}
template< typename Real,
diff --git a/src/UnitTests/Containers/Segments/SegmentsTest.hpp b/src/UnitTests/Containers/Segments/SegmentsTest.hpp
index 590b39881f..6d4692dbe7 100644
--- a/src/UnitTests/Containers/Segments/SegmentsTest.hpp
+++ b/src/UnitTests/Containers/Segments/SegmentsTest.hpp
@@ -127,17 +127,12 @@ void test_AllReduction_MaximumInSegments()
TNL::Containers::Vector< IndexType, DeviceType, IndexType > v( segments.getStorageSize() );
- /*IndexType k( 1 );
- for( IndexType i = 0; i < segmentsCount; i++ )
- for( IndexType j = 0; j < segmentSize; j++ )
- v.setElement( segments.getGlobalIndex( i, j ), k++ );*/
auto view = v.getView();
auto init = [=] __cuda_callable__ ( const IndexType segmentIdx, const IndexType localIdx, const IndexType globalIdx ) mutable -> bool {
view[ globalIdx ] = segmentIdx * 5 + localIdx + 1;
return true;
};
segments.forAll( init );
- std::cerr << v << std::endl;
TNL::Containers::Vector< IndexType, DeviceType, IndexType >result( segmentsCount );
diff --git a/src/UnitTests/Matrices/SparseMatrixCopyTest.h b/src/UnitTests/Matrices/SparseMatrixCopyTest.h
index 46777f6c00..053f1e9fb5 100644
--- a/src/UnitTests/Matrices/SparseMatrixCopyTest.h
+++ b/src/UnitTests/Matrices/SparseMatrixCopyTest.h
@@ -458,6 +458,7 @@ void tridiagonalMatrixAssignment()
RowCapacitiesType rowCapacities;
matrix.getCompressedRowLengths( rowCapacities );
RowCapacitiesType exactRowLengths{ 1, 3, 3, 3, 3, 3, 3, 3, 3, 2 };
+
EXPECT_EQ( rowCapacities, exactRowLengths );
for( IndexType i = 0; i < rows; i++ )
for( IndexType j = 0; j < columns; j++ )
@@ -510,6 +511,10 @@ void multidiagonalMatrixAssignment()
RowCapacitiesType rowCapacities;
matrix.getCompressedRowLengths( rowCapacities );
RowCapacitiesType exactRowLengths{ 3, 4, 5, 5, 6, 5, 5, 4, 4, 3 };
+ /*std::cerr << "hostMatrix " << hostMatrix << std::endl;
+ std::cerr << "matrix " << matrix << std::endl;
+ std::cerr << "rowCapacities " << rowCapacities << std::endl;*/
+
EXPECT_EQ( rowCapacities, exactRowLengths );
for( IndexType i = 0; i < rows; i++ )
for( IndexType j = 0; j < columns; j++ )
--
GitLab