From fe457c9765fffac0090a51fb48600a9238c24857 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1=C5=A1=20Oberhuber?= <oberhuber.tomas@gmail.com>
Date: Mon, 6 Jan 2020 22:04:40 +0100
Subject: [PATCH] Fixed segments reduction lambda function parameters.
---
src/TNL/Containers/Segments/CSRView.hpp | 5 +++--
src/TNL/Containers/Segments/EllpackView.hpp | 8 +++++---
src/TNL/Containers/Segments/SlicedEllpackView.hpp | 8 +++++---
src/TNL/Matrices/SparseMatrix.hpp | 6 +++---
src/UnitTests/Containers/Segments/SegmentsTest.hpp | 2 +-
src/UnitTests/Matrices/SparseMatrixTest.hpp | 4 ++--
6 files changed, 19 insertions(+), 14 deletions(-)
diff --git a/src/TNL/Containers/Segments/CSRView.hpp b/src/TNL/Containers/Segments/CSRView.hpp
index bbed8e3cb0..2d2b583317 100644
--- a/src/TNL/Containers/Segments/CSRView.hpp
+++ b/src/TNL/Containers/Segments/CSRView.hpp
@@ -195,15 +195,16 @@ void
CSRView< Device, Index >::
segmentsReduction( IndexType first, IndexType last, Fetch& fetch, Reduction& reduction, ResultKeeper& keeper, const Real& zero, Args... args ) const
{
- using RealType = decltype( fetch( IndexType(), IndexType(), std::declval< bool& >(), args... ) );
+ using RealType = decltype( fetch( IndexType(), IndexType(), IndexType(), std::declval< bool& >(), args... ) );
const auto offsetsView = this->offsets.getConstView();
auto l = [=] __cuda_callable__ ( const IndexType i, Args... args ) mutable {
const IndexType begin = offsetsView[ i ];
const IndexType end = offsetsView[ i + 1 ];
RealType aux( zero );
+ IndexType localIdx( 0 );
bool compute( true );
for( IndexType j = begin; j < end && compute; j++ )
- reduction( aux, fetch( i, j, compute, args... ) );
+ reduction( aux, fetch( i, localIdx++, j, compute, args... ) );
keeper( i, aux );
};
Algorithms::ParallelFor< Device >::exec( first, last, l, args... );
diff --git a/src/TNL/Containers/Segments/EllpackView.hpp b/src/TNL/Containers/Segments/EllpackView.hpp
index dc6bd485dd..21be886541 100644
--- a/src/TNL/Containers/Segments/EllpackView.hpp
+++ b/src/TNL/Containers/Segments/EllpackView.hpp
@@ -245,7 +245,7 @@ void
EllpackView< Device, Index, RowMajorOrder, Alignment >::
segmentsReduction( IndexType first, IndexType last, Fetch& fetch, Reduction& reduction, ResultKeeper& keeper, const Real& zero, Args... args ) const
{
- using RealType = decltype( fetch( IndexType(), IndexType(), std::declval< bool& >(), args... ) );
+ using RealType = decltype( fetch( IndexType(), IndexType(), IndexType(), std::declval< bool& >(), args... ) );
if( RowMajorOrder )
{
const IndexType segmentSize = this->segmentSize;
@@ -253,9 +253,10 @@ segmentsReduction( IndexType first, IndexType last, Fetch& fetch, Reduction& red
const IndexType begin = i * segmentSize;
const IndexType end = begin + segmentSize;
RealType aux( zero );
+ IndexType localIdx( 0 );
bool compute( true );
for( IndexType j = begin; j < end && compute; j++ )
- reduction( aux, fetch( i, j, compute, args... ) );
+ reduction( aux, fetch( i, localIdx++, j, compute, args... ) );
keeper( i, aux );
};
Algorithms::ParallelFor< Device >::exec( first, last, l, args... );
@@ -268,9 +269,10 @@ segmentsReduction( IndexType first, IndexType last, Fetch& fetch, Reduction& red
const IndexType begin = i;
const IndexType end = storageSize;
RealType aux( zero );
+ IndexType localIdx( 0 );
bool compute( true );
for( IndexType j = begin; j < end && compute; j += alignedSize )
- reduction( aux, fetch( i, j, compute, args... ) );
+ reduction( aux, fetch( i, localIdx++, j, compute, args... ) );
keeper( i, aux );
};
Algorithms::ParallelFor< Device >::exec( first, last, l, args... );
diff --git a/src/TNL/Containers/Segments/SlicedEllpackView.hpp b/src/TNL/Containers/Segments/SlicedEllpackView.hpp
index 82570664fa..5f9cbdee3e 100644
--- a/src/TNL/Containers/Segments/SlicedEllpackView.hpp
+++ b/src/TNL/Containers/Segments/SlicedEllpackView.hpp
@@ -293,7 +293,7 @@ void
SlicedEllpackView< Device, Index, RowMajorOrder, SliceSize >::
segmentsReduction( IndexType first, IndexType last, Fetch& fetch, Reduction& reduction, ResultKeeper& keeper, const Real& zero, Args... args ) const
{
- using RealType = decltype( fetch( IndexType(), IndexType(), std::declval< bool& >(), args... ) );
+ using RealType = decltype( fetch( IndexType(), IndexType(), IndexType(), std::declval< bool& >(), args... ) );
const auto sliceSegmentSizes_view = this->sliceSegmentSizes.getConstView();
const auto sliceOffsets_view = this->sliceOffsets.getConstView();
if( RowMajorOrder )
@@ -305,9 +305,10 @@ segmentsReduction( IndexType first, IndexType last, Fetch& fetch, Reduction& red
const IndexType begin = sliceOffsets_view[ sliceIdx ] + segmentInSliceIdx * segmentSize;
const IndexType end = begin + segmentSize;
RealType aux( zero );
+ IndexType localIdx( 0 );
bool compute( true );
for( IndexType globalIdx = begin; globalIdx< end; globalIdx++ )
- reduction( aux, fetch( segmentIdx, globalIdx, compute, args... ) );
+ reduction( aux, fetch( segmentIdx, localIdx++, globalIdx, compute, args... ) );
keeper( segmentIdx, aux );
};
Algorithms::ParallelFor< Device >::exec( first, last, l, args... );
@@ -321,9 +322,10 @@ segmentsReduction( IndexType first, IndexType last, Fetch& fetch, Reduction& red
const IndexType begin = sliceOffsets_view[ sliceIdx ] + segmentInSliceIdx;
const IndexType end = sliceOffsets_view[ sliceIdx + 1 ];
RealType aux( zero );
+ IndexType localIdx( 0 );
bool compute( true );
for( IndexType globalIdx = begin; globalIdx < end; globalIdx += SliceSize )
- reduction( aux, fetch( segmentIdx, globalIdx, compute, args... ) );
+ reduction( aux, fetch( segmentIdx, localIdx++, globalIdx, compute, args... ) );
keeper( segmentIdx, aux );
};
Algorithms::ParallelFor< Device >::exec( first, last, l, args... );
diff --git a/src/TNL/Matrices/SparseMatrix.hpp b/src/TNL/Matrices/SparseMatrix.hpp
index 6aa75995f5..8dbe53f4dc 100644
--- a/src/TNL/Matrices/SparseMatrix.hpp
+++ b/src/TNL/Matrices/SparseMatrix.hpp
@@ -417,12 +417,12 @@ vectorProduct( const InVector& inVector,
const auto valuesView = this->values.getConstView();
const auto columnIndexesView = this->columnIndexes.getConstView();
const IndexType paddingIndex = this->getPaddingIndex();
- auto fetch = [=] __cuda_callable__ ( IndexType row, IndexType offset, bool& compute ) -> RealType {
- const IndexType column = columnIndexesView[ offset ];
+ auto fetch = [=] __cuda_callable__ ( IndexType row, IndexType localIdx, IndexType globalIdx, bool& compute ) -> RealType {
+ const IndexType column = columnIndexesView[ globalIdx ];
compute = ( column != paddingIndex );
if( ! compute )
return 0.0;
- return valuesView[ offset ] * inVectorView[ column ];
+ return valuesView[ globalIdx ] * inVectorView[ column ];
};
auto reduction = [] __cuda_callable__ ( RealType& sum, const RealType& value ) {
sum += value;
diff --git a/src/UnitTests/Containers/Segments/SegmentsTest.hpp b/src/UnitTests/Containers/Segments/SegmentsTest.hpp
index 6189c2e9a4..8320fafe57 100644
--- a/src/UnitTests/Containers/Segments/SegmentsTest.hpp
+++ b/src/UnitTests/Containers/Segments/SegmentsTest.hpp
@@ -143,7 +143,7 @@ void test_AllReduction_MaximumInSegments()
const auto v_view = v.getConstView();
auto result_view = result.getView();
- auto fetch = [=] __cuda_callable__ ( IndexType segmentIdx, IndexType globalIdx, bool& compute ) -> IndexType {
+ auto fetch = [=] __cuda_callable__ ( IndexType segmentIdx, IndexType localIdx, IndexType globalIdx, bool& compute ) -> IndexType {
return v_view[ globalIdx ];
};
auto reduce = [] __cuda_callable__ ( IndexType& a, const IndexType b ) {
diff --git a/src/UnitTests/Matrices/SparseMatrixTest.hpp b/src/UnitTests/Matrices/SparseMatrixTest.hpp
index bf261aa846..b0a9fcb000 100644
--- a/src/UnitTests/Matrices/SparseMatrixTest.hpp
+++ b/src/UnitTests/Matrices/SparseMatrixTest.hpp
@@ -1214,7 +1214,7 @@ void test_RowsReduction()
// Compute number of non-zero elements in rows.
typename Matrix::RowsCapacitiesType rowLengths( rows );
auto rowLengths_view = rowLengths.getView();
- auto fetch = [] __cuda_callable__ ( IndexType row, IndexType column, const RealType& value ) -> IndexType {
+ auto fetch = [] __cuda_callable__ ( IndexType row, IndexType column, IndexType globalIdx, const RealType& value ) -> IndexType {
return ( value != 0.0 );
};
auto reduce = [] __cuda_callable__ ( IndexType& aux, const IndexType a ) {
@@ -1232,7 +1232,7 @@ void test_RowsReduction()
// Compute max norm
TNL::Containers::Vector< RealType, DeviceType, IndexType > rowSums( rows );
auto rowSums_view = rowSums.getView();
- auto max_fetch = [] __cuda_callable__ ( IndexType row, IndexType column, const RealType& value ) -> IndexType {
+ auto max_fetch = [] __cuda_callable__ ( IndexType row, IndexType column, IndexType globalIdx, const RealType& value ) -> IndexType {
return abs( value );
};
auto max_reduce = [] __cuda_callable__ ( IndexType& aux, const IndexType a ) {
--
GitLab