Renaming CSRScalarKernel to CSRKernelScalar and maiking separate source files. (69c12aea) · Commits · TNL / tnl-dev

src/TNL/Algorithms/Segments/CSR.h

+2 −2

Original line number	Diff line number	Diff line
		@@ -22,7 +22,7 @@ namespace TNL {

		template< typename Device,
		typename Index,
		typename Kernel = CSRScalarKernel< Index, Device >,
		typename Kernel = CSRKernelScalar< Index, Device >,
		typename IndexAllocator = typename Allocators::Default< Device >::template Allocator< Index > >
		class CSR
		{
		@@ -133,7 +133,7 @@ class CSR
		template< typename Device,
		typename Index,
		typename IndexAllocator = typename Allocators::Default< Device >::template Allocator< Index > >
		using CSRScalar = CSR< Device, Index, CSRScalarKernel< Index, Device >, IndexAllocator >;
		using CSRScalar = CSR< Device, Index, CSRKernelScalar< Index, Device >, IndexAllocator >;

		template< typename Device,
		typename Index,

src/TNL/Algorithms/Segments/CSRKernelScalar.h

0 → 100644

+61 −0

Original line number	Diff line number	Diff line
		/***************************************************************************
		CSRKernelScalar.h - description
		-------------------
		begin : Jan 23, 2021 -> Joe Biden inauguration
		copyright : (C) 2021 by Tomas Oberhuber
		email : tomas.oberhuber@fjfi.cvut.cz
		***************************************************************************/

		/* See Copyright Notice in tnl/Copyright */

		#pragma once

		#include <TNL/Assert.h>
		#include <TNL/Cuda/LaunchHelpers.h>
		#include <TNL/Containers/VectorView.h>
		#include <TNL/Algorithms/ParallelFor.h>
		#include <TNL/Algorithms/Segments/details/LambdaAdapter.h>

		namespace TNL {
		namespace Algorithms {
		namespace Segments {

		template< typename Index,
		typename Device >
		struct CSRKernelScalar
		{
		using IndexType = Index;
		using DeviceType = Device;
		using ViewType = CSRKernelScalar< Index, Device >;
		using ConstViewType = CSRKernelScalar< Index, Device >;

		template< typename Offsets >
		void init( const Offsets& offsets );

		void reset();

		ViewType getView();

		ConstViewType getConstView() const;

		template< typename OffsetsView,
		typename Fetch,
		typename Reduction,
		typename ResultKeeper,
		typename Real,
		typename... Args >
		static void segmentsReduction( const OffsetsView& offsets,
		Index first,
		Index last,
		Fetch& fetch,
		const Reduction& reduction,
		ResultKeeper& keeper,
		const Real& zero,
		Args... args );
		};

		} // namespace Segments
		} // namespace Algorithms
		} // namespace TNL

		#include <TNL/Algorithms/Segments/CSRKernelScalar.hpp>
		No newline at end of file

src/TNL/Algorithms/Segments/CSRKernelScalar.hpp

0 → 100644

+92 −0

Original line number	Diff line number	Diff line
		/***************************************************************************
		CSRKernelScalar.h - description
		-------------------
		begin : Jan 23, 2021 -> Joe Biden inauguration
		copyright : (C) 2021 by Tomas Oberhuber
		email : tomas.oberhuber@fjfi.cvut.cz
		***************************************************************************/

		/* See Copyright Notice in tnl/Copyright */

		#pragma once

		#include <TNL/Assert.h>
		#include <TNL/Cuda/LaunchHelpers.h>
		#include <TNL/Containers/VectorView.h>
		#include <TNL/Algorithms/ParallelFor.h>
		#include <TNL/Algorithms/Segments/CSRKernelScalar.h>
		#include <TNL/Algorithms/Segments/details/LambdaAdapter.h>

		namespace TNL {
		namespace Algorithms {
		namespace Segments {

		template< typename Index,
		typename Device >
		template< typename Offsets >
		void
		CSRKernelScalar< Index, Device >::
		init( const Offsets& offsets )
		{
		}

		template< typename Index,
		typename Device >
		void
		CSRKernelScalar< Index, Device >::
		reset()
		{
		}

		template< typename Index,
		typename Device >
		auto
		CSRKernelScalar< Index, Device >::
		getView() -> ViewType
		{
		return *this;
		}

		template< typename Index,
		typename Device >
		auto
		CSRKernelScalar< Index, Device >::
		getConstView() const -> ConstViewType
		{
		return *this;
		};

		template< typename Index,
		typename Device >
		template< typename OffsetsView,
		typename Fetch,
		typename Reduction,
		typename ResultKeeper,
		typename Real,
		typename... Args >
		void
		CSRKernelScalar< Index, Device >::
		segmentsReduction( const OffsetsView& offsets,
		Index first,
		Index last,
		Fetch& fetch,
		const Reduction& reduction,
		ResultKeeper& keeper,
		const Real& zero,
		Args... args )
		{
		auto l = [=] __cuda_callable__ ( const IndexType segmentIdx, Args... args ) mutable {
		const IndexType begin = offsets[ segmentIdx ];
		const IndexType end = offsets[ segmentIdx + 1 ];
		Real aux( zero );
		IndexType localIdx( 0 );
		bool compute( true );
		for( IndexType globalIdx = begin; globalIdx < end && compute; globalIdx++ )
		aux = reduction( aux, details::FetchLambdaAdapter< IndexType, Fetch >::call( fetch, segmentIdx, localIdx++, globalIdx, compute ) );
		keeper( segmentIdx, aux );
		};
		Algorithms::ParallelFor< Device >::exec( first, last, l, args... );
		}
		} // namespace Segments
		} // namespace Algorithms
		} // namespace TNL

src/TNL/Algorithms/Segments/CSRKernels.h

+0 −47

Original line number	Diff line number	Diff line
		@@ -20,53 +20,6 @@ namespace TNL {
		namespace Algorithms {
		namespace Segments {

		template< typename Index,
		typename Device >
		struct CSRScalarKernel
		{
		using IndexType = Index;
		using DeviceType = Device;
		using ViewType = CSRScalarKernel< Index, Device >;
		using ConstViewType = CSRScalarKernel< Index, Device >;

		template< typename Offsets >
		void init( const Offsets& offsets ) {};

		void reset(){};

		ViewType getView() { return *this; };

		ConstViewType getConstView() const { return *this; };

		template< typename OffsetsView,
		typename Fetch,
		typename Reduction,
		typename ResultKeeper,
		typename Real,
		typename... Args >
		static void segmentsReduction( const OffsetsView& offsets,
		Index first,
		Index last,
		Fetch& fetch,
		const Reduction& reduction,
		ResultKeeper& keeper,
		const Real& zero,
		Args... args )
		{
		auto l = [=] __cuda_callable__ ( const IndexType segmentIdx, Args... args ) mutable {
		const IndexType begin = offsets[ segmentIdx ];
		const IndexType end = offsets[ segmentIdx + 1 ];
		Real aux( zero );
		IndexType localIdx( 0 );
		bool compute( true );
		for( IndexType globalIdx = begin; globalIdx < end && compute; globalIdx++ )
		aux = reduction( aux, details::FetchLambdaAdapter< IndexType, Fetch >::call( fetch, segmentIdx, localIdx++, globalIdx, compute ) );
		keeper( segmentIdx, aux );
		};
		Algorithms::ParallelFor< Device >::exec( first, last, l, args... );
		}
		};

		#ifdef HAVE_CUDA
		template< typename Offsets,
		typename Index,

src/TNL/Algorithms/Segments/CSRView.h

+3 −2

Original line number	Diff line number	Diff line
		@@ -14,6 +14,7 @@

		#include <TNL/Containers/Vector.h>
		#include <TNL/Algorithms/Segments/SegmentView.h>
		#include <TNL/Algorithms/Segments/CSRKernelScalar.h>
		#include <TNL/Algorithms/Segments/CSRKernels.h>

		namespace TNL {
		@@ -22,7 +23,7 @@ namespace TNL {

		template< typename Device,
		typename Index,
		typename Kernel = CSRScalarKernel< Index, Device > >
		typename Kernel = CSRKernelScalar< Index, Device > >
		class CSRView
		{
		public:
		@@ -131,7 +132,7 @@ class CSRView

		template< typename Device,
		typename Index >
		using CSRViewScalar = CSRView< Device, Index, CSRScalarKernel< Index, Device > >;
		using CSRViewScalar = CSRView< Device, Index, CSRKernelScalar< Index, Device > >;

		template< typename Device,
		typename Index >