Added script for running traversers benchmark. (524483f7) · Commits · TNL / tnl-dev

src/Benchmarks/Benchmarks.h

+1 −1

Original line number	Diff line number	Diff line
		@@ -56,7 +56,7 @@ timeFunction( ComputeFunction compute,

		int i;
		for( i = 0;
		i < loops && ( ! minTime \|\| timer.getRealTime() < ( double ) minTime );
		i < loops \|\| timer.getRealTime() < ( double ) minTime;
		++i)
		{
		// abuse the monitor's "time" for loops

src/Benchmarks/Traversers/GridTraversersBenchmark.h

0 → 100644

+137 −0

Original line number	Diff line number	Diff line
		/***************************************************************************
		WriteOne.h - description
		-------------------
		begin : Dec 19, 2018
		copyright : (C) 2018 by oberhuber
		email : tomas.oberhuber@fjfi.cvut.cz
		***************************************************************************/

		/* See Copyright Notice in tnl/Copyright */

		// Implemented by: Tomas Oberhuber

		#pragma once

		#include <TNL/ParallelFor.h>
		#include <TNL/Devices/Host.h>
		#include <TNL/Devices/Cuda.h>
		#include <TNL/Containers/Vector.h>

		namespace TNL {
		namespace Benchmarks {


		template< int Dimension,
		typename Device,
		typename Real,
		typename Index >
		class GridTraversersBenchmark{};

		template< typename Device,
		typename Real,
		typename Index >
		class GridTraversersBenchmark< 1, Device, Real, Index >
		{
		public:

		using Vector = Containers::Vector< Real, Device, Index >;

		GridTraversersBenchmark( Index size )
		:v( size ), size( size )
		{}

		void writeOne()
		{

		auto f = [] __cuda_callable__ ( Index i, Real* data )
		{
		data[ i ] = i;
		};

		ParallelFor< Device >::exec( ( Index ) 0, size, f, v.getData() );
		}

		protected:

		Index size;
		Vector v;
		};


		template< typename Device,
		typename Real,
		typename Index >
		class GridTraversersBenchmark< 2, Device, Real, Index >
		{
		public:

		using Vector = Containers::Vector< Real, Device, Index >;

		GridTraversersBenchmark( Index size )
		:size( size ), v( size * size ) { }

		void writeOne()
		{
		Index _size = this->size;
		auto f = [=] __cuda_callable__ ( Index i, Index j, Real* data )
		{
		data[ i * _size + j ] = i + j;
		};

		ParallelFor2D< Device >::exec( ( Index ) 0,
		( Index ) 0,
		this->size,
		this->size,
		f, v.getData() );
		}

		protected:

		Index size;

		Vector v;

		};

		template< typename Device,
		typename Real,
		typename Index >
		class GridTraversersBenchmark< 3, Device, Real, Index >
		{
		public:

		using Vector = Containers::Vector< Real, Device, Index >;

		GridTraversersBenchmark( Index size )
		: size( size ), v( size * size * size ) {}

		void writeOne()
		{
		Index _size = this->size;
		auto f = [=] __cuda_callable__ ( Index i, Index j, Index k, Real* data )
		{
		data[ ( i * _size + j ) * _size + k ] = i + j + k;
		};

		ParallelFor3D< Device >::exec( ( Index ) 0,
		( Index ) 0,
		( Index ) 0,
		this->size,
		this->size,
		this->size,
		f, v.getData() );
		}

		protected:

		Index size;
		Vector v;

		};


		} // namespace Benchmarks
		} // namespace TNL

src/Benchmarks/Traversers/tnl-benchmark-traversers.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -107,6 +107,7 @@ bool setupBenchmark( const Config::ParameterContainer& parameters )


		Benchmark benchmark; //( loops, verbose );
		benchmark.setup( parameters );
		Benchmark::MetadataMap metadata = getHardwareMetadata();
		runBenchmark< Dimension >( parameters, benchmark, metadata );

src/Benchmarks/scripts/run-tnl-benchmark-traversers

0 → 100644

+5 −0

Original line number	Diff line number	Diff line
		#!/bin/bash

		tnl-benchmark-traversers --dimension 1 --loops 1 --min-size 16 --max-size 100000 --min-time 1
		tnl-benchmark-traversers --dimension 2 --loops 1 --min-size 16 --max-size 10000 --min-time 1 --output-mode append
		tnl-benchmark-traversers --dimension 3 --loops 1 --min-size 16 --max-size 1000 --min-time 1 --output-mode append