diff --git a/src/Examples/narrow-band/CMakeLists.txt b/src/Examples/narrow-band/CMakeLists.txt
deleted file mode 100644
index 158cd20132ed4f499228cda7400ba18776dc1503..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/CMakeLists.txt
+++ /dev/null
@@ -1,22 +0,0 @@
-set( tnl_narrow_band_SOURCES
-# MainBuildConfig.h
-# tnlNarrowBand2D_impl.h
-# tnlNarrowBand.h
-# narrowBandConfig.h
- main.cpp)
-
-
-IF( BUILD_CUDA )
- CUDA_ADD_EXECUTABLE(narrow-band main.cu)
-ELSE( BUILD_CUDA )
- ADD_EXECUTABLE(narrow-band main.cpp)
-ENDIF( BUILD_CUDA )
-target_link_libraries (narrow-band tnl )
-
-
-INSTALL( TARGETS narrow-band
- RUNTIME DESTINATION bin
- PERMISSIONS OWNER_READ OWNER_WRITE OWNER_EXECUTE GROUP_READ GROUP_EXECUTE WORLD_READ WORLD_EXECUTE )
-
-#INSTALL( FILES ${tnl_narrow_band_SOURCES}
-# DESTINATION ${TNL_TARGET_DATA_DIRECTORY}/examples/narrow-band )
diff --git a/src/Examples/narrow-band/MainBuildConfig.h b/src/Examples/narrow-band/MainBuildConfig.h
deleted file mode 100644
index ed3d686eb99379af1589d734eac9b5812cccdedf..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/MainBuildConfig.h
+++ /dev/null
@@ -1,64 +0,0 @@
-/***************************************************************************
- MainBuildConfig.h - description
- -------------------
- begin : Jul 7, 2014
- copyright : (C) 2014 by Tomas Oberhuber
- email : tomas.oberhuber@fjfi.cvut.cz
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef MAINBUILDCONFIG_H_
-#define MAINBUILDCONFIG_H_
-
-#include <solvers/tnlBuildConfigTags.h>
-
-class MainBuildConfig
-{
- public:
-
- static void print() {std::cerr << "MainBuildConfig" <<std::endl; }
-};
-
-/****
- * Turn off support for float and long double.
- */
-template<> struct tnlConfigTagReal< MainBuildConfig, float > { enum { enabled = false }; };
-template<> struct tnlConfigTagReal< MainBuildConfig, long double > { enum { enabled = false }; };
-
-/****
- * Turn off support for short int and long int indexing.
- */
-template<> struct tnlConfigTagIndex< MainBuildConfig, short int >{ enum { enabled = false }; };
-template<> struct tnlConfigTagIndex< MainBuildConfig, long int >{ enum { enabled = false }; };
-
-/****
- * Use of tnlGrid is enabled for allowed dimensions and Real, Device and Index types.
- */
-template< int Dimensions, typename Real, typename Device, typename Index >
- struct tnlConfigTagMesh< MainBuildConfig, tnlGrid< Dimensions, Real, Device, Index > >
- { enum { enabled = tnlConfigTagDimensions< MainBuildConfig, Dimensions >::enabled &&
- tnlConfigTagReal< MainBuildConfig, Real >::enabled &&
- tnlConfigTagDevice< MainBuildConfig, Device >::enabled &&
- tnlConfigTagIndex< MainBuildConfig, Index >::enabled }; };
-
-/****
- * Please, chose your preferred time discretisation here.
- */
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlExplicitTimeDiscretisationTag >{ enum { enabled = true }; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlSemiImplicitTimeDiscretisationTag >{ enum { enabled = false}; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlImplicitTimeDiscretisationTag >{ enum { enabled = false }; };
-
-/****
- * Only the Runge-Kutta-Merson solver is enabled by default.
- */
-template<> struct tnlConfigTagExplicitSolver< MainBuildConfig, tnlExplicitEulerSolverTag >{ enum { enabled = false }; };
-
-#endif /* MAINBUILDCONFIG_H_ */
diff --git a/src/Examples/narrow-band/main.cpp b/src/Examples/narrow-band/main.cpp
deleted file mode 100644
index 8849008ff630db0400a6d7d98e789099e5fbb5d9..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/main.cpp
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cpp - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/Examples/narrow-band/main.cu b/src/Examples/narrow-band/main.cu
deleted file mode 100644
index 8849008ff630db0400a6d7d98e789099e5fbb5d9..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/main.cu
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cpp - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/Examples/narrow-band/main.h b/src/Examples/narrow-band/main.h
deleted file mode 100644
index 51dbdac37cfc5ff76b5ad03b826bf3a4642b17b4..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/main.h
+++ /dev/null
@@ -1,88 +0,0 @@
-/***************************************************************************
- main.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-
-#include "MainBuildConfig.h"
- //for HOST versions:
-//#include "tnlNarrowBand.h"
- //for DEVICE versions:
-#include "tnlNarrowBand_CUDA.h"
-#include "narrowBandConfig.h"
-#include <solvers/tnlBuildConfigTags.h>
-
-#include <mesh/tnlGrid.h>
-#include <core/tnlDevice.h>
-#include <time.h>
-#include <ctime>
-
-typedef MainBuildConfig BuildConfig;
-
-int main( int argc, char* argv[] )
-{
- time_t start;
- time_t stop;
- time(&start);
- std::clock_t start2= std::clock();
- Config::ParameterContainer parameters;
- tnlConfigDescription configDescription;
- narrowBandConfig< BuildConfig >::configSetup( configDescription );
-
- if( ! parseCommandLine( argc, argv, configDescription, parameters ) )
- return false;
-
- const int& dim = parameters.getParameter< int >( "dim" );
-
- if(dim == 2)
- {
- tnlNarrowBand<tnlGrid<2,double,TNL::Devices::Host, int>, double, int> solver;
- if(!solver.init(parameters))
- {
- cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- TNL_CHECK_CUDA_DEVICE;
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver..." <<std::endl;
- solver.run();
- }
-// else if(dim == 3)
-// {
-// tnlNarrowBand<tnlGrid<3,double,TNL::Devices::Host, int>, double, int> solver;
-// if(!solver.init(parameters))
-// {
-// cerr << "Solver failed to initialize." <<std::endl;
-// return EXIT_FAILURE;
-// }
-// TNL_CHECK_CUDA_DEVICE;
-// std::cout << "-------------------------------------------------------------" <<std::endl;
-// std::cout << "Starting solver..." <<std::endl;
-// solver.run();
-// }
- else
- {
- std::cerr << "Unsupported number of dimensions: " << dim << "!" <<std::endl;
- return EXIT_FAILURE;
- }
-
-
- time(&stop);
- std::cout << "Solver stopped..." <<std::endl;
- std::cout <<std::endl;
- std::cout << "Running time was: " << difftime(stop,start) << " .... " << (std::clock() - start2) / (double)(CLOCKS_PER_SEC) <<std::endl;
- return EXIT_SUCCESS;
-}
-
-
diff --git a/src/Examples/narrow-band/narrowBandConfig.h b/src/Examples/narrow-band/narrowBandConfig.h
deleted file mode 100644
index bab58ceac46bf9c766b697ed79c2c676111323a2..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/narrowBandConfig.h
+++ /dev/null
@@ -1,40 +0,0 @@
-/***************************************************************************
- narrowBandConfig.h - description
- -------------------
- begin : Oct 15, 2015
- copyright : (C) 2015 by Tomas Sobotik
- email :
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef NARROWBANDCONFIG_H_
-#define NARROWBANDCONFIG_H_
-
-#include <config/tnlConfigDescription.h>
-
-template< typename ConfigTag >
-class narrowBandConfig
-{
- public:
- static void configSetup( tnlConfigDescription& config )
- {
- config.addDelimiter( "Narrow Band Solver solver settings:" );
- config.addEntry < String > ( "problem-name", "This defines particular problem.", "fast-sweeping" );
- config.addRequiredEntry < String > ( "initial-condition", "Initial condition for solver");
- config.addRequiredEntry < int > ( "dim", "Dimension of problem.");
- config.addRequiredEntry < double > ( "tau", "Time step.");
- config.addRequiredEntry < double > ( "final-time", "Final time.");
- config.addEntry < String > ( "mesh", "Name of mesh.", "mesh.tnl" );
- config.addEntry < String > ( "exact-input", "Are the function values near the curve equal to the SDF? (yes/no)", "no" );
- }
-};
-
-#endif /* NARROWBANDCONFIG_H_ */
diff --git a/src/Examples/narrow-band/tnlNarrowBand.h b/src/Examples/narrow-band/tnlNarrowBand.h
deleted file mode 100644
index 7d3d19bc03b43247f735cf04c5c82368360a748d..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/tnlNarrowBand.h
+++ /dev/null
@@ -1,186 +0,0 @@
-/***************************************************************************
- tnlNarrowBand.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLNARROWBAND_H_
-#define TNLNARROWBAND_H_
-
-#include <TNL/Config/ParameterContainer.h>
-#include <TNL/Containers/Vector.h>
-#include <TNL/Containers/StaticVector.h>
-#include <functions/tnlMeshFunction.h>
-#include <TNL/Devices/Host.h>
-#include <mesh/tnlGrid.h>
-#include <mesh/grids/tnlGridEntity.h>
-#include <limits.h>
-#include <core/tnlDevice.h>
-#include <ctime>
-#ifdef HAVE_OPENMP
-#include <omp.h>
-#endif
-
-
-
-
-template< typename Mesh,
- typename Real,
- typename Index >
-class tnlNarrowBand
-{};
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 2, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
-
- tnlNarrowBand();
-
- static String getType();
- bool init( const Config::ParameterContainer& parameters );
-
- bool initGrid();
- bool run();
-
- //for single core version use this implementation:
- void updateValue(const Index i, const Index j);
- //for parallel version use this one instead:
-// void updateValue(const Index i, const Index j, DofVectorType* grid);
-
-
- void setupSquare1000(Index i, Index j);
- void setupSquare1100(Index i, Index j);
- void setupSquare1010(Index i, Index j);
- void setupSquare1001(Index i, Index j);
- void setupSquare1110(Index i, Index j);
- void setupSquare1101(Index i, Index j);
- void setupSquare1011(Index i, Index j);
- void setupSquare1111(Index i, Index j);
- void setupSquare0000(Index i, Index j);
- void setupSquare0100(Index i, Index j);
- void setupSquare0010(Index i, Index j);
- void setupSquare0001(Index i, Index j);
- void setupSquare0110(Index i, Index j);
- void setupSquare0101(Index i, Index j);
- void setupSquare0011(Index i, Index j);
- void setupSquare0111(Index i, Index j);
-
- Real fabsMin(const Real x, const Real y);
-
-
-protected:
-
- MeshType Mesh;
-
- bool exactInput;
-
- tnlMeshFunction<MeshType> dofVector, dofVector2;
- DofVectorType data;
-
- RealType h;
-
- tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage > Entity;
-
-
-#ifdef HAVE_OPENMP
-// omp_lock_t* gridLock;
-#endif
-
-
-};
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 3, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
- tnlNarrowBand();
-
- static String getType();
- bool init( const Config::ParameterContainer& parameters );
-
- bool initGrid();
- bool run();
-
- //for single core version use this implementation:
- void updateValue(const Index i, const Index j, const Index k);
- //for parallel version use this one instead:
-// void updateValue(const Index i, const Index j, DofVectorType* grid);
-
- Real fabsMin(const Real x, const Real y);
-
-
-protected:
-
- MeshType Mesh;
-
- bool exactInput;
-
-
- tnlMeshFunction<MeshType> dofVector, dofVector2;
- DofVectorType data;
-
- RealType h;
-
- tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage > Entity;
-
-#ifdef HAVE_OPENMP
-// omp_lock_t* gridLock;
-#endif
-
-
-};
-
-
- //for single core version use this implementation:
-#include "tnlNarrowBand2D_impl.h"
- //for parallel version use this one instead:
-// #include "tnlNarrowBand2D_openMP_impl.h"
-
-#include "tnlNarrowBand3D_impl.h"
-
-#endif /* TNLNARROWBAND_H_ */
diff --git a/src/Examples/narrow-band/tnlNarrowBand2D_CUDA_v4_impl.h b/src/Examples/narrow-band/tnlNarrowBand2D_CUDA_v4_impl.h
deleted file mode 100644
index dff0b48c8d69cc6fa6ec932e24446a8c3b1b1417..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/tnlNarrowBand2D_CUDA_v4_impl.h
+++ /dev/null
@@ -1,1317 +0,0 @@
-/***************************************************************************
- tnlNarrowBand2D_CUDA_v4_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLNARROWBAND2D_IMPL_H_
-#define TNLNARROWBAND2D_IMPL_H_
-
-#define NARROWBAND_SUBGRID_SIZE 32
-
-#include "tnlNarrowBand.h"
-
-#ifdef HAVE_CUDA
-__device__
-double fabsMin( double x, double y)
-{
- double fx = abs(x);
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-}
-
-__device__
-double atomicFabsMin(double* address, double val)
-{
- unsigned long long int* address_as_ull =
- (unsigned long long int*)address;
- unsigned long long int old = *address_as_ull, assumed;
- do {
- assumed = old;
- old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(__longlong_as_double(assumed),val) ));
- } while (assumed != old);
- return __longlong_as_double(old);
-}
-#endif
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-#ifdef HAVE_CUDA
- __device__ __host__
-#endif
-Real tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >:: positivePart(const Real arg) const
-{
- if(arg > 0.0)
- return arg;
- return 0.0;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-#ifdef HAVE_CUDA
- __device__ __host__
-#endif
-Real tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: negativePart(const Real arg) const
-{
- if(arg < 0.0)
- return -arg;
- return 0.0;
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlNarrowBand< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: tnlNarrowBand()
-:dofVector(Mesh)
-{
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- h = Mesh.template getSpaceStepsProducts< 1, 0 >();
- //Entity.refresh();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
- tau = parameters.getParameter< double >( "tau" );
-
- finalTime = parameters.getParameter< double >( "final-time" );
-
- statusGridSize = ((Mesh.getDimensions().x() + NARROWBAND_SUBGRID_SIZE-1 ) / NARROWBAND_SUBGRID_SIZE);
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaStatusVector), statusGridSize*statusGridSize*sizeof(int));
-// cudaMemcpy(cudaDofVector, this->dofVector.getData().getData(), statusGridSize*statusGridSize* sizeof(int)), cudaMemcpyHostToDevice);
-
- cudaMalloc(&reinitialize, sizeof(int));
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-#endif
-
- int n = Mesh.getDimensions().x();
-
- dim3 threadsPerBlock2(NARROWBAND_SUBGRID_SIZE, NARROWBAND_SUBGRID_SIZE);
- dim3 numBlocks2(statusGridSize ,statusGridSize);
- initSetupGridCUDA<<<numBlocks2,threadsPerBlock2>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initSetupGrid2CUDA<<<numBlocks2,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
-
- /*dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 + 1 ,n/16 +1);*/
- initCUDA<<<numBlocks2,threadsPerBlock2>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
-
- cout << "Solver initialized." <<std::endl;
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlockFS(1, 512);
- dim3 numBlocksFS(4,1);
- dim3 threadsPerBlockNB(NARROWBAND_SUBGRID_SIZE, NARROWBAND_SUBGRID_SIZE);
- dim3 numBlocksNB(n/NARROWBAND_SUBGRID_SIZE + 1,n/NARROWBAND_SUBGRID_SIZE + 1);
-
- double time = 0.0;
- int reinit = 0;
-
- cout << "Hi!" <<std::endl;
- runCUDA<<<numBlocksFS,threadsPerBlockFS>>>(this->cudaSolver,0,0);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- cout << "Hi2!" <<std::endl;
- while(time < finalTime)
- {
- if(tau+time > finalTime)
- tau=finalTime-time;
-
- runNarrowBandCUDA<<<numBlocksNB,threadsPerBlockNB>>>(this->cudaSolver,tau);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- time += tau;
-
-
- cudaMemcpy(&reinit, this->reinitialize, sizeof(int), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- if(reinit != 0 /*&& time != finalTime */)
- {
- cout << time <<std::endl;
-
- initSetupGridCUDA<<<numBlocksNB,threadsPerBlockNB>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initSetupGrid2CUDA<<<numBlocksNB,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initCUDA<<<numBlocksNB,threadsPerBlockNB>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- runCUDA<<<numBlocksFS,threadsPerBlockFS>>>(this->cudaSolver,0,0);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- }
- }
-
- //data.setLike(dofVector.getData());
- //cudaMemcpy(data.getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaMemcpy(dofVector.getData().getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- //data.save("u-00001.tnl");
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
- // 1 - with curve, 2 - to the north of curve, 4 - to the south of curve,
- // 8 - to the east of curve, 16 - to the west of curve.
- int subgridID = i/NARROWBAND_SUBGRID_SIZE + (j/NARROWBAND_SUBGRID_SIZE) * statusGridSize;
- if(cudaStatusVector[subgridID] != 0 && i<Mesh.getDimensions().x() && j < Mesh.getDimensions().y())
- {
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real value = cudaDofVector2[Entity.getIndex()];
- Real a,b, tmp;
-
- if( i == 0 /*|| (i/NARROWBAND_SUBGRID_SIZE == 0 && !(cudaStatusVector[subgridID] & 9))*/ )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 /*|| (i/NARROWBAND_SUBGRID_SIZE == NARROWBAND_SUBGRID_SIZE - 1 && !(cudaStatusVector[subgridID] & 17))*/ )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- else
- {
- a = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] );
- }
-
- if( j == 0 /*|| (j/NARROWBAND_SUBGRID_SIZE == 0 && !(cudaStatusVector[subgridID] & 3))*/ )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()];
- else if( j == Mesh.getDimensions().y() - 1 /* || (j/NARROWBAND_SUBGRID_SIZE == NARROWBAND_SUBGRID_SIZE - 1 && !(cudaStatusVector[subgridID] & 5)) */)
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- else
- {
- b = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] );
- }
-
-
- if(abs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
- // cudaDofVector2[Entity.getIndex()] = fabsMin(value, tmp);
- atomicFabsMin(&(cudaDofVector2[Entity.getIndex()]), tmp);
- }
-
-}
-
-
-__global__ void initCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
-
-
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
- solver->initGrid();
- }
-
-
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- int j = blockDim.y*blockIdx.y + threadIdx.y;
-
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
-
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- int gid = Entity.getIndex();
-
- if(abs(cudaDofVector2[gid]) > 1.5*h)
- cudaDofVector2[gid] = INT_MAX*sign(cudaDofVector2[gid]);
-
-// if (i >0 && j > 0 && i+1 < Mesh.getDimensions().x() && j+1 < Mesh.getDimensions().y())
-// {
-// if(cudaDofVector2[gid]*cudaDofVector2[gid+1] <= 0 )
-// {
-// cudaDofVector2[gid] = sign(cudaDofVector2[gid])*0.5*h;
-// cudaDofVector2[gid+1] = sign(cudaDofVector2[gid+1])*0.5*h;
-// }
-// if( cudaDofVector2[gid]*cudaDofVector2[gid+Mesh.getDimensions().x()] <= 0 )
-// {
-// cudaDofVector2[gid] = sign(cudaDofVector2[gid])*0.5*h;
-// cudaDofVector2[gid+Mesh.getDimensions().x()] = sign(cudaDofVector2[gid+Mesh.getDimensions().x()])*0.5*h;
-// }
-//
-// if(cudaDofVector2[gid]*cudaDofVector2[gid-1] <= 0 )
-// {
-// cudaDofVector2[gid] = sign(cudaDofVector2[gid])*0.5*h;
-// cudaDofVector2[gid-1] = sign(cudaDofVector2[gid-1])*0.5*h;
-// }
-// if( cudaDofVector2[gid]*cudaDofVector2[gid-Mesh.getDimensions().x()] <= 0 )
-// {
-// cudaDofVector2[gid] = sign(cudaDofVector2[gid])*0.5*h;
-// cudaDofVector2[gid-Mesh.getDimensions().x()] = sign(cudaDofVector2[gid-Mesh.getDimensions().x()])*0.5*h;
-// }
-// }
-
-
-//
-
-
-
-
-
-
-// if(i+1 < Mesh.getDimensions().x() && j+1 < Mesh.getDimensions().y() )
-// {
-// if(cudaDofVector[Entity.getIndex()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1111(i,j);
-// else
-// setupSquare1110(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1101(i,j);
-// else
-// setupSquare1100(i,j);
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1011(i,j);
-// else
-// setupSquare1010(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1001(i,j);
-// else
-// setupSquare1000(i,j);
-// }
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0111(i,j);
-// else
-// setupSquare0110(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0101(i,j);
-// else
-// setupSquare0100(i,j);
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0011(i,j);
-// else
-// setupSquare0010(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0001(i,j);
-// else
-// setupSquare0000(i,j);
-// }
-// }
-// }
-//
-// }
-
- return true;
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
- //Real fy = abs(y);
-
- //Real tmpMin = Min(fx,abs(y));
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i)
-{
-
-
- int gx = 0;
- int gy = threadIdx.y;
- //if(solver->Mesh.getDimensions().x() <= gx || solver->Mesh.getDimensions().y() <= gy)
- // return;
- int n = solver->Mesh.getDimensions().x();
- int blockCount = n/blockDim.y +1;
- //int gid = solver->Mesh.getDimensions().x() * gy + gx;
- //int max = solver->Mesh.getDimensions().x()*solver->Mesh.getDimensions().x();
-
- //int id1 = gx+gy;
- //int id2 = (solver->Mesh.getDimensions().x() - gx - 1) + gy;
-
- if(blockIdx.x==0)
- {
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==1)
- {
- gx=n-1;
- gy=threadIdx.y;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==2)
- {
- gx=0;
- gy=n-threadIdx.y-1;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==3)
- {
- gx=n-1;
- gy=n-threadIdx.y-1;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
-
-}
-
-
-
-
-__global__ void initSetupGridCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
- __shared__ double u0;
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
-
-// printf("Hello from block = %d, thread = %d, x = %d, y = %d\n", blockIdx.x + gridDim.x*blockIdx.y,(blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + threadIdx.x, threadIdx.x, threadIdx.y);
- if(threadIdx.x+threadIdx.y == 0)
- {
-// printf("Hello from block = %d, thread = %d, x = %d, y = %d\n", blockIdx.x + gridDim.x*blockIdx.y,(blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + threadIdx.x, threadIdx.x, threadIdx.y);
-
- if(blockIdx.x+blockIdx.y == 0)
- *(solver->reinitialize) = 0;
-
- solver->cudaStatusVector[blockIdx.x + gridDim.x*blockIdx.y] = 0;
-
- u0 = solver->cudaDofVector2[(blockDim.y*blockIdx.y + 0)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + 0];
- }
- __syncthreads();
-
- double u = solver->cudaDofVector2[(blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + threadIdx.x];
-
- if(u*u0 <=0.0)
- atomicMax(&(solver->cudaStatusVector[blockIdx.x + gridDim.x*blockIdx.y]),1);
- }
-// if(threadIdx.x+threadIdx.y == 0)
-
-// printf("Bye from block = %d, thread = %d, x = %d, y = %d\n", blockIdx.x + gridDim.x*blockIdx.y,(blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + threadIdx.x, threadIdx.x, threadIdx.y);
-
-
-}
-
-
-
-// run this with one thread per block
-__global__ void initSetupGrid2CUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
-// printf("Hello\n");
- if(solver->cudaStatusVector[blockIdx.x + gridDim.x*blockIdx.y] == 1)
- {
-// 1 - with curve, 2 - to the north of curve, 4 - to the south of curve,
-// 8 - to the east of curve, 16 - to the west of curve.
- if(blockIdx.x > 0)
- {
- atomicAdd(&(solver->cudaStatusVector[blockIdx.x - 1 + gridDim.x*blockIdx.y]), 16);
- }
-
- if(blockIdx.x < gridDim.x - 1)
- atomicAdd(&(solver->cudaStatusVector[blockIdx.x + 1 + gridDim.x*blockIdx.y]), 8);
-
- if(blockIdx.y > 0 )
- atomicAdd(&(solver->cudaStatusVector[blockIdx.x + gridDim.x*(blockIdx.y - 1)]), 4);
-
- if(blockIdx.y < gridDim.y - 1)
- atomicAdd(&(solver->cudaStatusVector[blockIdx.x + gridDim.x*(blockIdx.y + 1)]), 2);
- }
-
-
-}
-
-
-
-
-
-__global__ void runNarrowBandCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, double tau)
-{
- int gid = (blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x()+ threadIdx.x;
- int i = threadIdx.x + blockIdx.x*blockDim.x;
- int j = threadIdx.y + blockIdx.y*blockDim.y;
-
-// if(i+j == 0)
-// printf("Hello\n");
-
- int blockID = blockIdx.x + blockIdx.y*gridDim.x; /*i/NARROWBAND_SUBGRID_SIZE + (j/NARROWBAND_SUBGRID_SIZE) * ((Mesh.getDimensions().x() + NARROWBAND_SUBGRID_SIZE-1 ) / NARROWBAND_SUBGRID_SIZE);*/
-
- int status = solver->cudaStatusVector[blockID];
-
- if(solver->Mesh.getDimensions().x() > i && solver->Mesh.getDimensions().y() > j)
- {
-
- if(status != 0)
- {
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(solver->Mesh);
- Entity.setCoordinates(Containers::StaticVector<2,double>(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- double value = solver->cudaDofVector2[Entity.getIndex()];
- double xf,xb,yf,yb, grad, fu, a,b;
- a = b = 0.0;
-
- if( i == 0 || (threadIdx.x == 0 && !(status & 9)) )
- {
- xb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- xf = solver->cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] - value;
- }
- else if( i == solver->Mesh.getDimensions().x() - 1 || (threadIdx.x == blockDim.x - 1 && !(status & 17)) )
- {
- xb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- xf = solver->cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()] - value;
- }
- else
- {
- xb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- xf = solver-> cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] - value;
- }
-
- if( j == 0 || (threadIdx.y == 0 && !(status & 3)) )
- {
- yb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] ;
- yf = solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] - value;
- }
- else if( j == solver->Mesh.getDimensions().y() - 1 || (threadIdx.y == blockDim.y - 1 && !(status & 5)) )
- {
- yb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- yf = solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()] - value;
- }
- else
- {
- yb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- yf = solver-> cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] - value;
- }
- __syncthreads();
-
-
-
-
-
- if(sign(value) >= 0.0)
- {
- xf = solver->negativePart(xf);
-
- xb = solver->positivePart(xb);
-
- yf = solver->negativePart(yf);
-
- yb = solver->positivePart(yb);
-
- }
- else
- {
-
- xb = solver->negativePart(xb);
-
- xf = solver->positivePart(xf);
-
- yb = solver->negativePart(yb);
-
- yf = solver->positivePart(yf);
- }
-
-
- if(xb > xf)
- a = xb*solver->Mesh.template getSpaceStepsProducts< -1, 0 >();
- else
- a = xf*solver->Mesh.template getSpaceStepsProducts< -1, 0 >();
-
- if(yb > yf)
- b = yb*solver->Mesh.template getSpaceStepsProducts< 0, -1 >();
- else
- b = yf*solver->Mesh.template getSpaceStepsProducts< 0, -1 >();
-
-
-
-// grad = sqrt(0.5 * (xf*xf + xb*xb + yf*yf + yb*yb ) )*solver->Mesh.template getSpaceStepsProducts< -1, 0 >();
-
- grad = sqrt(/*0.5 **/ (a*a + b*b ) );
-
- fu = -1.0 * grad;
-
- if((tau*fu+value)*value <=0 )
- {
- // 1 - with curve, 2 - to the north of curve, 4 - to the south of curve,
- // 8 - to the east of curve, 16 - to the west of curve.
-
- if((threadIdx.x == 6 && !(status & 9)) && (blockIdx.x > 0) )
- atomicMax(solver->reinitialize,1);
- else if((threadIdx.x == blockDim.x - 7 && !(status & 17)) && (blockIdx.x < gridDim.x - 1) )
- atomicMax(solver->reinitialize,1);
- else if((threadIdx.y == 6 && !(status & 3)) && (blockIdx.y > 0) )
- atomicMax(solver->reinitialize,1);
- else if((threadIdx.y == blockDim.y - 7 && !(status & 5)) && (blockIdx.y < gridDim.y - 1) )
- atomicMax(solver->reinitialize,1);
- }
-
- solver->cudaDofVector2[Entity.getIndex()] += tau*fu;
- }
- }
-}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(INT_MAX,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(-INT_MAX,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(cudaDofVector[Entity.getIndex()],cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(cudaDofVector[Entity.getIndex()],cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-}
-#endif
-
-
-
-
-#endif /* TNLNARROWBAND_IMPL_H_ */
diff --git a/src/Examples/narrow-band/tnlNarrowBand2D_CUDA_v5_impl.h b/src/Examples/narrow-band/tnlNarrowBand2D_CUDA_v5_impl.h
deleted file mode 100644
index c92810490069164d9f170249c328304576773d20..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/tnlNarrowBand2D_CUDA_v5_impl.h
+++ /dev/null
@@ -1,1313 +0,0 @@
-/***************************************************************************
- tnlNarrowBand2D_CUDA_v4_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLNARROWBAND2D_IMPL_H_
-#define TNLNARROWBAND2D_IMPL_H_
-
-#define NARROWBAND_SUBGRID_SIZE 32
-
-#include "tnlNarrowBand.h"
-
-__device__
-double fabsMin( double x, double y)
-{
- double fx = abs(x);
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-}
-
-__device__
-double atomicFabsMin(double* address, double val)
-{
- unsigned long long int* address_as_ull =
- (unsigned long long int*)address;
- unsigned long long int old = *address_as_ull, assumed;
- do {
- assumed = old;
- old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(__longlong_as_double(assumed),val) ));
- } while (assumed != old);
- return __longlong_as_double(old);
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-#ifdef HAVE_CUDA
- __device__ __host__
-#endif
-Real tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >:: positivePart(const Real arg) const
-{
- if(arg > 0.0)
- return arg;
- return 0.0;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-#ifdef HAVE_CUDA
- __device__ __host__
-#endif
-Real tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: negativePart(const Real arg) const
-{
- if(arg < 0.0)
- return -arg;
- return 0.0;
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlNarrowBand< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: tnlNarrowBand()
-:dofVector(Mesh)
-{
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- h = Mesh.template getSpaceStepsProducts< 1, 0 >();
- //Entity.refresh();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
- tau = parameters.getParameter< double >( "tau" );
-
- finalTime = parameters.getParameter< double >( "final-time" );
-
- statusGridSize = ((Mesh.getDimensions().x() + NARROWBAND_SUBGRID_SIZE-1 ) / NARROWBAND_SUBGRID_SIZE);
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaStatusVector), statusGridSize*statusGridSize*sizeof(int));
-// cudaMemcpy(cudaDofVector, this->dofVector.getData().getData(), statusGridSize*statusGridSize* sizeof(int)), cudaMemcpyHostToDevice);
-
- cudaMalloc(&reinitialize, sizeof(int));
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-#endif
-
- int n = Mesh.getDimensions().x();
-
- dim3 threadsPerBlock2(NARROWBAND_SUBGRID_SIZE, NARROWBAND_SUBGRID_SIZE);
- dim3 numBlocks2(statusGridSize ,statusGridSize);
- initSetupGridCUDA<<<numBlocks2,threadsPerBlock2>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initSetupGrid2CUDA<<<numBlocks2,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
-
- /*dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 + 1 ,n/16 +1);*/
- initCUDA<<<numBlocks2,threadsPerBlock2>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
-
- cout << "Solver initialized." <<std::endl;
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlockFS(1, 512);
- dim3 numBlocksFS(4,1);
- dim3 threadsPerBlockNB(NARROWBAND_SUBGRID_SIZE, NARROWBAND_SUBGRID_SIZE);
- dim3 numBlocksNB(n/NARROWBAND_SUBGRID_SIZE + 1,n/NARROWBAND_SUBGRID_SIZE + 1);
-
- double time = 0.0;
- int reinit = 0;
-
- cout << "Hi!" <<std::endl;
- runCUDA<<<numBlocksFS,threadsPerBlockFS>>>(this->cudaSolver,0,0);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- cout << "Hi2!" <<std::endl;
- while(time < finalTime)
- {
- if(tau+time > finalTime)
- tau=finalTime-time;
-
- runNarrowBandCUDA<<<numBlocksNB,threadsPerBlockNB>>>(this->cudaSolver,tau);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- time += tau;
-
-
- cudaMemcpy(&reinit, this->reinitialize, sizeof(int), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- if(reinit != 0 /*&& time != finalTime */)
- {
- cout << time <<std::endl;
-
- initSetupGridCUDA<<<numBlocksNB,threadsPerBlockNB>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initSetupGrid2CUDA<<<numBlocksNB,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initCUDA<<<numBlocksNB,threadsPerBlockNB>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- runCUDA<<<numBlocksFS,threadsPerBlockFS>>>(this->cudaSolver,0,0);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- }
- }
-
- //data.setLike(dofVector.getData());
- //cudaMemcpy(data.getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaMemcpy(dofVector.getData().getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- //data.save("u-00001.tnl");
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
- // 1 - with curve, 2 - to the north of curve, 4 - to the south of curve,
- // 8 - to the east of curve, 16 - to the west of curve.
- int subgridID = i/NARROWBAND_SUBGRID_SIZE + (j/NARROWBAND_SUBGRID_SIZE) * ((Mesh.getDimensions().x() + NARROWBAND_SUBGRID_SIZE-1 ) / NARROWBAND_SUBGRID_SIZE);
- if(/*cudaStatusVector[subgridID] != 0 &&*/ i<Mesh.getDimensions().x() && Mesh.getDimensions().y())
- {
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real value = cudaDofVector2[Entity.getIndex()];
- Real a,b, tmp;
-
- if( i == 0 /*|| (i/NARROWBAND_SUBGRID_SIZE == 0 && !(cudaStatusVector[subgridID] & 9)) */)
- a = cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 /*|| (i/NARROWBAND_SUBGRID_SIZE == NARROWBAND_SUBGRID_SIZE - 1 && !(cudaStatusVector[subgridID] & 17)) */)
- a = cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- else
- {
- a = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] );
- }
-
- if( j == 0/* || (j/NARROWBAND_SUBGRID_SIZE == 0 && !(cudaStatusVector[subgridID] & 3)) */)
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()];
- else if( j == Mesh.getDimensions().y() - 1 /* || (j/NARROWBAND_SUBGRID_SIZE == NARROWBAND_SUBGRID_SIZE - 1 && !(cudaStatusVector[subgridID] & 5))*/ )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- else
- {
- b = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] );
- }
-
-
- if(abs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
- // cudaDofVector2[Entity.getIndex()] = fabsMin(value, tmp);
- atomicFabsMin(&(cudaDofVector2[Entity.getIndex()]), tmp);
- }
-
-}
-
-
-__global__ void initCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
-
-
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
- solver->initGrid();
- }
-
-
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- int j = blockDim.y*blockIdx.y + threadIdx.y;
-
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
-
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- int gid = Entity.getIndex();
-
- cudaDofVector2[gid] = INT_MAX*sign(cudaDofVector2[gid]);
-
- if (i >0 && j > 0 && i+1 < Mesh.getDimensions().x() && j+1 < Mesh.getDimensions().y())
- {
- if(cudaDofVector2[gid]*cudaDofVector2[gid+1] <= 0 )
- {
- cudaDofVector2[gid] = sign(cudaDofVector2[gid])*0.5*h;
- cudaDofVector2[gid+1] = sign(cudaDofVector2[gid+1])*0.5*h;
- }
- if( cudaDofVector2[gid]*cudaDofVector2[gid+Mesh.getDimensions().x()] <= 0 )
- {
- cudaDofVector2[gid] = sign(cudaDofVector2[gid])*0.5*h;
- cudaDofVector2[gid+Mesh.getDimensions().x()] = sign(cudaDofVector2[gid+Mesh.getDimensions().x()])*0.5*h;
- }
-
- if(cudaDofVector2[gid]*cudaDofVector2[gid-1] <= 0 )
- {
- cudaDofVector2[gid] = sign(cudaDofVector2[gid])*0.5*h;
- cudaDofVector2[gid-1] = sign(cudaDofVector2[gid-1])*0.5*h;
- }
- if( cudaDofVector2[gid]*cudaDofVector2[gid-Mesh.getDimensions().x()] <= 0 )
- {
- cudaDofVector2[gid] = sign(cudaDofVector2[gid])*0.5*h;
- cudaDofVector2[gid-Mesh.getDimensions().x()] = sign(cudaDofVector2[gid-Mesh.getDimensions().x()])*0.5*h;
- }
- }
-
-
-//
-
-
-
-
-
-
-// if(i+1 < Mesh.getDimensions().x() && j+1 < Mesh.getDimensions().y() )
-// {
-// if(cudaDofVector[Entity.getIndex()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1111(i,j);
-// else
-// setupSquare1110(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1101(i,j);
-// else
-// setupSquare1100(i,j);
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1011(i,j);
-// else
-// setupSquare1010(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1001(i,j);
-// else
-// setupSquare1000(i,j);
-// }
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0111(i,j);
-// else
-// setupSquare0110(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0101(i,j);
-// else
-// setupSquare0100(i,j);
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0011(i,j);
-// else
-// setupSquare0010(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0001(i,j);
-// else
-// setupSquare0000(i,j);
-// }
-// }
-// }
-//
-// }
-
- return true;
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
- //Real fy = abs(y);
-
- //Real tmpMin = Min(fx,abs(y));
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i)
-{
-
-
- int gx = 0;
- int gy = threadIdx.y;
- //if(solver->Mesh.getDimensions().x() <= gx || solver->Mesh.getDimensions().y() <= gy)
- // return;
- int n = solver->Mesh.getDimensions().x();
- int blockCount = n/blockDim.y +1;
- //int gid = solver->Mesh.getDimensions().x() * gy + gx;
- //int max = solver->Mesh.getDimensions().x()*solver->Mesh.getDimensions().x();
-
- //int id1 = gx+gy;
- //int id2 = (solver->Mesh.getDimensions().x() - gx - 1) + gy;
-
- if(blockIdx.x==0)
- {
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==1)
- {
- gx=n-1;
- gy=threadIdx.y;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==2)
- {
- gx=0;
- gy=n-threadIdx.y-1;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==3)
- {
- gx=n-1;
- gy=n-threadIdx.y-1;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
-
-}
-
-
-
-
-__global__ void initSetupGridCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
- __shared__ double u0;
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
-
-// printf("Hello from block = %d, thread = %d, x = %d, y = %d\n", blockIdx.x + gridDim.x*blockIdx.y,(blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + threadIdx.x, threadIdx.x, threadIdx.y);
- if(threadIdx.x+threadIdx.y == 0)
- {
-// printf("Hello from block = %d, thread = %d, x = %d, y = %d\n", blockIdx.x + gridDim.x*blockIdx.y,(blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + threadIdx.x, threadIdx.x, threadIdx.y);
-
- if(blockIdx.x+blockIdx.y == 0)
- *(solver->reinitialize) = 0;
-
- solver->cudaStatusVector[blockIdx.x + gridDim.x*blockIdx.y] = 0;
-
- u0 = solver->cudaDofVector2[(blockDim.y*blockIdx.y + 0)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + 0];
- }
- __syncthreads();
-
- double u = solver->cudaDofVector2[(blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + threadIdx.x];
-
- if(u*u0 <=0.0)
- atomicMax(&(solver->cudaStatusVector[blockIdx.x + gridDim.x*blockIdx.y]),1);
- }
-// if(threadIdx.x+threadIdx.y == 0)
-
-// printf("Bye from block = %d, thread = %d, x = %d, y = %d\n", blockIdx.x + gridDim.x*blockIdx.y,(blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x() + blockDim.x*blockIdx.x + threadIdx.x, threadIdx.x, threadIdx.y);
-
-
-}
-
-
-
-// run this with one thread per block
-__global__ void initSetupGrid2CUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
-// printf("Hello\n");
- if(solver->cudaStatusVector[blockIdx.x + gridDim.x*blockIdx.y] == 1)
- {
-// 1 - with curve, 2 - to the north of curve, 4 - to the south of curve,
-// 8 - to the east of curve, 16 - to the west of curve.
- if(blockIdx.x > 0)
- atomicAdd(&(solver->cudaStatusVector[blockIdx.x - 1 + gridDim.x*blockIdx.y]), 16);
-
- if(blockIdx.x < gridDim.x - 1)
- atomicAdd(&(solver->cudaStatusVector[blockIdx.x + 1 + gridDim.x*blockIdx.y]), 8);
-
- if(blockIdx.y > 0 )
- atomicAdd(&(solver->cudaStatusVector[blockIdx.x + gridDim.x*(blockIdx.y - 1)]), 4);
-
- if(blockIdx.y < gridDim.y - 1)
- atomicAdd(&(solver->cudaStatusVector[blockIdx.x + gridDim.x*(blockIdx.y + 1)]), 2);
- }
-
-
-}
-
-
-
-
-
-__global__ void runNarrowBandCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, double tau)
-{
- int gid = (blockDim.y*blockIdx.y + threadIdx.y)*solver->Mesh.getDimensions().x()+ threadIdx.x;
- int i = threadIdx.x + blockIdx.x*blockDim.x;
- int j = threadIdx.y + blockIdx.y*blockDim.y;
-
-// if(i+j == 0)
-// printf("Hello\n");
-
- int blockID = blockIdx.x + blockIdx.y*gridDim.x; /*i/NARROWBAND_SUBGRID_SIZE + (j/NARROWBAND_SUBGRID_SIZE) * ((Mesh.getDimensions().x() + NARROWBAND_SUBGRID_SIZE-1 ) / NARROWBAND_SUBGRID_SIZE);*/
-
- int status = solver->cudaStatusVector[blockID];
-
- if(solver->Mesh.getDimensions().x() > i && solver->Mesh.getDimensions().y() > j)
- {
-
-// if(status != 0)
- {
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(solver->Mesh);
- Entity.setCoordinates(Containers::StaticVector<2,double>(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- double value = solver->cudaDofVector2[Entity.getIndex()];
- double xf,xb,yf,yb, grad, fu, a,b;
- a = b = 0.0;
-
- if( i == 0 /*|| (threadIdx.x == 0 && !(status & 9)) */)
- {
- xb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- xf = solver->cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] - value;
- }
- else if( i == solver->Mesh.getDimensions().x() - 1 /*|| (threadIdx.x == blockDim.x - 1 && !(status & 17)) */)
- {
- xb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- xf = solver->cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()] - value;
- }
- else
- {
- xb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- xf = solver-> cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] - value;
- }
-
- if( j == 0/* || (threadIdx.y == 0 && !(status & 3))*/ )
- {
- yb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] ;
- yf = solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] - value;
- }
- else if( j == solver->Mesh.getDimensions().y() - 1 /*|| (threadIdx.y == blockDim.y - 1 && !(status & 5)) */)
- {
- yb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- yf = solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()] - value;
- }
- else
- {
- yb = value - solver->cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- yf = solver-> cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] - value;
- }
- __syncthreads();
-
-
-
-
-
- if(sign(value) > 0.0)
- {
- xf = solver->negativePart(xf);
-
- xb = solver->positivePart(xb);
-
- yf = solver->negativePart(yf);
-
- yb = solver->positivePart(yb);
-
- }
- else
- {
-
- xb = solver->negativePart(xb);
-
- xf = solver->positivePart(xf);
-
- yb = solver->negativePart(yb);
-
- yf = solver->positivePart(yf);
- }
-
-
- if(xb > xf)
- a = xb*solver->Mesh.template getSpaceStepsProducts< -1, 0 >();
- else
- a = xf*solver->Mesh.template getSpaceStepsProducts< -1, 0 >();
-
- if(yb > yf)
- b = yb*solver->Mesh.template getSpaceStepsProducts< 0, -1 >();
- else
- b = yf*solver->Mesh.template getSpaceStepsProducts< 0, -1 >();
-
-
-
-// grad = sqrt(0.5 * (xf*xf + xb*xb + yf*yf + yb*yb ) )*solver->Mesh.template getSpaceStepsProducts< -1, 0 >();
-
- grad = sqrt(/*0.5 **/ (a*a + b*b ) );
-
- fu = -1.0 * grad;
-
-// if((tau*fu+value)*value <=0 )
-// {
-// // 1 - with curve, 2 - to the north of curve, 4 - to the south of curve,
-// // 8 - to the east of curve, 16 - to the west of curve.
-//
-// if((threadIdx.x == 1 && !(status & 9)) && (blockIdx.x > 0) )
-// atomicMax(solver->reinitialize,1);
-// else if((threadIdx.x == blockDim.x - 2 && !(status & 17)) && (blockIdx.x < gridDim.x - 1) )
-// atomicMax(solver->reinitialize,1);
-// else if((threadIdx.y == 1 && !(status & 3)) && (blockIdx.y > 0) )
-// atomicMax(solver->reinitialize,1);
-// else if((threadIdx.y == blockDim.y - 2 && !(status & 5)) && (blockIdx.y < gridDim.y - 1) )
-// atomicMax(solver->reinitialize,1);
-// }
-
- solver->cudaDofVector2[Entity.getIndex()] += tau*fu;
- }
- }
-}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(INT_MAX,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(-INT_MAX,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(cudaDofVector[Entity.getIndex()],cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(cudaDofVector[Entity.getIndex()],cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-}
-#endif
-
-
-
-
-#endif /* TNLNARROWBAND_IMPL_H_ */
diff --git a/src/Examples/narrow-band/tnlNarrowBand2D_impl.h b/src/Examples/narrow-band/tnlNarrowBand2D_impl.h
deleted file mode 100644
index d42bc2a7610d5bb02a94031d446a61ac5a2e6579..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/tnlNarrowBand2D_impl.h
+++ /dev/null
@@ -1,927 +0,0 @@
-/***************************************************************************
- tnlNarrowBand2D_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLNARROWBAND2D_IMPL_H_
-#define TNLNARROWBAND2D_IMPL_H_
-
-#include "tnlNarrowBand.h"
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlNarrowBand< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: tnlNarrowBand()
-:Entity(Mesh),
- dofVector(Mesh),
- dofVector2(Mesh)
-{
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
- dofVector2.load(initialCondition);
-
- h = Mesh.template getSpaceStepsProducts< 1, 0 >();
- Entity.refresh();
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
- cout << "a" <<std::endl;
- return initGrid();
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
-
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- for(int i=0; i< Mesh.getDimensions().x()*Mesh.getDimensions().x();i++)
- {
- dofVector2[i]=INT_MAX*sign(dofVector[i]);
- }
-
- for(int i = 0 ; i < Mesh.getDimensions().x()-1; i++)
- {
- for(int j = 0 ; j < Mesh.getDimensions().x()-1; j++)
- {
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- neighborEntities.refresh(Mesh,Entity.getIndex());
-
- if(dofVector[this->Entity.getIndex()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1111(i,j);
- else
- setupSquare1110(i,j);
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1101(i,j);
- else
- setupSquare1100(i,j);
- }
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1011(i,j);
- else
- setupSquare1010(i,j);
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1001(i,j);
- else
- setupSquare1000(i,j);
- }
- }
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0111(i,j);
- else
- setupSquare0110(i,j);
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0101(i,j);
- else
- setupSquare0100(i,j);
- }
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0011(i,j);
- else
- setupSquare0010(i,j);
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0001(i,j);
- else
- setupSquare0000(i,j);
- }
- }
- }
-
- }
- }
- cout << "a" <<std::endl;
-
-// Real tmp = 0.0;
-// Real ax=0.5/sqrt(2.0);
-//
-// if(!exactInput)
-// {
-// for(Index i = 0; i < Mesh.getDimensions().x()*Mesh.getDimensions().y(); i++)
-// dofVector[i]=0.5*h*sign(dofVector[i]);
-// }
-//
-//
-// for(Index i = 1; i < Mesh.getDimensions().x()-1; i++)
-// {
-// for(Index j = 1; j < Mesh.getDimensions().y()-1; j++)
-// {
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-// }
-//
-//
-//
-// for(int i = 1; i < Mesh.getDimensions().x()-1; i++)
-// {
-// Index j = 0;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-//
-// for(int i = 1; i < Mesh.getDimensions().x()-1; i++)
-// {
-// Index j = Mesh.getDimensions().y() - 1;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-//
-// for(int j = 1; j < Mesh.getDimensions().y()-1; j++)
-// {
-// Index i = 0;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-//
-// for(int j = 1; j < Mesh.getDimensions().y()-1; j++)
-// {
-// Index i = Mesh.getDimensions().x() - 1;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-//
-//
-// Index i = Mesh.getDimensions().x() - 1;
-// Index j = Mesh.getDimensions().y() - 1;
-//
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-// if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp > 0.0 &&
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp > 0.0)
-//
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-//
-//
-//
-// j = 0;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-// if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp > 0.0 &&
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp > 0.0)
-//
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-//
-//
-//
-// i = 0;
-// j = Mesh.getDimensions().y() -1;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-// if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp > 0.0 &&
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp > 0.0)
-//
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-//
-//
-//
-// j = 0;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-// if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp > 0.0 &&
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp > 0.0)
-//
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-
- //data.setLike(dofVector2.getData());
- //data=dofVector2.getData();
- //cout << data.getType() <<std::endl;
- dofVector2.save("u-00000.tnl");
- //dofVector2.getData().save("u-00000.tnl");
-
- return true;
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
-
-// data.setLike(dofVector2.getData());
-// data = dofVector2.getData();
-// cout << data.getType() <<std::endl;
- dofVector2.save("u-00001.tnl");
- //dofVector2.getData().save("u-00001.tnl");
-
- return true;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
-
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- Real value = dofVector2[Entity.getIndex()];
- Real a,b, tmp;
-
- if( i == 0 )
- a = dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = dofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- else
- {
- a = fabsMin( dofVector2[neighborEntities.template getEntityIndex< -1, 0 >()],
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] );
- }
-
- if( j == 0 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- else
- {
- b = fabsMin( dofVector2[neighborEntities.template getEntityIndex< 0, -1 >()],
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] );
- }
-
-
- if(fabs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
-
- dofVector2[Entity.getIndex()] = fabsMin(value, tmp);
-
-// if(dofVector2[Entity.getIndex()] > 1.0)
-// cout << value << " " << tmp << " " << dofVector2[Entity.getIndex()] <<std::endl;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-Real tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = fabs(x);
- Real fy = fabs(y);
-
- Real tmpMin = Min(fx,fy);
-
- if(tmpMin == fx)
- return x;
- else
- return y;
-
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-{
-// this->Entity.setCoordinates(CoordinatesType(i,j));
-// this->Entity.refresh();
-// auto neighborEntities = Entity.getNeighborEntities();
-// dofVector2[Entity.getIndex()]=fabsMin(INT_MAX,dofVector2[Entity.getIndex()]);
-// dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-{
-// this->Entity.setCoordinates(CoordinatesType(i,j));
-// this->Entity.refresh();
-// auto neighborEntities = Entity.getNeighborEntities();
-// dofVector2[Entity.getIndex()]=fabsMin(-INT_MAX,dofVector2[(Entity.getIndex())]);
-// dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- dofVector2[Entity.getIndex()]=fabsMin(dofVector[Entity.getIndex()],dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- dofVector2[Entity.getIndex()]=fabsMin(dofVector[Entity.getIndex()],dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-}
-
-
-
-
-#endif /* TNLNARROWBAND_IMPL_H_ */
diff --git a/src/Examples/narrow-band/tnlNarrowBand3D_CUDA_impl.h b/src/Examples/narrow-band/tnlNarrowBand3D_CUDA_impl.h
deleted file mode 100644
index d362f249a79112aa9b902f86cd1a304702292423..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/tnlNarrowBand3D_CUDA_impl.h
+++ /dev/null
@@ -1,961 +0,0 @@
-/***************************************************************************
- tnlNarrowBand2D_CUDA_v4_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLNARROWBAND3D_IMPL_H_
-#define TNLNARROWBAND3D_IMPL_H_
-
-#include "tnlNarrowBand.h"
-
-//__device__
-//double fabsMin( double x, double y)
-//{
-// double fx = abs(x);
-//
-// if(Min(fx,abs(y)) == fx)
-// return x;
-// else
-// return y;
-//}
-//
-//__device__
-//double atomicFabsMin(double* address, double val)
-//{
-// unsigned long long int* address_as_ull =
-// (unsigned long long int*)address;
-// unsigned long long int old = *address_as_ull, assumed;
-// do {
-// assumed = old;
-// old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(assumed,val) ));
-// } while (assumed != old);
-// return __longlong_as_double(old);
-//}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlNarrowBand< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- this->h = Mesh.template getSpaceStepsProducts< 1, 0, 0 >();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-#endif
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(8, 8,8);
- dim3 numBlocks(n/8 + 1, n/8 +1, n/8 +1);
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(1, 512);
- dim3 numBlocks(8,1);
-
-
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,0,0);
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- cudaMemcpy(this->dofVector.getData().getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j, Index k)
-{
- tnlGridEntity< tnlGrid< 3,double, TNL::Devices::Host, int >, 3, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j,k));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage >,3> neighborEntities(Entity);
- Real value = cudaDofVector2[Entity.getIndex()];
- Real a,b,c, tmp;
-
- if( i == 0 )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0, 0 >()];
- else
- {
- a = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0, 0 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0, 0 >()] );
- }
-
- if( j == 0 )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1, 0 >()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1, 0 >()];
- else
- {
- b = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1, 0 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1, 0 >()] );
- }
-
- if( k == 0 )
- c = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 0, 1 >()];
- else if( k == Mesh.getDimensions().z() - 1 )
- c = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 0, -1 >()];
- else
- {
- c = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< 0, 0, -1 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 0, 1 >()] );
- }
-
- Real hD = 3.0*h*h - 2.0*(a*a + b*b + c*c - a*b - a*c - b*c);
-
- if(hD < 0.0)
- tmp = fabsMin(a,fabsMin(b,c)) + sign(value)*h;
- else
- tmp = (1.0/3.0) * ( a + b + c + sign(value)*sqrt(hD) );
-
- atomicFabsMin(&cudaDofVector2[Entity.getIndex()],tmp);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid(int i, int j, int k)
-{
- tnlGridEntity< tnlGrid< 3,double, TNL::Devices::Host, int >, 3, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j,k));
- Entity.refresh();
- int gid = Entity.getIndex();
-
- if(abs(cudaDofVector[gid]) < 1.8*h)
- cudaDofVector2[gid] = cudaDofVector[gid];
- else
- cudaDofVector2[gid] = INT_MAX*sign(cudaDofVector[gid]);
-
- return true;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlNarrowBand< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i)
-{
-
- int gx = 0;
- int gy = threadIdx.y;
-
- int n = solver->Mesh.getDimensions().x();
- int blockCount = n/blockDim.y +1;
-
- if(blockIdx.x==0)
- {
- for(int gz = 0; gz < n;gz++)
- {
- gx = 0;
- gy = threadIdx.y;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,gz);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- __syncthreads();
- }
- }
- else if(blockIdx.x==1)
- {
- for(int gz = 0; gz < n;gz++)
- {
- gx=n-1;
- gy=threadIdx.y;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,gz);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==2)
- {
-
- for(int gz = 0; gz < n;gz++)
- {
- gx=0;
- gy=n-threadIdx.y-1;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,gz);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==3)
- {
- for(int gz = 0; gz < n;gz++)
- {
- gx=n-1;
- gy=n-threadIdx.y-1;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,gz);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
-
-
-
-
- else if(blockIdx.x==4)
- {
- for(int gz = n-1; gz > -1;gz--)
- {
- gx = 0;
- gy = threadIdx.y;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,gz);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==5)
- {
- for(int gz = n-1; gz > -1;gz--)
- {
- gx=n-1;
- gy=threadIdx.y;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,gz);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==6)
- {
-
- for(int gz = n-1; gz > -1;gz--)
- {
- gx=0;
- gy=n-threadIdx.y-1;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,gz);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==7)
- {
- for(int gz = n-1; gz > -1;gz--)
- {
- gx=n-1;
- gy=n-threadIdx.y-1;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,gz);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
-
-
-
-
-}
-
-
-__global__ void initCUDA(tnlNarrowBand< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver)
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
- int gz = blockDim.z*blockIdx.z + threadIdx.z;
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy && solver->Mesh.getDimensions().z() > gz)
- {
- solver->initGrid(gx,gy,gz);
- }
-
-
-}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// cudaDofVector2[index]=fabsMin(INT_MAX,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// cudaDofVector2[index]=fabsMin(-INT_MAX,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// a = al-be;
-// b=1.0;
-// c=-al;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// a = al-be;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// cudaDofVector2[index]=fabsMin(cudaDofVector[index],cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)],cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)],cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)],cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//
-//
-//
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// a = al-be;
-// b=1.0;
-// c=-al;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// a = al-be;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// cudaDofVector2[index]=fabsMin(cudaDofVector[index],cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)],cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)],cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)],cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//}
-#endif
-
-
-
-
-#endif /* TNLNARROWBAND_IMPL_H_ */
diff --git a/src/Examples/narrow-band/tnlNarrowBand3D_impl.h b/src/Examples/narrow-band/tnlNarrowBand3D_impl.h
deleted file mode 100644
index 6e63d527b92e0b5c7907a5a2e8b24cba7ed432f2..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/tnlNarrowBand3D_impl.h
+++ /dev/null
@@ -1,307 +0,0 @@
-/***************************************************************************
- tnlNarrowBand2D_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLNARROWBAND3D_IMPL_H_
-#define TNLNARROWBAND3D_IMPL_H_
-
-#include "tnlNarrowBand.h"
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlNarrowBand< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: tnlNarrowBand()
-:Entity(Mesh),
- dofVector(Mesh),
- dofVector2(Mesh)
-{
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
- dofVector2.load(initialCondition);
-
- h = Mesh.template getSpaceStepsProducts< 1, 0, 0 >();
- Entity.refresh();
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-// cout << "bla "<<endl;
- return initGrid();
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- for(int i=0; i< Mesh.getDimensions().x()*Mesh.getDimensions().y()*Mesh.getDimensions().z();i++)
- {
-
- if (abs(dofVector[i]) < 1.8*h)
- dofVector2[i]=dofVector[i];
- else
- dofVector2[i]=INT_MAX*sign(dofVector[i]);
- }
-
- return true;
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- for(Index k = 0; k < Mesh.getDimensions().z(); k++)
- {
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index k = 0; k < Mesh.getDimensions().z(); k++)
- {
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index k = 0; k < Mesh.getDimensions().z(); k++)
- {
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
- for(Index k = 0; k < Mesh.getDimensions().z(); k++)
- {
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
-
-
-
-
-
-
-
- for(Index k = Mesh.getDimensions().z() -1; k > -1; k--)
- {
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index k = Mesh.getDimensions().z() -1; k > -1; k--)
- {
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index k = Mesh.getDimensions().z() -1; k > -1; k--)
- {
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
- for(Index k = Mesh.getDimensions().z() -1; k > -1; k--)
- {
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
-
- dofVector2.save("u-00001.tnl");
-
- cout << "bla 3"<<endl;
- return true;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j, Index k)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j,k));
- this->Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage >,3> neighborEntities(Entity);
- Real value = dofVector2[Entity.getIndex()];
- Real a,b,c, tmp;
-
- if( i == 0 )
- a = dofVector2[neighborEntities.template getEntityIndex< 1, 0, 0>()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = dofVector2[neighborEntities.template getEntityIndex< -1, 0, 0 >()];
- else
- {
- a = fabsMin( dofVector2[neighborEntities.template getEntityIndex< -1, 0, 0>()],
- dofVector2[neighborEntities.template getEntityIndex< 1, 0, 0>()] );
- }
-
- if( j == 0 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, 1, 0>()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, -1, 0>()];
- else
- {
- b = fabsMin( dofVector2[neighborEntities.template getEntityIndex< 0, -1, 0>()],
- dofVector2[neighborEntities.template getEntityIndex< 0, 1, 0>()] );
- }
-
- if( k == 0 )
- c = dofVector2[neighborEntities.template getEntityIndex< 0, 0, 1>()];
- else if( k == Mesh.getDimensions().z() - 1 )
- c = dofVector2[neighborEntities.template getEntityIndex< 0, 0, -1>()];
- else
- {
- c = fabsMin( dofVector2[neighborEntities.template getEntityIndex< 0, 0, -1>()],
- dofVector2[neighborEntities.template getEntityIndex< 0, 0, 1>()] );
- }
-
- Real hD = 3.0*h*h - 2.0*(a*a+b*b+c*c-a*b-a*c-b*c);
-
- if(hD < 0.0)
- tmp = fabsMin(a,fabsMin(b,c)) + sign(value)*h;
- else
- tmp = (1.0/3.0) * ( a + b + c + sign(value)*sqrt(hD) );
-
-
- dofVector2[Entity.getIndex()] = fabsMin(value, tmp);
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-Real tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = fabs(x);
- Real fy = fabs(y);
-
- Real tmpMin = Min(fx,fy);
-
- if(tmpMin == fx)
- return x;
- else
- return y;
-
-}
-
-
-
-#endif /* TNLNARROWBAND_IMPL_H_ */
diff --git a/src/Examples/narrow-band/tnlNarrowBand_CUDA.h b/src/Examples/narrow-band/tnlNarrowBand_CUDA.h
deleted file mode 100644
index ca9b1da2cc6e26b14bc003532b6eea75e89d907d..0000000000000000000000000000000000000000
--- a/src/Examples/narrow-band/tnlNarrowBand_CUDA.h
+++ /dev/null
@@ -1,203 +0,0 @@
-/***************************************************************************
- tnlNarrowBand_CUDA.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLNARROWBAND_H_
-#define TNLNARROWBAND_H_
-
-#include <TNL/Config/ParameterContainer.h>
-#include <TNL/Containers/Vector.h>
-#include <TNL/Containers/StaticVector.h>
-#include <TNL/Devices/Host.h>
-#include <mesh/tnlGrid.h>
-#include <mesh/grids/tnlGridEntity.h>
-
-#include <functions/tnlMeshFunction.h>
-#include <limits.h>
-#include <core/tnlDevice.h>
-#include <ctime>
-
-
-
-
-
-template< typename Mesh,
- typename Real,
- typename Index >
-class tnlNarrowBand
-{};
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 2, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
- tnlNarrowBand();
-
- static String getType();
- bool init( const Config::ParameterContainer& parameters );
- bool run();
-#ifdef HAVE_CUDA
- __device__ __host__
-#endif
- RealType positivePart(const RealType arg) const;
-#ifdef HAVE_CUDA
- __device__ __host__
-#endif
- RealType negativePart(const RealType arg) const;
-
-#ifdef HAVE_CUDA
- __device__ bool initGrid();
- __device__ void updateValue(const Index i, const Index j);
- __device__ void updateValue(const Index i, const Index j, double** sharedMem, const int k3);
- __device__ Real fabsMin(const Real x, const Real y);
-
- tnlNarrowBand< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >* cudaSolver;
- double* cudaDofVector;
- double* cudaDofVector2;
- int* cudaStatusVector;
- int counter;
- int* reinitialize;
- __device__ void setupSquare1000(Index i, Index j);
- __device__ void setupSquare1100(Index i, Index j);
- __device__ void setupSquare1010(Index i, Index j);
- __device__ void setupSquare1001(Index i, Index j);
- __device__ void setupSquare1110(Index i, Index j);
- __device__ void setupSquare1101(Index i, Index j);
- __device__ void setupSquare1011(Index i, Index j);
- __device__ void setupSquare1111(Index i, Index j);
- __device__ void setupSquare0000(Index i, Index j);
- __device__ void setupSquare0100(Index i, Index j);
- __device__ void setupSquare0010(Index i, Index j);
- __device__ void setupSquare0001(Index i, Index j);
- __device__ void setupSquare0110(Index i, Index j);
- __device__ void setupSquare0101(Index i, Index j);
- __device__ void setupSquare0011(Index i, Index j);
- __device__ void setupSquare0111(Index i, Index j);
-#endif
-
- MeshType Mesh;
-
-protected:
-
- int statusGridSize;
- bool exactInput;
-
- tnlMeshFunction<MeshType> dofVector;
- DofVectorType data;
-
-
- RealType h, tau, finalTime;
-
-
-};
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 3, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
-
-
- static String getType();
- bool init( const Config::ParameterContainer& parameters );
- bool run();
-
-#ifdef HAVE_CUDA
- __device__ bool initGrid(int i, int j, int k);
- __device__ void updateValue(const Index i, const Index j, const Index k);
- __device__ void updateValue(const Index i, const Index j, const Index k, double** sharedMem, const int k3);
- __device__ Real fabsMin(const Real x, const Real y);
-
- tnlNarrowBand< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >* cudaSolver;
- double* cudaDofVector;
- double* cudaDofVector2;
- int counter;
-#endif
-
- MeshType Mesh;
-
-protected:
-
-
-
- bool exactInput;
-
- tnlMeshFunction<MeshType> dofVector;
- DofVectorType data;
-
- RealType h;
-
-
-};
-
-
-
-
-
-
-
-#ifdef HAVE_CUDA
-//template<int sweep_t>
-__global__ void runCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i);
-//__global__ void runCUDA(tnlNarrowBand< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i);
-
-__global__ void initCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver);
-
-__global__ void initSetupGridCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver);
-__global__ void initSetupGrid2CUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver);
-__global__ void initSetupGrid1_2CUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver);
-__global__ void runNarrowBandCUDA(tnlNarrowBand< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, double tau);
-//__global__ void initCUDA(tnlNarrowBand< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver);
-#endif
-
-
-
-#include "tnlNarrowBand2D_CUDA_v4_impl.h"
-// #include "tnlNarrowBand3D_CUDA_impl.h"
-
-#endif /* TNLNARROWBAND_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/CMakeLists.txt b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/CMakeLists.txt
deleted file mode 100644
index 3f9db0da04526a8d5fc935f4079cc0bca91b3e56..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/CMakeLists.txt
+++ /dev/null
@@ -1,22 +0,0 @@
-set( tnl_fast_sweeping_map_SOURCES
-# MainBuildConfig.h
-# tnlFastSweepingMap2D_impl.h
-# tnlFastSweepingMap.h
-# fastSweepingMapConfig.h
- main.cpp)
-
-
-IF( BUILD_CUDA )
- CUDA_ADD_EXECUTABLE(fast-sweeping-map main.cu)
-ELSE( BUILD_CUDA )
- ADD_EXECUTABLE(fast-sweeping-map main.cpp)
-ENDIF( BUILD_CUDA )
-target_link_libraries (fast-sweeping-map tnl )
-
-
-INSTALL( TARGETS fast-sweeping-map
- RUNTIME DESTINATION bin
- PERMISSIONS OWNER_READ OWNER_WRITE OWNER_EXECUTE GROUP_READ GROUP_EXECUTE WORLD_READ WORLD_EXECUTE )
-
-#INSTALL( FILES ${tnl_fast_sweeping_map_SOURCES}
-# DESTINATION ${TNL_TARGET_DATA_DIRECTORY}/examples/fast-sweeping-map )
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/MainBuildConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/MainBuildConfig.h
deleted file mode 100644
index ed3d686eb99379af1589d734eac9b5812cccdedf..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/MainBuildConfig.h
+++ /dev/null
@@ -1,64 +0,0 @@
-/***************************************************************************
- MainBuildConfig.h - description
- -------------------
- begin : Jul 7, 2014
- copyright : (C) 2014 by Tomas Oberhuber
- email : tomas.oberhuber@fjfi.cvut.cz
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef MAINBUILDCONFIG_H_
-#define MAINBUILDCONFIG_H_
-
-#include <solvers/tnlBuildConfigTags.h>
-
-class MainBuildConfig
-{
- public:
-
- static void print() {std::cerr << "MainBuildConfig" <<std::endl; }
-};
-
-/****
- * Turn off support for float and long double.
- */
-template<> struct tnlConfigTagReal< MainBuildConfig, float > { enum { enabled = false }; };
-template<> struct tnlConfigTagReal< MainBuildConfig, long double > { enum { enabled = false }; };
-
-/****
- * Turn off support for short int and long int indexing.
- */
-template<> struct tnlConfigTagIndex< MainBuildConfig, short int >{ enum { enabled = false }; };
-template<> struct tnlConfigTagIndex< MainBuildConfig, long int >{ enum { enabled = false }; };
-
-/****
- * Use of tnlGrid is enabled for allowed dimensions and Real, Device and Index types.
- */
-template< int Dimensions, typename Real, typename Device, typename Index >
- struct tnlConfigTagMesh< MainBuildConfig, tnlGrid< Dimensions, Real, Device, Index > >
- { enum { enabled = tnlConfigTagDimensions< MainBuildConfig, Dimensions >::enabled &&
- tnlConfigTagReal< MainBuildConfig, Real >::enabled &&
- tnlConfigTagDevice< MainBuildConfig, Device >::enabled &&
- tnlConfigTagIndex< MainBuildConfig, Index >::enabled }; };
-
-/****
- * Please, chose your preferred time discretisation here.
- */
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlExplicitTimeDiscretisationTag >{ enum { enabled = true }; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlSemiImplicitTimeDiscretisationTag >{ enum { enabled = false}; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlImplicitTimeDiscretisationTag >{ enum { enabled = false }; };
-
-/****
- * Only the Runge-Kutta-Merson solver is enabled by default.
- */
-template<> struct tnlConfigTagExplicitSolver< MainBuildConfig, tnlExplicitEulerSolverTag >{ enum { enabled = false }; };
-
-#endif /* MAINBUILDCONFIG_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/fastSweepingMapConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/fastSweepingMapConfig.h
deleted file mode 100644
index 9251deca876e821dace59682ca1a151555095c69..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/fastSweepingMapConfig.h
+++ /dev/null
@@ -1,39 +0,0 @@
-/***************************************************************************
- fastSweepingConfig.h - description
- -------------------
- begin : Oct 15, 2015
- copyright : (C) 2015 by Tomas Sobotik
- email :
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef FASTSWEEPINGCONFIG_H_
-#define FASTSWEEPINGCONFIG_H_
-
-#include <config/tnlConfigDescription.h>
-
-template< typename ConfigTag >
-class fastSweepingMapConfig
-{
- public:
- static void configSetup( tnlConfigDescription& config )
- {
- config.addDelimiter( "Parallel Eikonal solver settings:" );
- config.addEntry < String > ( "problem-name", "This defines particular problem.", "fast-sweeping" );
- config.addRequiredEntry < String > ( "initial-condition", "Initial condition for solver");
- config.addRequiredEntry < int > ( "dim", "Dimension of problem.");
- config.addEntry < String > ( "mesh", "Name of mesh.", "mesh.tnl" );
- config.addEntry < String > ( "exact-input", "Are the function values near the curve equal to the SDF? (yes/no)", "no" );
- config.addRequiredEntry < String > ( "map", "Gradient map for solver");
- }
-};
-
-#endif /* FASTSWEEPINGCONFIG_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.cpp b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.cpp
deleted file mode 100644
index 8849008ff630db0400a6d7d98e789099e5fbb5d9..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.cpp
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cpp - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.cu b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.cu
deleted file mode 100644
index 8849008ff630db0400a6d7d98e789099e5fbb5d9..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.cu
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cpp - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.h
deleted file mode 100644
index 6f23851c2ea111712b9d65cfdbb613b04c1e1cdd..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/main.h
+++ /dev/null
@@ -1,88 +0,0 @@
-/***************************************************************************
- main.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-
-#include "MainBuildConfig.h"
- //for HOST versions:
-#include "tnlFastSweepingMap.h"
- //for DEVICE versions:
-//#include "tnlFastSweepingMap_CUDA.h"
-#include "fastSweepingMapConfig.h"
-#include <solvers/tnlBuildConfigTags.h>
-
-#include <mesh/tnlGrid.h>
-#include <core/tnlDevice.h>
-#include <time.h>
-#include <ctime>
-
-typedef MainBuildConfig BuildConfig;
-
-int main( int argc, char* argv[] )
-{
- time_t start;
- time_t stop;
- time(&start);
- std::clock_t start2= std::clock();
- Config::ParameterContainer parameters;
- tnlConfigDescription configDescription;
- fastSweepingMapConfig< BuildConfig >::configSetup( configDescription );
-
- if( ! parseCommandLine( argc, argv, configDescription, parameters ) )
- return false;
-
- const int& dim = parameters.getParameter< int >( "dim" );
-
- if(dim == 2)
- {
- tnlFastSweepingMap<tnlGrid<2,double,TNL::Devices::Host, int>, double, int> solver;
- if(!solver.init(parameters))
- {
- cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- TNL_CHECK_CUDA_DEVICE;
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver..." <<std::endl;
- solver.run();
- }
-// else if(dim == 3)
-// {
-// tnlFastSweepingMap<tnlGrid<3,double,TNL::Devices::Host, int>, double, int> solver;
-// if(!solver.init(parameters))
-// {
-// cerr << "Solver failed to initialize." <<std::endl;
-// return EXIT_FAILURE;
-// }
-// TNL_CHECK_CUDA_DEVICE;
-// std::cout << "-------------------------------------------------------------" <<std::endl;
-// std::cout << "Starting solver..." <<std::endl;
-// solver.run();
-// }
- else
- {
- std::cerr << "Unsupported number of dimensions: " << dim << "!" <<std::endl;
- return EXIT_FAILURE;
- }
-
-
- time(&stop);
- std::cout << "Solver stopped..." <<std::endl;
- std::cout <<std::endl;
- std::cout << "Running time was: " << difftime(stop,start) << " .... " << (std::clock() - start2) / (double)(CLOCKS_PER_SEC) <<std::endl;
- return EXIT_SUCCESS;
-}
-
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap.h
deleted file mode 100644
index c568329ba2aa5fdb8fed303d43b25e73f210c014..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap.h
+++ /dev/null
@@ -1,188 +0,0 @@
-/***************************************************************************
- tnlFastSweepingMap.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING_H_
-#define TNLFASTSWEEPING_H_
-
-#include <TNL/Config/ParameterContainer.h>
-#include <TNL/Containers/Vector.h>
-#include <TNL/Containers/StaticVector.h>
-#include <functions/tnlMeshFunction.h>
-#include <TNL/Devices/Host.h>
-#include <mesh/tnlGrid.h>
-#include <mesh/grids/tnlGridEntity.h>
-#include <limits.h>
-#include <core/tnlDevice.h>
-#include <ctime>
-#ifdef HAVE_OPENMP
-#include <omp.h>
-#endif
-
-
-
-
-template< typename Mesh,
- typename Real,
- typename Index >
-class tnlFastSweepingMap
-{};
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 2, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
-
- tnlFastSweepingMap();
-
- static String getType();
- bool init( const Config::ParameterContainer& parameters );
-
- bool initGrid();
- bool run();
-
- //for single core version use this implementation:
- void updateValue(const Index i, const Index j);
- //for parallel version use this one instead:
-// void updateValue(const Index i, const Index j, DofVectorType* grid);
-
-
- void setupSquare1000(Index i, Index j);
- void setupSquare1100(Index i, Index j);
- void setupSquare1010(Index i, Index j);
- void setupSquare1001(Index i, Index j);
- void setupSquare1110(Index i, Index j);
- void setupSquare1101(Index i, Index j);
- void setupSquare1011(Index i, Index j);
- void setupSquare1111(Index i, Index j);
- void setupSquare0000(Index i, Index j);
- void setupSquare0100(Index i, Index j);
- void setupSquare0010(Index i, Index j);
- void setupSquare0001(Index i, Index j);
- void setupSquare0110(Index i, Index j);
- void setupSquare0101(Index i, Index j);
- void setupSquare0011(Index i, Index j);
- void setupSquare0111(Index i, Index j);
-
- Real fabsMin(const Real x, const Real y);
-
-
-protected:
-
- MeshType Mesh;
-
- bool exactInput;
-
- int something_changed;
-
- tnlMeshFunction<MeshType> dofVector, dofVector2;
- DofVectorType data,map;
-
- RealType h;
-
- tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage > Entity;
-
-
-#ifdef HAVE_OPENMP
-// omp_lock_t* gridLock;
-#endif
-
-
-};
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlFastSweepingMap< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 3, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
- tnlFastSweepingMap();
-
- static String getType();
- bool init( const Config::ParameterContainer& parameters );
-
- bool initGrid();
- bool run();
-
- //for single core version use this implementation:
- void updateValue(const Index i, const Index j, const Index k);
- //for parallel version use this one instead:
-// void updateValue(const Index i, const Index j, DofVectorType* grid);
-
- Real fabsMin(const Real x, const Real y);
-
-
-protected:
-
- MeshType Mesh;
-
- bool exactInput;
-
-
- tnlMeshFunction<MeshType> dofVector, dofVector2;
- DofVectorType data;
-
- RealType h;
-
- tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage > Entity;
-
-#ifdef HAVE_OPENMP
-// omp_lock_t* gridLock;
-#endif
-
-
-};
-
-
- //for single core version use this implementation:
-#include "tnlFastSweepingMap2D_impl.h"
- //for parallel version use this one instead:
-// #include "tnlFastSweepingMap2D_openMP_impl.h"
-
-// #include "tnlFastSweepingMap3D_impl.h"
-
-#endif /* TNLFASTSWEEPING_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap2D_CUDA_v4_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap2D_CUDA_v4_impl.h
deleted file mode 100644
index d02b8d6c5d40f8f581e160201c952777ba15aefe..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap2D_CUDA_v4_impl.h
+++ /dev/null
@@ -1,1051 +0,0 @@
-/***************************************************************************
- tnlFastSweepingMap2D_CUDA_v4_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-#include "tnlFastSweepingMap.h"
-
-#define MAP_SOLVER_MAX_VALUE 3
-
-__device__
-double fabsMin( double x, double y)
-{
- double fx = abs(x);
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-}
-
-__device__
-double atomicFabsMin(double* address, double val)
-{
- unsigned long long int* address_as_ull =
- (unsigned long long int*)address;
- unsigned long long int old = *address_as_ull, assumed;
- do {
- assumed = old;
- old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(__longlong_as_double(assumed),val) ));
- } while (assumed != old);
- return __longlong_as_double(old);
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweepingMap< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: tnlFastSweepingMap()
-:dofVector(Mesh)
-{
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- const String& mapFile = parameters.getParameter <String>("map");
- if(! this->map.load( mapFile ))
- cout << "Failed to load map file : " << mapFile <<std::endl;
-
- h = Mesh.template getSpaceStepsProducts< 1, 0 >();
- //Entity.refresh();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(map_cuda), this->map.getSize()*sizeof(double));
- cudaMemcpy(map_cuda, this->map.getData(), this->map.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(changed), sizeof(int));
- //counter == 0 --> setting changed to 0
- cudaMemcpy(changed, &counter, sizeof(int), cudaMemcpyHostToDevice);
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-#endif
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 + 1 ,n/16 +1);
-
-
- initCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(1, 1024);
- dim3 numBlocks(4,1);
-
- int run = 1;
- int zero = 0;
- int cntr = 0;
-
- while(run != 0)
- {
- cudaMemcpy(this->changed, &zero, sizeof(int), cudaMemcpyHostToDevice);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,0,0, this->changed);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- cudaMemcpy(&run, this->changed,sizeof(int), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- cntr++;
- cout << "Finished set of sweeps #" << cntr << " " << run <<std::endl;
- }
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- //data.setLike(dofVector.getData());
- //cudaMemcpy(data.getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaMemcpy(dofVector.getData().getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- //data.save("u-00001.tnl");
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j, Index* something_changed)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
-
- if(map_cuda[Entity.getIndex()] != 0.0)
- {
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real value = cudaDofVector2[Entity.getIndex()];
- Real im = abs(1.0/map_cuda[Entity.getIndex()]);
- Real a,b, tmp;
-
- if( i == 0 )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- else
- {
- a = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] );
- }
-
- if( j == 0 )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- else
- {
- b = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] );
- }
-
-
- if(abs(a-b) >= im*h)
- tmp = fabsMin(a,b) + sign(value)*im*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * im * h * im * h - (a - b) * (a - b) ) );
-
- // cudaDofVector2[Entity.getIndex()] = fabsMin(value, tmp);
- atomicFabsMin(&(cudaDofVector2[Entity.getIndex()]), tmp);
-
- if(abs(value)-abs(tmp) > 0.0)
- atomicMax(something_changed,1);
- }
- else
- {
- atomicFabsMin(&(cudaDofVector2[Entity.getIndex()]), MAP_SOLVER_MAX_VALUE);
- }
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- int j = blockDim.y*blockIdx.y + threadIdx.y;
-
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
-
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- int gid = Entity.getIndex();
-
- cudaDofVector2[gid] = INT_MAX*sign(cudaDofVector[gid]);
-
- if(abs(cudaDofVector[gid]) < 1.01*h)
- {
- cudaDofVector2[gid] = cudaDofVector[gid];
- if(map_cuda[gid] != 0.0)
- cudaDofVector2[gid] /=map_cuda[gid];
- }
-
-
-
-
-
-// if(i+1 < Mesh.getDimensions().x() && j+1 < Mesh.getDimensions().y() )
-// {
-// if(cudaDofVector[Entity.getIndex()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1111(i,j);
-// else
-// setupSquare1110(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1101(i,j);
-// else
-// setupSquare1100(i,j);
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1011(i,j);
-// else
-// setupSquare1010(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1001(i,j);
-// else
-// setupSquare1000(i,j);
-// }
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0111(i,j);
-// else
-// setupSquare0110(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0101(i,j);
-// else
-// setupSquare0100(i,j);
-// }
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0011(i,j);
-// else
-// setupSquare0010(i,j);
-// }
-// else
-// {
-// if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0001(i,j);
-// else
-// setupSquare0000(i,j);
-// }
-// }
-// }
-//
-// }
-
- return true;
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
- //Real fy = abs(y);
-
- //Real tmpMin = Min(fx,abs(y));
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlFastSweepingMap< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i, int* changed)
-{
-
- __shared__ int something_changed;
- if(threadIdx.x+threadIdx.y == 0)
- something_changed = 0;
-
- int gx = 0;
- int gy = threadIdx.y;
- //if(solver->Mesh.getDimensions().x() <= gx || solver->Mesh.getDimensions().y() <= gy)
- // return;
- int n = solver->Mesh.getDimensions().x();
- int blockCount = n/blockDim.y +1;
- //int gid = solver->Mesh.getDimensions().x() * gy + gx;
- //int max = solver->Mesh.getDimensions().x()*solver->Mesh.getDimensions().x();
-
- //int id1 = gx+gy;
- //int id2 = (solver->Mesh.getDimensions().x() - gx - 1) + gy;
-
- __syncthreads();
- if(blockIdx.x==0)
- {
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,&something_changed);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==1)
- {
- gx=n-1;
- gy=threadIdx.y;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,&something_changed);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==2)
- {
- gx=0;
- gy=n-threadIdx.y-1;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,&something_changed);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==3)
- {
- gx=n-1;
- gy=n-threadIdx.y-1;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,&something_changed);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
-
-
- if(threadIdx.x+threadIdx.y == 0)
- atomicMax(changed, something_changed);
-
-
-
-
-}
-
-
-__global__ void initCUDA(tnlFastSweepingMap< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
-
-
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
- solver->initGrid();
- }
-
-
-}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(INT_MAX,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(-INT_MAX,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(cudaDofVector[Entity.getIndex()],cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=fabsMin(cudaDofVector[Entity.getIndex()],cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-}
-#endif
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap2D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap2D_impl.h
deleted file mode 100644
index 4bd9e17c5626c7fcbbe0b747a2011d0d74ac9809..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap2D_impl.h
+++ /dev/null
@@ -1,823 +0,0 @@
-/***************************************************************************
- tnlFastSweepingMap2D_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-
-#define MAP_SOLVER_MAX_VALUE 3
-
-
-#include "tnlFastSweepingMap.h"
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweepingMap< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: tnlFastSweepingMap()
-:Entity(Mesh),
- dofVector(Mesh),
- dofVector2(Mesh)
-{
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
- dofVector2.load(initialCondition);
-
- const String& mapFile = parameters.getParameter <String>("map");
- if(! this->map.load( mapFile ))
- cout << "Failed to load map file : " << mapFile <<std::endl;
-
- h = Mesh.template getSpaceStepsProducts< 1, 0 >();
- Entity.refresh();
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
- cout << "a" <<std::endl;
-
- something_changed = 1;
- return initGrid();
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
-
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- for(int i=0; i< Mesh.getDimensions().x()*Mesh.getDimensions().x();i++)
- {
- dofVector2[i]=INT_MAX*sign(dofVector[i]);
-
- if(abs(dofVector[i]) < 1.01*h)
- {
- dofVector2[i] = dofVector[i];
- if(map[i] != 0.0)
- dofVector2[i] /= map[i];
- }
- }
-
-// for(int i = 0 ; i < Mesh.getDimensions().x()-1; i++)
-// {
-// for(int j = 0 ; j < Mesh.getDimensions().x()-1; j++)
-// {
-// this->Entity.setCoordinates(CoordinatesType(i,j));
-// this->Entity.refresh();
-// neighborEntities.refresh(Mesh,Entity.getIndex());
-//
-// if(dofVector[this->Entity.getIndex()] > 0)
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1111(i,j);
-// else
-// setupSquare1110(i,j);
-// }
-// else
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1101(i,j);
-// else
-// setupSquare1100(i,j);
-// }
-// }
-// else
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1011(i,j);
-// else
-// setupSquare1010(i,j);
-// }
-// else
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare1001(i,j);
-// else
-// setupSquare1000(i,j);
-// }
-// }
-// }
-// else
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0111(i,j);
-// else
-// setupSquare0110(i,j);
-// }
-// else
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0101(i,j);
-// else
-// setupSquare0100(i,j);
-// }
-// }
-// else
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0011(i,j);
-// else
-// setupSquare0010(i,j);
-// }
-// else
-// {
-// if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
-// setupSquare0001(i,j);
-// else
-// setupSquare0000(i,j);
-// }
-// }
-// }
-//
-// }
-// }
- cout << "a" <<std::endl;
-
- //data.setLike(dofVector2.getData());
- //data=dofVector2.getData();
- //cout << data.getType() <<std::endl;
- dofVector2.save("u-00000.tnl");
- //dofVector2.getData().save("u-00000.tnl");
-
- return true;
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
- int cntr = 0;
- while(something_changed != 0)
- {
- something_changed = 0;
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j);
- }
- }
-
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j);
- }
- }
-
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j);
- }
- }
-
- /*---------------------------------------------------------------------------------------------------------------------------*/
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j);
- }
- }
-
- /*---------------------------------------------------------------------------------------------------------------------------*/
- cntr++;
- cout << "Finished set of sweeps #" << cntr << " " << something_changed <<std::endl;
- }
-
-
-
- dofVector2.save("u-00001.tnl");
-
- return true;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
-
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- if(map[Entity.getIndex()] != 0.0)
- {
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- Real value = dofVector2[Entity.getIndex()];
- Real im = abs(1.0/map[Entity.getIndex()]);
- Real a,b, tmp;
-
- if( i == 0 )
- a = dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = dofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- else
- {
- a = fabsMin( dofVector2[neighborEntities.template getEntityIndex< -1, 0 >()],
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] );
- }
-
- if( j == 0 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- else
- {
- b = fabsMin( dofVector2[neighborEntities.template getEntityIndex< 0, -1 >()],
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] );
- }
-
-
- if(fabs(a-b) >= im*h)
- tmp = fabsMin(a,b) + sign(value)*im*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * im * h * im * h - (a - b) * (a - b) ) );
-
- if(abs(value)-abs(tmp) > 0.0)
- something_changed = 1;
-
- dofVector2[Entity.getIndex()] = fabsMin(value, tmp);
-
- }
- else
- {
- dofVector2[Entity.getIndex()] = MAP_SOLVER_MAX_VALUE;
- }
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-Real tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = fabs(x);
- Real fy = fabs(y);
-
- Real tmpMin = Min(fx,fy);
-
- if(tmpMin == fx)
- return x;
- else
- return y;
-
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-{
-// this->Entity.setCoordinates(CoordinatesType(i,j));
-// this->Entity.refresh();
-// auto neighborEntities = Entity.getNeighborEntities();
-// dofVector2[Entity.getIndex()]=fabsMin(INT_MAX,dofVector2[Entity.getIndex()]);
-// dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-{
-// this->Entity.setCoordinates(CoordinatesType(i,j));
-// this->Entity.refresh();
-// auto neighborEntities = Entity.getNeighborEntities();
-// dofVector2[Entity.getIndex()]=fabsMin(-INT_MAX,dofVector2[(Entity.getIndex())]);
-// dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- dofVector2[Entity.getIndex()]=fabsMin(dofVector[Entity.getIndex()],dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- dofVector2[Entity.getIndex()]=fabsMin(dofVector[Entity.getIndex()],dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-}
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap_CUDA.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap_CUDA.h
deleted file mode 100644
index a23057e78c745e74467db4c4190d6f217024bc5a..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping-map/tnlFastSweepingMap_CUDA.h
+++ /dev/null
@@ -1,196 +0,0 @@
-/***************************************************************************
- tnlFastSweepingMap_CUDA.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING_H_
-#define TNLFASTSWEEPING_H_
-
-#include <TNL/Config/ParameterContainer.h>
-#include <TNL/Containers/Vector.h>
-#include <TNL/Containers/StaticVector.h>
-#include <TNL/Devices/Host.h>
-#include <mesh/tnlGrid.h>
-#include <mesh/grids/tnlGridEntity.h>
-
-#include <functions/tnlMeshFunction.h>
-#include <limits.h>
-#include <core/tnlDevice.h>
-#include <ctime>
-
-
-
-
-
-template< typename Mesh,
- typename Real,
- typename Index >
-class tnlFastSweepingMap
-{};
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 2, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
- tnlFastSweepingMap();
-
- __host__ static String getType();
- __host__ bool init( const Config::ParameterContainer& parameters );
- __host__ bool run();
-
-#ifdef HAVE_CUDA
- __device__ bool initGrid();
- __device__ void updateValue(const Index i, const Index j, Index* something_changed);
- __device__ void updateValue(const Index i, const Index j, double** sharedMem, const int k3);
- __device__ Real fabsMin(const Real x, const Real y);
-
- tnlFastSweepingMap< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >* cudaSolver;
- double* cudaDofVector;
- double* cudaDofVector2;
- double* map_cuda;
- int counter;
- int* changed;
- __device__ void setupSquare1000(Index i, Index j);
- __device__ void setupSquare1100(Index i, Index j);
- __device__ void setupSquare1010(Index i, Index j);
- __device__ void setupSquare1001(Index i, Index j);
- __device__ void setupSquare1110(Index i, Index j);
- __device__ void setupSquare1101(Index i, Index j);
- __device__ void setupSquare1011(Index i, Index j);
- __device__ void setupSquare1111(Index i, Index j);
- __device__ void setupSquare0000(Index i, Index j);
- __device__ void setupSquare0100(Index i, Index j);
- __device__ void setupSquare0010(Index i, Index j);
- __device__ void setupSquare0001(Index i, Index j);
- __device__ void setupSquare0110(Index i, Index j);
- __device__ void setupSquare0101(Index i, Index j);
- __device__ void setupSquare0011(Index i, Index j);
- __device__ void setupSquare0111(Index i, Index j);
-#endif
-
- MeshType Mesh;
-
-protected:
-
-
-
- bool exactInput;
-
- tnlMeshFunction<MeshType> dofVector;
- DofVectorType data, map;
-
-
- RealType h;
-
-
-};
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlFastSweepingMap< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 3, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
-
-
- __host__ static String getType();
- __host__ bool init( const Config::ParameterContainer& parameters );
- __host__ bool run();
-
-#ifdef HAVE_CUDA
- __device__ bool initGrid(int i, int j, int k);
- __device__ void updateValue(const Index i, const Index j, const Index k);
- __device__ void updateValue(const Index i, const Index j, const Index k, double** sharedMem, const int k3);
- __device__ Real fabsMin(const Real x, const Real y);
-
- tnlFastSweepingMap< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >* cudaSolver;
- double* cudaDofVector;
- double* cudaDofVector2;
- int counter;
-#endif
-
- MeshType Mesh;
-
-protected:
-
-
-
- bool exactInput;
-
- tnlMeshFunction<MeshType> dofVector;
- DofVectorType data;
-
- RealType h;
-
-
-};
-
-
-
-
-
-
-
-#ifdef HAVE_CUDA
-//template<int sweep_t>
-__global__ void runCUDA(tnlFastSweepingMap< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i, int* changed);
-//__global__ void runCUDA(tnlFastSweepingMap< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i);
-
-__global__ void initCUDA(tnlFastSweepingMap< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver);
-//__global__ void initCUDA(tnlFastSweepingMap< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver);
-#endif
-
-/*various implementtions.... choose one*/
-//#include "tnlFastSweepingMap2D_CUDA_impl.h"
-//#include "tnlFastSweepingMap2D_CUDA_v2_impl.h"
-//#include "tnlFastSweepingMap2D_CUDA_v3_impl.h"
-#include "tnlFastSweepingMap2D_CUDA_v4_impl.h"
-//#include "tnlFastSweepingMap2D_CUDA_v5_impl.h"
-
-
-// #include "tnlFastSweepingMap3D_CUDA_impl.h"
-
-#endif /* TNLFASTSWEEPING_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/CMakeLists.txt b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/CMakeLists.txt
deleted file mode 100644
index 1a23d646a43090c0a63216b43c317c14ab0903d3..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/CMakeLists.txt
+++ /dev/null
@@ -1,22 +0,0 @@
-set( tnl_fast_sweeping_SOURCES
-# MainBuildConfig.h
-# tnlFastSweeping2D_impl.h
-# tnlFastSweeping.h
-# fastSweepingConfig.h
- main.cpp)
-
-
-IF( BUILD_CUDA )
- CUDA_ADD_EXECUTABLE(fast-sweeping main.cu)
-ELSE( BUILD_CUDA )
- ADD_EXECUTABLE(fast-sweeping main.cpp)
-ENDIF( BUILD_CUDA )
-target_link_libraries (fast-sweeping tnl )
-
-
-INSTALL( TARGETS fast-sweeping
- RUNTIME DESTINATION bin
- PERMISSIONS OWNER_READ OWNER_WRITE OWNER_EXECUTE GROUP_READ GROUP_EXECUTE WORLD_READ WORLD_EXECUTE )
-
-#INSTALL( FILES ${tnl_fast_sweeping_SOURCES}
-# DESTINATION ${TNL_TARGET_DATA_DIRECTORY}/examples/fast-sweeping )
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/MainBuildConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/MainBuildConfig.h
deleted file mode 100644
index ed3d686eb99379af1589d734eac9b5812cccdedf..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/MainBuildConfig.h
+++ /dev/null
@@ -1,64 +0,0 @@
-/***************************************************************************
- MainBuildConfig.h - description
- -------------------
- begin : Jul 7, 2014
- copyright : (C) 2014 by Tomas Oberhuber
- email : tomas.oberhuber@fjfi.cvut.cz
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef MAINBUILDCONFIG_H_
-#define MAINBUILDCONFIG_H_
-
-#include <solvers/tnlBuildConfigTags.h>
-
-class MainBuildConfig
-{
- public:
-
- static void print() {std::cerr << "MainBuildConfig" <<std::endl; }
-};
-
-/****
- * Turn off support for float and long double.
- */
-template<> struct tnlConfigTagReal< MainBuildConfig, float > { enum { enabled = false }; };
-template<> struct tnlConfigTagReal< MainBuildConfig, long double > { enum { enabled = false }; };
-
-/****
- * Turn off support for short int and long int indexing.
- */
-template<> struct tnlConfigTagIndex< MainBuildConfig, short int >{ enum { enabled = false }; };
-template<> struct tnlConfigTagIndex< MainBuildConfig, long int >{ enum { enabled = false }; };
-
-/****
- * Use of tnlGrid is enabled for allowed dimensions and Real, Device and Index types.
- */
-template< int Dimensions, typename Real, typename Device, typename Index >
- struct tnlConfigTagMesh< MainBuildConfig, tnlGrid< Dimensions, Real, Device, Index > >
- { enum { enabled = tnlConfigTagDimensions< MainBuildConfig, Dimensions >::enabled &&
- tnlConfigTagReal< MainBuildConfig, Real >::enabled &&
- tnlConfigTagDevice< MainBuildConfig, Device >::enabled &&
- tnlConfigTagIndex< MainBuildConfig, Index >::enabled }; };
-
-/****
- * Please, chose your preferred time discretisation here.
- */
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlExplicitTimeDiscretisationTag >{ enum { enabled = true }; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlSemiImplicitTimeDiscretisationTag >{ enum { enabled = false}; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlImplicitTimeDiscretisationTag >{ enum { enabled = false }; };
-
-/****
- * Only the Runge-Kutta-Merson solver is enabled by default.
- */
-template<> struct tnlConfigTagExplicitSolver< MainBuildConfig, tnlExplicitEulerSolverTag >{ enum { enabled = false }; };
-
-#endif /* MAINBUILDCONFIG_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/fastSweepingConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/fastSweepingConfig.h
deleted file mode 100644
index 3df2c1e889050448fc07baf7dcd0e32feab3f778..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/fastSweepingConfig.h
+++ /dev/null
@@ -1,38 +0,0 @@
-/***************************************************************************
- fastSweepingConfig.h - description
- -------------------
- begin : Oct 15, 2015
- copyright : (C) 2015 by Tomas Sobotik
- email :
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef FASTSWEEPINGCONFIG_H_
-#define FASTSWEEPINGCONFIG_H_
-
-#include <config/tnlConfigDescription.h>
-
-template< typename ConfigTag >
-class fastSweepingConfig
-{
- public:
- static void configSetup( tnlConfigDescription& config )
- {
- config.addDelimiter( "Parallel Eikonal solver settings:" );
- config.addEntry < String > ( "problem-name", "This defines particular problem.", "fast-sweeping" );
- config.addRequiredEntry < String > ( "initial-condition", "Initial condition for solver");
- config.addRequiredEntry < int > ( "dim", "Dimension of problem.");
- config.addEntry < String > ( "mesh", "Name of mesh.", "mesh.tnl" );
- config.addEntry < String > ( "exact-input", "Are the function values near the curve equal to the SDF? (yes/no)", "no" );
- }
-};
-
-#endif /* FASTSWEEPINGCONFIG_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.cpp b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.cpp
deleted file mode 100644
index 8849008ff630db0400a6d7d98e789099e5fbb5d9..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.cpp
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cpp - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.cu b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.cu
deleted file mode 100644
index 8849008ff630db0400a6d7d98e789099e5fbb5d9..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.cu
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cpp - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.h
deleted file mode 100644
index e5ac15fede2281abbd31320985de93671d63d178..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/main.h
+++ /dev/null
@@ -1,88 +0,0 @@
-/***************************************************************************
- main.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-
-#include "MainBuildConfig.h"
- //for HOST versions:
-#include "tnlFastSweeping.h"
- //for DEVICE versions:
-//#include "tnlFastSweeping_CUDA.h"
-#include "fastSweepingConfig.h"
-#include <solvers/tnlBuildConfigTags.h>
-
-#include <mesh/tnlGrid.h>
-#include <core/tnlDevice.h>
-#include <time.h>
-#include <ctime>
-
-typedef MainBuildConfig BuildConfig;
-
-int main( int argc, char* argv[] )
-{
- time_t start;
- time_t stop;
- time(&start);
- std::clock_t start2= std::clock();
- Config::ParameterContainer parameters;
- tnlConfigDescription configDescription;
- fastSweepingConfig< BuildConfig >::configSetup( configDescription );
-
- if( ! parseCommandLine( argc, argv, configDescription, parameters ) )
- return false;
-
- const int& dim = parameters.getParameter< int >( "dim" );
-
- if(dim == 2)
- {
- tnlFastSweeping<tnlGrid<2,double,TNL::Devices::Host, int>, double, int> solver;
- if(!solver.init(parameters))
- {
- cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- TNL_CHECK_CUDA_DEVICE;
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver..." <<std::endl;
- solver.run();
- }
- else if(dim == 3)
- {
- tnlFastSweeping<tnlGrid<3,double,TNL::Devices::Host, int>, double, int> solver;
- if(!solver.init(parameters))
- {
- cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- TNL_CHECK_CUDA_DEVICE;
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver..." <<std::endl;
- solver.run();
- }
- else
- {
- std::cerr << "Unsupported number of dimensions: " << dim << "!" <<std::endl;
- return EXIT_FAILURE;
- }
-
-
- time(&stop);
- std::cout << "Solver stopped..." <<std::endl;
- std::cout <<std::endl;
- std::cout << "Running time was: " << difftime(stop,start) << " .... " << (std::clock() - start2) / (double)(CLOCKS_PER_SEC) <<std::endl;
- return EXIT_SUCCESS;
-}
-
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping.h
deleted file mode 100644
index 96d26db7b5a2077d8e2199292f0e888b0171a5c2..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping.h
+++ /dev/null
@@ -1,186 +0,0 @@
-/***************************************************************************
- tnlFastSweeping.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING_H_
-#define TNLFASTSWEEPING_H_
-
-#include <TNL/Config/ParameterContainer.h>
-#include <TNL/Containers/Vector.h>
-#include <TNL/Containers/StaticVector.h>
-#include <functions/tnlMeshFunction.h>
-#include <TNL/Devices/Host.h>
-#include <mesh/tnlGrid.h>
-#include <mesh/grids/tnlGridEntity.h>
-#include <limits.h>
-#include <core/tnlDevice.h>
-#include <ctime>
-#ifdef HAVE_OPENMP
-#include <omp.h>
-#endif
-
-
-
-
-template< typename Mesh,
- typename Real,
- typename Index >
-class tnlFastSweeping
-{};
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 2, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
-
- tnlFastSweeping();
-
- static String getType();
- bool init( const Config::ParameterContainer& parameters );
-
- bool initGrid();
- bool run();
-
- //for single core version use this implementation:
- void updateValue(const Index i, const Index j);
- //for parallel version use this one instead:
-// void updateValue(const Index i, const Index j, DofVectorType* grid);
-
-
- void setupSquare1000(Index i, Index j);
- void setupSquare1100(Index i, Index j);
- void setupSquare1010(Index i, Index j);
- void setupSquare1001(Index i, Index j);
- void setupSquare1110(Index i, Index j);
- void setupSquare1101(Index i, Index j);
- void setupSquare1011(Index i, Index j);
- void setupSquare1111(Index i, Index j);
- void setupSquare0000(Index i, Index j);
- void setupSquare0100(Index i, Index j);
- void setupSquare0010(Index i, Index j);
- void setupSquare0001(Index i, Index j);
- void setupSquare0110(Index i, Index j);
- void setupSquare0101(Index i, Index j);
- void setupSquare0011(Index i, Index j);
- void setupSquare0111(Index i, Index j);
-
- Real fabsMin(const Real x, const Real y);
-
-
-protected:
-
- MeshType Mesh;
-
- bool exactInput;
-
- tnlMeshFunction<MeshType> dofVector, dofVector2;
- DofVectorType data;
-
- RealType h;
-
- tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage > Entity;
-
-
-#ifdef HAVE_OPENMP
-// omp_lock_t* gridLock;
-#endif
-
-
-};
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 3, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
- tnlFastSweeping();
-
- static String getType();
- bool init( const Config::ParameterContainer& parameters );
-
- bool initGrid();
- bool run();
-
- //for single core version use this implementation:
- void updateValue(const Index i, const Index j, const Index k);
- //for parallel version use this one instead:
-// void updateValue(const Index i, const Index j, DofVectorType* grid);
-
- Real fabsMin(const Real x, const Real y);
-
-
-protected:
-
- MeshType Mesh;
-
- bool exactInput;
-
-
- tnlMeshFunction<MeshType> dofVector, dofVector2;
- DofVectorType data;
-
- RealType h;
-
- tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage > Entity;
-
-#ifdef HAVE_OPENMP
-// omp_lock_t* gridLock;
-#endif
-
-
-};
-
-
- //for single core version use this implementation:
-#include "tnlFastSweeping2D_impl.h"
- //for parallel version use this one instead:
-// #include "tnlFastSweeping2D_openMP_impl.h"
-
-#include "tnlFastSweeping3D_impl.h"
-
-#endif /* TNLFASTSWEEPING_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_impl.h
deleted file mode 100644
index bc1da169c01466a69c00b24e450e5eba09aacd1a..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_impl.h
+++ /dev/null
@@ -1,522 +0,0 @@
-/***************************************************************************
- tnlFastSweeping2D_CUDA_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- h = Mesh.getSpaceSteps().x();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData(), this->dofVector.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData(), this->dofVector.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-#endif
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 + 1 ,n/16 +1);
-
- initCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-//
-// for(Index i = 0; i < Mesh.getDimensions().x(); i++)
-// {
-// for(Index j = 0; j < Mesh.getDimensions().y(); j++)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
-// {
-// for(Index j = 0; j < Mesh.getDimensions().y(); j++)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
-// {
-// for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-// for(Index i = 0; i < Mesh.getDimensions().x(); i++)
-// {
-// for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-//
-// dofVector.save("u-00001.tnl");
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(32, 32);
- dim3 numBlocks(n/32 + 1 ,n/32 +1);
-
- for(int i = 2*n - 1; i > -1; i--)
- {
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,4,i);
- cudaDeviceSynchronize();
- }
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- for(int i = 0; i < 2*n ; i++)
- {
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,1,i);
- cudaDeviceSynchronize();
- }
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- for(int i = 0; i < 2*n ; i++)
- {
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,2,i);
- cudaDeviceSynchronize();
- }
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- for(int i = 2*n - 1; i > -1; i--)
- {
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,3,i);
- cudaDeviceSynchronize();
- }
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- cudaMemcpy(this->dofVector.getData(), cudaDofVector, this->dofVector.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
- Index index = Mesh.getCellIndex(CoordinatesType(i,j));
- Real value = cudaDofVector[index];
- Real a,b, tmp;
-
- if( i == 0 )
- a = cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = cudaDofVector[Mesh.template getCellNextToCell<-1,0>(index)];
- else
- {
- a = fabsMin( cudaDofVector[Mesh.template getCellNextToCell<-1,0>(index)],
- cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)] );
- }
-
- if( j == 0 )
- b = cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = cudaDofVector[Mesh.template getCellNextToCell<0,-1>(index)];
- else
- {
- b = fabsMin( cudaDofVector[Mesh.template getCellNextToCell<0,-1>(index)],
- cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)] );
- }
-
-
- if(abs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
- cudaDofVector[index] = fabsMin(value, tmp);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
- int gid = Mesh.getCellIndex(CoordinatesType(gx,gy));
-
- int total = blockDim.x*gridDim.x;
-
-
-
- Real tmp = 0.0;
- int flag = 0;
- counter = 0;
- tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
-
-
- if(!exactInput)
- {
- cudaDofVector[gid]=cudaDofVector[gid]=0.5*h*sign(cudaDofVector[gid]);
- }
- __threadfence();
-// printf("-----------------------------------------------------------------------------------\n");
-
- __threadfence();
-
- if(gx > 0 && gx < Mesh.getDimensions().x()-1)
- {
- if(gy > 0 && gy < Mesh.getDimensions().y()-1)
- {
-
- Index j = gy;
- Index i = gx;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag=1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
- }
-
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
-// printf("****************************************************************\n");
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
- if(gx > 0 && gx < Mesh.getDimensions().x()-1 && gy == 0)
- {
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
- Index j = 0;
- Index i = gx;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\n");
- if(gx > 0 && gx < Mesh.getDimensions().x()-1 && gy == Mesh.getDimensions().y() - 1)
- {
- Index i = gx;
- Index j = Mesh.getDimensions().y() - 1;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n");
- if(gy > 0 && gy < Mesh.getDimensions().y()-1 && gx == 0)
- {
- Index j = gy;
- Index i = 0;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-// printf("@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n");
- if(gy > 0 && gy < Mesh.getDimensions().y()-1 && gx == Mesh.getDimensions().x() - 1)
- {
- Index j = gy;
- Index i = Mesh.getDimensions().x() - 1;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("##################################################################################################\n");
- if(gx == Mesh.getDimensions().x() - 1 &&
- gy == Mesh.getDimensions().y() - 1)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx-1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy-1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
- if(gx == Mesh.getDimensions().x() - 1 &&
- gy == 0)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx-1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy+1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
-// printf("$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$\n");
- if(gx == 0 &&
- gy == Mesh.getDimensions().y() - 1)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx+1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy-1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
- if(gx == 0 &&
- gy == 0)
- {
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx+1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy+1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
-
- __threadfence();
-
- if(flag==1)
- cudaDofVector[gid] = tmp*3;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
- Real fy = abs(y);
-
- Real tmpMin = Min(fx,fy);
-
- if(tmpMin == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i)
-{
-
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
- if(solver->Mesh.getDimensions().x() <= gx || solver->Mesh.getDimensions().y() <= gy)
- return;
- int total = solver->Mesh.getDimensions().x();
- //int gid = solver->Mesh.getDimensions().x() * gy + gx;
- int max = solver->Mesh.getDimensions().x()*solver->Mesh.getDimensions().x();
-
- int id1 = gx+gy;
- int id2 = (solver->Mesh.getDimensions().x() - gx - 1) + gy;
-
- /*---------------------------------------------------------------------------------------------------------------------------*/
- if(sweep == 1)
-// for(int i = 0; i < 2*total - 1; i++)
- {
- if(id1 == i)
- {
- solver->updateValue(gx,gy);
- return;
- }
-
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
- else if(sweep == 2)
-// for(int i = 0; i < 2*total - 1; i++)
- {
- if(id2 == i)
- {
- solver->updateValue(gx,gy);
- return;
- }
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
- else if(sweep == 3)
-// for(int i = 2*total - 2; i > -1; i--)
- {
- if(id1 == i)
- {
- solver->updateValue(gx,gy);
- return;
- }
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
- else if(sweep == 4)
-// for(int i = 2*total - 2; i > -1; i--)
- {
- if(id2 == i)
- {
- solver->updateValue(gx,gy);
- return;
- }
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
-
-
-
-}
-
-
-__global__ void initCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
- solver->initGrid();
- }
-
-
-}
-#endif
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v2_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v2_impl.h
deleted file mode 100644
index 3ad5b7944f839f794f695240e7abba59e23d16b4..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v2_impl.h
+++ /dev/null
@@ -1,588 +0,0 @@
-/***************************************************************************
- tnlFastSweeping2D_CUDA_v2_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- h = Mesh.getSpaceSteps().x();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData(), this->dofVector.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData(), this->dofVector.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-#endif
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 + 1 ,n/16 +1);
-
- initCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-//
-// for(Index i = 0; i < Mesh.getDimensions().x(); i++)
-// {
-// for(Index j = 0; j < Mesh.getDimensions().y(); j++)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
-// {
-// for(Index j = 0; j < Mesh.getDimensions().y(); j++)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
-// {
-// for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-// for(Index i = 0; i < Mesh.getDimensions().x(); i++)
-// {
-// for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-//
-// dofVector.save("u-00001.tnl");
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(27, 27);
- dim3 numBlocks(1 ,1);
-
-// for(int i = 2*n - 1; i > -1; i--)
- {
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,4,0);
- cudaDeviceSynchronize();
- }
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-//// for(int i = 0; i < 2*n ; i++)
-// {
-// runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,1,0);
-// cudaDeviceSynchronize();
-// }
-// cudaDeviceSynchronize();
-// TNL_CHECK_CUDA_DEVICE;
-//// for(int i = 0; i < 2*n ; i++)
-// {
-// runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,2,0);
-// cudaDeviceSynchronize();
-// }
-// cudaDeviceSynchronize();
-// TNL_CHECK_CUDA_DEVICE;
-//// for(int i = 2*n - 1; i > -1; i--)
-// {
-// runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,3,0);
-// cudaDeviceSynchronize();
-// }
-//
-// cudaDeviceSynchronize();
-// TNL_CHECK_CUDA_DEVICE;
-
- cudaMemcpy(this->dofVector.getData(), cudaDofVector, this->dofVector.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
- Index index = Mesh.getCellIndex(CoordinatesType(i,j));
- Real value = cudaDofVector[index];
- Real a,b, tmp;
-
- if( i == 0 )
- a = cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = cudaDofVector[Mesh.template getCellNextToCell<-1,0>(index)];
- else
- {
- a = fabsMin( cudaDofVector[Mesh.template getCellNextToCell<-1,0>(index)],
- cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)] );
- }
-
- if( j == 0 )
- b = cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = cudaDofVector[Mesh.template getCellNextToCell<0,-1>(index)];
- else
- {
- b = fabsMin( cudaDofVector[Mesh.template getCellNextToCell<0,-1>(index)],
- cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)] );
- }
-
-
- if(abs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
- cudaDofVector[index] = fabsMin(value, tmp);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
- int gid = Mesh.getCellIndex(CoordinatesType(gx,gy));
-
- int total = blockDim.x*gridDim.x;
-
-
-
- Real tmp = 0.0;
- int flag = 0;
- counter = 0;
- tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
-
-
- if(!exactInput)
- {
- cudaDofVector[gid]=cudaDofVector[gid]=0.5*h*sign(cudaDofVector[gid]);
- }
- __threadfence();
-// printf("-----------------------------------------------------------------------------------\n");
-
- __threadfence();
-
- if(gx > 0 && gx < Mesh.getDimensions().x()-1)
- {
- if(gy > 0 && gy < Mesh.getDimensions().y()-1)
- {
-
- Index j = gy;
- Index i = gx;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag=1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
- }
-
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
-// printf("****************************************************************\n");
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
- if(gx > 0 && gx < Mesh.getDimensions().x()-1 && gy == 0)
- {
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
- Index j = 0;
- Index i = gx;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\n");
- if(gx > 0 && gx < Mesh.getDimensions().x()-1 && gy == Mesh.getDimensions().y() - 1)
- {
- Index i = gx;
- Index j = Mesh.getDimensions().y() - 1;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n");
- if(gy > 0 && gy < Mesh.getDimensions().y()-1 && gx == 0)
- {
- Index j = gy;
- Index i = 0;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-// printf("@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n");
- if(gy > 0 && gy < Mesh.getDimensions().y()-1 && gx == Mesh.getDimensions().x() - 1)
- {
- Index j = gy;
- Index i = Mesh.getDimensions().x() - 1;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("##################################################################################################\n");
- if(gx == Mesh.getDimensions().x() - 1 &&
- gy == Mesh.getDimensions().y() - 1)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx-1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy-1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
- if(gx == Mesh.getDimensions().x() - 1 &&
- gy == 0)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx-1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy+1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
-// printf("$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$\n");
- if(gx == 0 &&
- gy == Mesh.getDimensions().y() - 1)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx+1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy-1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
- if(gx == 0 &&
- gy == 0)
- {
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx+1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy+1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
-
- __threadfence();
-
- if(flag==1)
- cudaDofVector[gid] = tmp*3;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
-
- Real tmpMin = Min(fx,abs(y));
-
- if(tmpMin == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
-{
-
- //int gx = threadIdx.x;
- //int gy = threadIdx.y;
- int id1,id2;
- int nx = solver->Mesh.getDimensions().x()+ threadIdx.x;
- int ny = solver->Mesh.getDimensions().y()+ threadIdx.y;
-
- int blockCount = solver->Mesh.getDimensions().x()/blockDim.x + 1;
-
- for(int gy = threadIdx.y; gy < ny;gy+=blockDim.y)
- {
- for(int gx = threadIdx.x; gx < nx;gx+=blockDim.x)
- {
-// if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy && gy > -1&& gx > -1)
- {
- id1 = threadIdx.x+threadIdx.y;
-
- for(int l = 0; l < 2*blockDim.x - 1; l++)
- {
- if(id1 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- }
- __syncthreads();
- }
-
- }
- //gx+=blockDim.x;
- //__syncthreads();
- }
- //gx = threadIdx.x;
- //gy+=blockDim.y;
- //__syncthreads();
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-// gx = blockDim.x*(blockCount-1) + threadIdx.x;
-// gy = threadIdx.y;
- for(int gy = threadIdx.y; gy < ny;gy+=blockDim.y)
- {
- for(int gx = blockDim.x*(blockCount-1) + threadIdx.x; gx >- 1;gx-=blockDim.x)
- {
-// if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy && gy > -1&& gx > -1)
- {
- id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-
- for(int l = 0; l < 2*blockDim.x - 1; l++)
- {
- if(id2 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- }
- __syncthreads();
- }
- }
- //gx-=blockDim.x;
- //__syncthreads();
- }
- //gx = blockDim.x*(blockCount-1) + threadIdx.x;
- //gy+=blockDim.y;
- //__syncthreads();
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-// gx = blockDim.x*(blockCount-1) + threadIdx.x;
-// gy = blockDim.x*(blockCount-1) + threadIdx.y;
- for(int gy = blockDim.x*(blockCount-1) +threadIdx.y; gy >- 1;gy-=blockDim.y)
- {
- for(int gx = blockDim.x*(blockCount-1) + threadIdx.x; gx >- 1;gx-=blockDim.x)
- {
-// if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy && gy > -1&& gx > -1)
- {
- id1 = threadIdx.x+threadIdx.y;
-
- for(int l = 2*blockDim.x - 2; l > -1; l--)
- {
- if(id1 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- }
- __syncthreads();
- }
- }
- //gx-=blockDim.x;
- //__syncthreads();
- }
- //gx = blockDim.x*(blockCount-1) + threadIdx.x;
- //gy-=blockDim.y;
- //__syncthreads();
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
- //gx = threadIdx.x;
- //gy = blockDim.x*(blockCount-1) +threadIdx.y;
- for(int gy = blockDim.x*(blockCount-1) +threadIdx.y; gy >- 1;gy-=blockDim.y)
- {
- for(int gx = threadIdx.x; gx < nx;gx+=blockDim.x)
- {
-// if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy && gy > -1&& gx > -1)
- {
- id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-
- for(int l = 2*blockDim.x - 2; l > -1; l--)
- {
- if(id2 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- }
- __syncthreads();
- }
- }
- //gx+=blockDim.x;
- //__syncthreads();
- }
- //gx = threadIdx.x;
- //gy-=blockDim.y;
- ///__syncthreads();
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
-
-
-
-
-}
-
-
-__global__ void initCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
- solver->initGrid();
- }
-
-
-}
-#endif
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v3_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v3_impl.h
deleted file mode 100644
index ff36d3f8e0a73e5a0987b06f81be2650b277fa25..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v3_impl.h
+++ /dev/null
@@ -1,920 +0,0 @@
-/***************************************************************************
- tnlFastSweeping2D_CUDA_v3_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-
-
-
-__device__ double atomicSet(double* address, double val)
-{
- unsigned long long int* address_as_ull =
- (unsigned long long int*)address;
- unsigned long long int old = *address_as_ull, assumed;
- do {
- assumed = old;
- old = atomicCAS(address_as_ull, assumed,__double_as_longlong(val ));
- } while (assumed != old);
- return __longlong_as_double(old);
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- h = Mesh.getSpaceSteps().x();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData(), this->dofVector.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData(), this->dofVector.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-#endif
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 + 1 ,n/16 +1);
-
- initCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-//
-// for(Index i = 0; i < Mesh.getDimensions().x(); i++)
-// {
-// for(Index j = 0; j < Mesh.getDimensions().y(); j++)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
-// {
-// for(Index j = 0; j < Mesh.getDimensions().y(); j++)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
-// {
-// for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-// for(Index i = 0; i < Mesh.getDimensions().x(); i++)
-// {
-// for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-//
-// dofVector.save("u-00001.tnl");
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 +1 ,n/16 +1);
- int m =n/16 +1;
-
- for(int i = 0; i < 2*m -1; i++)
- {
- runCUDA<15><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,1,i);
- //cudaDeviceSynchronize();
- }
-// cudaDeviceSynchronize();
-// TNL_CHECK_CUDA_DEVICE;
-// for(int i = 0; i < 2*m -1; i++)
-// {
-// runCUDA<2><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,2,i);
-// cudaDeviceSynchronize();
-// }
-// cudaDeviceSynchronize();
-// TNL_CHECK_CUDA_DEVICE;
-// for(int i = 0; i < 2*m -1; i++)
-// {
-// runCUDA<4><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,4,i);
-// cudaDeviceSynchronize();
-// }
-// cudaDeviceSynchronize();
-// TNL_CHECK_CUDA_DEVICE;
-// for(int i = 0; i < 2*m -1; i++)
-// {
-// runCUDA<8><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,8,i);
-// cudaDeviceSynchronize();
-// }
-
-
-
-
-// for(int i = 0; i < (2*m -1)/4 -1; i++)
-// {
-// runCUDA<15><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,15,i);//all
-// cudaDeviceSynchronize();
-// }
-// for(int i = (2*m -1)/4 -1; i < (2*m -1)/2 -1; i++)
-// {
-// runCUDA<5><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,5,i); //two
-// cudaDeviceSynchronize();
-// runCUDA<10><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,10,i); //two
-// cudaDeviceSynchronize();
-// }
-// for(int i = (2*m -1)/2 -1; i < (2*m -1)/2 +1; i++)
-// {
-// runCUDA<1><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,1,i); //separate
-// cudaDeviceSynchronize();
-// runCUDA<2><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,2,i); //separate
-// cudaDeviceSynchronize();
-// runCUDA<4><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,4,i); //separate
-// cudaDeviceSynchronize();
-// runCUDA<8><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,8,i); //separate
-// cudaDeviceSynchronize();
-// }
-// for(int i = (2*m -1)/2 +1; i < (2*m -1/4)*3 +1; i++)
-// {
-// runCUDA<5><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,5,i); //two
-// cudaDeviceSynchronize();
-// runCUDA<10><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,10,i); //two
-// cudaDeviceSynchronize();
-// }
-// for(int i = (2*m -1/4)*3 +1; i < 2*m -1; i++)
-// {
-// runCUDA<15><<<numBlocks,threadsPerBlock>>>(this->cudaSolver,15,i);//all
-// cudaDeviceSynchronize();
-// }
-cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- cudaMemcpy(this->dofVector.getData(), cudaDofVector, this->dofVector.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
- Index index = Mesh.getCellIndex(CoordinatesType(i,j));
- Real value = cudaDofVector[index];
- Real a,b, tmp;
-
- if( i == 0 )
- a = cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = cudaDofVector[Mesh.template getCellNextToCell<-1,0>(index)];
- else
- {
- a = fabsMin( cudaDofVector[Mesh.template getCellNextToCell<-1,0>(index)],
- cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)] );
- }
-
- if( j == 0 )
- b = cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = cudaDofVector[Mesh.template getCellNextToCell<0,-1>(index)];
- else
- {
- b = fabsMin( cudaDofVector[Mesh.template getCellNextToCell<0,-1>(index)],
- cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)] );
- }
-
-
- if(abs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
- atomicSet(&cudaDofVector[index],fabsMin(value, tmp));
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
- int gid = Mesh.getCellIndex(CoordinatesType(gx,gy));
-
- int total = blockDim.x*gridDim.x;
-
-
-
- Real tmp = 0.0;
- int flag = 0;
- counter = 0;
- tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
-
-
- if(!exactInput)
- {
- cudaDofVector[gid]=cudaDofVector[gid]=0.5*h*sign(cudaDofVector[gid]);
- }
- __threadfence();
-// printf("-----------------------------------------------------------------------------------\n");
-
- __threadfence();
-
- if(gx > 0 && gx < Mesh.getDimensions().x()-1)
- {
- if(gy > 0 && gy < Mesh.getDimensions().y()-1)
- {
-
- Index j = gy;
- Index i = gx;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag=1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
- }
-
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
-// printf("****************************************************************\n");
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
- if(gx > 0 && gx < Mesh.getDimensions().x()-1 && gy == 0)
- {
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
- Index j = 0;
- Index i = gx;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\n");
- if(gx > 0 && gx < Mesh.getDimensions().x()-1 && gy == Mesh.getDimensions().y() - 1)
- {
- Index i = gx;
- Index j = Mesh.getDimensions().y() - 1;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n");
- if(gy > 0 && gy < Mesh.getDimensions().y()-1 && gx == 0)
- {
- Index j = gy;
- Index i = 0;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-// printf("@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n");
- if(gy > 0 && gy < Mesh.getDimensions().y()-1 && gx == Mesh.getDimensions().x() - 1)
- {
- Index j = gy;
- Index i = Mesh.getDimensions().x() - 1;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("##################################################################################################\n");
- if(gx == Mesh.getDimensions().x() - 1 &&
- gy == Mesh.getDimensions().y() - 1)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx-1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy-1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
- if(gx == Mesh.getDimensions().x() - 1 &&
- gy == 0)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx-1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy+1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
-// printf("$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$\n");
- if(gx == 0 &&
- gy == Mesh.getDimensions().y() - 1)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx+1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy-1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
- if(gx == 0 &&
- gy == 0)
- {
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx+1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy+1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
-
- __threadfence();
-
- if(flag==1)
- cudaDofVector[gid] = tmp*3;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
-// Real fx = abs(x);
-//
-// Real tmpMin = Min(fx,abs(y));
-
- if(abs(y) > abs(x))
- return x;
- else
- return y;
-
-
-}
-
-
-template<>
-__global__ void runCUDA<1>(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
-{
-
- if(blockIdx.x+blockIdx.y == k)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id1 = threadIdx.x+threadIdx.y;
-
- for(int l = 0; l < 2*blockDim.x - 1; l++)
- {
- if(id1 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- }
- __syncthreads();
- }
-
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-}
- template<>
- __global__ void runCUDA<2>(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
- {
- if((gridDim.x - blockIdx.x - 1)+blockIdx.y == k)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-
- for(int l = 0; l < 2*blockDim.x - 1; l++)
- {
- if(id2 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- }
- __syncthreads();
- }
-
- }
- } /*---------------------------------------------------------------------------------------------------------------------------*/
- template<>
- __global__ void runCUDA<4>(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
- {
- if(blockIdx.x+blockIdx.y == gridDim.x+gridDim.y-k-2)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id1 = threadIdx.x+threadIdx.y;
-
- for(int l = 2*blockDim.x - 2; l > -1; l--)
- {
- if(id1 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
- }
-
- template<>
- __global__ void runCUDA<8>(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
- {
- if((gridDim.x - blockIdx.x - 1)+blockIdx.y == gridDim.x+gridDim.y-k-2)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-
- for(int l = 2*blockDim.x - 2; l > -1; l--)
- {
- if(id2 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
-
-
-
-
-}
-
-
- template<>
- __global__ void runCUDA<5>(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
- {
-
- if(blockIdx.x+blockIdx.y == k)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id1 = threadIdx.x+threadIdx.y;
-
- for(int l = 0; l < 2*blockDim.x - 1; l++)
- {
- if(id1 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
- else if(blockIdx.x+blockIdx.y == gridDim.x+gridDim.y-k-2)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id1 = threadIdx.x+threadIdx.y;
-
- for(int l = 2*blockDim.x - 2; l > -1; l--)
- {
- if(id1 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
- }
-
-
- template<>
- __global__ void runCUDA<10>(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
- {
- if((gridDim.x - blockIdx.x - 1)+blockIdx.y == k)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-
- for(int l = 0; l < 2*blockDim.x - 1; l++)
- {
- if(id2 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
-
- else if((gridDim.x - blockIdx.x - 1)+blockIdx.y == gridDim.x+gridDim.y-k-2)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-
- for(int l = 2*blockDim.x - 2; l > -1; l--)
- {
- if(id2 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
-
- }
-
-
-
- template<>
- __global__ void runCUDA<15>(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
- {
-
- if(blockIdx.x+blockIdx.y == k)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id1 = threadIdx.x+threadIdx.y;
-
- for(int l = 0; l < 2*blockDim.x - 1; l++)
- {
- if(id1 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
- if((gridDim.x - blockIdx.x - 1)+blockIdx.y == k)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-
- for(int l = 0; l < 2*blockDim.x - 1; l++)
- {
- if(id2 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
- if(blockIdx.x+blockIdx.y == gridDim.x+gridDim.y-k-2)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id1 = threadIdx.x+threadIdx.y;
-
- for(int l = 2*blockDim.x - 2; l > -1; l--)
- {
- if(id1 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
- if((gridDim.x - blockIdx.x - 1)+blockIdx.y == gridDim.x+gridDim.y-k-2)
- {
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = threadIdx.y + blockDim.y*blockIdx.y;
-
- int id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-
- for(int l = 2*blockDim.x - 2; l > -1; l--)
- {
- if(id2 == l)
- {
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
- solver->updateValue(gx,gy);
- return;
- }
- __syncthreads();
- }
-
- }
- /*---------------------------------------------------------------------------------------------------------------------------*/
-
-
-
-
-
- }
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-__global__ void initCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
- solver->initGrid();
- }
-
-
-}
-#endif
-
-
-
-
-
-
-//__global__ void runCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int k)
-//{
-//
-// if(sweep==1 && blockIdx.x+blockIdx.y == k)
-// {
-// int gx = threadIdx.x + blockDim.x*blockIdx.x;
-// int gy = threadIdx.y + blockDim.y*blockIdx.y;
-//
-// int id1 = threadIdx.x+threadIdx.y;
-//
-// for(int l = 0; l < 2*blockDim.x - 1; l++)
-// {
-// if(id1 == l)
-// {
-// if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
-// solver->updateValue(gx,gy);
-// }
-// __syncthreads();
-// }
-//
-// }
-// /*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// else if(sweep==2 && (gridDim.x - blockIdx.x - 1)+blockIdx.y == k)
-// {
-// int gx = threadIdx.x + blockDim.x*blockIdx.x;
-// int gy = threadIdx.y + blockDim.y*blockIdx.y;
-//
-// int id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-//
-// for(int l = 0; l < 2*blockDim.x - 1; l++)
-// {
-// if(id2 == l)
-// {
-// if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
-// solver->updateValue(gx,gy);
-// }
-// __syncthreads();
-// }
-//
-// }
-// /*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// else if(sweep==4 && blockIdx.x+blockIdx.y == gridDim.x+gridDim.y-k-2)
-// {
-// int gx = threadIdx.x + blockDim.x*blockIdx.x;
-// int gy = threadIdx.y + blockDim.y*blockIdx.y;
-//
-// int id1 = threadIdx.x+threadIdx.y;
-//
-// for(int l = 2*blockDim.x - 2; l > -1; l--)
-// {
-// if(id1 == l)
-// {
-// if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
-// solver->updateValue(gx,gy);
-// return;
-// }
-// __syncthreads();
-// }
-//
-// }
-// /*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// else if(sweep==8 && (gridDim.x - blockIdx.x - 1)+blockIdx.y == gridDim.x+gridDim.y-k-2)
-// {
-// int gx = threadIdx.x + blockDim.x*blockIdx.x;
-// int gy = threadIdx.y + blockDim.y*blockIdx.y;
-//
-// int id2 = (blockDim.x - threadIdx.x - 1) + threadIdx.y;
-//
-// for(int l = 2*blockDim.x - 2; l > -1; l--)
-// {
-// if(id2 == l)
-// {
-// if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy /*&& gy > -1&& gx > -1*/)
-// solver->updateValue(gx,gy);
-// return;
-// }
-// __syncthreads();
-// }
-//
-// }
-// /*---------------------------------------------------------------------------------------------------------------------------*/
-//
-//
-//
-//
-//
-//}
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v4_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v4_impl.h
deleted file mode 100644
index e0a9697c2e5ea4097232b4d4a8f3c6da6fa41e50..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v4_impl.h
+++ /dev/null
@@ -1,1003 +0,0 @@
-/***************************************************************************
- tnlFastSweeping2D_CUDA_v4_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-__device__
-double fabsMin( double x, double y)
-{
- double fx = abs(x);
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-}
-
-__device__
-double atomicFabsMin(double* address, double val)
-{
- unsigned long long int* address_as_ull =
- (unsigned long long int*)address;
- unsigned long long int old = *address_as_ull, assumed;
- do {
- assumed = old;
- old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(__longlong_as_double(assumed),val) ));
- } while (assumed != old);
- return __longlong_as_double(old);
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: tnlFastSweeping()
-:dofVector(Mesh)
-{
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- h = Mesh.template getSpaceStepsProducts< 1, 0 >();
- //Entity.refresh();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-#endif
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 + 1 ,n/16 +1);
-
-
- initCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(1, 1024);
- dim3 numBlocks(4,1);
-
-
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,0,0);
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- //data.setLike(dofVector.getData());
- //cudaMemcpy(data.getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaMemcpy(dofVector.getData().getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- //data.save("u-00001.tnl");
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- Real value = cudaDofVector2[Entity.getIndex()];
- Real a,b, tmp;
-
- if( i == 0 )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- else
- {
- a = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] );
- }
-
- if( j == 0 )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- else
- {
- b = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] );
- }
-
-
- if(abs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
-// cudaDofVector2[Entity.getIndex()] = fabsMin(value, tmp);
- atomicFabsMin(&(cudaDofVector2[Entity.getIndex()]), tmp);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- int j = blockDim.y*blockIdx.y + threadIdx.y;
-
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
-
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- int gid = Entity.getIndex();
-
- cudaDofVector2[gid] = INT_MAX*sign(cudaDofVector[gid]);
-//
-// if(abs(cudaDofVector[gid]) < 1.01*h)
-// cudaDofVector2[gid] = cudaDofVector[gid];
-
-
-
-
-
- if(i+1 < Mesh.getDimensions().x() && j+1 < Mesh.getDimensions().y() && !exactInput)
- {
- if(cudaDofVector[Entity.getIndex()] > 0)
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1111(i,j);
- else
- setupSquare1110(i,j);
- }
- else
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1101(i,j);
- else
- setupSquare1100(i,j);
- }
- }
- else
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1011(i,j);
- else
- setupSquare1010(i,j);
- }
- else
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1001(i,j);
- else
- setupSquare1000(i,j);
- }
- }
- }
- else
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0111(i,j);
- else
- setupSquare0110(i,j);
- }
- else
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0101(i,j);
- else
- setupSquare0100(i,j);
- }
- }
- else
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0011(i,j);
- else
- setupSquare0010(i,j);
- }
- else
- {
- if(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0001(i,j);
- else
- setupSquare0000(i,j);
- }
- }
- }
-
- }
- if(exactInput)
- {
- if(abs(cudaDofVector[gid]) < 1.5*h)
- cudaDofVector2[gid] = cudaDofVector[gid];
- }
-
-
- return true;
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
- //Real fy = abs(y);
-
- //Real tmpMin = Min(fx,abs(y));
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i)
-{
-
- int gx = 0;
- int gy = threadIdx.y;
- //if(solver->Mesh.getDimensions().x() <= gx || solver->Mesh.getDimensions().y() <= gy)
- // return;
- int n = solver->Mesh.getDimensions().x();
- int blockCount = n/blockDim.y +1;
- //int gid = solver->Mesh.getDimensions().x() * gy + gx;
- //int max = solver->Mesh.getDimensions().x()*solver->Mesh.getDimensions().x();
-
- //int id1 = gx+gy;
- //int id2 = (solver->Mesh.getDimensions().x() - gx - 1) + gy;
-
- if(blockIdx.x==0)
- {
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==1)
- {
- gx=n-1;
- gy=threadIdx.y;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==2)
- {
- gx=0;
- gy=n-threadIdx.y-1;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- else if(blockIdx.x==3)
- {
- gx=n-1;
- gy=n-threadIdx.y-1;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
-
-
-
-
-
-}
-
-
-__global__ void initCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
-
-
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
- solver->initGrid();
- }
-
-
-}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-{
-// tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
-// Entity.setCoordinates(CoordinatesType(i,j));
-// Entity.refresh();
-// tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-// cudaDofVector2[Entity.getIndex()]=fabsMin(INT_MAX,cudaDofVector2[Entity.getIndex()]);
-// cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
-// cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
-// cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-{
-// tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
-// Entity.setCoordinates(CoordinatesType(i,j));
-// Entity.refresh();
-// tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-// cudaDofVector2[Entity.getIndex()]=fabsMin(-INT_MAX,cudaDofVector2[Entity.getIndex()]);
-// cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
-// cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
-// cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-INT_MAX,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=INT_MAX; //fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=0.5*h; //fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=-0.5*h; //fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=0.5*h; //fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=0.5*h; //fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=0.5*h; //fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=0.5*h; //fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=INT_MAX; //fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=0.5*h; //fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=0.5*h; //fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=INT_MAX; //fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=-0.5*h; //fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=-0.5*h; //fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=0.5*h; //fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=INT_MAX; //fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=0.5*h; //fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=-INT_MAX; //fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=-0.5*h; //fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=0.5*h; //fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=-0.5*h; //fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=-0.5*h; //fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=0.5*h; //fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=-0.5*h; //fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=-INT_MAX; //fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[Entity.getIndex()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=-0.5*h; //fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=INT_MAX; //fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=-0.5*h; //fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=0.5*h; //fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=0.5*h; //fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=-0.5*h; //fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=-INT_MAX; //fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=-0.5*h; //fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=0.5*h; //fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=-0.5*h; //fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=-0.5*h; //fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=0.5*h; //fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=0.5*h; //fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=0.5*h; //fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=-0.5*h; //fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=-0.5*h; //fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=0.5*h; //fabsMin(cudaDofVector[Entity.getIndex()],cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=-0.5*h; //fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=0.5*h; //fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=-0.5*h; //fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=-0.5*h; //fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=0.5*h; //fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=0.5*h; //fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=-0.5*h; //fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Real al,be, a,b,c,s;
- al=abs(cudaDofVector[Entity.getIndex()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- cudaDofVector[Entity.getIndex()]));
-
- be=abs(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b = 1.0;
- c = -be;
- s = h/sqrt(a*a+b*b);
-
-
- cudaDofVector2[Entity.getIndex()]=-0.5*h; //fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=-0.5*h; //fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=0.5*h; //fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=0.5*h; //fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-{
- tnlGridEntity< tnlGrid< 2,double, TNL::Devices::Host, int >, 2, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- cudaDofVector2[Entity.getIndex()]=-0.5*h; //fabsMin(cudaDofVector[Entity.getIndex()],cudaDofVector2[Entity.getIndex()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=0.5*h; //fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 0, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=-0.5*h; //fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 1 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=0.5*h; //fabsMin(cudaDofVector[neighborEntities.template getEntityIndex< 1, 0 >()],cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-}
-#endif
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v5_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v5_impl.h
deleted file mode 100644
index 1591bb6137007b88e636a83bb113030ecd529f9c..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_CUDA_v5_impl.h
+++ /dev/null
@@ -1,697 +0,0 @@
-/***************************************************************************
- tnlFastSweeping2D_CUDA_v5_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-__device__
-double fabsMin( double x, double y)
-{
- double fx = abs(x);
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-}
-
-__device__
-double atomicFabsMin(double* address, double val)
-{
- unsigned long long int* address_as_ull =
- (unsigned long long int*)address;
- unsigned long long int old = *address_as_ull, assumed;
- do {
- assumed = old;
- old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(assumed,val) ));
- } while (assumed != old);
- return __longlong_as_double(old);
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- h = Mesh.getSpaceSteps().x();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData(), this->dofVector.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData(), this->dofVector.getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-#endif
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(16, 16);
- dim3 numBlocks(n/16 + 1 ,n/16 +1);
-
- initCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-//
-// for(Index i = 0; i < Mesh.getDimensions().x(); i++)
-// {
-// for(Index j = 0; j < Mesh.getDimensions().y(); j++)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
-// {
-// for(Index j = 0; j < Mesh.getDimensions().y(); j++)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-// for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
-// {
-// for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-// for(Index i = 0; i < Mesh.getDimensions().x(); i++)
-// {
-// for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
-// {
-// updateValue(i,j);
-// }
-// }
-//
-///*---------------------------------------------------------------------------------------------------------------------------*/
-//
-//
-// dofVector.save("u-00001.tnl");
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(1, 512);
- dim3 numBlocks(4,1);
-
-
- runCUDA<<<numBlocks,threadsPerBlock,3*(512+1)*sizeof(double)>>>(this->cudaSolver,0,0);
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- cudaMemcpy(this->dofVector.getData(), cudaDofVector, this->dofVector.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
- Index index = Mesh.getCellIndex(CoordinatesType(i,j));
- Real value = cudaDofVector[index];
- Real a,b, tmp;
-
- if( i == 0 )
- a = cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = cudaDofVector[Mesh.template getCellNextToCell<-1,0>(index)];
- else
- {
- a = fabsMin( cudaDofVector[Mesh.template getCellNextToCell<-1,0>(index)],
- cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)] );
- }
-
- if( j == 0 )
- b = cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = cudaDofVector[Mesh.template getCellNextToCell<0,-1>(index)];
- else
- {
- b = fabsMin( cudaDofVector[Mesh.template getCellNextToCell<0,-1>(index)],
- cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)] );
- }
-
-
- if(abs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
- cudaDofVector[index] = fabsMin(value, tmp);
-
-}
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j, double** sharedMem, int k3)
-{
- Index index = Mesh.getCellIndex(CoordinatesType(i,j));
- Real value = sharedMem[k3+1][threadIdx.y];
- Real a,b, tmp;
-
- if( i == 0 )
- a = sharedMem[k3][threadIdx.y];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = sharedMem[k3+2][threadIdx.y];
- else
- {
- a = fabsMin( sharedMem[k3][threadIdx.y],
- sharedMem[k3+2][threadIdx.y] );
- }
-
- if( j == 0 )
- b = sharedMem[k3][threadIdx.y+1];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = sharedMem[k3+2][threadIdx.y-1];
- else
- {
- b = fabsMin( sharedMem[k3][threadIdx.y+1],
- sharedMem[k3+2][threadIdx.y-1] );
- }
-
-
- if(abs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
-// sharedMem[k3+1][threadIdx.y] = this->fabsMin(value, tmp);
-// atomicFabsMin(&(cudaDofVector[index]), tmp);
- cudaDofVector[index] = tmp;
- this->fabsMin(value, tmp);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
- int gid = Mesh.getCellIndex(CoordinatesType(gx,gy));
-
- int total = blockDim.x*gridDim.x;
-
-
-
- Real tmp = 0.0;
- int flag = 0;
- counter = 0;
- tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
-
-
- if(!exactInput)
- {
- cudaDofVector[gid]=cudaDofVector[gid]=0.5*h*sign(cudaDofVector[gid]);
- }
- __threadfence();
-// printf("-----------------------------------------------------------------------------------\n");
-
- __threadfence();
-
- if(gx > 0 && gx < Mesh.getDimensions().x()-1)
- {
- if(gy > 0 && gy < Mesh.getDimensions().y()-1)
- {
-
- Index j = gy;
- Index i = gx;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag=1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
- }
-
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
-// printf("****************************************************************\n");
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
- if(gx > 0 && gx < Mesh.getDimensions().x()-1 && gy == 0)
- {
-// printf("gx: %d, gy: %d, gid: %d \n", gx, gy,gid);
- Index j = 0;
- Index i = gx;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\n");
- if(gx > 0 && gx < Mesh.getDimensions().x()-1 && gy == Mesh.getDimensions().y() - 1)
- {
- Index i = gx;
- Index j = Mesh.getDimensions().y() - 1;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n");
- if(gy > 0 && gy < Mesh.getDimensions().y()-1 && gx == 0)
- {
- Index j = gy;
- Index i = 0;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-// printf("@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n");
- if(gy > 0 && gy < Mesh.getDimensions().y()-1 && gx == Mesh.getDimensions().x() - 1)
- {
- Index j = gy;
- Index i = Mesh.getDimensions().x() - 1;
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-// printf("##################################################################################################\n");
- if(gx == Mesh.getDimensions().x() - 1 &&
- gy == Mesh.getDimensions().y() - 1)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx-1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy-1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
- if(gx == Mesh.getDimensions().x() - 1 &&
- gy == 0)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx-1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy+1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
-// printf("$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$\n");
- if(gx == 0 &&
- gy == Mesh.getDimensions().y() - 1)
- {
-
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx+1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy-1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
- if(gx == 0 &&
- gy == 0)
- {
-// tmp = sign(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))]);
- if(cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx+1,gy))]*tmp > 0.0 &&
- cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy+1))]*tmp > 0.0)
-
- flag = 1;//cudaDofVector[Mesh.getCellIndex(CoordinatesType(gx,gy))] = tmp*INT_MAX;
- }
-
- __threadfence();
-
- if(flag==1)
- cudaDofVector[gid] = tmp*3;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
- //Real fy = abs(y);
-
- //Real tmpMin = Min(fx,abs(y));
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i)
-{
-
- extern __shared__ double u[];
- double* sharedMem[5];
- sharedMem[0] = u;
- sharedMem[1] = &(u[blockDim.y+1]);
- sharedMem[2] = &(sharedMem[1][blockDim.y+1]);
- sharedMem[3] = sharedMem[1];
- sharedMem[4] = sharedMem[2];
-
- int gx = 0;
- int gy = threadIdx.y;
- //if(solver->Mesh.getDimensions().x() <= gx || solver->Mesh.getDimensions().y() <= gy)
- // return;
- int n = solver->Mesh.getDimensions().x();
- int blockCount = n/blockDim.y +1;
- //int gid = solver->Mesh.getDimensions().x() * gy + gx;
- //int max = solver->Mesh.getDimensions().x()*solver->Mesh.getDimensions().x();
-
- //int id1 = gx+gy;
- //int id2 = (solver->Mesh.getDimensions().x() - gx - 1) + gy;
-
-
- if(blockIdx.x==0)
- {
- if(threadIdx.y==0)
- sharedMem[1][0]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(0,0))];
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- int k3=k%3;
-
- if(threadIdx.y==0)
- {
- if(gx==n-1)
- sharedMem[k3][0]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(0,gy+blockDim.y))];
- else
- sharedMem[k3][0]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx+1,gy))];
- }
-// else
-// solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx,gy-1))]=sharedMem[k3+2][threadIdx.y-1];
-
- if(gy<n-1)
- sharedMem[k3][threadIdx.y+1]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx,gy+1))];
-
- solver->updateValue(gx,gy,sharedMem,k3);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
-// else if(blockIdx.x==1)
-// {
-// gx=n-1;
-// gy=threadIdx.y;
-//
-// if(threadIdx.y==0)
-// sharedMem[1][0]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(n-1,0))];
-//
-// for(int k = 0; k < n*blockCount + blockDim.y; k++)
-// {
-// if(threadIdx.y < k+1 && gy < n)
-// {
-// int k3=k%3;
-//
-// if(threadIdx.y==0)
-// if(gx==0)
-// sharedMem[k3+2][0]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(n-1,gy+blockDim.y))];
-// else
-// sharedMem[k3+2][0]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx-1,gy))];
-// else
-// solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx,gy-1))]=sharedMem[k3][threadIdx.y-1];
-//
-// if(gy<n-1)
-// sharedMem[k3+2][threadIdx.y+1]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx,gy+1))];
-//
-//
-// solver->updateValue(gx,gy,sharedMem,k3);
-// gx--;
-// if(gx==-1)
-// {
-// gx=n-1;
-// gy+=blockDim.y;
-// }
-// }
-//
-//
-// __syncthreads();
-// }
-// }
-// else if(blockIdx.x==2)
-// {
-// gx=0;
-// gy=n-blockDim.y-1+threadIdx.y;
-//
-// if(threadIdx.y==0)
-// sharedMem[1][0]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(0,n-1))];
-//
-// for(int k = 0; k < n*blockCount + blockDim.y; k++)
-// {
-// if(blockDim.y-threadIdx.y < k+1 && gy > -1)
-// {
-// int k3=k%3;
-//
-// if(threadIdx.y==blockDim.y-1)
-// if(gx==n-1)
-// sharedMem[k3][n-1]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(0,gy-blockDim.y))];
-// else
-// sharedMem[k3][n-1]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx+1,gy))];
-// else
-// solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx,gy-1))]=sharedMem[k3+2][threadIdx.y-1];
-//
-// if(gy<n-1)
-// sharedMem[k3][threadIdx.y+1]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx,gy+1))];
-//
-//
-// solver->updateValue(gx,gy,sharedMem,k3);
-// gx++;
-// if(gx==n)
-// {
-// gx=0;
-// gy-=blockDim.y;
-// }
-// }
-//
-//
-// __syncthreads();
-// }
-// }
-// else if(blockIdx.x==3)
-// {
-// gx=n-1;
-// gy=n-blockDim.y-1+threadIdx.y;
-//
-// if(threadIdx.y==0)
-// sharedMem[1][0]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(n-1,n-1))];
-//
-//
-// for(int k = 0; k < n*blockCount + blockDim.y; k++)
-// {
-// if(blockDim.y-threadIdx.y < k+1 && gy > -1)
-// {
-// int k3=k%3;
-//
-// if(threadIdx.y==blockDim.y-1)
-// if(gx==n-1)
-// sharedMem[k3+2][n-1]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(n-1,gy-blockDim.y))];
-// else
-// sharedMem[k3+2][n-1]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx+1,gy))];
-// else
-// solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx,gy-1))]=sharedMem[k3][threadIdx.y-1];
-//
-// if(gy<n-1)
-// sharedMem[k3+2][threadIdx.y+1]=solver->cudaDofVector[solver->Mesh.getCellIndex(Containers::StaticVector<2,int>(gx,gy+1))];
-//
-//
-// solver->updateValue(gx,gy,sharedMem,k3);
-// gx--;
-// if(gx==-1)
-// {
-// gx=n-1;
-// gy-=blockDim.y;
-// }
-// }
-//
-//
-// __syncthreads();
-// }
-// }
-
-
-
-
-}
-
-
-__global__ void initCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver)
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy)
- {
- solver->initGrid();
- }
-
-
-}
-#endif
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_impl.h
deleted file mode 100644
index c4ce8fe6b2ee80131c0bc2a945bc8cef4377f106..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_impl.h
+++ /dev/null
@@ -1,927 +0,0 @@
-/***************************************************************************
- tnlFastSweeping2D_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: tnlFastSweeping()
-:Entity(Mesh),
- dofVector(Mesh),
- dofVector2(Mesh)
-{
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
- dofVector2.load(initialCondition);
-
- h = Mesh.template getSpaceStepsProducts< 1, 0 >();
- Entity.refresh();
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
- cout << "a" <<std::endl;
- return initGrid();
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
-
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- for(int i=0; i< Mesh.getDimensions().x()*Mesh.getDimensions().x();i++)
- {
- dofVector2[i]=INT_MAX*sign(dofVector[i]);
- }
-
- for(int i = 0 ; i < Mesh.getDimensions().x()-1; i++)
- {
- for(int j = 0 ; j < Mesh.getDimensions().x()-1; j++)
- {
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- neighborEntities.refresh(Mesh,Entity.getIndex());
-
- if(dofVector[this->Entity.getIndex()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1111(i,j);
- else
- setupSquare1110(i,j);
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1101(i,j);
- else
- setupSquare1100(i,j);
- }
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1011(i,j);
- else
- setupSquare1010(i,j);
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare1001(i,j);
- else
- setupSquare1000(i,j);
- }
- }
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0111(i,j);
- else
- setupSquare0110(i,j);
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0101(i,j);
- else
- setupSquare0100(i,j);
- }
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()] > 0)
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0011(i,j);
- else
- setupSquare0010(i,j);
- }
- else
- {
- if(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()] > 0)
- setupSquare0001(i,j);
- else
- setupSquare0000(i,j);
- }
- }
- }
-
- }
- }
- cout << "a" <<std::endl;
-
-// Real tmp = 0.0;
-// Real ax=0.5/sqrt(2.0);
-//
-// if(!exactInput)
-// {
-// for(Index i = 0; i < Mesh.getDimensions().x()*Mesh.getDimensions().y(); i++)
-// dofVector[i]=0.5*h*sign(dofVector[i]);
-// }
-//
-//
-// for(Index i = 1; i < Mesh.getDimensions().x()-1; i++)
-// {
-// for(Index j = 1; j < Mesh.getDimensions().y()-1; j++)
-// {
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-// }
-//
-//
-//
-// for(int i = 1; i < Mesh.getDimensions().x()-1; i++)
-// {
-// Index j = 0;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-//
-// for(int i = 1; i < Mesh.getDimensions().x()-1; i++)
-// {
-// Index j = Mesh.getDimensions().y() - 1;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-//
-// for(int j = 1; j < Mesh.getDimensions().y()-1; j++)
-// {
-// Index i = 0;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-//
-// for(int j = 1; j < Mesh.getDimensions().y()-1; j++)
-// {
-// Index i = Mesh.getDimensions().x() - 1;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-//
-//
-// if(tmp == 0.0)
-// {}
-// else if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
-// {}
-// else
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-// }
-//
-//
-// Index i = Mesh.getDimensions().x() - 1;
-// Index j = Mesh.getDimensions().y() - 1;
-//
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-// if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp > 0.0 &&
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp > 0.0)
-//
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-//
-//
-//
-// j = 0;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-// if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp > 0.0 &&
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp > 0.0)
-//
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-//
-//
-//
-// i = 0;
-// j = Mesh.getDimensions().y() -1;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-// if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp > 0.0 &&
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp > 0.0)
-//
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-//
-//
-//
-// j = 0;
-// tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-// if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp > 0.0 &&
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp > 0.0)
-//
-// dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-
- //data.setLike(dofVector2.getData());
- //data=dofVector2.getData();
- //cout << data.getType() <<std::endl;
- dofVector2.save("u-00000.tnl");
- //dofVector2.getData().save("u-00000.tnl");
-
- return true;
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
-
-// data.setLike(dofVector2.getData());
-// data = dofVector2.getData();
-// cout << data.getType() <<std::endl;
- dofVector2.save("u-00001.tnl");
- //dofVector2.getData().save("u-00001.tnl");
-
- return true;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j)
-{
-
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- Real value = dofVector2[Entity.getIndex()];
- Real a,b, tmp;
-
- if( i == 0 )
- a = dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = dofVector2[neighborEntities.template getEntityIndex< -1, 0 >()];
- else
- {
- a = fabsMin( dofVector2[neighborEntities.template getEntityIndex< -1, 0 >()],
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()] );
- }
-
- if( j == 0 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, -1 >()];
- else
- {
- b = fabsMin( dofVector2[neighborEntities.template getEntityIndex< 0, -1 >()],
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()] );
- }
-
-
- if(fabs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
-
- dofVector2[Entity.getIndex()] = fabsMin(value, tmp);
-
-// if(dofVector2[Entity.getIndex()] > 1.0)
-// cout << value << " " << tmp << " " << dofVector2[Entity.getIndex()] <<std::endl;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-Real tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = fabs(x);
- Real fy = fabs(y);
-
- Real tmpMin = Min(fx,fy);
-
- if(tmpMin == fx)
- return x;
- else
- return y;
-
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-{
-// this->Entity.setCoordinates(CoordinatesType(i,j));
-// this->Entity.refresh();
-// auto neighborEntities = Entity.getNeighborEntities();
-// dofVector2[Entity.getIndex()]=fabsMin(INT_MAX,dofVector2[Entity.getIndex()]);
-// dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-{
-// this->Entity.setCoordinates(CoordinatesType(i,j));
-// this->Entity.refresh();
-// auto neighborEntities = Entity.getNeighborEntities();
-// dofVector2[Entity.getIndex()]=fabsMin(-INT_MAX,dofVector2[(Entity.getIndex())]);
-// dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
-// dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-INT_MAX,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[Entity.getIndex()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- a = be/al;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(-abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- dofVector2[Entity.getIndex()]=fabsMin(dofVector[Entity.getIndex()],dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]));
-
- a = al-be;
- b=1.0;
- c=-al;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- Real al,be, a,b,c,s;
- al=abs(dofVector[Entity.getIndex()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 0 >()]-
- dofVector[Entity.getIndex()]));
-
- be=abs(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]/
- (dofVector[neighborEntities.template getEntityIndex< 1, 1 >()]-
- dofVector[neighborEntities.template getEntityIndex< 0, 1 >()]));
-
- a = al-be;
- b=1.0;
- c=-be;
- s= h/sqrt(a*a+b*b);
-
-
- dofVector2[Entity.getIndex()]=fabsMin(-abs(a*0+b*0+c)*s,dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(-abs(a*1+b*0+c)*s,dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(abs(a*1+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(abs(a*0+b*1+c)*s,dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j));
- this->Entity.refresh();
- auto neighborEntities = Entity.getNeighborEntities();
- dofVector2[Entity.getIndex()]=fabsMin(dofVector[Entity.getIndex()],dofVector2[(Entity.getIndex())]);
- dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 0, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 0, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 1 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 1 >()]);
- dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]=fabsMin(dofVector[neighborEntities.template getEntityIndex< 1, 0 >()],dofVector2[neighborEntities.template getEntityIndex< 1, 0 >()]);
-}
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_openMP_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_openMP_impl.h
deleted file mode 100644
index 54bbe931e0ae305ac300027d7da850d705b4c309..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping2D_openMP_impl.h
+++ /dev/null
@@ -1,399 +0,0 @@
-/***************************************************************************
- tnlFastSweeping_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING2D_IMPL_H_
-#define TNLFASTSWEEPING2D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- h = Mesh.getSpaceSteps().x();
-
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-#ifdef HAVE_OPENMP
-// gridLock = (omp_lock_t*) malloc(sizeof(omp_lock_t)*Mesh.getDimensions().x()*Mesh.getDimensions().y());
-//
-// for(int i = 0; i < Mesh.getDimensions().x()*Mesh.getDimensions().y(); i++)
-// omp_init_lock(&gridLock[i]);
-#endif
-
- return initGrid();
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
-
- Real tmp = 0.0;
-
- if(!exactInput)
- {
- for(Index i = 0; i < Mesh.getDimensions().x()*Mesh.getDimensions().y(); i++)
- dofVector[i]=0.5*h*sign(dofVector[i]);
- }
-
-
- for(Index i = 1; i < Mesh.getDimensions().x()-1; i++)
- {
- for(Index j = 1; j < Mesh.getDimensions().y()-1; j++)
- {
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
- }
-
-
-
- for(int i = 1; i < Mesh.getDimensions().x()-1; i++)
- {
- Index j = 0;
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 )
- {}
- else
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
- for(int i = 1; i < Mesh.getDimensions().x()-1; i++)
- {
- Index j = Mesh.getDimensions().y() - 1;
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
- for(int j = 1; j < Mesh.getDimensions().y()-1; j++)
- {
- Index i = 0;
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
- for(int j = 1; j < Mesh.getDimensions().y()-1; j++)
- {
- Index i = Mesh.getDimensions().x() - 1;
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
-
-
- if(tmp == 0.0)
- {}
- else if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp < 0.0 ||
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp < 0.0 )
- {}
- else
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
- }
-
-
- Index i = Mesh.getDimensions().x() - 1;
- Index j = Mesh.getDimensions().y() - 1;
-
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
- if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp > 0.0 &&
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp > 0.0)
-
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-
-
-
- j = 0;
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
- if(dofVector[Mesh.getCellIndex(CoordinatesType(i-1,j))]*tmp > 0.0 &&
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp > 0.0)
-
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-
-
-
- i = 0;
- j = Mesh.getDimensions().y() -1;
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
- if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp > 0.0 &&
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j-1))]*tmp > 0.0)
-
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-
-
-
- j = 0;
- tmp = sign(dofVector[Mesh.getCellIndex(CoordinatesType(i,j))]);
- if(dofVector[Mesh.getCellIndex(CoordinatesType(i+1,j))]*tmp > 0.0 &&
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j+1))]*tmp > 0.0)
-
- dofVector[Mesh.getCellIndex(CoordinatesType(i,j))] = tmp*INT_MAX;
-
-
- dofVector.save("u-00000.tnl");
-
- return true;
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- DofVectorType d2,d3,d4;
- d2.setLike(dofVector);
- d2=dofVector;
- d3.setLike(dofVector);
- d3=dofVector;
- d4.setLike(dofVector);
- d4=dofVector;
-
-
-#ifdef HAVE_OPENMP
-#pragma omp parallel sections num_threads(4)
- {
- {
-#endif
-
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,&dofVector);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-#ifdef HAVE_OPENMP
- }
-#pragma omp section
- {
-#endif
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,&d2);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-#ifdef HAVE_OPENMP
- }
-#pragma omp section
- {
-#endif
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j, &d3);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-#ifdef HAVE_OPENMP
- }
-#pragma omp section
- {
-#endif
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j, &d4);
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-#ifdef HAVE_OPENMP
- }
- }
-#endif
-
-
-#ifdef HAVE_OPENMP
-#pragma omp parallel for num_threads(4) schedule(dynamic)
-#endif
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- int index = Mesh.getCellIndex(CoordinatesType(i,j));
- dofVector[index] = fabsMin(dofVector[index], d2[index]);
- dofVector[index] = fabsMin(dofVector[index], d3[index]);
- dofVector[index] = fabsMin(dofVector[index], d4[index]);
- }
- }
-
- dofVector.save("u-00001.tnl");
-
- return true;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j, DofVectorType* grid)
-{
- Index index = Mesh.getCellIndex(CoordinatesType(i,j));
- Real value = (*grid)[index];
- Real a,b, tmp;
-
- if( i == 0 )
- a = (*grid)[Mesh.template getCellNextToCell<1,0>(index)];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = (*grid)[Mesh.template getCellNextToCell<-1,0>(index)];
- else
- {
- a = fabsMin( (*grid)[Mesh.template getCellNextToCell<-1,0>(index)],
- (*grid)[Mesh.template getCellNextToCell<1,0>(index)] );
- }
-
- if( j == 0 )
- b = (*grid)[Mesh.template getCellNextToCell<0,1>(index)];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = (*grid)[Mesh.template getCellNextToCell<0,-1>(index)];
- else
- {
- b = fabsMin( (*grid)[Mesh.template getCellNextToCell<0,-1>(index)],
- (*grid)[Mesh.template getCellNextToCell<0,1>(index)] );
- }
-
-
- if(fabs(a-b) >= h)
- tmp = fabsMin(a,b) + sign(value)*h;
- else
- tmp = 0.5 * (a + b + sign(value)*sqrt(2.0 * h * h - (a - b) * (a - b) ) );
-
-#ifdef HAVE_OPENMP
-// omp_set_lock(&gridLock[index]);
-#endif
- (*grid)[index] = fabsMin(value, tmp);
-#ifdef HAVE_OPENMP
-// omp_unset_lock(&gridLock[index]);
-#endif
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-Real tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = fabs(x);
- Real fy = fabs(y);
-
- Real tmpMin = Min(fx,fy);
-
- if(tmpMin == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping3D_CUDA_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping3D_CUDA_impl.h
deleted file mode 100644
index 6a5195cfe4edda5754b8826158d6d9faa5701fa0..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping3D_CUDA_impl.h
+++ /dev/null
@@ -1,961 +0,0 @@
-/***************************************************************************
- tnlFastSweeping2D_CUDA_v4_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING3D_IMPL_H_
-#define TNLFASTSWEEPING3D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-//__device__
-//double fabsMin( double x, double y)
-//{
-// double fx = abs(x);
-//
-// if(Min(fx,abs(y)) == fx)
-// return x;
-// else
-// return y;
-//}
-//
-//__device__
-//double atomicFabsMin(double* address, double val)
-//{
-// unsigned long long int* address_as_ull =
-// (unsigned long long int*)address;
-// unsigned long long int old = *address_as_ull, assumed;
-// do {
-// assumed = old;
-// old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(assumed,val) ));
-// } while (assumed != old);
-// return __longlong_as_double(old);
-//}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
- this->h = Mesh.template getSpaceStepsProducts< 1, 0, 0 >();
- counter = 0;
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-
-
-#ifdef HAVE_CUDA
-
- cudaMalloc(&(cudaDofVector), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
- cudaMalloc(&(cudaDofVector2), this->dofVector.getData().getSize()*sizeof(double));
- cudaMemcpy(cudaDofVector2, this->dofVector.getData().getData(), this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyHostToDevice);
-
-
- cudaMalloc(&(this->cudaSolver), sizeof(tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >), cudaMemcpyHostToDevice);
-
-#endif
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(8, 8,8);
- dim3 numBlocks(n/8 + 1, n/8 +1, n/8 +1);
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- return true;
-}
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- int n = Mesh.getDimensions().x();
- dim3 threadsPerBlock(1, 1024);
- dim3 numBlocks(8,1);
-
-
- runCUDA<<<numBlocks,threadsPerBlock>>>(this->cudaSolver,0,0);
-
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- cudaMemcpy(this->dofVector.getData().getData(), cudaDofVector2, this->dofVector.getData().getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- cudaFree(cudaDofVector);
- cudaFree(cudaDofVector2);
- cudaFree(cudaSolver);
- dofVector.save("u-00001.tnl");
- cudaDeviceSynchronize();
- return true;
-}
-
-
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j, Index k)
-{
- tnlGridEntity< tnlGrid< 3,double, TNL::Devices::Host, int >, 3, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j,k));
- Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage >,3> neighborEntities(Entity);
- Real value = cudaDofVector2[Entity.getIndex()];
- Real a,b,c, tmp;
-
- if( i == 0 )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0, 0 >()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0, 0 >()];
- else
- {
- a = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< -1, 0, 0 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 1, 0, 0 >()] );
- }
-
- if( j == 0 )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1, 0 >()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1, 0 >()];
- else
- {
- b = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< 0, -1, 0 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 1, 0 >()] );
- }
-
- if( k == 0 )
- c = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 0, 1 >()];
- else if( k == Mesh.getDimensions().z() - 1 )
- c = cudaDofVector2[neighborEntities.template getEntityIndex< 0, 0, -1 >()];
- else
- {
- c = fabsMin( cudaDofVector2[neighborEntities.template getEntityIndex< 0, 0, -1 >()],
- cudaDofVector2[neighborEntities.template getEntityIndex< 0, 0, 1 >()] );
- }
-
- Real hD = 3.0*h*h - 2.0*(a*a + b*b + c*c - a*b - a*c - b*c);
-
- if(hD < 0.0)
- tmp = fabsMin(a,fabsMin(b,c)) + sign(value)*h;
- else
- tmp = (1.0/3.0) * ( a + b + c + sign(value)*sqrt(hD) );
-
- atomicFabsMin(&cudaDofVector2[Entity.getIndex()],tmp);
-
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-bool tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid(int i, int j, int k)
-{
- tnlGridEntity< tnlGrid< 3,double, TNL::Devices::Host, int >, 3, tnlGridEntityNoStencilStorage > Entity(Mesh);
- Entity.setCoordinates(CoordinatesType(i,j,k));
- Entity.refresh();
- int gid = Entity.getIndex();
-
- if(abs(cudaDofVector[gid]) < 1.0*h)
- cudaDofVector2[gid] = 0.5*h;//cudaDofVector[gid];
- else
- cudaDofVector2[gid] = INT_MAX*sign(cudaDofVector[gid]);
-
- return true;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-__device__
-Real tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = abs(x);
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-
-
-}
-
-
-
-__global__ void runCUDA(tnlFastSweeping< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i)
-{
-
- int gx = 0;
- int gy = threadIdx.y;
-
- int n = solver->Mesh.getDimensions().x();
- int blockCount = n/blockDim.y +1;
-
- if(blockIdx.x==0)
- {
- for(int gz = 0; gz < n;gz++)
- {
- gx = 0;
- gy = threadIdx.y;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,gz);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- __syncthreads();
- }
- }
- else if(blockIdx.x==1)
- {
- for(int gz = 0; gz < n;gz++)
- {
- gx=n-1;
- gy=threadIdx.y;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,gz);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==2)
- {
-
- for(int gz = 0; gz < n;gz++)
- {
- gx=0;
- gy=n-threadIdx.y-1;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,gz);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==3)
- {
- for(int gz = 0; gz < n;gz++)
- {
- gx=n-1;
- gy=n-threadIdx.y-1;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,gz);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
-
-
-
-
- else if(blockIdx.x==4)
- {
- for(int gz = n-1; gz > -1;gz--)
- {
- gx = 0;
- gy = threadIdx.y;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,gz);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==5)
- {
- for(int gz = n-1; gz > -1;gz--)
- {
- gx=n-1;
- gy=threadIdx.y;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy < n)
- {
- solver->updateValue(gx,gy,gz);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy+=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==6)
- {
-
- for(int gz = n-1; gz > -1;gz--)
- {
- gx=0;
- gy=n-threadIdx.y-1;
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,gz);
- gx++;
- if(gx==n)
- {
- gx=0;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
- else if(blockIdx.x==7)
- {
- for(int gz = n-1; gz > -1;gz--)
- {
- gx=n-1;
- gy=n-threadIdx.y-1;
-
- for(int k = 0; k < n*blockCount + blockDim.y; k++)
- {
- if(threadIdx.y < k+1 && gy > -1)
- {
- solver->updateValue(gx,gy,gz);
- gx--;
- if(gx==-1)
- {
- gx=n-1;
- gy-=blockDim.y;
- }
- }
-
-
- __syncthreads();
- }
- }
- }
-
-
-
-
-}
-
-
-__global__ void initCUDA(tnlFastSweeping< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver)
-{
- int gx = threadIdx.x + blockDim.x*blockIdx.x;
- int gy = blockDim.y*blockIdx.y + threadIdx.y;
- int gz = blockDim.z*blockIdx.z + threadIdx.z;
-
- if(solver->Mesh.getDimensions().x() > gx && solver->Mesh.getDimensions().y() > gy && solver->Mesh.getDimensions().z() > gz)
- {
- solver->initGrid(gx,gy,gz);
- }
-
-
-}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1111( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// cudaDofVector2[index]=fabsMin(INT_MAX,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0000( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// cudaDofVector2[index]=fabsMin(-INT_MAX,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-INT_MAX,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1110( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1101( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1011( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0111( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0001( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0010( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0100( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1000( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// a = be/al;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1100( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// a = al-be;
-// b=1.0;
-// c=-al;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1010( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// a = al-be;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(-abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare1001( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// cudaDofVector2[index]=fabsMin(cudaDofVector[index],cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)],cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)],cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)],cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//
-//
-//
-//
-//
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0011( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]));
-//
-// a = al-be;
-// b=1.0;
-// c=-al;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0101( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// Real al,be, a,b,c,s;
-// al=abs(cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,0>(index)]));
-//
-// be=abs(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]/
-// (cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)]-
-// cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)]));
-//
-// a = al-be;
-// b=1.0;
-// c=-be;
-// s= h/sqrt(a*a+b*b);
-//
-//
-// cudaDofVector2[index]=fabsMin(-abs(a*0+b*0+c)*s,cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(-abs(a*1+b*0+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(abs(a*1+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(abs(a*0+b*1+c)*s,cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//
-//}
-//
-//template< typename MeshReal,
-// typename Device,
-// typename MeshIndex,
-// typename Real,
-// typename Index >
-//void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: setupSquare0110( Index i, Index j)
-//{
-// Index index = Mesh.getCellIndex(CoordinatesType(i,j));
-// cudaDofVector2[index]=fabsMin(cudaDofVector[index],cudaDofVector2[(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<0,1>(index)],cudaDofVector2[Mesh.template getCellNextToCell<0,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<1,1>(index)],cudaDofVector2[Mesh.template getCellNextToCell<1,1>(index)]);
-// cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]=fabsMin(cudaDofVector[Mesh.template getCellNextToCell<1,0>(index)],cudaDofVector2[Mesh.template getCellNextToCell<1,0>(index)]);
-//}
-#endif
-
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping3D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping3D_impl.h
deleted file mode 100644
index e22de0ab85f3814004b072ef0ebdb403b7f970c7..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping3D_impl.h
+++ /dev/null
@@ -1,307 +0,0 @@
-/***************************************************************************
- tnlFastSweeping2D_impl.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING3D_IMPL_H_
-#define TNLFASTSWEEPING3D_IMPL_H_
-
-#include "tnlFastSweeping.h"
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-String tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: getType()
-{
- return String( "tnlFastSweeping< " ) +
- MeshType::getType() + ", " +
- ::getType< Real >() + ", " +
- ::getType< Index >() + " >";
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: tnlFastSweeping()
-:Entity(Mesh),
- dofVector(Mesh),
- dofVector2(Mesh)
-{
-}
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: init( const Config::ParameterContainer& parameters )
-{
- const String& meshFile = parameters.getParameter< String >( "mesh" );
-
- if( ! Mesh.load( meshFile ) )
- {
- std::cerr << "I am not able to load the mesh from the file " << meshFile << "." <<std::endl;
- return false;
- }
-
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- if( ! dofVector.load( initialCondition ) )
- {
- std::cerr << "I am not able to load the initial condition from the file " << meshFile << "." <<std::endl;
- return false;
- }
- dofVector2.load(initialCondition);
-
- h = Mesh.template getSpaceStepsProducts< 1, 0, 0 >();
- Entity.refresh();
-
- const String& exact_input = parameters.getParameter< String >( "exact-input" );
-
- if(exact_input == "no")
- exactInput=false;
- else
- exactInput=true;
-// cout << "bla "<<endl;
- return initGrid();
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: initGrid()
-{
- for(int i=0; i< Mesh.getDimensions().x()*Mesh.getDimensions().y()*Mesh.getDimensions().z();i++)
- {
-
- if (abs(dofVector[i]) < 1.8*h)
- dofVector2[i]=dofVector[i];
- else
- dofVector2[i]=INT_MAX*sign(dofVector[i]);
- }
-
- return true;
-}
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-bool tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: run()
-{
-
- for(Index k = 0; k < Mesh.getDimensions().z(); k++)
- {
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index k = 0; k < Mesh.getDimensions().z(); k++)
- {
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index k = 0; k < Mesh.getDimensions().z(); k++)
- {
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
- for(Index k = 0; k < Mesh.getDimensions().z(); k++)
- {
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
-
-
-
-
-
-
-
- for(Index k = Mesh.getDimensions().z() -1; k > -1; k--)
- {
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index k = Mesh.getDimensions().z() -1; k > -1; k--)
- {
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = 0; j < Mesh.getDimensions().y(); j++)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
- for(Index k = Mesh.getDimensions().z() -1; k > -1; k--)
- {
- for(Index i = Mesh.getDimensions().x() - 1; i > -1; i--)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
- for(Index k = Mesh.getDimensions().z() -1; k > -1; k--)
- {
- for(Index i = 0; i < Mesh.getDimensions().x(); i++)
- {
- for(Index j = Mesh.getDimensions().y() - 1; j > -1; j--)
- {
- updateValue(i,j,k);
- }
- }
- }
-
-/*---------------------------------------------------------------------------------------------------------------------------*/
-
-
- dofVector2.save("u-00001.tnl");
-
- cout << "bla 3"<<endl;
- return true;
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-void tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: updateValue( Index i, Index j, Index k)
-{
- this->Entity.setCoordinates(CoordinatesType(i,j,k));
- this->Entity.refresh();
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage >,3> neighborEntities(Entity);
- Real value = dofVector2[Entity.getIndex()];
- Real a,b,c, tmp;
-
- if( i == 0 )
- a = dofVector2[neighborEntities.template getEntityIndex< 1, 0, 0>()];
- else if( i == Mesh.getDimensions().x() - 1 )
- a = dofVector2[neighborEntities.template getEntityIndex< -1, 0, 0 >()];
- else
- {
- a = fabsMin( dofVector2[neighborEntities.template getEntityIndex< -1, 0, 0>()],
- dofVector2[neighborEntities.template getEntityIndex< 1, 0, 0>()] );
- }
-
- if( j == 0 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, 1, 0>()];
- else if( j == Mesh.getDimensions().y() - 1 )
- b = dofVector2[neighborEntities.template getEntityIndex< 0, -1, 0>()];
- else
- {
- b = fabsMin( dofVector2[neighborEntities.template getEntityIndex< 0, -1, 0>()],
- dofVector2[neighborEntities.template getEntityIndex< 0, 1, 0>()] );
- }
-
- if( k == 0 )
- c = dofVector2[neighborEntities.template getEntityIndex< 0, 0, 1>()];
- else if( k == Mesh.getDimensions().z() - 1 )
- c = dofVector2[neighborEntities.template getEntityIndex< 0, 0, -1>()];
- else
- {
- c = fabsMin( dofVector2[neighborEntities.template getEntityIndex< 0, 0, -1>()],
- dofVector2[neighborEntities.template getEntityIndex< 0, 0, 1>()] );
- }
-
- Real hD = 3.0*h*h - 2.0*(a*a+b*b+c*c-a*b-a*c-b*c);
-
- if(hD < 0.0)
- tmp = fabsMin(a,fabsMin(b,c)) + sign(value)*h;
- else
- tmp = (1.0/3.0) * ( a + b + c + sign(value)*sqrt(hD) );
-
-
- dofVector2[Entity.getIndex()] = fabsMin(value, tmp);
-}
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-Real tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index > :: fabsMin( Real x, Real y)
-{
- Real fx = fabs(x);
- Real fy = fabs(y);
-
- Real tmpMin = Min(fx,fy);
-
- if(tmpMin == fx)
- return x;
- else
- return y;
-
-}
-
-
-
-#endif /* TNLFASTSWEEPING_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweepingSolver.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweepingSolver.h
deleted file mode 100644
index fc9eb545987dfa14bb17b28f4d104d4c6e6fa2e7..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweepingSolver.h
+++ /dev/null
@@ -1,36 +0,0 @@
-/*
- * File: tnlFastSweepingSolver.h
- * Author: oberhuber
- *
- * Created on July 12, 2016, 6:04 PM
- */
-
-#pragma once
-
-#include <functions/tnlConstantFunction.h>
-#include <problems/tnlPDEProblem.h>
-
-template< typename Mesh,
- typename Communicator,
- typename Anisotropy = tnlConstanstFunction< Mesh > >
-class tnlFastSweepingSolver : public tnlPDEProblem< Mesh,
- Communicator,
- typename Mesh::RealType,
- typename Mesh::DeviceType,
- typename Mesh::IndexType >
-{
- public:
-
- typedef typename DifferentialOperator::RealType RealType;
- typedef typename Mesh::DeviceType DeviceType;
- typedef typename DifferentialOperator::IndexType IndexType;
-
- typedef tnlMeshFunction< Mesh > MeshFunctionType;
- typedef tnlPDEProblem< Mesh, TimeDependentProblem, RealType, DeviceType, IndexType > BaseType;
-
- using typename BaseType::MeshType;
- using typename BaseType::DofVectorType;
- using typename BaseType::MeshDependentDataType;
-};
-
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping_CUDA.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping_CUDA.h
deleted file mode 100644
index f531da431bfec5d16da8ea7deabe6595031a0873..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/fast-sweeping/tnlFastSweeping_CUDA.h
+++ /dev/null
@@ -1,194 +0,0 @@
-/***************************************************************************
- tnlFastSweeping_CUDA.h - description
- -------------------
- begin : Oct 15 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-#ifndef TNLFASTSWEEPING_H_
-#define TNLFASTSWEEPING_H_
-
-#include <TNL/Config/ParameterContainer.h>
-#include <TNL/Containers/Vector.h>
-#include <TNL/Containers/StaticVector.h>
-#include <TNL/Devices/Host.h>
-#include <mesh/tnlGrid.h>
-#include <mesh/grids/tnlGridEntity.h>
-
-#include <functions/tnlMeshFunction.h>
-#include <limits.h>
-#include <core/tnlDevice.h>
-#include <ctime>
-
-
-
-
-
-template< typename Mesh,
- typename Real,
- typename Index >
-class tnlFastSweeping
-{};
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 2, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
- tnlFastSweeping();
-
- __host__ static String getType();
- __host__ bool init( const Config::ParameterContainer& parameters );
- __host__ bool run();
-
-#ifdef HAVE_CUDA
- __device__ bool initGrid();
- __device__ void updateValue(const Index i, const Index j);
- __device__ void updateValue(const Index i, const Index j, double** sharedMem, const int k3);
- __device__ Real fabsMin(const Real x, const Real y);
-
- tnlFastSweeping< tnlGrid< 2,MeshReal, Device, MeshIndex >, Real, Index >* cudaSolver;
- double* cudaDofVector;
- double* cudaDofVector2;
- int counter;
- __device__ void setupSquare1000(Index i, Index j);
- __device__ void setupSquare1100(Index i, Index j);
- __device__ void setupSquare1010(Index i, Index j);
- __device__ void setupSquare1001(Index i, Index j);
- __device__ void setupSquare1110(Index i, Index j);
- __device__ void setupSquare1101(Index i, Index j);
- __device__ void setupSquare1011(Index i, Index j);
- __device__ void setupSquare1111(Index i, Index j);
- __device__ void setupSquare0000(Index i, Index j);
- __device__ void setupSquare0100(Index i, Index j);
- __device__ void setupSquare0010(Index i, Index j);
- __device__ void setupSquare0001(Index i, Index j);
- __device__ void setupSquare0110(Index i, Index j);
- __device__ void setupSquare0101(Index i, Index j);
- __device__ void setupSquare0011(Index i, Index j);
- __device__ void setupSquare0111(Index i, Index j);
-#endif
-
- MeshType Mesh;
-
-protected:
-
-
-
- bool exactInput;
-
- tnlMeshFunction<MeshType> dofVector;
- DofVectorType data;
-
-
- RealType h;
-
-
-};
-
-
-
-
-
-
-
-
-
-template< typename MeshReal,
- typename Device,
- typename MeshIndex,
- typename Real,
- typename Index >
-class tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >
-{
-
-public:
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef tnlGrid< 3, Real, Device, Index > MeshType;
- typedef TNL::Containers::Vector< RealType, DeviceType, IndexType> DofVectorType;
- typedef typename MeshType::CoordinatesType CoordinatesType;
-
-
-
- __host__ static String getType();
- __host__ bool init( const Config::ParameterContainer& parameters );
- __host__ bool run();
-
-#ifdef HAVE_CUDA
- __device__ bool initGrid(int i, int j, int k);
- __device__ void updateValue(const Index i, const Index j, const Index k);
- __device__ void updateValue(const Index i, const Index j, const Index k, double** sharedMem, const int k3);
- __device__ Real fabsMin(const Real x, const Real y);
-
- tnlFastSweeping< tnlGrid< 3,MeshReal, Device, MeshIndex >, Real, Index >* cudaSolver;
- double* cudaDofVector;
- double* cudaDofVector2;
- int counter;
-#endif
-
- MeshType Mesh;
-
-protected:
-
-
-
- bool exactInput;
-
- tnlMeshFunction<MeshType> dofVector;
- DofVectorType data;
-
- RealType h;
-
-
-};
-
-
-
-
-
-
-
-#ifdef HAVE_CUDA
-//template<int sweep_t>
-__global__ void runCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i);
-__global__ void runCUDA(tnlFastSweeping< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver, int sweep, int i);
-
-__global__ void initCUDA(tnlFastSweeping< tnlGrid< 2,double, TNL::Devices::Host, int >, double, int >* solver);
-__global__ void initCUDA(tnlFastSweeping< tnlGrid< 3,double, TNL::Devices::Host, int >, double, int >* solver);
-#endif
-
-/*various implementtions.... choose one*/
-//#include "tnlFastSweeping2D_CUDA_impl.h"
-//#include "tnlFastSweeping2D_CUDA_v2_impl.h"
-//#include "tnlFastSweeping2D_CUDA_v3_impl.h"
-#include "tnlFastSweeping2D_CUDA_v4_impl.h"
-//#include "tnlFastSweeping2D_CUDA_v5_impl.h"
-
-
-#include "tnlFastSweeping3D_CUDA_impl.h"
-
-#endif /* TNLFASTSWEEPING_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/CMakeLists.txt b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/CMakeLists.txt
deleted file mode 100644
index 48382df82de6e7e175e47d24fc3ce69e13c217b5..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/CMakeLists.txt
+++ /dev/null
@@ -1,23 +0,0 @@
-set( tnl_hamilton_jacobi_parallel_map_SOURCES
-# MainBuildConfig.h
-# tnlParallelMapSolver2D_impl.h
-# tnlParallelMapSolver.h
-# parallelMapConfig.h
-# main.cu
- main.cpp)
-
-
-IF( BUILD_CUDA )
- CUDA_ADD_EXECUTABLE(hamilton-jacobi-parallel-map main.cu)
-ELSE( BUILD_CUDA )
- ADD_EXECUTABLE(hamilton-jacobi-parallel-map main.cpp)
-ENDIF( BUILD_CUDA )
-target_link_libraries (hamilton-jacobi-parallel-map tnl )
-
-
-INSTALL( TARGETS hamilton-jacobi-parallel-map
- RUNTIME DESTINATION bin
- PERMISSIONS OWNER_READ OWNER_WRITE OWNER_EXECUTE GROUP_READ GROUP_EXECUTE WORLD_READ WORLD_EXECUTE )
-
-#INSTALL( FILES ${tnl_hamilton_jacobi_parallel_map_SOURCES}
-# DESTINATION ${TNL_TARGET_DATA_DIRECTORY}/examples/hamilton-jacobi-parallel-map )
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/MainBuildConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/MainBuildConfig.h
deleted file mode 100644
index ed3d686eb99379af1589d734eac9b5812cccdedf..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/MainBuildConfig.h
+++ /dev/null
@@ -1,64 +0,0 @@
-/***************************************************************************
- MainBuildConfig.h - description
- -------------------
- begin : Jul 7, 2014
- copyright : (C) 2014 by Tomas Oberhuber
- email : tomas.oberhuber@fjfi.cvut.cz
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef MAINBUILDCONFIG_H_
-#define MAINBUILDCONFIG_H_
-
-#include <solvers/tnlBuildConfigTags.h>
-
-class MainBuildConfig
-{
- public:
-
- static void print() {std::cerr << "MainBuildConfig" <<std::endl; }
-};
-
-/****
- * Turn off support for float and long double.
- */
-template<> struct tnlConfigTagReal< MainBuildConfig, float > { enum { enabled = false }; };
-template<> struct tnlConfigTagReal< MainBuildConfig, long double > { enum { enabled = false }; };
-
-/****
- * Turn off support for short int and long int indexing.
- */
-template<> struct tnlConfigTagIndex< MainBuildConfig, short int >{ enum { enabled = false }; };
-template<> struct tnlConfigTagIndex< MainBuildConfig, long int >{ enum { enabled = false }; };
-
-/****
- * Use of tnlGrid is enabled for allowed dimensions and Real, Device and Index types.
- */
-template< int Dimensions, typename Real, typename Device, typename Index >
- struct tnlConfigTagMesh< MainBuildConfig, tnlGrid< Dimensions, Real, Device, Index > >
- { enum { enabled = tnlConfigTagDimensions< MainBuildConfig, Dimensions >::enabled &&
- tnlConfigTagReal< MainBuildConfig, Real >::enabled &&
- tnlConfigTagDevice< MainBuildConfig, Device >::enabled &&
- tnlConfigTagIndex< MainBuildConfig, Index >::enabled }; };
-
-/****
- * Please, chose your preferred time discretisation here.
- */
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlExplicitTimeDiscretisationTag >{ enum { enabled = true }; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlSemiImplicitTimeDiscretisationTag >{ enum { enabled = false}; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlImplicitTimeDiscretisationTag >{ enum { enabled = false }; };
-
-/****
- * Only the Runge-Kutta-Merson solver is enabled by default.
- */
-template<> struct tnlConfigTagExplicitSolver< MainBuildConfig, tnlExplicitEulerSolverTag >{ enum { enabled = false }; };
-
-#endif /* MAINBUILDCONFIG_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/gnuplot.txt b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/gnuplot.txt
deleted file mode 100644
index d4ae61983910a676269a23e3d992f5f46ea83a8f..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/gnuplot.txt
+++ /dev/null
@@ -1,32 +0,0 @@
-tomas@tomas-linux:~/Desktop/VU_CPU_MAPA/work_dir$ gnuplot
-
- G N U P L O T
- Version 4.6 patchlevel 4 last modified 2013-10-02
- Build System: Linux x86_64
-
- Copyright (C) 1986-1993, 1998, 2004, 2007-2013
- Thomas Williams, Colin Kelley and many others
-
- gnuplot home: http://www.gnuplot.info
- faq, bugs, etc: type "help FAQ"
- immediate help: type "help" (plot window: hit 'h')
-
-Terminal type set to 'wxt'
-gnuplot> set cntrparam levels 15
-gnuplot> set cntrparam bspline
-gnuplot> set contour
-gnuplot> splot 'u-00001.gplt'
-
-gnuplot> unset surface
-gnuplot> splot 'u-00001.gplt'
-
-gnuplot> set table "test.gplt"
-gnuplot> splot 'u-00001.gplt'
-gnuplot> unset table
-
-gnuplot> set table "test2.gplt"
-gnuplot> plot 'test.gplt' index 10
-gnuplot> unset table
-
-gnuplot> plot 'test2.gplt'
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.cpp b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.cpp
deleted file mode 100644
index b13498e17330fae7bb00a0bdc2abcc7a19f8e7a8..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.cpp
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cpp - description
- -------------------
- begin : Jul 8 , 2014
- copyright : (C) 2014 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.cu b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.cu
deleted file mode 100644
index 7101976712e153d73c5f0979b211164a36ec648d..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.cu
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cu - description
- -------------------
- begin : Mar 30 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.h
deleted file mode 100644
index fff21c77eb5980a3f3f86c28170b4169dd6f7917..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/main.h
+++ /dev/null
@@ -1,98 +0,0 @@
-/***************************************************************************
- main.h - description
- -------------------
- begin : Mar 22 , 2016
- copyright : (C) 2016 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "tnlParallelMapSolver.h"
-#include "parallelMapConfig.h"
-#include "MainBuildConfig.h"
-#include <solvers/tnlBuildConfigTags.h>
-#include <operators/hamilton-jacobi/godunov-eikonal/parallelGodunovMap.h>
-#include <mesh/tnlGrid.h>
-#include <core/tnlDevice.h>
-#include <time.h>
-#include <ctime>
-
-typedef MainBuildConfig BuildConfig;
-
-int main( int argc, char* argv[] )
-{
- time_t start;
- time_t stop;
- time(&start);
- std::clock_t start2= std::clock();
- Config::ParameterContainer parameters;
- tnlConfigDescription configDescription;
- parallelMapConfig< BuildConfig >::configSetup( configDescription );
-
- if( ! parseCommandLine( argc, argv, configDescription, parameters ) )
- return false;
-
-
- tnlDeviceEnum device;
- device = TNL::Devices::HostDevice;
-
- const int& dim = parameters.getParameter< int >( "dim" );
-
- if(dim == 2)
- {
-
- typedef parallelGodunovMapScheme< tnlGrid<2,double,TNL::Devices::Host, int>, double, int > SchemeTypeHost;
-/*#ifdef HAVE_CUDA
- typedef parallelGodunovMapScheme< tnlGrid<2,double,tnlCuda, int>, double, int > SchemeTypeDevice;
-#endif
-#ifndef HAVE_CUDA*/
- typedef parallelGodunovMapScheme< tnlGrid<2,double,TNL::Devices::Host, int>, double, int > SchemeTypeDevice;
-/*#endif*/
-
- if(device==TNL::Devices::HostDevice)
- {
- typedef TNL::Devices::Host Device;
-
-
- tnlParallelMapSolver<2,SchemeTypeHost,SchemeTypeDevice, Device> solver;
- if(!solver.init(parameters))
- {
- std::cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver loop..." <<std::endl;
- solver.run();
- }
- else if(device==tnlCudaDevice )
- {
- typedef tnlCuda Device;
-//typedef parallelGodunovMapScheme< tnlGrid<2,double,Device, int>, double, int > SchemeType;
-
- tnlParallelMapSolver<2,SchemeTypeHost,SchemeTypeDevice, Device> solver;
- if(!solver.init(parameters))
- {
- std::cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver loop..." <<std::endl;
- solver.run();
- }
- }
-
-
- time(&stop);
- cout <<std::endl;
- cout << "Running time was: " << difftime(stop,start) << " .... " << (std::clock() - start2) / (double)(CLOCKS_PER_SEC) <<std::endl;
- return EXIT_SUCCESS;
-}
-
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/mapa_png.png b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/mapa_png.png
deleted file mode 100644
index 668b6fe24b17b2fec486db28505b41e3beb2091a..0000000000000000000000000000000000000000
Binary files a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/mapa_png.png and /dev/null differ
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/no-Makefile b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/no-Makefile
deleted file mode 100644
index bfdc1ef236ca02ecfe6bc88f81d872e9524ec621..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/no-Makefile
+++ /dev/null
@@ -1,41 +0,0 @@
-TNL_VERSION=0.1
-TNL_INSTALL_DIR=${HOME}/local/lib
-TNL_INCLUDE_DIR=${HOME}/local/include/tnl-${TNL_VERSION}
-
-TARGET = hamiltonJacobiParallelSolver
-#CONFIG_FILE = $(TARGET).cfg.desc
-INSTALL_DIR = ${HOME}/local
-CXX = g++
-CUDA_CXX = nvcc
-OMP_FLAGS = -DHAVE_OPENMP -fopenmp
-CXX_FLAGS = -std=gnu++0x -I$(TNL_INCLUDE_DIR) -O3 $(OMP_FLAGS) -DDEBUG
-LD_FLAGS = -L$(TNL_INSTALL_DIR) -ltnl-0.1 -lgomp
-
-SOURCES = main.cpp
-HEADERS =
-OBJECTS = main.o
-DIST = $(SOURCES) Makefile
-
-all: $(TARGET)
-clean:
- rm -f $(OBJECTS)
- rm -f $(TARGET)-conf.h
-
-dist: $(DIST)
- tar zcvf $(TARGET).tgz $(DIST)
-
-install: $(TARGET)
- cp $(TARGET) $(INSTALL_DIR)/bin
- cp $(CONFIG_FILE) $(INSTALL_DIR)/share
-
-uninstall: $(TARGET)
- rm -f $(INSTALL_DIR)/bin/$(TARGET)
- rm -f $(CONFIG_FILE) $(INSTALL_DIR)/share
-
-$(TARGET): $(OBJECTS)
- $(CXX) -o $(TARGET) $(OBJECTS) $(LD_FLAGS)
-
-%.o: %.cpp $(HEADERS)
- $(CXX) -c -o $@ $(CXX_FLAGS) $<
-
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/parallelMapConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/parallelMapConfig.h
deleted file mode 100644
index c07ee95aa04bb8c7a1f3cc376aabd859ff7bc5be..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/parallelMapConfig.h
+++ /dev/null
@@ -1,47 +0,0 @@
-/***************************************************************************
- parallelMapConfig.h - description
- -------------------
- begin : Mar 22 , 2016
- copyright : (C) 2016 by Tomas Sobotik
- email :
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef HAMILTONJACOBIPARALLELMAPPROBLEMCONFIG_H_
-#define HAMILTONJACOBIPARALLELMAPPROBLEMCONFIG_H_
-
-#include <config/tnlConfigDescription.h>
-
-template< typename ConfigTag >
-class parallelMapConfig
-{
- public:
- static void configSetup( tnlConfigDescription& config )
- {
- config.addDelimiter( "Parallel Eikonal solver settings:" );
- config.addEntry < String > ( "problem-name", "This defines particular problem.", "hamilton-jacobi-parallel" );
- config.addEntry < String > ( "scheme", "This defines scheme used for discretization.", "godunov" );
- config.addEntryEnum( "godunov" );
- config.addEntryEnum( "upwind" );
- config.addRequiredEntry < String > ( "initial-condition", "Initial condition for solver");
- config.addRequiredEntry < String > ( "map", "Gradient map for solver");
- config.addEntry < String > ( "mesh", "Name of mesh.", "mesh.tnl" );
- config.addEntry < double > ( "epsilon", "This defines epsilon for smoothening of sign().", 0.0 );
- config.addEntry < double > ( "delta", " Allowed difference on subgrid boundaries", 0.0 );
- config.addRequiredEntry < double > ( "stop-time", " Final time for solver");
- config.addRequiredEntry < double > ( "initial-tau", " initial tau for solver" );
- config.addEntry < double > ( "cfl-condition", " CFL condition", 0.0 );
- config.addEntry < int > ( "subgrid-size", "Subgrid size.", 16 );
- config.addRequiredEntry < int > ( "dim", "Dimension of problem.");
- }
-};
-
-#endif /* HAMILTONJACOBIPARALLELMAPPROBLEMCONFIG_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/run b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/run
deleted file mode 100755
index 48441996274633f8d391d9b32978b05b2e4fa263..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/run
+++ /dev/null
@@ -1,43 +0,0 @@
-#!/bin/bash
-
-dimensions=2
-
-size=2
-
-time=50
-
-rm -r work_dir
-mkdir work_dir
-cp mapa_png.png work_dir/mapa_png.png
-cd work_dir
-
-tnl-image-converter --image-format png\
- --input-images mapa_png.png
-
-
-tnl-init --test-function sdf-para \
- --x-centre 0.5 \
- --y-centre 1.0 \
- --offset 0.05 \
- --output-file init.tnl \
- --final-time 0.0 \
- --snapshot-period 0.1
-
-hamilton-jacobi-parallel-map-dbg --initial-condition init.tnl \
- --map mapa_png.tnl \
- --cfl-condition 50 \
- --mesh mesh.tnl \
- --initial-tau 1.0e-3 \
- --epsilon 4.0 \
- --delta 0.0 \
- --stop-time $time \
- --scheme godunov \
- --subgrid-size 8 \
- --dim $dimensions
-
-
-#cp ../template.dat1 template.dat1
-#cp ../template.dat2 template.dat2
-#cp ../gplt2eps.py gplt2eps.py
-cd ..
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnl-err2eoc-2.py b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnl-err2eoc-2.py
deleted file mode 100755
index f8cde3768e9b76156507e133f8bc3ecaa526fc71..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnl-err2eoc-2.py
+++ /dev/null
@@ -1,141 +0,0 @@
-#!/usr/bin/env python
-
-import sys, string, math
-
-arguments = sys. argv[1:]
-format = "txt"
-output_file_name = "eoc-table.txt"
-input_files = []
-verbose = 1
-size = 1.0
-
-i = 0
-while i < len( arguments ):
- if arguments[ i ] == "--format":
- format = arguments[ i + 1 ]
- i = i + 2
- continue
- if arguments[ i ] == "--output-file":
- output_file_name = arguments[ i + 1 ]
- i = i + 2
- continue
- if arguments[ i ] == "--verbose":
- verbose = float( arguments[ i + 1 ] )
- i = i +2
- continue
- if arguments[ i ] == "--size":
- size = float( arguments[ i + 1 ] )
- i = i +2
- continue
- input_files. append( arguments[ i ] )
- i = i + 1
-
-if not verbose == 0:
- print "Writing to " + output_file_name + " in " + format + "."
-
-h_list = []
-l1_norm_list = []
-l2_norm_list = []
-max_norm_list = []
-items = 0
-
-for file_name in input_files:
- if not verbose == 0:
- print "Processing file " + file_name
- file = open( file_name, "r" )
-
- l1_max = 0.0
- l_max_max = 0.0
- file.readline();
- file.readline();
- for line in file. readlines():
- data = string. split( line )
- h_list. append( size/(float(file_name[0:len(file_name)-5] ) - 1.0) )
- l1_norm_list. append( float( data[ 1 ] ) )
- l2_norm_list. append( float( data[ 2 ] ) )
- max_norm_list. append( float( data[ 3 ] ) )
- items = items + 1
- if not verbose == 0:
- print line
- file. close()
-
-h_width = 12
-err_width = 15
-file = open( output_file_name, "w" )
-if format == "latex":
- file. write( "\\begin{tabular}{|r|l|l|l|l|l|l|}\\hline\n" )
- file. write( "\\raisebox{-1ex}[0ex]{$h$}& \n" )
- file. write( "\\multicolumn{2}{|c|}{\\raisebox{1ex}[3.5ex]{$\\left\| \\cdot \\right\\|_{L_1\\left(\\omega_h;\\left[0,T\\right]\\right)}^{h,\\tau}$}}& \n" )
- file. write( "\\multicolumn{2}{|c|}{\\raisebox{1ex}[3.5ex]{$\\left\| \\cdot \\right\\|_{L_2\\left(\\omega_h;\left[0,T\\right]\\right)}^{h,\\tau}$}}& \n" )
- file. write( "\\multicolumn{2}{|c|}{\\raisebox{1ex}[3.5ex]{$\\left\| \\cdot \\right\\|_{L_\\infty\\left(\\omega_h;\\left[0,T\\right]\\right)}^{h,\\tau}$}}\\\\ \\cline{2-7} \n" )
- file. write( " " + string. rjust( " ", h_width ) + "&" +
- string. rjust( "Error", err_width ) + "&" +
- string. rjust( "{\\bf EOC}", err_width ) + "&" +
- string. rjust( "Error", err_width ) + "&" +
- string. rjust( "{\\bf EOC}", err_width ) + "&" +
- string. rjust( "Error.", err_width ) + "&" +
- string. rjust( "{\\bf EOC}", err_width ) +
- "\\\\ \\hline \\hline \n")
-if format == "txt":
- file. write( "+--------------+----------------+----------------+----------------+----------------+----------------+----------------+\n" )
- file. write( "| h | L1 Err. | L1 EOC. | L2 Err. | L2 EOC | MAX Err. | MAX EOC |\n" )
- file. write( "+==============+================+================+================+================+================+================+\n" )
-
-
-i = 0
-while i < items:
- if i == 0:
- if format == "latex":
- file. write( " " + string. ljust( str( h_list[ i ] ), h_width ) + "&" +
- string. rjust( "%.2g" % l1_norm_list[ i ], err_width ) + "&" +
- string. rjust( " ", err_width ) + "&"+
- string. rjust( "%.2g" % l2_norm_list[ i ], err_width ) + "&" +
- string. rjust( " ", err_width ) + "&" +
- string. rjust( "%.2g" % max_norm_list[ i ], err_width ) + "&" +
- string. rjust( " ", err_width ) + "\\\\\n" )
- if format == "txt":
- file. write( "| " + string. ljust( str( h_list[ i ] ), h_width ) + " |" +
- string. rjust( "%.2g" % l1_norm_list[ i ], err_width ) + " |" +
- string. rjust( " ", err_width ) + " |" +
- string. rjust( "%.2g" % l2_norm_list[ i ], err_width ) + " |" +
- string. rjust( " ", err_width ) + " |" +
- string. rjust( "%.2g" % max_norm_list[ i ], err_width ) + " |" +
- string. rjust( " ", err_width ) + " |\n" )
- file. write( "+--------------+----------------+----------------+----------------+----------------+----------------+----------------+\n" )
- i = i + 1;
- continue
- if h_list[ i ] == h_list[ i - 1 ]:
- print "Unable to count eoc since h[ " + \
- str( i ) + " ] = h[ " + str( i - 1 ) + \
- " ] = " + str( h_list[ i ] ) + ". \n"
- file. write( " eoc error: h[ " + \
- str( i ) + " ] = h[ " + str( i - 1 ) + \
- " ] = " + str( h_list[ i ] ) + ". \n" )
- else:
- h_ratio = math. log( h_list[ i ] / h_list[ i - 1 ] )
- l1_ratio = math. log( l1_norm_list[ i ] / l1_norm_list[ i - 1 ] )
- l2_ratio = math. log( l2_norm_list[ i ] / l2_norm_list[ i - 1 ] )
- max_ratio = math. log( max_norm_list[ i ] / max_norm_list[ i - 1 ] )
- if format == "latex":
- file. write( " " + string. ljust( str( h_list[ i ] ), h_width ) + "&" +
- string. rjust( "%.2g" % l1_norm_list[ i ], err_width ) + "&" +
- string. rjust( "{\\bf " + "%.2g" % ( l1_ratio / h_ratio ) + "}", err_width ) + "&" +
- string. rjust( "%.2g" % l2_norm_list[ i ], err_width ) + "&" +
- string. rjust( "{\\bf " + "%.2g" % ( l2_ratio / h_ratio ) + "}", err_width ) + "&" +
- string. rjust( "%.2g" % max_norm_list[ i ], err_width ) + "&" +
- string. rjust( "{\\bf " + "%.2g" % ( max_ratio / h_ratio ) + "}", err_width ) + "\\\\\n" )
- if format == "txt":
- file. write( "| " + string. ljust( str( h_list[ i ] ), h_width ) + " |" +
- string. rjust( "%.2g" % l1_norm_list[ i ], err_width ) + " |" +
- string. rjust( "**" + "%.2g" % ( l1_ratio / h_ratio ) + "**", err_width ) + " |" +
- string. rjust( "%.2g" % l2_norm_list[ i ], err_width ) + " |" +
- string. rjust( "**" + "%.2g" % ( l2_ratio / h_ratio ) + "**", err_width ) + " |" +
- string. rjust( "%.2g" % max_norm_list[ i ], err_width ) + " |" +
- string. rjust( "**" + "%.2g" % ( max_ratio / h_ratio ) + "**", err_width ) + " |\n" )
- file. write( "+--------------+----------------+----------------+----------------+----------------+----------------+----------------+\n" )
- i = i + 1
-
-if format == "latex":
- file. write( "\\hline \n" )
- file. write( "\\end{tabular} \n" )
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnlParallelMapSolver.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnlParallelMapSolver.h
deleted file mode 100644
index 400e163c9dcc8d536a478a0952aabf8ccbb1a2d8..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnlParallelMapSolver.h
+++ /dev/null
@@ -1,217 +0,0 @@
-/***************************************************************************
- tnlParallelMapSolver.h - description
- -------------------
- begin : Mar 22 , 2016
- copyright : (C) 2016 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef TNLPARALLELMAPSOLVER_H_
-#define TNLPARALLELMAPSOLVER_H_
-
-#include <TNL/Config/ParameterContainer.h>
-#include <TNL/Containers/Vector.h>
-#include <TNL/Containers/StaticVector.h>
-#include <functions/tnlMeshFunction.h>
-#include <TNL/Devices/Host.h>
-#include <mesh/tnlGrid.h>
-#include <mesh/grids/tnlGridEntity.h>
-#include <limits.h>
-#include <core/tnlDevice.h>
-
-
-#include <ctime>
-
-#ifdef HAVE_CUDA
-#include <core/tnlCuda.h>
-#endif
-
-
-template< int Dimension,
- typename SchemeHost,
- typename SchemeDevice,
- typename Device,
- typename RealType = double,
- typename IndexType = int >
-class tnlParallelMapSolver
-{};
-
-template<typename SchemeHost, typename SchemeDevice, typename Device>
-class tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >
-{
-public:
-
- typedef SchemeDevice SchemeTypeDevice;
- typedef SchemeHost SchemeTypeHost;
- typedef Device DeviceType;
- typedef TNL::Containers::Vector< double, TNL::Devices::Host, int > VectorType;
- typedef TNL::Containers::Vector< int, TNL::Devices::Host, int > IntVectorType;
- typedef tnlGrid< 2, double, TNL::Devices::Host, int > MeshType;
-#ifdef HAVE_CUDA
- typedef TNL::Containers::Vector< double, TNL::Devices::Host, int > VectorTypeCUDA;
- typedef TNL::Containers::Vector< int, TNL::Devices::Host, int > IntVectorTypeCUDA;
- typedef tnlGrid< 2, double, TNL::Devices::Host, int > MeshTypeCUDA;
-#endif
- tnlParallelMapSolver();
- bool init( const Config::ParameterContainer& parameters );
- void run();
-
- void test();
-
-/*private:*/
-
-
- void synchronize();
-
- int getOwner( int i) const;
-
- int getSubgridValue( int i ) const;
-
- void setSubgridValue( int i, int value );
-
- int getBoundaryCondition( int i ) const;
-
- void setBoundaryCondition( int i, int value );
-
- void stretchGrid();
-
- void contractGrid();
-
- VectorType getSubgrid( const int i ) const;
-
- void insertSubgrid( VectorType u, const int i );
-
- VectorType runSubgrid( int boundaryCondition, VectorType u, int subGridID,VectorType map);
-
-
- tnlMeshFunction<MeshType> u0;
- VectorType work_u, map_stretched, map;
- IntVectorType subgridValues, boundaryConditions, unusedCell, calculationsCount;
- MeshType mesh, subMesh;
-
-// tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage > Entity;
-
- SchemeHost schemeHost;
- SchemeDevice schemeDevice;
- double delta, tau0, stopTime,cflCondition;
- int gridRows, gridCols, gridLevels, currentStep, n;
-
- std::clock_t start;
- double time_diff;
-
-
- tnlDeviceEnum device;
-
- tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* getSelf()
- {
- return this;
- };
-
-#ifdef HAVE_CUDA
-
- tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver;
-
- double* work_u_cuda;
- double* map_stretched_cuda;
-
- int* subgridValues_cuda;
- int* boundaryConditions_cuda;
- int* unusedCell_cuda;
- int* calculationsCount_cuda;
- double* tmpw;
- double* tmp_map;
-
-
- int* runcuda;
- int run_host;
-
-
- __device__ void getSubgridCUDA2D( const int i, tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* caller, double* a);
-
- __device__ void updateSubgridCUDA2D( const int i, tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* caller, double* a);
-
- __device__ void insertSubgridCUDA2D( double u, const int i );
-
- __device__ void runSubgridCUDA2D( int boundaryCondition, double* u, int subGridID);
-
- __device__ int getOwnerCUDA2D( int i) const;
-
- __device__ int getSubgridValueCUDA2D( int i ) const;
-
- __device__ void setSubgridValueCUDA2D( int i, int value );
-
- __device__ int getBoundaryConditionCUDA2D( int i ) const;
-
- __device__ void setBoundaryConditionCUDA2D( int i, int value );
-
-#endif
-
-};
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-#ifdef HAVE_CUDA
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void runCUDA2D(tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* caller);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void initRunCUDA2D(tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* caller);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void initCUDA2D( tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver, double* ptr, int * ptr2, int* ptr3, double* tmp_map_ptr);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void synchronizeCUDA2D(tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void synchronize2CUDA2D(tnlParallelMapSolver<2, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver);
-
-
-
-__device__
-double fabsMin( double x, double y)
-{
- double fx = abs(x);
-
- if(Min(fx,abs(y)) == fx)
- return x;
- else
- return y;
-}
-
-__device__
-double atomicFabsMin(double* address, double val)
-{
- unsigned long long int* address_as_ull =
- (unsigned long long int*)address;
- unsigned long long int old = *address_as_ull, assumed;
- do {
- assumed = old;
- old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(__longlong_as_double(assumed),val) ));
- } while (assumed != old);
- return __longlong_as_double(old);
-}
-
-#endif
-
-#include "tnlParallelMapSolver2D_impl.h"
-#endif /* TNLPARALLELMAPSOLVER_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnlParallelMapSolver2D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnlParallelMapSolver2D_impl.h
deleted file mode 100644
index e8cbc6fc1619c8a936a0239a5b4e0056361412e3..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel-map/tnlParallelMapSolver2D_impl.h
+++ /dev/null
@@ -1,1315 +0,0 @@
-/***************************************************************************
- tnlParallelMapSolver2D_impl.h - description
- -------------------
- begin : Mar 22 , 2016
- copyright : (C) 2016 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef TNLPARALLELMAPSOLVER2D_IMPL_H_
-#define TNLPARALLELMAPSOLVER2D_IMPL_H_
-
-
-#include "tnlParallelMapSolver.h"
-#include <core/mfilename.h>
-
-
-
-
-#define MAP_SOLVER_MAX_VALUE 3
-
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::tnlParallelMapSolver()
-{
- this->device = TNL::Devices::HostDevice; /////////////// tnlCuda Device --- vypocet na GPU, TNL::Devices::HostDevice --- vypocet na CPU
-
-#ifdef HAVE_CUDA
- if(this->device == tnlCudaDevice)
- {
- run_host = 1;
- }
-#endif
-
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::test()
-{
-/*
- for(int i =0; i < this->subgridValues.getSize(); i++ )
- {
- insertSubgrid(getSubgrid(i), i);
- }
-*/
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-
-bool tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::init( const Config::ParameterContainer& parameters )
-{
- cout << "Initializating solver..." <<std::endl;
- const String& meshLocation = parameters.getParameter <String>("mesh");
- this->mesh.load( meshLocation );
-
- this->n = parameters.getParameter <int>("subgrid-size");
- cout << "Setting N to " << this->n <<std::endl;
-
- this->subMesh.setDimensions( this->n, this->n );
- this->subMesh.setDomain( Containers::StaticVector<2,double>(0.0, 0.0),
- Containers::StaticVector<2,double>(mesh.template getSpaceStepsProducts< 1, 0 >()*(double)(this->n), mesh.template getSpaceStepsProducts< 0, 1 >()*(double)(this->n)) );
-
- this->subMesh.save("submesh.tnl");
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- this->u0.load( initialCondition );
-
- /* LOAD MAP */
- const String& mapFile = parameters.getParameter <String>("map");
- if(! this->map.load( mapFile ))
- cout << "Failed to load map file : " << mapFile <<std::endl;
-
-
- this->delta = parameters.getParameter <double>("delta");
- this->delta *= mesh.template getSpaceStepsProducts< 1, 0 >()*mesh.template getSpaceStepsProducts< 0, 1 >();
-
- cout << "Setting delta to " << this->delta <<std::endl;
-
- this->tau0 = parameters.getParameter <double>("initial-tau");
- cout << "Setting initial tau to " << this->tau0 <<std::endl;
- this->stopTime = parameters.getParameter <double>("stop-time");
-
- this->cflCondition = parameters.getParameter <double>("cfl-condition");
- this -> cflCondition *= sqrt(mesh.template getSpaceStepsProducts< 1, 0 >()*mesh.template getSpaceStepsProducts< 0, 1 >());
- cout << "Setting CFL to " << this->cflCondition <<std::endl;
-
- stretchGrid();
- this->stopTime /= (double)(this->gridCols);
- this->stopTime *= (1.0+1.0/((double)(this->n) - 2.0));
- cout << "Setting stopping time to " << this->stopTime <<std::endl;
-
- cout << "Initializating scheme..." <<std::endl;
- if(!this->schemeHost.init(parameters))
- {
- cerr << "SchemeHost failed to initialize." <<std::endl;
- return false;
- }
- cout << "Scheme initialized." <<std::endl;
-
- test();
-
- VectorType* tmp = new VectorType[subgridValues.getSize()];
- bool containsCurve = false;
-
-#ifdef HAVE_CUDA
-
- if(this->device == tnlCudaDevice)
- {
- cudaMalloc(&(this->cudaSolver), sizeof(tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >));
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >), cudaMemcpyHostToDevice);
-
- double** tmpdev = NULL;
- cudaMalloc(&tmpdev, sizeof(double*));
- cudaMalloc(&(this->tmpw), this->work_u.getSize()*sizeof(double));
- cudaMalloc(&(this->tmp_map), this->map_stretched.getSize()*sizeof(double));
- cudaMalloc(&(this->runcuda), sizeof(int));
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- int* tmpUC;
- cudaMalloc(&(tmpUC), this->work_u.getSize()*sizeof(int));
- cudaMemcpy(tmpUC, this->unusedCell.getData(), this->unusedCell.getSize()*sizeof(int), cudaMemcpyHostToDevice);
-
- initCUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<1,1>>>(this->cudaSolver, (this->tmpw), (this->runcuda),tmpUC, tmp_map);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- double* tmpu = NULL;
- cudaMemcpy(&tmpu, tmpdev,sizeof(double*), cudaMemcpyDeviceToHost);
- cudaMemcpy((this->tmpw), this->work_u.getData(), this->work_u.getSize()*sizeof(double), cudaMemcpyHostToDevice);
- cudaMemcpy((this->tmp_map), this->map_stretched.getData(), this->map_stretched.getSize()*sizeof(double), cudaMemcpyHostToDevice);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- }
-#endif
-
- if(this->device == TNL::Devices::HostDevice)
- {
- VectorType tmp_map;
- tmp_map.setSize(this->n * this->n);
- for(int i = 0; i < this->subgridValues.getSize(); i++)
- {
-
- if(! tmp[i].setSize(this->n * this->n))
- cout << "Could not allocate tmp["<< i <<"] array." <<std::endl;
- tmp[i] = getSubgrid(i);
- containsCurve = false;
-
- for(int j = 0; j < tmp[i].getSize(); j++)
- {
- if(tmp[i][0]*tmp[i][j] <= 0.0)
- {
- containsCurve = true;
- j=tmp[i].getSize();
- }
-
- }
- if(containsCurve)
- {
- for( int j = 0; j < tmp_map.getSize(); j++)
- {
- tmp_map[j] = this->map_stretched[ (i / this->gridCols) * this->n*this->n*this->gridCols
- + (i % this->gridCols) * this->n
- + (j/this->n) * this->n*this->gridCols
- + (j % this->n) ];
- }
- //cout << "Computing initial SDF on subgrid " << i << "." <<std::endl;
- tmp[i] = runSubgrid(0, tmp[i],i,tmp_map);
- insertSubgrid(tmp[i], i);
- setSubgridValue(i, 4);
- //cout << "Computed initial SDF on subgrid " << i << "." <<std::endl;
- }
- containsCurve = false;
-
- }
- }
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- dim3 threadsPerBlock(this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initRunCUDA2D<SchemeTypeHost,SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock,3*this->n*this->n*sizeof(double)>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- }
-#endif
-
-
- this->currentStep = 1;
- if(this->device == TNL::Devices::HostDevice)
- synchronize();
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
- dim3 threadsPerBlock(this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows);
-
- synchronizeCUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- synchronize2CUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- }
-
-#endif
- cout << "Solver initialized." <<std::endl;
-
- return true;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::run()
-{
- if(this->device == TNL::Devices::HostDevice)
- {
- while ((this->boundaryConditions.max() > 0 )/* || !end*/)
- {
-
-#ifdef HAVE_OPENMP
-#pragma omp parallel for num_threads(4) schedule(dynamic)
-#endif
- for(int i = 0; i < this->subgridValues.getSize(); i++)
- {
- if(getSubgridValue(i) != INT_MAX)
- {
- VectorType tmp, tmp_map;
- tmp.setSize(this->n * this->n);
- tmp_map.setSize(this->n * this->n);
- for( int j = 0; j < tmp_map.getSize(); j++)
- {
- tmp_map[j] = this->map_stretched[ (i / this->gridCols) * this->n*this->n*this->gridCols
- + (i % this->gridCols) * this->n
- + (j/this->n) * this->n*this->gridCols
- + (j % this->n) ];
- }
-
- if(getSubgridValue(i) == currentStep+4)
- {
-
- if(getBoundaryCondition(i) & 1)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(1, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 2)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(2, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 4)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(4, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 8)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(8, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- }
- else
- {
-
- if(getBoundaryCondition(i) == 1)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(1, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) == 2)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(2, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) == 4)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(4, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) == 8)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(8, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- }
-
- if(getBoundaryCondition(i) & 3)
- {
- //cout << "3 @ " << getBoundaryCondition(i) <<std::endl;
- tmp = getSubgrid(i);
- tmp = runSubgrid(3, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- }
- if(getBoundaryCondition(i) & 5)
- {
- //cout << "5 @ " << getBoundaryCondition(i) <<std::endl;
- tmp = getSubgrid(i);
- tmp = runSubgrid(5, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- }
- if(getBoundaryCondition(i) & 10)
- {
- //cout << "10 @ " << getBoundaryCondition(i) <<std::endl;
- tmp = getSubgrid(i);
- tmp = runSubgrid(10, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- }
- if(getBoundaryCondition(i) & 12)
- {
- //cout << "12 @ " << getBoundaryCondition(i) <<std::endl;
- tmp = getSubgrid(i);
- tmp = runSubgrid(12, tmp ,i,tmp_map);
- insertSubgrid( tmp, i);
- }
-
-
- setBoundaryCondition(i, 0);
-
- setSubgridValue(i, getSubgridValue(i)-1);
-
- }
- }
- synchronize();
- }
- }
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
- bool end_cuda = false;
- dim3 threadsPerBlock(this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- bool* tmpb;
- cudaMemcpy(&(this->run_host),this->runcuda,sizeof(int), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
-
- int i = 1;
- time_diff = 0.0;
- while (run_host || !end_cuda)
- {
- cout << "Computing at step "<< i++ <<std::endl;
- if(run_host != 0 )
- end_cuda = true;
- else
- end_cuda = false;
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- start = std::clock();
- runCUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock,3*this->n*this->n*sizeof(double)>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- time_diff += (std::clock() - start) / (double)(CLOCKS_PER_SEC);
-
- //start = std::clock();
- synchronizeCUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- synchronize2CUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //time_diff += (std::clock() - start) / (double)(CLOCKS_PER_SEC);
-
- cudaMemcpy(&run_host, (this->runcuda),sizeof(int), cudaMemcpyDeviceToHost);
- }
- cout << "Solving time was: " << time_diff <<std::endl;
-
- cudaMemcpy(this->work_u.getData()/* test*/, (this->tmpw), this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
-
- cudaDeviceSynchronize();
- }
-#endif
- contractGrid();
- this->u0.save("u-00001.tnl");
- cout << "Maximum number of calculations on one subgrid was " << this->calculationsCount.absMax() <<std::endl;
- cout << "Average number of calculations on one subgrid was " << ( (double) this->calculationsCount.sum() / (double) this->calculationsCount.getSize() ) <<std::endl;
- cout << "Solver finished" <<std::endl;
-
-#ifdef HAVE_CUDA
- if(this->device == tnlCudaDevice)
- {
- cudaFree(this->runcuda);
- cudaFree(this->tmpw);
- cudaFree(this->tmp_map);
- cudaFree(this->cudaSolver);
- }
-#endif
-
-}
-
-//north - 1, east - 2, west - 4, south - 8
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::synchronize() //needs fix ---- maybe not anymore --- but frankly: yeah, it does -- aaaa-and maybe fixed now
-{
- cout << "Synchronizig..." <<std::endl;
- int tmp1, tmp2;
- int grid1, grid2;
-
-// if(this->currentStep & 1)
-// {
- for(int j = 0; j < this->gridRows - 1; j++)
- {
- for (int i = 0; i < this->gridCols*this->n; i++)
- {
- tmp1 = this->gridCols*this->n*((this->n-1)+j*this->n) + i;
- tmp2 = this->gridCols*this->n*((this->n)+j*this->n) + i;
- grid1 = getSubgridValue(getOwner(tmp1));
- grid2 = getSubgridValue(getOwner(tmp2));
- if(getOwner(tmp1)==getOwner(tmp2))
- cout << "i, j" << i << "," << j <<std::endl;
- if ((fabs(this->work_u[tmp1]) < fabs(this->work_u[tmp2]) - this->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- this->work_u[tmp2] = this->work_u[tmp1];
- this->unusedCell[tmp2] = 0;
- if(grid2 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp2), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp2)) & 8) )
- setBoundaryCondition(getOwner(tmp2), getBoundaryCondition(getOwner(tmp2))+8);
- }
- else if ((fabs(this->work_u[tmp1]) > fabs(this->work_u[tmp2]) + this->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- this->work_u[tmp1] = this->work_u[tmp2];
- this->unusedCell[tmp1] = 0;
- if(grid1 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp1), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp1)) & 1) )
- setBoundaryCondition(getOwner(tmp1), getBoundaryCondition(getOwner(tmp1))+1);
- }
- }
- }
-
-// }
-// else
-// {
- for(int i = 1; i < this->gridCols; i++)
- {
- for (int j = 0; j < this->gridRows*this->n; j++)
- {
- tmp1 = this->gridCols*this->n*j + i*this->n - 1;
- tmp2 = this->gridCols*this->n*j + i*this->n ;
- grid1 = getSubgridValue(getOwner(tmp1));
- grid2 = getSubgridValue(getOwner(tmp2));
- if(getOwner(tmp1)==getOwner(tmp2))
- cout << "i, j" << i << "," << j <<std::endl;
- if ((fabs(this->work_u[tmp1]) < fabs(this->work_u[tmp2]) - this->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- this->work_u[tmp2] = this->work_u[tmp1];
- this->unusedCell[tmp2] = 0;
- if(grid2 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp2), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp2)) & 4) )
- setBoundaryCondition(getOwner(tmp2), getBoundaryCondition(getOwner(tmp2))+4);
- }
- else if ((fabs(this->work_u[tmp1]) > fabs(this->work_u[tmp2]) + this->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- this->work_u[tmp1] = this->work_u[tmp2];
- this->unusedCell[tmp1] = 0;
- if(grid1 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp1), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp1)) & 2) )
- setBoundaryCondition(getOwner(tmp1), getBoundaryCondition(getOwner(tmp1))+2);
- }
- }
- }
-// }
-
-
- this->currentStep++;
- int stepValue = this->currentStep + 4;
- for (int i = 0; i < this->subgridValues.getSize(); i++)
- {
- if( getSubgridValue(i) == -INT_MAX )
- setSubgridValue(i, stepValue);
- }
-
- cout << "Grid synchronized at step " << (this->currentStep - 1 ) <<std::endl;
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getOwner(int i) const
-{
-
- return (i / (this->gridCols*this->n*this->n))*this->gridCols + (i % (this->gridCols*this->n))/this->n;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getSubgridValue( int i ) const
-{
- return this->subgridValues[i];
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::setSubgridValue(int i, int value)
-{
- this->subgridValues[i] = value;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getBoundaryCondition( int i ) const
-{
- return this->boundaryConditions[i];
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::setBoundaryCondition(int i, int value)
-{
- this->boundaryConditions[i] = value;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::stretchGrid()
-{
- cout << "Stretching grid..." <<std::endl;
-
-
- this->gridCols = ceil( ((double)(this->mesh.getDimensions().x()-1)) / ((double)(this->n-1)) );
- this->gridRows = ceil( ((double)(this->mesh.getDimensions().y()-1)) / ((double)(this->n-1)) );
-
-
- cout << "Setting gridCols to " << this->gridCols << "." <<std::endl;
- cout << "Setting gridRows to " << this->gridRows << "." <<std::endl;
-
- this->subgridValues.setSize(this->gridCols*this->gridRows);
- this->subgridValues.setValue(0);
- this->boundaryConditions.setSize(this->gridCols*this->gridRows);
- this->boundaryConditions.setValue(0);
- this->calculationsCount.setSize(this->gridCols*this->gridRows);
- this->calculationsCount.setValue(0);
-
- for(int i = 0; i < this->subgridValues.getSize(); i++ )
- {
- this->subgridValues[i] = INT_MAX;
- this->boundaryConditions[i] = 0;
- }
-
- int stretchedSize = this->n*this->n*this->gridCols*this->gridRows;
-
- if(!this->work_u.setSize(stretchedSize))
- cerr << "Could not allocate memory for stretched grid." <<std::endl;
- if(!this->map_stretched.setSize(stretchedSize))
- cerr << "Could not allocate memory for stretched map." <<std::endl;
- if(!this->unusedCell.setSize(stretchedSize))
- cerr << "Could not allocate memory for supporting stretched grid." <<std::endl;
- int idealStretch =this->mesh.getDimensions().x() + (this->mesh.getDimensions().x()-2)/(this->n-1);
- cout << idealStretch <<std::endl;
-
- for(int i = 0; i < stretchedSize; i++)
- {
- this->unusedCell[i] = 1;
- int diff =(this->n*this->gridCols) - idealStretch ;
- int k = i/this->n - i/(this->n*this->gridCols) + this->mesh.getDimensions().x()*(i/(this->n*this->n*this->gridCols)) + (i/(this->n*this->gridCols))*diff;
-
- if(i%(this->n*this->gridCols) - idealStretch >= 0)
- {
- k+= i%(this->n*this->gridCols) - idealStretch +1 ;
- }
-
- if(i/(this->n*this->gridCols) - idealStretch + 1 > 0)
- {
- k+= (i/(this->n*this->gridCols) - idealStretch +1 )* this->mesh.getDimensions().x() ;
- }
-
-
- if(fabs(this->u0[i-k]) < mesh.template getSpaceStepsProducts< 1, 0 >()+mesh.template getSpaceStepsProducts< 0, 1 >() )
- this->work_u[i] = this->u0[i-k];
- else
- this->work_u[i] = sign(this->u0[i-k])*MAP_SOLVER_MAX_VALUE;
-
- this->map_stretched[i] = this->map[i-k];
- }
-
-
- cout << "Grid stretched." <<std::endl;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::contractGrid()
-{
- cout << "Contracting grid..." <<std::endl;
- int stretchedSize = this->n*this->n*this->gridCols*this->gridRows;
-
- int idealStretch =this->mesh.getDimensions().x() + (this->mesh.getDimensions().x()-2)/(this->n-1);
- cout << idealStretch <<std::endl;
-
- for(int i = 0; i < stretchedSize; i++)
- {
- int diff =(this->n*this->gridCols) - idealStretch ;
- int k = i/this->n - i/(this->n*this->gridCols) + this->mesh.getDimensions().x()*(i/(this->n*this->n*this->gridCols)) + (i/(this->n*this->gridCols))*diff;
-
- if((i%(this->n*this->gridCols) - idealStretch < 0) && (i/(this->n*this->gridCols) - idealStretch + 1 <= 0))
- {
- this->u0[i-k] = this->work_u[i];
- }
-
- }
-
- cout << "Grid contracted" <<std::endl;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-typename tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::VectorType
-tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getSubgrid( const int i ) const
-{
- VectorType u;
- u.setSize(this->n*this->n);
-
- for( int j = 0; j < u.getSize(); j++)
- {
- u[j] = this->work_u[ (i / this->gridCols) * this->n*this->n*this->gridCols
- + (i % this->gridCols) * this->n
- + (j/this->n) * this->n*this->gridCols
- + (j % this->n) ];
- }
- return u;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::insertSubgrid( VectorType u, const int i )
-{
-
- for( int j = 0; j < this->n*this->n; j++)
- {
- int index = (i / this->gridCols)*this->n*this->n*this->gridCols + (i % this->gridCols)*this->n + (j/this->n)*this->n*this->gridCols + (j % this->n);
- if( (fabs(this->work_u[index]) > fabs(u[j])) || (this->unusedCell[index] == 1) )
- {
- this->work_u[index] = u[j];
- this->unusedCell[index] = 0;
- }
- }
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-typename tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::VectorType
-tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::runSubgrid( int boundaryCondition, VectorType u, int subGridID,VectorType map)
-{
-
- VectorType fu;
-
- fu.setLike(u);
- fu.setValue( 0.0 );
-
-
-
- bool tmp = false;
- for(int i = 0; i < u.getSize(); i++)
- {
- if(u[0]*u[i] <= 0.0)
- tmp=true;
- int centerGID = (this->n*(subGridID / this->gridRows)+ (this->n >> 1))*(this->n*this->gridCols) + this->n*(subGridID % this->gridRows) + (this->n >> 1);
- if(this->unusedCell[centerGID] == 0 || boundaryCondition == 0)
- tmp = true;
- }
-
-
- double value = sign(u[0]) * u.absMax();
-
- if(tmp)
- {}
-
-
- //north - 1, east - 2, west - 4, south - 8
- else if(boundaryCondition == 4)
- {
- for(int i = 0; i < this->n; i++)
- for(int j = 1;j < this->n; j++)
- //if(fabs(u[i*this->n + j]) < fabs(u[i*this->n]))
- u[i*this->n + j] = value;// u[i*this->n];
- }
- else if(boundaryCondition == 2)
- {
- for(int i = 0; i < this->n; i++)
- for(int j =0 ;j < this->n -1; j++)
- //if(fabs(u[i*this->n + j]) < fabs(u[(i+1)*this->n - 1]))
- u[i*this->n + j] = value;// u[(i+1)*this->n - 1];
- }
- else if(boundaryCondition == 1)
- {
- for(int j = 0; j < this->n; j++)
- for(int i = 0;i < this->n - 1; i++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j + this->n*(this->n - 1)]))
- u[i*this->n + j] = value;// u[j + this->n*(this->n - 1)];
- }
- else if(boundaryCondition == 8)
- {
- for(int j = 0; j < this->n; j++)
- for(int i = 1;i < this->n; i++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j]))
- u[i*this->n + j] = value;// u[j];
- }
-
-
-
- double time = 0.0;
- double currentTau = this->tau0;
- double finalTime = this->stopTime;// + 3.0*(u.max() - u.min());
- if( time + currentTau > finalTime ) currentTau = finalTime - time;
-
- double maxResidue( 1.0 );
- tnlGridEntity<MeshType, 2, tnlGridEntityNoStencilStorage > Entity(subMesh);
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
-
- for( int i = 0; i < u.getSize(); i ++ )
- {
- if(map[i] == 0.0)
- {
- u[i] = /*sign(u[l])**/MAP_SOLVER_MAX_VALUE;
- }
- }
-
- while( time < finalTime )
- {
- /****
- * Compute the RHS
- */
-
- for( int i = 0; i < fu.getSize(); i ++ )
- {
- Entity.setCoordinates(Containers::StaticVector<2,int>(i % subMesh.getDimensions().x(),i / subMesh.getDimensions().x()));
- Entity.refresh();
- neighborEntities.refresh(subMesh,Entity.getIndex());
- if(map[i] != 0.0)
- fu[ i ] = schemeHost.getValue( this->subMesh, i, Containers::StaticVector<2,int>(i % subMesh.getDimensions().x(),i / subMesh.getDimensions().x()), u, time, boundaryCondition,neighborEntities,map);
- }
- maxResidue = fu. absMax();
-
-
- if(maxResidue != 0.0)
- currentTau = fabs(this -> cflCondition / maxResidue);
-
-
- if(currentTau > 1.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >())
- {
- currentTau = 1.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >();
- }
-
-
- if( time + currentTau > finalTime ) currentTau = finalTime - time;
-
-
-
- for( int i = 0; i < fu.getSize(); i ++ )
- {
- if(map[i] != 0.0)
- u[ i ] += currentTau * fu[ i ];
- }
- time += currentTau;
-
- }
- return u;
-}
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getSubgridCUDA2D( const int i ,tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >* caller, double* a)
-{
- int th = (blockIdx.y) * caller->n*caller->n*caller->gridCols
- + (blockIdx.x) * caller->n
- + threadIdx.y * caller->n*caller->gridCols
- + threadIdx.x;
-
- *a = caller->work_u_cuda[th];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::updateSubgridCUDA2D( const int i ,tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >* caller, double* a)
-{
- int index = (blockIdx.y) * caller->n*caller->n*caller->gridCols
- + (blockIdx.x) * caller->n
- + threadIdx.y * caller->n*caller->gridCols
- + threadIdx.x;
-
- if( (fabs(caller->work_u_cuda[index]) > fabs(*a)) || (caller->unusedCell_cuda[index] == 1) )
- {
- caller->work_u_cuda[index] = *a;
- caller->unusedCell_cuda[index] = 0;
-
- }
-
- *a = caller->work_u_cuda[index];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::insertSubgridCUDA2D( double u, const int i )
-{
- int index = (blockIdx.y)*this->n*this->n*this->gridCols
- + (blockIdx.x)*this->n
- + threadIdx.y*this->n*this->gridCols
- + threadIdx.x;
-
- if( (fabs(this->work_u_cuda[index]) > fabs(u)) || (this->unusedCell_cuda[index] == 1) )
- {
- this->work_u_cuda[index] = u;
- this->unusedCell_cuda[index] = 0;
-
- }
-
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::runSubgridCUDA2D( int boundaryCondition, double* u, int subGridID)
-{
-
- __shared__ int tmp;
- __shared__ double value;
- volatile double* sharedTau = &u[blockDim.x*blockDim.y];
- double* map_local = &u[2*blockDim.x*blockDim.y];
-
- int i = threadIdx.x;
- int j = threadIdx.y;
- int l = threadIdx.y * blockDim.x + threadIdx.x;
- int gid = (blockDim.y*blockIdx.y + threadIdx.y)*blockDim.x*gridDim.x + blockDim.x*blockIdx.x + threadIdx.x;
-
- /* LOAD MAP */
- map_local[l]=this->map_stretched_cuda[gid];
- if(map_local[l] != 0.0)
- map_local[l] = 1.0/map_local[l];
- /* LOADED */
-
- bool computeFU = !((i == 0 && (boundaryCondition & 4)) or
- (i == blockDim.x - 1 && (boundaryCondition & 2)) or
- (j == 0 && (boundaryCondition & 8)) or
- (j == blockDim.y - 1 && (boundaryCondition & 1)));
-
- if(l == 0)
- {
- tmp = 0;
- int centerGID = (blockDim.y*blockIdx.y + (blockDim.y>>1))*(blockDim.x*gridDim.x) + blockDim.x*blockIdx.x + (blockDim.x>>1);
- if(this->unusedCell_cuda[centerGID] == 0 || boundaryCondition == 0)
- tmp = 1;
- }
- __syncthreads();
-
-
- if(tmp !=1)
- {
- if(computeFU)
- {
- if(boundaryCondition == 4)
- u[l] = u[threadIdx.y * blockDim.x] ;//+ sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(threadIdx.x);
- else if(boundaryCondition == 2)
- u[l] = u[threadIdx.y * blockDim.x + blockDim.x - 1] ;//+ sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(this->n - 1 - threadIdx.x);
- else if(boundaryCondition == 8)
- u[l] = u[threadIdx.x] ;//+ sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(threadIdx.y);
- else if(boundaryCondition == 1)
- u[l] = u[(blockDim.y - 1)* blockDim.x + threadIdx.x] ;//+ sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(this->n - 1 - threadIdx.y);
- }
- }
-
- double time = 0.0;
- __shared__ double currentTau;
- double cfl = this->cflCondition;
- double fu = 0.0;
-
- double finalTime = this->stopTime;
- if(boundaryCondition == 0)
- finalTime*=2.0;
- __syncthreads();
-
- tnlGridEntity<MeshType, 2, tnlGridEntityNoStencilStorage > Entity(subMesh);
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Entity.setCoordinates(Containers::StaticVector<2,int>(i,j));
- Entity.refresh();
- neighborEntities.refresh(subMesh,Entity.getIndex());
-
-
- if(map_local[l] == 0.0)
- {
- u[l] = /*sign(u[l])**/MAP_SOLVER_MAX_VALUE;
- computeFU = false;
- }
- __syncthreads();
-
-
- while( time < finalTime )
- {
- sharedTau[l] = finalTime;
-
- if(computeFU)
- {
- fu = schemeHost.getValueDev( this->subMesh, l, Containers::StaticVector<2,int>(i,j), u, time, boundaryCondition, neighborEntities, map_local);
- sharedTau[l]=abs(cfl/fu);
- }
-
-
-
- if(l == 0)
- {
- if(sharedTau[0] > 1.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >()) sharedTau[0] = 1.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >();
- }
- else if(l == blockDim.x*blockDim.y - 1)
- if( time + sharedTau[l] > finalTime ) sharedTau[l] = finalTime - time;
-
-
- if((blockDim.x == 16) && (l < 128)) sharedTau[l] = Min(sharedTau[l],sharedTau[l+128]);
- __syncthreads();
- if((blockDim.x == 16) && (l < 64)) sharedTau[l] = Min(sharedTau[l],sharedTau[l+64]);
- __syncthreads();
- if(l < 32) sharedTau[l] = Min(sharedTau[l],sharedTau[l+32]);
- if(l < 16) sharedTau[l] = Min(sharedTau[l],sharedTau[l+16]);
- if(l < 8) sharedTau[l] = Min(sharedTau[l],sharedTau[l+8]);
- if(l < 4) sharedTau[l] = Min(sharedTau[l],sharedTau[l+4]);
- if(l < 2) sharedTau[l] = Min(sharedTau[l],sharedTau[l+2]);
- if(l < 1) currentTau = Min(sharedTau[l],sharedTau[l+1]);
- __syncthreads();
-
- u[l] += currentTau * fu;
- time += currentTau;
- }
-
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getOwnerCUDA2D(int i) const
-{
-
- return ((i / (this->gridCols*this->n*this->n))*this->gridCols
- + (i % (this->gridCols*this->n))/this->n);
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getSubgridValueCUDA2D( int i ) const
-{
- return this->subgridValues_cuda[i];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::setSubgridValueCUDA2D(int i, int value)
-{
- this->subgridValues_cuda[i] = value;
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getBoundaryConditionCUDA2D( int i ) const
-{
- return this->boundaryConditions_cuda[i];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int>::setBoundaryConditionCUDA2D(int i, int value)
-{
- this->boundaryConditions_cuda[i] = value;
-}
-
-
-
-//north - 1, east - 2, west - 4, south - 8
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void synchronizeCUDA2D(tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver) //needs fix ---- maybe not anymore --- but frankly: yeah, it does -- aaaa-and maybe fixed now
-{
-
- __shared__ int boundary[4]; // north,east,west,south
- __shared__ int subgridValue;
- __shared__ int newSubgridValue;
-
-
- int gid = (blockDim.y*blockIdx.y + threadIdx.y)*blockDim.x*gridDim.x + blockDim.x*blockIdx.x + threadIdx.x;
- double u = cudaSolver->work_u_cuda[gid];
- double u_cmp;
- int subgridValue_cmp=INT_MAX;
- int boundary_index=0;
-
-
- if(threadIdx.x+threadIdx.y == 0)
- {
- subgridValue = cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x);
- boundary[0] = 0;
- boundary[1] = 0;
- boundary[2] = 0;
- boundary[3] = 0;
- newSubgridValue = 0;
- }
- __syncthreads();
-
-
-
- if( (threadIdx.x == 0 /* && !(cudaSolver->currentStep & 1)*/) ||
- (threadIdx.y == 0 /* && (cudaSolver->currentStep & 1)*/) ||
- (threadIdx.x == blockDim.x - 1 /* && !(cudaSolver->currentStep & 1)*/) ||
- (threadIdx.y == blockDim.y - 1 /* && (cudaSolver->currentStep & 1)*/) )
- {
- if(threadIdx.x == 0 && (blockIdx.x != 0)/* && !(cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid - 1];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x - 1);
- boundary_index = 2;
- }
-
- if(threadIdx.x == blockDim.x - 1 && (blockIdx.x != gridDim.x - 1)/* && !(cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid + 1];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x + 1);
- boundary_index = 1;
- }
-
- __threadfence();
- if((subgridValue == INT_MAX || fabs(u_cmp) + cudaSolver->delta < fabs(u) ) && (subgridValue_cmp != INT_MAX && subgridValue_cmp != -INT_MAX))
- {
- cudaSolver->unusedCell_cuda[gid] = 0;
- atomicMax(&newSubgridValue, INT_MAX);
- atomicMax(&boundary[boundary_index], 1);
- cudaSolver->work_u_cuda[gid] = u_cmp;
- u=u_cmp;
- }
- __threadfence();
- if(threadIdx.y == 0 && (blockIdx.y != 0)/* && (cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid - blockDim.x*gridDim.x];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA2D((blockIdx.y - 1)*gridDim.x + blockIdx.x);
- boundary_index = 3;
- }
- if(threadIdx.y == blockDim.y - 1 && (blockIdx.y != gridDim.y - 1)/* && (cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid + blockDim.x*gridDim.x];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA2D((blockIdx.y + 1)*gridDim.x + blockIdx.x);
- boundary_index = 0;
- }
-
- if((subgridValue == INT_MAX || fabs(u_cmp) + cudaSolver->delta < fabs(u) ) && (subgridValue_cmp != INT_MAX && subgridValue_cmp != -INT_MAX))
- {
- cudaSolver->unusedCell_cuda[gid] = 0;
- atomicMax(&newSubgridValue, INT_MAX);
- atomicMax(&boundary[boundary_index], 1);
- cudaSolver->work_u_cuda[gid] = u_cmp;
- }
- }
- __threadfence();
- __syncthreads();
-
- if(threadIdx.x+threadIdx.y == 0)
- {
- if(subgridValue == INT_MAX && newSubgridValue !=0)
- cudaSolver->setSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x, -INT_MAX);
-
- cudaSolver->setBoundaryConditionCUDA2D(blockIdx.y*gridDim.x + blockIdx.x, boundary[0] +
- 2 * boundary[1] +
- 4 * boundary[2] +
- 8 * boundary[3]);
-
-
- if(blockIdx.x+blockIdx.y ==0)
- {
- cudaSolver->currentStep += 1;
- *(cudaSolver->runcuda) = 0;
- }
- }
-
-}
-
-
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void synchronize2CUDA2D(tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver)
-{
-
-
- int stepValue = cudaSolver->currentStep + 4;
- if( cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x) == -INT_MAX )
- cudaSolver->setSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x, stepValue);
-
- atomicMax((cudaSolver->runcuda),cudaSolver->getBoundaryConditionCUDA2D(blockIdx.y*gridDim.x + blockIdx.x));
-}
-
-
-
-
-
-
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void initCUDA2D( tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver, double* ptr , int* ptr2, int* ptr3, double* tmp_map_ptr)
-{
-
-
- cudaSolver->work_u_cuda = ptr;
- cudaSolver->map_stretched_cuda = tmp_map_ptr;
- cudaSolver->unusedCell_cuda = ptr3;
- cudaSolver->subgridValues_cuda =(int*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*sizeof(int));
- cudaSolver->boundaryConditions_cuda =(int*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*sizeof(int));
- cudaSolver->runcuda = ptr2;
- *(cudaSolver->runcuda) = 1;
-
-/* CHANGED !!!!!! from 1 to 0*/ cudaSolver->currentStep = 0;
-
- printf("GPU memory allocated.\n");
-
- for(int i = 0; i < cudaSolver->gridCols*cudaSolver->gridRows; i++)
- {
- cudaSolver->subgridValues_cuda[i] = INT_MAX;
- cudaSolver->boundaryConditions_cuda[i] = 0;
- }
-
- printf("GPU memory initialized.\n");
-}
-
-
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device >
-__global__
-void initRunCUDA2D(tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >* caller)
-
-{
- extern __shared__ double u[];
-
- int i = blockIdx.y * gridDim.x + blockIdx.x;
- int l = threadIdx.y * blockDim.x + threadIdx.x;
-
- __shared__ int containsCurve;
- if(l == 0)
- containsCurve = 0;
-
-
- caller->getSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
-
- if(u[0] * u[l] <= 0.0)
- atomicMax( &containsCurve, 1);
-
- __syncthreads();
- if(containsCurve == 1)
- {
- caller->runSubgridCUDA2D(0,u,i);
- caller->insertSubgridCUDA2D(u[l],i);
- __syncthreads();
- if(l == 0)
- caller->setSubgridValueCUDA2D(i, 4);
- }
-
-
-}
-
-
-
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device >
-__global__
-void runCUDA2D(tnlParallelMapSolver<2,SchemeHost, SchemeDevice, Device, double, int >* caller)
-{
- extern __shared__ double u[];
- int i = blockIdx.y * gridDim.x + blockIdx.x;
- int l = threadIdx.y * blockDim.x + threadIdx.x;
- int bound = caller->getBoundaryConditionCUDA2D(i);
-
- if(caller->getSubgridValueCUDA2D(i) != INT_MAX && bound != 0 && caller->getSubgridValueCUDA2D(i) > 0)
- {
- caller->getSubgridCUDA2D(i,caller, &u[l]);
-
-
- if(caller->getSubgridValueCUDA2D(i) == caller->currentStep+4)
- {
- if(bound & 1)
- {
- caller->runSubgridCUDA2D(1,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 2)
- {
- caller->runSubgridCUDA2D(2,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 4)
- {
- caller->runSubgridCUDA2D(4,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 8)
- {
- caller->runSubgridCUDA2D(8,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- }
- else
- {
-
- if(bound == 1)
- {
- caller->runSubgridCUDA2D(1,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound == 2)
- {
- caller->runSubgridCUDA2D(2,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound == 4)
- {
- caller->runSubgridCUDA2D(4,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound == 8)
- {
- caller->runSubgridCUDA2D(8,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- }
-
- if(bound & 3)
- {
- caller->runSubgridCUDA2D(3,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 5)
- {
- caller->runSubgridCUDA2D(5,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 10)
- {
- caller->runSubgridCUDA2D(10,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 12)
- {
- caller->runSubgridCUDA2D(12,u,i);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
-
-
- if(l==0)
- {
- caller->setBoundaryConditionCUDA2D(i, 0);
- caller->setSubgridValueCUDA2D(i, caller->getSubgridValueCUDA2D(i) - 1 );
- }
-
-
- }
-
-
-
-}
-
-#endif /*HAVE_CUDA*/
-
-#endif /* TNLPARALLELMAPSOLVER2D_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/CMakeLists.txt b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/CMakeLists.txt
deleted file mode 100644
index f6a00127c7f79344a0c9303c1a0f4b2a8ad84832..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/CMakeLists.txt
+++ /dev/null
@@ -1,23 +0,0 @@
-set( tnl_hamilton_jacobi_parallel_SOURCES
-# MainBuildConfig.h
-# tnlParallelEikonalSolver2D_impl.h
-# tnlParallelEikonalSolver3D_impl.h
-# tnlParallelEikonalSolver.h
-# parallelEikonalConfig.h
- main.cpp)
-
-
-IF( BUILD_CUDA )
- CUDA_ADD_EXECUTABLE(hamilton-jacobi-parallel main.cu)
-ELSE( BUILD_CUDA )
- ADD_EXECUTABLE(hamilton-jacobi-parallel main.cpp)
-ENDIF( BUILD_CUDA )
-target_link_libraries (hamilton-jacobi-parallel tnl )
-
-
-INSTALL( TARGETS hamilton-jacobi-parallel
- RUNTIME DESTINATION bin
- PERMISSIONS OWNER_READ OWNER_WRITE OWNER_EXECUTE GROUP_READ GROUP_EXECUTE WORLD_READ WORLD_EXECUTE )
-
-#INSTALL( FILES ${tnl_hamilton_jacobi_parallel_SOURCES}
-# DESTINATION ${TNL_TARGET_DATA_DIRECTORY}/examples/hamilton-jacobi-parallel )
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/MainBuildConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/MainBuildConfig.h
deleted file mode 100644
index ed3d686eb99379af1589d734eac9b5812cccdedf..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/MainBuildConfig.h
+++ /dev/null
@@ -1,64 +0,0 @@
-/***************************************************************************
- MainBuildConfig.h - description
- -------------------
- begin : Jul 7, 2014
- copyright : (C) 2014 by Tomas Oberhuber
- email : tomas.oberhuber@fjfi.cvut.cz
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef MAINBUILDCONFIG_H_
-#define MAINBUILDCONFIG_H_
-
-#include <solvers/tnlBuildConfigTags.h>
-
-class MainBuildConfig
-{
- public:
-
- static void print() {std::cerr << "MainBuildConfig" <<std::endl; }
-};
-
-/****
- * Turn off support for float and long double.
- */
-template<> struct tnlConfigTagReal< MainBuildConfig, float > { enum { enabled = false }; };
-template<> struct tnlConfigTagReal< MainBuildConfig, long double > { enum { enabled = false }; };
-
-/****
- * Turn off support for short int and long int indexing.
- */
-template<> struct tnlConfigTagIndex< MainBuildConfig, short int >{ enum { enabled = false }; };
-template<> struct tnlConfigTagIndex< MainBuildConfig, long int >{ enum { enabled = false }; };
-
-/****
- * Use of tnlGrid is enabled for allowed dimensions and Real, Device and Index types.
- */
-template< int Dimensions, typename Real, typename Device, typename Index >
- struct tnlConfigTagMesh< MainBuildConfig, tnlGrid< Dimensions, Real, Device, Index > >
- { enum { enabled = tnlConfigTagDimensions< MainBuildConfig, Dimensions >::enabled &&
- tnlConfigTagReal< MainBuildConfig, Real >::enabled &&
- tnlConfigTagDevice< MainBuildConfig, Device >::enabled &&
- tnlConfigTagIndex< MainBuildConfig, Index >::enabled }; };
-
-/****
- * Please, chose your preferred time discretisation here.
- */
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlExplicitTimeDiscretisationTag >{ enum { enabled = true }; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlSemiImplicitTimeDiscretisationTag >{ enum { enabled = false}; };
-template<> struct tnlConfigTagTimeDiscretisation< MainBuildConfig, tnlImplicitTimeDiscretisationTag >{ enum { enabled = false }; };
-
-/****
- * Only the Runge-Kutta-Merson solver is enabled by default.
- */
-template<> struct tnlConfigTagExplicitSolver< MainBuildConfig, tnlExplicitEulerSolverTag >{ enum { enabled = false }; };
-
-#endif /* MAINBUILDCONFIG_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.cpp b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.cpp
deleted file mode 100644
index b13498e17330fae7bb00a0bdc2abcc7a19f8e7a8..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.cpp
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cpp - description
- -------------------
- begin : Jul 8 , 2014
- copyright : (C) 2014 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.cu b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.cu
deleted file mode 100644
index 7101976712e153d73c5f0979b211164a36ec648d..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.cu
+++ /dev/null
@@ -1,17 +0,0 @@
-/***************************************************************************
- main.cu - description
- -------------------
- begin : Mar 30 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "main.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.h
deleted file mode 100644
index dbaebdcebd3b2bdf0509eda61729c1b11579716a..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/main.h
+++ /dev/null
@@ -1,142 +0,0 @@
-/***************************************************************************
- main.h - description
- -------------------
- begin : Mar 30 , 2015
- copyright : (C) 2015 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#include "tnlParallelEikonalSolver.h"
-#include "parallelEikonalConfig.h"
-#include "MainBuildConfig.h"
-#include <solvers/tnlBuildConfigTags.h>
-#include <operators/hamilton-jacobi/godunov-eikonal/parallelGodunovEikonal.h>
-#include <mesh/tnlGrid.h>
-#include <core/tnlDevice.h>
-#include <time.h>
-#include <ctime>
-
-typedef MainBuildConfig BuildConfig;
-
-int main( int argc, char* argv[] )
-{
- time_t start;
- time_t stop;
- time(&start);
- std::clock_t start2= std::clock();
- Config::ParameterContainer parameters;
- tnlConfigDescription configDescription;
- parallelEikonalConfig< BuildConfig >::configSetup( configDescription );
-
- if( ! parseCommandLine( argc, argv, configDescription, parameters ) )
- return false;
-
- //if (parameters.GetParameter <String>("scheme") == "godunov")
- //{
- tnlDeviceEnum device;
- device = TNL::Devices::HostDevice;
-
- const int& dim = parameters.getParameter< int >( "dim" );
-
- if(dim == 2)
- {
-
- typedef parallelGodunovEikonalScheme< tnlGrid<2,double,TNL::Devices::Host, int>, double, int > SchemeTypeHost;
- /*#ifdef HAVE_CUDA
- typedef parallelGodunovEikonalScheme< tnlGrid<2,double,tnlCuda, int>, double, int > SchemeTypeDevice;
- #endif
- #ifndef HAVE_CUDA*/
- typedef parallelGodunovEikonalScheme< tnlGrid<2,double,TNL::Devices::Host, int>, double, int > SchemeTypeDevice;
- /*#endif*/
-
- if(device==TNL::Devices::HostDevice)
- {
- typedef TNL::Devices::Host Device;
-
-
- tnlParallelEikonalSolver<2,SchemeTypeHost,SchemeTypeDevice, Device> solver;
- if(!solver.init(parameters))
- {
- std::cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver loop..." <<std::endl;
- solver.run();
- }
- else if(device==tnlCudaDevice )
- {
- typedef tnlCuda Device;
- //typedef parallelGodunovEikonalScheme< tnlGrid<2,double,Device, int>, double, int > SchemeType;
-
- tnlParallelEikonalSolver<2,SchemeTypeHost,SchemeTypeDevice, Device> solver;
- if(!solver.init(parameters))
- {
- std::cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver loop..." <<std::endl;
- solver.run();
- }
- // }
- }
- else if(dim == 3)
- {
-
- typedef parallelGodunovEikonalScheme< tnlGrid<3,double,TNL::Devices::Host, int>, double, int > SchemeTypeHost;
- /*#ifdef HAVE_CUDA
- typedef parallelGodunovEikonalScheme< tnlGrid<2,double,tnlCuda, int>, double, int > SchemeTypeDevice;
- #endif
- #ifndef HAVE_CUDA*/
- typedef parallelGodunovEikonalScheme< tnlGrid<3,double,TNL::Devices::Host, int>, double, int > SchemeTypeDevice;
- /*#endif*/
-
- if(device==TNL::Devices::HostDevice)
- {
- typedef TNL::Devices::Host Device;
-
-
- tnlParallelEikonalSolver<3,SchemeTypeHost,SchemeTypeDevice, Device> solver;
- if(!solver.init(parameters))
- {
- std::cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver loop..." <<std::endl;
- solver.run();
- }
- else if(device==tnlCudaDevice )
- {
- typedef tnlCuda Device;
- //typedef parallelGodunovEikonalScheme< tnlGrid<2,double,Device, int>, double, int > SchemeType;
-
- tnlParallelEikonalSolver<3,SchemeTypeHost,SchemeTypeDevice, Device> solver;
- if(!solver.init(parameters))
- {
- std::cerr << "Solver failed to initialize." <<std::endl;
- return EXIT_FAILURE;
- }
- std::cout << "-------------------------------------------------------------" <<std::endl;
- std::cout << "Starting solver loop..." <<std::endl;
- solver.run();
- }
- // }
- }
-
- time(&stop);
- std::cout <<std::endl;
- std::cout << "Running time was: " << difftime(stop,start) << " .... " << (std::clock() - start2) / (double)(CLOCKS_PER_SEC) <<std::endl;
- return EXIT_SUCCESS;
-}
-
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/no-Makefile b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/no-Makefile
deleted file mode 100644
index bfdc1ef236ca02ecfe6bc88f81d872e9524ec621..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/no-Makefile
+++ /dev/null
@@ -1,41 +0,0 @@
-TNL_VERSION=0.1
-TNL_INSTALL_DIR=${HOME}/local/lib
-TNL_INCLUDE_DIR=${HOME}/local/include/tnl-${TNL_VERSION}
-
-TARGET = hamiltonJacobiParallelSolver
-#CONFIG_FILE = $(TARGET).cfg.desc
-INSTALL_DIR = ${HOME}/local
-CXX = g++
-CUDA_CXX = nvcc
-OMP_FLAGS = -DHAVE_OPENMP -fopenmp
-CXX_FLAGS = -std=gnu++0x -I$(TNL_INCLUDE_DIR) -O3 $(OMP_FLAGS) -DDEBUG
-LD_FLAGS = -L$(TNL_INSTALL_DIR) -ltnl-0.1 -lgomp
-
-SOURCES = main.cpp
-HEADERS =
-OBJECTS = main.o
-DIST = $(SOURCES) Makefile
-
-all: $(TARGET)
-clean:
- rm -f $(OBJECTS)
- rm -f $(TARGET)-conf.h
-
-dist: $(DIST)
- tar zcvf $(TARGET).tgz $(DIST)
-
-install: $(TARGET)
- cp $(TARGET) $(INSTALL_DIR)/bin
- cp $(CONFIG_FILE) $(INSTALL_DIR)/share
-
-uninstall: $(TARGET)
- rm -f $(INSTALL_DIR)/bin/$(TARGET)
- rm -f $(CONFIG_FILE) $(INSTALL_DIR)/share
-
-$(TARGET): $(OBJECTS)
- $(CXX) -o $(TARGET) $(OBJECTS) $(LD_FLAGS)
-
-%.o: %.cpp $(HEADERS)
- $(CXX) -c -o $@ $(CXX_FLAGS) $<
-
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/parallelEikonalConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/parallelEikonalConfig.h
deleted file mode 100644
index c27f5ebb39e5c4db31ed13d1a8e80b8ca8915d51..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/parallelEikonalConfig.h
+++ /dev/null
@@ -1,46 +0,0 @@
-/***************************************************************************
- parallelEikonalConfig.h - description
- -------------------
- begin : Oct 5, 2014
- copyright : (C) 2014 by Tomas Sobotik
- email :
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef HAMILTONJACOBIPARALLELEIKONALPROBLEMCONFIG_H_
-#define HAMILTONJACOBIPARALLELEIKONALPROBLEMCONFIG_H_
-
-#include <config/tnlConfigDescription.h>
-
-template< typename ConfigTag >
-class parallelEikonalConfig
-{
- public:
- static void configSetup( tnlConfigDescription& config )
- {
- config.addDelimiter( "Parallel Eikonal solver settings:" );
- config.addEntry < String > ( "problem-name", "This defines particular problem.", "hamilton-jacobi-parallel" );
- config.addEntry < String > ( "scheme", "This defines scheme used for discretization.", "godunov" );
- config.addEntryEnum( "godunov" );
- config.addEntryEnum( "upwind" );
- config.addRequiredEntry < String > ( "initial-condition", "Initial condition for solver");
- config.addEntry < String > ( "mesh", "Name of mesh.", "mesh.tnl" );
- config.addEntry < double > ( "epsilon", "This defines epsilon for smoothening of sign().", 0.0 );
- config.addEntry < double > ( "delta", " Allowed difference on subgrid boundaries", 0.0 );
- config.addRequiredEntry < double > ( "stop-time", " Final time for solver");
- config.addRequiredEntry < double > ( "initial-tau", " initial tau for solver" );
- config.addEntry < double > ( "cfl-condition", " CFL condition", 0.0 );
- config.addEntry < int > ( "subgrid-size", "Subgrid size.", 16 );
- config.addRequiredEntry < int > ( "dim", "Dimension of problem.");
- }
-};
-
-#endif /* HAMILTONJACOBIPARALLELEIKONALPROBLEMCONFIG_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/run b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/run
deleted file mode 100755
index 3aece294a9c1189cd885acbe459dba20be713716..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/run
+++ /dev/null
@@ -1,64 +0,0 @@
-#!/bin/bash
-
-#GRID_SIZES="0897"
-GRID_SIZES="0008 0015 0029 0057 0113 0225 0449"
-#GRID_SIZES="1793"
-
-dimensions=2
-
-size=2
-
-time=3
-
-for grid_size in $GRID_SIZES;
-
-do
-
- rm -r grid-${grid_size}
- mkdir grid-${grid_size}
- cd grid-${grid_size}
-
- tnl-grid-setup --dimensions $dimensions \
- --origin-x -1.0 \
- --origin-y -1.0 \
- --origin-z -1.0 \
- --proportions-x $size \
- --proportions-y $size \
- --proportions-z $size \
- --size-x ${grid_size} \
- --size-y ${grid_size} \
- --size-z ${grid_size}
-
- tnl-init --test-function sdf-para \
- --offset 0.25 \
- --output-file init.tnl \
- --final-time 0.0 \
- --snapshot-period 0.1 \
-
-
- tnl-init --test-function sdf-para-sdf \
- --offset 0.25 \
- --output-file sdf.tnl \
- --final-time 0.0 \
- --snapshot-period 0.1
-
- hamilton-jacobi-parallel --initial-condition init.tnl \
- --cfl-condition 1.0e-1 \
- --mesh mesh.tnl \
- --initial-tau 1.0e-3 \
- --epsilon 1.0 \
- --delta 0.0 \
- --stop-time $time \
- --scheme godunov \
- --subgrid-size 8
-
- tnl-diff --mesh mesh.tnl --mode sequence --input-files sdf.tnl u-00001.tnl --write-difference yes --output-file ../${grid_size}.diff
-
- cd ..
-
-done
-
-
-./tnl-err2eoc-2.py --format txt --size $size *.diff
-
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnl-err2eoc-2.py b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnl-err2eoc-2.py
deleted file mode 100755
index f8cde3768e9b76156507e133f8bc3ecaa526fc71..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnl-err2eoc-2.py
+++ /dev/null
@@ -1,141 +0,0 @@
-#!/usr/bin/env python
-
-import sys, string, math
-
-arguments = sys. argv[1:]
-format = "txt"
-output_file_name = "eoc-table.txt"
-input_files = []
-verbose = 1
-size = 1.0
-
-i = 0
-while i < len( arguments ):
- if arguments[ i ] == "--format":
- format = arguments[ i + 1 ]
- i = i + 2
- continue
- if arguments[ i ] == "--output-file":
- output_file_name = arguments[ i + 1 ]
- i = i + 2
- continue
- if arguments[ i ] == "--verbose":
- verbose = float( arguments[ i + 1 ] )
- i = i +2
- continue
- if arguments[ i ] == "--size":
- size = float( arguments[ i + 1 ] )
- i = i +2
- continue
- input_files. append( arguments[ i ] )
- i = i + 1
-
-if not verbose == 0:
- print "Writing to " + output_file_name + " in " + format + "."
-
-h_list = []
-l1_norm_list = []
-l2_norm_list = []
-max_norm_list = []
-items = 0
-
-for file_name in input_files:
- if not verbose == 0:
- print "Processing file " + file_name
- file = open( file_name, "r" )
-
- l1_max = 0.0
- l_max_max = 0.0
- file.readline();
- file.readline();
- for line in file. readlines():
- data = string. split( line )
- h_list. append( size/(float(file_name[0:len(file_name)-5] ) - 1.0) )
- l1_norm_list. append( float( data[ 1 ] ) )
- l2_norm_list. append( float( data[ 2 ] ) )
- max_norm_list. append( float( data[ 3 ] ) )
- items = items + 1
- if not verbose == 0:
- print line
- file. close()
-
-h_width = 12
-err_width = 15
-file = open( output_file_name, "w" )
-if format == "latex":
- file. write( "\\begin{tabular}{|r|l|l|l|l|l|l|}\\hline\n" )
- file. write( "\\raisebox{-1ex}[0ex]{$h$}& \n" )
- file. write( "\\multicolumn{2}{|c|}{\\raisebox{1ex}[3.5ex]{$\\left\| \\cdot \\right\\|_{L_1\\left(\\omega_h;\\left[0,T\\right]\\right)}^{h,\\tau}$}}& \n" )
- file. write( "\\multicolumn{2}{|c|}{\\raisebox{1ex}[3.5ex]{$\\left\| \\cdot \\right\\|_{L_2\\left(\\omega_h;\left[0,T\\right]\\right)}^{h,\\tau}$}}& \n" )
- file. write( "\\multicolumn{2}{|c|}{\\raisebox{1ex}[3.5ex]{$\\left\| \\cdot \\right\\|_{L_\\infty\\left(\\omega_h;\\left[0,T\\right]\\right)}^{h,\\tau}$}}\\\\ \\cline{2-7} \n" )
- file. write( " " + string. rjust( " ", h_width ) + "&" +
- string. rjust( "Error", err_width ) + "&" +
- string. rjust( "{\\bf EOC}", err_width ) + "&" +
- string. rjust( "Error", err_width ) + "&" +
- string. rjust( "{\\bf EOC}", err_width ) + "&" +
- string. rjust( "Error.", err_width ) + "&" +
- string. rjust( "{\\bf EOC}", err_width ) +
- "\\\\ \\hline \\hline \n")
-if format == "txt":
- file. write( "+--------------+----------------+----------------+----------------+----------------+----------------+----------------+\n" )
- file. write( "| h | L1 Err. | L1 EOC. | L2 Err. | L2 EOC | MAX Err. | MAX EOC |\n" )
- file. write( "+==============+================+================+================+================+================+================+\n" )
-
-
-i = 0
-while i < items:
- if i == 0:
- if format == "latex":
- file. write( " " + string. ljust( str( h_list[ i ] ), h_width ) + "&" +
- string. rjust( "%.2g" % l1_norm_list[ i ], err_width ) + "&" +
- string. rjust( " ", err_width ) + "&"+
- string. rjust( "%.2g" % l2_norm_list[ i ], err_width ) + "&" +
- string. rjust( " ", err_width ) + "&" +
- string. rjust( "%.2g" % max_norm_list[ i ], err_width ) + "&" +
- string. rjust( " ", err_width ) + "\\\\\n" )
- if format == "txt":
- file. write( "| " + string. ljust( str( h_list[ i ] ), h_width ) + " |" +
- string. rjust( "%.2g" % l1_norm_list[ i ], err_width ) + " |" +
- string. rjust( " ", err_width ) + " |" +
- string. rjust( "%.2g" % l2_norm_list[ i ], err_width ) + " |" +
- string. rjust( " ", err_width ) + " |" +
- string. rjust( "%.2g" % max_norm_list[ i ], err_width ) + " |" +
- string. rjust( " ", err_width ) + " |\n" )
- file. write( "+--------------+----------------+----------------+----------------+----------------+----------------+----------------+\n" )
- i = i + 1;
- continue
- if h_list[ i ] == h_list[ i - 1 ]:
- print "Unable to count eoc since h[ " + \
- str( i ) + " ] = h[ " + str( i - 1 ) + \
- " ] = " + str( h_list[ i ] ) + ". \n"
- file. write( " eoc error: h[ " + \
- str( i ) + " ] = h[ " + str( i - 1 ) + \
- " ] = " + str( h_list[ i ] ) + ". \n" )
- else:
- h_ratio = math. log( h_list[ i ] / h_list[ i - 1 ] )
- l1_ratio = math. log( l1_norm_list[ i ] / l1_norm_list[ i - 1 ] )
- l2_ratio = math. log( l2_norm_list[ i ] / l2_norm_list[ i - 1 ] )
- max_ratio = math. log( max_norm_list[ i ] / max_norm_list[ i - 1 ] )
- if format == "latex":
- file. write( " " + string. ljust( str( h_list[ i ] ), h_width ) + "&" +
- string. rjust( "%.2g" % l1_norm_list[ i ], err_width ) + "&" +
- string. rjust( "{\\bf " + "%.2g" % ( l1_ratio / h_ratio ) + "}", err_width ) + "&" +
- string. rjust( "%.2g" % l2_norm_list[ i ], err_width ) + "&" +
- string. rjust( "{\\bf " + "%.2g" % ( l2_ratio / h_ratio ) + "}", err_width ) + "&" +
- string. rjust( "%.2g" % max_norm_list[ i ], err_width ) + "&" +
- string. rjust( "{\\bf " + "%.2g" % ( max_ratio / h_ratio ) + "}", err_width ) + "\\\\\n" )
- if format == "txt":
- file. write( "| " + string. ljust( str( h_list[ i ] ), h_width ) + " |" +
- string. rjust( "%.2g" % l1_norm_list[ i ], err_width ) + " |" +
- string. rjust( "**" + "%.2g" % ( l1_ratio / h_ratio ) + "**", err_width ) + " |" +
- string. rjust( "%.2g" % l2_norm_list[ i ], err_width ) + " |" +
- string. rjust( "**" + "%.2g" % ( l2_ratio / h_ratio ) + "**", err_width ) + " |" +
- string. rjust( "%.2g" % max_norm_list[ i ], err_width ) + " |" +
- string. rjust( "**" + "%.2g" % ( max_ratio / h_ratio ) + "**", err_width ) + " |\n" )
- file. write( "+--------------+----------------+----------------+----------------+----------------+----------------+----------------+\n" )
- i = i + 1
-
-if format == "latex":
- file. write( "\\hline \n" )
- file. write( "\\end{tabular} \n" )
-
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver.h
deleted file mode 100644
index 19cdd949359d4349172af820def49169146c8717..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver.h
+++ /dev/null
@@ -1,366 +0,0 @@
-/***************************************************************************
- tnlParallelEikonalSolver.h - description
- -------------------
- begin : Nov 28 , 2014
- copyright : (C) 2014 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef TNLPARALLELEIKONALSOLVER_H_
-#define TNLPARALLELEIKONALSOLVER_H_
-
-#include <TNL/Config/ParameterContainer.h>
-#include <TNL/Containers/Vector.h>
-#include <TNL/Containers/StaticVector.h>
-#include <functions/tnlMeshFunction.h>
-#include <TNL/Devices/Host.h>
-#include <mesh/tnlGrid.h>
-#include <mesh/grids/tnlGridEntity.h>
-#include <limits.h>
-#include <core/tnlDevice.h>
- #include <omp.h>
-
-
-#include <ctime>
-
-#ifdef HAVE_CUDA
-#include <core/tnlCuda.h>
-#endif
-
-
-template< int Dimension,
- typename SchemeHost,
- typename SchemeDevice,
- typename Device,
- typename RealType = double,
- typename IndexType = int >
-class tnlParallelEikonalSolver
-{};
-
-template<typename SchemeHost, typename SchemeDevice, typename Device>
-class tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >
-{
-public:
-
- typedef SchemeDevice SchemeTypeDevice;
- typedef SchemeHost SchemeTypeHost;
- typedef Device DeviceType;
- typedef TNL::Containers::Vector< double, TNL::Devices::Host, int > VectorType;
- typedef TNL::Containers::Vector< int, TNL::Devices::Host, int > IntVectorType;
- typedef tnlGrid< 2, double, TNL::Devices::Host, int > MeshType;
-#ifdef HAVE_CUDA
- typedef TNL::Containers::Vector< double, TNL::Devices::Host, int > VectorTypeCUDA;
- typedef TNL::Containers::Vector< int, TNL::Devices::Host, int > IntVectorTypeCUDA;
- typedef tnlGrid< 2, double, TNL::Devices::Host, int > MeshTypeCUDA;
-#endif
- tnlParallelEikonalSolver();
- bool init( const Config::ParameterContainer& parameters );
- void run();
-
- void test();
-
-/*private:*/
-
-
- void synchronize();
-
- int getOwner( int i) const;
-
- int getSubgridValue( int i ) const;
-
- void setSubgridValue( int i, int value );
-
- int getBoundaryCondition( int i ) const;
-
- void setBoundaryCondition( int i, int value );
-
- void stretchGrid();
-
- void contractGrid();
-
- VectorType getSubgrid( const int i ) const;
-
- void insertSubgrid( VectorType u, const int i );
-
- VectorType runSubgrid( int boundaryCondition, VectorType u, int subGridID);
-
-
- tnlMeshFunction<MeshType> u0;
- VectorType work_u;
- IntVectorType subgridValues, boundaryConditions, unusedCell, calculationsCount;
- MeshType mesh, subMesh;
-
-// tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage > Entity;
-
- SchemeHost schemeHost;
- SchemeDevice schemeDevice;
- double delta, tau0, stopTime,cflCondition;
- int gridRows, gridCols, gridLevels, currentStep, n;
-
- std::clock_t start;
- double time_diff;
-
-
- tnlDeviceEnum device;
-
- tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* getSelf()
- {
- return this;
- };
-
-#ifdef HAVE_CUDA
-
- tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver;
-
- double* work_u_cuda;
-
- int* subgridValues_cuda;
- int*boundaryConditions_cuda;
- int* unusedCell_cuda;
- int* calculationsCount_cuda;
- double* tmpw;
- //MeshTypeCUDA mesh_cuda, subMesh_cuda;
- //SchemeDevice scheme_cuda;
- //double delta_cuda, tau0_cuda, stopTime_cuda,cflCondition_cuda;
- //int gridRows_cuda, gridCols_cuda, currentStep_cuda, n_cuda;
-
- int* runcuda;
- int run_host;
-
-
- __device__ void getSubgridCUDA2D( const int i, tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* caller, double* a);
-
- __device__ void updateSubgridCUDA2D( const int i, tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* caller, double* a);
-
- __device__ void insertSubgridCUDA2D( double u, const int i );
-
- __device__ void runSubgridCUDA2D( int boundaryCondition, double* u, int subGridID);
-
- /*__global__ void runCUDA();*/
-
- //__device__ void synchronizeCUDA();
-
- __device__ int getOwnerCUDA2D( int i) const;
-
- __device__ int getSubgridValueCUDA2D( int i ) const;
-
- __device__ void setSubgridValueCUDA2D( int i, int value );
-
- __device__ int getBoundaryConditionCUDA2D( int i ) const;
-
- __device__ void setBoundaryConditionCUDA2D( int i, int value );
-
- //__device__ bool initCUDA( tnlParallelEikonalSolver<SchemeHost, SchemeDevice, Device, double, int >* cudaSolver);
-
- /*__global__ void initRunCUDA(tnlParallelEikonalSolver<Scheme, double, TNL::Devices::Host, int >* caller);*/
-
-#endif
-
-};
-
-
-
-
-
-
-
- template<typename SchemeHost, typename SchemeDevice, typename Device>
- class tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >
- {
- public:
-
- typedef SchemeDevice SchemeTypeDevice;
- typedef SchemeHost SchemeTypeHost;
- typedef Device DeviceType;
- typedef TNL::Containers::Vector< double, TNL::Devices::Host, int > VectorType;
- typedef TNL::Containers::Vector< int, TNL::Devices::Host, int > IntVectorType;
- typedef tnlGrid< 3, double, TNL::Devices::Host, int > MeshType;
- #ifdef HAVE_CUDA
- typedef TNL::Containers::Vector< double, TNL::Devices::Host, int > VectorTypeCUDA;
- typedef TNL::Containers::Vector< int, TNL::Devices::Host, int > IntVectorTypeCUDA;
- typedef tnlGrid< 3, double, TNL::Devices::Host, int > MeshTypeCUDA;
- #endif
- tnlParallelEikonalSolver();
- bool init( const Config::ParameterContainer& parameters );
- void run();
-
- void test();
-
- /*private:*/
-
-
- void synchronize();
-
- int getOwner( int i) const;
-
- int getSubgridValue( int i ) const;
-
- void setSubgridValue( int i, int value );
-
- int getBoundaryCondition( int i ) const;
-
- void setBoundaryCondition( int i, int value );
-
- void stretchGrid();
-
- void contractGrid();
-
- VectorType getSubgrid( const int i ) const;
-
- void insertSubgrid( VectorType u, const int i );
-
- VectorType runSubgrid( int boundaryCondition, VectorType u, int subGridID);
-
-
- tnlMeshFunction<MeshType> u0;
- VectorType work_u;
- IntVectorType subgridValues, boundaryConditions, unusedCell, calculationsCount;
- MeshType mesh, subMesh;
- SchemeHost schemeHost;
- SchemeDevice schemeDevice;
- double delta, tau0, stopTime,cflCondition;
- int gridRows, gridCols, gridLevels, currentStep, n;
-
- std::clock_t start;
- double time_diff;
-
-
- tnlDeviceEnum device;
-
- tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* getSelf()
- {
- return this;
- };
-
-#ifdef HAVE_CUDA
-
- tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver;
-
- double* work_u_cuda;
-
- int* subgridValues_cuda;
- int*boundaryConditions_cuda;
- int* unusedCell_cuda;
- int* calculationsCount_cuda;
- double* tmpw;
- //MeshTypeCUDA mesh_cuda, subMesh_cuda;
- //SchemeDevice scheme_cuda;
- //double delta_cuda, tau0_cuda, stopTime_cuda,cflCondition_cuda;
- //int gridRows_cuda, gridCols_cuda, currentStep_cuda, n_cuda;
-
- int* runcuda;
- int run_host;
-
-
- __device__ void getSubgridCUDA3D( const int i, tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* caller, double* a);
-
- __device__ void updateSubgridCUDA3D( const int i, tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* caller, double* a);
-
- __device__ void insertSubgridCUDA3D( double u, const int i );
-
- __device__ void runSubgridCUDA3D( int boundaryCondition, double* u, int subGridID);
-
- /*__global__ void runCUDA();*/
-
- //__device__ void synchronizeCUDA();
-
- __device__ int getOwnerCUDA3D( int i) const;
-
- __device__ int getSubgridValueCUDA3D( int i ) const;
-
- __device__ void setSubgridValueCUDA3D( int i, int value );
-
- __device__ int getBoundaryConditionCUDA3D( int i ) const;
-
- __device__ void setBoundaryConditionCUDA3D( int i, int value );
-
- //__device__ bool initCUDA( tnlParallelEikonalSolver<SchemeHost, SchemeDevice, Device, double, int >* cudaSolver);
-
- /*__global__ void initRunCUDA(tnlParallelEikonalSolver<Scheme, double, TNL::Devices::Host, int >* caller);*/
-
-#endif
-
-};
-
-
-
-
-
-
-#ifdef HAVE_CUDA
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void runCUDA2D(tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* caller);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void initRunCUDA2D(tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* caller);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void initCUDA2D( tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver, double* ptr, int * ptr2, int* ptr3);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void synchronizeCUDA2D(tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void synchronize2CUDA2D(tnlParallelEikonalSolver<2, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver);
-
-
-
-
-
-
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void runCUDA3D(tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* caller);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void initRunCUDA3D(tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* caller);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void initCUDA3D( tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver, double* ptr, int * ptr2, int* ptr3);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void synchronizeCUDA3D(tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver);
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__ void synchronize2CUDA3D(tnlParallelEikonalSolver<3, SchemeHost, SchemeDevice, Device, double, int >* cudaSolver);
-#endif
-
-
-#ifdef HAVE_CUDA
-__cuda_callable__
-double fabsMin( double x, double y)
-{
- double fx = fabs(x);
-
- if(Min(fx,fabs(y)) == fx)
- return x;
- else
- return y;
-}
-
-__cuda_callable__
-double atomicFabsMin(double* address, double val)
-{
- unsigned long long int* address_as_ull =
- (unsigned long long int*)address;
- unsigned long long int old = *address_as_ull, assumed;
- do {
- assumed = old;
- old = atomicCAS(address_as_ull, assumed,__double_as_longlong( fabsMin(__longlong_as_double(assumed),val) ));
- } while (assumed != old);
- return __longlong_as_double(old);
-}
-
-#endif
-
-#include "tnlParallelEikonalSolver2D_impl.h"
-#include "tnlParallelEikonalSolver3D_impl.h"
-#endif /* TNLPARALLELEIKONALSOLVER_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver2D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver2D_impl.h
deleted file mode 100644
index 76cf49bc8aa28890d598fe010aa777acb2c6edfd..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver2D_impl.h
+++ /dev/null
@@ -1,1928 +0,0 @@
-/***************************************************************************
- tnlParallelEikonalSolver2D_impl.h - description
- -------------------
- begin : Nov 28 , 2014
- copyright : (C) 2014 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef TNLPARALLELEIKONALSOLVER2D_IMPL_H_
-#define TNLPARALLELEIKONALSOLVER2D_IMPL_H_
-
-
-#include "tnlParallelEikonalSolver.h"
-#include <core/mfilename.h>
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::tnlParallelEikonalSolver()
-{
- cout << "a" <<std::endl;
- this->device = tnlCudaDevice; /////////////// tnlCuda Device --- vypocet na GPU, TNL::Devices::HostDevice --- vypocet na CPU
-
-#ifdef HAVE_CUDA
- if(this->device == tnlCudaDevice)
- {
- run_host = 1;
- }
-#endif
-
- cout << "b" <<std::endl;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::test()
-{
-/*
- for(int i =0; i < this->subgridValues.getSize(); i++ )
- {
- insertSubgrid(getSubgrid(i), i);
- }
-*/
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-
-bool tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::init( const Config::ParameterContainer& parameters )
-{
- cout << "Initializating solver..." <<std::endl;
- const String& meshLocation = parameters.getParameter <String>("mesh");
- this->mesh.load( meshLocation );
-
- this->n = parameters.getParameter <int>("subgrid-size");
- cout << "Setting N to " << this->n <<std::endl;
-
- this->subMesh.setDimensions( this->n, this->n );
- this->subMesh.setDomain( Containers::StaticVector<2,double>(0.0, 0.0),
- Containers::StaticVector<2,double>(mesh.template getSpaceStepsProducts< 1, 0 >()*(double)(this->n), mesh.template getSpaceStepsProducts< 0, 1 >()*(double)(this->n)) );
-
- this->subMesh.save("submesh.tnl");
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- this->u0.load( initialCondition );
-
- //cout << this->mesh.getCellCenter(0) <<std::endl;
-
- this->delta = parameters.getParameter <double>("delta");
- this->delta *= mesh.template getSpaceStepsProducts< 1, 0 >()*mesh.template getSpaceStepsProducts< 0, 1 >();
-
- cout << "Setting delta to " << this->delta <<std::endl;
-
- this->tau0 = parameters.getParameter <double>("initial-tau");
- cout << "Setting initial tau to " << this->tau0 <<std::endl;
- this->stopTime = parameters.getParameter <double>("stop-time");
-
- this->cflCondition = parameters.getParameter <double>("cfl-condition");
- this -> cflCondition *= sqrt(mesh.template getSpaceStepsProducts< 1, 0 >()*mesh.template getSpaceStepsProducts< 0, 1 >());
- cout << "Setting CFL to " << this->cflCondition <<std::endl;
-
- stretchGrid();
- this->stopTime /= (double)(this->gridCols);
- this->stopTime *= (1.0+1.0/((double)(this->n) - 2.0));
- cout << "Setting stopping time to " << this->stopTime <<std::endl;
- //this->stopTime = 1.5*((double)(this->n))*parameters.getParameter <double>("stop-time")*this->mesh.template getSpaceStepsProducts< 1, 0 >();
- //cout << "Setting stopping time to " << this->stopTime <<std::endl;
-
- cout << "Initializating scheme..." <<std::endl;
- if(!this->schemeHost.init(parameters))
- {
- cerr << "SchemeHost failed to initialize." <<std::endl;
- return false;
- }
- cout << "Scheme initialized." <<std::endl;
-
- test();
-
- VectorType* tmp = new VectorType[subgridValues.getSize()];
- bool containsCurve = false;
-
-#ifdef HAVE_CUDA
-
- if(this->device == tnlCudaDevice)
- {
- /*cout << "Testing... " <<std::endl;
- if(this->device == tnlCudaDevice)
- {
- if( !initCUDA2D(parameters, gridRows, gridCols) )
- return false;
- }*/
- //cout << "s" <<std::endl;
- cudaMalloc(&(this->cudaSolver), sizeof(tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >));
- //cout << "s" <<std::endl;
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >), cudaMemcpyHostToDevice);
- //cout << "s" <<std::endl;
- double** tmpdev = NULL;
- cudaMalloc(&tmpdev, sizeof(double*));
- //double* tmpw;
- cudaMalloc(&(this->tmpw), this->work_u.getSize()*sizeof(double));
- cudaMalloc(&(this->runcuda), sizeof(int));
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- int* tmpUC;
- cudaMalloc(&(tmpUC), this->work_u.getSize()*sizeof(int));
- cudaMemcpy(tmpUC, this->unusedCell.getData(), this->unusedCell.getSize()*sizeof(int), cudaMemcpyHostToDevice);
-
- initCUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<1,1>>>(this->cudaSolver, (this->tmpw), (this->runcuda),tmpUC);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cout << "s " <<std::endl;
- //cudaMalloc(&(cudaSolver->work_u_cuda), this->work_u.getSize()*sizeof(double));
- double* tmpu = NULL;
-
- cudaMemcpy(&tmpu, tmpdev,sizeof(double*), cudaMemcpyDeviceToHost);
- //printf("%p %p \n",tmpu,tmpw);
- cudaMemcpy((this->tmpw), this->work_u.getData(), this->work_u.getSize()*sizeof(double), cudaMemcpyHostToDevice);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cout << "s "<<std::endl;
-
- }
-#endif
-
- if(this->device == TNL::Devices::HostDevice)
- {
- for(int i = 0; i < this->subgridValues.getSize(); i++)
- {
-
- if(! tmp[i].setSize(this->n * this->n))
- cout << "Could not allocate tmp["<< i <<"] array." <<std::endl;
- tmp[i] = getSubgrid(i);
- containsCurve = false;
-
- for(int j = 0; j < tmp[i].getSize(); j++)
- {
- if(tmp[i][0]*tmp[i][j] <= 0.0)
- {
- containsCurve = true;
- j=tmp[i].getSize();
- }
-
- }
- if(containsCurve)
- {
- //cout << "Computing initial SDF on subgrid " << i << "." <<std::endl;
- tmp[i] = runSubgrid(0, tmp[i],i);
- insertSubgrid(tmp[i], i);
- setSubgridValue(i, 4);
- //cout << "Computed initial SDF on subgrid " << i << "." <<std::endl;
- }
- containsCurve = false;
-
- }
- }
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
-// cout << "pre 1 kernel" <<std::endl;
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- dim3 threadsPerBlock(this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initRunCUDA2D<SchemeTypeHost,SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock,3*this->n*this->n*sizeof(double)>>>(this->cudaSolver);
- cudaDeviceSynchronize();
-// cout << "post 1 kernel" <<std::endl;
-
- }
-#endif
-
-
- this->currentStep = 1;
- if(this->device == TNL::Devices::HostDevice)
- synchronize();
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
- dim3 threadsPerBlock(this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows);
- //double * test = (double*)malloc(this->work_u.getSize()*sizeof(double));
- //cout << test[0] <<" " << test[1] <<" " << test[2] <<" " << test[3] <<std::endl;
- //cudaMemcpy(/*this->work_u.getData()*/ test, (this->tmpw), this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- //cout << this->tmpw << " " << test[0] <<" " << test[1] << " " <<test[2] << " " <<test[3] <<std::endl;
-
- TNL_CHECK_CUDA_DEVICE;
-
- synchronizeCUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- synchronize2CUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cout << test[0] << " " <<test[1] <<" " << test[2] << " " <<test[3] <<std::endl;
- //cudaMemcpy(/*this->work_u.getData()*/ test, (this->tmpw), this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- //TNL_CHECK_CUDA_DEVICE;
- //cout << this->tmpw << " " << test[0] << " " <<test[1] << " " <<test[2] <<" " << test[3] <<std::endl;
- //free(test);
-
- }
-
-#endif
- cout << "Solver initialized." <<std::endl;
-
- return true;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::run()
-{
- if(this->device == TNL::Devices::HostDevice)
- {
-
- bool end = false;
- while ((this->boundaryConditions.max() > 0 ) || !end)
- {
- if(this->boundaryConditions.max() == 0 )
- end=true;
- else
- end=false;
-#ifdef HAVE_OPENMP
-#pragma omp parallel for num_threads(4) schedule(dynamic)
-#endif
- for(int i = 0; i < this->subgridValues.getSize(); i++)
- {
- if(getSubgridValue(i) != INT_MAX)
- {
- VectorType tmp;
- tmp.setSize(this->n * this->n);
- //cout << "subMesh: " << i << ", BC: " << getBoundaryCondition(i) <<std::endl;
-
- if(getSubgridValue(i) == currentStep+4)
- {
-
- if(getBoundaryCondition(i) & 1)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(1, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 2)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(1, tmp ,i);
- insertSubgrid( tmp, 2);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 4)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(4, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 8)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(8, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- }
-
- if( ((getBoundaryCondition(i) & 2) )|| (getBoundaryCondition(i) & 1)//)
- /* &&(!(getBoundaryCondition(i) & 5) && !(getBoundaryCondition(i) & 10)) */)
- {
- //cout << "3 @ " << getBoundaryCondition(i) <<std::endl;
- tmp = getSubgrid(i);
- tmp = runSubgrid(1, tmp ,i);
- insertSubgrid( tmp, 3);
- }
- if( ((getBoundaryCondition(i) & 4) )|| (getBoundaryCondition(i) & 1)//)
- /* &&(!(getBoundaryCondition(i) & 3) && !(getBoundaryCondition(i) & 12)) */)
- {
- //cout << "5 @ " << getBoundaryCondition(i) <<std::endl;
- tmp = getSubgrid(i);
- tmp = runSubgrid(5, tmp ,i);
- insertSubgrid( tmp, i);
- }
- if( ((getBoundaryCondition(i) & 2) )|| (getBoundaryCondition(i) & 8)//)
- /* &&(!(getBoundaryCondition(i) & 12) && !(getBoundaryCondition(i) & 3))*/ )
- {
- //cout << "10 @ " << getBoundaryCondition(i) <<std::endl;
- tmp = getSubgrid(i);
- tmp = runSubgrid(10, tmp ,i);
- insertSubgrid( tmp, i);
- }
- if( ((getBoundaryCondition(i) & 4) )|| (getBoundaryCondition(i) & 8)//)
- /*&&(!(getBoundaryCondition(i) & 10) && !(getBoundaryCondition(i) & 5)) */)
- {
- //cout << "12 @ " << getBoundaryCondition(i) <<std::endl;
- tmp = getSubgrid(i);
- tmp = runSubgrid(12, tmp ,i);
- insertSubgrid( tmp, i);
- }
-
-
- /*if(getBoundaryCondition(i))
- {
- insertSubgrid( runSubgrid(15, getSubgrid(i),i), i);
- }*/
-
- setBoundaryCondition(i, 0);
-
- setSubgridValue(i, getSubgridValue(i)-1);
-
- }
- }
- synchronize();
- }
- }
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
- //cout << "fn" <<std::endl;
- bool end_cuda = false;
- dim3 threadsPerBlock(this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cudaMalloc(&runcuda,sizeof(bool));
- //cudaMemcpy(runcuda, &run_host, sizeof(bool), cudaMemcpyHostToDevice);
- //cout << "fn" <<std::endl;
- bool* tmpb;
- //cudaMemcpy(tmpb, &(cudaSolver->runcuda),sizeof(bool*), cudaMemcpyDeviceToHost);
- //cudaDeviceSynchronize();
- //TNL_CHECK_CUDA_DEVICE;
- cudaMemcpy(&(this->run_host),this->runcuda,sizeof(int), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cout << "fn" <<std::endl;
- int i = 1;
- time_diff = 0.0;
- while (run_host || !end_cuda)
- {
- cout << "Computing at step "<< i++ <<std::endl;
- if(run_host != 0 )
- end_cuda = true;
- else
- end_cuda = false;
- //cout << "a" <<std::endl;
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- start = std::clock();
- runCUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock,3*this->n*this->n*sizeof(double)>>>(this->cudaSolver);
- //cout << "a" <<std::endl;
- cudaDeviceSynchronize();
- time_diff += (std::clock() - start) / (double)(CLOCKS_PER_SEC);
-
- //start = std::clock();
- synchronizeCUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- synchronize2CUDA2D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //time_diff += (std::clock() - start) / (double)(CLOCKS_PER_SEC);
-
-
- //cout << "a" <<std::endl;
- //run_host = false;
- //cout << "in kernel loop" << run_host <<std::endl;
- //cudaMemcpy(tmpb, &(cudaSolver->runcuda),sizeof(bool*), cudaMemcpyDeviceToHost);
- cudaMemcpy(&run_host, (this->runcuda),sizeof(int), cudaMemcpyDeviceToHost);
- //cout << "in kernel loop" << run_host <<std::endl;
- }
- cout << "Solving time was: " << time_diff <<std::endl;
- //cout << "b" <<std::endl;
-
- //double* tmpu;
- //cudaMemcpy(tmpu, &(cudaSolver->work_u_cuda),sizeof(double*), cudaMemcpyHostToDevice);
- //cudaMemcpy(this->work_u.getData(), tmpu, this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- //cout << this->work_u.getData()[0] <<std::endl;
-
- //double * test = (double*)malloc(this->work_u.getSize()*sizeof(double));
- //cout << test[0] << test[1] << test[2] << test[3] <<std::endl;
- cudaMemcpy(this->work_u.getData()/* test*/, (this->tmpw), this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- //cout << this->tmpw << " " << test[0] << test[1] << test[2] << test[3] <<std::endl;
- //free(test);
-
- cudaDeviceSynchronize();
- }
-#endif
- contractGrid();
- this->u0.save("u-00001.tnl");
- cout << "Maximum number of calculations on one subgrid was " << this->calculationsCount.absMax() <<std::endl;
- cout << "Average number of calculations on one subgrid was " << ( (double) this->calculationsCount.sum() / (double) this->calculationsCount.getSize() ) <<std::endl;
- cout << "Solver finished" <<std::endl;
-
-#ifdef HAVE_CUDA
- if(this->device == tnlCudaDevice)
- {
- cudaFree(this->runcuda);
- cudaFree(this->tmpw);
- cudaFree(this->cudaSolver);
- }
-#endif
-
-}
-
-//north - 1, east - 2, west - 4, south - 8
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::synchronize() //needs fix ---- maybe not anymore --- but frankly: yeah, it does -- aaaa-and maybe fixed now
-{
- cout << "Synchronizig..." <<std::endl;
- int tmp1, tmp2;
- int grid1, grid2;
-
- if(this->currentStep & 1)
- {
- for(int j = 0; j < this->gridRows - 1; j++)
- {
- for (int i = 0; i < this->gridCols*this->n; i++)
- {
- tmp1 = this->gridCols*this->n*((this->n-1)+j*this->n) + i;
- tmp2 = this->gridCols*this->n*((this->n)+j*this->n) + i;
- grid1 = getSubgridValue(getOwner(tmp1));
- grid2 = getSubgridValue(getOwner(tmp2));
- if(getOwner(tmp1)==getOwner(tmp2))
- cout << "i, j" << i << "," << j <<std::endl;
- if ((fabs(this->work_u[tmp1]) < fabs(this->work_u[tmp2]) - this->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- this->work_u[tmp2] = this->work_u[tmp1];
- this->unusedCell[tmp2] = 0;
- if(grid2 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp2), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp2)) & 8) )
- setBoundaryCondition(getOwner(tmp2), getBoundaryCondition(getOwner(tmp2))+8);
- }
- else if ((fabs(this->work_u[tmp1]) > fabs(this->work_u[tmp2]) + this->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- this->work_u[tmp1] = this->work_u[tmp2];
- this->unusedCell[tmp1] = 0;
- if(grid1 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp1), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp1)) & 1) )
- setBoundaryCondition(getOwner(tmp1), getBoundaryCondition(getOwner(tmp1))+1);
- }
- }
- }
-
- }
- else
- {
- for(int i = 1; i < this->gridCols; i++)
- {
- for (int j = 0; j < this->gridRows*this->n; j++)
- {
- tmp1 = this->gridCols*this->n*j + i*this->n - 1;
- tmp2 = this->gridCols*this->n*j + i*this->n ;
- grid1 = getSubgridValue(getOwner(tmp1));
- grid2 = getSubgridValue(getOwner(tmp2));
- if(getOwner(tmp1)==getOwner(tmp2))
- cout << "i, j" << i << "," << j <<std::endl;
- if ((fabs(this->work_u[tmp1]) < fabs(this->work_u[tmp2]) - this->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- this->work_u[tmp2] = this->work_u[tmp1];
- this->unusedCell[tmp2] = 0;
- if(grid2 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp2), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp2)) & 4) )
- setBoundaryCondition(getOwner(tmp2), getBoundaryCondition(getOwner(tmp2))+4);
- }
- else if ((fabs(this->work_u[tmp1]) > fabs(this->work_u[tmp2]) + this->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- this->work_u[tmp1] = this->work_u[tmp2];
- this->unusedCell[tmp1] = 0;
- if(grid1 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp1), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp1)) & 2) )
- setBoundaryCondition(getOwner(tmp1), getBoundaryCondition(getOwner(tmp1))+2);
- }
- }
- }
- }
-
-
- this->currentStep++;
- int stepValue = this->currentStep + 4;
- for (int i = 0; i < this->subgridValues.getSize(); i++)
- {
- if( getSubgridValue(i) == -INT_MAX )
- setSubgridValue(i, stepValue);
- }
-
- cout << "Grid synchronized at step " << (this->currentStep - 1 ) <<std::endl;
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getOwner(int i) const
-{
-
- return (i / (this->gridCols*this->n*this->n))*this->gridCols + (i % (this->gridCols*this->n))/this->n;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getSubgridValue( int i ) const
-{
- return this->subgridValues[i];
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::setSubgridValue(int i, int value)
-{
- this->subgridValues[i] = value;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getBoundaryCondition( int i ) const
-{
- return this->boundaryConditions[i];
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::setBoundaryCondition(int i, int value)
-{
- this->boundaryConditions[i] = value;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::stretchGrid()
-{
- cout << "Stretching grid..." <<std::endl;
-
-
- this->gridCols = ceil( ((double)(this->mesh.getDimensions().x()-1)) / ((double)(this->n-1)) );
- this->gridRows = ceil( ((double)(this->mesh.getDimensions().y()-1)) / ((double)(this->n-1)) );
-
- //this->gridCols = (this->mesh.getDimensions().x()-1) / (this->n-1) ;
- //this->gridRows = (this->mesh.getDimensions().y()-1) / (this->n-1) ;
-
- cout << "Setting gridCols to " << this->gridCols << "." <<std::endl;
- cout << "Setting gridRows to " << this->gridRows << "." <<std::endl;
-
- this->subgridValues.setSize(this->gridCols*this->gridRows);
- this->subgridValues.setValue(0);
- this->boundaryConditions.setSize(this->gridCols*this->gridRows);
- this->boundaryConditions.setValue(0);
- this->calculationsCount.setSize(this->gridCols*this->gridRows);
- this->calculationsCount.setValue(0);
-
- for(int i = 0; i < this->subgridValues.getSize(); i++ )
- {
- this->subgridValues[i] = INT_MAX;
- this->boundaryConditions[i] = 0;
- }
-
- int stretchedSize = this->n*this->n*this->gridCols*this->gridRows;
-
- if(!this->work_u.setSize(stretchedSize))
- cerr << "Could not allocate memory for stretched grid." <<std::endl;
- if(!this->unusedCell.setSize(stretchedSize))
- cerr << "Could not allocate memory for supporting stretched grid." <<std::endl;
- int idealStretch =this->mesh.getDimensions().x() + (this->mesh.getDimensions().x()-2)/(this->n-1);
- cout << idealStretch <<std::endl;
-
- for(int i = 0; i < stretchedSize; i++)
- {
- this->unusedCell[i] = 1;
- int diff =(this->n*this->gridCols) - idealStretch ;
- //cout << "diff = " << diff <<endl;
- int k = i/this->n - i/(this->n*this->gridCols) + this->mesh.getDimensions().x()*(i/(this->n*this->n*this->gridCols)) + (i/(this->n*this->gridCols))*diff;
-
- if(i%(this->n*this->gridCols) - idealStretch >= 0)
- {
- //cout << i%(this->n*this->gridCols) - idealStretch +1 <<std::endl;
- k+= i%(this->n*this->gridCols) - idealStretch +1 ;
- }
-
- if(i/(this->n*this->gridCols) - idealStretch + 1 > 0)
- {
- //cout << i/(this->n*this->gridCols) - idealStretch + 1 <<std::endl;
- k+= (i/(this->n*this->gridCols) - idealStretch +1 )* this->mesh.getDimensions().x() ;
- }
-
- //cout << "i = " << i << " : i-k = " << i-k <<std::endl;
- /*int j=(i % (this->n*this->gridCols)) - ( (this->mesh.getDimensions().x() - this->n)/(this->n - 1) + this->mesh.getDimensions().x() - 1)
- + (this->n*this->gridCols - this->mesh.getDimensions().x())*(i/(this->n*this->n*this->gridCols)) ;
-
- if(j > 0)
- k += j;
-
- int l = i-k - (this->u0.getSize() - 1);
- int m = (l % this->mesh.getDimensions().x());
-
- if(l>0)
- k+= l + ( (l / this->mesh.getDimensions().x()) + 1 )*this->mesh.getDimensions().x() - (l % this->mesh.getDimensions().x());*/
-
- this->work_u[i] = this->u0[i-k];
- //cout << (i-k) <<endl;
- }
-
-
- cout << "Grid stretched." <<std::endl;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::contractGrid()
-{
- cout << "Contracting grid..." <<std::endl;
- int stretchedSize = this->n*this->n*this->gridCols*this->gridRows;
-
- int idealStretch =this->mesh.getDimensions().x() + (this->mesh.getDimensions().x()-2)/(this->n-1);
- cout << idealStretch <<std::endl;
-
- for(int i = 0; i < stretchedSize; i++)
- {
- int diff =(this->n*this->gridCols) - idealStretch ;
- int k = i/this->n - i/(this->n*this->gridCols) + this->mesh.getDimensions().x()*(i/(this->n*this->n*this->gridCols)) + (i/(this->n*this->gridCols))*diff;
-
- if((i%(this->n*this->gridCols) - idealStretch < 0) && (i/(this->n*this->gridCols) - idealStretch + 1 <= 0))
- {
- //cout << i <<" : " <<i-k<<std::endl;
- this->u0[i-k] = this->work_u[i];
- }
-
- }
-
- cout << "Grid contracted" <<std::endl;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-typename tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::VectorType
-tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getSubgrid( const int i ) const
-{
- VectorType u;
- u.setSize(this->n*this->n);
-
- for( int j = 0; j < u.getSize(); j++)
- {
- u[j] = this->work_u[ (i / this->gridCols) * this->n*this->n*this->gridCols
- + (i % this->gridCols) * this->n
- + (j/this->n) * this->n*this->gridCols
- + (j % this->n) ];
- }
- return u;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::insertSubgrid( VectorType u, const int i )
-{
-
- for( int j = 0; j < this->n*this->n; j++)
- {
- int index = (i / this->gridCols)*this->n*this->n*this->gridCols
- + (i % this->gridCols)*this->n
- + (j/this->n)*this->n*this->gridCols
- + (j % this->n);
- //OMP LOCK index
- if( (fabs(this->work_u[index]) > fabs(u[j])) || (this->unusedCell[index] == 1) )
- {
- this->work_u[index] = u[j];
- this->unusedCell[index] = 0;
- }
- //OMP UNLOCK index
- }
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-typename tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::VectorType
-tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::runSubgrid( int boundaryCondition, VectorType u, int subGridID)
-{
-
- VectorType fu;
-
- fu.setLike(u);
- fu.setValue( 0.0 );
-
-/*
- * Insert Euler-Solver Here
- */
-
- /**/
-
- /*for(int i = 0; i < u.getSize(); i++)
- {
- int x = this->subMesh.getCellCoordinates(i).x();
- int y = this->subMesh.getCellCoordinates(i).y();
-
- if(x == 0 && (boundaryCondition & 4) && y ==0)
- {
- if((u[subMesh.getCellYSuccessor( i )] - u[i])/subMesh.template getSpaceStepsProducts< 0, 1 >() > 1.0)
- {
- //cout << "x = 0; y = 0" <<std::endl;
- u[i] = u[subMesh.getCellYSuccessor( i )] - subMesh.template getSpaceStepsProducts< 0, 1 >();
- }
- }
- else if(x == 0 && (boundaryCondition & 4) && y == subMesh.getDimensions().y() - 1)
- {
- if((u[subMesh.getCellYPredecessor( i )] - u[i])/subMesh.template getSpaceStepsProducts< 0, 1 >() > 1.0)
- {
- //cout << "x = 0; y = n" <<std::endl;
- u[i] = u[subMesh.getCellYPredecessor( i )] - subMesh.template getSpaceStepsProducts< 0, 1 >();
- }
- }
-
-
- else if(x == subMesh.getDimensions().x() - 1 && (boundaryCondition & 2) && y ==0)
- {
- if((u[subMesh.getCellYSuccessor( i )] - u[i])/subMesh.template getSpaceStepsProducts< 0, 1 >() > 1.0)
- {
- //cout << "x = n; y = 0" <<std::endl;
- u[i] = u[subMesh.getCellYSuccessor( i )] - subMesh.template getSpaceStepsProducts< 0, 1 >();
- }
- }
- else if(x == subMesh.getDimensions().x() - 1 && (boundaryCondition & 2) && y == subMesh.getDimensions().y() - 1)
- {
- if((u[subMesh.getCellYPredecessor( i )] - u[i])/subMesh.template getSpaceStepsProducts< 0, 1 >() > 1.0)
- {
- //cout << "x = n; y = n" <<std::endl;
- u[i] = u[subMesh.getCellYPredecessor( i )] - subMesh.template getSpaceStepsProducts< 0, 1 >();
- }
- }
-
-
- else if(y == 0 && (boundaryCondition & 8) && x ==0)
- {
- if((u[subMesh.getCellXSuccessor( i )] - u[i])/subMesh.template getSpaceStepsProducts< 1, 0 >() > 1.0)
- {
- //cout << "y = 0; x = 0" <<std::endl;
- u[i] = u[subMesh.getCellXSuccessor( i )] - subMesh.template getSpaceStepsProducts< 1, 0 >();
- }
- }
- else if(y == 0 && (boundaryCondition & 8) && x == subMesh.getDimensions().x() - 1)
- {
- if((u[subMesh.getCellXPredecessor( i )] - u[i])/subMesh.template getSpaceStepsProducts< 1, 0 >() > 1.0)
- {
- //cout << "y = 0; x = n" <<std::endl;
- u[i] = u[subMesh.getCellXPredecessor( i )] - subMesh.template getSpaceStepsProducts< 1, 0 >();
- }
- }
-
-
- else if(y == subMesh.getDimensions().y() - 1 && (boundaryCondition & 1) && x ==0)
- {
- if((u[subMesh.getCellXSuccessor( i )] - u[i])/subMesh.template getSpaceStepsProducts< 1, 0 >() > 1.0) {
- //cout << "y = n; x = 0" <<std::endl;
- u[i] = u[subMesh.getCellXSuccessor( i )] - subMesh.template getSpaceStepsProducts< 1, 0 >();
- }
- }
- else if(y == subMesh.getDimensions().y() - 1 && (boundaryCondition & 1) && x == subMesh.getDimensions().x() - 1)
- {
- if((u[subMesh.getCellXPredecessor( i )] - u[i])/subMesh.template getSpaceStepsProducts< 1, 0 >() > 1.0)
- {
- //cout << "y = n; x = n" <<std::endl;
- u[i] = u[subMesh.getCellXPredecessor( i )] - subMesh.template getSpaceStepsProducts< 1, 0 >();
- }
- }
- }*/
-
- /**/
-
-
-/* bool tmp = false;
- for(int i = 0; i < u.getSize(); i++)
- {
- if(u[0]*u[i] <= 0.0)
- tmp=true;
- }
-
-
- if(tmp)
- {}
- else if(boundaryCondition == 4)
- {
- int i;
- for(i = 0; i < u.getSize() - subMesh.getDimensions().x() ; i=subMesh.getCellYSuccessor(i))
- {
- int j;
- for(j = i; j < subMesh.getDimensions().x() - 1; j=subMesh.getCellXSuccessor(j))
- {
- u[j] = u[i];
- }
- u[j] = u[i];
- }
- int j;
- for(j = i; j < subMesh.getDimensions().x() - 1; j=subMesh.getCellXSuccessor(j))
- {
- u[j] = u[i];
- }
- u[j] = u[i];
- }
- else if(boundaryCondition == 8)
- {
- int i;
- for(i = 0; i < subMesh.getDimensions().x() - 1; i=subMesh.getCellXSuccessor(i))
- {
- int j;
- for(j = i; j < u.getSize() - subMesh.getDimensions().x(); j=subMesh.getCellYSuccessor(j))
- {
- u[j] = u[i];
- }
- u[j] = u[i];
- }
- int j;
- for(j = i; j < u.getSize() - subMesh.getDimensions().x(); j=subMesh.getCellYSuccessor(j))
- {
- u[j] = u[i];
- }
- u[j] = u[i];
-
- }
- else if(boundaryCondition == 2)
- {
- int i;
- for(i = subMesh.getDimensions().x() - 1; i < u.getSize() - subMesh.getDimensions().x() ; i=subMesh.getCellYSuccessor(i))
- {
- int j;
- for(j = i; j > (i-1)*subMesh.getDimensions().x(); j=subMesh.getCellXPredecessor(j))
- {
- u[j] = u[i];
- }
- u[j] = u[i];
- }
- int j;
- for(j = i; j > (i-1)*subMesh.getDimensions().x(); j=subMesh.getCellXPredecessor(j))
- {
- u[j] = u[i];
- }
- u[j] = u[i];
- }
- else if(boundaryCondition == 1)
- {
- int i;
- for(i = (subMesh.getDimensions().y() - 1)*subMesh.getDimensions().x(); i < u.getSize() - 1; i=subMesh.getCellXSuccessor(i))
- {
- int j;
- for(j = i; j >=subMesh.getDimensions().x(); j=subMesh.getCellYPredecessor(j))
- {
- u[j] = u[i];
- }
- u[j] = u[i];
- }
- int j;
- for(j = i; j >=subMesh.getDimensions().x(); j=subMesh.getCellYPredecessor(j))
- {
- u[j] = u[i];
- }
- u[j] = u[i];
- }
-*/
- /**/
-
-
-
- bool tmp = false;
- for(int i = 0; i < u.getSize(); i++)
- {
- if(u[0]*u[i] <= 0.0)
- tmp=true;
- int centerGID = (this->n*(subGridID / this->gridRows)+ (this->n >> 1))*(this->n*this->gridCols) + this->n*(subGridID % this->gridRows) + (this->n >> 1);
- if(this->unusedCell[centerGID] == 0 || boundaryCondition == 0)
- tmp = true;
- }
- //if(this->currentStep + 3 < getSubgridValue(subGridID))
- //tmp = true;
-
-
- double value = sign(u[0]) * u.absMax();
-
- if(tmp)
- {}
-
-
- //north - 1, east - 2, west - 4, south - 8
- else if(boundaryCondition == 4)
- {
- for(int i = 0; i < this->n; i++)
- for(int j = 1;j < this->n; j++)
- //if(fabs(u[i*this->n + j]) < fabs(u[i*this->n]))
- u[i*this->n + j] = value;// u[i*this->n];
- }
- else if(boundaryCondition == 2)
- {
- for(int i = 0; i < this->n; i++)
- for(int j =0 ;j < this->n -1; j++)
- //if(fabs(u[i*this->n + j]) < fabs(u[(i+1)*this->n - 1]))
- u[i*this->n + j] = value;// u[(i+1)*this->n - 1];
- }
- else if(boundaryCondition == 1)
- {
- for(int j = 0; j < this->n; j++)
- for(int i = 0;i < this->n - 1; i++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j + this->n*(this->n - 1)]))
- u[i*this->n + j] = value;// u[j + this->n*(this->n - 1)];
- }
- else if(boundaryCondition == 8)
- {
- for(int j = 0; j < this->n; j++)
- for(int i = 1;i < this->n; i++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j]))
- u[i*this->n + j] = value;// u[j];
- }
-
-/*
-
- else if(boundaryCondition == 5)
- {
- for(int i = 0; i < this->n - 1; i++)
- for(int j = 1;j < this->n; j++)
- //if(fabs(u[i*this->n + j]) < fabs(u[i*this->n]))
- u[i*this->n + j] = value;// u[i*this->n];
- }
- else if(boundaryCondition == 10)
- {
- for(int i = 1; i < this->n; i++)
- for(int j =0 ;j < this->n -1; j++)
- //if(fabs(u[i*this->n + j]) < fabs(u[(i+1)*this->n - 1]))
- u[i*this->n + j] = value;// u[(i+1)*this->n - 1];
- }
- else if(boundaryCondition == 3)
- {
- for(int j = 0; j < this->n - 1; j++)
- for(int i = 0;i < this->n - 1; i++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j + this->n*(this->n - 1)]))
- u[i*this->n + j] = value;// u[j + this->n*(this->n - 1)];
- }
- else if(boundaryCondition == 12)
- {
- for(int j = 1; j < this->n; j++)
- for(int i = 1;i < this->n; i++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j]))
- u[i*this->n + j] = value;// u[j];
- }
-*/
-
-
- /**/
-
- /*if (u.max() > 0.0)
- this->stopTime *=(double) this->gridCols;*/
-
-
- double time = 0.0;
- double currentTau = this->tau0;
- double finalTime = this->stopTime;// + 3.0*(u.max() - u.min());
- if( time + currentTau > finalTime ) currentTau = finalTime - time;
-
- double maxResidue( 1.0 );
- //double lastResidue( 10000.0 );
- tnlGridEntity<MeshType, 2, tnlGridEntityNoStencilStorage > Entity(subMesh);
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- while( time < finalTime /*|| maxResidue > subMesh.template getSpaceStepsProducts< 1, 0 >()*/)
- {
- /****
- * Compute the RHS
- */
-
- for( int i = 0; i < fu.getSize(); i ++ )
- {
- Entity.setCoordinates(Containers::StaticVector<2,int>(i % subMesh.getDimensions().x(),i / subMesh.getDimensions().x()));
- Entity.refresh();
- neighborEntities.refresh(subMesh,Entity.getIndex());
- fu[ i ] = schemeHost.getValue( this->subMesh, i, Containers::StaticVector<2,int>(i % subMesh.getDimensions().x(),i / subMesh.getDimensions().x()), u, time, boundaryCondition,neighborEntities);
- }
- maxResidue = fu. absMax();
-
-
- if( this -> cflCondition * maxResidue != 0.0)
- currentTau = this -> cflCondition / maxResidue;
-
- /* if (maxResidue < 0.05)
- std::cout << "Max < 0.05" <<std::endl;*/
- if(currentTau > 1.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >())
- {
- //cout << currentTau << " >= " << 2.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >() <<std::endl;
- currentTau = 1.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >();
- }
- /*if(maxResidue > lastResidue)
- currentTau *=(1.0/10.0);*/
-
-
- if( time + currentTau > finalTime ) currentTau = finalTime - time;
-// for( int i = 0; i < fu.getSize(); i ++ )
-// {
-// //cout << "Too big RHS! i = " << i << ", fu = " << fu[i] << ", u = " << u[i] <<std::endl;
-// if((u[i]+currentTau * fu[ i ])*u[i] < 0.0 && fu[i] != 0.0 && u[i] != 0.0 )
-// currentTau = fabs(u[i]/(2.0*fu[i]));
-//
-// }
-
-
- for( int i = 0; i < fu.getSize(); i ++ )
- {
- double add = u[i] + currentTau * fu[ i ];
- //if( fabs(u[i]) < fabs(add) or (this->subgridValues[subGridID] == this->currentStep +4) )
- u[ i ] = add;
- }
- time += currentTau;
-
- //cout << '\r' << flush;
- //cout << maxResidue << " " << currentTau << " @ " << time << flush;
- //lastResidue = maxResidue;
- }
- //cout << "Time: " << time << ", Res: " << maxResidue <<endl;
- /*if (u.max() > 0.0)
- this->stopTime /=(double) this->gridCols;*/
-
- VectorType solution;
- solution.setLike(u);
- for( int i = 0; i < u.getSize(); i ++ )
- {
- solution[i]=u[i];
- }
- return solution;
-}
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getSubgridCUDA2D( const int i ,tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >* caller, double* a)
-{
- //int j = threadIdx.x + threadIdx.y * blockDim.x;
- int th = (blockIdx.y) * caller->n*caller->n*caller->gridCols
- + (blockIdx.x) * caller->n
- + threadIdx.y * caller->n*caller->gridCols
- + threadIdx.x;
- //printf("i= %d,j= %d,th= %d\n",i,j,th);
- *a = caller->work_u_cuda[th];
- //printf("Hi %f \n", *a);
- //return ret;
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::updateSubgridCUDA2D( const int i ,tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >* caller, double* a)
-{
-// int j = threadIdx.x + threadIdx.y * blockDim.x;
- int index = (blockIdx.y) * caller->n*caller->n*caller->gridCols
- + (blockIdx.x) * caller->n
- + threadIdx.y * caller->n*caller->gridCols
- + threadIdx.x;
-
- if( (fabs(caller->work_u_cuda[index]) > fabs(*a)) || (caller->unusedCell_cuda[index] == 1) )
- {
- caller->work_u_cuda[index] = *a;
- caller->unusedCell_cuda[index] = 0;
-
- }
-
- *a = caller->work_u_cuda[index];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::insertSubgridCUDA2D( double u, const int i )
-{
-
-
-// int j = threadIdx.x + threadIdx.y * blockDim.x;
- //printf("j = %d, u = %f\n", j,u);
-
- int index = (blockIdx.y)*this->n*this->n*this->gridCols
- + (blockIdx.x)*this->n
- + threadIdx.y*this->n*this->gridCols
- + threadIdx.x;
-
- //printf("i= %d,j= %d,index= %d\n",i,j,index);
- if( (fabs(this->work_u_cuda[index]) > fabs(u)) || (this->unusedCell_cuda[index] == 1) )
- {
- this->work_u_cuda[index] = u;
- this->unusedCell_cuda[index] = 0;
-
- }
-
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::runSubgridCUDA2D( int boundaryCondition, double* u, int subGridID)
-{
-
- __shared__ int tmp;
- __shared__ double value;
- //double tmpRes = 0.0;
- volatile double* sharedTau = &u[blockDim.x*blockDim.y];
- volatile double* absVal = &u[2*blockDim.x*blockDim.y];
- int i = threadIdx.x;
- int j = threadIdx.y;
- int l = threadIdx.y * blockDim.x + threadIdx.x;
- bool computeFU = !((i == 0 && (boundaryCondition & 4)) or
- (i == blockDim.x - 1 && (boundaryCondition & 2)) or
- (j == 0 && (boundaryCondition & 8)) or
- (j == blockDim.y - 1 && (boundaryCondition & 1)));
-
- if(l == 0)
- {
- tmp = 0;
- int centerGID = (blockDim.y*blockIdx.y + (blockDim.y>>1))*(blockDim.x*gridDim.x) + blockDim.x*blockIdx.x + (blockDim.x>>1);
- if(this->unusedCell_cuda[centerGID] == 0 || boundaryCondition == 0)
- tmp = 1;
- }
- __syncthreads();
-
- /*if(!tmp && (u[0]*u[l] <= 0.0))
- atomicMax( &tmp, 1);*/
-
- __syncthreads();
- if(tmp !=1)
- {
-// if(computeFU)
-// absVal[l]=0.0;
-// else
-// absVal[l] = fabs(u[l]);
-//
-// __syncthreads();
-//
-// if((blockDim.x == 16) && (l < 128)) absVal[l] = Max(absVal[l],absVal[l+128]);
-// __syncthreads();
-// if((blockDim.x == 16) && (l < 64)) absVal[l] = Max(absVal[l],absVal[l+64]);
-// __syncthreads();
-// if(l < 32) absVal[l] = Max(absVal[l],absVal[l+32]);
-// if(l < 16) absVal[l] = Max(absVal[l],absVal[l+16]);
-// if(l < 8) absVal[l] = Max(absVal[l],absVal[l+8]);
-// if(l < 4) absVal[l] = Max(absVal[l],absVal[l+4]);
-// if(l < 2) absVal[l] = Max(absVal[l],absVal[l+2]);
-// if(l < 1) value = sign(u[0])*Max(absVal[l],absVal[l+1]);
-// __syncthreads();
-//
-// if(computeFU)
-// u[l] = value;
- if(computeFU)
- {
- if(boundaryCondition == 4)
- u[l] = u[threadIdx.y * blockDim.x] + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(threadIdx.x) ;//+ 2*sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(threadIdx.x+this->n);
- else if(boundaryCondition == 2)
- u[l] = u[threadIdx.y * blockDim.x + blockDim.x - 1] + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(this->n - 1 - threadIdx.x);//+ 2*sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(blockDim.x - threadIdx.x - 1+this->n);
- else if(boundaryCondition == 8)
- u[l] = u[threadIdx.x] + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(threadIdx.y) ;//+ 2*sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(threadIdx.y+this->n);
- else if(boundaryCondition == 1)
- u[l] = u[(blockDim.y - 1)* blockDim.x + threadIdx.x] + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(this->n - 1 - threadIdx.y) ;//+ sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0 >()*(blockDim.y - threadIdx.y - 1 +this->n);
- }
- }
-
- double time = 0.0;
- __shared__ double currentTau;
- double cfl = this->cflCondition;
- double fu = 0.0;
-// if(threadIdx.x * threadIdx.y == 0)
-// {
-// currentTau = finalTime;
-// }
- double finalTime = this->stopTime;
- __syncthreads();
-// if( time + currentTau > finalTime ) currentTau = finalTime - time;
-
- tnlGridEntity<MeshType, 2, tnlGridEntityNoStencilStorage > Entity(subMesh);
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 2, tnlGridEntityNoStencilStorage >,2> neighborEntities(Entity);
- Entity.setCoordinates(Containers::StaticVector<2,int>(i,j));
- Entity.refresh();
- neighborEntities.refresh(subMesh,Entity.getIndex());
-
-
- while( time < finalTime )
- {
- if(computeFU)
- fu = schemeHost.getValueDev( this->subMesh, l, Containers::StaticVector<2,int>(i,j)/*this->subMesh.getCellCoordinates(l)*/, u, time, boundaryCondition, neighborEntities);
-
- sharedTau[l]=abs(cfl/fu);
-
- if(l == 0)
- {
- if(sharedTau[0] > 1.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >()) sharedTau[0] = 1.0 * this->subMesh.template getSpaceStepsProducts< 1, 0 >();
- }
- else if(l == blockDim.x*blockDim.y - 1)
- if( time + sharedTau[l] > finalTime ) sharedTau[l] = finalTime - time;
-
-
-// if( (sign(u[l]+sharedTau[l]*fu) != sign(u[l])) && fu != 0.0 && fu != -0.0)
-// {
-// printf("orig: %10f", sharedTau[l]);
-// sharedTau[l]=abs(u[l]/(1.1*fu)) ;
-// printf(" new: %10f\n", sharedTau[l]);
-// }
-
-
-
- if((blockDim.x == 16) && (l < 128)) sharedTau[l] = Min(sharedTau[l],sharedTau[l+128]);
- __syncthreads();
- if((blockDim.x == 16) && (l < 64)) sharedTau[l] = Min(sharedTau[l],sharedTau[l+64]);
- __syncthreads();
- if(l < 32) sharedTau[l] = Min(sharedTau[l],sharedTau[l+32]);
- if(l < 16) sharedTau[l] = Min(sharedTau[l],sharedTau[l+16]);
- if(l < 8) sharedTau[l] = Min(sharedTau[l],sharedTau[l+8]);
- if(l < 4) sharedTau[l] = Min(sharedTau[l],sharedTau[l+4]);
- if(l < 2) sharedTau[l] = Min(sharedTau[l],sharedTau[l+2]);
- if(l < 1) currentTau = Min(sharedTau[l],sharedTau[l+1]);
- __syncthreads();
-
- u[l] += currentTau * fu;
- time += currentTau;
- }
-
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getOwnerCUDA2D(int i) const
-{
-
- return ((i / (this->gridCols*this->n*this->n))*this->gridCols
- + (i % (this->gridCols*this->n))/this->n);
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getSubgridValueCUDA2D( int i ) const
-{
- return this->subgridValues_cuda[i];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::setSubgridValueCUDA2D(int i, int value)
-{
- this->subgridValues_cuda[i] = value;
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::getBoundaryConditionCUDA2D( int i ) const
-{
- return this->boundaryConditions_cuda[i];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::setBoundaryConditionCUDA2D(int i, int value)
-{
- this->boundaryConditions_cuda[i] = value;
-}
-
-
-
-//north - 1, east - 2, west - 4, south - 8
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void /*tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::*/synchronizeCUDA2D(tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver) //needs fix ---- maybe not anymore --- but frankly: yeah, it does -- aaaa-and maybe fixed now
-{
-
- __shared__ int boundary[4]; // north,east,west,south
- __shared__ int subgridValue;
- __shared__ int newSubgridValue;
-
-
- int gid = (blockDim.y*blockIdx.y + threadIdx.y)*blockDim.x*gridDim.x + blockDim.x*blockIdx.x + threadIdx.x;
- double u = cudaSolver->work_u_cuda[gid];
- double u_cmp;
- int subgridValue_cmp=INT_MAX;
- int boundary_index=0;
-
-
- if(threadIdx.x+threadIdx.y == 0)
- {
- subgridValue = cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x);
- boundary[0] = 0;
- boundary[1] = 0;
- boundary[2] = 0;
- boundary[3] = 0;
- newSubgridValue = 0;
- //printf("%d %d\n", blockDim.x, gridDim.x);
- }
- __syncthreads();
-
-
-
- if( (threadIdx.x == 0 /* && !(cudaSolver->currentStep & 1)*/) ||
- (threadIdx.y == 0 /* && (cudaSolver->currentStep & 1)*/) ||
- (threadIdx.x == blockDim.x - 1 /* && !(cudaSolver->currentStep & 1)*/) ||
- (threadIdx.y == blockDim.y - 1 /* && (cudaSolver->currentStep & 1)*/) )
- {
- if(threadIdx.x == 0 && (blockIdx.x != 0)/* && !(cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid - 1];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x - 1);
- boundary_index = 2;
- }
-
- if(threadIdx.x == blockDim.x - 1 && (blockIdx.x != gridDim.x - 1)/* && !(cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid + 1];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x + 1);
- boundary_index = 1;
- }
-
- __threadfence();
- if((subgridValue == INT_MAX || fabs(u_cmp) + cudaSolver->delta < fabs(u) ) && (subgridValue_cmp != INT_MAX && subgridValue_cmp != -INT_MAX))
- {
- cudaSolver->unusedCell_cuda[gid] = 0;
- atomicMax(&newSubgridValue, INT_MAX);
- atomicMax(&boundary[boundary_index], 1);
- cudaSolver->work_u_cuda[gid] = u_cmp;
- u=u_cmp;
- }
- __threadfence();
- if(threadIdx.y == 0 && (blockIdx.y != 0)/* && (cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid - blockDim.x*gridDim.x];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA2D((blockIdx.y - 1)*gridDim.x + blockIdx.x);
- boundary_index = 3;
- }
- if(threadIdx.y == blockDim.y - 1 && (blockIdx.y != gridDim.y - 1)/* && (cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid + blockDim.x*gridDim.x];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA2D((blockIdx.y + 1)*gridDim.x + blockIdx.x);
- boundary_index = 0;
- }
-
-// __threadfence();
- if((subgridValue == INT_MAX || fabs(u_cmp) + cudaSolver->delta < fabs(u) ) && (subgridValue_cmp != INT_MAX && subgridValue_cmp != -INT_MAX))
- {
- cudaSolver->unusedCell_cuda[gid] = 0;
- atomicMax(&newSubgridValue, INT_MAX);
- atomicMax(&boundary[boundary_index], 1);
- cudaSolver->work_u_cuda[gid] = u_cmp;
- }
- }
- __threadfence();
- __syncthreads();
-
- if(threadIdx.x+threadIdx.y == 0)
- {
- if(subgridValue == INT_MAX && newSubgridValue !=0)
- cudaSolver->setSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x, -INT_MAX);
-
- cudaSolver->setBoundaryConditionCUDA2D(blockIdx.y*gridDim.x + blockIdx.x, boundary[0] +
- 2 * boundary[1] +
- 4 * boundary[2] +
- 8 * boundary[3]);
-
-
- if(blockIdx.x+blockIdx.y ==0)
- {
- cudaSolver->currentStep = cudaSolver->currentStep + 1;
- *(cudaSolver->runcuda) = 0;
- }
-//
-// int stepValue = cudaSolver->currentStep + 4;
-// if( cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x) == -INT_MAX )
-// cudaSolver->setSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x, stepValue);
-//
-// atomicMax((cudaSolver->runcuda),cudaSolver->getBoundaryConditionCUDA2D(blockIdx.y*gridDim.x + blockIdx.x));
- }
-
-
- /*
- //printf("I am not an empty kernel!\n");
- //cout << "Synchronizig..." <<std::endl;
- int tmp1, tmp2;
- int grid1, grid2;
-
- if(cudaSolver->currentStep & 1)
- {
- //printf("I am not an empty kernel! 1\n");
- for(int j = 0; j < cudaSolver->gridRows - 1; j++)
- {
- //printf("I am not an empty kernel! 3\n");
- for (int i = 0; i < cudaSolver->gridCols*cudaSolver->n; i++)
- {
- tmp1 = cudaSolver->gridCols*cudaSolver->n*((cudaSolver->n-1)+j*cudaSolver->n) + i;
- tmp2 = cudaSolver->gridCols*cudaSolver->n*((cudaSolver->n)+j*cudaSolver->n) + i;
- grid1 = cudaSolver->getSubgridValueCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1));
- grid2 = cudaSolver->getSubgridValueCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2));
-
- if ((fabs(cudaSolver->work_u_cuda[tmp1]) < fabs(cudaSolver->work_u_cuda[tmp2]) - cudaSolver->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- //printf("%d %d %d %d \n",tmp1,tmp2,cudaSolver->getOwnerCUDA2D(tmp1),cudaSolver->getOwnerCUDA2D(tmp2));
- cudaSolver->work_u_cuda[tmp2] = cudaSolver->work_u_cuda[tmp1];
- cudaSolver->unusedCell_cuda[tmp2] = 0;
- if(grid2 == INT_MAX)
- {
- cudaSolver->setSubgridValueCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2), -INT_MAX);
- }
- if(! (cudaSolver->getBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2)) & 8) )
- cudaSolver->setBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2), cudaSolver->getBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2))+8);
- }
- else if ((fabs(cudaSolver->work_u_cuda[tmp1]) > fabs(cudaSolver->work_u_cuda[tmp2]) + cudaSolver->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- //printf("%d %d %d %d \n",tmp1,tmp2,cudaSolver->getOwnerCUDA2D(tmp1),cudaSolver->getOwnerCUDA2D(tmp2));
- cudaSolver->work_u_cuda[tmp1] = cudaSolver->work_u_cuda[tmp2];
- cudaSolver->unusedCell_cuda[tmp1] = 0;
- if(grid1 == INT_MAX)
- {
- cudaSolver->setSubgridValueCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1), -INT_MAX);
- }
- if(! (cudaSolver->getBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1)) & 1) )
- cudaSolver->setBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1), cudaSolver->getBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1))+1);
- }
- }
- }
-
- }
- else
- {
- //printf("I am not an empty kernel! 2\n");
- for(int i = 1; i < cudaSolver->gridCols; i++)
- {
- //printf("I am not an empty kernel! 4\n");
- for (int j = 0; j < cudaSolver->gridRows*cudaSolver->n; j++)
- {
-
- tmp1 = cudaSolver->gridCols*cudaSolver->n*j + i*cudaSolver->n - 1;
- tmp2 = cudaSolver->gridCols*cudaSolver->n*j + i*cudaSolver->n ;
- grid1 = cudaSolver->getSubgridValueCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1));
- grid2 = cudaSolver->getSubgridValueCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2));
-
- if ((fabs(cudaSolver->work_u_cuda[tmp1]) < fabs(cudaSolver->work_u_cuda[tmp2]) - cudaSolver->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- //printf("%d %d %d %d \n",tmp1,tmp2,cudaSolver->getOwnerCUDA2D(tmp1),cudaSolver->getOwnerCUDA2D(tmp2));
- cudaSolver->work_u_cuda[tmp2] = cudaSolver->work_u_cuda[tmp1];
- cudaSolver->unusedCell_cuda[tmp2] = 0;
- if(grid2 == INT_MAX)
- {
- cudaSolver->setSubgridValueCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2), -INT_MAX);
- }
- if(! (cudaSolver->getBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2)) & 4) )
- cudaSolver->setBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2), cudaSolver->getBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp2))+4);
- }
- else if ((fabs(cudaSolver->work_u_cuda[tmp1]) > fabs(cudaSolver->work_u_cuda[tmp2]) + cudaSolver->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- //printf("%d %d %d %d \n",tmp1,tmp2,cudaSolver->getOwnerCUDA2D(tmp1),cudaSolver->getOwnerCUDA2D(tmp2));
- cudaSolver->work_u_cuda[tmp1] = cudaSolver->work_u_cuda[tmp2];
- cudaSolver->unusedCell_cuda[tmp1] = 0;
- if(grid1 == INT_MAX)
- {
- cudaSolver->setSubgridValueCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1), -INT_MAX);
- }
- if(! (cudaSolver->getBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1)) & 2) )
- cudaSolver->setBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1), cudaSolver->getBoundaryConditionCUDA2D(cudaSolver->getOwnerCUDA2D(tmp1))+2);
- }
- }
- }
- }
- //printf("I am not an empty kernel! 5 cudaSolver->currentStep : %d \n", cudaSolver->currentStep);
-
- cudaSolver->currentStep = cudaSolver->currentStep + 1;
- int stepValue = cudaSolver->currentStep + 4;
- for (int i = 0; i < cudaSolver->gridRows * cudaSolver->gridCols; i++)
- {
- if( cudaSolver->getSubgridValueCUDA2D(i) == -INT_MAX )
- cudaSolver->setSubgridValueCUDA2D(i, stepValue);
- }
-
- int maxi = 0;
- for(int q=0; q < cudaSolver->gridRows*cudaSolver->gridCols;q++)
- {
- //printf("%d : %d\n", q, cudaSolver->boundaryConditions_cuda[q]);
- maxi=Max(maxi,cudaSolver->getBoundaryConditionCUDA2D(q));
- }
- //printf("I am not an empty kernel! %d\n", maxi);
- *(cudaSolver->runcuda) = (maxi > 0);
- //printf("I am not an empty kernel! 7 %d\n", cudaSolver->boundaryConditions_cuda[0]);
- //cout << "Grid synchronized at step " << (this->currentStep - 1 ) <<std::endl;
-*/
-}
-
-
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void synchronize2CUDA2D(tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver)
-{
-// if(blockIdx.x+blockIdx.y ==0)
-// {
-// cudaSolver->currentStep = cudaSolver->currentStep + 1;
-// *(cudaSolver->runcuda) = 0;
-// }
-
- int stepValue = cudaSolver->currentStep + 4;
- if( cudaSolver->getSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x) == -INT_MAX )
- cudaSolver->setSubgridValueCUDA2D(blockIdx.y*gridDim.x + blockIdx.x, stepValue);
-
- atomicMax((cudaSolver->runcuda),cudaSolver->getBoundaryConditionCUDA2D(blockIdx.y*gridDim.x + blockIdx.x));
-}
-
-
-
-
-
-
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void /*tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::*/initCUDA2D( tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver, double* ptr , int* ptr2, int* ptr3)
-{
- //cout << "Initializating solver..." <<std::endl;
- //const String& meshLocation = parameters.getParameter <String>("mesh");
- //this->mesh_cuda.load( meshLocation );
-
- //this->n_cuda = parameters.getParameter <int>("subgrid-size");
- //cout << "Setting N << this->n_cuda <<std::endl;
-
- //this->subMesh_cuda.setDimensions( this->n_cuda, this->n_cuda );
- //this->subMesh_cuda.setDomain( Containers::StaticVector<2,double>(0.0, 0.0),
- //Containers::StaticVector<2,double>(this->mesh_cuda.template getSpaceStepsProducts< 1, 0 >()*(double)(this->n_cuda), this->mesh_cuda.template getSpaceStepsProducts< 0, 1 >()*(double)(this->n_cuda)) );
-
- //this->subMesh_cuda.save("submesh.tnl");
-
-// const String& initialCondition = parameters.getParameter <String>("initial-condition");
-// this->u0.load( initialCondition );
-
- //cout << this->mesh.getCellCenter(0) <<std::endl;
-
- //this->delta_cuda = parameters.getParameter <double>("delta");
- //this->delta_cuda *= this->mesh_cuda.template getSpaceStepsProducts< 1, 0 >()*this->mesh_cuda.template getSpaceStepsProducts< 0, 1 >();
-
- //cout << "Setting delta to " << this->delta <<std::endl;
-
- //this->tau0_cuda = parameters.getParameter <double>("initial-tau");
- //cout << "Setting initial tau to " << this->tau0_cuda <<std::endl;
- //this->stopTime_cuda = parameters.getParameter <double>("stop-time");
-
- //this->cflCondition_cuda = parameters.getParameter <double>("cfl-condition");
- //this -> cflCondition_cuda *= sqrt(this->mesh_cuda.template getSpaceStepsProducts< 1, 0 >()*this->mesh_cuda.template getSpaceStepsProducts< 0, 1 >());
- //cout << "Setting CFL to " << this->cflCondition <<std::endl;
-////
-////
-
-// this->gridRows_cuda = gridRows;
-// this->gridCols_cuda = gridCols;
-
- cudaSolver->work_u_cuda = ptr;//(double*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*cudaSolver->n*cudaSolver->n*sizeof(double));
- cudaSolver->unusedCell_cuda = ptr3;//(int*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*cudaSolver->n*cudaSolver->n*sizeof(int));
- cudaSolver->subgridValues_cuda =(int*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*sizeof(int));
- cudaSolver->boundaryConditions_cuda =(int*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*sizeof(int));
- cudaSolver->runcuda = ptr2;//(bool*)malloc(sizeof(bool));
- *(cudaSolver->runcuda) = 1;
- cudaSolver->currentStep = 1;
- //cudaMemcpy(ptr,&(cudaSolver->work_u_cuda), sizeof(double*),cudaMemcpyDeviceToHost);
- //ptr = cudaSolver->work_u_cuda;
- printf("GPU memory allocated.\n");
-
- for(int i = 0; i < cudaSolver->gridCols*cudaSolver->gridRows; i++)
- {
- cudaSolver->subgridValues_cuda[i] = INT_MAX;
- cudaSolver->boundaryConditions_cuda[i] = 0;
- }
-
- /*for(long int j = 0; j < cudaSolver->n*cudaSolver->n*cudaSolver->gridCols*cudaSolver->gridRows; j++)
- {
- printf("%d\n",j);
- cudaSolver->unusedCell_cuda[ j] = 1;
- }*/
- printf("GPU memory initialized.\n");
-
-
- //cudaSolver->work_u_cuda[50] = 32.153438;
-////
-////
- //stretchGrid();
- //this->stopTime_cuda /= (double)(this->gridCols_cuda);
- //this->stopTime_cuda *= (1.0+1.0/((double)(this->n_cuda) - 1.0));
- //cout << "Setting stopping time to " << this->stopTime <<std::endl;
- //this->stopTime_cuda = 1.5*((double)(this->n_cuda))*parameters.getParameter <double>("stop-time")*this->mesh_cuda.template getSpaceStepsProducts< 1, 0 >();
- //cout << "Setting stopping time to " << this->stopTime <<std::endl;
-
- //cout << "Initializating scheme..." <<std::endl;
- //if(!this->schemeDevice.init(parameters))
-// {
- //cerr << "Scheme failed to initialize." <<std::endl;
-// return false;
-// }
- //cout << "Scheme initialized." <<std::endl;
-
- //test();
-
-// this->currentStep_cuda = 1;
- //return true;
-}
-
-
-
-
-//extern __shared__ double array[];
-template< typename SchemeHost, typename SchemeDevice, typename Device >
-__global__
-void /*tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::*/initRunCUDA2D(tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >* caller)
-
-{
-
-
- extern __shared__ double u[];
- //printf("%p\n",caller->work_u_cuda);
-
- int i = blockIdx.y * gridDim.x + blockIdx.x;
- int l = threadIdx.y * blockDim.x + threadIdx.x;
-
- __shared__ int containsCurve;
- if(l == 0)
- containsCurve = 0;
-
- //double a;
- caller->getSubgridCUDA2D(i,caller, &u[l]);
- //printf("%f %f\n",a , u[l]);
- //u[l] = a;
- //printf("Hi %f \n", u[l]);
- __syncthreads();
- //printf("hurewrwr %f \n", u[l]);
- if(u[0] * u[l] <= 0.0)
- {
- //printf("contains %d \n",i);
- atomicMax( &containsCurve, 1);
- }
-
- __syncthreads();
- //printf("hu");
- //printf("%d : %f\n", l, u[l]);
- if(containsCurve == 1)
- {
- //printf("have curve \n");
- caller->runSubgridCUDA2D(0,u,i);
- //printf("%d : %f\n", l, u[l]);
- __syncthreads();
- caller->insertSubgridCUDA2D(u[l],i);
- __syncthreads();
- if(l == 0)
- caller->setSubgridValueCUDA2D(i, 4);
- }
-
-
-}
-
-
-
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device >
-__global__
-void /*tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int>::*/runCUDA2D(tnlParallelEikonalSolver<2,SchemeHost, SchemeDevice, Device, double, int >* caller)
-{
- extern __shared__ double u[];
- int i = blockIdx.y * gridDim.x + blockIdx.x;
- int l = threadIdx.y * blockDim.x + threadIdx.x;
- int bound = caller->getBoundaryConditionCUDA2D(i);
-
- if(caller->getSubgridValueCUDA2D(i) != INT_MAX && bound != 0 && caller->getSubgridValueCUDA2D(i) > 0)
- {
- caller->getSubgridCUDA2D(i,caller, &u[l]);
-
- //if(l == 0)
- //printf("i = %d, bound = %d\n",i,caller->getSubgridValueCUDA2D(i));
- if(caller->getSubgridValueCUDA2D(i) == caller->currentStep+4)
- {
- if(bound & 1)
- {
- caller->runSubgridCUDA2D(1,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 2 )
- {
- caller->runSubgridCUDA2D(2,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 4)
- {
- caller->runSubgridCUDA2D(4,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 8)
- {
- caller->runSubgridCUDA2D(8,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
-
-
-
-
-
- if( ((bound & 3 )))
- {
- caller->runSubgridCUDA2D(3,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound & 5 )))
- {
- caller->runSubgridCUDA2D(5,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound & 10 )))
- {
- caller->runSubgridCUDA2D(10,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( (bound & 12 ))
- {
- caller->runSubgridCUDA2D(12,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
-
-
-
-
-
- }
-
-
- else
- {
-
-
-
-
-
-
-
-
-
- if( ((bound == 2)))
- {
- caller->runSubgridCUDA2D(2,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound == 1) ))
- {
- caller->runSubgridCUDA2D(1,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound == 8) ))
- {
- caller->runSubgridCUDA2D(8,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( (bound == 4))
- {
- caller->runSubgridCUDA2D(4,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
-
-
-
-
-
-
-
-
-
-
- if( ((bound & 3) ))
- {
- caller->runSubgridCUDA2D(3,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound & 5) ))
- {
- caller->runSubgridCUDA2D(5,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound & 10) ))
- {
- caller->runSubgridCUDA2D(10,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
- if( (bound & 12) )
- {
- caller->runSubgridCUDA2D(12,u,i);
- //__syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }
-
-
-
-
-
-
-
-
-
-
-
-
- }
- /*if( bound )
- {
- caller->runSubgridCUDA2D(15,u,i);
- __syncthreads();
- //caller->insertSubgridCUDA2D(u[l],i);
- //__syncthreads();
- //caller->getSubgridCUDA2D(i,caller, &u[l]);
- caller->updateSubgridCUDA2D(i,caller, &u[l]);
- __syncthreads();
- }*/
-
- if(l==0)
- {
- caller->setBoundaryConditionCUDA2D(i, 0);
- caller->setSubgridValueCUDA2D(i, caller->getSubgridValueCUDA2D(i) - 1 );
- }
-
-
- }
-
-
-
-}
-
-#endif /*HAVE_CUDA*/
-
-#endif /* TNLPARALLELEIKONALSOLVER2D_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver3D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver3D_impl.h
deleted file mode 100644
index dc3fd54679bc5c5b6be44841e56e1c32167b5226..0000000000000000000000000000000000000000
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi-parallel/tnlParallelEikonalSolver3D_impl.h
+++ /dev/null
@@ -1,1706 +0,0 @@
-/***************************************************************************
- tnlParallelEikonalSolver2D_impl.h - description
- -------------------
- begin : Nov 28 , 2014
- copyright : (C) 2014 by Tomas Sobotik
- ***************************************************************************/
-
-/***************************************************************************
- * *
- * This program is free software; you can redistribute it and/or modify *
- * it under the terms of the GNU General Public License as published by *
- * the Free Software Foundation; either version 2 of the License, or *
- * (at your option) any later version. *
- * *
- ***************************************************************************/
-
-#ifndef TNLPARALLELEIKONALSOLVER3D_IMPL_H_
-#define TNLPARALLELEIKONALSOLVER3D_IMPL_H_
-
-
-#include "tnlParallelEikonalSolver.h"
-#include <core/mfilename.h>
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::tnlParallelEikonalSolver()
-{
- cout << "a" <<std::endl;
- this->device = TNL::Devices::HostDevice; /////////////// tnlCuda Device --- vypocet na GPU, TNL::Devices::HostDevice --- vypocet na CPU
-
-#ifdef HAVE_CUDA
- if(this->device == tnlCudaDevice)
- {
- run_host = 1;
- }
-#endif
-
- cout << "b" <<std::endl;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::test()
-{
-/*
- for(int i =0; i < this->subgridValues.getSize(); i++ )
- {
- insertSubgrid(getSubgrid(i), i);
- }
-*/
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-
-bool tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::init( const Config::ParameterContainer& parameters )
-{
- cout << "Initializating solver..." <<std::endl;
- const String& meshLocation = parameters.getParameter <String>("mesh");
- this->mesh.load( meshLocation );
-
- this->n = parameters.getParameter <int>("subgrid-size");
- cout << "Setting N to " << this->n <<std::endl;
-
- this->subMesh.setDimensions( this->n, this->n, this->n );
- this->subMesh.setDomain( Containers::StaticVector<3,double>(0.0, 0.0, 0.0),
- Containers::StaticVector<3,double>(mesh.template getSpaceStepsProducts< 1, 0, 0 >()*(double)(this->n), mesh.template getSpaceStepsProducts< 0, 1, 0 >()*(double)(this->n),mesh.template getSpaceStepsProducts< 0, 0, 1 >()*(double)(this->n)) );
-
- this->subMesh.save("submesh.tnl");
-
- const String& initialCondition = parameters.getParameter <String>("initial-condition");
- this->u0.load( initialCondition );
-
- //cout << this->mesh.getCellCenter(0) <<std::endl;
-
- this->delta = parameters.getParameter <double>("delta");
- this->delta *= mesh.template getSpaceStepsProducts< 1, 0, 0 >()*mesh.template getSpaceStepsProducts< 0, 1, 0 >();
-
- cout << "Setting delta to " << this->delta <<std::endl;
-
- this->tau0 = parameters.getParameter <double>("initial-tau");
- cout << "Setting initial tau to " << this->tau0 <<std::endl;
- this->stopTime = parameters.getParameter <double>("stop-time");
-
- this->cflCondition = parameters.getParameter <double>("cfl-condition");
- this -> cflCondition *= sqrt(mesh.template getSpaceStepsProducts< 1, 0, 0 >()*mesh.template getSpaceStepsProducts< 0, 1, 0 >());
- cout << "Setting CFL to " << this->cflCondition <<std::endl;
-
- stretchGrid();
- this->stopTime /= (double)(this->gridCols);
- this->stopTime *= (1.0+1.0/((double)(this->n) - 2.0));
- cout << "Setting stopping time to " << this->stopTime <<std::endl;
- //this->stopTime = 1.5*((double)(this->n))*parameters.getParameter <double>("stop-time")*mesh.template getSpaceStepsProducts< 1, 0, 0 >();
- //cout << "Setting stopping time to " << this->stopTime <<std::endl;
-
- cout << "Initializating scheme..." <<std::endl;
- if(!this->schemeHost.init(parameters))
- {
- cerr << "SchemeHost failed to initialize." <<std::endl;
- return false;
- }
- cout << "Scheme initialized." <<std::endl;
-
- test();
-
- VectorType* tmp = new VectorType[subgridValues.getSize()];
-
-
-#ifdef HAVE_CUDA
-
- if(this->device == tnlCudaDevice)
- {
- /*cout << "Testing... " <<std::endl;
- if(this->device == tnlCudaDevice)
- {
- if( !initCUDA3D(parameters, gridRows, gridCols) )
- return false;
- }*/
- //cout << "s" <<std::endl;
- cudaMalloc(&(this->cudaSolver), sizeof(tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >));
- //cout << "s" <<std::endl;
- cudaMemcpy(this->cudaSolver, this,sizeof(tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >), cudaMemcpyHostToDevice);
- //cout << "s" <<std::endl;
- double** tmpdev = NULL;
- cudaMalloc(&tmpdev, sizeof(double*));
- //double* tmpw;
- cudaMalloc(&(this->tmpw), this->work_u.getSize()*sizeof(double));
- cudaMalloc(&(this->runcuda), sizeof(int));
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- int* tmpUC;
- cudaMalloc(&(tmpUC), this->work_u.getSize()*sizeof(int));
- cudaMemcpy(tmpUC, this->unusedCell.getData(), this->unusedCell.getSize()*sizeof(int), cudaMemcpyHostToDevice);
-
- initCUDA3D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<1,1>>>(this->cudaSolver, (this->tmpw), (this->runcuda),tmpUC);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cout << "s " <<std::endl;
- //cudaMalloc(&(cudaSolver->work_u_cuda), this->work_u.getSize()*sizeof(double));
- double* tmpu = NULL;
-
- cudaMemcpy(&tmpu, tmpdev,sizeof(double*), cudaMemcpyDeviceToHost);
- //printf("%p %p \n",tmpu,tmpw);
- cudaMemcpy((this->tmpw), this->work_u.getData(), this->work_u.getSize()*sizeof(double), cudaMemcpyHostToDevice);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cout << "s "<<std::endl;
-
- }
-#endif
-
- if(this->device == TNL::Devices::HostDevice)
- {
-#ifdef HAVE_OPENMP
-#pragma omp parallel for num_threads(4) schedule(dynamic)
-#endif
- for(int i = 0; i < this->subgridValues.getSize(); i++)
- {
- bool containsCurve = false;
-// cout << "Working on subgrid " << i <<" --- check 1" <<std::endl;
-
- if(! tmp[i].setSize(this->n*this->n*this->n))
- cout << "Could not allocate tmp["<< i <<"] array." <<std::endl;
-// cout << "Working on subgrid " << i <<" --- check 2" <<std::endl;
-
- tmp[i] = getSubgrid(i);
- containsCurve = false;
-// cout << "Working on subgrid " << i <<" --- check 3" <<std::endl;
-
-
- for(int j = 0; j < tmp[i].getSize(); j++)
- {
- if(tmp[i][0]*tmp[i][j] <= 0.0)
- {
- containsCurve = true;
- j=tmp[i].getSize();
-// cout << tmp[i][0] << " " << tmp[i][j] <<std::endl;
- }
-
- }
-// cout << "Working on subgrid " << i <<" --- check 4" <<std::endl;
-
- if(containsCurve)
- {
-// cout << "Computing initial SDF on subgrid " << i << "." <<std::endl;
- tmp[i] = runSubgrid(0, tmp[i] ,i);
- insertSubgrid( tmp[i], i);
- setSubgridValue(i, 4);
-// cout << "Computed initial SDF on subgrid " << i << "." <<std::endl;
- }
- containsCurve = false;
-
- }
-// cout << "CPU: Curve found" <<std::endl;
- }
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
-// cout << "pre 1 kernel" <<std::endl;
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- dim3 threadsPerBlock(this->n, this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows,this->gridLevels);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- initRunCUDA3D<SchemeTypeHost,SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock,2*this->n*this->n*this->n*sizeof(double)>>>(this->cudaSolver);
- cudaDeviceSynchronize();
-// cout << "post 1 kernel" <<std::endl;
-
- }
-#endif
-
-
- this->currentStep = 1;
- if(this->device == TNL::Devices::HostDevice)
- synchronize();
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
- dim3 threadsPerBlock(this->n, this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows,this->gridLevels);
- //double * test = (double*)malloc(this->work_u.getSize()*sizeof(double));
- //cout << test[0] <<" " << test[1] <<" " << test[2] <<" " << test[3] <<std::endl;
- //cudaMemcpy(/*this->work_u.getData()*/ test, (this->tmpw), this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- //cout << this->tmpw << " " << test[0] <<" " << test[1] << " " <<test[2] << " " <<test[3] <<std::endl;
-
- TNL_CHECK_CUDA_DEVICE;
-
- synchronizeCUDA3D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cout << cudaGetErrorString(cudaDeviceSynchronize()) <<std::endl;
- TNL_CHECK_CUDA_DEVICE;
- synchronize2CUDA3D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cout << test[0] << " " <<test[1] <<" " << test[2] << " " <<test[3] <<std::endl;
- //cudaMemcpy(/*this->work_u.getData()*/ test, (this->tmpw), this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- //TNL_CHECK_CUDA_DEVICE;
- //cout << this->tmpw << " " << test[0] << " " <<test[1] << " " <<test[2] <<" " << test[3] <<std::endl;
- //free(test);
-
- }
-
-#endif
- cout << "Solver initialized." <<std::endl;
-
- return true;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::run()
-{
- if(this->device == TNL::Devices::HostDevice)
- {
-
- bool end = false;
- while (/*(this->boundaryConditions.max() > 0 ) ||*/ !end)
- {
- if(this->boundaryConditions.max() == 0 || this->subgridValues.max() < 0)
- end=true;
- else
- end=false;
-#ifdef HAVE_OPENMP
-#pragma omp parallel for num_threads(4) schedule(dynamic)
-#endif
- for(int i = 0; i < this->subgridValues.getSize(); i++)
- {
- VectorType tmp;
- tmp.setSize(this->n*this->n*this->n);
- if(getSubgridValue(i) != INT_MAX)
- {
- //cout << "subMesh: " << i << ", BC: " << getBoundaryCondition(i) <<std::endl;
-
- if(getSubgridValue(i) == currentStep+4)
- {
-
- if(getBoundaryCondition(i) & 1)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(1, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 2)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(2, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 4)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(4, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 8)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(8, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 16)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(16, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- if(getBoundaryCondition(i) & 32)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(32, tmp ,i);
- insertSubgrid( tmp, i);
- this->calculationsCount[i]++;
- }
- }
-
- if( getBoundaryCondition(i) & 19)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(19, tmp ,i);
- insertSubgrid( tmp, i);
- }
- if( getBoundaryCondition(i) & 21)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(21, tmp ,i);
- insertSubgrid( tmp, i);
- }
- if( getBoundaryCondition(i) & 26)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(26, tmp ,i);
- insertSubgrid( tmp, i);
- }
- if( getBoundaryCondition(i) & 28)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(28, tmp ,i);
- insertSubgrid( tmp, i);
- }
-
- if( getBoundaryCondition(i) & 35)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(35, tmp ,i);
- insertSubgrid( tmp, i);
- }
- if( getBoundaryCondition(i) & 37)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(37, tmp ,i);
- insertSubgrid( tmp, i);
- }
- if( getBoundaryCondition(i) & 42)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(42, tmp ,i);
- insertSubgrid( tmp, i);
- }
- if( getBoundaryCondition(i) & 44)
- {
- tmp = getSubgrid(i);
- tmp = runSubgrid(44, tmp ,i);
- insertSubgrid( tmp, i);
- }
-
-
- setBoundaryCondition(i, 0);
- setSubgridValue(i, getSubgridValue(i)-1);
-
- }
- }
- synchronize();
- }
- }
-#ifdef HAVE_CUDA
- else if(this->device == tnlCudaDevice)
- {
- //cout << "fn" <<std::endl;
- bool end_cuda = false;
- dim3 threadsPerBlock(this->n, this->n, this->n);
- dim3 numBlocks(this->gridCols,this->gridRows,this->gridLevels);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cudaMalloc(&runcuda,sizeof(bool));
- //cudaMemcpy(runcuda, &run_host, sizeof(bool), cudaMemcpyHostToDevice);
- //cout << "fn" <<std::endl;
- bool* tmpb;
- //cudaMemcpy(tmpb, &(cudaSolver->runcuda),sizeof(bool*), cudaMemcpyDeviceToHost);
- //cudaDeviceSynchronize();
- //TNL_CHECK_CUDA_DEVICE;
- cudaMemcpy(&(this->run_host),this->runcuda,sizeof(int), cudaMemcpyDeviceToHost);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //cout << "fn" <<std::endl;
- int i = 1;
- time_diff = 0.0;
- while (run_host || !end_cuda)
- {
- cout << "Computing at step "<< i++ <<std::endl;
- if(run_host != 0 )
- end_cuda = true;
- else
- end_cuda = false;
- //cout << "a" <<std::endl;
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- start = std::clock();
- runCUDA3D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock,2*this->n*this->n*this->n*sizeof(double)>>>(this->cudaSolver);
- //cout << "a" <<std::endl;
- cudaDeviceSynchronize();
- time_diff += (std::clock() - start) / (double)(CLOCKS_PER_SEC);
-
- //start = std::clock();
- synchronizeCUDA3D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,threadsPerBlock>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- synchronize2CUDA3D<SchemeTypeHost, SchemeTypeDevice, DeviceType><<<numBlocks,1>>>(this->cudaSolver);
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- //time_diff += (std::clock() - start) / (double)(CLOCKS_PER_SEC);
-
-
- //cout << "a" <<std::endl;
- //run_host = false;
- //cout << "in kernel loop" << run_host <<std::endl;
- //cudaMemcpy(tmpb, &(cudaSolver->runcuda),sizeof(bool*), cudaMemcpyDeviceToHost);
- cudaMemcpy(&run_host, (this->runcuda),sizeof(int), cudaMemcpyDeviceToHost);
- //cout << "in kernel loop" << run_host <<std::endl;
- }
- cout << "Solving time was: " << time_diff <<std::endl;
- //cout << "b" <<std::endl;
-
- //double* tmpu;
- //cudaMemcpy(tmpu, &(cudaSolver->work_u_cuda),sizeof(double*), cudaMemcpyHostToDevice);
- //cudaMemcpy(this->work_u.getData(), tmpu, this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- //cout << this->work_u.getData()[0] <<std::endl;
-
- //double * test = (double*)malloc(this->work_u.getSize()*sizeof(double));
- //cout << test[0] << test[1] << test[2] << test[3] <<std::endl;
- cudaMemcpy(this->work_u.getData()/* test*/, (this->tmpw), this->work_u.getSize()*sizeof(double), cudaMemcpyDeviceToHost);
- //cout << this->tmpw << " " << test[0] << test[1] << test[2] << test[3] <<std::endl;
- //free(test);
-
- cudaDeviceSynchronize();
- }
-#endif
- contractGrid();
- this->u0.save("u-00001.tnl");
- cout << "Maximum number of calculations on one subgrid was " << this->calculationsCount.absMax() <<std::endl;
- cout << "Average number of calculations on one subgrid was " << ( (double) this->calculationsCount.sum() / (double) this->calculationsCount.getSize() ) <<std::endl;
- cout << "Solver finished" <<std::endl;
-
-#ifdef HAVE_CUDA
- if(this->device == tnlCudaDevice)
- {
- cudaFree(this->runcuda);
- cudaFree(this->tmpw);
- cudaFree(this->cudaSolver);
- }
-#endif
-
-}
-
-//north - 1, east - 2, west - 4, south - 8
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::synchronize() //needs fix ---- maybe not anymore --- but frankly: yeah, it does -- aaaa-and maybe fixed now
-{
- cout << "Synchronizig..." <<std::endl;
- int tmp1, tmp2;
- int grid1, grid2;
-
-// if(this->currentStep & 1)
-// {
- for(int j = 0; j < this->gridRows - 1; j++)
- {
- for (int i = 0; i < this->gridCols*this->n; i++)
- {
- for (int k = 0; k < this->gridLevels*this->n; k++)
- {
-// cout << "a" <<std::endl;
- tmp1 = this->gridCols*this->n*((this->n-1)+j*this->n) + i + k*this->gridCols*this->n*this->gridRows*this->n;
-// cout << "b" <<std::endl;
- tmp2 = this->gridCols*this->n*((this->n)+j*this->n) + i + k*this->gridCols*this->n*this->gridRows*this->n;
-// cout << "c" <<std::endl;
- if(tmp1 > work_u.getSize())
- cout << "tmp1: " << tmp1 << " x: " << j <<" y: " << i <<" z: " << k <<std::endl;
- if(tmp2 > work_u.getSize())
- cout << "tmp2: " << tmp2 << " x: " << j <<" y: " << i <<" z: " << k <<std::endl;
- grid1 = getSubgridValue(getOwner(tmp1));
-// cout << "d" <<std::endl;
- grid2 = getSubgridValue(getOwner(tmp2));
-// cout << "e" <<std::endl;
- if(getOwner(tmp1)==getOwner(tmp2))
- cout << "i, j, k" << i << "," << j << "," << k <<std::endl;
- if ((fabs(this->work_u[tmp1]) < fabs(this->work_u[tmp2]) - this->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- this->work_u[tmp2] = this->work_u[tmp1];
-// cout << "f" <<std::endl;
- this->unusedCell[tmp2] = 0;
-// cout << "g" <<std::endl;
- if(grid2 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp2), -INT_MAX);
- }
-// cout << "h" <<std::endl;
- if(! (getBoundaryCondition(getOwner(tmp2)) & 8) )
- setBoundaryCondition(getOwner(tmp2), getBoundaryCondition(getOwner(tmp2))+8);
-// cout << "i" <<std::endl;
- }
- else if ((fabs(this->work_u[tmp1]) > fabs(this->work_u[tmp2]) + this->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- this->work_u[tmp1] = this->work_u[tmp2];
-// cout << "j" <<std::endl;
- this->unusedCell[tmp1] = 0;
-// cout << "k" <<std::endl;
- if(grid1 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp1), -INT_MAX);
- }
-// cout << "l" <<std::endl;
- if(! (getBoundaryCondition(getOwner(tmp1)) & 1) )
- setBoundaryCondition(getOwner(tmp1), getBoundaryCondition(getOwner(tmp1))+1);
-// cout << "m" <<std::endl;
- }
- }
- }
- }
-
-// }
-// else
-// {
-
- cout << "sync 2" <<std::endl;
- for(int i = 1; i < this->gridCols; i++)
- {
- for (int j = 0; j < this->gridRows*this->n; j++)
- {
- for (int k = 0; k < this->gridLevels*this->n; k++)
- {
- tmp1 = this->gridCols*this->n*j + i*this->n - 1 + k*this->gridCols*this->n*this->gridRows*this->n;
- tmp2 = this->gridCols*this->n*j + i*this->n + k*this->gridCols*this->n*this->gridRows*this->n;
- grid1 = getSubgridValue(getOwner(tmp1));
- grid2 = getSubgridValue(getOwner(tmp2));
- if(getOwner(tmp1)==getOwner(tmp2))
- cout << "i, j, k" << i << "," << j << "," << k <<std::endl;
- if ((fabs(this->work_u[tmp1]) < fabs(this->work_u[tmp2]) - this->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- this->work_u[tmp2] = this->work_u[tmp1];
- this->unusedCell[tmp2] = 0;
- if(grid2 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp2), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp2)) & 4) )
- setBoundaryCondition(getOwner(tmp2), getBoundaryCondition(getOwner(tmp2))+4);
- }
- else if ((fabs(this->work_u[tmp1]) > fabs(this->work_u[tmp2]) + this->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- this->work_u[tmp1] = this->work_u[tmp2];
- this->unusedCell[tmp1] = 0;
- if(grid1 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp1), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp1)) & 2) )
- setBoundaryCondition(getOwner(tmp1), getBoundaryCondition(getOwner(tmp1))+2);
- }
- }
- }
- }
-
- cout << "sync 3" <<std::endl;
-
- for(int k = 1; k < this->gridLevels; k++)
- {
- for (int j = 0; j < this->gridRows*this->n; j++)
- {
- for (int i = 0; i < this->gridCols*this->n; i++)
- {
- tmp1 = this->gridCols*this->n*j + i + (k*this->n-1)*this->gridCols*this->n*this->gridRows*this->n;
- tmp2 = this->gridCols*this->n*j + i + k*this->n*this->gridCols*this->n*this->gridRows*this->n;
- grid1 = getSubgridValue(getOwner(tmp1));
- grid2 = getSubgridValue(getOwner(tmp2));
- if(getOwner(tmp1)==getOwner(tmp2))
- cout << "i, j, k" << i << "," << j << "," << k <<std::endl;
- if ((fabs(this->work_u[tmp1]) < fabs(this->work_u[tmp2]) - this->delta || grid2 == INT_MAX || grid2 == -INT_MAX) && (grid1 != INT_MAX && grid1 != -INT_MAX))
- {
- this->work_u[tmp2] = this->work_u[tmp1];
- this->unusedCell[tmp2] = 0;
- if(grid2 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp2), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp2)) & 32) )
- setBoundaryCondition(getOwner(tmp2), getBoundaryCondition(getOwner(tmp2))+32);
- }
- else if ((fabs(this->work_u[tmp1]) > fabs(this->work_u[tmp2]) + this->delta || grid1 == INT_MAX || grid1 == -INT_MAX) && (grid2 != INT_MAX && grid2 != -INT_MAX))
- {
- this->work_u[tmp1] = this->work_u[tmp2];
- this->unusedCell[tmp1] = 0;
- if(grid1 == INT_MAX)
- {
- setSubgridValue(getOwner(tmp1), -INT_MAX);
- }
- if(! (getBoundaryCondition(getOwner(tmp1)) & 16) )
- setBoundaryCondition(getOwner(tmp1), getBoundaryCondition(getOwner(tmp1))+16);
- }
- }
- }
- }
-// }
-
-
-
- this->currentStep++;
- int stepValue = this->currentStep + 4;
- for (int i = 0; i < this->subgridValues.getSize(); i++)
- {
- if( getSubgridValue(i) == -INT_MAX )
- setSubgridValue(i, stepValue);
- }
-
- cout << "Grid synchronized at step " << (this->currentStep - 1 ) <<std::endl;
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::getOwner(int i) const
-{
-
- int j = i % (this->gridCols*this->gridRows*this->n*this->n);
-
- return ( (i / (this->gridCols*this->gridRows*this->n*this->n*this->n))*this->gridCols*this->gridRows
- + (j / (this->gridCols*this->n*this->n))*this->gridCols
- + (j % (this->gridCols*this->n))/this->n);
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::getSubgridValue( int i ) const
-{
- return this->subgridValues[i];
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::setSubgridValue(int i, int value)
-{
- this->subgridValues[i] = value;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-int tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::getBoundaryCondition( int i ) const
-{
- return this->boundaryConditions[i];
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::setBoundaryCondition(int i, int value)
-{
- this->boundaryConditions[i] = value;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::stretchGrid()
-{
- cout << "Stretching grid..." <<std::endl;
-
-
- this->gridCols = ceil( ((double)(this->mesh.getDimensions().x()-1)) / ((double)(this->n-1)) );
- this->gridRows = ceil( ((double)(this->mesh.getDimensions().y()-1)) / ((double)(this->n-1)) );
- this->gridLevels = ceil( ((double)(this->mesh.getDimensions().z()-1)) / ((double)(this->n-1)) );
-
- //this->gridCols = (this->mesh.getDimensions().x()-1) / (this->n-1) ;
- //this->gridRows = (this->mesh.getDimensions().y()-1) / (this->n-1) ;
-
- cout << "Setting gridCols to " << this->gridCols << "." <<std::endl;
- cout << "Setting gridRows to " << this->gridRows << "." <<std::endl;
- cout << "Setting gridLevels to " << this->gridLevels << "." <<std::endl;
-
- this->subgridValues.setSize(this->gridCols*this->gridRows*this->gridLevels);
- this->subgridValues.setValue(0);
- this->boundaryConditions.setSize(this->gridCols*this->gridRows*this->gridLevels);
- this->boundaryConditions.setValue(0);
- this->calculationsCount.setSize(this->gridCols*this->gridRows*this->gridLevels);
- this->calculationsCount.setValue(0);
-
- for(int i = 0; i < this->subgridValues.getSize(); i++ )
- {
- this->subgridValues[i] = INT_MAX;
- this->boundaryConditions[i] = 0;
- }
-
- int levelSize = this->n*this->n*this->gridCols*this->gridRows;
- int stretchedSize = this->n*levelSize*this->gridLevels;
-
- if(!this->work_u.setSize(stretchedSize))
- cerr << "Could not allocate memory for stretched grid." <<std::endl;
- if(!this->unusedCell.setSize(stretchedSize))
- cerr << "Could not allocate memory for supporting stretched grid." <<std::endl;
- int idealStretch =this->mesh.getDimensions().x() + (this->mesh.getDimensions().x()-2)/(this->n-1);
- cout << idealStretch <<std::endl;
-
-
-
-
- for(int i = 0; i < levelSize; i++)
- {
- int diff =(this->n*this->gridCols) - idealStretch ;
-
- int k = i/this->n - i/(this->n*this->gridCols) + this->mesh.getDimensions().x()*(i/(this->n*this->n*this->gridCols)) + (i/(this->n*this->gridCols))*diff;
-
- if(i%(this->n*this->gridCols) - idealStretch >= 0)
- {
- k+= i%(this->n*this->gridCols) - idealStretch +1 ;
- }
-
- if(i/(this->n*this->gridCols) - idealStretch + 1 > 0)
- {
- k+= (i/(this->n*this->gridCols) - idealStretch +1 )* this->mesh.getDimensions().x() ;
- }
-
- for( int j = 0; j<this->n*this->gridLevels; j++)
- {
- this->unusedCell[i+j*levelSize] = 1;
- int l = j/this->n;
-
- if(j - idealStretch >= 0)
- {
- l+= j - idealStretch + 1;
- }
-
- this->work_u[i+j*levelSize] = this->u0[i+(j-l)*mesh.getDimensions().x()*mesh.getDimensions().y()-k];
- }
-
- }
-
-
-
- cout << "Grid stretched." <<std::endl;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::contractGrid()
-{
- cout << "Contracting grid..." <<std::endl;
- int levelSize = this->n*this->n*this->gridCols*this->gridRows;
- int stretchedSize = this->n*levelSize*this->gridLevels;
-
- int idealStretch =this->mesh.getDimensions().x() + (this->mesh.getDimensions().x()-2)/(this->n-1);
- cout << idealStretch <<std::endl;
-
-
- for(int i = 0; i < levelSize; i++)
- {
- int diff =(this->n*this->gridCols) - idealStretch ;
- int k = i/this->n - i/(this->n*this->gridCols) + this->mesh.getDimensions().x()*(i/(this->n*this->n*this->gridCols)) + (i/(this->n*this->gridCols))*diff;
-
- if((i%(this->n*this->gridCols) - idealStretch < 0) && (i/(this->n*this->gridCols) - idealStretch + 1 <= 0) )
- {
- for( int j = 0; j<this->n*this->gridLevels; j++)
- {
- int l = j/this->n;
- if(j - idealStretch < 0)
- this->u0[i+(j-l)*mesh.getDimensions().x()*mesh.getDimensions().y()-k] = this->work_u[i+j*levelSize];
- }
- }
-
- }
-
- cout << "Grid contracted" <<std::endl;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-typename tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::VectorType
-tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::getSubgrid( const int i ) const
-{
-
- VectorType u;
- u.setSize(this->n*this->n*this->n);
-
- int idx, idy, idz;
- idz = i / (gridRows*this->gridCols);
- idy = (i % (this->gridRows*this->gridCols)) / this->gridCols;
- idx = i % (this->gridCols);
-
- for( int j = 0; j < this->n; j++)
- {
- // int index = (i / this->gridCols)*this->n*this->n*this->gridCols + (i % this->gridCols)*this->n + (j/this->n)*this->n*this->gridCols + (j % this->n);
- for( int k = 0; k < this->n; k++)
- {
- for( int l = 0; l < this->n; l++)
- {
- int index = (idz*this->n + l) * this->n*this->n*this->gridCols*this->gridRows
- + (idy) * this->n*this->n*this->gridCols
- + (idx) * this->n
- + k * this->n*this->gridCols
- + j;
-
- u[j + k*this->n + l*this->n*this->n] = this->work_u[ index ];
- }
- }
- }
- return u;
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::insertSubgrid( VectorType u, const int i )
-{
- int idx, idy, idz;
- idz = i / (this->gridRows*this->gridCols);
- idy = (i % (this->gridRows*this->gridCols)) / this->gridCols;
- idx = i % (this->gridCols);
-
- for( int j = 0; j < this->n; j++)
- {
- // int index = (i / this->gridCols)*this->n*this->n*this->gridCols + (i % this->gridCols)*this->n + (j/this->n)*this->n*this->gridCols + (j % this->n);
- for( int k = 0; k < this->n; k++)
- {
- for( int l = 0; l < this->n; l++)
- {
-
- int index = (idz*this->n + l) * this->n*this->n*this->gridCols*this->gridRows
- + (idy) * this->n*this->n*this->gridCols
- + (idx) * this->n
- + k * this->n*this->gridCols
- + j;
-
- //OMP LOCK index
-// cout<< idx << " " << idy << " " << idz << " " << j << " " << k << " " << l << " " << idz << " " << unusedCell.getSize() << " " << u.getSize() << " " << index <<endl;
- if( (fabs(this->work_u[index]) > fabs(u[j + k*this->n + l*this->n*this->n])) || (this->unusedCell[index] == 1) )
- {
- this->work_u[index] = u[j + k*this->n + l*this->n*this->n];
- this->unusedCell[index] = 0;
- }
- //OMP UNLOCK index
- }
- }
- }
-}
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-typename tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::VectorType
-tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::runSubgrid( int boundaryCondition, VectorType u, int subGridID)
-{
-
- VectorType fu;
-
- fu.setLike(u);
- fu.setValue( 0.0 );
-
-
- bool tmp = false;
- for(int i = 0; i < u.getSize(); i++)
- {
- if(u[0]*u[i] <= 0.0)
- tmp=true;
- }
- int idx,idy,idz;
- idz = subGridID / (this->gridRows*this->gridCols);
- idy = (subGridID % (this->gridRows*this->gridCols)) / this->gridCols;
- idx = subGridID % (this->gridCols);
- int centerGID = (this->n*idy + (this->n>>1) )*(this->n*this->gridCols) + this->n*idx + (this->n>>1)
- + ((this->n>>1)+this->n*idz)*this->n*this->n*this->gridRows*this->gridCols;
- if(this->unusedCell[centerGID] == 0 || boundaryCondition == 0)
- tmp = true;
- //if(this->currentStep + 3 < getSubgridValue(subGridID))
- //tmp = true;
-
-
- double value = sign(u[0]) * u.absMax();
-
- if(tmp)
- {}
-
-
- //north - 1, east - 2, west - 4, south - 8
- else if(boundaryCondition == 4)
- {
- for(int i = 0; i < this->n; i++)
- for(int j = 1;j < this->n; j++)
- for(int k = 0;k < this->n; k++)
- //if(fabs(u[i*this->n + j]) < fabs(u[i*this->n]))
- u[k*this->n*this->n + i*this->n + j] = value;// u[i*this->n];
- }
- else if(boundaryCondition == 2)
- {
- for(int i = 0; i < this->n; i++)
- for(int j =0 ;j < this->n -1; j++)
- for(int k = 0;k < this->n; k++)
- //if(fabs(u[i*this->n + j]) < fabs(u[(i+1)*this->n - 1]))
- u[k*this->n*this->n + i*this->n + j] = value;// u[(i+1)*this->n - 1];
- }
- else if(boundaryCondition == 1)
- {
- for(int j = 0; j < this->n; j++)
- for(int i = 0;i < this->n - 1; i++)
- for(int k = 0;k < this->n; k++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j + this->n*(this->n - 1)]))
- u[k*this->n*this->n + i*this->n + j] = value;// u[j + this->n*(this->n - 1)];
- }
- else if(boundaryCondition == 8)
- {
- for(int j = 0; j < this->n; j++)
- for(int i = 1;i < this->n; i++)
- for(int k = 0;k < this->n; k++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j]))
- u[k*this->n*this->n + i*this->n + j] = value;// u[j];
- }
- else if(boundaryCondition == 16)
- {
- for(int j = 0; j < this->n; j++)
- for(int i = 0;i < this->n ; i++)
- for(int k = 0;k < this->n-1; k++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j + this->n*(this->n - 1)]))
- u[k*this->n*this->n + i*this->n + j] = value;// u[j + this->n*(this->n - 1)];
- }
- else if(boundaryCondition == 32)
- {
- for(int j = 0; j < this->n; j++)
- for(int i = 0;i < this->n; i++)
- for(int k = 1;k < this->n; k++)
- //if(fabs(u[i*this->n + j]) < fabs(u[j]))
- u[k*this->n*this->n + i*this->n + j] = value;// u[j];
- }
-
-
- double time = 0.0;
- double currentTau = this->tau0;
- double finalTime = this->stopTime;// + 3.0*(u.max() - u.min());
- if(boundaryCondition == 0) finalTime *= 2.0;
- if( time + currentTau > finalTime ) currentTau = finalTime - time;
-
- double maxResidue( 1.0 );
- //double lastResidue( 10000.0 );
- tnlGridEntity<MeshType, 3, tnlGridEntityNoStencilStorage > Entity(subMesh);
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage >,3> neighborEntities(Entity);
- while( time < finalTime /*|| maxResidue > subMesh.template getSpaceStepsProducts< 1, 0, 0 >()*/)
- {
- /****
- * Compute the RHS
- */
-
- for( int i = 0; i < fu.getSize(); i ++ )
- {
-// std::cout << "i: " << i << ", time: " << time <<endl;
- Containers::StaticVector<3,int> coords(i % subMesh.getDimensions().x(),
- (i % (subMesh.getDimensions().x()*subMesh.getDimensions().y())) / subMesh.getDimensions().x(),
- i / (subMesh.getDimensions().x()*subMesh.getDimensions().y()));
-// cout << "b " << i << " " << i % subMesh.getDimensions().x() << " " << (i % (subMesh.getDimensions().x()*subMesh.getDimensions().y())) << " " << (i % subMesh.getDimensions().x()*subMesh.getDimensions().y()) / subMesh.getDimensions().x() << " " << subMesh.getDimensions().x()*subMesh.getDimensions().y() << " " <<endl;
- Entity.setCoordinates(coords);
-// cout <<"c" << coords <<std::endl;
- Entity.refresh();
-// cout << "d" <<endl;
- neighborEntities.refresh(subMesh,Entity.getIndex());
-// cout << "e" <<endl;
- fu[ i ] = schemeHost.getValue( this->subMesh, i, coords,u, time, boundaryCondition, neighborEntities );
-// std::cout << "f" <<endl;
- }
- maxResidue = fu. absMax();
-
-
- if( this -> cflCondition * maxResidue != 0.0)
- currentTau = this -> cflCondition / maxResidue;
-
- /* if (maxResidue < 0.05)
- std::cout << "Max < 0.05" <<std::endl;*/
- if(currentTau > 0.5 * this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >())
- currentTau = 0.5 * this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >();
- /*if(maxResidue > lastResidue)
- currentTau *=(1.0/10.0);*/
-
-
- if( time + currentTau > finalTime ) currentTau = finalTime - time;
-// for( int i = 0; i < fu.getSize(); i ++ )
-// {
-// //cout << "Too big RHS! i = " << i << ", fu = " << fu[i] << ", u = " << u[i] <<std::endl;
-// if((u[i]+currentTau * fu[ i ])*u[i] < 0.0 && fu[i] != 0.0 && u[i] != 0.0 )
-// currentTau = fabs(u[i]/(2.0*fu[i]));
-//
-// }
-
-
- for( int i = 0; i < fu.getSize(); i ++ )
- {
- double add = u[i] + currentTau * fu[ i ];
- //if( fabs(u[i]) < fabs(add) or (this->subgridValues[subGridID] == this->currentStep +4) )
- u[ i ] = add;
- }
- time += currentTau;
-
- //cout << '\r' << flush;
- //cout << maxResidue << " " << currentTau << " @ " << time << flush;
- //lastResidue = maxResidue;
- }
- //cout << "Time: " << time << ", Res: " << maxResidue <<endl;
- /*if (u.max() > 0.0)
- this->stopTime /=(double) this->gridCols;*/
-
-// VectorType solution;
-// solution.setLike(u);
-// for( int i = 0; i < u.getSize(); i ++ )
-// {
-// solution[i]=u[i];
-// }
-// return solution;
- return u;
-}
-
-
-#ifdef HAVE_CUDA
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::getSubgridCUDA3D( const int i ,tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >* caller, double* a)
-{
- //int j = threadIdx.x + threadIdx.y * blockDim.x;
-// int index = (blockIdx.z*this->n + threadIdx.z) * this->n*this->n*this->gridCols*this->gridRows
-// + (blockIdx.y) * this->n*this->n*this->gridCols
-// + (blockIdx.x) * this->n
-// + threadIdx.y * this->n*this->gridCols
-// + threadIdx.x;
-
-
- int index = blockDim.x*blockIdx.x + threadIdx.x +
- (blockDim.y*blockIdx.y + threadIdx.y)*blockDim.x*gridDim.x +
- (blockDim.z*blockIdx.z + threadIdx.z)*blockDim.x*gridDim.x*blockDim.y*gridDim.y;
-
- //printf("i= %d,j= %d,th= %d\n",i,j,th);
- *a = caller->work_u_cuda[index];
- //printf("Hi %f \n", *a);
- //return ret;
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::updateSubgridCUDA3D( const int i ,tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >* caller, double* a)
-{
-// int j = threadIdx.x + threadIdx.y * blockDim.x;
-// int index = (blockIdx.z*this->n + threadIdx.z) * this->n*this->n*this->gridCols*this->gridRows
-// + (blockIdx.y) * this->n*this->n*this->gridCols
-// + (blockIdx.x) * this->n
-// + threadIdx.y * this->n*this->gridCols
-// + threadIdx.x;
-
- int index = blockDim.x*blockIdx.x + threadIdx.x +
- (blockDim.y*blockIdx.y + threadIdx.y)*blockDim.x*gridDim.x +
- (blockDim.z*blockIdx.z + threadIdx.z)*blockDim.x*gridDim.x*blockDim.y*gridDim.y;
-
- if( (fabs(caller->work_u_cuda[index]) > fabs(*a)) || (caller->unusedCell_cuda[index] == 1) )
- {
- caller->work_u_cuda[index] = *a;
- caller->unusedCell_cuda[index] = 0;
-
- }
-
- *a = caller->work_u_cuda[index];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::insertSubgridCUDA3D( double u, const int i )
-{
-
-
-// int j = threadIdx.x + threadIdx.y * blockDim.x;
- //printf("j = %d, u = %f\n", j,u);
-
-// int index = (blockIdx.z*this->n + threadIdx.z) * this->n*this->n*this->gridCols*this->gridRows
-// + (blockIdx.y) * this->n*this->n*this->gridCols
-// + (blockIdx.x) * this->n
-// + threadIdx.y * this->n*this->gridCols
-// + threadIdx.x;
-
- int index = blockDim.x*blockIdx.x + threadIdx.x +
- (blockDim.y*blockIdx.y + threadIdx.y)*blockDim.x*gridDim.x +
- (blockDim.z*blockIdx.z + threadIdx.z)*blockDim.x*gridDim.x*blockDim.y*gridDim.y;
-
- //printf("i= %d,j= %d,index= %d\n",i,j,index);
- if( (fabs(this->work_u_cuda[index]) > fabs(u)) || (this->unusedCell_cuda[index] == 1) )
- {
- this->work_u_cuda[index] = u;
- this->unusedCell_cuda[index] = 0;
-
- }
-
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::runSubgridCUDA3D( int boundaryCondition, double* u, int subGridID)
-{
-
- __shared__ int tmp;
- __shared__ double value;
- //double tmpRes = 0.0;
- volatile double* sharedTau = &u[blockDim.x*blockDim.y*blockDim.z];
-// volatile double* absVal = &u[2*blockDim.x*blockDim.y*blockDim.z];
- int i = threadIdx.x;
- int j = threadIdx.y;
- int k = threadIdx.z;
- int l = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z*blockDim.x*blockDim.y;
- bool computeFU = !((i == 0 && (boundaryCondition & 4)) or
- (i == blockDim.x - 1 && (boundaryCondition & 2)) or
- (j == 0 && (boundaryCondition & 8)) or
- (j == blockDim.y - 1 && (boundaryCondition & 1))or
- (k == 0 && (boundaryCondition & 32)) or
- (k == blockDim.z - 1 && (boundaryCondition & 16)));
-
- if(l == 0)
- {
- tmp = 0;
- int centerGID = (blockDim.y*blockIdx.y + (blockDim.y>>1) )*(blockDim.x*gridDim.x) + blockDim.x*blockIdx.x + (blockDim.x>>1)
- + ((blockDim.z>>1)+blockDim.z*blockIdx.z)*blockDim.x*blockDim.y*gridDim.x*gridDim.y;
- if(this->unusedCell_cuda[centerGID] == 0 || boundaryCondition == 0)
- tmp = 1;
- }
- __syncthreads();
-
-
- __syncthreads();
- if(tmp !=1)
- {
-// if(computeFU)
-// absVal[l]=0.0;
-// else
-// absVal[l] = fabs(u[l]);
-//
-// __syncthreads();
-//
-// if((blockDim.x == 16) && (l < 128)) absVal[l] = Max(absVal[l],absVal[l+128]);
-// __syncthreads();
-// if((blockDim.x == 16) && (l < 64)) absVal[l] = Max(absVal[l],absVal[l+64]);
-// __syncthreads();
-// if(l < 32) absVal[l] = Max(absVal[l],absVal[l+32]);
-// if(l < 16) absVal[l] = Max(absVal[l],absVal[l+16]);
-// if(l < 8) absVal[l] = Max(absVal[l],absVal[l+8]);
-// if(l < 4) absVal[l] = Max(absVal[l],absVal[l+4]);
-// if(l < 2) absVal[l] = Max(absVal[l],absVal[l+2]);
-// if(l < 1) value = sign(u[0])*Max(absVal[l],absVal[l+1]);
-// __syncthreads();
-//
-// if(computeFU)
-// u[l] = value;
- if(computeFU)
- {
- tnlGridEntity<MeshType, 3, tnlGridEntityNoStencilStorage > Ent(subMesh);
- if(boundaryCondition == 4)
- {
- Ent.setCoordinates(Containers::StaticVector<3,int>(0,j,k));
- Ent.refresh();
- u[l] = u[Ent.getIndex()];// + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >()*(threadIdx.x) ;//+ 2*sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >()*(threadIdx.x+this->n);
- }
- else if(boundaryCondition == 2)
- {
- Ent.setCoordinates(Containers::StaticVector<3,int>(blockDim.x - 1,j,k));
- Ent.refresh();
- u[l] = u[Ent.getIndex()];// + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >()*(this->n - 1 - threadIdx.x);//+ 2*sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >()*(blockDim.x - threadIdx.x - 1+this->n);
- }
- else if(boundaryCondition == 8)
- {
- Ent.setCoordinates(Containers::StaticVector<3,int>(i,0,k));
- Ent.refresh();
- u[l] = u[Ent.getIndex()];// + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 0, 1, 0 >()*(threadIdx.y) ;//+ 2*sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >()*(threadIdx.y+this->n);
- }
- else if(boundaryCondition == 1)
- {
- Ent.setCoordinates(Containers::StaticVector<3,int>(i,blockDim.y - 1,k));
- Ent.refresh();
- u[l] = u[Ent.getIndex()];// + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 0, 1, 0 >()*(this->n - 1 - threadIdx.y) ;//+ sign(u[0])*this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >()*(blockDim.y - threadIdx.y - 1 +this->n);
- }
- else if(boundaryCondition == 32)
- {
- Ent.setCoordinates(Containers::StaticVector<3,int>(i,j,0));
- Ent.refresh();
- u[l] = u[Ent.getIndex()];// + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 0, 0, 1 >()*(threadIdx.z);
- }
- else if(boundaryCondition == 16)
- {
- Ent.setCoordinates(Containers::StaticVector<3,int>(i,j,blockDim.z - 1));
- Ent.refresh();
- u[l] = u[Ent.getIndex()];// + sign(u[0])*this->subMesh.template getSpaceStepsProducts< 0, 0, 1 >()*(this->n - 1 - threadIdx.z) ;
- }
- }
- }
-
- double time = 0.0;
- __shared__ double currentTau;
- double cfl = this->cflCondition;
- double fu = 0.0;
-// if(threadIdx.x * threadIdx.y * threadIdx.z == 0)
-// {
-// currentTau = this->tau0;
-// }
- double finalTime = this->stopTime;
- __syncthreads();
- if( boundaryCondition == 0 ) finalTime *= 2.0;
-
- tnlGridEntity<MeshType, 3, tnlGridEntityNoStencilStorage > Entity(subMesh);
- tnlNeighborGridEntityGetter<tnlGridEntity< MeshType, 3, tnlGridEntityNoStencilStorage >,3> neighborEntities(Entity);
- Entity.setCoordinates(Containers::StaticVector<3,int>(i,j,k));
- Entity.refresh();
- neighborEntities.refresh(subMesh,Entity.getIndex());
-
-
- while( time < finalTime )
- {
- sharedTau[l]=finalTime;
-
- if(computeFU)
- {
- fu = schemeHost.getValueDev( this->subMesh, l, Containers::StaticVector<3,int>(i,j,k), u, time, boundaryCondition, neighborEntities);
- if(abs(fu) > 0.0)
- sharedTau[l]=abs(cfl/fu);
- }
-
- if(l == 0)
- {
- if(sharedTau[0] > 0.5 * this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >()) sharedTau[0] = 0.5 * this->subMesh.template getSpaceStepsProducts< 1, 0, 0 >();
- }
- else if(l == blockDim.x*blockDim.y*blockDim.z - 1)
- {
- if( time + sharedTau[l] > finalTime ) sharedTau[l] = finalTime - time;
- }
-
- __syncthreads();
- if(l < 256) sharedTau[l] = Min(sharedTau[l],sharedTau[l+256]);
- __syncthreads();
- if(l < 128) sharedTau[l] = Min(sharedTau[l],sharedTau[l+128]);
- __syncthreads();
- if(l < 64) sharedTau[l] = Min(sharedTau[l],sharedTau[l+64]);
- __syncthreads();
- if(l < 32) sharedTau[l] = Min(sharedTau[l],sharedTau[l+32]);
- __syncthreads();
- if(l < 16) sharedTau[l] = Min(sharedTau[l],sharedTau[l+16]);
- if(l < 8) sharedTau[l] = Min(sharedTau[l],sharedTau[l+8]);
- if(l < 4) sharedTau[l] = Min(sharedTau[l],sharedTau[l+4]);
- if(l < 2) sharedTau[l] = Min(sharedTau[l],sharedTau[l+2]);
- if(l < 1) currentTau = Min(sharedTau[l],sharedTau[l+1]);
- __syncthreads();
-
-// if(abs(fu) < 10000.0)
-// printf("bla");
- if(computeFU)
- u[l] += currentTau * fu;
- time += currentTau;
- __syncthreads();
- }
-
-
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::getOwnerCUDA3D(int i) const
-{
- int j = i % (this->gridCols*this->gridRows*this->n*this->n);
-
- return ( (i / (this->gridCols*this->gridRows*this->n*this->n))*this->gridCols*this->gridRows
- + (j / (this->gridCols*this->n*this->n))*this->gridCols
- + (j % (this->gridCols*this->n))/this->n);
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::getSubgridValueCUDA3D( int i ) const
-{
- return this->subgridValues_cuda[i];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::setSubgridValueCUDA3D(int i, int value)
-{
- this->subgridValues_cuda[i] = value;
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-int tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::getBoundaryConditionCUDA3D( int i ) const
-{
- return this->boundaryConditions_cuda[i];
-}
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__device__
-void tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::setBoundaryConditionCUDA3D(int i, int value)
-{
- this->boundaryConditions_cuda[i] = value;
-}
-
-
-
-//north - 1, east - 2, west - 4, south - 8, up -16, down - 32
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void /*tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int>::*/synchronizeCUDA3D(tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver) //needs fix ---- maybe not anymore --- but frankly: yeah, it does -- aaaa-and maybe fixed now
-{
-
- __shared__ int boundary[6]; // north,east,west,south
- __shared__ int subgridValue;
- __shared__ int newSubgridValue;
-
-
- int gid = blockDim.x*blockIdx.x + threadIdx.x +
- (blockDim.y*blockIdx.y + threadIdx.y)*blockDim.x*gridDim.x +
- (blockDim.z*blockIdx.z + threadIdx.z)*blockDim.x*gridDim.x*blockDim.y*gridDim.y;
- double u = cudaSolver->work_u_cuda[gid];
- double u_cmp;
- int subgridValue_cmp=INT_MAX;
- int boundary_index=0;
-
-
- if(threadIdx.x+threadIdx.y+threadIdx.z == 0)
- {
- subgridValue = cudaSolver->getSubgridValueCUDA3D(blockIdx.y*gridDim.x + blockIdx.x + blockIdx.z*gridDim.x*gridDim.y);
- boundary[0] = 0;
- boundary[1] = 0;
- boundary[2] = 0;
- boundary[3] = 0;
- boundary[4] = 0;
- boundary[5] = 0;
- newSubgridValue = 0;
-// printf("aaa z = %d, y = %d, x = %d\n",blockIdx.z,blockIdx.y,blockIdx.x);
- }
- __syncthreads();
-
-
-
- if( (threadIdx.x == 0 /* && !(cudaSolver->currentStep & 1)*/) ||
- (threadIdx.y == 0 /* && (cudaSolver->currentStep & 1)*/) ||
- (threadIdx.z == 0 /* && !(cudaSolver->currentStep & 1)*/) ||
- (threadIdx.x == blockDim.x - 1 /* && !(cudaSolver->currentStep & 1)*/) ||
- (threadIdx.y == blockDim.y - 1 /* && (cudaSolver->currentStep & 1)*/) ||
- (threadIdx.z == blockDim.z - 1 /* && (cudaSolver->currentStep & 1)*/) )
- {
- if(threadIdx.x == 0 && (blockIdx.x != 0)/* && !(cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid - 1];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA3D(blockIdx.y*gridDim.x + blockIdx.x + blockIdx.z*gridDim.x*gridDim.y - 1);
- boundary_index = 2;
- }
-
- if(threadIdx.x == blockDim.x - 1 && (blockIdx.x != gridDim.x - 1)/* && !(cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid + 1];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA3D(blockIdx.y*gridDim.x + blockIdx.x + blockIdx.z*gridDim.x*gridDim.y + 1);
- boundary_index = 1;
- }
-
- __threadfence();
- if((subgridValue == INT_MAX || fabs(u_cmp) + cudaSolver->delta < fabs(u) ) && (subgridValue_cmp != INT_MAX && subgridValue_cmp != -INT_MAX))
- {
- cudaSolver->unusedCell_cuda[gid] = 0;
- atomicMax(&newSubgridValue, INT_MAX);
- atomicMax(&boundary[boundary_index], 1);
- cudaSolver->work_u_cuda[gid] = u_cmp;
- u=u_cmp;
- }
- __threadfence();
- if(threadIdx.y == 0 && (blockIdx.y != 0)/* && (cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid - blockDim.x*gridDim.x];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA3D((blockIdx.y - 1)*gridDim.x + blockIdx.x + blockIdx.z*gridDim.x*gridDim.y);
- boundary_index = 3;
- }
- if(threadIdx.y == blockDim.y - 1 && (blockIdx.y != gridDim.y - 1)/* && (cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid + blockDim.x*gridDim.x];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA3D((blockIdx.y + 1)*gridDim.x + blockIdx.x + blockIdx.z*gridDim.x*gridDim.y);
- boundary_index = 0;
- }
-
- __threadfence();
- if((subgridValue == INT_MAX || fabs(u_cmp) + cudaSolver->delta < fabs(u) ) && (subgridValue_cmp != INT_MAX && subgridValue_cmp != -INT_MAX))
- {
- cudaSolver->unusedCell_cuda[gid] = 0;
- atomicMax(&newSubgridValue, INT_MAX);
- atomicMax(&boundary[boundary_index], 1);
- cudaSolver->work_u_cuda[gid] = u_cmp;
- u=u_cmp;
- }
- __threadfence();
-
- if(threadIdx.z == 0 && (blockIdx.z != 0)/* && (cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid - blockDim.x*gridDim.x*blockDim.y*gridDim.y];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA3D(blockIdx.y*gridDim.x + blockIdx.x + (blockIdx.z - 1)*gridDim.x*gridDim.y);
- boundary_index = 5;
- }
- if(threadIdx.z == blockDim.z - 1 && (blockIdx.z != gridDim.z - 1)/* && (cudaSolver->currentStep & 1)*/)
- {
- u_cmp = cudaSolver->work_u_cuda[gid + blockDim.x*gridDim.x*blockDim.y*gridDim.y];
- subgridValue_cmp = cudaSolver->getSubgridValueCUDA3D(blockIdx.y*gridDim.x + blockIdx.x + (blockIdx.z + 1)*gridDim.x*gridDim.y);
- boundary_index = 4;
- }
- __threadfence();
-
- if((subgridValue == INT_MAX || fabs(u_cmp) + cudaSolver->delta < fabs(u) ) && (subgridValue_cmp != INT_MAX && subgridValue_cmp != -INT_MAX))
- {
- cudaSolver->unusedCell_cuda[gid] = 0;
- atomicMax(&newSubgridValue, INT_MAX);
- atomicMax(&boundary[boundary_index], 1);
- cudaSolver->work_u_cuda[gid] = u_cmp;
- }
- __threadfence();
-
- }
- __syncthreads();
-
- if(threadIdx.x+threadIdx.y+threadIdx.z == 0)
- {
-
- if(subgridValue == INT_MAX && newSubgridValue != 0)
- cudaSolver->setSubgridValueCUDA3D(blockIdx.y*gridDim.x + blockIdx.x + blockIdx.z*gridDim.x*gridDim.y, -INT_MAX);
-
- cudaSolver->setBoundaryConditionCUDA3D(blockIdx.y*gridDim.x + blockIdx.x + blockIdx.z*gridDim.x*gridDim.y, 1 * boundary[0] +
- 2 * boundary[1] +
- 4 * boundary[2] +
- 8 * boundary[3] +
- 16 * boundary[4] +
- 32 * boundary[5] );
- if(blockIdx.x+blockIdx.y+blockIdx.z == 0)
- {
- cudaSolver->currentStep = cudaSolver->currentStep + 1;
- *(cudaSolver->runcuda) = 0;
- }
- }
-}
-
-
-
-template <typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void synchronize2CUDA3D(tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver)
-{
- int stepValue = cudaSolver->currentStep + 4;
- if( cudaSolver->getSubgridValueCUDA3D(blockIdx.z*gridDim.x*gridDim.y + blockIdx.y*gridDim.x + blockIdx.x) == -INT_MAX )
- cudaSolver->setSubgridValueCUDA3D(blockIdx.z*gridDim.x*gridDim.y + blockIdx.y*gridDim.x + blockIdx.x, stepValue);
-
- atomicMax((cudaSolver->runcuda),cudaSolver->getBoundaryConditionCUDA3D(blockIdx.z*gridDim.x*gridDim.y + blockIdx.y*gridDim.x + blockIdx.x));
-}
-
-
-
-
-
-
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device>
-__global__
-void initCUDA3D( tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >* cudaSolver, double* ptr , int* ptr2, int* ptr3)
-{
-
-
- cudaSolver->work_u_cuda = ptr;//(double*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*cudaSolver->n*cudaSolver->n*sizeof(double));
- cudaSolver->unusedCell_cuda = ptr3;//(int*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*cudaSolver->n*cudaSolver->n*sizeof(int));
- cudaSolver->subgridValues_cuda =(int*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*cudaSolver->gridLevels*sizeof(int));
- cudaSolver->boundaryConditions_cuda =(int*)malloc(cudaSolver->gridCols*cudaSolver->gridRows*cudaSolver->gridLevels*sizeof(int));
- cudaSolver->runcuda = ptr2;//(bool*)malloc(sizeof(bool));
- *(cudaSolver->runcuda) = 1;
- cudaSolver->currentStep = 1;
- //cudaMemcpy(ptr,&(cudaSolver->work_u_cuda), sizeof(double*),cudaMemcpyDeviceToHost);
- //ptr = cudaSolver->work_u_cuda;
- printf("GPU memory allocated.\n");
-
- for(int i = 0; i < cudaSolver->gridCols*cudaSolver->gridRows*cudaSolver->gridLevels; i++)
- {
- cudaSolver->subgridValues_cuda[i] = INT_MAX;
- cudaSolver->boundaryConditions_cuda[i] = 0;
- }
-
- /*for(long int j = 0; j < cudaSolver->n*cudaSolver->n*cudaSolver->gridCols*cudaSolver->gridRows; j++)
- {
- printf("%d\n",j);
- cudaSolver->unusedCell_cuda[ j] = 1;
- }*/
- printf("GPU memory initialized.\n");
-}
-
-
-
-
-//extern __shared__ double array[];
-template< typename SchemeHost, typename SchemeDevice, typename Device >
-__global__
-void initRunCUDA3D(tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >* caller)
-
-{
-
-
- extern __shared__ double u[];
-
- int i = blockIdx.z * gridDim.x * gridDim.y + blockIdx.y * gridDim.x + blockIdx.x;
- int l = threadIdx.z * blockDim.x * blockDim.y + threadIdx.y * blockDim.x + threadIdx.x;
-
- __shared__ int containsCurve;
- if(l == 0)
- {
-// printf("z = %d, y = %d, x = %d\n",blockIdx.z,blockIdx.y,blockIdx.x);
- containsCurve = 0;
- }
-
- caller->getSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- if(u[0] * u[l] <= 0.0)
- {
- atomicMax( &containsCurve, 1);
- }
-
- __syncthreads();
- if(containsCurve == 1)
- {
- caller->runSubgridCUDA3D(0,u,i);
- __syncthreads();
-// caller->insertSubgridCUDA3D(u[l],i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
-
- __syncthreads();
- if(l == 0)
- caller->setSubgridValueCUDA3D(i, 4);
- }
-
-
-}
-
-
-
-
-
-template< typename SchemeHost, typename SchemeDevice, typename Device >
-__global__
-void runCUDA3D(tnlParallelEikonalSolver<3,SchemeHost, SchemeDevice, Device, double, int >* caller)
-{
- extern __shared__ double u[];
- int i = blockIdx.z * gridDim.x * gridDim.y + blockIdx.y * gridDim.x + blockIdx.x;
- int l = threadIdx.z * blockDim.x * blockDim.y + threadIdx.y * blockDim.x + threadIdx.x;
- int bound = caller->getBoundaryConditionCUDA3D(i);
-
- if(caller->getSubgridValueCUDA3D(i) != INT_MAX && bound != 0 && caller->getSubgridValueCUDA3D(i) > 0)
- {
- caller->getSubgridCUDA3D(i,caller, &u[l]);
-
- //if(l == 0)
- //printf("i = %d, bound = %d\n",i,caller->getSubgridValueCUDA3D(i));
- if(caller->getSubgridValueCUDA3D(i) == caller->currentStep+4)
- {
- if(bound & 1)
- {
- caller->runSubgridCUDA3D(1,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 2 )
- {
- caller->runSubgridCUDA3D(2,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 4)
- {
- caller->runSubgridCUDA3D(4,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 8)
- {
- caller->runSubgridCUDA3D(8,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 16)
- {
- caller->runSubgridCUDA3D(16,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound & 32)
- {
- caller->runSubgridCUDA3D(32,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
-
- }
- else
- {
- if( ((bound == 2)))
- {
- caller->runSubgridCUDA3D(2,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound == 1) ))
- {
- caller->runSubgridCUDA3D(1,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound == 8) ))
- {
- caller->runSubgridCUDA3D(8,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if((bound == 4))
- {
- caller->runSubgridCUDA3D(4,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound == 16)
- {
- caller->runSubgridCUDA3D(16,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if(bound == 32)
- {
- caller->runSubgridCUDA3D(32,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- }
- /* 1 2 4 8 16 32 */
-
- if( ((bound & 19 ))) /* 1 1 0 0 1 0 */
- {
- caller->runSubgridCUDA3D(19,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound & 21 ))) /* 1 0 1 0 1 0 */
- {
- caller->runSubgridCUDA3D(21,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound & 26 ))) /* 0 1 0 1 1 0 */
- {
- caller->runSubgridCUDA3D(26,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if( (bound & 28 )) /* 0 0 1 1 1 0 */
- {
- caller->runSubgridCUDA3D(28,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
-
-
-
- if( ((bound & 35 ))) /* 1 0 1 0 0 1 */
- {
- caller->runSubgridCUDA3D(35,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound & 37 ))) /* 1 0 1 0 0 1 */
- {
- caller->runSubgridCUDA3D(37,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if( ((bound & 42 ))) /* 0 1 0 1 0 1 */
- {
- caller->runSubgridCUDA3D(42,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
- if( (bound & 44 )) /* 0 0 1 1 0 1 */
- {
- caller->runSubgridCUDA3D(44,u,i);
- caller->updateSubgridCUDA3D(i,caller, &u[l]);
- __syncthreads();
- }
-
- if(l==0)
- {
- caller->setBoundaryConditionCUDA3D(i, 0);
- caller->setSubgridValueCUDA3D(i, caller->getSubgridValueCUDA3D(i) - 1 );
- }
-
-
- }
-
-
-
-}
-
-#endif /*HAVE_CUDA*/
-
-#endif /* TNLPARALLELEIKONALSOLVER3D_IMPL_H_ */
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/MainBuildConfig.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/MainBuildConfig.h
index f8f9187fa514cc9d836bc8072a7adbfddd5f8216..a2a1d7372caadb2ec725a1051e3b9975d0fa82df 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/MainBuildConfig.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/MainBuildConfig.h
@@ -23,7 +23,7 @@ namespace Solvers {
/****
* Turn off support for float and long double.
*/
-template<> struct ConfigTagReal< HamiltonJacobiBuildConfig, float > { enum { enabled = false }; };
+template<> struct ConfigTagReal< HamiltonJacobiBuildConfig, float > { enum { enabled = true }; };
template<> struct ConfigTagReal< HamiltonJacobiBuildConfig, long double > { enum { enabled = false }; };
/****
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
index b981a92a8cb5c7d495736ab1c12cb4b891167bee..f712ce2cc9ff3de5701a8550722b8904d33c4229 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -19,92 +19,112 @@ class tnlDirectEikonalMethodsBase
};
template< typename Real,
- typename Device,
- typename Index >
+ typename Device,
+ typename Index >
class tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >
{
- public:
-
- typedef Meshes::Grid< 1, Real, Device, Index > MeshType;
- typedef Real RealType;
- typedef Device DevcieType;
- typedef Index IndexType;
- typedef Functions::MeshFunction< MeshType > MeshFunctionType;
- typedef Functions::MeshFunction< MeshType, 1, bool > InterfaceMapType;
- using MeshFunctionPointer = Pointers::SharedPointer< MeshFunctionType >;
- using InterfaceMapPointer = Pointers::SharedPointer< InterfaceMapType >;
-
- void initInterface( const MeshFunctionPointer& input,
- MeshFunctionPointer& output,
- InterfaceMapPointer& interfaceMap );
-
- template< typename MeshEntity >
- __cuda_callable__ void updateCell( MeshFunctionType& u,
- const MeshEntity& cell,
- const RealType velocity = 1.0 );
-
- __cuda_callable__ bool updateCell( volatile Real sArray[18],
- int thri, const Real h,
- const Real velocity = 1.0 );
+ public:
+
+ typedef Meshes::Grid< 1, Real, Device, Index > MeshType;
+ typedef Real RealType;
+ typedef Device DevcieType;
+ typedef Index IndexType;
+ typedef Functions::MeshFunction< MeshType > MeshFunctionType;
+ typedef Functions::MeshFunction< MeshType, 1, bool > InterfaceMapType;
+ using MeshFunctionPointer = Pointers::SharedPointer< MeshFunctionType >;
+ using InterfaceMapPointer = Pointers::SharedPointer< InterfaceMapType >;
+
+ void initInterface( const MeshFunctionPointer& input,
+ MeshFunctionPointer& output,
+ InterfaceMapPointer& interfaceMap );
+
+ template< typename MeshEntity >
+ __cuda_callable__ void updateCell( MeshFunctionType& u,
+ const MeshEntity& cell,
+ const RealType velocity = 1.0 );
+
+ __cuda_callable__ bool updateCell( volatile Real sArray[18],
+ int thri, const Real h,
+ const Real velocity = 1.0 );
};
template< typename Real,
- typename Device,
- typename Index >
+ typename Device,
+ typename Index >
class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
{
- public:
- typedef Meshes::Grid< 2, Real, Device, Index > MeshType;
- typedef Real RealType;
- typedef Device DevcieType;
- typedef Index IndexType;
- typedef Functions::MeshFunction< MeshType > MeshFunctionType;
- typedef Functions::MeshFunction< MeshType, 2, bool > InterfaceMapType;
- using MeshFunctionPointer = Pointers::SharedPointer< MeshFunctionType >;
- using InterfaceMapPointer = Pointers::SharedPointer< InterfaceMapType >;
-
- void initInterface( const MeshFunctionPointer& input,
- MeshFunctionPointer& output,
- InterfaceMapPointer& interfaceMap );
-
- template< typename MeshEntity >
- __cuda_callable__ void updateCell( MeshFunctionType& u,
- const MeshEntity& cell,
- const RealType velocity = 1.0 );
-
- __cuda_callable__ bool updateCell( volatile Real sArray[18][18],
- int thri, int thrj, const Real hx, const Real hy,
- const Real velocity = 1.0 );
+ public:
+ typedef Meshes::Grid< 2, Real, Device, Index > MeshType;
+ typedef Real RealType;
+ typedef Device DevcieType;
+ typedef Index IndexType;
+ typedef Functions::MeshFunction< MeshType > MeshFunctionType;
+ typedef Functions::MeshFunction< MeshType, 2, bool > InterfaceMapType;
+ typedef TNL::Containers::Array< int, Device, IndexType > ArrayContainer;
+ using MeshFunctionPointer = Pointers::SharedPointer< MeshFunctionType >;
+ using InterfaceMapPointer = Pointers::SharedPointer< InterfaceMapType >;
+
+ void initInterface( const MeshFunctionPointer& input,
+ MeshFunctionPointer& output,
+ InterfaceMapPointer& interfaceMap );
+
+ template< typename MeshEntity >
+ __cuda_callable__ void updateCell( MeshFunctionType& u,
+ const MeshEntity& cell,
+ const RealType velocity = 1.0 );
+
+ template< int sizeSArray >
+ __cuda_callable__ bool updateCell( volatile Real *sArray,
+ int thri, int thrj, const Real hx, const Real hy,
+ const Real velocity = 1.0 );
+
+ template< int sizeSArray >
+ void updateBlocks( InterfaceMapType interfaceMap,
+ MeshFunctionType aux,
+ MeshFunctionType helpFunc,
+ ArrayContainer BlockIterHost, int numThreadsPerBlock/*, Real **sArray*/ );
+
+ void getNeighbours( ArrayContainer BlockIterHost, int numBlockX, int numBlockY );
};
template< typename Real,
- typename Device,
- typename Index >
+ typename Device,
+ typename Index >
class tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
{
- public:
- typedef Meshes::Grid< 3, Real, Device, Index > MeshType;
- typedef Real RealType;
- typedef Device DevcieType;
- typedef Index IndexType;
- typedef Functions::MeshFunction< MeshType > MeshFunctionType;
- typedef Functions::MeshFunction< MeshType, 3, bool > InterfaceMapType;
- using MeshFunctionPointer = Pointers::SharedPointer< MeshFunctionType >;
- using InterfaceMapPointer = Pointers::SharedPointer< InterfaceMapType >;
-
- void initInterface( const MeshFunctionPointer& input,
- MeshFunctionPointer& output,
- InterfaceMapPointer& interfaceMap );
-
- template< typename MeshEntity >
- __cuda_callable__ void updateCell( MeshFunctionType& u,
- const MeshEntity& cell,
- const RealType velocity = 1.0);
-
- __cuda_callable__ bool updateCell( volatile Real sArray[10][10][10],
- int thri, int thrj, int thrk, const Real hx, const Real hy, const Real hz,
- const Real velocity = 1.0 );
+ public:
+ typedef Meshes::Grid< 3, Real, Device, Index > MeshType;
+ typedef Real RealType;
+ typedef Device DevcieType;
+ typedef Index IndexType;
+ typedef Functions::MeshFunction< MeshType > MeshFunctionType;
+ typedef Functions::MeshFunction< MeshType, 3, bool > InterfaceMapType;
+ typedef TNL::Containers::Array< int, Device, IndexType > ArrayContainer;
+ using MeshFunctionPointer = Pointers::SharedPointer< MeshFunctionType >;
+ using InterfaceMapPointer = Pointers::SharedPointer< InterfaceMapType >;
+
+ void initInterface( const MeshFunctionPointer& input,
+ MeshFunctionPointer& output,
+ InterfaceMapPointer& interfaceMap );
+
+ template< typename MeshEntity >
+ __cuda_callable__ void updateCell( MeshFunctionType& u,
+ const MeshEntity& cell,
+ const RealType velocity = 1.0);
+
+ template< int sizeSArray >
+ void updateBlocks( const InterfaceMapType interfaceMap,
+ const MeshFunctionType aux,
+ MeshFunctionType& helpFunc,
+ ArrayContainer BlockIterHost, int numThreadsPerBlock/*, Real **sArray*/ );
+
+ void getNeighbours( ArrayContainer BlockIterHost, int numBlockX, int numBlockY, int numBlockZ );
+
+ template< int sizeSArray >
+ __cuda_callable__ bool updateCell3D( volatile Real *sArray,
+ int thri, int thrj, int thrk, const Real hx, const Real hy, const Real hz,
+ const Real velocity = 1.0 );
};
template < typename T1, typename T2 >
@@ -113,44 +133,54 @@ T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double
template < typename T1 >
__cuda_callable__ void sortMinims( T1 pom[] );
-
#ifdef HAVE_CUDA
template < typename Real, typename Device, typename Index >
__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index > >& input,
- Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index > >& output,
- Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index >, 1, bool >& interfaceMap );
+ Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index > >& output,
+ Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index >, 1, bool >& interfaceMap );
template < typename Real, typename Device, typename Index >
__global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > > ptr,
- const Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index >, 1, bool >& interfaceMap,
- Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index > >& aux,
- bool *BlockIterDevice );
+ const Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index >, 1, bool >& interfaceMap,
+ Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index > >& aux,
+ bool *BlockIterDevice );
-template < typename Real, typename Device, typename Index >
+template < int sizeSArray, typename Real, typename Device, typename Index >
__global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr,
- const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
- Real *aux,
- int *BlockIterDevice);
-__global__ void CudaParallelReduc( int *BlockIterDevice, int *dBlock, int nBlocks );
+ const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
+ const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
+ Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& helpFunc,
+ TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice, int oddEvenBlock =0);
-template < typename Real, typename Device, typename Index >
-__global__ void aux1( Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux, Real *dAux, int a );
+template < typename Index >
+__global__ void CudaParallelReduc( TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice,
+ TNL::Containers::Array< int, Devices::Cuda, Index > dBlock, int nBlocks );
+
+template < typename Index >
+__global__ void GetNeighbours( TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice,
+ TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterPom, int numBlockX, int numBlockY );
template < typename Real, typename Device, typename Index >
__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& input,
- Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& output,
- Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap );
+ Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& output,
+ Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap );
template < typename Real, typename Device, typename Index >
__global__ void CudaInitCaller3d( const Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& input,
- Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& output,
- Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index >, 3, bool >& interfaceMap );
+ Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& output,
+ Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index >, 3, bool >& interfaceMap );
-template < typename Real, typename Device, typename Index >
+template < int sizeSArray, typename Real, typename Device, typename Index >
__global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr,
- const Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index >, 3, bool >& interfaceMap,
- Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& aux,
- int *BlockIterDevice );
+ const Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index >, 3, bool >& interfaceMap,
+ const Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& aux,
+ Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& helpFunc,
+ TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice );
+
+template < typename Index >
+__global__ void GetNeighbours3D( TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice,
+ TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterPom,
+ int numBlockX, int numBlockY, int numBlockZ );
#endif
#include "tnlDirectEikonalMethodsBase_impl.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase_impl.h
index 649a5ad43a3041fcd479ab7828f7fb1a85634b1c..8f7937541d4708f519cd73a1cb50f0b582fbad48 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase_impl.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase_impl.h
@@ -1,4 +1,4 @@
- /*
+/*
* File: tnlDirectEikonalMethodsBase_impl.h
* Author: oberhuber
*
@@ -11,1061 +11,1649 @@
#include <iostream>
#include "tnlFastSweepingMethod.h"
+#include "tnlDirectEikonalMethodsBase.h"
template< typename Real,
- typename Device,
- typename Index >
+ typename Device,
+ typename Index >
void
tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >::
initInterface( const MeshFunctionPointer& _input,
- MeshFunctionPointer& _output,
- InterfaceMapPointer& _interfaceMap )
+ MeshFunctionPointer& _output,
+ InterfaceMapPointer& _interfaceMap )
{
- if( std::is_same< Device, Devices::Cuda >::value )
- {
+ if( std::is_same< Device, Devices::Cuda >::value )
+ {
#ifdef HAVE_CUDA
- const MeshType& mesh = _input->getMesh();
-
- const int cudaBlockSize( 16 );
- int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().x(), cudaBlockSize );
- dim3 blockSize( cudaBlockSize );
- dim3 gridSize( numBlocksX );
- Devices::Cuda::synchronizeDevice();
- CudaInitCaller<<< gridSize, blockSize >>>( _input.template getData< Device >(),
- _output.template modifyData< Device >(),
- _interfaceMap.template modifyData< Device >() );
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
+ const MeshType& mesh = _input->getMesh();
+
+ const int cudaBlockSize( 16 );
+ int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().x(), cudaBlockSize );
+ dim3 blockSize( cudaBlockSize );
+ dim3 gridSize( numBlocksX );
+ Devices::Cuda::synchronizeDevice();
+ CudaInitCaller<<< gridSize, blockSize >>>( _input.template getData< Device >(),
+ _output.template modifyData< Device >(),
+ _interfaceMap.template modifyData< Device >() );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
#endif
- }
- if( std::is_same< Device, Devices::Host >::value )
- {
- const MeshType& mesh = _input->getMesh();
- typedef typename MeshType::Cell Cell;
- const MeshFunctionType& input = _input.getData();
- MeshFunctionType& output = _output.modifyData();
- InterfaceMapType& interfaceMap = _interfaceMap.modifyData();
- Cell cell( mesh );
- for( cell.getCoordinates().x() = 0;
+ }
+ if( std::is_same< Device, Devices::Host >::value )
+ {
+ const MeshType& mesh = _input->getMesh();
+ typedef typename MeshType::Cell Cell;
+ const MeshFunctionType& input = _input.getData();
+ MeshFunctionType& output = _output.modifyData();
+ InterfaceMapType& interfaceMap = _interfaceMap.modifyData();
+ Cell cell( mesh );
+ for( cell.getCoordinates().x() = 0;
cell.getCoordinates().x() < mesh.getDimensions().x();
cell.getCoordinates().x() ++ )
- {
- cell.refresh();
- output[ cell.getIndex() ] =
- input( cell ) >= 0 ? std::numeric_limits< RealType >::max() :
- -std::numeric_limits< RealType >::max();
- interfaceMap[ cell.getIndex() ] = false;
- }
+ {
+ cell.refresh();
+ output[ cell.getIndex() ] =
+ input( cell ) >= 0 ? std::numeric_limits< RealType >::max() :
+ -std::numeric_limits< RealType >::max();
+ interfaceMap[ cell.getIndex() ] = false;
+ }
+
+
+ const RealType& h = mesh.getSpaceSteps().x();
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh.getDimensions().x() - 1;
+ cell.getCoordinates().x() ++ )
+ {
+ cell.refresh();
+ const RealType& c = input( cell );
+ if( ! cell.isBoundaryEntity() )
+ {
+ const auto& neighbors = cell.getNeighborEntities();
+ Real pom = 0;
+ //const IndexType& c = cell.getIndex();
+ const IndexType e = neighbors.template getEntityIndex< 1 >();
+ if( c * input[ e ] <= 0 )
+ {
+ pom = TNL::sign( c )*( h * c )/( c - input[ e ]);
+ if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) )
+ output[ cell.getIndex() ] = pom;
+
+ pom = pom - TNL::sign( c )*h; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
+ if( TNL::abs( output[ e ] ) > TNL::abs( pom ) )
+ output[ e ] = pom;
+
+ interfaceMap[ cell.getIndex() ] = true;
+ interfaceMap[ e ] = true;
+ }
+ }
+ }
+ }
+}
+
+template< typename Real,
+ typename Device,
+ typename Index >
+template< int sizeSArray >
+void
+tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
+updateBlocks( InterfaceMapType interfaceMap,
+ MeshFunctionType aux,
+ MeshFunctionType helpFunc,
+ ArrayContainer BlockIterHost, int numThreadsPerBlock/*, Real **sArray*/ )
+{
+#pragma omp parallel for schedule( dynamic )
+ for( int i = 0; i < BlockIterHost.getSize(); i++ )
+ {
+ if( BlockIterHost[ i ] )
+ {
+ MeshType mesh = interfaceMap.template getMesh< Devices::Host >();
+
+ int dimX = mesh.getDimensions().x(); int dimY = mesh.getDimensions().y();
+ //std::cout << "dimX = " << dimX << " ,dimY = " << dimY << std::endl;
+ int numOfBlocky = dimY/numThreadsPerBlock + ((dimY%numThreadsPerBlock != 0) ? 1:0);
+ int numOfBlockx = dimX/numThreadsPerBlock + ((dimX%numThreadsPerBlock != 0) ? 1:0);
+ //std::cout << "numOfBlockx = " << numOfBlockx << " ,numOfBlocky = " << numOfBlocky << std::endl;
+ int xkolik = numThreadsPerBlock + 1;
+ int ykolik = numThreadsPerBlock + 1;
+
+ int blIdx = i%numOfBlockx;
+ int blIdy = i/numOfBlockx;
+ //std::cout << "blIdx = " << blIdx << " ,blIdy = " << blIdy << std::endl;
+
+ if( numOfBlockx - 1 == blIdx )
+ xkolik = dimX - (blIdx)*numThreadsPerBlock+1;
+
+ if( numOfBlocky -1 == blIdy )
+ ykolik = dimY - (blIdy)*numThreadsPerBlock+1;
+ //std::cout << "xkolik = " << xkolik << " ,ykolik = " << ykolik << std::endl;
+
+
+ /*bool changed[numThreadsPerBlock*numThreadsPerBlock];
+ changed[ 0 ] = 1;*/
+ Real hx = mesh.getSpaceSteps().x();
+ Real hy = mesh.getSpaceSteps().y();
+
+ bool changed = false;
+ BlockIterHost[ blIdy * numOfBlockx + blIdx ] = 0;
+
+
+ Real *sArray;
+ sArray = new Real[ sizeSArray * sizeSArray ];
+ if( sArray == nullptr )
+ std::cout << "Error while allocating memory for sArray." << std::endl;
+
+ for( int thri = 0; thri < sizeSArray; thri++ ){
+ for( int thrj = 0; thrj < sizeSArray; thrj++ )
+ sArray[ thri * sizeSArray + thrj ] = std::numeric_limits< Real >::max();
+ }
+
+
+ //printf("numThreadsPerBlock = %d\n", numThreadsPerBlock);
+ for( int thrj = 0; thrj < numThreadsPerBlock + 1; thrj++ )
+ {
+ if( dimX > (blIdx+1) * numThreadsPerBlock && thrj+1 < ykolik )
+ sArray[ ( thrj+1 )* sizeSArray +xkolik] = aux[ blIdy*numThreadsPerBlock*dimX - dimX + blIdx*numThreadsPerBlock - 1 + (thrj+1)*dimX + xkolik ];
- const RealType& h = mesh.getSpaceSteps().x();
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh.getDimensions().x() - 1;
- cell.getCoordinates().x() ++ )
+ if( blIdx != 0 && thrj+1 < ykolik )
+ sArray[(thrj+1)* sizeSArray] = aux[ blIdy*numThreadsPerBlock*dimX - dimX + blIdx*numThreadsPerBlock - 1 + (thrj+1)*dimX ];
+
+ if( dimY > (blIdy+1) * numThreadsPerBlock && thrj+1 < xkolik )
+ sArray[ykolik * sizeSArray + thrj+1] = aux[ blIdy*numThreadsPerBlock*dimX - dimX + blIdx*numThreadsPerBlock - 1 + ykolik*dimX + thrj+1 ];
+
+ if( blIdy != 0 && thrj+1 < xkolik )
+ sArray[thrj+1] = aux[ blIdy*numThreadsPerBlock*dimX - dimX + blIdx*numThreadsPerBlock - 1 + thrj+1 ];
+ }
+
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX )
+ sArray[(k+1) * sizeSArray + l+1] = aux[ blIdy * numThreadsPerBlock * dimX + numThreadsPerBlock * blIdx + k*dimX + l ];
+ }
+
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX ){
+ //std::cout << "proslo i = " << k * numThreadsPerBlock + l << std::endl;
+ if( ! interfaceMap[ blIdy * numThreadsPerBlock * dimX + numThreadsPerBlock * blIdx + k*dimX + l ] )
+ {
+ changed = this->template updateCell< sizeSArray >( sArray, l+1, k+1, hx,hy) || changed;
+
+ }
+ }
+ }
+ }
+ /*aux.save( "aux-1pruch.tnl" );
+ for( int k = 0; k < sizeSArray; k++ ){
+ for( int l = 0; l < sizeSArray; l++ ) {
+ std::cout << sArray[ k * sizeSArray + l] << " ";
+ }
+ std::cout << std::endl;
+ }*/
+
+ for( int k = 0; k < numThreadsPerBlock; k++ )
+ for( int l = numThreadsPerBlock-1; l >-1; l-- ) {
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX )
+ {
+ if( ! interfaceMap[ blIdy * numThreadsPerBlock * dimX + numThreadsPerBlock * blIdx + k*dimX + l ] )
+ {
+ this->template updateCell< sizeSArray >( sArray, l+1, k+1, hx,hy);
+ }
+ }
+ }
+ /*aux.save( "aux-2pruch.tnl" );
+ for( int k = 0; k < sizeSArray; k++ ){
+ for( int l = 0; l < sizeSArray; l++ ) {
+ std::cout << sArray[ k * sizeSArray + l] << " ";
+ }
+ std::cout << std::endl;
+ }*/
+
+ for( int k = numThreadsPerBlock-1; k > -1; k-- )
+ for( int l = 0; l < numThreadsPerBlock; l++ ) {
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX )
+ {
+ if( ! interfaceMap[ blIdy * numThreadsPerBlock * dimX + numThreadsPerBlock * blIdx + k*dimX + l ] )
+ {
+ this->template updateCell< sizeSArray >( sArray, l+1, k+1, hx,hy);
+ }
+ }
+ }
+ /*aux.save( "aux-3pruch.tnl" );
+ for( int k = 0; k < sizeSArray; k++ ){
+ for( int l = 0; l < sizeSArray; l++ ) {
+ std::cout << sArray[ k * sizeSArray + l] << " ";
+ }
+ std::cout << std::endl;
+ }*/
+
+ for( int k = numThreadsPerBlock-1; k > -1; k-- ){
+ for( int l = numThreadsPerBlock-1; l >-1; l-- ) {
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX )
+ {
+ if( ! interfaceMap[ blIdy * numThreadsPerBlock * dimX + numThreadsPerBlock * blIdx + k*dimX + l ] )
+ {
+ this->template updateCell< sizeSArray >( sArray, l+1, k+1, hx, hy, 1.0);
+ }
+ }
+ }
+ }
+ /*aux.save( "aux-4pruch.tnl" );
+ for( int k = 0; k < sizeSArray; k++ ){
+ for( int l = 0; l < sizeSArray; l++ ) {
+ std::cout << sArray[ k * sizeSArray + l] << " ";
+ }
+ std::cout << std::endl;
+ }*/
+
+
+ if( changed ){
+ BlockIterHost[ blIdy * numOfBlockx + blIdx ] = 1;
+ }
+
+
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ ) {
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX )
+ helpFunc[ blIdy * numThreadsPerBlock * dimX + numThreadsPerBlock * blIdx + k*dimX + l ] = sArray[ (k + 1)* sizeSArray + l + 1 ];
+ //std::cout<< sArray[k+1][l+1];
+ }
+ //std::cout<<std::endl;
+ }
+ delete []sArray;
+ }
+ }
+}
+template< typename Real,
+ typename Device,
+ typename Index >
+template< int sizeSArray >
+void
+tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
+updateBlocks( const InterfaceMapType interfaceMap,
+ const MeshFunctionType aux,
+ MeshFunctionType& helpFunc,
+ ArrayContainer BlockIterHost, int numThreadsPerBlock/*, Real **sArray*/ )
+{
+//#pragma omp parallel for schedule( dynamic )
+ for( int i = 0; i < BlockIterHost.getSize(); i++ )
+ {
+ if( BlockIterHost[ i ] )
+ {
+ MeshType mesh = interfaceMap.template getMesh< Devices::Host >();
+
+ int dimX = mesh.getDimensions().x(); int dimY = mesh.getDimensions().y();
+ int dimZ = mesh.getDimensions().z();
+ //std::cout << "dimX = " << dimX << " ,dimY = " << dimY << std::endl;
+ int numOfBlocky = dimY/numThreadsPerBlock + ((dimY%numThreadsPerBlock != 0) ? 1:0);
+ int numOfBlockx = dimX/numThreadsPerBlock + ((dimX%numThreadsPerBlock != 0) ? 1:0);
+ int numOfBlockz = dimZ/numThreadsPerBlock + ((dimZ%numThreadsPerBlock != 0) ? 1:0);
+ //std::cout << "numOfBlockx = " << numOfBlockx << " ,numOfBlocky = " << numOfBlocky << std::endl;
+ int xkolik = numThreadsPerBlock + 1;
+ int ykolik = numThreadsPerBlock + 1;
+ int zkolik = numThreadsPerBlock + 1;
+
+
+ int blIdz = i/( numOfBlockx * numOfBlocky );
+ int blIdy = (i-blIdz*numOfBlockx * numOfBlocky )/(numOfBlockx );
+ int blIdx = (i-blIdz*numOfBlockx * numOfBlocky )%( numOfBlockx );
+ //std::cout << "blIdx = " << blIdx << " ,blIdy = " << blIdy << std::endl;
+
+ if( numOfBlockx - 1 == blIdx )
+ xkolik = dimX - (blIdx)*numThreadsPerBlock+1;
+ if( numOfBlocky -1 == blIdy )
+ ykolik = dimY - (blIdy)*numThreadsPerBlock+1;
+ if( numOfBlockz-1 == blIdz )
+ zkolik = dimZ - (blIdz)*numThreadsPerBlock+1;
+ //std::cout << "xkolik = " << xkolik << " ,ykolik = " << ykolik << std::endl;
+
+
+ /*bool changed[numThreadsPerBlock*numThreadsPerBlock];
+ changed[ 0 ] = 1;*/
+ Real hx = mesh.getSpaceSteps().x();
+ Real hy = mesh.getSpaceSteps().y();
+ Real hz = mesh.getSpaceSteps().z();
+
+ bool changed = false;
+ BlockIterHost[ i ] = 0;
+
+
+ Real *sArray;
+ sArray = new Real[ sizeSArray * sizeSArray * sizeSArray ];
+ if( sArray == nullptr )
+ std::cout << "Error while allocating memory for sArray." << std::endl;
+
+ for( int k = 0; k < sizeSArray; k++ )
+ for( int l = 0; l < sizeSArray; l++ )
+ for( int m = 0; m < sizeSArray; m++ ){
+ sArray[ m * sizeSArray * sizeSArray + k * sizeSArray + l ] = std::numeric_limits< Real >::max();
+ }
+
+
+ for( int thrk = 0; thrk < numThreadsPerBlock; thrk++ )
+ for( int thrj = 0; thrj < numThreadsPerBlock; thrj++ )
{
- cell.refresh();
- const RealType& c = input( cell );
- if( ! cell.isBoundaryEntity() )
- {
- const auto& neighbors = cell.getNeighborEntities();
- Real pom = 0;
- //const IndexType& c = cell.getIndex();
- const IndexType e = neighbors.template getEntityIndex< 1 >();
- if( c * input[ e ] <= 0 )
+ if( blIdx != 0 && thrj+1 < ykolik && thrk+1 < zkolik )
+ sArray[(thrk+1 )* sizeSArray * sizeSArray + (thrj+1)*sizeSArray + 0] =
+ aux[ blIdz*numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock*dimX + blIdx*numThreadsPerBlock + thrj * dimX -1 + thrk*dimX*dimY ];
+
+ if( dimX > (blIdx+1) * numThreadsPerBlock && thrj+1 < ykolik && thrk+1 < zkolik )
+ sArray[ (thrk+1) * sizeSArray * sizeSArray + (thrj+1) *sizeSArray + xkolik ] =
+ aux[ blIdz*numThreadsPerBlock * dimX * dimY + blIdy *numThreadsPerBlock*dimX+ blIdx*numThreadsPerBlock + numThreadsPerBlock + thrj * dimX + thrk*dimX*dimY ];
+
+ if( blIdy != 0 && thrj+1 < xkolik && thrk+1 < zkolik )
+ sArray[ (thrk+1) * sizeSArray * sizeSArray +0*sizeSArray + thrj+1] =
+ aux[ blIdz*numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock*dimX + blIdx*numThreadsPerBlock - dimX + thrj + thrk*dimX*dimY ];
+
+ if( dimY > (blIdy+1) * numThreadsPerBlock && thrj+1 < xkolik && thrk+1 < zkolik )
+ sArray[ (thrk+1) * sizeSArray * sizeSArray + ykolik*sizeSArray + thrj+1] =
+ aux[ blIdz*numThreadsPerBlock * dimX * dimY + (blIdy+1) * numThreadsPerBlock*dimX + blIdx*numThreadsPerBlock + thrj + thrk*dimX*dimY ];
+
+ if( blIdz != 0 && thrj+1 < ykolik && thrk+1 < xkolik )
+ sArray[ 0 * sizeSArray * sizeSArray +(thrj+1 )* sizeSArray + thrk+1] =
+ aux[ blIdz*numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock*dimX + blIdx*numThreadsPerBlock - dimX * dimY + thrj * dimX + thrk ];
+
+ if( dimZ > (blIdz+1) * numThreadsPerBlock && thrj+1 < ykolik && thrk+1 < xkolik )
+ sArray[zkolik * sizeSArray * sizeSArray + (thrj+1) * sizeSArray + thrk+1] =
+ aux[ (blIdz+1)*numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock*dimX + blIdx*numThreadsPerBlock + thrj * dimX + thrk ];
+ }
+
+ for( int m = 0; m < numThreadsPerBlock; m++ ){
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ sArray[(m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1] =
+ aux[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ];
+ }
+ }
+ }
+ /*string s;
+ int numWhile = 0;
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );*/
+
+ for( int m = 0; m < numThreadsPerBlock; m++ ){
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ ){
+ //std::cout << "proslo i = " << k * numThreadsPerBlock + l << std::endl;
+ if( ! interfaceMap[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] )
{
- pom = TNL::sign( c )*( h * c )/( c - input[ e ]);
- if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) )
- output[ cell.getIndex() ] = pom;
-
- pom = pom - TNL::sign( c )*h; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
- if( TNL::abs( output[ e ] ) > TNL::abs( pom ) )
- output[ e ] = pom;
-
- interfaceMap[ cell.getIndex() ] = true;
- interfaceMap[ e ] = true;
+ //printf("In with point m = %d, k = %d, l = %d\n", m, k, l);
+ changed = this->template updateCell3D< sizeSArray >( sArray, l+1, k+1, m+1, hx,hy,hz) || changed;
+
}
- }
+ }
+ }
+ }
+ }
+ /*for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );*/
+
+ for( int m = numThreadsPerBlock-1; m >-1; m-- ){
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l <numThreadsPerBlock; l++ ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ {
+ if( ! interfaceMap[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] )
+ {
+ this->template updateCell3D< sizeSArray >( sArray, l+1, k+1, m+1, hx,hy,hz);
+ }
+ }
+ }
}
+ }
+ /*for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );*/
+
+ for( int m = 0; m < numThreadsPerBlock; m++ ){
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = numThreadsPerBlock-1; l >-1; l-- ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ {
+ if( ! interfaceMap[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] )
+ {
+ this->template updateCell3D< sizeSArray >( sArray, l+1, k+1, m+1, hx,hy,hz);
+ }
+ }
+ }
+ }
+ }
+ /*for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );
+ */
+ for( int m = numThreadsPerBlock-1; m >-1; m-- ){
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = numThreadsPerBlock-1; l >-1; l-- ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ {
+ if( ! interfaceMap[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] )
+ {
+ this->template updateCell3D< sizeSArray >( sArray, l+1, k+1, m+1, hx,hy,hz);
+ }
+ }
+ }
+ }
+ }
+ /*for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );*/
+
+ for( int m = 0; m < numThreadsPerBlock; m++ ){
+ for( int k = numThreadsPerBlock-1; k > -1; k-- ){
+ for( int l = 0; l <numThreadsPerBlock; l++ ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ {
+ if( ! interfaceMap[blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] )
+ {
+ this->template updateCell3D< sizeSArray >( sArray, l+1, k+1, m+1, hx,hy,hz);
+ }
+ }
+ }
+ }
+ }
+ /*for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );*/
+
+ for( int m = numThreadsPerBlock-1; m >-1; m-- ){
+ for( int k = numThreadsPerBlock-1; k > -1; k-- ){
+ for( int l = 0; l <numThreadsPerBlock; l++ ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ {
+ if( ! interfaceMap[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] )
+ {
+ this->template updateCell3D< sizeSArray >( sArray, l+1, k+1, m+1, hx,hy,hz);
+ }
+ }
+ }
+ }
+ }
+ /*for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );*/
+
+ for( int m = 0; m < numThreadsPerBlock; m++ ){
+ for( int k = numThreadsPerBlock-1; k > -1; k-- ){
+ for( int l = numThreadsPerBlock-1; l >-1; l-- ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ {
+ if( ! interfaceMap[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] )
+ {
+ this->template updateCell3D< sizeSArray >( sArray, l+1, k+1, m+1, hx,hy,hz);
+ }
+ }
+ }
+ }
+ }
+ /*for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );*/
+
+
+ for( int m = numThreadsPerBlock-1; m >-1; m-- ){
+ for( int k = numThreadsPerBlock-1; k > -1; k-- ){
+ for( int l = numThreadsPerBlock-1; l >-1; l-- ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ {
+ if( ! interfaceMap[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] )
+ {
+ this->template updateCell3D< sizeSArray >( sArray, l+1, k+1, m+1, hx,hy,hz);
+ }
+ }
+ }
+ }
+ }
+ /*for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ )
+ for( int m = 0; m < numThreadsPerBlock; m++ )
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ )
+ helpFunc[ m*dimX*dimY + k*dimX + l ] = sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ }
+ numWhile++;
+ s = "helpFunc-"+ std::to_string(numWhile) + ".tnl";
+ helpFunc.save( s );*/
+
+ if( changed ){
+ BlockIterHost[ i ] = 1;
+ }
+
+
+ for( int k = 0; k < numThreadsPerBlock; k++ ){
+ for( int l = 0; l < numThreadsPerBlock; l++ ) {
+ for( int m = 0; m < numThreadsPerBlock; m++ ){
+ if( blIdy * numThreadsPerBlock + k < dimY && blIdx * numThreadsPerBlock + l < dimX && blIdz * numThreadsPerBlock + m < dimZ ){
+ helpFunc[ blIdz * numThreadsPerBlock * dimX * dimY + blIdy * numThreadsPerBlock * dimX + blIdx*numThreadsPerBlock + m*dimX*dimY + k*dimX + l ] =
+ sArray[ (m+1) * sizeSArray * sizeSArray + (k+1) *sizeSArray + l+1 ];
+ //std::cout << helpFunc[ m*dimX*dimY + k*dimX + l ] << " ";
+ }
+ }
+ //std::cout << std::endl;
+ }
+ //std::cout << std::endl;
+ }
+ //helpFunc.save( "helpF.tnl");
+ delete []sArray;
}
+ }
+}
+template< typename Real,
+ typename Device,
+ typename Index >
+void
+tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
+getNeighbours( ArrayContainer BlockIterHost, int numBlockX, int numBlockY, int numBlockZ )
+{
+ int* BlockIterPom;
+ BlockIterPom = new int [ numBlockX * numBlockY * numBlockZ ];
+
+ for( int i = 0; i< BlockIterHost.getSize(); i++)
+ {
+ BlockIterPom[ i ] = 0;
+
+ int m=0, l=0, k=0;
+ l = i/( numBlockX * numBlockY );
+ k = (i-l*numBlockX * numBlockY )/(numBlockX );
+ m = (i-l*numBlockX * numBlockY )%( numBlockX );
+
+ if( m > 0 && BlockIterHost[ i - 1 ] ){
+ BlockIterPom[ i ] = 1;
+ }else if( m < numBlockX -1 && BlockIterHost[ i + 1 ] ){
+ BlockIterPom[ i ] = 1;
+ }else if( k > 0 && BlockIterHost[ i - numBlockX ] ){
+ BlockIterPom[ i ] = 1;
+ }else if( k < numBlockY -1 && BlockIterHost[ i + numBlockX ] ){
+ BlockIterPom[ i ] = 1;
+ }else if( l > 0 && BlockIterHost[ i - numBlockX*numBlockY ] ){
+ BlockIterPom[ i ] = 1;
+ }else if( l < numBlockZ-1 && BlockIterHost[ i + numBlockX*numBlockY ] ){
+ BlockIterPom[ i ] = 1;
+ }
+ }
+ for( int i = 0; i< BlockIterHost.getSize(); i++)
+ {
+ BlockIterHost[ i ] = BlockIterPom[ i ];
+ }
+}
+
+
+template< typename Real,
+ typename Device,
+ typename Index >
+void
+tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
+getNeighbours( ArrayContainer BlockIterHost, int numBlockX, int numBlockY )
+{
+ int* BlockIterPom;
+ BlockIterPom = new int [numBlockX * numBlockY];
+
+ for(int i = 0; i < numBlockX * numBlockY; i++)
+ {
+ BlockIterPom[ i ] = 0;//BlockIterPom[ i ] = 0;
+ int m=0, k=0;
+ m = i%numBlockX;
+ k = i/numBlockX;
+ if( m > 0 && BlockIterHost[ i - 1 ] ){
+ BlockIterPom[ i ] = 1;
+ }else if( m < numBlockX -1 && BlockIterHost[ i + 1 ] ){
+ BlockIterPom[ i ] = 1;
+ }else if( k > 0 && BlockIterHost[ i - numBlockX ] ){
+ BlockIterPom[ i ] = 1;
+ }else if( k < numBlockY -1 && BlockIterHost[ i + numBlockX ] ){
+ BlockIterPom[ i ] = 1;
+ }
+ //BlockIterPom[ i ];
+ }
+
+ for(int i = 0; i < numBlockX * numBlockY; i++)
+ {
+ if( !BlockIterHost[ i ] )
+ BlockIterHost[ i ] = BlockIterPom[ i ];
+ }
+ /*else
+ BlockIter[ i ] = 0;*/
+ /*for( int i = numBlockX-1; i > -1; i-- )
+ {
+ for( int j = 0; j< numBlockY; j++ )
+ std::cout << BlockIterHost[ i*numBlockY + j ];
+ std::cout << std::endl;
+ }
+ std::cout << std::endl;*/
+ delete[] BlockIterPom;
}
template< typename Real,
- typename Device,
- typename Index >
- template< typename MeshEntity >
+ typename Device,
+ typename Index >
+template< typename MeshEntity >
void
tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >::
updateCell( MeshFunctionType& u,
- const MeshEntity& cell,
- const RealType v )
+ const MeshEntity& cell,
+ const RealType v )
{
- const auto& neighborEntities = cell.template getNeighborEntities< 1 >();
- const MeshType& mesh = cell.getMesh();
- const RealType& h = mesh.getSpaceSteps().x();
- const RealType value = u( cell );
- RealType a, tmp = std::numeric_limits< RealType >::max();
-
- if( cell.getCoordinates().x() == 0 )
- a = u[ neighborEntities.template getEntityIndex< 1 >() ];
- else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 1 )
- a = u[ neighborEntities.template getEntityIndex< -1 >() ];
- else
- {
- a = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< -1 >() ],
- u[ neighborEntities.template getEntityIndex< 1 >() ] );
- }
-
- if( fabs( a ) == std::numeric_limits< RealType >::max() )
- return;
-
- tmp = a + TNL::sign( value ) * h/v;
-
- u[ cell.getIndex() ] = argAbsMin( value, tmp );
+ const auto& neighborEntities = cell.template getNeighborEntities< 1 >();
+ const MeshType& mesh = cell.getMesh();
+ const RealType& h = mesh.getSpaceSteps().x();
+ const RealType value = u( cell );
+ RealType a, tmp = std::numeric_limits< RealType >::max();
+
+ if( cell.getCoordinates().x() == 0 )
+ a = u[ neighborEntities.template getEntityIndex< 1 >() ];
+ else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 1 )
+ a = u[ neighborEntities.template getEntityIndex< -1 >() ];
+ else
+ {
+ a = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< -1 >() ],
+ u[ neighborEntities.template getEntityIndex< 1 >() ] );
+ }
+
+ if( fabs( a ) == std::numeric_limits< RealType >::max() )
+ return;
+
+ tmp = a + TNL::sign( value ) * h/v;
+
+ u[ cell.getIndex() ] = argAbsMin( value, tmp );
}
template< typename Real,
- typename Device,
- typename Index >
+ typename Device,
+ typename Index >
void
tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
initInterface( const MeshFunctionPointer& _input,
- MeshFunctionPointer& _output,
- InterfaceMapPointer& _interfaceMap )
+ MeshFunctionPointer& _output,
+ InterfaceMapPointer& _interfaceMap )
{
-
- if( std::is_same< Device, Devices::Cuda >::value )
- {
+
+ if( std::is_same< Device, Devices::Cuda >::value )
+ {
#ifdef HAVE_CUDA
- const MeshType& mesh = _input->getMesh();
-
- const int cudaBlockSize( 16 );
- int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().x(), cudaBlockSize );
- int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().y(), cudaBlockSize );
- dim3 blockSize( cudaBlockSize, cudaBlockSize );
- dim3 gridSize( numBlocksX, numBlocksY );
- Devices::Cuda::synchronizeDevice();
- CudaInitCaller<<< gridSize, blockSize >>>( _input.template getData< Device >(),
- _output.template modifyData< Device >(),
- _interfaceMap.template modifyData< Device >() );
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
+ const MeshType& mesh = _input->getMesh();
+
+ const int cudaBlockSize( 16 );
+ int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().x(), cudaBlockSize );
+ int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().y(), cudaBlockSize );
+ dim3 blockSize( cudaBlockSize, cudaBlockSize );
+ dim3 gridSize( numBlocksX, numBlocksY );
+ Devices::Cuda::synchronizeDevice();
+ CudaInitCaller<<< gridSize, blockSize >>>( _input.template getData< Device >(),
+ _output.template modifyData< Device >(),
+ _interfaceMap.template modifyData< Device >() );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
#endif
- }
- if( std::is_same< Device, Devices::Host >::value )
- {
- MeshFunctionType input = _input.getData();
-
- /*double A[320][320];
- std::ifstream fileInit("/home/maty/Downloads/initData.txt");
-
- for (int i = 0; i < 320; i++)
- for (int j = 0; j < 320; j++)
- fileInit >> A[i][j];
- fileInit.close();
- for (int i = 0; i < 320; i++)
- for (int j = 0; j < 320; j++)
- input[i*320 + j] = A[i][j];*/
-
-
- MeshFunctionType& output = _output.modifyData();
- InterfaceMapType& interfaceMap = _interfaceMap.modifyData();
- const MeshType& mesh = input.getMesh();
- typedef typename MeshType::Cell Cell;
- Cell cell( mesh );
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh.getDimensions().y();
- cell.getCoordinates().y() ++ )
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh.getDimensions().x();
- cell.getCoordinates().x() ++ )
- {
- cell.refresh();
- output[ cell.getIndex() ] =
- input( cell ) >= 0 ? std::numeric_limits< RealType >::max() :
- - std::numeric_limits< RealType >::max();
- interfaceMap[ cell.getIndex() ] = false;
- }
-
- const RealType& hx = mesh.getSpaceSteps().x();
- const RealType& hy = mesh.getSpaceSteps().y();
- for( cell.getCoordinates().y() = 0;
+ }
+ if( std::is_same< Device, Devices::Host >::value )
+ {
+ MeshFunctionType input = _input.getData();
+
+ /*double A[320][320];
+ std::ifstream fileInit("/home/maty/Downloads/initData.txt");
+
+ for (int i = 0; i < 320; i++)
+ for (int j = 0; j < 320; j++)
+ fileInit >> A[j];
+ fileInit.close();
+ for (int i = 0; i < 320; i++)
+ for (int j = 0; j < 320; j++)
+ input[i*320 + j] = A[j];*/
+
+
+ MeshFunctionType& output = _output.modifyData();
+ InterfaceMapType& interfaceMap = _interfaceMap.modifyData();
+ const MeshType& mesh = input.getMesh();
+ typedef typename MeshType::Cell Cell;
+ Cell cell( mesh );
+ for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh.getDimensions().y();
+ cell.getCoordinates().y() ++ )
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh.getDimensions().x();
+ cell.getCoordinates().x() ++ )
+ {
+ cell.refresh();
+ output[ cell.getIndex() ] =
+ input( cell ) >= 0 ? std::numeric_limits< RealType >::max() :
+ - std::numeric_limits< RealType >::max();
+ interfaceMap[ cell.getIndex() ] = false;
+ }
+
+ const RealType& hx = mesh.getSpaceSteps().x();
+ const RealType& hy = mesh.getSpaceSteps().y();
+ for( cell.getCoordinates().y() = 0;
cell.getCoordinates().y() < mesh.getDimensions().y();
cell.getCoordinates().y() ++ )
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh.getDimensions().x();
- cell.getCoordinates().x() ++ )
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh.getDimensions().x();
+ cell.getCoordinates().x() ++ )
+ {
+ cell.refresh();
+ const RealType& c = input( cell );
+ if( ! cell.isBoundaryEntity() )
+ {
+ auto neighbors = cell.getNeighborEntities();
+ Real pom = 0;
+ const IndexType e = neighbors.template getEntityIndex< 1, 0 >();
+ const IndexType n = neighbors.template getEntityIndex< 0, 1 >();
+ //Try init with exact data:
+ /*if( c * input[ n ] <= 0 )
+ {
+ output[ cell.getIndex() ] = c;
+ output[ n ] = input[ n ];
+ interfaceMap[ cell.getIndex() ] = true;
+ interfaceMap[ n ] = true;
+ }
+ if( c * input[ e ] <= 0 )
+ {
+ output[ cell.getIndex() ] = c;
+ output[ e ] = input[ e ];
+ interfaceMap[ cell.getIndex() ] = true;
+ interfaceMap[ e ] = true;
+ }*/
+ if( c * input[ n ] <= 0 )
{
- cell.refresh();
- const RealType& c = input( cell );
- if( ! cell.isBoundaryEntity() )
+ /*if( c >= 0 )
+ {*/
+ pom = TNL::sign( c )*( hy * c )/( c - input[ n ]);
+ if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) )
+ output[ cell.getIndex() ] = pom;
+ pom = pom - TNL::sign( c )*hy;
+ if( TNL::abs( output[ n ] ) > TNL::abs( pom ) )
+ output[ n ] = pom; //( hy * c )/( c - input[ n ]) - hy;
+ /*}else
{
- auto neighbors = cell.getNeighborEntities();
- Real pom = 0;
- const IndexType e = neighbors.template getEntityIndex< 1, 0 >();
- const IndexType n = neighbors.template getEntityIndex< 0, 1 >();
- //Try init with exact data:
- /*if( c * input[ n ] <= 0 )
- {
- output[ cell.getIndex() ] = c;
- output[ n ] = input[ n ];
- interfaceMap[ cell.getIndex() ] = true;
- interfaceMap[ n ] = true;
- }
- if( c * input[ e ] <= 0 )
- {
- output[ cell.getIndex() ] = c;
- output[ e ] = input[ e ];
- interfaceMap[ cell.getIndex() ] = true;
- interfaceMap[ e ] = true;
- }*/
- if( c * input[ n ] <= 0 )
- {
- /*if( c >= 0 )
- {*/
- pom = TNL::sign( c )*( hy * c )/( c - input[ n ]);
- if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) )
- output[ cell.getIndex() ] = pom;
- pom = pom - TNL::sign( c )*hy;
- if( TNL::abs( output[ n ] ) > TNL::abs( pom ) )
- output[ n ] = pom; //( hy * c )/( c - input[ n ]) - hy;
- /*}else
- {
- pom = - ( hy * c )/( c - input[ n ]);
- if( output[ cell.getIndex() ] < pom )
- output[ cell.getIndex() ] = pom;
- if( output[ n ] > hy + pom )
- output[ n ] = hy + pom; //hy - ( hy * c )/( c - input[ n ]);
- }*/
- interfaceMap[ cell.getIndex() ] = true;
- interfaceMap[ n ] = true;
- }
- if( c * input[ e ] <= 0 )
- {
- /*if( c >= 0 )
- {*/
- pom = TNL::sign( c )*( hx * c )/( c - input[ e ]);
- if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) )
- output[ cell.getIndex() ] = pom;
-
- pom = pom - TNL::sign( c )*hx; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
- if( TNL::abs( output[ e ] ) > TNL::abs( pom ) )
- output[ e ] = pom;
- /*}else
- {
- pom = - (hx * c)/( c - input[ e ]);
- if( output[ cell.getIndex() ] < pom )
- output[ cell.getIndex() ] = pom;
-
- pom = pom + hx; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
- if( output[ e ] > pom )
- output[ e ] = pom;
- }*/
- interfaceMap[ cell.getIndex() ] = true;
- interfaceMap[ e ] = true;
- }
- }
+ pom = - ( hy * c )/( c - input[ n ]);
+ if( output[ cell.getIndex() ] < pom )
+ output[ cell.getIndex() ] = pom;
+ if( output[ n ] > hy + pom )
+ output[ n ] = hy + pom; //hy - ( hy * c )/( c - input[ n ]);
+ }*/
+ interfaceMap[ cell.getIndex() ] = true;
+ interfaceMap[ n ] = true;
}
+ if( c * input[ e ] <= 0 )
+ {
+ /*if( c >= 0 )
+ {*/
+ pom = TNL::sign( c )*( hx * c )/( c - input[ e ]);
+ if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) )
+ output[ cell.getIndex() ] = pom;
+
+ pom = pom - TNL::sign( c )*hx; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
+ if( TNL::abs( output[ e ] ) > TNL::abs( pom ) )
+ output[ e ] = pom;
+ /*}else
+ {
+ pom = - (hx * c)/( c - input[ e ]);
+ if( output[ cell.getIndex() ] < pom )
+ output[ cell.getIndex() ] = pom;
+
+ pom = pom + hx; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
+ if( output[ e ] > pom )
+ output[ e ] = pom;
+ }*/
+ interfaceMap[ cell.getIndex() ] = true;
+ interfaceMap[ e ] = true;
+ }
+ }
}
+ }
}
template< typename Real,
- typename Device,
- typename Index >
- template< typename MeshEntity >
+ typename Device,
+ typename Index >
+template< typename MeshEntity >
__cuda_callable__
void
tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
updateCell( MeshFunctionType& u,
- const MeshEntity& cell,
- const RealType v)
+ const MeshEntity& cell,
+ const RealType v)
{
- const auto& neighborEntities = cell.template getNeighborEntities< 2 >();
- const MeshType& mesh = cell.getMesh();
- const RealType& hx = mesh.getSpaceSteps().x();
- const RealType& hy = mesh.getSpaceSteps().y();
- const RealType value = u( cell );
- RealType a, b, tmp = std::numeric_limits< RealType >::max();
-
- if( cell.getCoordinates().x() == 0 )
- a = u[ neighborEntities.template getEntityIndex< 1, 0 >() ];
- else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 1 )
- a = u[ neighborEntities.template getEntityIndex< -1, 0 >() ];
- else
- {
- a = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< -1, 0 >() ],
- u[ neighborEntities.template getEntityIndex< 1, 0 >() ] );
- }
-
- if( cell.getCoordinates().y() == 0 )
- b = u[ neighborEntities.template getEntityIndex< 0, 1 >()];
- else if( cell.getCoordinates().y() == mesh.getDimensions().y() - 1 )
- b = u[ neighborEntities.template getEntityIndex< 0, -1 >() ];
- else
- {
- b = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< 0, -1 >() ],
- u[ neighborEntities.template getEntityIndex< 0, 1 >() ] );
- }
- if( fabs( a ) == std::numeric_limits< RealType >::max() &&
- fabs( b ) == std::numeric_limits< RealType >::max() )
- return;
- /*if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
- fabs( b ) == TypeInfo< Real >::getMaxValue() ||
- fabs( a - b ) >= TNL::sqrt( (hx * hx + hy * hy)/v ) )
+ const auto& neighborEntities = cell.template getNeighborEntities< 2 >();
+ const MeshType& mesh = cell.getMesh();
+ const RealType& hx = mesh.getSpaceSteps().x();
+ const RealType& hy = mesh.getSpaceSteps().y();
+ const RealType value = u( cell );
+ RealType a, b, tmp = std::numeric_limits< RealType >::max();
+
+ if( cell.getCoordinates().x() == 0 )
+ a = u[ neighborEntities.template getEntityIndex< 1, 0 >() ];
+ else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 1 )
+ a = u[ neighborEntities.template getEntityIndex< -1, 0 >() ];
+ else
+ {
+ a = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< -1, 0 >() ],
+ u[ neighborEntities.template getEntityIndex< 1, 0 >() ] );
+ }
+
+ if( cell.getCoordinates().y() == 0 )
+ b = u[ neighborEntities.template getEntityIndex< 0, 1 >()];
+ else if( cell.getCoordinates().y() == mesh.getDimensions().y() - 1 )
+ b = u[ neighborEntities.template getEntityIndex< 0, -1 >() ];
+ else
+ {
+ b = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< 0, -1 >() ],
+ u[ neighborEntities.template getEntityIndex< 0, 1 >() ] );
+ }
+ if( fabs( a ) == std::numeric_limits< RealType >::max() &&
+ fabs( b ) == std::numeric_limits< RealType >::max() )
+ return;
+ /*if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
+ fabs( b ) == TypeInfo< Real >::getMaxValue() ||
+ fabs( a - b ) >= TNL::sqrt( (hx * hx + hy * hy)/v ) )
{
- tmp =
- fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * hy :
- a + TNL::sign( value ) * hx;
+ tmp =
+ fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * hy :
+ a + TNL::sign( value ) * hx;
}*/
- /*if( fabs( a ) != TypeInfo< Real >::getMaxValue() &&
- fabs( b ) != TypeInfo< Real >::getMaxValue() &&
- fabs( a - b ) < TNL::sqrt( (hx * hx + hy * hy)/v ) )
+ /*if( fabs( a ) != TypeInfo< Real >::getMaxValue() &&
+ fabs( b ) != TypeInfo< Real >::getMaxValue() &&
+ fabs( a - b ) < TNL::sqrt( (hx * hx + hy * hy)/v ) )
{
- tmp = ( hx * hx * b + hy * hy * a +
- sign( value ) * hx * hy * TNL::sqrt( ( hx * hx + hy * hy )/v -
- ( a - b ) * ( a - b ) ) )/( hx * hx + hy * hy );
- u[ cell.getIndex() ] = tmp;
+ tmp = ( hx * hx * b + hy * hy * a +
+ sign( value ) * hx * hy * TNL::sqrt( ( hx * hx + hy * hy )/v -
+ ( a - b ) * ( a - b ) ) )/( hx * hx + hy * hy );
+ u[ cell.getIndex() ] = tmp;
}
else
{
- tmp =
- fabs( a ) > fabs( b ) ? b + TNL::sign( value ) * hy/v :
- a + TNL::sign( value ) * hx/v;
- u[ cell.getIndex() ] = argAbsMin( value, tmp );
- //tmp = TypeInfo< RealType >::getMaxValue();
+ tmp =
+ fabs( a ) > fabs( b ) ? b + TNL::sign( value ) * hy/v :
+ a + TNL::sign( value ) * hx/v;
+ u[ cell.getIndex() ] = argAbsMin( value, tmp );
+ //tmp = TypeInfo< RealType >::getMaxValue();
}*/
- RealType pom[6] = { a, b, std::numeric_limits< RealType >::max(), (RealType)hx, (RealType)hy, 0.0 };
- sortMinims( pom );
- tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]/v;
-
-
- if( fabs( tmp ) < fabs( pom[ 1 ] ) )
- u[ cell.getIndex() ] = argAbsMin( value, tmp );
- else
- {
- tmp = ( pom[ 3 ] * pom[ 3 ] * pom[ 1 ] + pom[ 4 ] * pom[ 4 ] * pom[ 0 ] +
+ RealType pom[6] = { a, b, std::numeric_limits< RealType >::max(), (RealType)hx, (RealType)hy, 0.0 };
+ sortMinims( pom );
+ tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]/v;
+
+
+ if( fabs( tmp ) < fabs( pom[ 1 ] ) )
+ u[ cell.getIndex() ] = argAbsMin( value, tmp );
+ else
+ {
+ tmp = ( pom[ 3 ] * pom[ 3 ] * pom[ 1 ] + pom[ 4 ] * pom[ 4 ] * pom[ 0 ] +
TNL::sign( value ) * pom[ 3 ] * pom[ 4 ] * TNL::sqrt( ( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] )/( v * v ) -
( pom[ 1 ] - pom[ 0 ] ) * ( pom[ 1 ] - pom[ 0 ] ) ) )/( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] );
- u[ cell.getIndex() ] = argAbsMin( value, tmp );
- }
+ u[ cell.getIndex() ] = argAbsMin( value, tmp );
+ }
}
template< typename Real,
- typename Device,
- typename Index >
+ typename Device,
+ typename Index >
void
tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
initInterface( const MeshFunctionPointer& _input,
- MeshFunctionPointer& _output,
- InterfaceMapPointer& _interfaceMap )
+ MeshFunctionPointer& _output,
+ InterfaceMapPointer& _interfaceMap )
{
- if( std::is_same< Device, Devices::Cuda >::value )
- {
+ if( std::is_same< Device, Devices::Cuda >::value )
+ {
#ifdef HAVE_CUDA
- const MeshType& mesh = _input->getMesh();
-
- const int cudaBlockSize( 8 );
- int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().x(), cudaBlockSize );
- int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().y(), cudaBlockSize );
- int numBlocksZ = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().z(), cudaBlockSize );
- if( cudaBlockSize * cudaBlockSize * cudaBlockSize > 1024 || numBlocksX > 1024 || numBlocksY > 1024 || numBlocksZ > 64 )
- std::cout << "Invalid kernel call. Dimensions of grid are max: [1024,1024,64], and maximum threads per block are 1024!" << std::endl;
- dim3 blockSize( cudaBlockSize, cudaBlockSize, cudaBlockSize );
- dim3 gridSize( numBlocksX, numBlocksY, numBlocksZ );
- Devices::Cuda::synchronizeDevice();
- CudaInitCaller3d<<< gridSize, blockSize >>>( _input.template getData< Device >(),
- _output.template modifyData< Device >(),
- _interfaceMap.template modifyData< Device >() );
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
+ const MeshType& mesh = _input->getMesh();
+
+ const int cudaBlockSize( 8 );
+ int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().x(), cudaBlockSize );
+ int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().y(), cudaBlockSize );
+ int numBlocksZ = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().z(), cudaBlockSize );
+ if( cudaBlockSize * cudaBlockSize * cudaBlockSize > 1024 || numBlocksX > 1024 || numBlocksY > 1024 || numBlocksZ > 64 )
+ std::cout << "Invalid kernel call. Dimensions of grid are max: [1024,1024,64], and maximum threads per block are 1024!" << std::endl;
+ dim3 blockSize( cudaBlockSize, cudaBlockSize, cudaBlockSize );
+ dim3 gridSize( numBlocksX, numBlocksY, numBlocksZ );
+ Devices::Cuda::synchronizeDevice();
+ CudaInitCaller3d<<< gridSize, blockSize >>>( _input.template getData< Device >(),
+ _output.template modifyData< Device >(),
+ _interfaceMap.template modifyData< Device >() );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
#endif
- }
- if( std::is_same< Device, Devices::Host >::value )
- {
- const MeshFunctionType& input = _input.getData();
- MeshFunctionType& output = _output.modifyData();
- InterfaceMapType& interfaceMap = _interfaceMap.modifyData();
- const MeshType& mesh = input.getMesh();
- typedef typename MeshType::Cell Cell;
- Cell cell( mesh );
- for( cell.getCoordinates().z() = 0;
- cell.getCoordinates().z() < mesh.getDimensions().z();
- cell.getCoordinates().z() ++ )
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh.getDimensions().y();
- cell.getCoordinates().y() ++ )
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh.getDimensions().x();
- cell.getCoordinates().x() ++ )
- {
- cell.refresh();
- output[ cell.getIndex() ] =
- input( cell ) > 0 ? std::numeric_limits< RealType >::max() :
- - std::numeric_limits< RealType >::max();
- interfaceMap[ cell.getIndex() ] = false;
- }
-
- const RealType& hx = mesh.getSpaceSteps().x();
- const RealType& hy = mesh.getSpaceSteps().y();
- const RealType& hz = mesh.getSpaceSteps().z();
- for( cell.getCoordinates().z() = 0;
- cell.getCoordinates().z() < mesh.getDimensions().z();
- cell.getCoordinates().z() ++ )
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh.getDimensions().y();
- cell.getCoordinates().y() ++ )
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh.getDimensions().x();
- cell.getCoordinates().x() ++ )
+ }
+ if( std::is_same< Device, Devices::Host >::value )
+ {
+ const MeshFunctionType& input = _input.getData();
+ MeshFunctionType& output = _output.modifyData();
+ InterfaceMapType& interfaceMap = _interfaceMap.modifyData();
+ const MeshType& mesh = input.getMesh();
+ typedef typename MeshType::Cell Cell;
+ Cell cell( mesh );
+ for( cell.getCoordinates().z() = 0;
+ cell.getCoordinates().z() < mesh.getDimensions().z();
+ cell.getCoordinates().z() ++ )
+ for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh.getDimensions().y();
+ cell.getCoordinates().y() ++ )
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh.getDimensions().x();
+ cell.getCoordinates().x() ++ )
+ {
+ cell.refresh();
+ output[ cell.getIndex() ] =
+ input( cell ) > 0 ? 10://std::numeric_limits< RealType >::max() :
+ -10;//- std::numeric_limits< RealType >::max();
+ interfaceMap[ cell.getIndex() ] = false;
+ }
+
+ const RealType& hx = mesh.getSpaceSteps().x();
+ const RealType& hy = mesh.getSpaceSteps().y();
+ const RealType& hz = mesh.getSpaceSteps().z();
+ for( cell.getCoordinates().z() = 0;
+ cell.getCoordinates().z() < mesh.getDimensions().z();
+ cell.getCoordinates().z() ++ )
+ for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh.getDimensions().y();
+ cell.getCoordinates().y() ++ )
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh.getDimensions().x();
+ cell.getCoordinates().x() ++ )
+ {
+ cell.refresh();
+ const RealType& c = input( cell );
+ if( ! cell.isBoundaryEntity() )
+ {
+ auto neighbors = cell.getNeighborEntities();
+ Real pom = 0;
+ const IndexType e = neighbors.template getEntityIndex< 1, 0, 0 >();
+ const IndexType n = neighbors.template getEntityIndex< 0, 1, 0 >();
+ const IndexType t = neighbors.template getEntityIndex< 0, 0, 1 >();
+ //Try exact initiation
+ /*const IndexType w = neighbors.template getEntityIndex< -1, 0, 0 >();
+ const IndexType s = neighbors.template getEntityIndex< 0, -1, 0 >();
+ const IndexType b = neighbors.template getEntityIndex< 0, 0, -1 >();
+ if( c * input[ e ] <= 0 )
+ {
+ output[ cell.getIndex() ] = c;
+ output[ e ] = input[ e ];
+ interfaceMap[ e ] = true;
+ interfaceMap[ cell.getIndex() ] = true;
+ }
+ else if( c * input[ n ] <= 0 )
+ {
+ output[ cell.getIndex() ] = c;
+ output[ n ] = input[ n ];
+ interfaceMap[ n ] = true;
+ interfaceMap[ cell.getIndex() ] = true;
+ }
+ else if( c * input[ t ] <= 0 )
+ {
+ output[ cell.getIndex() ] = c;
+ output[ t ] = input[ t ];
+ interfaceMap[ t ] = true;
+ interfaceMap[ cell.getIndex() ] = true;
+ }*/
+ if( c * input[ n ] <= 0 )
+ {
+ if( c >= 0 )
{
- cell.refresh();
- const RealType& c = input( cell );
- if( ! cell.isBoundaryEntity() )
- {
- auto neighbors = cell.getNeighborEntities();
- Real pom = 0;
- const IndexType e = neighbors.template getEntityIndex< 1, 0, 0 >();
- const IndexType n = neighbors.template getEntityIndex< 0, 1, 0 >();
- const IndexType t = neighbors.template getEntityIndex< 0, 0, 1 >();
- //Try exact initiation
- /*const IndexType w = neighbors.template getEntityIndex< -1, 0, 0 >();
- const IndexType s = neighbors.template getEntityIndex< 0, -1, 0 >();
- const IndexType b = neighbors.template getEntityIndex< 0, 0, -1 >();
- if( c * input[ e ] <= 0 )
- {
- output[ cell.getIndex() ] = c;
- output[ e ] = input[ e ];
- interfaceMap[ e ] = true;
- interfaceMap[ cell.getIndex() ] = true;
- }
- else if( c * input[ n ] <= 0 )
- {
- output[ cell.getIndex() ] = c;
- output[ n ] = input[ n ];
- interfaceMap[ n ] = true;
- interfaceMap[ cell.getIndex() ] = true;
- }
- else if( c * input[ t ] <= 0 )
- {
- output[ cell.getIndex() ] = c;
- output[ t ] = input[ t ];
- interfaceMap[ t ] = true;
- interfaceMap[ cell.getIndex() ] = true;
- }*/
- if( c * input[ n ] <= 0 )
- {
- if( c >= 0 )
- {
- pom = ( hy * c )/( c - input[ n ]);
- if( output[ cell.getIndex() ] > pom )
- output[ cell.getIndex() ] = pom;
-
- if ( output[ n ] < pom - hy)
- output[ n ] = pom - hy; // ( hy * c )/( c - input[ n ]) - hy;
-
- }else
- {
- pom = - ( hy * c )/( c - input[ n ]);
- if( output[ cell.getIndex() ] < pom )
- output[ cell.getIndex() ] = pom;
- if( output[ n ] > hy + pom )
- output[ n ] = hy + pom; //hy - ( hy * c )/( c - input[ n ]);
-
- }
- interfaceMap[ cell.getIndex() ] = true;
- interfaceMap[ n ] = true;
- }
- if( c * input[ e ] <= 0 )
- {
- if( c >= 0 )
- {
- pom = ( hx * c )/( c - input[ e ]);
- if( output[ cell.getIndex() ] > pom )
- output[ cell.getIndex() ] = pom;
-
- pom = pom - hx; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
- if( output[ e ] < pom )
- output[ e ] = pom;
-
- }else
- {
- pom = - (hx * c)/( c - input[ e ]);
- if( output[ cell.getIndex() ] < pom )
- output[ cell.getIndex() ] = pom;
-
- pom = pom + hx; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
- if( output[ e ] > pom )
- output[ e ] = pom;
- }
- interfaceMap[ cell.getIndex() ] = true;
- interfaceMap[ e ] = true;
- }
- if( c * input[ t ] <= 0 )
- {
- if( c >= 0 )
- {
- pom = ( hz * c )/( c - input[ t ]);
- if( output[ cell.getIndex() ] > pom )
- output[ cell.getIndex() ] = pom;
-
- pom = pom - hz; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
- if( output[ t ] < pom )
- output[ t ] = pom;
-
- }else
- {
- pom = - (hz * c)/( c - input[ t ]);
- if( output[ cell.getIndex() ] < pom )
- output[ cell.getIndex() ] = pom;
-
- pom = pom + hz; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
- if( output[ t ] > pom )
- output[ t ] = pom;
-
- }
- interfaceMap[ cell.getIndex() ] = true;
- interfaceMap[ t ] = true;
- }
- }
- /*output[ cell.getIndex() ] =
- c > 0 ? TypeInfo< RealType >::getMaxValue() :
- -TypeInfo< RealType >::getMaxValue();
- interfaceMap[ cell.getIndex() ] = false;*/ //is on line 245
+ pom = ( hy * c )/( c - input[ n ]);
+ if( output[ cell.getIndex() ] > pom )
+ output[ cell.getIndex() ] = pom;
+
+ if ( output[ n ] < pom - hy)
+ output[ n ] = pom - hy; // ( hy * c )/( c - input[ n ]) - hy;
+
+ }else
+ {
+ pom = - ( hy * c )/( c - input[ n ]);
+ if( output[ cell.getIndex() ] < pom )
+ output[ cell.getIndex() ] = pom;
+ if( output[ n ] > hy + pom )
+ output[ n ] = hy + pom; //hy - ( hy * c )/( c - input[ n ]);
+
}
- }
+ interfaceMap[ cell.getIndex() ] = true;
+ interfaceMap[ n ] = true;
+ }
+ if( c * input[ e ] <= 0 )
+ {
+ if( c >= 0 )
+ {
+ pom = ( hx * c )/( c - input[ e ]);
+ if( output[ cell.getIndex() ] > pom )
+ output[ cell.getIndex() ] = pom;
+
+ pom = pom - hx; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
+ if( output[ e ] < pom )
+ output[ e ] = pom;
+
+ }else
+ {
+ pom = - (hx * c)/( c - input[ e ]);
+ if( output[ cell.getIndex() ] < pom )
+ output[ cell.getIndex() ] = pom;
+
+ pom = pom + hx; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
+ if( output[ e ] > pom )
+ output[ e ] = pom;
+ }
+ interfaceMap[ cell.getIndex() ] = true;
+ interfaceMap[ e ] = true;
+ }
+ if( c * input[ t ] <= 0 )
+ {
+ if( c >= 0 )
+ {
+ pom = ( hz * c )/( c - input[ t ]);
+ if( output[ cell.getIndex() ] > pom )
+ output[ cell.getIndex() ] = pom;
+
+ pom = pom - hz; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
+ if( output[ t ] < pom )
+ output[ t ] = pom;
+
+ }else
+ {
+ pom = - (hz * c)/( c - input[ t ]);
+ if( output[ cell.getIndex() ] < pom )
+ output[ cell.getIndex() ] = pom;
+
+ pom = pom + hz; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
+ if( output[ t ] > pom )
+ output[ t ] = pom;
+
+ }
+ interfaceMap[ cell.getIndex() ] = true;
+ interfaceMap[ t ] = true;
+ }
+ }
+ /*output[ cell.getIndex() ] =
+ c > 0 ? TypeInfo< RealType >::getMaxValue() :
+ -TypeInfo< RealType >::getMaxValue();
+ interfaceMap[ cell.getIndex() ] = false;*/ //is on line 245
+ }
+ }
}
template< typename Real,
- typename Device,
- typename Index >
- template< typename MeshEntity >
+ typename Device,
+ typename Index >
+template< typename MeshEntity >
__cuda_callable__
void
tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
updateCell( MeshFunctionType& u,
- const MeshEntity& cell,
- const RealType v )
+ const MeshEntity& cell,
+ const RealType v )
{
- const auto& neighborEntities = cell.template getNeighborEntities< 3 >();
- const MeshType& mesh = cell.getMesh();
+ const auto& neighborEntities = cell.template getNeighborEntities< 3 >();
+ const MeshType& mesh = cell.getMesh();
- const RealType& hx = mesh.getSpaceSteps().x();
- const RealType& hy = mesh.getSpaceSteps().y();
- const RealType& hz = mesh.getSpaceSteps().z();
- const RealType value = u( cell );
- //std::cout << value << std::endl;
- RealType a, b, c, tmp = std::numeric_limits< RealType >::max();
-
-
- if( cell.getCoordinates().x() == 0 )
- a = u[ neighborEntities.template getEntityIndex< 1, 0, 0 >() ];
- else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 1 )
- a = u[ neighborEntities.template getEntityIndex< -1, 0, 0 >() ];
- else
+ const RealType& hx = mesh.getSpaceSteps().x();
+ const RealType& hy = mesh.getSpaceSteps().y();
+ const RealType& hz = mesh.getSpaceSteps().z();
+ const RealType value = u( cell );
+ //std::cout << value << std::endl;
+ RealType a, b, c, tmp = std::numeric_limits< RealType >::max();
+
+
+ if( cell.getCoordinates().x() == 0 )
+ a = u[ neighborEntities.template getEntityIndex< 1, 0, 0 >() ];
+ else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 1 )
+ a = u[ neighborEntities.template getEntityIndex< -1, 0, 0 >() ];
+ else
+ {
+ a = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< -1, 0, 0 >() ],
+ u[ neighborEntities.template getEntityIndex< 1, 0, 0 >() ] );
+ }
+ if( cell.getCoordinates().y() == 0 )
+ b = u[ neighborEntities.template getEntityIndex< 0, 1, 0 >() ];
+ else if( cell.getCoordinates().y() == mesh.getDimensions().y() - 1 )
+ b = u[ neighborEntities.template getEntityIndex< 0, -1, 0 >() ];
+ else
+ {
+ b = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< 0, -1, 0 >() ],
+ u[ neighborEntities.template getEntityIndex< 0, 1, 0 >() ] );
+ }if( cell.getCoordinates().z() == 0 )
+ c = u[ neighborEntities.template getEntityIndex< 0, 0, 1 >() ];
+ else if( cell.getCoordinates().z() == mesh.getDimensions().z() - 1 )
+ c = u[ neighborEntities.template getEntityIndex< 0, 0, -1 >() ];
+ else
+ {
+ c = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< 0, 0, -1 >() ],
+ u[ neighborEntities.template getEntityIndex< 0, 0, 1 >() ] );
+ }
+ if( fabs( a ) == std::numeric_limits< RealType >::max() &&
+ fabs( b ) == std::numeric_limits< RealType >::max() &&
+ fabs( c ) == std::numeric_limits< RealType >::max() )
+ return;
+
+
+ /*if( fabs( a ) != TypeInfo< Real >::getMaxValue() &&
+ fabs( b ) != TypeInfo< Real >::getMaxValue() &&
+ fabs( a - b ) >= TNL::sqrt( (hx * hx + hy * hy)/v ) )
{
- a = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< -1, 0, 0 >() ],
- u[ neighborEntities.template getEntityIndex< 1, 0, 0 >() ] );
+ tmp = ( hx * hx * a + hy * hy * b +
+ sign( value ) * hx * hy * sqrt( ( hx * hx + hy * hy )/v -
+ ( a - b ) * ( a - b ) ) )/( hx * hx + hy * hy );
}
- if( cell.getCoordinates().y() == 0 )
- b = u[ neighborEntities.template getEntityIndex< 0, 1, 0 >() ];
- else if( cell.getCoordinates().y() == mesh.getDimensions().y() - 1 )
- b = u[ neighborEntities.template getEntityIndex< 0, -1, 0 >() ];
- else
- {
- b = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< 0, -1, 0 >() ],
- u[ neighborEntities.template getEntityIndex< 0, 1, 0 >() ] );
- }if( cell.getCoordinates().z() == 0 )
- c = u[ neighborEntities.template getEntityIndex< 0, 0, 1 >() ];
- else if( cell.getCoordinates().z() == mesh.getDimensions().z() - 1 )
- c = u[ neighborEntities.template getEntityIndex< 0, 0, -1 >() ];
- else
+ if( fabs( a ) != TypeInfo< Real >::getMaxValue() &&
+ fabs( c ) != TypeInfo< Real >::getMaxValue() &&
+ fabs( a - c ) >= TNL::sqrt( (hx * hx + hz * hz)/v ) )
{
- c = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< 0, 0, -1 >() ],
- u[ neighborEntities.template getEntityIndex< 0, 0, 1 >() ] );
+ tmp = ( hx * hx * a + hz * hz * c +
+ sign( value ) * hx * hz * sqrt( ( hx * hx + hz * hz )/v -
+ ( a - c ) * ( a - c ) ) )/( hx * hx + hz * hz );
}
- if( fabs( a ) == std::numeric_limits< RealType >::max() &&
- fabs( b ) == std::numeric_limits< RealType >::max() &&
- fabs( c ) == std::numeric_limits< RealType >::max() )
- return;
-
-
- /*if( fabs( a ) != TypeInfo< Real >::getMaxValue() &&
- fabs( b ) != TypeInfo< Real >::getMaxValue() &&
- fabs( a - b ) >= TNL::sqrt( (hx * hx + hy * hy)/v ) )
- {
- tmp = ( hx * hx * a + hy * hy * b +
- sign( value ) * hx * hy * sqrt( ( hx * hx + hy * hy )/v -
- ( a - b ) * ( a - b ) ) )/( hx * hx + hy * hy );
- }
- if( fabs( a ) != TypeInfo< Real >::getMaxValue() &&
- fabs( c ) != TypeInfo< Real >::getMaxValue() &&
- fabs( a - c ) >= TNL::sqrt( (hx * hx + hz * hz)/v ) )
- {
- tmp = ( hx * hx * a + hz * hz * c +
- sign( value ) * hx * hz * sqrt( ( hx * hx + hz * hz )/v -
- ( a - c ) * ( a - c ) ) )/( hx * hx + hz * hz );
- }
- if( fabs( b ) != TypeInfo< Real >::getMaxValue() &&
- fabs( c ) != TypeInfo< Real >::getMaxValue() &&
- fabs( b - c ) >= TNL::sqrt( (hy * hy + hz * hz)/v ) )
- {
- tmp = ( hy * hy * b + hz * hz * c +
- sign( value ) * hy * hz * sqrt( ( hy * hy + hz * hz )/v -
- ( b - c ) * ( b - c ) ) )/( hy * hy + hz * hz );
- }*/
- RealType pom[6] = { a, b, c, (RealType)hx, (RealType)hy, (RealType)hz};
- sortMinims( pom );
- tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ];
- if( fabs( tmp ) < fabs( pom[ 1 ] ) )
+ if( fabs( b ) != TypeInfo< Real >::getMaxValue() &&
+ fabs( c ) != TypeInfo< Real >::getMaxValue() &&
+ fabs( b - c ) >= TNL::sqrt( (hy * hy + hz * hz)/v ) )
+ {
+ tmp = ( hy * hy * b + hz * hz * c +
+ sign( value ) * hy * hz * sqrt( ( hy * hy + hz * hz )/v -
+ ( b - c ) * ( b - c ) ) )/( hy * hy + hz * hz );
+ }*/
+ RealType pom[6] = { a, b, c, (RealType)hx, (RealType)hy, (RealType)hz};
+ sortMinims( pom );
+ tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ];
+ if( fabs( tmp ) < fabs( pom[ 1 ] ) )
+ {
+ u[ cell.getIndex() ] = argAbsMin( value, tmp );
+ }
+ else
+ {
+ tmp = ( pom[ 3 ] * pom[ 3 ] * pom[ 1 ] + pom[ 4 ] * pom[ 4 ] * pom[ 0 ] +
+ TNL::sign( value ) * pom[ 3 ] * pom[ 4 ] * TNL::sqrt( ( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] )/( v * v ) -
+ ( pom[ 1 ] - pom[ 0 ] ) * ( pom[ 1 ] - pom[ 0 ] ) ) )/( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] );
+ if( fabs( tmp ) < fabs( pom[ 2 ]) )
{
- u[ cell.getIndex() ] = argAbsMin( value, tmp );
+ u[ cell.getIndex() ] = argAbsMin( value, tmp );
}
else
{
- tmp = ( pom[ 3 ] * pom[ 3 ] * pom[ 1 ] + pom[ 4 ] * pom[ 4 ] * pom[ 0 ] +
- TNL::sign( value ) * pom[ 3 ] * pom[ 4 ] * TNL::sqrt( ( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] )/( v * v ) -
- ( pom[ 1 ] - pom[ 0 ] ) * ( pom[ 1 ] - pom[ 0 ] ) ) )/( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] );
- if( fabs( tmp ) < fabs( pom[ 2 ]) )
- {
- u[ cell.getIndex() ] = argAbsMin( value, tmp );
- }
- else
- {
- tmp = ( hy * hy * hz * hz * a + hx * hx * hz * hz * b + hx * hx * hy * hy * c +
- TNL::sign( value ) * hx * hy * hz * TNL::sqrt( ( hx * hx * hz * hz + hy * hy * hz * hz + hx * hx * hy * hy)/( v * v ) -
- hz * hz * ( a - b ) * ( a - b ) - hy * hy * ( a - c ) * ( a - c ) -
- hx * hx * ( b - c ) * ( b - c ) ) )/( hx * hx * hy * hy + hy * hy * hz * hz + hz * hz * hx *hx );
- u[ cell.getIndex() ] = argAbsMin( value, tmp );
- }
+ tmp = ( hy * hy * hz * hz * a + hx * hx * hz * hz * b + hx * hx * hy * hy * c +
+ TNL::sign( value ) * hx * hy * hz * TNL::sqrt( ( hx * hx * hz * hz + hy * hy * hz * hz + hx * hx * hy * hy)/( v * v ) -
+ hz * hz * ( a - b ) * ( a - b ) - hy * hy * ( a - c ) * ( a - c ) -
+ hx * hx * ( b - c ) * ( b - c ) ) )/( hx * hx * hy * hy + hy * hy * hz * hz + hz * hz * hx *hx );
+ u[ cell.getIndex() ] = argAbsMin( value, tmp );
}
+ }
}
template < typename T1, typename T2 >
T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double v)
{
- T1 tmp;
- if( fabs( a ) != std::numeric_limits< T1 >::max &&
- fabs( b ) != std::numeric_limits< T1 >::max &&
- fabs( a - b ) < ha/v )//TNL::sqrt( (ha * ha + hb * hb)/2 )/v )
- {
- tmp = ( ha * ha * b + hb * hb * a +
+ T1 tmp;
+ if( fabs( a ) != std::numeric_limits< T1 >::max &&
+ fabs( b ) != std::numeric_limits< T1 >::max &&
+ fabs( a - b ) < ha/v )//TNL::sqrt( (ha * ha + hb * hb)/2 )/v )
+ {
+ tmp = ( ha * ha * b + hb * hb * a +
TNL::sign( value ) * ha * hb * TNL::sqrt( ( ha * ha + hb * hb )/( v * v ) -
( a - b ) * ( a - b ) ) )/( ha * ha + hb * hb );
- }
- else
- {
- tmp = std::numeric_limits< T1 >::max;
- }
-
- return tmp;
+ }
+ else
+ {
+ tmp = std::numeric_limits< T1 >::max;
+ }
+
+ return tmp;
}
template < typename T1 >
__cuda_callable__ void sortMinims( T1 pom[] )
{
- T1 tmp[6] = {0.0,0.0,0.0,0.0,0.0,0.0};
- if( fabs(pom[0]) <= fabs(pom[1]) && fabs(pom[1]) <= fabs(pom[2])){
- tmp[0] = pom[0]; tmp[1] = pom[1]; tmp[2] = pom[2];
- tmp[3] = pom[3]; tmp[4] = pom[4]; tmp[5] = pom[5];
-
- }
- else if( fabs(pom[0]) <= fabs(pom[2]) && fabs(pom[2]) <= fabs(pom[1]) ){
- tmp[0] = pom[0]; tmp[1] = pom[2]; tmp[2] = pom[1];
- tmp[3] = pom[3]; tmp[4] = pom[5]; tmp[5] = pom[4];
- }
- else if( fabs(pom[1]) <= fabs(pom[0]) && fabs(pom[0]) <= fabs(pom[2]) ){
- tmp[0] = pom[1]; tmp[1] = pom[0]; tmp[2] = pom[2];
- tmp[3] = pom[4]; tmp[4] = pom[3]; tmp[5] = pom[5];
- }
- else if( fabs(pom[1]) <= fabs(pom[2]) && fabs(pom[2]) <= fabs(pom[0]) ){
- tmp[0] = pom[1]; tmp[1] = pom[2]; tmp[2] = pom[0];
- tmp[3] = pom[4]; tmp[4] = pom[5]; tmp[5] = pom[3];
- }
- else if( fabs(pom[2]) <= fabs(pom[0]) && fabs(pom[0]) <= fabs(pom[1]) ){
- tmp[0] = pom[2]; tmp[1] = pom[0]; tmp[2] = pom[1];
- tmp[3] = pom[5]; tmp[4] = pom[3]; tmp[5] = pom[4];
- }
- else if( fabs(pom[2]) <= fabs(pom[1]) && fabs(pom[1]) <= fabs(pom[0]) ){
- tmp[0] = pom[2]; tmp[1] = pom[1]; tmp[2] = pom[0];
- tmp[3] = pom[5]; tmp[4] = pom[4]; tmp[5] = pom[3];
- }
+ T1 tmp[6] = {0.0,0.0,0.0,0.0,0.0,0.0};
+ if( fabs(pom[0]) <= fabs(pom[1]) && fabs(pom[1]) <= fabs(pom[2])){
+ tmp[0] = pom[0]; tmp[1] = pom[1]; tmp[2] = pom[2];
+ tmp[3] = pom[3]; tmp[4] = pom[4]; tmp[5] = pom[5];
- for( int i = 0; i < 6; i++ )
- {
- pom[ i ] = tmp[ i ];
- }
+ }
+ else if( fabs(pom[0]) <= fabs(pom[2]) && fabs(pom[2]) <= fabs(pom[1]) ){
+ tmp[0] = pom[0]; tmp[1] = pom[2]; tmp[2] = pom[1];
+ tmp[3] = pom[3]; tmp[4] = pom[5]; tmp[5] = pom[4];
+ }
+ else if( fabs(pom[1]) <= fabs(pom[0]) && fabs(pom[0]) <= fabs(pom[2]) ){
+ tmp[0] = pom[1]; tmp[1] = pom[0]; tmp[2] = pom[2];
+ tmp[3] = pom[4]; tmp[4] = pom[3]; tmp[5] = pom[5];
+ }
+ else if( fabs(pom[1]) <= fabs(pom[2]) && fabs(pom[2]) <= fabs(pom[0]) ){
+ tmp[0] = pom[1]; tmp[1] = pom[2]; tmp[2] = pom[0];
+ tmp[3] = pom[4]; tmp[4] = pom[5]; tmp[5] = pom[3];
+ }
+ else if( fabs(pom[2]) <= fabs(pom[0]) && fabs(pom[0]) <= fabs(pom[1]) ){
+ tmp[0] = pom[2]; tmp[1] = pom[0]; tmp[2] = pom[1];
+ tmp[3] = pom[5]; tmp[4] = pom[3]; tmp[5] = pom[4];
+ }
+ else if( fabs(pom[2]) <= fabs(pom[1]) && fabs(pom[1]) <= fabs(pom[0]) ){
+ tmp[0] = pom[2]; tmp[1] = pom[1]; tmp[2] = pom[0];
+ tmp[3] = pom[5]; tmp[4] = pom[4]; tmp[5] = pom[3];
+ }
+
+ for( int i = 0; i < 6; i++ )
+ {
+ pom[ i ] = tmp[ i ];
+ }
}
-
+template< typename Real,
+ typename Device,
+ typename Index >
+template< int sizeSArray >
+__cuda_callable__
+bool
+tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
+updateCell( volatile Real *sArray, int thri, int thrj, const Real hx, const Real hy,
+ const Real v )
+{
+ const RealType value = sArray[ thrj * sizeSArray + thri ];
+ RealType a, b, tmp = std::numeric_limits< RealType >::max();
+
+ b = TNL::argAbsMin( sArray[ (thrj+1) * sizeSArray + thri ],
+ sArray[ (thrj-1) * sizeSArray + thri ] );
+
+ a = TNL::argAbsMin( sArray[ thrj * sizeSArray + thri+1 ],
+ sArray[ thrj * sizeSArray + thri-1 ] );
+
+ if( fabs( a ) == std::numeric_limits< RealType >::max() &&
+ fabs( b ) == std::numeric_limits< RealType >::max() )
+ return false;
+
+ RealType pom[6] = { a, b, std::numeric_limits< RealType >::max(), (RealType)hx, (RealType)hy, 0.0 };
+ sortMinims( pom );
+ tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]/v;
+
+
+ if( fabs( tmp ) < fabs( pom[ 1 ] ) )
+ {
+ sArray[ thrj * sizeSArray + thri ] = argAbsMin( value, tmp );
+ tmp = value - sArray[ thrj * sizeSArray + thri ];
+ if ( fabs( tmp ) > 0.001*hx )
+ return true;
+ else
+ return false;
+ }
+ else
+ {
+ tmp = ( pom[ 3 ] * pom[ 3 ] * pom[ 1 ] + pom[ 4 ] * pom[ 4 ] * pom[ 0 ] +
+ TNL::sign( value ) * pom[ 3 ] * pom[ 4 ] * TNL::sqrt( ( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] )/( v * v ) -
+ ( pom[ 1 ] - pom[ 0 ] ) * ( pom[ 1 ] - pom[ 0 ] ) ) )/( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] );
+ sArray[ thrj * sizeSArray + thri ] = argAbsMin( value, tmp );
+ tmp = value - sArray[ thrj * sizeSArray + thri ];
+ if ( fabs( tmp ) > 0.001*hx )
+ return true;
+ else
+ return false;
+ }
+
+ return false;
+}
+template< typename Real,
+ typename Device,
+ typename Index >
+template< int sizeSArray >
+__cuda_callable__
+bool
+tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
+updateCell3D( volatile Real *sArray, int thri, int thrj, int thrk,
+ const Real hx, const Real hy, const Real hz, const Real v )
+{
+ const RealType value = sArray[thrk *sizeSArray * sizeSArray + thrj * sizeSArray + thri];
+
+ RealType a, b, c, tmp = std::numeric_limits< RealType >::max();
+
+ c = TNL::argAbsMin( sArray[ (thrk+1)* sizeSArray*sizeSArray + thrj * sizeSArray + thri ],
+ sArray[ (thrk-1) * sizeSArray *sizeSArray + thrj* sizeSArray + thri ] );
+
+ b = TNL::argAbsMin( sArray[ thrk* sizeSArray*sizeSArray + (thrj+1) * sizeSArray + thri ],
+ sArray[ thrk* sizeSArray * sizeSArray + (thrj-1)* sizeSArray +thri ] );
+
+ a = TNL::argAbsMin( sArray[ thrk* sizeSArray* sizeSArray + thrj* sizeSArray + thri+1 ],
+ sArray[ thrk* sizeSArray * sizeSArray + thrj* sizeSArray +thri-1 ] );
+
+ /*if( thrk == 8 )
+ printf("Calculating a = %f, b = %f, c = %f\n" , a, b, c );*/
+
+ if( fabs( a ) == 10&& //std::numeric_limits< RealType >::max() &&
+ fabs( b ) == 10&&//std::numeric_limits< RealType >::max() &&
+ fabs( c ) == 10)//std::numeric_limits< RealType >::max() )
+ return false;
+
+ RealType pom[6] = { a, b, c, (RealType)hx, (RealType)hy, (RealType)hz};
+
+ sortMinims( pom );
+
+ tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ];
+ if( fabs( tmp ) < fabs( pom[ 1 ] ) )
+ {
+ sArray[ thrk* sizeSArray* sizeSArray + thrj* sizeSArray + thri ] = argAbsMin( value, tmp );
+ tmp = value - sArray[ thrk* sizeSArray* sizeSArray + thrj* sizeSArray + thri ];
+ if ( fabs( tmp ) > 0.001*hx )
+ return true;
+ else
+ return false;
+ }
+ else
+ {
+ tmp = ( pom[ 3 ] * pom[ 3 ] * pom[ 1 ] + pom[ 4 ] * pom[ 4 ] * pom[ 0 ] +
+ TNL::sign( value ) * pom[ 3 ] * pom[ 4 ] * TNL::sqrt( ( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] )/( v * v ) -
+ ( pom[ 1 ] - pom[ 0 ] ) * ( pom[ 1 ] - pom[ 0 ] ) ) )/( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] );
+ if( fabs( tmp ) < fabs( pom[ 2 ]) )
+ {
+ sArray[ thrk* sizeSArray* sizeSArray + thrj* sizeSArray + thri ] = argAbsMin( value, tmp );
+ tmp = value - sArray[ thrk* sizeSArray* sizeSArray + thrj* sizeSArray + thri ];
+ if ( fabs( tmp ) > 0.001*hx )
+ return true;
+ else
+ return false;
+ }
+ else
+ {
+ tmp = ( hy * hy * hz * hz * a + hx * hx * hz * hz * b + hx * hx * hy * hy * c +
+ TNL::sign( value ) * hx * hy * hz * TNL::sqrt( ( hx * hx * hz * hz + hy * hy * hz * hz + hx * hx * hy * hy)/( v * v ) -
+ hz * hz * ( a - b ) * ( a - b ) - hy * hy * ( a - c ) * ( a - c ) -
+ hx * hx * ( b - c ) * ( b - c ) ) )/( hx * hx * hy * hy + hy * hy * hz * hz + hz * hz * hx *hx );
+ sArray[ thrk* sizeSArray* sizeSArray + thrj* sizeSArray + thri ] = argAbsMin( value, tmp );
+ tmp = value - sArray[ thrk* sizeSArray* sizeSArray + thrj* sizeSArray + thri ];
+ if ( fabs( tmp ) > 0.001*hx )
+ return true;
+ else
+ return false;
+ }
+ }
+
+ return false;
+}
#ifdef HAVE_CUDA
template < typename Real, typename Device, typename Index >
__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index > >& input,
- Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index > >& output,
- Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index >, 1, bool >& interfaceMap )
+ Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index > >& output,
+ Functions::MeshFunction< Meshes::Grid< 1, Real, Device, Index >, 1, bool >& interfaceMap )
{
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- const Meshes::Grid< 1, Real, Device, Index >& mesh = input.template getMesh< Devices::Cuda >();
+ int i = threadIdx.x + blockDim.x*blockIdx.x;
+ const Meshes::Grid< 1, Real, Device, Index >& mesh = input.template getMesh< Devices::Cuda >();
+
+ if( i < mesh.getDimensions().x() )
+ {
+ typedef typename Meshes::Grid< 1, Real, Device, Index >::Cell Cell;
+ Cell cell( mesh );
+ cell.getCoordinates().x() = i;
+ cell.refresh();
+ const Index cind = cell.getIndex();
+
- if( i < mesh.getDimensions().x() )
+ output[ cind ] =
+ input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
+ - std::numeric_limits< Real >::max();
+ interfaceMap[ cind ] = false;
+
+ const Real& h = mesh.getSpaceSteps().x();
+ cell.refresh();
+ const Real& c = input( cell );
+ if( ! cell.isBoundaryEntity() )
{
- typedef typename Meshes::Grid< 1, Real, Device, Index >::Cell Cell;
- Cell cell( mesh );
- cell.getCoordinates().x() = i;
- cell.refresh();
- const Index cind = cell.getIndex();
-
-
- output[ cind ] =
- input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
- - std::numeric_limits< Real >::max();
- interfaceMap[ cind ] = false;
-
- const Real& h = mesh.getSpaceSteps().x();
- cell.refresh();
- const Real& c = input( cell );
- if( ! cell.isBoundaryEntity() )
- {
- auto neighbors = cell.getNeighborEntities();
- Real pom = 0;
- const Index e = neighbors.template getEntityIndex< 1 >();
- const Index w = neighbors.template getEntityIndex< -1 >();
- if( c * input[ e ] <= 0 )
- {
- pom = TNL::sign( c )*( h * c )/( c - input[ e ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- if( c * input[ w ] <= 0 )
- {
- pom = TNL::sign( c )*( h * c )/( c - input[ w ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- }
+ auto neighbors = cell.getNeighborEntities();
+ Real pom = 0;
+ const Index e = neighbors.template getEntityIndex< 1 >();
+ const Index w = neighbors.template getEntityIndex< -1 >();
+ if( c * input[ e ] <= 0 )
+ {
+ pom = TNL::sign( c )*( h * c )/( c - input[ e ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
+ if( c * input[ w ] <= 0 )
+ {
+ pom = TNL::sign( c )*( h * c )/( c - input[ w ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
}
-
+ }
+
}
template < typename Real, typename Device, typename Index >
__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& input,
- Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& output,
- Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap )
+ Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& output,
+ Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap )
{
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- int j = blockDim.y*blockIdx.y + threadIdx.y;
- const Meshes::Grid< 2, Real, Device, Index >& mesh = input.template getMesh< Devices::Cuda >();
+ int i = threadIdx.x + blockDim.x*blockIdx.x;
+ int j = blockDim.y*blockIdx.y + threadIdx.y;
+ const Meshes::Grid< 2, Real, Device, Index >& mesh = input.template getMesh< Devices::Cuda >();
+
+ if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+ {
+ typedef typename Meshes::Grid< 2, Real, Device, Index >::Cell Cell;
+ Cell cell( mesh );
+ cell.getCoordinates().x() = i; cell.getCoordinates().y() = j;
+ cell.refresh();
+ const Index cind = cell.getIndex();
+
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+ output[ cind ] =
+ input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
+ - std::numeric_limits< Real >::max();
+ interfaceMap[ cind ] = false;
+
+ const Real& hx = mesh.getSpaceSteps().x();
+ const Real& hy = mesh.getSpaceSteps().y();
+ cell.refresh();
+ const Real& c = input( cell );
+ if( ! cell.isBoundaryEntity() )
{
- typedef typename Meshes::Grid< 2, Real, Device, Index >::Cell Cell;
- Cell cell( mesh );
- cell.getCoordinates().x() = i; cell.getCoordinates().y() = j;
- cell.refresh();
- const Index cind = cell.getIndex();
-
-
- output[ cind ] =
- input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
- - std::numeric_limits< Real >::max();
- interfaceMap[ cind ] = false;
-
- const Real& hx = mesh.getSpaceSteps().x();
- const Real& hy = mesh.getSpaceSteps().y();
- cell.refresh();
- const Real& c = input( cell );
- if( ! cell.isBoundaryEntity() )
- {
- auto neighbors = cell.getNeighborEntities();
- Real pom = 0;
- const Index e = neighbors.template getEntityIndex< 1, 0 >();
- const Index w = neighbors.template getEntityIndex< -1, 0 >();
- const Index n = neighbors.template getEntityIndex< 0, 1 >();
- const Index s = neighbors.template getEntityIndex< 0, -1 >();
-
- if( c * input[ n ] <= 0 )
- {
- pom = TNL::sign( c )*( hy * c )/( c - input[ n ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cell.getIndex() ] = true;
- }
- if( c * input[ e ] <= 0 )
- {
- pom = TNL::sign( c )*( hx * c )/( c - input[ e ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- if( c * input[ w ] <= 0 )
- {
- pom = TNL::sign( c )*( hx * c )/( c - input[ w ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- if( c * input[ s ] <= 0 )
- {
- pom = TNL::sign( c )*( hy * c )/( c - input[ s ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- }
+ auto neighbors = cell.getNeighborEntities();
+ Real pom = 0;
+ const Index e = neighbors.template getEntityIndex< 1, 0 >();
+ const Index w = neighbors.template getEntityIndex< -1, 0 >();
+ const Index n = neighbors.template getEntityIndex< 0, 1 >();
+ const Index s = neighbors.template getEntityIndex< 0, -1 >();
+
+ if( c * input[ n ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hy * c )/( c - input[ n ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cell.getIndex() ] = true;
+ }
+ if( c * input[ e ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hx * c )/( c - input[ e ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
+ if( c * input[ w ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hx * c )/( c - input[ w ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
+ if( c * input[ s ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hy * c )/( c - input[ s ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
}
+ }
}
template < typename Real, typename Device, typename Index >
__global__ void CudaInitCaller3d( const Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& input,
- Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& output,
- Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index >, 3, bool >& interfaceMap )
+ Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& output,
+ Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index >, 3, bool >& interfaceMap )
{
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- int j = blockDim.y*blockIdx.y + threadIdx.y;
- int k = blockDim.z*blockIdx.z + threadIdx.z;
- const Meshes::Grid< 3, Real, Device, Index >& mesh = input.template getMesh< Devices::Cuda >();
+ int i = threadIdx.x + blockDim.x*blockIdx.x;
+ int j = blockDim.y*blockIdx.y + threadIdx.y;
+ int k = blockDim.z*blockIdx.z + threadIdx.z;
+ const Meshes::Grid< 3, Real, Device, Index >& mesh = input.template getMesh< Devices::Cuda >();
+
+ if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && k < mesh.getDimensions().z() )
+ {
+ typedef typename Meshes::Grid< 3, Real, Device, Index >::Cell Cell;
+ Cell cell( mesh );
+ cell.getCoordinates().x() = i; cell.getCoordinates().y() = j; cell.getCoordinates().z() = k;
+ cell.refresh();
+ const Index cind = cell.getIndex();
+
+
+ output[ cind ] =
+ input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
+ - std::numeric_limits< Real >::max();
+ interfaceMap[ cind ] = false;
+ cell.refresh();
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && k < mesh.getDimensions().z() )
+ const Real& hx = mesh.getSpaceSteps().x();
+ const Real& hy = mesh.getSpaceSteps().y();
+ const Real& hz = mesh.getSpaceSteps().z();
+ const Real& c = input( cell );
+ if( ! cell.isBoundaryEntity() )
{
- typedef typename Meshes::Grid< 3, Real, Device, Index >::Cell Cell;
- Cell cell( mesh );
- cell.getCoordinates().x() = i; cell.getCoordinates().y() = j; cell.getCoordinates().z() = k;
- cell.refresh();
- const Index cind = cell.getIndex();
-
-
- output[ cind ] =
- input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
- - std::numeric_limits< Real >::max();
- interfaceMap[ cind ] = false;
- cell.refresh();
-
- const Real& hx = mesh.getSpaceSteps().x();
- const Real& hy = mesh.getSpaceSteps().y();
- const Real& hz = mesh.getSpaceSteps().z();
- const Real& c = input( cell );
- if( ! cell.isBoundaryEntity() )
- {
- auto neighbors = cell.getNeighborEntities();
- Real pom = 0;
- const Index e = neighbors.template getEntityIndex< 1, 0, 0 >();
- const Index w = neighbors.template getEntityIndex< -1, 0, 0 >();
- const Index n = neighbors.template getEntityIndex< 0, 1, 0 >();
- const Index s = neighbors.template getEntityIndex< 0, -1, 0 >();
- const Index t = neighbors.template getEntityIndex< 0, 0, 1 >();
- const Index b = neighbors.template getEntityIndex< 0, 0, -1 >();
-
- if( c * input[ n ] <= 0 )
- {
- pom = TNL::sign( c )*( hy * c )/( c - input[ n ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- if( c * input[ e ] <= 0 )
- {
- pom = TNL::sign( c )*( hx * c )/( c - input[ e ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- if( c * input[ w ] <= 0 )
- {
- pom = TNL::sign( c )*( hx * c )/( c - input[ w ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- if( c * input[ s ] <= 0 )
- {
- pom = TNL::sign( c )*( hy * c )/( c - input[ s ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- if( c * input[ b ] <= 0 )
- {
- pom = TNL::sign( c )*( hz * c )/( c - input[ b ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- if( c * input[ t ] <= 0 )
- {
- pom = TNL::sign( c )*( hz * c )/( c - input[ t ]);
- if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
- output[ cind ] = pom;
-
- interfaceMap[ cind ] = true;
- }
- }
+ auto neighbors = cell.getNeighborEntities();
+ Real pom = 0;
+ const Index e = neighbors.template getEntityIndex< 1, 0, 0 >();
+ const Index w = neighbors.template getEntityIndex< -1, 0, 0 >();
+ const Index n = neighbors.template getEntityIndex< 0, 1, 0 >();
+ const Index s = neighbors.template getEntityIndex< 0, -1, 0 >();
+ const Index t = neighbors.template getEntityIndex< 0, 0, 1 >();
+ const Index b = neighbors.template getEntityIndex< 0, 0, -1 >();
+
+ if( c * input[ n ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hy * c )/( c - input[ n ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
+ if( c * input[ e ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hx * c )/( c - input[ e ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
+ if( c * input[ w ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hx * c )/( c - input[ w ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
+ if( c * input[ s ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hy * c )/( c - input[ s ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
+ if( c * input[ b ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hz * c )/( c - input[ b ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
+ if( c * input[ t ] <= 0 )
+ {
+ pom = TNL::sign( c )*( hz * c )/( c - input[ t ]);
+ if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+ output[ cind ] = pom;
+
+ interfaceMap[ cind ] = true;
+ }
}
+ }
}
-template< typename Real,
- typename Device,
- typename Index >
-__cuda_callable__
-bool
-tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
-updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, const Real hy,
- const Real v )
-{
- const RealType value = sArray[ thrj ][ thri ];
- RealType a, b, tmp = std::numeric_limits< RealType >::max();
-
- b = TNL::argAbsMin( sArray[ thrj+1 ][ thri ],
- sArray[ thrj-1 ][ thri ] );
-
- a = TNL::argAbsMin( sArray[ thrj ][ thri+1 ],
- sArray[ thrj ][ thri-1 ] );
- if( fabs( a ) == std::numeric_limits< RealType >::max() &&
- fabs( b ) == std::numeric_limits< RealType >::max() )
- return false;
-
- RealType pom[6] = { a, b, std::numeric_limits< RealType >::max(), (RealType)hx, (RealType)hy, 0.0 };
- sortMinims( pom );
- tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]/v;
-
-
- if( fabs( tmp ) < fabs( pom[ 1 ] ) )
- {
- sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
- tmp = value - sArray[ thrj ][ thri ];
- if ( fabs( tmp ) > 0.01*hx )
- return true;
- else
- return false;
- }
- else
- {
- tmp = ( pom[ 3 ] * pom[ 3 ] * pom[ 1 ] + pom[ 4 ] * pom[ 4 ] * pom[ 0 ] +
- TNL::sign( value ) * pom[ 3 ] * pom[ 4 ] * TNL::sqrt( ( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] )/( v * v ) -
- ( pom[ 1 ] - pom[ 0 ] ) * ( pom[ 1 ] - pom[ 0 ] ) ) )/( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] );
- sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
- tmp = value - sArray[ thrj ][ thri ];
- if ( fabs( tmp ) > 0.01*hx )
- return true;
- else
- return false;
- }
-
- return false;
-}
template< typename Real,
- typename Device,
- typename Index >
+ typename Device,
+ typename Index >
__cuda_callable__
bool
tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >::
updateCell( volatile Real sArray[18], int thri, const Real h, const Real v )
{
- const RealType value = sArray[ thri ];
- RealType a, tmp = std::numeric_limits< RealType >::max();
-
- a = TNL::argAbsMin( sArray[ thri+1 ],
- sArray[ thri-1 ] );
-
- if( fabs( a ) == std::numeric_limits< RealType >::max() )
- return false;
-
- tmp = a + TNL::sign( value ) * h/v;
-
-
- sArray[ thri ] = argAbsMin( value, tmp );
-
- tmp = value - sArray[ thri ];
- if ( fabs( tmp ) > 0.01*h )
- return true;
- else
- return false;
-}
-
-template< typename Real,
- typename Device,
- typename Index >
-__cuda_callable__
-bool
-tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
-updateCell( volatile Real sArray[10][10][10], int thri, int thrj, int thrk,
- const Real hx, const Real hy, const Real hz, const Real v )
-{
- const RealType value = sArray[thrk][thrj][thri];
- //std::cout << value << std::endl;
- RealType a, b, c, tmp = std::numeric_limits< RealType >::max();
-
- c = TNL::argAbsMin( sArray[ thrk+1 ][ thrj ][ thri ],
- sArray[ thrk-1 ][ thrj ][ thri ] );
-
- b = TNL::argAbsMin( sArray[ thrk ][ thrj+1 ][ thri ],
- sArray[ thrk ][ thrj-1 ][ thri ] );
-
- a = TNL::argAbsMin( sArray[ thrk ][ thrj ][ thri+1 ],
- sArray[ thrk ][ thrj ][ thri-1 ] );
-
-
- if( fabs( a ) == std::numeric_limits< RealType >::max() &&
- fabs( b ) == std::numeric_limits< RealType >::max() &&
- fabs( c ) == std::numeric_limits< RealType >::max() )
- return false;
-
- RealType pom[6] = { a, b, c, (RealType)hx, (RealType)hy, (RealType)hz};
-
- sortMinims( pom );
-
- tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ];
- if( fabs( tmp ) < fabs( pom[ 1 ] ) )
- {
- sArray[ thrk ][ thrj ][ thri ] = argAbsMin( value, tmp );
- tmp = value - sArray[ thrk ][ thrj ][ thri ];
- if ( fabs( tmp ) > 0.01*hx )
- return true;
- else
- return false;
- }
- else
- {
- tmp = ( pom[ 3 ] * pom[ 3 ] * pom[ 1 ] + pom[ 4 ] * pom[ 4 ] * pom[ 0 ] +
- TNL::sign( value ) * pom[ 3 ] * pom[ 4 ] * TNL::sqrt( ( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] )/( v * v ) -
- ( pom[ 1 ] - pom[ 0 ] ) * ( pom[ 1 ] - pom[ 0 ] ) ) )/( pom[ 3 ] * pom[ 3 ] + pom[ 4 ] * pom[ 4 ] );
- if( fabs( tmp ) < fabs( pom[ 2 ]) )
- {
- sArray[ thrk ][ thrj ][ thri ] = argAbsMin( value, tmp );
- tmp = value - sArray[ thrk ][ thrj ][ thri ];
- if ( fabs( tmp ) > 0.01*hx )
- return true;
- else
- return false;
- }
- else
- {
- tmp = ( hy * hy * hz * hz * a + hx * hx * hz * hz * b + hx * hx * hy * hy * c +
- TNL::sign( value ) * hx * hy * hz * TNL::sqrt( ( hx * hx * hz * hz + hy * hy * hz * hz + hx * hx * hy * hy)/( v * v ) -
- hz * hz * ( a - b ) * ( a - b ) - hy * hy * ( a - c ) * ( a - c ) -
- hx * hx * ( b - c ) * ( b - c ) ) )/( hx * hx * hy * hy + hy * hy * hz * hz + hz * hz * hx *hx );
- sArray[ thrk ][ thrj ][ thri ] = argAbsMin( value, tmp );
- tmp = value - sArray[ thrk ][ thrj ][ thri ];
- if ( fabs( tmp ) > 0.01*hx )
- return true;
- else
- return false;
- }
- }
-
+ const RealType value = sArray[ thri ];
+ RealType a, tmp = std::numeric_limits< RealType >::max();
+
+ a = TNL::argAbsMin( sArray[ thri+1 ],
+ sArray[ thri-1 ] );
+
+ if( fabs( a ) == std::numeric_limits< RealType >::max() )
+ return false;
+
+ tmp = a + TNL::sign( value ) * h/v;
+
+
+ sArray[ thri ] = argAbsMin( value, tmp );
+
+ tmp = value - sArray[ thri ];
+ if ( fabs( tmp ) > 0.001*h )
+ return true;
+ else
return false;
}
#endif
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
index fa807742735f8beaa034b9a2555d3bc4a57f9f8e..57b1886e8b67c943051623c0a07f301faa76cad8 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
@@ -1,9 +1,9 @@
/***************************************************************************
- FastSweepingMethod.h - description
- -------------------
- begin : Jul 14, 2016
- copyright : (C) 2017 by Tomas Oberhuber
- email : tomas.oberhuber@fjfi.cvut.cz
+ FastSweepingMethod.h - description
+ -------------------
+ begin : Jul 14, 2016
+ copyright : (C) 2017 by Tomas Oberhuber
+ email : tomas.oberhuber@fjfi.cvut.cz
***************************************************************************/
/* See Copyright Notice in tnl/Copyright */
@@ -17,131 +17,134 @@
template< typename Mesh,
- typename Anisotropy = Functions::Analytic::Constant< Mesh::getMeshDimension(), typename Mesh::RealType > >
+ typename Anisotropy = Functions::Analytic::Constant< Mesh::getMeshDimension(), typename Mesh::RealType > >
class FastSweepingMethod
{
};
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
class FastSweepingMethod< Meshes::Grid< 1, Real, Device, Index >, Anisotropy >
- : public tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >
+: public tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >
{
- //static_assert( std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
-
- public:
-
- typedef Meshes::Grid< 1, Real, Device, Index > MeshType;
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef Anisotropy AnisotropyType;
- typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > > BaseType;
- using MeshPointer = Pointers::SharedPointer< MeshType >;
- using AnisotropyPointer = Pointers::SharedPointer< AnisotropyType, DeviceType >;
-
-
- using typename BaseType::InterfaceMapType;
- using typename BaseType::MeshFunctionType;
- using typename BaseType::InterfaceMapPointer;
- using typename BaseType::MeshFunctionPointer;
-
-
- FastSweepingMethod();
-
- const IndexType& getMaxIterations() const;
-
- void setMaxIterations( const IndexType& maxIterations );
-
- void solve( const MeshPointer& mesh,
- const AnisotropyPointer& anisotropy,
- MeshFunctionPointer& u );
-
-
- protected:
+ //static_assert( std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
+
+ public:
+
+ typedef Meshes::Grid< 1, Real, Device, Index > MeshType;
+ typedef Real RealType;
+ typedef Device DeviceType;
+ typedef Index IndexType;
+ typedef Anisotropy AnisotropyType;
+ typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > > BaseType;
+ using MeshPointer = Pointers::SharedPointer< MeshType >;
+ using AnisotropyPointer = Pointers::SharedPointer< AnisotropyType, DeviceType >;
+
+
+ using typename BaseType::InterfaceMapType;
+ using typename BaseType::MeshFunctionType;
+ using typename BaseType::InterfaceMapPointer;
+ using typename BaseType::MeshFunctionPointer;
+
+
+ FastSweepingMethod();
+
+ const IndexType& getMaxIterations() const;
+
+ void setMaxIterations( const IndexType& maxIterations );
+
+ void solve( const MeshPointer& mesh,
+ const AnisotropyPointer& anisotropy,
+ MeshFunctionPointer& u );
+
+
+ protected:
const IndexType maxIterations;
};
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
class FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >
- : public tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
+: public tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
{
- //static_assert( std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
-
- public:
-
- typedef Meshes::Grid< 2, Real, Device, Index > MeshType;
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef Anisotropy AnisotropyType;
- typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > BaseType;
- using MeshPointer = Pointers::SharedPointer< MeshType >;
- using AnisotropyPointer = Pointers::SharedPointer< AnisotropyType, DeviceType >;
-
- using typename BaseType::InterfaceMapType;
- using typename BaseType::MeshFunctionType;
- using typename BaseType::InterfaceMapPointer;
- using typename BaseType::MeshFunctionPointer;
-
- FastSweepingMethod();
-
- const IndexType& getMaxIterations() const;
-
- void setMaxIterations( const IndexType& maxIterations );
-
- void solve( const MeshPointer& mesh,
- const AnisotropyPointer& anisotropy,
- MeshFunctionPointer& u );
-
- protected:
+ //static_assert( std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
+
+ public:
+
+ typedef Meshes::Grid< 2, Real, Device, Index > MeshType;
+ typedef Real RealType;
+ typedef Device DeviceType;
+ typedef Index IndexType;
+ typedef Anisotropy AnisotropyType;
+ typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > BaseType;
+ using MeshPointer = Pointers::SharedPointer< MeshType >;
+ using AnisotropyPointer = Pointers::SharedPointer< AnisotropyType, DeviceType >;
+
+ using typename BaseType::InterfaceMapType;
+ using typename BaseType::MeshFunctionType;
+ using typename BaseType::InterfaceMapPointer;
+ using typename BaseType::MeshFunctionPointer;
+ using typename BaseType::ArrayContainer;
+
+ FastSweepingMethod();
+
+ const IndexType& getMaxIterations() const;
+
+ void setMaxIterations( const IndexType& maxIterations );
+
+ void solve( const MeshPointer& mesh,
+ const AnisotropyPointer& anisotropy,
+ MeshFunctionPointer& u );
+
+ protected:
const IndexType maxIterations;
};
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
class FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >
- : public tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
+: public tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
{
- //static_assert( std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
-
- public:
-
- typedef Meshes::Grid< 3, Real, Device, Index > MeshType;
- typedef Real RealType;
- typedef Device DeviceType;
- typedef Index IndexType;
- typedef Anisotropy AnisotropyType;
- typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > BaseType;
- using MeshPointer = Pointers::SharedPointer< MeshType >;
- using AnisotropyPointer = Pointers::SharedPointer< AnisotropyType, DeviceType >;
-
- using typename BaseType::InterfaceMapType;
- using typename BaseType::MeshFunctionType;
- using typename BaseType::InterfaceMapPointer;
- using typename BaseType::MeshFunctionPointer;
-
- FastSweepingMethod();
-
- const IndexType& getMaxIterations() const;
-
- void setMaxIterations( const IndexType& maxIterations );
-
- void solve( const MeshPointer& mesh,
- const AnisotropyPointer& anisotropy,
- MeshFunctionPointer& u );
-
-
- protected:
+ //static_assert( std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
+
+ public:
+
+ typedef Meshes::Grid< 3, Real, Device, Index > MeshType;
+ typedef Real RealType;
+ typedef Device DeviceType;
+ typedef Index IndexType;
+ typedef Anisotropy AnisotropyType;
+ typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > BaseType;
+ using MeshPointer = Pointers::SharedPointer< MeshType >;
+ using AnisotropyPointer = Pointers::SharedPointer< AnisotropyType, DeviceType >;
+
+ using typename BaseType::InterfaceMapType;
+ using typename BaseType::MeshFunctionType;
+ using typename BaseType::InterfaceMapPointer;
+ using typename BaseType::MeshFunctionPointer;
+ using typename BaseType::ArrayContainer;
+
+
+ FastSweepingMethod();
+
+ const IndexType& getMaxIterations() const;
+
+ void setMaxIterations( const IndexType& maxIterations );
+
+ void solve( const MeshPointer& mesh,
+ const AnisotropyPointer& anisotropy,
+ MeshFunctionPointer& u );
+
+
+ protected:
const IndexType maxIterations;
};
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod2D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod2D_impl.h
index 6703843c179ea0215c40cf2d21b9cd66fde99de6..d5ce1efe164bc835866dc7fc95909b3defdf846f 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod2D_impl.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod2D_impl.h
@@ -15,511 +15,733 @@
#include "tnlFastSweepingMethod.h"
#include <TNL/Devices/Cuda.h>
+#include <TNL/Communicators/MpiDefs.h>
+
+
+#include <string.h>
#include <iostream>
#include <fstream>
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >::
FastSweepingMethod()
-: maxIterations( 100 )
+: maxIterations( 1 )
{
-
+
}
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
const Index&
FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >::
getMaxIterations() const
{
-
+
}
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
void
FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >::
setMaxIterations( const IndexType& maxIterations )
{
-
+
}
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
void
FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >::
solve( const MeshPointer& mesh,
- const AnisotropyPointer& anisotropy,
- MeshFunctionPointer& u )
-{
- /*MeshFunctionType v;
- v.setMesh(mesh);
- double A[320][320];
- for (int i = 0; i < 320; i++)
- for (int j = 0; j < 320; j++)
- A[i][j] = 0;
-
- std::ifstream file("/home/maty/Downloads/mapa2.txt");
-
- for (int i = 0; i < 320; i++)
- for (int j = 0; j < 320; j++)
- file >> A[i][j];
- file.close();
- for (int i = 0; i < 320; i++)
- for (int j = 0; j < 320; j++)
- v[i*320 + j] = A[i][j];
- v.save("mapa.tnl");*/
-
-
- MeshFunctionPointer auxPtr;
- InterfaceMapPointer interfaceMapPtr;
- auxPtr->setMesh( mesh );
- interfaceMapPtr->setMesh( mesh );
- std::cout << "Initiating the interface cells ..." << std::endl;
- BaseType::initInterface( u, auxPtr, interfaceMapPtr );
+ const AnisotropyPointer& anisotropy,
+ MeshFunctionPointer& u )
+{
+ MeshFunctionPointer auxPtr;
+ InterfaceMapPointer interfaceMapPtr;
+ auxPtr->setMesh( mesh );
+ interfaceMapPtr->setMesh( mesh );
+ std::cout << "Initiating the interface cells ..." << std::endl;
+ BaseType::initInterface( u, auxPtr, interfaceMapPtr );
+
+ auxPtr->save( "aux-ini.tnl" );
+
+ typename MeshType::Cell cell( *mesh );
+
+ IndexType iteration( 0 );
+ InterfaceMapType interfaceMap = *interfaceMapPtr;
+ MeshFunctionType aux = *auxPtr;
+
+
+//#ifdef HAVE_MPI
+ bool a = Communicators::MpiCommunicator::IsInitialized();
+ if( a )
+ printf("Je Init\n");
+ else
+ printf("Neni Init\n");
+//#endif
+
+ while( iteration < this->maxIterations )
+ {
+ if( std::is_same< DeviceType, Devices::Host >::value )
+ {
+ int numThreadsPerBlock = -1;
+
+ numThreadsPerBlock = ( mesh->getDimensions().x()/2 + (mesh->getDimensions().x() % 2 != 0 ? 1:0));
+ //printf("numThreadsPerBlock = %d\n", numThreadsPerBlock);
+ if( numThreadsPerBlock <= 16 )
+ numThreadsPerBlock = 16;
+ else if(numThreadsPerBlock <= 32 )
+ numThreadsPerBlock = 32;
+ else if(numThreadsPerBlock <= 64 )
+ numThreadsPerBlock = 64;
+ else if(numThreadsPerBlock <= 128 )
+ numThreadsPerBlock = 128;
+ else if(numThreadsPerBlock <= 256 )
+ numThreadsPerBlock = 256;
+ else if(numThreadsPerBlock <= 512 )
+ numThreadsPerBlock = 512;
+ else
+ numThreadsPerBlock = 1024;
+ //printf("numThreadsPerBlock = %d\n", numThreadsPerBlock);
+
+ if( numThreadsPerBlock == -1 ){
+ printf("Fail in setting numThreadsPerBlock.\n");
+ break;
+ }
+
+
+
+ int numBlocksX = mesh->getDimensions().x() / numThreadsPerBlock + (mesh->getDimensions().x() % numThreadsPerBlock != 0 ? 1:0);
+ int numBlocksY = mesh->getDimensions().y() / numThreadsPerBlock + (mesh->getDimensions().y() % numThreadsPerBlock != 0 ? 1:0);
+
+ //std::cout << "numBlocksX = " << numBlocksX << std::endl;
+
+ /*Real **sArray = new Real*[numBlocksX*numBlocksY];
+ for( int i = 0; i < numBlocksX * numBlocksY; i++ )
+ sArray[ i ] = new Real [ (numThreadsPerBlock + 2)*(numThreadsPerBlock + 2)];*/
+
+ ArrayContainer BlockIterHost;
+ BlockIterHost.setSize( numBlocksX * numBlocksY );
+ BlockIterHost.setValue( 1 );
+ int IsCalculationDone = 1;
+
+ MeshFunctionPointer helpFunc( mesh );
+ MeshFunctionPointer helpFunc1( mesh );
+ helpFunc1 = auxPtr;
+ auxPtr = helpFunc;
+ helpFunc = helpFunc1;
+ //std::cout<< "Size = " << BlockIterHost.getSize() << std::endl;
+ /*for( int k = numBlocksX-1; k >-1; k-- ){
+ for( int l = 0; l < numBlocksY; l++ ){
+ std::cout<< BlockIterHost[ l*numBlocksX + k ];
+ }
+ std::cout<<std::endl;
+ }
+ std::cout<<std::endl;*/
+ unsigned int numWhile = 0;
+ while( IsCalculationDone )
+ {
+ IsCalculationDone = 0;
+ helpFunc1 = auxPtr;
+ auxPtr = helpFunc;
+ helpFunc = helpFunc1;
+ switch ( numThreadsPerBlock ){
+ case 16:
+ this->template updateBlocks< 18 >( interfaceMap, *auxPtr, *helpFunc, BlockIterHost, numThreadsPerBlock/*, sArray*/ );
+ case 32:
+ this->template updateBlocks< 34 >( interfaceMap, *auxPtr, *helpFunc, BlockIterHost, numThreadsPerBlock/*, sArray*/ );
+ case 64:
+ this->template updateBlocks< 66 >( interfaceMap, *auxPtr, *helpFunc, BlockIterHost, numThreadsPerBlock/*, sArray*/ );
+ case 128:
+ this->template updateBlocks< 130 >( interfaceMap, *auxPtr, *helpFunc, BlockIterHost, numThreadsPerBlock/*, sArray*/ );
+ case 256:
+ this->template updateBlocks< 258 >( interfaceMap, *auxPtr, *helpFunc, BlockIterHost, numThreadsPerBlock/*, sArray*/ );
+ case 512:
+ this->template updateBlocks< 514 >( interfaceMap, *auxPtr, *helpFunc, BlockIterHost, numThreadsPerBlock/*, sArray*/ );
+ default:
+ this->template updateBlocks< 1028 >( interfaceMap, *auxPtr, *helpFunc, BlockIterHost, numThreadsPerBlock/*, sArray*/ );
+ }
- auxPtr->save( "aux-ini.tnl" );
-
- typename MeshType::Cell cell( *mesh );
-
- IndexType iteration( 0 );
- InterfaceMapType interfaceMap = *interfaceMapPtr;
- MeshFunctionType aux = *auxPtr;
- while( iteration < this->maxIterations )
- {
- if( std::is_same< DeviceType, Devices::Host >::value )
- {
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh->getDimensions().y();
- cell.getCoordinates().y()++ )
- {
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh->getDimensions().x();
- cell.getCoordinates().x()++ )
- {
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
-
- //aux.save( "aux-1.tnl" );
-
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh->getDimensions().y();
- cell.getCoordinates().y()++ )
- {
- for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
- cell.getCoordinates().x() >= 0 ;
- cell.getCoordinates().x()-- )
- {
- //std::cerr << "2 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
-
- //aux.save( "aux-2.tnl" );
-
- for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
- cell.getCoordinates().y() >= 0 ;
- cell.getCoordinates().y()-- )
- {
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh->getDimensions().x();
- cell.getCoordinates().x()++ )
- {
- //std::cerr << "3 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
-
- //aux.save( "aux-3.tnl" );
-
- for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
- cell.getCoordinates().y() >= 0;
- cell.getCoordinates().y()-- )
- {
- for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
- cell.getCoordinates().x() >= 0 ;
- cell.getCoordinates().x()-- )
- {
- //std::cerr << "4 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
-
- //aux.save( "aux-4.tnl" );
-
- /*for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh->getDimensions().y();
- cell.getCoordinates().x()++ )
- {
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh->getDimensions().x();
- cell.getCoordinates().y()++ )
- {
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
-
-
- aux.save( "aux-5.tnl" );
-
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh->getDimensions().y();
- cell.getCoordinates().x()++ )
- {
- for( cell.getCoordinates().y() = mesh->getDimensions().x() - 1;
- cell.getCoordinates().y() >= 0 ;
- cell.getCoordinates().y()-- )
- {
- //std::cerr << "2 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
+
+ //Reduction
+ for( int i = 0; i < BlockIterHost.getSize(); i++ ){
+ if( IsCalculationDone == 0 ){
+ IsCalculationDone = IsCalculationDone || BlockIterHost[ i ];
+ //break;
+ }
+ }
+ numWhile++;
+ /*std::cout <<"numWhile = "<< numWhile <<std::endl;
+
+ for( int j = numBlocksY-1; j>-1; j-- ){
+ for( int i = 0; i < numBlocksX; i++ )
+ std::cout << BlockIterHost[ j * numBlocksX + i ];
+ std::cout << std::endl;
+ }
+ std::cout << std::endl;*/
+
+ this->getNeighbours( BlockIterHost, numBlocksX, numBlocksY );
+
+ /*for( int j = numBlocksY-1; j>-1; j-- ){
+ for( int i = 0; i < numBlocksX; i++ )
+ std::cout << "BlockIterHost = "<< j*numBlocksX + i<< " ," << BlockIterHost[ j * numBlocksX + i ];
+ std::cout << std::endl;
}
- aux.save( "aux-6.tnl" );
-
- for( cell.getCoordinates().x() = mesh->getDimensions().y() - 1;
- cell.getCoordinates().x() >= 0 ;
- cell.getCoordinates().x()-- )
- {
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh->getDimensions().x();
- cell.getCoordinates().y()++ )
- {
- //std::cerr << "3 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- aux.save( "aux-7.tnl" );
-
- for( cell.getCoordinates().x() = mesh->getDimensions().y() - 1;
- cell.getCoordinates().x() >= 0;
- cell.getCoordinates().x()-- )
- {
- for( cell.getCoordinates().y() = mesh->getDimensions().x() - 1;
- cell.getCoordinates().y() >= 0 ;
- cell.getCoordinates().y()-- )
- {
- //std::cerr << "4 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }*/
+ std::cout << std::endl;*/
+
+ //std::cout<<std::endl;
+ //string s( "aux-"+ std::to_string(numWhile) + ".tnl");
+ //aux.save( s );
}
- if( std::is_same< DeviceType, Devices::Cuda >::value )
- {
- // TODO: CUDA code
+ if( numWhile == 1 ){
+ auxPtr = helpFunc;
+ }
+ /*for( int i = 0; i < numBlocksX * numBlocksY; i++ )
+ delete []sArray[i];*/
+
+
+ /*for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh->getDimensions().y();
+ cell.getCoordinates().y()++ )
+ {
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh->getDimensions().x();
+ cell.getCoordinates().x()++ )
+ {
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+
+ //aux.save( "aux-1.tnl" );
+
+ for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh->getDimensions().y();
+ cell.getCoordinates().y()++ )
+ {
+ for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
+ cell.getCoordinates().x() >= 0 ;
+ cell.getCoordinates().x()-- )
+ {
+ //std::cerr << "2 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+
+ //aux.save( "aux-2.tnl" );
+
+ for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
+ cell.getCoordinates().y() >= 0 ;
+ cell.getCoordinates().y()-- )
+ {
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh->getDimensions().x();
+ cell.getCoordinates().x()++ )
+ {
+ //std::cerr << "3 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+
+ //aux.save( "aux-3.tnl" );
+
+ for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
+ cell.getCoordinates().y() >= 0;
+ cell.getCoordinates().y()-- )
+ {
+ for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
+ cell.getCoordinates().x() >= 0 ;
+ cell.getCoordinates().x()-- )
+ {
+ //std::cerr << "4 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+
+ for( int j = 0;
+ j < mesh->getDimensions().y();
+ j++ )
+ {
+ for( int i = 0;
+ i < mesh->getDimensions().x();
+ i++ )
+ {
+ std::cout << aux[ i * mesh->getDimensions().y() + j ] << " ";
+ }
+ std::cout << std::endl;
+ }*/
+
+ }
+ if( std::is_same< DeviceType, Devices::Cuda >::value )
+ {
+ // TODO: CUDA code
#ifdef HAVE_CUDA
-
- Real *dAux;
- cudaMalloc(&dAux, ( mesh->getDimensions().x() * mesh->getDimensions().y() ) * sizeof( Real ) );
-
-
-
-
- const int cudaBlockSize( 16 );
- int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().x(), cudaBlockSize );
- int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().y(), cudaBlockSize );
- dim3 blockSize( cudaBlockSize, cudaBlockSize );
- dim3 gridSize( numBlocksX, numBlocksY );
-
- tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
-
- aux1<<< gridSize, blockSize >>>( auxPtr.template modifyData< Device>(), dAux,1 );
-
- //int BlockIter = 1;// = (bool*)malloc( ( numBlocksX * numBlocksY ) * sizeof( bool ) );
-
- int *BlockIterDevice;
- int BlockIterD = 1;
-
- cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( int ) );
- int nBlocks = ( numBlocksX * numBlocksY )/512 + ((( numBlocksX * numBlocksY )%512 != 0) ? 1:0);
- int *dBlock;
- cudaMalloc(&dBlock, nBlocks * sizeof( int ) );
-
- while( BlockIterD )
- {
- /*for( int i = 0; i < numBlocksX * numBlocksY; i++ )
- BlockIter[ i ] = false;*/
-
- CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
- interfaceMapPtr.template getData< Device >(),
- dAux,
- BlockIterDevice );
-
- CudaParallelReduc<<< nBlocks , 512 >>>( BlockIterDevice, dBlock, ( numBlocksX * numBlocksY ) );
- CudaParallelReduc<<< 1, nBlocks >>>( dBlock, dBlock, nBlocks );
-
- cudaMemcpy(&BlockIterD, &dBlock[0], sizeof( int ), cudaMemcpyDeviceToHost);
-
- /*for( int i = 1; i < numBlocksX * numBlocksY; i++ )
- BlockIter[ 0 ] = BlockIter[ 0 ] || BlockIter[ i ];*/
-
- }
- aux1<<<gridSize,blockSize>>>( auxPtr.template modifyData< Device>(), dAux, 0 );
- cudaFree( dAux );
- cudaFree( BlockIterDevice );
- cudaFree( dBlock );
- cudaDeviceSynchronize();
-
- TNL_CHECK_CUDA_DEVICE;
-
- //aux = *auxPtr;
- //interfaceMap = *interfaceMapPtr;
-#endif
+ TNL_CHECK_CUDA_DEVICE;
+ // Maximum cudaBlockSite is 32. Because of maximum num. of threads in kernel.
+ const int cudaBlockSize( 16 );
+
+ int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().x(), cudaBlockSize );
+ int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().y(), cudaBlockSize );
+ dim3 blockSize( cudaBlockSize, cudaBlockSize );
+ dim3 gridSize( numBlocksX, numBlocksY );
+
+ tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
+
+ int BlockIterD = 1;
+
+ TNL::Containers::Array< int, Devices::Cuda, IndexType > BlockIterDevice;
+ BlockIterDevice.setSize( numBlocksX * numBlocksY );
+ BlockIterDevice.setValue( 1 );
+ TNL_CHECK_CUDA_DEVICE;
+
+
+ TNL::Containers::Array< int, Devices::Cuda, IndexType > BlockIterPom;
+ BlockIterPom.setSize( numBlocksX * numBlocksY );
+ BlockIterPom.setValue( 0 );
+ /*TNL::Containers::Array< int, Devices::Host, IndexType > BlockIterPom1;
+ BlockIterPom1.setSize( numBlocksX * numBlocksY );
+ BlockIterPom1.setValue( 0 );*/
+ /*int *BlockIterDevice;
+ cudaMalloc((void**) &BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( int ) );*/
+ int nBlocksNeigh = ( numBlocksX * numBlocksY )/1024 + ((( numBlocksX * numBlocksY )%1024 != 0) ? 1:0);
+ //std::cout << "nBlocksNeigh = " << nBlocksNeigh << std::endl;
+ //free( BlockIter );
+ /*int *BlockIterPom;
+ cudaMalloc((void**) &BlockIterPom, ( numBlocksX * numBlocksY ) * sizeof( int ) );*/
+
+ int nBlocks = ( numBlocksX * numBlocksY )/1024 + ((( numBlocksX * numBlocksY )%1024 != 0) ? 1:0);
+
+ TNL::Containers::Array< int, Devices::Cuda, IndexType > dBlock;
+ dBlock.setSize( nBlocks );
+ TNL_CHECK_CUDA_DEVICE;
+ /*int *dBlock;
+ cudaMalloc((void**) &dBlock, nBlocks * sizeof( int ) );*/
+
+
+ MeshFunctionPointer helpFunc1( mesh );
+ MeshFunctionPointer helpFunc( mesh );
+
+ helpFunc1 = auxPtr;
+ auxPtr = helpFunc;
+ helpFunc = helpFunc1;
+
+ int numIter = 0;
+
+ //int oddEvenBlock = 0;
+ while( BlockIterD )
+ {
+ /** HERE IS CHESS METHOD **/
+
+ /*auxPtr = helpFunc;
+
+ CudaUpdateCellCaller<18><<< gridSize, blockSize >>>( ptr,
+ interfaceMapPtr.template getData< Device >(),
+ auxPtr.template getData< Device>(),
+ helpFunc.template modifyData< Device>(),
+ BlockIterDevice,
+ oddEvenBlock );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+ auxPtr = helpFunc;
+
+ oddEvenBlock= (oddEvenBlock == 0) ? 1: 0;
+
+ CudaUpdateCellCaller<18><<< gridSize, blockSize >>>( ptr,
+ interfaceMapPtr.template getData< Device >(),
+ auxPtr.template getData< Device>(),
+ helpFunc.template modifyData< Device>(),
+ BlockIterDevice,
+ oddEvenBlock );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+ auxPtr = helpFunc;
+
+ oddEvenBlock= (oddEvenBlock == 0) ? 1: 0;
+
+ CudaParallelReduc<<< nBlocks , 1024 >>>( BlockIterDevice, dBlock, ( numBlocksX * numBlocksY ) );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+ CudaParallelReduc<<< 1, nBlocks >>>( dBlock, dBlock, nBlocks );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+
+ BlockIterD = dBlock.getElement( 0 );*/
+
+ /**------------------------------------------------------------------------------------------------*/
+
+
+ /** HERE IS FIM **/
+
+ helpFunc1 = auxPtr;
+ auxPtr = helpFunc;
+ helpFunc = helpFunc1;
+ TNL_CHECK_CUDA_DEVICE;
+
+ //int pocBloku = 0;
+ Devices::Cuda::synchronizeDevice();
+ CudaUpdateCellCaller<18><<< gridSize, blockSize >>>( ptr,
+ interfaceMapPtr.template getData< Device >(),
+ auxPtr.template modifyData< Device>(),
+ helpFunc.template modifyData< Device>(),
+ BlockIterDevice );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+
+ //std::cout << "Pocet aktivnich bloku = " << pocBloku << std::endl;
+ //BlockIterPom1 = BlockIterDevice;
+ ///for( int i =0; i< numBlocksX; i++ ){
+ // for( int j = 0; j < numBlocksY; j++ )
+ // {
+ // std::cout << BlockIterPom1[j*numBlocksX + i];
+ // }
+ // std::cout << std::endl;
+ //}
+ //std::cout << std::endl;
+
+ GetNeighbours<<< nBlocksNeigh, 1024 >>>( BlockIterDevice, BlockIterPom, numBlocksX, numBlocksY );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+ BlockIterDevice = BlockIterPom;
+
+ //std::cout<< "Probehlo" << std::endl;
+
+ //TNL::swap( auxPtr, helpFunc );
+
+
+ CudaParallelReduc<<< nBlocks , 1024 >>>( BlockIterDevice, dBlock, ( numBlocksX * numBlocksY ) );
+ TNL_CHECK_CUDA_DEVICE;
+
+ CudaParallelReduc<<< 1, nBlocks >>>( dBlock, dBlock, nBlocks );
+ TNL_CHECK_CUDA_DEVICE;
+
+
+ BlockIterD = dBlock.getElement( 0 );
+ //cudaMemcpy( &BlockIterD, &dBlock[0], sizeof( int ), cudaMemcpyDeviceToHost);
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+
+
+ /**-----------------------------------------------------------------------------------------------------------*/
+ /*for( int i = 1; i < numBlocksX * numBlocksY; i++ )
+ BlockIter[ 0 ] = BlockIter[ 0 ] || BlockIter[ i ];*/
+ numIter ++;
}
- iteration++;
- }
- aux.save("aux-final.tnl");
+ if( numIter == 1 ){
+ auxPtr = helpFunc;
+ }
+ /*cudaFree( BlockIterDevice );
+ cudaFree( dBlock );
+ delete BlockIter;*/
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+
+ aux = *auxPtr;
+ interfaceMap = *interfaceMapPtr;
+#endif
+ }
+ iteration++;
+ }
+ //#endif
+ aux.save("aux-final.tnl");
}
+
#ifdef HAVE_CUDA
-template < typename Real, typename Device, typename Index >
-__global__ void aux1( Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux, Real *dAux, int a )
+
+
+template < typename Index >
+__global__ void GetNeighbours( TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice,
+ TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterPom, int numBlockX, int numBlockY )
{
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- int j = blockDim.y*blockIdx.y + threadIdx.y;
- const Meshes::Grid< 2, Real, Device, Index >& mesh = aux.template getMesh< Devices::Cuda >();
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && a == 1 )
- {
- dAux[ j*mesh.getDimensions().x() + i ] = aux[ j*mesh.getDimensions().x() + i ];
- }
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && a == 0 )
- {
- aux[ j*mesh.getDimensions().x() + i ] = dAux[ j*mesh.getDimensions().x() + i ];
+ int i = blockIdx.x * 1024 + threadIdx.x;
+
+ if( i < numBlockX * numBlockY )
+ {
+ int pom = 0;//BlockIterPom[ i ] = 0;
+ int m=0, k=0;
+ m = i%numBlockX;
+ k = i/numBlockX;
+ if( m > 0 && BlockIterDevice[ i - 1 ] ){
+ pom = 1;//BlockIterPom[ i ] = 1;
+ }else if( m < numBlockX -1 && BlockIterDevice[ i + 1 ] ){
+ pom = 1;//BlockIterPom[ i ] = 1;
+ }else if( k > 0 && BlockIterDevice[ i - numBlockX ] ){
+ pom = 1;// BlockIterPom[ i ] = 1;
+ }else if( k < numBlockY -1 && BlockIterDevice[ i + numBlockX ] ){
+ pom = 1;//BlockIterPom[ i ] = 1;
}
+ BlockIterPom[ i ] = pom;//BlockIterPom[ i ];
+ }
}
-__global__ void CudaParallelReduc( int *BlockIterDevice, int *dBlock, int nBlocks )
+template < typename Index >
+__global__ void CudaParallelReduc( TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice,
+ TNL::Containers::Array< int, Devices::Cuda, Index > dBlock, int nBlocks )
{
- int i = threadIdx.x;
- int blId = blockIdx.x;
- __shared__ volatile int sArray[ 512 ];
- sArray[ i ] = false;
- if(blId * 1024 + i < nBlocks )
- sArray[ i ] = BlockIterDevice[ blId * 1024 + i ];
-
- if (blockDim.x * blockDim.y == 1024) {
- if (i < 512)
- sArray[ i ] += sArray[ i ];
- }
- __syncthreads();
- if (blockDim.x * blockDim.y >= 512) {
- if (i < 256) {
- sArray[ i ] += sArray[ i ];
- }
- }
- if (blockDim.x * blockDim.y >= 256) {
- if (i < 128) {
- sArray[ i ] += sArray[ i + 128 ];
- }
+ int i = threadIdx.x;
+ int blId = blockIdx.x;
+ int blockSize = blockDim.x;
+ /*if ( i == 0 && blId == 0 ){
+ printf( "nBlocks = %d \n", nBlocks );
+ for( int j = nBlocks-1; j > -1 ; j--){
+ printf( "cislo = %d \n", BlockIterDevice[ j ] );
+ }
+ }*/
+ __shared__ int sArray[ 1024 ];
+ sArray[ i ] = 0;
+ if( blId * 1024 + i < nBlocks )
+ sArray[ i ] = BlockIterDevice[ blId * 1024 + i ];
+ __syncthreads();
+ /*extern __shared__ volatile int sArray[];
+ unsigned int i = threadIdx.x;
+ unsigned int gid = blockIdx.x * blockSize * 2 + threadIdx.x;
+ unsigned int gridSize = blockSize * 2 * gridDim.x;
+ sArray[ i ] = 0;
+ while( gid < nBlocks )
+ {
+ sArray[ i ] += BlockIterDevice[ gid ] + BlockIterDevice[ gid + blockSize ];
+ gid += gridSize;
+ }
+ __syncthreads();*/
+
+ if ( blockSize == 1024) {
+ if (i < 512)
+ sArray[ i ] += sArray[ i + 512 ];
+ }
+ __syncthreads();
+ if (blockSize >= 512) {
+ if (i < 256) {
+ sArray[ i ] += sArray[ i + 256 ];
}
- __syncthreads();
- if (blockDim.x * blockDim.y >= 128) {
- if (i < 64) {
- sArray[ i ] += sArray[ i + 64 ];
- }
+ }
+ __syncthreads();
+ if (blockSize >= 256) {
+ if (i < 128) {
+ sArray[ i ] += sArray[ i + 128 ];
}
- __syncthreads();
- if (i < 32 )
- {
- if( blockDim.x * blockDim.y >= 64 ) sArray[ i ] += sArray[ i + 32 ];
- if( blockDim.x * blockDim.y >= 32 ) sArray[ i ] += sArray[ i + 16 ];
- if( blockDim.x * blockDim.y >= 16 ) sArray[ i ] += sArray[ i + 8 ];
- if( blockDim.x * blockDim.y >= 8 ) sArray[ i ] += sArray[ i + 4 ];
- if( blockDim.x * blockDim.y >= 4 ) sArray[ i ] += sArray[ i + 2 ];
- if( blockDim.x * blockDim.y >= 2 ) sArray[ i ] += sArray[ i + 1 ];
+ }
+ __syncthreads();
+ if (blockSize >= 128) {
+ if (i < 64) {
+ sArray[ i ] += sArray[ i + 64 ];
}
-
- if( i == 0 )
- dBlock[ blId ] = sArray[ 0 ];
+ }
+ __syncthreads();
+ if (i < 32 )
+ {
+ if( blockSize >= 64 ) sArray[ i ] += sArray[ i + 32 ];
+ if( blockSize >= 32 ) sArray[ i ] += sArray[ i + 16 ];
+ if( blockSize >= 16 ) sArray[ i ] += sArray[ i + 8 ];
+ if( blockSize >= 8 ) sArray[ i ] += sArray[ i + 4 ];
+ if( blockSize >= 4 ) sArray[ i ] += sArray[ i + 2 ];
+ if( blockSize >= 2 ) sArray[ i ] += sArray[ i + 1 ];
+ }
+
+ if( i == 0 )
+ dBlock[ blId ] = sArray[ 0 ];
}
-template < typename Real, typename Device, typename Index >
+template < int sizeSArray, typename Real, typename Device, typename Index >
__global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr,
- const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
- Real *aux,
- int *BlockIterDevice )
+ const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
+ const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
+ Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& helpFunc,
+ TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice, int oddEvenBlock )
{
- int thri = threadIdx.x; int thrj = threadIdx.y;
- int blIdx = blockIdx.x; int blIdy = blockIdx.y;
- int i = thri + blockDim.x*blIdx;
- int j = blockDim.y*blIdy + thrj;
- int currentIndex = thrj * blockDim.x + thri;
-
- //__shared__ volatile bool changed[ blockDim.x*blockDim.y ];
- __shared__ volatile bool changed[16*16];
- changed[ currentIndex ] = false;
-
- if( thrj == 0 && thri == 0 )
- changed[ 0 ] = true;
-
+ int thri = threadIdx.x; int thrj = threadIdx.y;
+ int i = threadIdx.x + blockDim.x*blockIdx.x;
+ int j = blockDim.y*blockIdx.y + threadIdx.y;
+ /** FOR CHESS METHOD */
+ //if( (blockIdx.y%2 + blockIdx.x) % 2 == oddEvenBlock )
+ //{
+ /**------------------------------------------*/
+
+
+ /** FOR FIM METHOD */
+
+ if( BlockIterDevice[ blockIdx.y * gridDim.x + blockIdx.x ] )
+ {
+ __syncthreads();
+ /**-----------------------------------------*/
const Meshes::Grid< 2, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
+ __shared__ int dimX;
+ __shared__ int dimY;
__shared__ Real hx;
__shared__ Real hy;
- if( thrj == 1 && thri == 1 )
+ if( thri==0 && thrj == 0)
{
- hx = mesh.getSpaceSteps().x();
- hy = mesh.getSpaceSteps().y();
+ dimX = mesh.getDimensions().x();
+ dimY = mesh.getDimensions().y();
+ hx = mesh.getSpaceSteps().x();
+ hy = mesh.getSpaceSteps().y();
+ BlockIterDevice[ blockIdx.y * gridDim.x + blockIdx.x ] = 0;
}
+ __syncthreads();
+ int numOfBlockx;
+ int numOfBlocky;
+ int xkolik;
+ int ykolik;
- //__shared__ volatile Real sArray[ blockDim.y+2 ][ blockDim.x+2 ];
- __shared__ volatile Real sArray[18][18];
- sArray[thrj][thri] = std::numeric_limits< Real >::max();
+ xkolik = blockDim.x + 1;
+ ykolik = blockDim.y + 1;
+ numOfBlocky = dimY/blockDim.y + ((dimY%blockDim.y != 0) ? 1:0);
+ numOfBlockx = dimX/blockDim.x + ((dimX%blockDim.x != 0) ? 1:0);
- //filling sArray edges
- int dimX = mesh.getDimensions().x(); int dimY = mesh.getDimensions().y();
- __shared__ volatile int numOfBlockx;
- __shared__ volatile int numOfBlocky;
- __shared__ int xkolik;
- __shared__ int ykolik;
- if( thri == 0 && thrj == 0 )
- {
- xkolik = blockDim.x + 1;
- ykolik = blockDim.y + 1;
- numOfBlocky = dimY/blockDim.y + ((dimY%blockDim.y != 0) ? 1:0);
- numOfBlockx = dimX/blockDim.x + ((dimX%blockDim.x != 0) ? 1:0);
+ if( numOfBlockx - 1 == blockIdx.x )
+ xkolik = dimX - (blockIdx.x)*blockDim.x+1;
- if( numOfBlockx - 1 == blIdx )
- xkolik = dimX - (blIdx)*blockDim.x+1;
-
- if( numOfBlocky -1 == blIdy )
- ykolik = dimY - (blIdy)*blockDim.y+1;
- BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = 0;
- }
+ if( numOfBlocky -1 == blockIdx.y )
+ ykolik = dimY - (blockIdx.y)*blockDim.y+1;
__syncthreads();
+ int currentIndex = thrj * blockDim.x + thri;
+ //__shared__ volatile bool changed[ blockDim.x*blockDim.y ];
+ __shared__ volatile bool changed[ (sizeSArray-2)*(sizeSArray-2)];
+ changed[ currentIndex ] = false;
+ if( thrj == 0 && thri == 0 )
+ changed[ 0 ] = true;
+
+
+ //__shared__ volatile Real sArray[ blockDim.y+2 ][ blockDim.x+2 ];
+ __shared__ volatile Real sArray[ sizeSArray * sizeSArray ];
+ sArray[ thrj * sizeSArray + thri ] = std::numeric_limits< Real >::max();
+
+ //filling sArray edges
if( thri == 0 )
- {
- if( dimX > (blIdx+1) * blockDim.x && thrj+1 < ykolik )
- sArray[thrj+1][xkolik] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + (thrj+1)*dimX + xkolik ];
- else
- sArray[thrj+1][xkolik] = std::numeric_limits< Real >::max();
+ {
+ if( dimX > (blockIdx.x+1) * blockDim.x && thrj+1 < ykolik )
+ sArray[(thrj+1)*sizeSArray + xkolik] = aux[ blockIdx.y*blockDim.y*dimX - dimX + blockIdx.x*blockDim.x - 1 + (thrj+1)*dimX + xkolik ];
+ else
+ sArray[(thrj+1)*sizeSArray + xkolik] = std::numeric_limits< Real >::max();
}
if( thri == 1 )
{
- if( blIdx != 0 && thrj+1 < ykolik )
- sArray[thrj+1][0] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + (thrj+1)*dimX ];
- else
- sArray[thrj+1][0] = std::numeric_limits< Real >::max();
+ if( blockIdx.x != 0 && thrj+1 < ykolik )
+ sArray[(thrj+1)*sizeSArray + 0] = aux[ blockIdx.y*blockDim.y*dimX - dimX + blockIdx.x*blockDim.x - 1 + (thrj+1)*dimX ];
+ else
+ sArray[(thrj+1)*sizeSArray + 0] = std::numeric_limits< Real >::max();
}
if( thri == 2 )
{
- if( dimY > (blIdy+1) * blockDim.y && thri+1 < xkolik )
- sArray[ykolik][thrj+1] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + ykolik*dimX + thrj+1 ];
- else
- sArray[ykolik][thrj+1] = std::numeric_limits< Real >::max();
+ if( dimY > (blockIdx.y+1) * blockDim.y && thrj+1 < xkolik )
+ sArray[ ykolik*sizeSArray + thrj+1 ] = aux[ blockIdx.y*blockDim.y*dimX - dimX + blockIdx.x*blockDim.x - 1 + ykolik*dimX + thrj+1 ];
+ else
+ sArray[ykolik*sizeSArray + thrj+1] = std::numeric_limits< Real >::max();
+
}
if( thri == 3 )
{
- if( blIdy != 0 && thrj+1 < xkolik )
- sArray[0][thrj+1] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + thrj+1 ];
- else
- sArray[0][thrj+1] = std::numeric_limits< Real >::max();
+ if( blockIdx.y != 0 && thrj+1 < xkolik )
+ sArray[0*sizeSArray + thrj+1] = aux[ blockIdx.y*blockDim.y*dimX - dimX + blockIdx.x*blockDim.x - 1 + thrj+1 ];
+ else
+ sArray[0*sizeSArray + thrj+1] = std::numeric_limits< Real >::max();
}
-
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+ if( i < dimX && j < dimY )
{
- sArray[thrj+1][thri+1] = aux[ j*mesh.getDimensions().x() + i ];
+ sArray[(thrj+1)*sizeSArray + thri+1] = aux[ j*dimX + i ];
}
__syncthreads();
-
+
while( changed[ 0 ] )
{
- __syncthreads();
-
- changed[ currentIndex] = false;
-
- //calculation of update cell
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
- {
- if( ! interfaceMap[ j * mesh.getDimensions().x() + i ] )
- {
- changed[ currentIndex ] = ptr.updateCell( sArray, thri+1, thrj+1, hx,hy);
- }
- }
- __syncthreads();
-
- //pyramid reduction
- if( blockDim.x*blockDim.y == 1024 )
+ __syncthreads();
+
+ changed[ currentIndex] = false;
+
+ //calculation of update cell
+ if( i < dimX && j < dimY )
+ {
+ if( ! interfaceMap[ j * dimX + i ] )
{
- if( currentIndex < 512 )
- {
- changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 512 ];
- }
+ changed[ currentIndex ] = ptr.updateCell<sizeSArray>( sArray, thri+1, thrj+1, hx,hy);
}
- __syncthreads();
- if( blockDim.x*blockDim.y >= 512 )
+ }
+ __syncthreads();
+
+ //pyramid reduction
+ if( blockDim.x*blockDim.y == 1024 )
+ {
+ if( currentIndex < 512 )
{
- if( currentIndex < 256 )
- {
- changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 256 ];
- }
+ changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 512 ];
}
- __syncthreads();
- if( blockDim.x*blockDim.y >= 256 )
+ }
+ __syncthreads();
+ if( blockDim.x*blockDim.y >= 512 )
+ {
+ if( currentIndex < 256 )
{
- if( currentIndex < 128 )
- {
- changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 128 ];
- }
+ changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 256 ];
}
- __syncthreads();
- if( blockDim.x*blockDim.y >= 128 )
+ }
+ __syncthreads();
+ if( blockDim.x*blockDim.y >= 256 )
+ {
+ if( currentIndex < 128 )
{
- if( currentIndex < 64 )
- {
- changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 64 ];
- }
+ changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 128 ];
}
- __syncthreads();
- if( currentIndex < 32 ) //POUZE IF JSOU SINCHRONNI NA JEDNOM WARPU
+ }
+ __syncthreads();
+ if( blockDim.x*blockDim.y >= 128 )
+ {
+ if( currentIndex < 64 )
{
- if( true ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 32 ];
- if( currentIndex < 16 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 16 ];
- if( currentIndex < 8 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 8 ];
- if( currentIndex < 4 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 4 ];
- if( currentIndex < 2 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 2 ];
- if( currentIndex < 1 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 1 ];
+ changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 64 ];
}
- if( changed[ 0 ] && thri == 0 && thrj == 0 )
- BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = 1;
- __syncthreads();
+ }
+ __syncthreads();
+ if( currentIndex < 32 )
+ {
+ if( true ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 32 ];
+ if( currentIndex < 16 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 16 ];
+ if( currentIndex < 8 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 8 ];
+ if( currentIndex < 4 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 4 ];
+ if( currentIndex < 2 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 2 ];
+ if( currentIndex < 1 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 1 ];
+ }
+ if( thri == 0 && thrj == 0 && changed[ 0 ] ){
+ BlockIterDevice[ blockIdx.y * gridDim.x + blockIdx.x ] = 1;
+ }
+ /*if( thri==0 && thrj == 0 && blockIdx.x == 0 && blockIdx.y == 0 )
+ {
+ for( int k = 15; k>-1; k-- ){
+ for( int l = 0; l < 16; l++ )
+ printf( "%f\t", sArray[k * 16 + l]);
+ printf( "\n");
+ }
+ printf( "\n");
+ }*/
+ __syncthreads();
}
-
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && (!interfaceMap[ j * mesh.getDimensions().x() + i ]) )
- aux[ j * mesh.getDimensions().x() + i ] = sArray[ thrj + 1 ][ thri + 1 ];
+ if( i < dimX && j < dimY )
+ helpFunc[ j * dimX + i ] = sArray[ ( thrj + 1 ) * sizeSArray + thri + 1 ];
+
+ }
}
#endif
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod3D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod3D_impl.h
index b024979cc6247022bd3ee156c877c2b47f71f99a..5af33cf29605ce983d75b61846ceeafeada7fde0 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod3D_impl.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod3D_impl.h
@@ -16,496 +16,649 @@
#include "tnlFastSweepingMethod.h"
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >::
FastSweepingMethod()
: maxIterations( 1 )
{
-
+
}
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
const Index&
FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >::
getMaxIterations() const
{
-
+
}
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
void
FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >::
setMaxIterations( const IndexType& maxIterations )
{
-
+
}
template< typename Real,
- typename Device,
- typename Index,
- typename Anisotropy >
+ typename Device,
+ typename Index,
+ typename Anisotropy >
void
FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >::
solve( const MeshPointer& mesh,
- const AnisotropyPointer& anisotropy,
- MeshFunctionPointer& u )
+ const AnisotropyPointer& anisotropy,
+ MeshFunctionPointer& u )
{
- MeshFunctionPointer auxPtr;
- InterfaceMapPointer interfaceMapPtr;
- auxPtr->setMesh( mesh );
- interfaceMapPtr->setMesh( mesh );
- std::cout << "Initiating the interface cells ..." << std::endl;
- BaseType::initInterface( u, auxPtr, interfaceMapPtr );
-#ifdef HAVE_CUDA
- cudaDeviceSynchronize();
-#endif
- auxPtr->save( "aux-ini.tnl" );
-
- typename MeshType::Cell cell( *mesh );
-
- IndexType iteration( 0 );
- MeshFunctionType aux = *auxPtr;
- InterfaceMapType interfaceMap = * interfaceMapPtr;
- while( iteration < this->maxIterations )
+ MeshFunctionPointer auxPtr;
+ InterfaceMapPointer interfaceMapPtr;
+ auxPtr->setMesh( mesh );
+ interfaceMapPtr->setMesh( mesh );
+ std::cout << "Initiating the interface cells ..." << std::endl;
+ BaseType::initInterface( u, auxPtr, interfaceMapPtr );
+ auxPtr->save( "aux-ini.tnl" );
+
+ typename MeshType::Cell cell( *mesh );
+
+ IndexType iteration( 0 );
+ MeshFunctionType aux = *auxPtr;
+ InterfaceMapType interfaceMap = * interfaceMapPtr;
+ while( iteration < this->maxIterations )
+ {
+ if( std::is_same< DeviceType, Devices::Host >::value )
{
- if( std::is_same< DeviceType, Devices::Host >::value )
- {
- for( cell.getCoordinates().z() = 0;
- cell.getCoordinates().z() < mesh->getDimensions().z();
- cell.getCoordinates().z()++ )
- {
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh->getDimensions().y();
- cell.getCoordinates().y()++ )
- {
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh->getDimensions().x();
- cell.getCoordinates().x()++ )
- {
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- }
- //aux.save( "aux-1.tnl" );
-
- for( cell.getCoordinates().z() = 0;
- cell.getCoordinates().z() < mesh->getDimensions().z();
- cell.getCoordinates().z()++ )
- {
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh->getDimensions().y();
- cell.getCoordinates().y()++ )
- {
- for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
- cell.getCoordinates().x() >= 0 ;
- cell.getCoordinates().x()-- )
- {
- //std::cerr << "2 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- }
- //aux.save( "aux-2.tnl" );
- for( cell.getCoordinates().z() = 0;
- cell.getCoordinates().z() < mesh->getDimensions().z();
- cell.getCoordinates().z()++ )
- {
- for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
- cell.getCoordinates().y() >= 0 ;
- cell.getCoordinates().y()-- )
- {
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh->getDimensions().x();
- cell.getCoordinates().x()++ )
- {
- //std::cerr << "3 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- }
- //aux.save( "aux-3.tnl" );
-
- for( cell.getCoordinates().z() = 0;
- cell.getCoordinates().z() < mesh->getDimensions().z();
- cell.getCoordinates().z()++ )
- {
- for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
- cell.getCoordinates().y() >= 0;
- cell.getCoordinates().y()-- )
- {
- for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
- cell.getCoordinates().x() >= 0 ;
- cell.getCoordinates().x()-- )
- {
- //std::cerr << "4 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- }
- //aux.save( "aux-4.tnl" );
-
- for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
- cell.getCoordinates().z() >= 0;
- cell.getCoordinates().z()-- )
- {
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh->getDimensions().y();
- cell.getCoordinates().y()++ )
- {
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh->getDimensions().x();
- cell.getCoordinates().x()++ )
- {
- //std::cerr << "5 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- }
- //aux.save( "aux-5.tnl" );
-
- for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
- cell.getCoordinates().z() >= 0;
- cell.getCoordinates().z()-- )
- {
- for( cell.getCoordinates().y() = 0;
- cell.getCoordinates().y() < mesh->getDimensions().y();
- cell.getCoordinates().y()++ )
- {
- for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
- cell.getCoordinates().x() >= 0 ;
- cell.getCoordinates().x()-- )
- {
- //std::cerr << "6 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- }
- //aux.save( "aux-6.tnl" );
-
- for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
- cell.getCoordinates().z() >= 0;
- cell.getCoordinates().z()-- )
- {
- for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
- cell.getCoordinates().y() >= 0 ;
- cell.getCoordinates().y()-- )
- {
- for( cell.getCoordinates().x() = 0;
- cell.getCoordinates().x() < mesh->getDimensions().x();
- cell.getCoordinates().x()++ )
- {
- //std::cerr << "7 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- }
- //aux.save( "aux-7.tnl" );
-
- for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
- cell.getCoordinates().z() >= 0;
- cell.getCoordinates().z()-- )
- {
- for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
- cell.getCoordinates().y() >= 0;
- cell.getCoordinates().y()-- )
- {
- for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
- cell.getCoordinates().x() >= 0 ;
- cell.getCoordinates().x()-- )
- {
- //std::cerr << "8 -> ";
- cell.refresh();
- if( ! interfaceMap( cell ) )
- this->updateCell( aux, cell );
- }
- }
- }
- }
- if( std::is_same< DeviceType, Devices::Cuda >::value )
- {
- // TODO: CUDA code
-#ifdef HAVE_CUDA
- const int cudaBlockSize( 8 );
- int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().x(), cudaBlockSize );
- int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().y(), cudaBlockSize );
- int numBlocksZ = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().z(), cudaBlockSize );
- if( cudaBlockSize * cudaBlockSize * cudaBlockSize > 1024 || numBlocksX > 1024 || numBlocksY > 1024 || numBlocksZ > 64 )
- std::cout << "Invalid kernel call. Dimensions of grid are max: [1024,1024,64], and maximum threads per block are 1024!" << std::endl;
- dim3 blockSize( cudaBlockSize, cudaBlockSize, cudaBlockSize );
- dim3 gridSize( numBlocksX, numBlocksY, numBlocksZ );
-
- tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr;
-
- int *BlockIterDevice;
- int BlockIterD = 1;
-
- cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY * numBlocksZ ) * sizeof( int ) );
- int nBlocks = ( numBlocksX * numBlocksY * numBlocksZ )/512 + ((( numBlocksX * numBlocksY * numBlocksZ )%512 != 0) ? 1:0);
- int *dBlock;
- cudaMalloc(&dBlock, nBlocks * sizeof( int ) );
-
- while( BlockIterD )
- {
- CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
- interfaceMapPtr.template getData< Device >(),
- auxPtr.template modifyData< Device>(),
- BlockIterDevice );
- CudaParallelReduc<<< nBlocks , 512 >>>( BlockIterDevice, dBlock, ( numBlocksX * numBlocksY * numBlocksZ ) );
- CudaParallelReduc<<< 1, nBlocks >>>( dBlock, dBlock, nBlocks );
-
- cudaMemcpy(&BlockIterD, &dBlock[0], sizeof( int ), cudaMemcpyDeviceToHost);
-
- /*for( int i = 1; i < numBlocksX * numBlocksY; i++ )
- BlockIter[ 0 ] = BlockIter[ 0 ] || BlockIter[ i ];*/
-
+ int numThreadsPerBlock = 64;
+
+
+ int numBlocksX = mesh->getDimensions().x() / numThreadsPerBlock + (mesh->getDimensions().x() % numThreadsPerBlock != 0 ? 1:0);
+ int numBlocksY = mesh->getDimensions().y() / numThreadsPerBlock + (mesh->getDimensions().y() % numThreadsPerBlock != 0 ? 1:0);
+ int numBlocksZ = mesh->getDimensions().z() / numThreadsPerBlock + (mesh->getDimensions().z() % numThreadsPerBlock != 0 ? 1:0);
+ //std::cout << "numBlocksX = " << numBlocksX << std::endl;
+
+ /*Real **sArray = new Real*[numBlocksX*numBlocksY];
+ for( int i = 0; i < numBlocksX * numBlocksY; i++ )
+ sArray[ i ] = new Real [ (numThreadsPerBlock + 2)*(numThreadsPerBlock + 2)];*/
+
+ ArrayContainer BlockIterHost;
+ BlockIterHost.setSize( numBlocksX * numBlocksY * numBlocksZ );
+ BlockIterHost.setValue( 1 );
+ int IsCalculationDone = 1;
+
+ MeshFunctionPointer helpFunc( mesh );
+ MeshFunctionPointer helpFunc1( mesh );
+ helpFunc1 = auxPtr;
+ auxPtr = helpFunc;
+ helpFunc = helpFunc1;
+ //std::cout<< "Size = " << BlockIterHost.getSize() << std::endl;
+ /*for( int k = numBlocksX-1; k >-1; k-- ){
+ for( int l = 0; l < numBlocksY; l++ ){
+ std::cout<< BlockIterHost[ l*numBlocksX + k ];
+ }
+ std::cout<<std::endl;
+ }
+ std::cout<<std::endl;*/
+ unsigned int numWhile = 0;
+ while( IsCalculationDone )
+ {
+ IsCalculationDone = 0;
+ helpFunc1 = auxPtr;
+ auxPtr = helpFunc;
+ helpFunc = helpFunc1;
+ this->template updateBlocks< 66 >( interfaceMap, *auxPtr, *helpFunc, BlockIterHost, numThreadsPerBlock/*, sArray*/ );
+
+ //Reduction
+ for( int i = 0; i < BlockIterHost.getSize(); i++ ){
+ if( IsCalculationDone == 0 ){
+ IsCalculationDone = IsCalculationDone || BlockIterHost[ i ];
+ //break;
}
- cudaFree( BlockIterDevice );
- cudaFree( dBlock );
- cudaDeviceSynchronize();
- TNL_CHECK_CUDA_DEVICE;
- aux = *auxPtr;
- interfaceMap = *interfaceMapPtr;
-#endif
- }
+ }
+ numWhile++;
+ std::cout <<"numWhile = "<< numWhile <<std::endl;
+ /*for( int k = 0; k < numBlocksZ; k++ ){
+ for( int j = numBlocksY-1; j>-1; j-- ){
+ for( int i = 0; i < numBlocksX; i++ ){
+ //std::cout << (*auxPtr)[ k * numBlocksX * numBlocksY + j * numBlocksX + i ] << " ";
+ std::cout << BlockIterHost[ k * numBlocksX * numBlocksY + j * numBlocksX + i ];
+ }
+ std::cout << std::endl;
+ }
+ std::cout << std::endl;
+ }
+ std::cout << std::endl;*/
+
+ this->getNeighbours( BlockIterHost, numBlocksX, numBlocksY, numBlocksZ );
- //aux.save( "aux-8.tnl" );
- iteration++;
+ /*for( int k = 0; k < numBlocksZ; k++ ){
+ for( int j = numBlocksY-1; j>-1; j-- ){
+ for( int i = 0; i < numBlocksX; i++ ){
+ //std::cout << (*auxPtr)[ k * numBlocksX * numBlocksY + j * numBlocksX + i ] << " ";
+ std::cout << BlockIterHost[ k * numBlocksX * numBlocksY + j * numBlocksX + i ];
+ }
+ std::cout << std::endl;
+ }
+ std::cout << std::endl;
+ }*/
+
+ /*for( int j = numBlocksY-1; j>-1; j-- ){
+ for( int i = 0; i < numBlocksX; i++ )
+ std::cout << "BlockIterHost = "<< j*numBlocksX + i<< " ," << BlockIterHost[ j * numBlocksX + i ];
+ std::cout << std::endl;
+ }
+ std::cout << std::endl;*/
+
+ //std::cout<<std::endl;
+ //string s( "aux-"+ std::to_string(numWhile) + ".tnl");
+ //aux.save( s );
+ }
+ if( numWhile == 1 ){
+ auxPtr = helpFunc;
+ }
+ aux = *auxPtr;
+
+ /*for( cell.getCoordinates().z() = 0;
+ cell.getCoordinates().z() < mesh->getDimensions().z();
+ cell.getCoordinates().z()++ )
+ {
+ for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh->getDimensions().y();
+ cell.getCoordinates().y()++ )
+ {
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh->getDimensions().x();
+ cell.getCoordinates().x()++ )
+ {
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+ }
+ //aux.save( "aux-1.tnl" );
+
+ for( cell.getCoordinates().z() = 0;
+ cell.getCoordinates().z() < mesh->getDimensions().z();
+ cell.getCoordinates().z()++ )
+ {
+ for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh->getDimensions().y();
+ cell.getCoordinates().y()++ )
+ {
+ for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
+ cell.getCoordinates().x() >= 0 ;
+ cell.getCoordinates().x()-- )
+ {
+ //std::cerr << "2 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+ }
+ //aux.save( "aux-2.tnl" );
+ for( cell.getCoordinates().z() = 0;
+ cell.getCoordinates().z() < mesh->getDimensions().z();
+ cell.getCoordinates().z()++ )
+ {
+ for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
+ cell.getCoordinates().y() >= 0 ;
+ cell.getCoordinates().y()-- )
+ {
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh->getDimensions().x();
+ cell.getCoordinates().x()++ )
+ {
+ //std::cerr << "3 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+ }
+ //aux.save( "aux-3.tnl" );
+
+ for( cell.getCoordinates().z() = 0;
+ cell.getCoordinates().z() < mesh->getDimensions().z();
+ cell.getCoordinates().z()++ )
+ {
+ for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
+ cell.getCoordinates().y() >= 0;
+ cell.getCoordinates().y()-- )
+ {
+ for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
+ cell.getCoordinates().x() >= 0 ;
+ cell.getCoordinates().x()-- )
+ {
+ //std::cerr << "4 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+ }
+ //aux.save( "aux-4.tnl" );
+
+ for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
+ cell.getCoordinates().z() >= 0;
+ cell.getCoordinates().z()-- )
+ {
+ for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh->getDimensions().y();
+ cell.getCoordinates().y()++ )
+ {
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh->getDimensions().x();
+ cell.getCoordinates().x()++ )
+ {
+ //std::cerr << "5 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+ }
+ //aux.save( "aux-5.tnl" );
+
+ for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
+ cell.getCoordinates().z() >= 0;
+ cell.getCoordinates().z()-- )
+ {
+ for( cell.getCoordinates().y() = 0;
+ cell.getCoordinates().y() < mesh->getDimensions().y();
+ cell.getCoordinates().y()++ )
+ {
+ for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
+ cell.getCoordinates().x() >= 0 ;
+ cell.getCoordinates().x()-- )
+ {
+ //std::cerr << "6 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+ }
+ //aux.save( "aux-6.tnl" );
+
+ for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
+ cell.getCoordinates().z() >= 0;
+ cell.getCoordinates().z()-- )
+ {
+ for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
+ cell.getCoordinates().y() >= 0 ;
+ cell.getCoordinates().y()-- )
+ {
+ for( cell.getCoordinates().x() = 0;
+ cell.getCoordinates().x() < mesh->getDimensions().x();
+ cell.getCoordinates().x()++ )
+ {
+ //std::cerr << "7 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+ }
+ //aux.save( "aux-7.tnl" );
+
+ for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
+ cell.getCoordinates().z() >= 0;
+ cell.getCoordinates().z()-- )
+ {
+ for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
+ cell.getCoordinates().y() >= 0;
+ cell.getCoordinates().y()-- )
+ {
+ for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
+ cell.getCoordinates().x() >= 0 ;
+ cell.getCoordinates().x()-- )
+ {
+ //std::cerr << "8 -> ";
+ cell.refresh();
+ if( ! interfaceMap( cell ) )
+ this->updateCell( aux, cell );
+ }
+ }
+ }*/
+ }
+ if( std::is_same< DeviceType, Devices::Cuda >::value )
+ {
+ // TODO: CUDA code
+#ifdef HAVE_CUDA
+ const int cudaBlockSize( 8 );
+ int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().x(), cudaBlockSize );
+ int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().y(), cudaBlockSize );
+ int numBlocksZ = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().z(), cudaBlockSize );
+ if( cudaBlockSize * cudaBlockSize * cudaBlockSize > 1024 || numBlocksX > 1024 || numBlocksY > 1024 || numBlocksZ > 64 )
+ std::cout << "Invalid kernel call. Dimensions of grid are max: [1024,1024,64], and maximum threads per block are 1024!" << std::endl;
+ dim3 blockSize( cudaBlockSize, cudaBlockSize, cudaBlockSize );
+ dim3 gridSize( numBlocksX, numBlocksY, numBlocksZ );
+
+ tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr;
+
+
+ int BlockIterD = 1;
+
+ TNL::Containers::Array< int, Devices::Cuda, IndexType > BlockIterDevice;
+ BlockIterDevice.setSize( numBlocksX * numBlocksY * numBlocksZ );
+ BlockIterDevice.setValue( 1 );
+ TNL::Containers::Array< int, Devices::Cuda, IndexType > BlockIterPom;
+ BlockIterPom.setSize( numBlocksX * numBlocksY * numBlocksZ );
+ BlockIterPom.setValue( 0 );
+ /*int *BlockIterDevice;
+ cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY * numBlocksZ ) * sizeof( int ) );*/
+ int nBlocks = ( numBlocksX * numBlocksY * numBlocksZ )/512 + ((( numBlocksX * numBlocksY * numBlocksZ )%512 != 0) ? 1:0);
- }
- aux.save("aux-final.tnl");
+ TNL::Containers::Array< int, Devices::Cuda, IndexType > dBlock;
+ dBlock.setSize( nBlocks );
+ dBlock.setValue( 0 );
+
+ int nBlocksNeigh = ( numBlocksX * numBlocksY * numBlocksZ )/1024 + ((( numBlocksX * numBlocksY * numBlocksZ )%1024 != 0) ? 1:0);
+ /*int *dBlock;
+ cudaMalloc(&dBlock, nBlocks * sizeof( int ) );*/
+ MeshFunctionPointer helpFunc1( mesh );
+ MeshFunctionPointer helpFunc( mesh );
+
+ helpFunc1 = auxPtr;
+ auxPtr = helpFunc;
+ helpFunc = helpFunc1;
+ int numIter = 0;
+
+ while( BlockIterD )
+ {
+ helpFunc1 = auxPtr;
+ auxPtr = helpFunc;
+ helpFunc = helpFunc1;
+ TNL_CHECK_CUDA_DEVICE;
+
+ CudaUpdateCellCaller< 10 ><<< gridSize, blockSize >>>( ptr,
+ interfaceMapPtr.template getData< Device >(),
+ auxPtr.template getData< Device>(),
+ helpFunc.template modifyData< Device>(),
+ BlockIterDevice );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+
+ GetNeighbours3D<<< nBlocksNeigh, 1024 >>>( BlockIterDevice, BlockIterPom, numBlocksX, numBlocksY, numBlocksZ );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+ BlockIterDevice = BlockIterPom;
+
+ CudaParallelReduc<<< nBlocks , 512 >>>( BlockIterDevice, dBlock, ( numBlocksX * numBlocksY * numBlocksZ ) );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+
+ CudaParallelReduc<<< 1, nBlocks >>>( dBlock, dBlock, nBlocks );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+ cudaMemcpy(&BlockIterD, &dBlock[0], sizeof( int ), cudaMemcpyDeviceToHost);
+ numIter++;
+ /*for( int i = 1; i < numBlocksX * numBlocksY; i++ )
+ BlockIter[ 0 ] = BlockIter[ 0 ] || BlockIter[ i ];*/
+
+ }
+ if( numIter == 1 ){
+ auxPtr = helpFunc;
+ }
+ //cudaFree( BlockIterDevice );
+ //cudaFree( dBlock );
+ cudaDeviceSynchronize();
+ TNL_CHECK_CUDA_DEVICE;
+ aux = *auxPtr;
+ interfaceMap = *interfaceMapPtr;
+#endif
+ }
+
+ //aux.save( "aux-8.tnl" );
+ iteration++;
+
+ }
+ aux.save("aux-final.tnl");
}
#ifdef HAVE_CUDA
-template < typename Real, typename Device, typename Index >
+template < typename Index >
+__global__ void GetNeighbours3D( TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice,
+ TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterPom,
+ int numBlockX, int numBlockY, int numBlockZ )
+{
+ int i = blockIdx.x * 1024 + threadIdx.x;
+
+ if( i < numBlockX * numBlockY * numBlockZ )
+ {
+ int pom = 0;//BlockIterPom[ i ] = 0;
+ int m=0, l=0, k=0;
+ l = i/( numBlockX * numBlockY );
+ k = (i-l*numBlockX * numBlockY )/(numBlockX );
+ m = (i-l*numBlockX * numBlockY )%( numBlockX );
+ if( m > 0 && BlockIterDevice[ i - 1 ] ){
+ pom = 1;//BlockIterPom[ i ] = 1;
+ }else if( m < numBlockX -1 && BlockIterDevice[ i + 1 ] ){
+ pom = 1;//BlockIterPom[ i ] = 1;
+ }else if( k > 0 && BlockIterDevice[ i - numBlockX ] ){
+ pom = 1;// BlockIterPom[ i ] = 1;
+ }else if( k < numBlockY -1 && BlockIterDevice[ i + numBlockX ] ){
+ pom = 1;//BlockIterPom[ i ] = 1;
+ }else if( l > 0 && BlockIterDevice[ i - numBlockX*numBlockY ] ){
+ pom = 1;
+ }else if( l < numBlockZ-1 && BlockIterDevice[ i + numBlockX*numBlockY ] ){
+ pom = 1;
+ }
+
+ BlockIterPom[ i ] = pom;//BlockIterPom[ i ];
+ }
+}
+
+template < int sizeSArray, typename Real, typename Device, typename Index >
__global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr,
- const Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index >, 3, bool >& interfaceMap,
- Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& aux,
- int *BlockIterDevice )
+ const Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index >, 3, bool >& interfaceMap,
+ const Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& aux,
+ Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& helpFunc,
+ TNL::Containers::Array< int, Devices::Cuda, Index > BlockIterDevice )
{
- int thri = threadIdx.x; int thrj = threadIdx.y; int thrk = threadIdx.z;
- int blIdx = blockIdx.x; int blIdy = blockIdx.y; int blIdz = blockIdx.z;
- int i = threadIdx.x + blockDim.x*blockIdx.x;
- int j = blockDim.y*blockIdx.y + threadIdx.y;
- int k = blockDim.z*blockIdx.z + threadIdx.z;
- int currentIndex = thrk * blockDim.x * blockDim.y + thrj * blockDim.x + thri;
+ int thri = threadIdx.x; int thrj = threadIdx.y; int thrk = threadIdx.z;
+ int blIdx = blockIdx.x; int blIdy = blockIdx.y; int blIdz = blockIdx.z;
+ int i = threadIdx.x + blockDim.x*blockIdx.x;
+ int j = blockDim.y*blockIdx.y + threadIdx.y;
+ int k = blockDim.z*blockIdx.z + threadIdx.z;
+ int currentIndex = thrk * blockDim.x * blockDim.y + thrj * blockDim.x + thri;
+
+ if( BlockIterDevice[ blockIdx.z * gridDim.x * gridDim.y + blockIdx.y * gridDim.x + blockIdx.x ] )
+ {
+ __syncthreads();
- __shared__ volatile bool changed[8*8*8];
- changed[ currentIndex ] = false;
+ __shared__ volatile bool changed[ 8*8*8/*(sizeSArray - 2)*(sizeSArray - 2)*(sizeSArray - 2)*/];
+ changed[ currentIndex ] = false;
if( thrj == 0 && thri == 0 && thrk == 0 )
- changed[ 0 ] = true;
+ changed[ 0 ] = true;
const Meshes::Grid< 3, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
- __shared__ Real hx;
- __shared__ Real hy;
- __shared__ Real hz;
+ __shared__ Real hx; __shared__ int dimX;
+ __shared__ Real hy; __shared__ int dimY;
+ __shared__ Real hz; __shared__ int dimZ;
+
if( thrj == 1 && thri == 1 && thrk == 1 )
{
- hx = mesh.getSpaceSteps().x();
- hy = mesh.getSpaceSteps().y();
- hz = mesh.getSpaceSteps().z();
+ //printf( "We are in the calculation. Block = %d.\n",blockIdx.z * gridDim.x * gridDim.y + blockIdx.y * gridDim.x + blockIdx.x );
+ hx = mesh.getSpaceSteps().x();
+ hy = mesh.getSpaceSteps().y();
+ hz = mesh.getSpaceSteps().z();
+ dimX = mesh.getDimensions().x();
+ dimY = mesh.getDimensions().y();
+ dimZ = mesh.getDimensions().z();
+ BlockIterDevice[ blockIdx.z * gridDim.x * gridDim.y + blockIdx.y * gridDim.x + blockIdx.x ] = 0;
}
- __shared__ volatile Real sArray[10][10][10];
- sArray[thrk][thrj][thri] = std::numeric_limits< Real >::max();
- if(thri == 0 )
- {
- sArray[8][thrj+1][thrk+1] = std::numeric_limits< Real >::max();
- sArray[9][thrj+1][thrk+1] = std::numeric_limits< Real >::max();
- sArray[thrk+1][thrj+1][8] = std::numeric_limits< Real >::max();
- sArray[thrk+1][thrj+1][9] = std::numeric_limits< Real >::max();
- sArray[thrj+1][8][thrk+1] = std::numeric_limits< Real >::max();
- sArray[thrj+1][9][thrk+1] = std::numeric_limits< Real >::max();
- }
-
+ __shared__ volatile Real sArray[ 10*10*10/*sizeSArray * sizeSArray * sizeSArray*/ ];
+ sArray[(thrk+1)* sizeSArray * sizeSArray + (thrj+1) *sizeSArray + thri+1] = std::numeric_limits< Real >::max();
+
//filling sArray edges
- int dimX = mesh.getDimensions().x(); int dimY = mesh.getDimensions().y();
- int dimZ = mesh.getDimensions().z();
- __shared__ volatile int numOfBlockx;
- __shared__ volatile int numOfBlocky;
- __shared__ volatile int numOfBlockz;
- __shared__ int xkolik;
- __shared__ int ykolik;
- __shared__ int zkolik;
- if( thri == 0 && thrj == 0 && thrk == 0 )
- {
- xkolik = blockDim.x + 1;
- ykolik = blockDim.y + 1;
- zkolik = blockDim.z + 1;
- numOfBlocky = dimY/blockDim.y + ((dimY%blockDim.y != 0) ? 1:0);
- numOfBlockx = dimX/blockDim.x + ((dimX%blockDim.x != 0) ? 1:0);
- numOfBlockz = dimZ/blockDim.z + ((dimZ%blockDim.z != 0) ? 1:0);
-
- if( numOfBlockx - 1 == blIdx )
- xkolik = dimX - (blIdx)*blockDim.x+1;
-
- if( numOfBlocky -1 == blIdy )
- ykolik = dimY - (blIdy)*blockDim.y+1;
- if( numOfBlockz-1 == blIdz )
- zkolik = dimZ - (blIdz)*blockDim.z+1;
-
- BlockIterDevice[ blIdz * numOfBlockx * numOfBlocky + blIdy * numOfBlockx + blIdx ] = 0;
- }
+ int numOfBlockx;
+ int numOfBlocky;
+ int numOfBlockz;
+ int xkolik;
+ int ykolik;
+ int zkolik;
+ xkolik = blockDim.x + 1;
+ ykolik = blockDim.y + 1;
+ zkolik = blockDim.z + 1;
+ numOfBlockx = gridDim.x;
+ numOfBlocky = gridDim.y;
+ numOfBlockz = gridDim.z;
+ __syncthreads();
+
+ if( numOfBlockx - 1 == blIdx )
+ xkolik = dimX - (blIdx)*blockDim.x+1;
+ if( numOfBlocky -1 == blIdy )
+ ykolik = dimY - (blIdy)*blockDim.y+1;
+ if( numOfBlockz-1 == blIdz )
+ zkolik = dimZ - (blIdz)*blockDim.z+1;
__syncthreads();
if( thri == 0 )
{
- if( blIdx != 0 && thrj+1 < ykolik && thrk+1 < zkolik )
- sArray[thrk+1][thrj+1][0] = aux[ blIdz*blockDim.z * dimX * dimY + blIdy * blockDim.y*dimX + blIdx*blockDim.x + thrj * dimX -1 + thrk*dimX*dimY ];
- else
- sArray[thrk+1][thrj+1][0] = std::numeric_limits< Real >::max();
+ if( blIdx != 0 && thrj+1 < ykolik && thrk+1 < zkolik )
+ sArray[(thrk+1 )* sizeSArray * sizeSArray + (thrj+1)*sizeSArray + 0] = aux[ blIdz*blockDim.z * dimX * dimY + blIdy * blockDim.y*dimX + blIdx*blockDim.x + thrj * dimX -1 + thrk*dimX*dimY ];
+ else
+ sArray[(thrk+1)* sizeSArray * sizeSArray + (thrj+1)*sizeSArray + 0] = std::numeric_limits< Real >::max();
}
if( thri == 1 )
{
- if( dimX > (blIdx+1) * blockDim.x && thrj+1 < ykolik && thrk+1 < zkolik )
- sArray[thrk+1][thrj+1][9] = aux[ blIdz*blockDim.z * dimX * dimY + blIdy *blockDim.y*dimX+ blIdx*blockDim.x + blockDim.x + thrj * dimX + thrk*dimX*dimY ];
- else
- sArray[thrk+1][thrj+1][9] = std::numeric_limits< Real >::max();
+ if( dimX > (blIdx+1) * blockDim.x && thrj+1 < ykolik && thrk+1 < zkolik )
+ sArray[ (thrk+1) * sizeSArray * sizeSArray + (thrj+1) *sizeSArray + xkolik ] = aux[ blIdz*blockDim.z * dimX * dimY + blIdy *blockDim.y*dimX+ blIdx*blockDim.x + blockDim.x + thrj * dimX + thrk*dimX*dimY ];
+ else
+ sArray[ (thrk+1) * sizeSArray * sizeSArray + (thrj+1)*sizeSArray + xkolik] = std::numeric_limits< Real >::max();
}
if( thri == 2 )
{
- if( blIdy != 0 && thrj+1 < xkolik && thrk+1 < zkolik )
- sArray[thrk+1][0][thrj+1] = aux[ blIdz*blockDim.z * dimX * dimY + blIdy * blockDim.y*dimX + blIdx*blockDim.x - dimX + thrj + thrk*dimX*dimY ];
- else
- sArray[thrk+1][0][thrj+1] = std::numeric_limits< Real >::max();
+ if( blIdy != 0 && thrj+1 < xkolik && thrk+1 < zkolik )
+ sArray[ (thrk+1) * sizeSArray * sizeSArray +0*sizeSArray + thrj+1] = aux[ blIdz*blockDim.z * dimX * dimY + blIdy * blockDim.y*dimX + blIdx*blockDim.x - dimX + thrj + thrk*dimX*dimY ];
+ else
+ sArray[ (thrk+1) * sizeSArray * sizeSArray + 0*sizeSArray + thrj+1] = std::numeric_limits< Real >::max();
}
if( thri == 3 )
{
- if( dimY > (blIdy+1) * blockDim.y && thrj+1 < xkolik && thrk+1 < zkolik )
- sArray[thrk+1][9][thrj+1] = aux[ blIdz*blockDim.z * dimX * dimY + (blIdy+1) * blockDim.y*dimX + blIdx*blockDim.x + thrj + thrk*dimX*dimY ];
- else
- sArray[thrk+1][9][thrj+1] = std::numeric_limits< Real >::max();
+ if( dimY > (blIdy+1) * blockDim.y && thrj+1 < xkolik && thrk+1 < zkolik )
+ sArray[ (thrk+1) * sizeSArray * sizeSArray + ykolik*sizeSArray + thrj+1] = aux[ blIdz*blockDim.z * dimX * dimY + (blIdy+1) * blockDim.y*dimX + blIdx*blockDim.x + thrj + thrk*dimX*dimY ];
+ else
+ sArray[ (thrk+1) * sizeSArray * sizeSArray + ykolik*sizeSArray + thrj+1] = std::numeric_limits< Real >::max();
}
if( thri == 4 )
{
- if( blIdz != 0 && thrj+1 < ykolik && thrk+1 < xkolik )
- sArray[0][thrj+1][thrk+1] = aux[ blIdz*blockDim.z * dimX * dimY + blIdy * blockDim.y*dimX + blIdx*blockDim.x - dimX * dimY + thrj * dimX + thrk ];
- else
- sArray[0][thrj+1][thrk+1] = std::numeric_limits< Real >::max();
+ if( blIdz != 0 && thrj+1 < ykolik && thrk+1 < xkolik )
+ sArray[ 0 * sizeSArray * sizeSArray +(thrj+1 )* sizeSArray + thrk+1] = aux[ blIdz*blockDim.z * dimX * dimY + blIdy * blockDim.y*dimX + blIdx*blockDim.x - dimX * dimY + thrj * dimX + thrk ];
+ else
+ sArray[0 * sizeSArray * sizeSArray + (thrj+1) *sizeSArray + thrk+1] = std::numeric_limits< Real >::max();
}
if( thri == 5 )
{
- if( dimZ > (blIdz+1) * blockDim.z && thrj+1 < ykolik && thrk+1 < xkolik )
- sArray[9][thrj+1][thrk+1] = aux[ (blIdz+1)*blockDim.z * dimX * dimY + blIdy * blockDim.y*dimX + blIdx*blockDim.x + thrj * dimX + thrk ];
- else
- sArray[9][thrj+1][thrk+1] = std::numeric_limits< Real >::max();
+ if( dimZ > (blIdz+1) * blockDim.z && thrj+1 < ykolik && thrk+1 < xkolik )
+ sArray[zkolik * sizeSArray * sizeSArray + (thrj+1) * sizeSArray + thrk+1] = aux[ (blIdz+1)*blockDim.z * dimX * dimY + blIdy * blockDim.y*dimX + blIdx*blockDim.x + thrj * dimX + thrk ];
+ else
+ sArray[zkolik * sizeSArray * sizeSArray + (thrj+1) * sizeSArray + thrk+1] = std::numeric_limits< Real >::max();
}
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && k < mesh.getDimensions().z() )
+ if( i < dimX && j < dimY && k < dimZ )
{
- sArray[thrk+1][thrj+1][thri+1] = aux[ k*dimX*dimY + j*dimX + i ];
+ sArray[(thrk+1) * sizeSArray * sizeSArray + (thrj+1) *sizeSArray + thri+1] = aux[ k*dimX*dimY + j*dimX + i ];
}
- __shared__ volatile int loopcounter;
- loopcounter = 0;
__syncthreads();
+
while( changed[ 0 ] )
{
- __syncthreads();
-
- changed[ currentIndex ] = false;
-
- //calculation of update cell
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && k < dimZ )
- {
- if( ! interfaceMap[ k*dimX*dimY + j * mesh.getDimensions().x() + i ] )
- {
- changed[ currentIndex ] = ptr.updateCell( sArray, thri+1, thrj+1, thrk+1, hx,hy,hz);
- }
- }
- __syncthreads();
-
- //pyramid reduction
- if( blockDim.x*blockDim.y*blockDim.z == 1024 )
- {
- if( currentIndex < 512 )
- {
- changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 512 ];
- }
- }
- __syncthreads();
- if( blockDim.x*blockDim.y*blockDim.z >= 512 )
+ __syncthreads();
+
+ changed[ currentIndex ] = false;
+
+ //calculation of update cell
+ if( i < dimX && j < dimY && k < dimZ )
+ {
+ if( ! interfaceMap[ k*dimX*dimY + j * dimX + i ] )
{
- if( currentIndex < 256 )
- {
- changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 256 ];
- }
+ changed[ currentIndex ] = ptr.updateCell3D< sizeSArray >( sArray, thri+1, thrj+1, thrk+1, hx,hy,hz);
}
- __syncthreads();
- if( blockDim.x*blockDim.y*blockDim.z >= 256 )
+ }
+ __syncthreads();
+
+ //pyramid reduction
+ if( blockDim.x*blockDim.y*blockDim.z == 1024 )
+ {
+ if( currentIndex < 512 )
{
- if( currentIndex < 128 )
- {
- changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 128 ];
- }
+ changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 512 ];
}
- __syncthreads();
- if( blockDim.x*blockDim.y*blockDim.z >= 128 )
+ }
+ __syncthreads();
+ if( blockDim.x*blockDim.y*blockDim.z >= 512 )
+ {
+ if( currentIndex < 256 )
{
- if( currentIndex < 64 )
- {
- changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 64 ];
- }
+ changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 256 ];
}
- __syncthreads();
- if( currentIndex < 32 ) //POUZE IF JSOU SINCHRONNI NA JEDNOM WARPU
+ }
+ __syncthreads();
+ if( blockDim.x*blockDim.y*blockDim.z >= 256 )
+ {
+ if( currentIndex < 128 )
{
- if( true ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 32 ];
- if( currentIndex < 16 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 16 ];
- if( currentIndex < 8 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 8 ];
- if( currentIndex < 4 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 4 ];
- if( currentIndex < 2 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 2 ];
- if( currentIndex < 1 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 1 ];
+ changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 128 ];
}
- __syncthreads();
-
- /*if(thri == 0 && thrj ==0 && thrk ==0 && blIdx == 0 && blIdy == 0 && blIdz == 0)
- {
- for(int m = 0; m < 8; m++){
- for(int n = 0; n<8; n++){
- for(int b=0; b<8; b++)
- printf(" %i ", changed[m*64 + n*8 + b]);
- printf("\n");
- }
- printf("\n \n");
- }
- }*/
- if( changed[ 0 ] && thri == 0 && thrj == 0 && thrk == 0 )
+ }
+ __syncthreads();
+ if( blockDim.x*blockDim.y*blockDim.z >= 128 )
+ {
+ if( currentIndex < 64 )
{
- //loopcounter++;
- BlockIterDevice[ blIdz * numOfBlockx * numOfBlocky + blIdy * numOfBlockx + blIdx ] = 1;
+ changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 64 ];
}
- __syncthreads();
- /*if(thri == 0 && thrj==0 && thrk==0)
- printf("%i \n",loopcounter);
- if(loopcounter == 500)
- break;*/
+ }
+ __syncthreads();
+ if( currentIndex < 32 )
+ {
+ if( true ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 32 ];
+ if( currentIndex < 16 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 16 ];
+ if( currentIndex < 8 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 8 ];
+ if( currentIndex < 4 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 4 ];
+ if( currentIndex < 2 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 2 ];
+ if( currentIndex < 1 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 1 ];
+ }
+ __syncthreads();
+
+ /*if(thri == 0 && thrj ==0 && thrk ==0 && blIdx == 0 && blIdy == 0 && blIdz == 0)
+ {
+ //for(int m = 0; m < 8; m++){
+ int m = 4;
+ for(int n = 0; n<8; n++){
+ for(int b=0; b<8; b++)
+ printf(" %i ", changed[m*64 + n*8 + b]);
+ printf("\n");
+ }
+ printf("\n \n");
+ }
+ //}*/
+
+ if( changed[ 0 ] && thri == 0 && thrj == 0 && thrk == 0 )
+ {
+ //printf( "Setting block calculation. Block = %d.\n",blockIdx.z * gridDim.x * gridDim.y + blockIdx.y * gridDim.x + blockIdx.x );
+ BlockIterDevice[ blIdz * gridDim.x * gridDim.y + blIdy * gridDim.x + blIdx ] = 1;
+ }
+ __syncthreads();
}
-
- if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && k < dimZ && (!interfaceMap[ k*dimX*dimY+j * mesh.getDimensions().x() + i ]) )
- aux[ k*dimX*dimY + j * mesh.getDimensions().x() + i ] = sArray[thrk+1][ thrj + 1 ][ thri + 1 ];
-}
+
+ if( i < dimX && j < dimY && k < dimZ )
+ helpFunc[ k*dimX*dimY + j * dimX + i ] = sArray[ (thrk+1) * sizeSArray * sizeSArray + (thrj+1) * sizeSArray + thri+1 ];
+
+ }
+}
#endif