From 0b3e87ab4c55865a3bedefcdd645d49f2974e362 Mon Sep 17 00:00:00 2001
From: fencl <fencl@geraldine.fjif.cvut.cz>
Date: Sat, 25 Nov 2017 12:48:46 +0100
Subject: [PATCH 01/28] zmena inicializace v tnlDirectEikonalMethodsBase_impl.h

---
 .../tnlDirectEikonalMethodsBase_impl.h        | 30 +++++++++++++++++--
 1 file changed, 28 insertions(+), 2 deletions(-)

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 b924f7233..88ed4ddf4 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
@@ -31,15 +31,38 @@ initInterface( const MeshFunctionType& input,
       if( ! cell.isBoundaryEntity()  )
       {
          const auto& neighbors = cell.getNeighborEntities();
+         const RealType& h = mesh.getSpaceSteps().x();
          //const IndexType& c = cell.getIndex();
          const IndexType e = neighbors.template getEntityIndex<  1 >();
          const IndexType w = neighbors.template getEntityIndex< -1 >();
 
-         if( c * input[ e ] <= 0 || c * input[ w ] <= 0 )
+         /*if( c * input[ e ] <= 0 || c * input[ w ] <= 0 )
          {
             output[ cell.getIndex() ] = c;
             interfaceMap[ cell.getIndex() ] = true;
             continue;
+         }*/
+         if( c * input[ e ] <=0 )
+         {
+             /*if( c >= 0 )
+             {
+                output[ cell.getIndex() ] = ( h * c )/( c - input[ e ] );
+             }
+             else
+             {
+                 output[ cell.getIndex() ] = - ( h * c )/( c - input[ e ] );
+             }*/
+             c >= 0 ? output[ cell.getIndex() ] = ( h * c )/( c - input[ e ] ) :
+                      output[ cell.getIndex() ] = - ( h * c )/( c - input[ e ] );
+             interfaceMap[ cell.getIndex() ] = true;
+             continue;
+         }
+         else if( c * input[ w ] <=0 )
+         {
+             c <= 0 ? output[ cell.getIndex() ] = ( h * c )/( c - input[ e ] ) :
+                      output[ cell.getIndex() ] = - ( h * c )/( c - input[ e ] );
+             interfaceMap[ cell.getIndex() ] = true;
+             continue;
          }
       }
       output[ cell.getIndex() ] =
@@ -85,6 +108,9 @@ initInterface( const MeshFunctionType& input,
          if( ! cell.isBoundaryEntity()  )
          {
             auto neighbors = cell.getNeighborEntities();
+            //const RealType& hx = mesh.getSpaceSteps().x();
+            //const RealType& hy = mesh.getSpaceSteps().y();
+            //const RealType& pom;
             const IndexType e = neighbors.template getEntityIndex<  1,  0 >();
             const IndexType w = neighbors.template getEntityIndex< -1,  0 >();
             const IndexType n = neighbors.template getEntityIndex<  0,  1 >();
@@ -92,7 +118,7 @@ initInterface( const MeshFunctionType& input,
             if( c * input[ e ] <= 0 || c * input[ w ] <= 0 ||
                 c * input[ n ] <= 0 || c * input[ s ] <= 0 )
             {
-               output[ cell.getIndex() ] = c;
+               output[ cell.getIndex() ] = tnlTypeInfo< RealType >::getMaxValue();
                interfaceMap[ cell.getIndex() ] = true;
                continue;
             }
-- 
GitLab


From 1c287e95fdad52f8ed165ffb935931c551d3791d Mon Sep 17 00:00:00 2001
From: Tomas Oberhuber <tomas.oberhuber@fjfi.cvut.cz>
Date: Mon, 27 Nov 2017 11:52:53 +0100
Subject: [PATCH 02/28] "Fixed" tnl-run-fsm-eoc-test.

---
 .../Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test              | 4 ++--
 .../hamilton-jacobi/tnlDirectEikonalMethodsBase_impl.h        | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test
index 6d997ce23..57a40528b 100755
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test
@@ -172,8 +172,8 @@ runTest()
                echo "========================================================================="
                echo "===                   STARTING THE SOLVER                             ==="
                echo "========================================================================="
-               solve explicit merson
-               #solve semi-implicit gmres
+               #solve explicit merson
+               solve semi-implicit cg
                mv computation-in-progress computation-done
             fi            
             echo "========================================================================="
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 904253116..414a2135c 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
@@ -117,7 +117,7 @@ initInterface( const MeshFunctionType& input,
             if( c * input[ e ] <= 0 || c * input[ w ] <= 0 ||
                 c * input[ n ] <= 0 || c * input[ s ] <= 0 )
             {
-               output[ cell.getIndex() ] = tnlTypeInfo< RealType >::getMaxValue();
+               output[ cell.getIndex() ] = TypeInfo< RealType >::getMaxValue();
                interfaceMap[ cell.getIndex() ] = true;
                continue;
             }
-- 
GitLab


From e7de9a4d04ce2cc59f7be35e0a34fd1048117555 Mon Sep 17 00:00:00 2001
From: fencl <fencl@geraldine.fjif.cvut.cz>
Date: Mon, 27 Nov 2017 18:55:03 +0100
Subject: [PATCH 03/28] finall change of inicialization for 2D

---
 .../tnlDirectEikonalMethodsBase_impl.h        | 82 ++++++++++++++++---
 1 file changed, 70 insertions(+), 12 deletions(-)

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 904253116..c76d988c3 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
@@ -51,21 +51,23 @@ initInterface( const MeshFunctionType& input,
              {
                  output[ cell.getIndex() ] = - ( h * c )/( c - input[ e ] );
              }*/
-             c >= 0 ? output[ cell.getIndex() ] = ( h * c )/( c - input[ e ] ) :
-                      output[ cell.getIndex() ] = - ( h * c )/( c - input[ e ] );
+             output[ cell.getIndex() ] =
+             c >= 0 ? ( h * c )/( c - input[ e ] ) :
+                      - ( h * c )/( c - input[ e ] );
              interfaceMap[ cell.getIndex() ] = true;
              continue;
          }
-         else if( c * input[ w ] <=0 )
+         if( c * input[ w ] <=0 )
          {
-             c <= 0 ? output[ cell.getIndex() ] = ( h * c )/( c - input[ e ] ) :
-                      output[ cell.getIndex() ] = - ( h * c )/( c - input[ e ] );
+             output[ cell.getIndex() ] =
+             c >= 0 ? ( h * c )/( c - input[ w ] ) :
+                      - ( h * c )/( c - input[ w ] );
              interfaceMap[ cell.getIndex() ] = true;
              continue;
          }
       }
       output[ cell.getIndex() ] =
-      c > 0 ? TypeInfo< RealType >::getMaxValue() :
+      c > 0 ? TNL::TypeInfo< RealType >::getMaxValue() :
              -TypeInfo< RealType >::getMaxValue();
       interfaceMap[ cell.getIndex() ] = false;
    }
@@ -95,6 +97,16 @@ initInterface( const MeshFunctionType& input,
    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() ++ )
+          output[ cell.getIndex() ] =
+              input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
+                                   - TypeInfo< RealType >::getMaxValue();
+   
    for( cell.getCoordinates().y() = 0;
         cell.getCoordinates().y() < mesh.getDimensions().y();
         cell.getCoordinates().y() ++ )
@@ -107,19 +119,65 @@ initInterface( const MeshFunctionType& input,
          if( ! cell.isBoundaryEntity()  )
          {
             auto neighbors = cell.getNeighborEntities();
-            //const RealType& hx = mesh.getSpaceSteps().x();
-            //const RealType& hy = mesh.getSpaceSteps().y();
-            //const RealType& pom;
+            const RealType& hx = mesh.getSpaceSteps().x();
+            const RealType& hy = mesh.getSpaceSteps().y();
+            Real pom = 0;
             const IndexType e = neighbors.template getEntityIndex<  1,  0 >();
-            const IndexType w = neighbors.template getEntityIndex< -1,  0 >();
+            //const IndexType w = neighbors.template getEntityIndex< -1,  0 >();
             const IndexType n = neighbors.template getEntityIndex<  0,  1 >();
-            const IndexType s = neighbors.template getEntityIndex<  0, -1 >();            
-            if( c * input[ e ] <= 0 || c * input[ w ] <= 0 ||
+            //const IndexType s = neighbors.template getEntityIndex<  0, -1 >();            
+            /*if( c * input[ e ] <= 0 || c * input[ w ] <= 0 ||
                 c * input[ n ] <= 0 || c * input[ s ] <= 0 )
             {
                output[ cell.getIndex() ] = tnlTypeInfo< RealType >::getMaxValue();
                interfaceMap[ cell.getIndex() ] = true;
                continue;
+            }*/
+            if( c * input[ n ] <= 0 )
+            {
+                if( c >= 0 )
+                {
+                    pom = ( hy * c )/( c - input[ n ]);
+                    if( output[ cell.getIndex() ] > pom ) 
+                        output[ cell.getIndex() ] = pom;
+                    
+                    output[ n ] = ( hy * c )/( c - input[ n ]) - hy;
+                }else
+                {
+                    pom = - ( hy * c )/( c - input[ n ]);
+                    if( output[ cell.getIndex() ] < pom )
+                        output[ cell.getIndex() ] = pom;
+                    
+                    output[ n ] = hy - ( hy * c )/( c - input[ n ]);
+                }
+                interfaceMap[ cell.getIndex() ] = true;
+                interfaceMap[ n ] = true;
+                continue;
+            }
+            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 ] != 0 && 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;
+                continue;
             }
          }
          output[ cell.getIndex() ] =
-- 
GitLab


From f575dd05e9e95af16c4ef3923352ec4c7e374a90 Mon Sep 17 00:00:00 2001
From: fencl <fencl@geraldine.fjif.cvut.cz>
Date: Mon, 4 Dec 2017 12:21:21 +0100
Subject: [PATCH 04/28] Implementation of 2D FSM updateCell

---
 .../tnlDirectEikonalMethodsBase_impl.h        | 65 +++++++++++++++++--
 1 file changed, 61 insertions(+), 4 deletions(-)

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 65f4d0ffe..00b61d2d5 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
@@ -187,7 +187,7 @@ initInterface( const MeshFunctionType& input,
       }
 }
 
-template< typename Real,
+/*template< typename Real,
           typename Device,
           typename Index >
    template< typename MeshEntity >
@@ -230,7 +230,7 @@ updateCell( MeshFunctionType& u,
        fabs( b ) == TypeInfo< Real >::getMaxValue() ||
        fabs( a - b ) >= h )
    {
-      tmp = argAbsMin( a, b ) + sign( value ) * h;
+      tmp = argAbsMin( a, b ) + sign( value ) * h;*/
       /*   std::cerr << "a = " << a << " b = " << b << " h = " << h 
              << " ArgAbsMin( a, b ) = " << ArgAbsMin( a, b ) << " sign( value ) = " << sign( value )
              << " sign( value ) * h = " << sign( value ) * h
@@ -243,14 +243,71 @@ updateCell( MeshFunctionType& u,
       std::cerr << " tmp = " << tmp << std::endl;
       std::cerr << " res = " << res << std::endl;*/
 
-   }
+   /*}
    else
       tmp = 0.5 * ( a + b + sign( value ) * sqrt( 2.0 * h * h - ( a - b ) * ( a - b ) ) );
 
    u[ cell.getIndex() ] = argAbsMin( value, tmp );
    //std::cerr << ArgAbsMin( value, tmp ) << " ";   
-}
+}*/
 
+template< typename Real,
+          typename Device,
+          typename Index >
+   template< typename MeshEntity >
+void
+tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
+updateCell( MeshFunctionType& u,
+            const MeshEntity& cell)
+            //MeshFunctionType& 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 );
+   //const RealType permeability = v( cell );
+   Real a, b, tmp;
+   
+   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 ) == TypeInfo< Real >::getMaxValue() && 
+       fabs( b ) == TypeInfo< Real >::getMaxValue() )
+      return;
+   if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
+       fabs( b ) == TypeInfo< Real >::getMaxValue() ||
+       fabs( a - b ) >= TNL::sqrt( (hx * hx + hy * hy)/1 ) ) //permeability ) )
+   {
+      tmp = 
+        fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * hy :
+                                 a + TNL::sign( value ) * hx;
+   }
+   else
+      tmp = ( hx * hx * a + hy * hy * b + 
+            sign( value ) * hx * hy * sqrt( ( hx * hx + hy * hy )/1 - 
+            ( a - b ) * ( a - b ) ) )/( hx * hx + hy * hy ); //permeability
+            
+   
+   u[ cell.getIndex() ] = argAbsMin( value, tmp );
+}
 
 template< typename Real,
           typename Device,
-- 
GitLab


From 1b4d886525908c612240d1b2f3ecb04d3481f533 Mon Sep 17 00:00:00 2001
From: fencl <fencl@geraldine.fjif.cvut.cz>
Date: Sun, 17 Dec 2017 17:46:30 +0100
Subject: [PATCH 05/28] remade initialization in 3D and made updateCell for 3D

---
 .../tnlDirectEikonalMethodsBase.h             |  19 +-
 .../tnlDirectEikonalMethodsBase_impl.h        | 388 ++++++++++++------
 2 files changed, 278 insertions(+), 129 deletions(-)

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 c6385f64a..1182ac574 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -37,7 +37,7 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >
       
       template< typename MeshEntity >
       void updateCell( MeshFunctionType& u,
-                       const MeshEntity& cell );
+                       const MeshEntity& cell);
       
 };
 
@@ -62,7 +62,8 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
       
       template< typename MeshEntity >
       void updateCell( MeshFunctionType& u,
-                       const MeshEntity& cell );
+                       const MeshEntity& cell,
+                       double velocity = 1.0 );
 };
 
 template< typename Real,
@@ -85,7 +86,19 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
       
       template< typename MeshEntity >
       void updateCell( MeshFunctionType& u,
-                       const MeshEntity& cell );      
+                       const MeshEntity& cell,
+                       double velocity = 1.0);
+      
+      Real sort( Real a, Real b, Real c,
+                 const RealType& ha,
+                 const RealType& hb,
+                 const RealType& hc ); 
 };
 
+template < typename T1, typename T2 >
+T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double v = 1);
+
+template < typename T1 >
+T1 sortMinims( T1 a, T1 b, T1 c, T1 ha, T1 hb, T1 hc);
+
 #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 00b61d2d5..7456a2fa1 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
@@ -34,23 +34,8 @@ initInterface( const MeshFunctionType& input,
          //const IndexType& c = cell.getIndex();
          const IndexType e = neighbors.template getEntityIndex<  1 >();
          const IndexType w = neighbors.template getEntityIndex< -1 >();
-
-         /*if( c * input[ e ] <= 0 || c * input[ w ] <= 0 )
-         {
-            output[ cell.getIndex() ] = c;
-            interfaceMap[ cell.getIndex() ] = true;
-            continue;
-         }*/
          if( c * input[ e ] <=0 )
          {
-             /*if( c >= 0 )
-             {
-                output[ cell.getIndex() ] = ( h * c )/( c - input[ e ] );
-             }
-             else
-             {
-                 output[ cell.getIndex() ] = - ( h * c )/( c - input[ e ] );
-             }*/
              output[ cell.getIndex() ] =
              c >= 0 ? ( h * c )/( c - input[ e ] ) :
                       - ( h * c )/( c - input[ e ] );
@@ -94,19 +79,21 @@ initInterface( const MeshFunctionType& input,
                MeshFunctionType& output,
                InterfaceMapType& interfaceMap  )
 {
-   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() ++ )
-          output[ cell.getIndex() ] =
-              input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
+    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() ++ )
+            output[ cell.getIndex() ] =
+                input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
                                    - TypeInfo< RealType >::getMaxValue();
    
+   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() ++ )
@@ -119,20 +106,9 @@ initInterface( const MeshFunctionType& input,
          if( ! cell.isBoundaryEntity()  )
          {
             auto neighbors = cell.getNeighborEntities();
-            const RealType& hx = mesh.getSpaceSteps().x();
-            const RealType& hy = mesh.getSpaceSteps().y();
             Real pom = 0;
             const IndexType e = neighbors.template getEntityIndex<  1,  0 >();
-            //const IndexType w = neighbors.template getEntityIndex< -1,  0 >();
             const IndexType n = neighbors.template getEntityIndex<  0,  1 >();
-            //const IndexType s = neighbors.template getEntityIndex<  0, -1 >();            
-            /*if( c * input[ e ] <= 0 || c * input[ w ] <= 0 ||
-                c * input[ n ] <= 0 || c * input[ s ] <= 0 )
-            {
-               output[ cell.getIndex() ] = TypeInfo< RealType >::getMaxValue();
-               interfaceMap[ cell.getIndex() ] = true;
-               continue;
-            }*/
             if( c * input[ n ] <= 0 )
             {
                 if( c >= 0 )
@@ -152,7 +128,6 @@ initInterface( const MeshFunctionType& input,
                 }
                 interfaceMap[ cell.getIndex() ] = true;
                 interfaceMap[ n ] = true;
-                continue;
             }
             if( c * input[ e ] <= 0 )
             {
@@ -163,7 +138,7 @@ initInterface( const MeshFunctionType& input,
                         output[ cell.getIndex() ] = pom;
                    
                     pom = pom - hx; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
-                    if( output[ e ] != 0 && output[ e ] >= pom )
+                    if( output[ e ] >= pom )
                         output[ e ] = pom;                         
                 }else
                 {
@@ -177,8 +152,8 @@ initInterface( const MeshFunctionType& input,
                 }
                 interfaceMap[ cell.getIndex() ] = true;
                 interfaceMap[ e ] = true;
-                continue;
             }
+            continue;
          }
          output[ cell.getIndex() ] =
             c > 0 ? TypeInfo< RealType >::getMaxValue() :
@@ -187,70 +162,6 @@ initInterface( const MeshFunctionType& input,
       }
 }
 
-/*template< typename Real,
-          typename Device,
-          typename Index >
-   template< typename MeshEntity >
-void
-tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
-updateCell( MeshFunctionType& u,
-            const MeshEntity& cell )
-{
-   const auto& neighborEntities = cell.template getNeighborEntities< 2 >();
-   const MeshType& mesh = cell.getMesh();
-  
-   const RealType& h = mesh.getSpaceSteps().x(); 
-   const RealType value = u( cell );
-   Real a, b, tmp;
-
-   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 = 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 = argAbsMin( u[ neighborEntities.template getEntityIndex< 0,  -1 >() ],
-                     u[ neighborEntities.template getEntityIndex< 0,   1 >() ] );
-   }
-
-   if( fabs( a ) == TypeInfo< Real >::getMaxValue() && 
-       fabs( b ) == TypeInfo< Real >::getMaxValue() )
-      return;
-   if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
-       fabs( b ) == TypeInfo< Real >::getMaxValue() ||
-       fabs( a - b ) >= h )
-   {
-      tmp = argAbsMin( a, b ) + sign( value ) * h;*/
-      /*   std::cerr << "a = " << a << " b = " << b << " h = " << h 
-             << " ArgAbsMin( a, b ) = " << ArgAbsMin( a, b ) << " sign( value ) = " << sign( value )
-             << " sign( value ) * h = " << sign( value ) * h
-             << " ArgAbsMin( a, b ) + sign( value ) * h = " << ArgAbsMin( a, b ) + sign( value ) * h           
-             << " tmp = " << tmp << std::endl;
-      tmp = ArgAbsMin( a, b ) + sign( value ) * h;
-      tmp = ArgAbsMin( a, b ) + sign( value ) * h;
-      tmp = ArgAbsMin( a, b ) + sign( value ) * h;
-      res = ArgAbsMin( a, b ) + sign( value ) * h;
-      std::cerr << " tmp = " << tmp << std::endl;
-      std::cerr << " res = " << res << std::endl;*/
-
-   /*}
-   else
-      tmp = 0.5 * ( a + b + sign( value ) * sqrt( 2.0 * h * h - ( a - b ) * ( a - b ) ) );
-
-   u[ cell.getIndex() ] = argAbsMin( value, tmp );
-   //std::cerr << ArgAbsMin( value, tmp ) << " ";   
-}*/
-
 template< typename Real,
           typename Device,
           typename Index >
@@ -258,8 +169,8 @@ template< typename Real,
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
 updateCell( MeshFunctionType& u,
-            const MeshEntity& cell)
-            //MeshFunctionType& v
+            const MeshEntity& cell,
+            double v)
 {
    const auto& neighborEntities = cell.template getNeighborEntities< 2 >();
    const MeshType& mesh = cell.getMesh();
@@ -267,7 +178,6 @@ updateCell( MeshFunctionType& u,
    const RealType& hx = mesh.getSpaceSteps().x();
    const RealType& hy = mesh.getSpaceSteps().y();
    const RealType value = u( cell );
-   //const RealType permeability = v( cell );
    Real a, b, tmp;
    
    if( cell.getCoordinates().x() == 0 )
@@ -292,18 +202,19 @@ updateCell( MeshFunctionType& u,
    if( fabs( a ) == TypeInfo< Real >::getMaxValue() && 
        fabs( b ) == TypeInfo< Real >::getMaxValue() )
       return;
-   if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
+   /*if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
        fabs( b ) == TypeInfo< Real >::getMaxValue() ||
-       fabs( a - b ) >= TNL::sqrt( (hx * hx + hy * hy)/1 ) ) //permeability ) )
+       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;
-   }
-   else
-      tmp = ( hx * hx * a + hy * hy * b + 
-            sign( value ) * hx * hy * sqrt( ( hx * hx + hy * hy )/1 - 
-            ( a - b ) * ( a - b ) ) )/( hx * hx + hy * hy ); //permeability
+   }*/
+   tmp = meet2DCondition( a, b, hx, hy, value);
+   if( tmp == 0 )
+      tmp = 
+          fabs( a ) >= fabs( b ) ? fabs( b ) + TNL::sign( value ) * hy :
+                    fabs( a ) + TNL::sign( value ) * hx;
             
    
    u[ cell.getIndex() ] = argAbsMin( value, tmp );
@@ -321,14 +232,30 @@ initInterface( const MeshFunctionType& input,
    const MeshType& mesh = input.getMesh();
    typedef typename MeshType::Cell Cell;
    Cell cell( mesh );
-   for( cell.getCoordinates().z() = 1;
-        cell.getCoordinates().z() < mesh.getDimensions().z() - 1;
+   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() ++ )
+                    output[ cell.getIndex() ] =
+                    input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
+                                       - TypeInfo< RealType >::getMaxValue();
+    
+   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() = 1;
-           cell.getCoordinates().y() < mesh.getDimensions().y() - 1;
+      for( cell.getCoordinates().y() = 0;
+           cell.getCoordinates().y() < mesh.getDimensions().y();
            cell.getCoordinates().y() ++ )
-         for( cell.getCoordinates().x() = 1;
-              cell.getCoordinates().x() < mesh.getDimensions().x() - 1;
+         for( cell.getCoordinates().x() = 0;
+              cell.getCoordinates().x() < mesh.getDimensions().x();
               cell.getCoordinates().x() ++ )
          {
             cell.refresh();
@@ -336,27 +263,98 @@ initInterface( const MeshFunctionType& input,
             if( ! cell.isBoundaryEntity() )
             {
                auto neighbors = cell.getNeighborEntities();
+               Real pom = 0;
                //const IndexType& c = cell.getIndex();
                const IndexType e = neighbors.template getEntityIndex<  1,  0,  0 >();
-               const IndexType w = neighbors.template getEntityIndex< -1,  0,  0 >();
+               //const IndexType w = neighbors.template getEntityIndex< -1,  0,  0 >();
                const IndexType n = neighbors.template getEntityIndex<  0,  1,  0 >();
-               const IndexType s = neighbors.template getEntityIndex<  0, -1,  0 >();
+               //const IndexType s = neighbors.template getEntityIndex<  0, -1,  0 >();
                const IndexType t = neighbors.template getEntityIndex<  0,  0,  1 >();
-               const IndexType b = neighbors.template getEntityIndex<  0,  0, -1 >();
-
-               if( c * input[ e ] <= 0 || c * input[ w ] <= 0 ||
+               //const IndexType b = neighbors.template getEntityIndex<  0,  0, -1 >();
+               /*if( c * input[ e ] <= 0 || c * input[ w ] <= 0 ||
                    c * input[ n ] <= 0 || c * input[ s ] <= 0 ||
                    c * input[ t ] <= 0 || c * input[ b ] <= 0 )
                {
                   output[ cell.getIndex() ] = c;
                   interfaceMap[ cell.getIndex() ] = true;
                   continue;
+               }*/
+               if( c * input[ n ] <= 0 )
+               {
+                   if( c >= 0 )
+                   {
+                   pom = ( hy * c )/( c - input[ n ]);
+                   if( output[ cell.getIndex() ] > pom ) 
+                   output[ cell.getIndex() ] = pom;
+                    
+                   output[ n ] = ( hy * c )/( c - input[ n ]) - hy;
+                   }else
+                   {
+                     pom = - ( hy * c )/( c - input[ n ]);
+                     if( output[ cell.getIndex() ] < pom )
+                         output[ cell.getIndex() ] = pom;
+                    
+                     output[ n ] = hy - ( hy * c )/( c - input[ n ]);
+                   }
+               interfaceMap[ cell.getIndex() ] = true;
+               interfaceMap[ n ] = true;
+               continue;
                }
+               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;
+               continue;
+               }
+               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;
+               continue;
             }
             output[ cell.getIndex() ] =
                c > 0 ? TypeInfo< RealType >::getMaxValue() :
                       -TypeInfo< RealType >::getMaxValue();
             interfaceMap[ cell.getIndex() ] = false;
+            }
          }
 }
 
@@ -367,7 +365,145 @@ template< typename Real,
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
 updateCell( MeshFunctionType& u,
-            const MeshEntity& cell )
+            const MeshEntity& cell, 
+            double 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& hz = mesh.getSpaceSteps().z();
+   const RealType value = u( cell );
+   Real a, b, c, tmp = 0;
    
+   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().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().x() == 0 )
+      b = u[ neighborEntities.template getEntityIndex< 0, 1, 0 >() ];
+   else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 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().x() == 0 )
+      c = u[ neighborEntities.template getEntityIndex< 0, 0, 1 >() ];
+   else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 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 ) == TypeInfo< Real >::getMaxValue() && 
+       fabs( b ) == TypeInfo< Real >::getMaxValue() &&
+       fabs( c ) == TypeInfo< Real >::getMaxValue() )
+      return;
+   if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
+       fabs( b ) == TypeInfo< Real >::getMaxValue() ||
+       fabs( c ) == TypeInfo< Real >::getMaxValue() ||
+       hz * hz * fabs( a - b ) * fabs( a - b ) + hy * hy * fabs( a - c ) * fabs( a - c ) +
+           hx * hx * fabs( b - c ) * fabs( b - c ) >=  ( hx * hx * hy * hy + hx * hx * hz * hz + hy * hy * hz * hz )/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 * 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 );
+       }*/
+       Real pom = sortMinims( a, b, c, (Real)hx, (Real)hy, (Real)hz);
+       tmp = meet2DCondition( pom[0], pom[1], pom[3], pom[4]);
+       
+       if( tmp == 0 )
+           tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]; 
+        
+   }
+   else
+      tmp = ( hx * hx * a + hy * hy * b + hz * hz * c +
+            sign( value ) * hx * hy * hz * sqrt( ( hx * hx + hy * hy + hz * hz)/v - 
+            hz * hz * ( a - b ) * ( a - b ) + hy * hy * ( a - c ) * ( a - c ) +
+            hx * hx * ( b - c ) * ( b - c ) ) )/( hx * hx + hy * hy + hz * hz );
+            
+   
+   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 ) == TypeInfo< T1 >::getMaxValue() &&
+       fabs( b ) == TypeInfo< T1 >::getMaxValue() &&
+       fabs( a - b ) <= TNL::sqrt( (ha * ha + hb * hb)/v ) )
+   {
+      tmp = ( ha * ha * a + hb * hb * b + 
+            sign( value ) * hb * hb * sqrt( ( ha * ha + hb * hb )/v - 
+            ( a - b ) * ( a - b ) ) )/( ha * ha + hb * hb );
+   }
+   else
+       tmp = 0;
+   
+   return tmp;
 }
+
+template < typename T1 >
+T1 sortMinims( T1 a, T1 b, T1 c, T1 ha, T1 hb, T1 hc)
+{
+    T1 tmp[6];
+    if( a <= b && b <= c){
+        tmp = {a,b,c,ha,hb,hc};
+    }
+    else if( a <= c && c <= b ){
+        tmp = {a,c,b,ha,hc,hb};
+    }
+    else if( b<=a && a<=c ){
+        tmp = {b,a,c,hb,ha,hc};
+    }
+    else if( b <= c && c <= a ){
+        tmp = {b,c,a,hb,hc,ha};
+    }
+    else if( c <= a && a <= b ){
+        tmp = {c,a,b,hc,ha,hb};
+    }
+    else if( c <=b && b <= a ){
+        tmp = {c,b,a,hc,hb,ha};
+    }
+        
+    return tmp;
+}
\ No newline at end of file
-- 
GitLab


From 0577cc48dbc00a774e46e88c0a3c1e93734703c2 Mon Sep 17 00:00:00 2001
From: fencl <fencl@geraldine.fjif.cvut.cz>
Date: Tue, 20 Feb 2018 18:47:13 +0100
Subject: [PATCH 06/28] removed some mistakes in methods 2D and 3D

---
 .../tnlDirectEikonalMethodsBase_impl.h        | 38 +++++++++----------
 1 file changed, 19 insertions(+), 19 deletions(-)

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 7456a2fa1..461c70f55 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
@@ -40,7 +40,6 @@ initInterface( const MeshFunctionType& input,
              c >= 0 ? ( h * c )/( c - input[ e ] ) :
                       - ( h * c )/( c - input[ e ] );
              interfaceMap[ cell.getIndex() ] = true;
-             continue;
          }
          if( c * input[ w ] <=0 )
          {
@@ -48,7 +47,6 @@ initInterface( const MeshFunctionType& input,
              c >= 0 ? ( h * c )/( c - input[ w ] ) :
                       - ( h * c )/( c - input[ w ] );
              interfaceMap[ cell.getIndex() ] = true;
-             continue;
          }
       }
       output[ cell.getIndex() ] =
@@ -88,9 +86,12 @@ initInterface( const MeshFunctionType& input,
         for( cell.getCoordinates().x() = 0;
              cell.getCoordinates().x() < mesh.getDimensions().x();
              cell.getCoordinates().x() ++ )
+        {
             output[ cell.getIndex() ] =
                 input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
                                    - TypeInfo< RealType >::getMaxValue();
+            interfaceMap[ cell.getIndex() ] = false;
+        }
    
    const RealType& hx = mesh.getSpaceSteps().x();
    const RealType& hy = mesh.getSpaceSteps().y();     
@@ -153,12 +154,11 @@ initInterface( const MeshFunctionType& input,
                 interfaceMap[ cell.getIndex() ] = true;
                 interfaceMap[ e ] = true;
             }
-            continue;
          }
-         output[ cell.getIndex() ] =
+         /*output[ cell.getIndex() ] =
             c > 0 ? TypeInfo< RealType >::getMaxValue() :
-                   -TypeInfo< RealType >::getMaxValue();  
-         interfaceMap[ cell.getIndex() ] = false;
+                   -TypeInfo< RealType >::getMaxValue();*/  
+         //interfaceMap[ cell.getIndex() ] = false; //is on line 90
       }
 }
 
@@ -210,7 +210,7 @@ updateCell( MeshFunctionType& u,
         fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * hy :
                                  a + TNL::sign( value ) * hx;
    }*/
-   tmp = meet2DCondition( a, b, hx, hy, value);
+   tmp = meet2DCondition( a, b, hx, hy, value, v);
    if( tmp == 0 )
       tmp = 
           fabs( a ) >= fabs( b ) ? fabs( b ) + TNL::sign( value ) * hy :
@@ -241,9 +241,12 @@ initInterface( const MeshFunctionType& input,
             for( cell.getCoordinates().x() = 0;
                  cell.getCoordinates().x() < mesh.getDimensions().x();
                  cell.getCoordinates().x() ++ )
-                    output[ cell.getIndex() ] =
-                    input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
-                                       - TypeInfo< RealType >::getMaxValue();
+            {
+                output[ cell.getIndex() ] =
+                input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
+                                   - TypeInfo< RealType >::getMaxValue();
+                interfaceMap[ cell.getIndex() ] = false;
+            }
     
    const RealType& hx = mesh.getSpaceSteps().x();
    const RealType& hy = mesh.getSpaceSteps().y();
@@ -298,7 +301,6 @@ initInterface( const MeshFunctionType& input,
                    }
                interfaceMap[ cell.getIndex() ] = true;
                interfaceMap[ n ] = true;
-               continue;
                }
                if( c * input[ e ] <= 0 )
                {
@@ -323,7 +325,6 @@ initInterface( const MeshFunctionType& input,
                    }
                interfaceMap[ cell.getIndex() ] = true;
                interfaceMap[ e ] = true;
-               continue;
                }
                if( c * input[ t ] <= 0 )
                {
@@ -348,13 +349,12 @@ initInterface( const MeshFunctionType& input,
                    }
                interfaceMap[ cell.getIndex() ] = true;
                interfaceMap[ t ] = true;
-               continue;
+               }           
             }
-            output[ cell.getIndex() ] =
+            /*output[ cell.getIndex() ] =
                c > 0 ? TypeInfo< RealType >::getMaxValue() :
                       -TypeInfo< RealType >::getMaxValue();
-            interfaceMap[ cell.getIndex() ] = false;
-            }
+            interfaceMap[ cell.getIndex() ] = false;*/ //is on line 245
          }
 }
 
@@ -448,7 +448,7 @@ updateCell( MeshFunctionType& u,
                 ( b - c ) * ( b - c ) ) )/( hy * hy + hz * hz );
        }*/
        Real pom = sortMinims( a, b, c, (Real)hx, (Real)hy, (Real)hz);
-       tmp = meet2DCondition( pom[0], pom[1], pom[3], pom[4]);
+       tmp = meet2DCondition( pom[0], pom[1], pom[3], pom[4], value, v);
        
        if( tmp == 0 )
            tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]; 
@@ -468,8 +468,8 @@ 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 ) == TypeInfo< T1 >::getMaxValue() &&
-       fabs( b ) == TypeInfo< T1 >::getMaxValue() &&
+   if( fabs( a ) != TypeInfo< T1 >::getMaxValue() &&
+       fabs( b ) != TypeInfo< T1 >::getMaxValue() &&
        fabs( a - b ) <= TNL::sqrt( (ha * ha + hb * hb)/v ) )
    {
       tmp = ( ha * ha * a + hb * hb * b + 
-- 
GitLab


From 53db9650839998c21f68f7f9631d414803ceb29b Mon Sep 17 00:00:00 2001
From: fencl <fencl@geraldine.fjif.cvut.cz>
Date: Tue, 27 Feb 2018 17:10:22 +0100
Subject: [PATCH 07/28] tnl-run-fsm-eoc-test: Change of the final file for
 computeError tnlDirectEikonalMethodsBase_impl.h: 2D method repaired, 3D
 method ready for debug tnlDirectEikonalMethod2D_impl.h: Added save of aux-*
 as results from computation tnlDirectEikonalMethod3D_impl.h: Implemented all
 sweaps and saving results from computation

---
 .../hamilton-jacobi/tnl-run-fsm-eoc-test      |  11 +-
 .../tnlDirectEikonalMethodsBase.h             |   2 +-
 .../tnlDirectEikonalMethodsBase_impl.h        | 143 ++++++++------
 .../tnlFastSweepingMethod2D_impl.h            |  11 +-
 .../tnlFastSweepingMethod3D_impl.h            | 175 ++++++++++++++++++
 5 files changed, 274 insertions(+), 68 deletions(-)

diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test
index 57a40528b..0dc50246f 100755
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-run-fsm-eoc-test
@@ -2,7 +2,7 @@
 
 device="host"
 dimensions="2D 3D"
-dimensions="2D"
+#dimensions="3D"
 sizes1D="16 32 64 128 256 512 1024 2048 4096"
 #sizes1D="256"
 sizes2D="16 32 64 128 256 512 1024"
@@ -12,7 +12,7 @@ testFunctions="paraboloid"
 snapshotPeriod=0.1
 finalTime=1.5
 solverName="tnl-direct-eikonal-solver"
-#solverName="gdb --args tnl-direct-eikonal-solver-dbg"
+#solverName="gdb --args tnl-direct-eikonal-solver-dbg --catch-exceptions no"
 #
 
 setupTestFunction()
@@ -59,7 +59,7 @@ setupGrid()
 setInitialCondition()
 {
    testFunction=$1
-   tnl-init --test-function ${testFunction}-sdf \
+   tnl-init --test-function ${testFunction} \
             --output-file initial-u.tnl \
             --amplitude ${amplitude} \
             --wave-length ${waveLength} \
@@ -78,7 +78,7 @@ setInitialCondition()
             --radius ${radius}
    
     tnl-init --test-function ${testFunction}-sdf \
-            --output-file final-u.tnl \
+            --output-file exact-u.tnl \
             --amplitude ${amplitude} \
             --wave-length ${waveLength} \
             --wave-length-x ${waveLengthX} \
@@ -108,7 +108,6 @@ solve()
                  --time-step 10 \
                  --time-step-order 2 \
                  --discrete-solver ${discreteSolver} \
-                 --merson-adaptivity 1.0e-5 \
                  --min-iterations 20 \
                  --convergence-residue 1.0e-12 \
                  --snapshot-period ${snapshotPeriod} \
@@ -118,7 +117,7 @@ solve()
 computeError()
 {
    tnl-diff --mesh mesh.tnl \
-            --input-files final-u.tnl u-*.tnl \
+            --input-files aux-final.tnl exact-u.tnl \
             --mode sequence \
             --snapshot-period ${snapshotPeriod} \
             --output-file errors.txt \
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 1182ac574..a4a46396c 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -99,6 +99,6 @@ template < typename T1, typename T2 >
 T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double v = 1);
 
 template < typename T1 >
-T1 sortMinims( T1 a, T1 b, T1 c, T1 ha, T1 hb, T1 hc);
+void sortMinims( T1 pom[] );
 
 #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 461c70f55..4cb8327a1 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
@@ -86,12 +86,13 @@ initInterface( const MeshFunctionType& input,
         for( cell.getCoordinates().x() = 0;
              cell.getCoordinates().x() < mesh.getDimensions().x();
              cell.getCoordinates().x() ++ )
-        {
-            output[ cell.getIndex() ] =
+            {
+                cell.refresh();
+                output[ cell.getIndex() ] =
                 input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
                                    - TypeInfo< RealType >::getMaxValue();
-            interfaceMap[ cell.getIndex() ] = false;
-        }
+                interfaceMap[ cell.getIndex() ] = false;
+            }
    
    const RealType& hx = mesh.getSpaceSteps().x();
    const RealType& hy = mesh.getSpaceSteps().y();     
@@ -117,15 +118,15 @@ initInterface( const MeshFunctionType& input,
                     pom = ( hy * c )/( c - input[ n ]);
                     if( output[ cell.getIndex() ] > pom ) 
                         output[ cell.getIndex() ] = pom;
-                    
-                    output[ n ] = ( hy * c )/( c - input[ n ]) - hy;
+                    if( output[ n ] < pom-hx )
+                        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;
-                    
-                    output[ n ] = hy - ( hy * c )/( c - input[ n ]);
+                    if( output[ n ] > hy + pom )
+                        output[ n ] = hy + pom; //hy - ( hy * c )/( c - input[ n ]);
                 }
                 interfaceMap[ cell.getIndex() ] = true;
                 interfaceMap[ n ] = true;
@@ -139,7 +140,7 @@ initInterface( const MeshFunctionType& input,
                         output[ cell.getIndex() ] = pom;
                    
                     pom = pom - hx; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
-                    if( output[ e ] >= pom )
+                    if( output[ e ] < pom )
                         output[ e ] = pom;                         
                 }else
                 {
@@ -148,7 +149,7 @@ initInterface( const MeshFunctionType& input,
                         output[ cell.getIndex() ] = pom;
                     
                     pom = pom + hx; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
-                    if( output[ e ] < pom )
+                    if( output[ e ] > pom )
                         output[ e ] = pom;
                 }
                 interfaceMap[ cell.getIndex() ] = true;
@@ -157,7 +158,7 @@ initInterface( const MeshFunctionType& input,
          }
          /*output[ cell.getIndex() ] =
             c > 0 ? TypeInfo< RealType >::getMaxValue() :
-                   -TypeInfo< RealType >::getMaxValue();*/  
+                   -TypeInfo< RealType >::getMaxValue(); */
          //interfaceMap[ cell.getIndex() ] = false; //is on line 90
       }
 }
@@ -178,7 +179,7 @@ updateCell( MeshFunctionType& u,
    const RealType& hx = mesh.getSpaceSteps().x();
    const RealType& hy = mesh.getSpaceSteps().y();
    const RealType value = u( cell );
-   Real a, b, tmp;
+   Real a, b, tmp = TypeInfo< Real >::getMaxValue();
    
    if( cell.getCoordinates().x() == 0 )
       a = u[ neighborEntities.template getEntityIndex< 1,  0 >() ];
@@ -211,10 +212,10 @@ updateCell( MeshFunctionType& u,
                                  a + TNL::sign( value ) * hx;
    }*/
    tmp = meet2DCondition( a, b, hx, hy, value, v);
-   if( tmp == 0 )
+   if( tmp == TypeInfo< Real >::getMaxValue() )
       tmp = 
-          fabs( a ) >= fabs( b ) ? fabs( b ) + TNL::sign( value ) * hy :
-                    fabs( a ) + TNL::sign( value ) * hx;
+          fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * hy :
+                                   a + TNL::sign( value ) * hx;
             
    
    u[ cell.getIndex() ] = argAbsMin( value, tmp );
@@ -242,6 +243,7 @@ initInterface( const MeshFunctionType& input,
                  cell.getCoordinates().x() < mesh.getDimensions().x();
                  cell.getCoordinates().x() ++ )
             {
+                cell.refresh();
                 output[ cell.getIndex() ] =
                 input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
                                    - TypeInfo< RealType >::getMaxValue();
@@ -288,16 +290,19 @@ initInterface( const MeshFunctionType& input,
                    {
                    pom = ( hy * c )/( c - input[ n ]);
                    if( output[ cell.getIndex() ] > pom ) 
-                   output[ cell.getIndex() ] = pom;
+                       output[ cell.getIndex() ] = pom;
                     
-                   output[ n ] = ( hy * c )/( c - input[ n ]) - hy;
+                   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;
-                    
-                     output[ n ] = hy - ( hy * c )/( c - input[ n ]);
+                     if( output[ n ] > hy + pom )
+                         output[ n ] = hy + pom; //hy - ( hy * c )/( c - input[ n ]);
+                   
                    }
                interfaceMap[ cell.getIndex() ] = true;
                interfaceMap[ n ] = true;
@@ -308,11 +313,12 @@ initInterface( const MeshFunctionType& input,
                    {
                        pom = ( hx * c )/( c - input[ e ]);
                        if( output[ cell.getIndex() ] > pom )
-                       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;                         
+                       if( output[ e ] < pom )
+                           output[ e ] = pom;      
+                       
                    }else
                    {
                        pom = - (hx * c)/( c - input[ e ]);
@@ -320,7 +326,7 @@ initInterface( const MeshFunctionType& input,
                            output[ cell.getIndex() ] = pom;
                     
                        pom = pom + hx; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
-                       if( output[ e ] < pom )
+                       if( output[ e ] > pom )
                            output[ e ] = pom;
                    }
                interfaceMap[ cell.getIndex() ] = true;
@@ -335,8 +341,9 @@ initInterface( const MeshFunctionType& input,
                            output[ cell.getIndex() ] = pom;
                    
                        pom = pom - hz; //output[ e ] = (hx * c)/( c - input[ e ]) - hx;
-                       if( output[ t ] >= pom )
-                           output[ t ] = pom;                         
+                       if( output[ t ] < pom )
+                           output[ t ] = pom; 
+                       
                    }else
                    {
                        pom = - (hz * c)/( c - input[ t ]);
@@ -344,8 +351,9 @@ initInterface( const MeshFunctionType& input,
                            output[ cell.getIndex() ] = pom;
                     
                        pom = pom + hz; //output[ e ] = hx - (hx * c)/( c - input[ e ]);
-                       if( output[ t ] < pom )
+                       if( output[ t ] > pom )
                            output[ t ] = pom;
+                       
                    }
                interfaceMap[ cell.getIndex() ] = true;
                interfaceMap[ t ] = true;
@@ -368,24 +376,15 @@ updateCell( MeshFunctionType& u,
             const MeshEntity& cell, 
             double v )
 {
-   const auto& neighborEntities = cell.template getNeighborEntities< 2 >();
+   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 );
-   Real a, b, c, tmp = 0;
+   Real a, b, c, tmp = TypeInfo< Real >::getMaxValue();
    
-   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().x() == 0 )
       a = u[ neighborEntities.template getEntityIndex< 1, 0, 0 >() ];
@@ -396,17 +395,17 @@ updateCell( MeshFunctionType& u,
       a = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< -1, 0, 0 >() ],
                         u[ neighborEntities.template getEntityIndex< 1, 0, 0 >() ] );
    }
-   if( cell.getCoordinates().x() == 0 )
+   if( cell.getCoordinates().y() == 0 )
       b = u[ neighborEntities.template getEntityIndex< 0, 1, 0 >() ];
-   else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 1 )
+   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().x() == 0 )
+   }if( cell.getCoordinates().z() == 0 )
       c = u[ neighborEntities.template getEntityIndex< 0, 0, 1 >() ];
-   else if( cell.getCoordinates().x() == mesh.getDimensions().x() - 1 )
+   else if( cell.getCoordinates().z() == mesh.getDimensions().z() - 1 )
       c = u[ neighborEntities.template getEntityIndex< 0, 0, -1 >() ];
    else
    {
@@ -447,11 +446,12 @@ updateCell( MeshFunctionType& u,
                 sign( value ) * hy * hz * sqrt( ( hy * hy + hz * hz )/v - 
                 ( b - c ) * ( b - c ) ) )/( hy * hy + hz * hz );
        }*/
-       Real pom = sortMinims( a, b, c, (Real)hx, (Real)hy, (Real)hz);
-       tmp = meet2DCondition( pom[0], pom[1], pom[3], pom[4], value, v);
+       Real pom[6] = { a, b, c, (Real)hx, (Real)hy, (Real)hz};
+       sortMinims( pom );
+       tmp = meet2DCondition( pom[0], pom[1], pom[2], pom[3], value, v);
        
-       if( tmp == 0 )
-           tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]; 
+       if( tmp == TypeInfo< Real >::getMaxValue() )
+           tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 2 ]; 
         
    }
    else
@@ -477,33 +477,64 @@ T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double
             ( a - b ) * ( a - b ) ) )/( ha * ha + hb * hb );
    }
    else
-       tmp = 0;
+   {
+       tmp = TypeInfo< T1 >::getMaxValue();
+   }
    
    return tmp;
 }
 
 template < typename T1 >
-T1 sortMinims( T1 a, T1 b, T1 c, T1 ha, T1 hb, T1 hc)
+void sortMinims( T1 pom[])
+{
+    T1 tmp[6]; 
+    if( pom[0] <= pom[1] && pom[1] <= pom[2]){
+        tmp[0] = pom[0]; tmp[1] = pom[1]; tmp[2] = pom[3]; tmp[3] = pom[4];
+    }
+    else if( pom[0] <= pom[2] && pom[2] <= pom[1] ){
+        tmp[0] = pom[0]; tmp[1] = pom[2]; tmp[2] = pom[3]; tmp[3] = pom[5];
+    }
+    else if( pom[1]<=pom[0] && pom[0] <= pom[2] ){
+        tmp[0] = pom[1]; tmp[1] = pom[0]; tmp[2] = pom[4]; tmp[3] = pom[3];
+    }
+    else if( pom[1] <= pom[2] && pom[2] <= pom[0] ){
+        tmp[0] = pom[1]; tmp[1] = pom[2]; tmp[2] = pom[4]; tmp[3] = pom[5];
+    }
+    else if( pom[2] <= pom[0] && pom[0] <= pom[1] ){
+        tmp[0] = pom[2]; tmp[1] = pom[0]; tmp[2] = pom[5]; tmp[3] = pom[3];
+    }
+    else if( pom[2] <= pom[1] && pom[1] <= pom[0] ){
+        tmp[0] = pom[2]; tmp[1] = pom[1]; tmp[2] = pom[5]; tmp[3] = pom[4];
+    }
+    for( int i = 0; i < 6; i++ )
+        pom[ i ] = tmp[ i ];
+        
+}
+
+/*template < typename T1 >
+T1* sortMinims( T1 pom[], T1 a, T1 b, T1 c, T1 ha, T1 hb, T1 hc)
 {
-    T1 tmp[6];
+    T1 tmp[6] = pom; 
     if( a <= b && b <= c){
-        tmp = {a,b,c,ha,hb,hc};
+        tmp[0] = a; tmp[1] = b; tmp[2] = ha; tmp[3] = hb;
     }
     else if( a <= c && c <= b ){
-        tmp = {a,c,b,ha,hc,hb};
+        tmp[0] = a; tmp[1] = c; tmp[2] = ha; tmp[3] = hc;
     }
     else if( b<=a && a<=c ){
-        tmp = {b,a,c,hb,ha,hc};
+        tmp[0] = b; tmp[1] = a; tmp[2] = hb; tmp[3] = ha;
     }
     else if( b <= c && c <= a ){
-        tmp = {b,c,a,hb,hc,ha};
+        tmp[0] = b; tmp[1] = c; tmp[2] = hb; tmp[3] = hc;
     }
     else if( c <= a && a <= b ){
-        tmp = {c,a,b,hc,ha,hb};
+        tmp[0] = c; tmp[1] = a; tmp[2] = hc; tmp[3] = ha;
     }
     else if( c <=b && b <= a ){
-        tmp = {c,b,a,hc,hb,ha};
+        tmp[0] = c; tmp[1] = b; tmp[2] = hc; tmp[3] = hb;
     }
+    for( int i = 0; i < 6; i++ )
+        pom[ i ] = tmp[ i ];
         
-    return tmp;
-}
\ No newline at end of file
+    return pom;
+}*/
\ No newline at end of file
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 13e5c8249..bb1d4a018 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
@@ -64,7 +64,7 @@ solve( const MeshPointer& mesh,
    interfaceMap.setMesh( mesh );
    std::cout << "Initiating the interface cells ..." << std::endl;
    BaseType::initInterface( u, aux, interfaceMap );
-   //aux.save( "aux-ini.tnl" );
+   aux.save( "aux-ini.tnl" );
 
    typename MeshType::Cell cell( *mesh );
    
@@ -85,7 +85,7 @@ solve( const MeshPointer& mesh,
                   this->updateCell( aux, cell );
             }
       }
-      //aux.save( "aux-1.tnl" );
+      aux.save( "aux-1.tnl" );
 
       for( cell.getCoordinates().y() = 0;
            cell.getCoordinates().y() < mesh->getDimensions().y();
@@ -101,7 +101,7 @@ solve( const MeshPointer& mesh,
                   this->updateCell( aux, cell );
             }
       }
-      //aux.save( "aux-2.tnl" );
+      aux.save( "aux-2.tnl" );
 
       for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
            cell.getCoordinates().y() >= 0 ;
@@ -117,7 +117,7 @@ solve( const MeshPointer& mesh,
                   this->updateCell( aux, cell );
             }
          }
-      //aux.save( "aux-3.tnl" );
+      aux.save( "aux-3.tnl" );
 
 
       for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
@@ -134,8 +134,9 @@ solve( const MeshPointer& mesh,
                   this->updateCell( aux, cell );
             }
          }   
-      //aux.save( "aux-4.tnl" );
+      aux.save( "aux-4.tnl" );
       iteration++;
    }
+   aux.save("aux-final.tnl");
 }
 
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 6fad87c7e..2c281617f 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
@@ -65,5 +65,180 @@ solve( const MeshPointer& mesh,
    std::cout << "Initiating the interface cells ..." << std::endl;
    BaseType::initInterface( u, aux, interfaceMap );
    aux.save( "aux-ini.tnl" );   
+   
+   typename MeshType::Cell cell( *mesh );
+   
+   IndexType iteration( 0 );
+   while( iteration < this->maxIterations )
+   {
+      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 );
+            }
+         }
+      }
+      aux.save( "aux-8.tnl" );
+      iteration++;
+      
+   }
+   aux.save("aux-final.tnl");
 }
 
-- 
GitLab


From 9b293e449f94ce30a4c48325ca5ff9849473d346 Mon Sep 17 00:00:00 2001
From: fencl <fencl@geraldine.fjif.cvut.cz>
Date: Sun, 4 Mar 2018 08:31:43 +0100
Subject: [PATCH 08/28] final changes in 3D updateCell

---
 .../tnlDirectEikonalMethodsBase_impl.h        | 56 +++++--------------
 1 file changed, 14 insertions(+), 42 deletions(-)

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 4cb8327a1..42a1eed11 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
  *
@@ -118,7 +118,7 @@ initInterface( const MeshFunctionType& input,
                     pom = ( hy * c )/( c - input[ n ]);
                     if( output[ cell.getIndex() ] > pom ) 
                         output[ cell.getIndex() ] = pom;
-                    if( output[ n ] < pom-hx )
+                    if( output[ n ] < pom - hy )
                         output[ n ] = pom - hy; //( hy * c )/( c - input[ n ]) - hy;
                 }else
                 {
@@ -214,8 +214,8 @@ updateCell( MeshFunctionType& u,
    tmp = meet2DCondition( a, b, hx, hy, value, v);
    if( tmp == TypeInfo< Real >::getMaxValue() )
       tmp = 
-          fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * hy :
-                                   a + TNL::sign( value ) * hx;
+          fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * hy/v :
+                                   a + TNL::sign( value ) * hx/v;
             
    
    u[ cell.getIndex() ] = argAbsMin( value, tmp );
@@ -357,7 +357,7 @@ initInterface( const MeshFunctionType& input,
                    }
                interfaceMap[ cell.getIndex() ] = true;
                interfaceMap[ t ] = true;
-               }           
+               }        
             }
             /*output[ cell.getIndex() ] =
                c > 0 ? TypeInfo< RealType >::getMaxValue() :
@@ -383,7 +383,7 @@ updateCell( MeshFunctionType& u,
    const RealType& hy = mesh.getSpaceSteps().y();
    const RealType& hz = mesh.getSpaceSteps().z();
    const RealType value = u( cell );
-   Real a, b, c, tmp = TypeInfo< Real >::getMaxValue();
+   Real a, b, c, tmp = TypeInfo< RealType >::getMaxValue();
    
    
    if( cell.getCoordinates().x() == 0 )
@@ -451,7 +451,7 @@ updateCell( MeshFunctionType& u,
        tmp = meet2DCondition( pom[0], pom[1], pom[2], pom[3], value, v);
        
        if( tmp == TypeInfo< Real >::getMaxValue() )
-           tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 2 ]; 
+           tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 2 ]/v; 
         
    }
    else
@@ -488,53 +488,25 @@ template < typename T1 >
 void sortMinims( T1 pom[])
 {
     T1 tmp[6]; 
-    if( pom[0] <= pom[1] && pom[1] <= pom[2]){
+    if( fabs(pom[0]) <= fabs(pom[1]) && fabs(pom[1]) <= fabs(pom[2])){
         tmp[0] = pom[0]; tmp[1] = pom[1]; tmp[2] = pom[3]; tmp[3] = pom[4];
     }
-    else if( pom[0] <= pom[2] && pom[2] <= pom[1] ){
+    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[3]; tmp[3] = pom[5];
     }
-    else if( pom[1]<=pom[0] && pom[0] <= pom[2] ){
+    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[4]; tmp[3] = pom[3];
     }
-    else if( pom[1] <= pom[2] && pom[2] <= pom[0] ){
+    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[4]; tmp[3] = pom[5];
     }
-    else if( pom[2] <= pom[0] && pom[0] <= pom[1] ){
+    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[5]; tmp[3] = pom[3];
     }
-    else if( pom[2] <= pom[1] && pom[1] <= pom[0] ){
+    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[5]; tmp[3] = pom[4];
     }
     for( int i = 0; i < 6; i++ )
         pom[ i ] = tmp[ i ];
         
-}
-
-/*template < typename T1 >
-T1* sortMinims( T1 pom[], T1 a, T1 b, T1 c, T1 ha, T1 hb, T1 hc)
-{
-    T1 tmp[6] = pom; 
-    if( a <= b && b <= c){
-        tmp[0] = a; tmp[1] = b; tmp[2] = ha; tmp[3] = hb;
-    }
-    else if( a <= c && c <= b ){
-        tmp[0] = a; tmp[1] = c; tmp[2] = ha; tmp[3] = hc;
-    }
-    else if( b<=a && a<=c ){
-        tmp[0] = b; tmp[1] = a; tmp[2] = hb; tmp[3] = ha;
-    }
-    else if( b <= c && c <= a ){
-        tmp[0] = b; tmp[1] = c; tmp[2] = hb; tmp[3] = hc;
-    }
-    else if( c <= a && a <= b ){
-        tmp[0] = c; tmp[1] = a; tmp[2] = hc; tmp[3] = ha;
-    }
-    else if( c <=b && b <= a ){
-        tmp[0] = c; tmp[1] = b; tmp[2] = hc; tmp[3] = hb;
-    }
-    for( int i = 0; i < 6; i++ )
-        pom[ i ] = tmp[ i ];
-        
-    return pom;
-}*/
\ No newline at end of file
+}
\ No newline at end of file
-- 
GitLab


From 7d7e266ec1467f761c1f3a6fd11a3e7b85f9b22b Mon Sep 17 00:00:00 2001
From: fencl <fencl@geraldine.fjif.cvut.cz>
Date: Wed, 28 Mar 2018 16:06:09 +0200
Subject: [PATCH 09/28] final changes in algorithms of FastSweepingMethod
 (Zhao)

---
 .../tnlDirectEikonalMethodsBase.h             |  10 +-
 .../tnlDirectEikonalMethodsBase_impl.h        | 198 ++++++++++++------
 .../tnlFastSweepingMethod2D_impl.h            |  82 +++++++-
 .../tnlFastSweepingMethod3D_impl.h            |  16 +-
 4 files changed, 218 insertions(+), 88 deletions(-)

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 a4a46396c..4dc8e9228 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -37,7 +37,7 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >
       
       template< typename MeshEntity >
       void updateCell( MeshFunctionType& u,
-                       const MeshEntity& cell);
+                       const MeshEntity& cell );
       
 };
 
@@ -63,7 +63,7 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
       template< typename MeshEntity >
       void updateCell( MeshFunctionType& u,
                        const MeshEntity& cell,
-                       double velocity = 1.0 );
+                       const RealType velocity = 1.0 );
 };
 
 template< typename Real,
@@ -87,12 +87,12 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
       template< typename MeshEntity >
       void updateCell( MeshFunctionType& u,
                        const MeshEntity& cell,
-                       double velocity = 1.0);
+                       const RealType velocity = 1.0);
       
-      Real sort( Real a, Real b, Real c,
+      /*Real sort( Real a, Real b, Real c,
                  const RealType& ha,
                  const RealType& hb,
-                 const RealType& hc ); 
+                 const RealType& hc ); */
 };
 
 template < typename T1, typename T2 >
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 42a1eed11..caa2f533f 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
@@ -8,7 +8,7 @@
 #pragma once
 
 #include <TNL/TypeInfo.h>
-
+#include <iostream>
 template< typename Real,
           typename Device,
           typename Index >
@@ -111,7 +111,22 @@ initInterface( const MeshFunctionType& input,
             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 )
                 {
@@ -154,12 +169,8 @@ initInterface( const MeshFunctionType& input,
                 }
                 interfaceMap[ cell.getIndex() ] = true;
                 interfaceMap[ e ] = true;
-            }
+            }*/
          }
-         /*output[ cell.getIndex() ] =
-            c > 0 ? TypeInfo< RealType >::getMaxValue() :
-                   -TypeInfo< RealType >::getMaxValue(); */
-         //interfaceMap[ cell.getIndex() ] = false; //is on line 90
       }
 }
 
@@ -170,8 +181,8 @@ template< typename Real,
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
 updateCell( MeshFunctionType& u,
-            const MeshEntity& cell,
-            double v)
+            const MeshEntity& cell,   
+            const RealType v)
 {
    const auto& neighborEntities = cell.template getNeighborEntities< 2 >();
    const MeshType& mesh = cell.getMesh();
@@ -179,7 +190,7 @@ updateCell( MeshFunctionType& u,
    const RealType& hx = mesh.getSpaceSteps().x();
    const RealType& hy = mesh.getSpaceSteps().y();
    const RealType value = u( cell );
-   Real a, b, tmp = TypeInfo< Real >::getMaxValue();
+   RealType a, b, tmp1, tmp = TypeInfo< RealType >::getMaxValue();
    
    if( cell.getCoordinates().x() == 0 )
       a = u[ neighborEntities.template getEntityIndex< 1,  0 >() ];
@@ -188,7 +199,7 @@ updateCell( MeshFunctionType& u,
    else
    {
       a = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< -1,  0 >() ],
-                     u[ neighborEntities.template getEntityIndex<  1,  0 >() ] );
+                          u[ neighborEntities.template getEntityIndex<  1,  0 >() ] );
    }
 
    if( cell.getCoordinates().y() == 0 )
@@ -198,10 +209,10 @@ updateCell( MeshFunctionType& u,
    else
    {
       b = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< 0,  -1 >() ],
-                     u[ neighborEntities.template getEntityIndex< 0,   1 >() ] );
+                          u[ neighborEntities.template getEntityIndex< 0,   1 >() ] );
    }
-   if( fabs( a ) == TypeInfo< Real >::getMaxValue() && 
-       fabs( b ) == TypeInfo< Real >::getMaxValue() )
+   if( fabs( a ) == TypeInfo< RealType >::getMaxValue() && 
+       fabs( b ) == TypeInfo< RealType >::getMaxValue() )
       return;
    /*if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
        fabs( b ) == TypeInfo< Real >::getMaxValue() ||
@@ -211,14 +222,37 @@ updateCell( MeshFunctionType& u,
         fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * hy :
                                  a + TNL::sign( value ) * hx;
    }*/
-   tmp = meet2DCondition( a, b, hx, hy, value, v);
-   if( tmp == TypeInfo< Real >::getMaxValue() )
-      tmp = 
-          fabs( a ) >= fabs( b ) ? b + TNL::sign( value ) * 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;
+   }
+   else
+   {
+       tmp = 
+          fabs( a ) > fabs( b ) ? b + TNL::sign( value ) * hy/v :
                                    a + TNL::sign( value ) * hx/v;
-            
-   
-   u[ cell.getIndex() ] = argAbsMin( value, tmp );
+       u[ cell.getIndex() ] = argAbsMin( value, tmp );
+       //tmp = TypeInfo< RealType >::getMaxValue();
+   }*/
+    RealType pom[6] = { a, b, TypeInfo< Real >::getMaxValue(), (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 );
+    }
 }
 
 template< typename Real,
@@ -245,7 +279,7 @@ initInterface( const MeshFunctionType& input,
             {
                 cell.refresh();
                 output[ cell.getIndex() ] =
-                input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
+                input( cell ) > 0 ? TypeInfo< RealType >::getMaxValue() :
                                    - TypeInfo< RealType >::getMaxValue();
                 interfaceMap[ cell.getIndex() ] = false;
             }
@@ -269,20 +303,33 @@ initInterface( const MeshFunctionType& input,
             {
                auto neighbors = cell.getNeighborEntities();
                Real pom = 0;
-               //const IndexType& c = cell.getIndex();
                const IndexType e = neighbors.template getEntityIndex<  1,  0,  0 >();
-               //const IndexType w = neighbors.template getEntityIndex< -1,  0,  0 >();
                const IndexType n = neighbors.template getEntityIndex<  0,  1,  0 >();
-               //const IndexType s = neighbors.template getEntityIndex<  0, -1,  0 >();
                const IndexType t = neighbors.template getEntityIndex<  0,  0,  1 >();
-               //const IndexType b = neighbors.template getEntityIndex<  0,  0, -1 >();
-               /*if( c * input[ e ] <= 0 || c * input[ w ] <= 0 ||
-                   c * input[ n ] <= 0 || c * input[ s ] <= 0 ||
-                   c * input[ t ] <= 0 || c * input[ b ] <= 0 )
+               //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;
-                  continue;
                }*/
                if( c * input[ n ] <= 0 )
                {
@@ -357,7 +404,7 @@ initInterface( const MeshFunctionType& input,
                    }
                interfaceMap[ cell.getIndex() ] = true;
                interfaceMap[ t ] = true;
-               }        
+               }    
             }
             /*output[ cell.getIndex() ] =
                c > 0 ? TypeInfo< RealType >::getMaxValue() :
@@ -374,7 +421,7 @@ void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
 updateCell( MeshFunctionType& u,
             const MeshEntity& cell, 
-            double v )
+            const RealType v )
 {
    const auto& neighborEntities = cell.template getNeighborEntities< 3 >();
    const MeshType& mesh = cell.getMesh();
@@ -383,7 +430,8 @@ updateCell( MeshFunctionType& u,
    const RealType& hy = mesh.getSpaceSteps().y();
    const RealType& hz = mesh.getSpaceSteps().z();
    const RealType value = u( cell );
-   Real a, b, c, tmp = TypeInfo< RealType >::getMaxValue();
+   //std::cout << value << std::endl;
+   RealType a, b, c, tmp = TypeInfo< RealType >::getMaxValue();
    
    
    if( cell.getCoordinates().x() == 0 )
@@ -412,16 +460,12 @@ updateCell( MeshFunctionType& u,
       c = TNL::argAbsMin( u[ neighborEntities.template getEntityIndex< 0, 0, -1 >() ],
                          u[ neighborEntities.template getEntityIndex< 0, 0, 1 >() ] );
    }
-   if( fabs( a ) == TypeInfo< Real >::getMaxValue() && 
-       fabs( b ) == TypeInfo< Real >::getMaxValue() &&
-       fabs( c ) == TypeInfo< Real >::getMaxValue() )
+   if( fabs( a ) == TypeInfo< RealType >::getMaxValue() && 
+       fabs( b ) == TypeInfo< RealType >::getMaxValue() &&
+       fabs( c ) == TypeInfo< RealType >::getMaxValue() )
       return;
-   if( fabs( a ) == TypeInfo< Real >::getMaxValue() ||
-       fabs( b ) == TypeInfo< Real >::getMaxValue() ||
-       fabs( c ) == TypeInfo< Real >::getMaxValue() ||
-       hz * hz * fabs( a - b ) * fabs( a - b ) + hy * hy * fabs( a - c ) * fabs( a - c ) +
-           hx * hx * fabs( b - c ) * fabs( b - c ) >=  ( hx * hx * hy * hy + hx * hx * hz * hz + hy * hy * hz * hz )/v )
-   {
+   
+   
        /*if( fabs( a ) != TypeInfo< Real >::getMaxValue() &&
            fabs( b ) != TypeInfo< Real >::getMaxValue() &&
            fabs( a - b ) >= TNL::sqrt( (hx * hx + hy * hy)/v ) )
@@ -446,22 +490,31 @@ updateCell( MeshFunctionType& u,
                 sign( value ) * hy * hz * sqrt( ( hy * hy + hz * hz )/v - 
                 ( b - c ) * ( b - c ) ) )/( hy * hy + hz * hz );
        }*/
-       Real pom[6] = { a, b, c, (Real)hx, (Real)hy, (Real)hz};
-       sortMinims( pom );
-       tmp = meet2DCondition( pom[0], pom[1], pom[2], pom[3], value, v);
-       
-       if( tmp == TypeInfo< Real >::getMaxValue() )
-           tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 2 ]/v; 
-        
-   }
-   else
-      tmp = ( hx * hx * a + hy * hy * b + hz * hz * c +
-            sign( value ) * hx * hy * hz * sqrt( ( hx * hx + hy * hy + hz * hz)/v - 
-            hz * hz * ( a - b ) * ( a - b ) + hy * hy * ( a - c ) * ( a - c ) +
-            hx * hx * ( b - c ) * ( b - c ) ) )/( hx * hx + hy * hy + hz * hz );
-            
-   
-   u[ cell.getIndex() ] = argAbsMin( value, tmp );
+    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 );
+        }
+        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 );
+        }
+    }
 }
 
 template < typename T1, typename T2 >
@@ -470,10 +523,10 @@ T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double
    T1 tmp;
    if( fabs( a ) != TypeInfo< T1 >::getMaxValue() &&
        fabs( b ) != TypeInfo< T1 >::getMaxValue() &&
-       fabs( a - b ) <= TNL::sqrt( (ha * ha + hb * hb)/v ) )
+       fabs( a - b ) < ha/v )//TNL::sqrt( (ha * ha + hb * hb)/2 )/v )
    {
-      tmp = ( ha * ha * a + hb * hb * b + 
-            sign( value ) * hb * hb * sqrt( ( ha * ha + hb * hb )/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
@@ -487,26 +540,35 @@ T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double
 template < typename T1 >
 void sortMinims( T1 pom[])
 {
-    T1 tmp[6]; 
+    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[3]; tmp[3] = pom[4];
+        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[3]; tmp[3] = pom[5];
+        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[4]; tmp[3] = pom[3];
+        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[4]; tmp[3] = pom[5];
+        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[5]; tmp[3] = pom[3];
+        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[5]; tmp[3] = pom[4];
+        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 ];
-        
+    }   
 }
\ No newline at end of file
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 bb1d4a018..e90743a80 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
@@ -85,7 +85,8 @@ solve( const MeshPointer& mesh,
                   this->updateCell( aux, cell );
             }
       }
-      aux.save( "aux-1.tnl" );
+      
+      //aux.save( "aux-1.tnl" );
 
       for( cell.getCoordinates().y() = 0;
            cell.getCoordinates().y() < mesh->getDimensions().y();
@@ -101,7 +102,8 @@ solve( const MeshPointer& mesh,
                   this->updateCell( aux, cell );
             }
       }
-      aux.save( "aux-2.tnl" );
+      
+      //aux.save( "aux-2.tnl" );
 
       for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
            cell.getCoordinates().y() >= 0 ;
@@ -117,9 +119,9 @@ solve( const MeshPointer& mesh,
                   this->updateCell( aux, cell );
             }
          }
-      aux.save( "aux-3.tnl" );
-
-
+      
+      //aux.save( "aux-3.tnl" );
+      
       for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
            cell.getCoordinates().y() >= 0;
            cell.getCoordinates().y()-- )
@@ -133,8 +135,74 @@ solve( const MeshPointer& mesh,
                if( ! interfaceMap( cell ) )            
                   this->updateCell( aux, cell );
             }
-         }   
-      aux.save( "aux-4.tnl" );
+         }
+            
+      //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 );
+            }
+      }
+      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 );
+            }
+         }*/
+      
       iteration++;
    }
    aux.save("aux-final.tnl");
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 2c281617f..9844c4fca 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
@@ -89,7 +89,7 @@ solve( const MeshPointer& mesh,
             }
          }
       }
-      aux.save( "aux-1.tnl" );
+      //aux.save( "aux-1.tnl" );
 
       for( cell.getCoordinates().z() = 0;
            cell.getCoordinates().z() < mesh->getDimensions().z();
@@ -110,7 +110,7 @@ solve( const MeshPointer& mesh,
             }
          }
       }
-      aux.save( "aux-2.tnl" );
+      //aux.save( "aux-2.tnl" );
       for( cell.getCoordinates().z() = 0;
            cell.getCoordinates().z() < mesh->getDimensions().z();
            cell.getCoordinates().z()++ )
@@ -130,7 +130,7 @@ solve( const MeshPointer& mesh,
             }
          }
       }
-      aux.save( "aux-3.tnl" );
+      //aux.save( "aux-3.tnl" );
       
       for( cell.getCoordinates().z() = 0;
            cell.getCoordinates().z() < mesh->getDimensions().z();
@@ -151,7 +151,7 @@ solve( const MeshPointer& mesh,
             }
          }
       }     
-      aux.save( "aux-4.tnl" );
+      //aux.save( "aux-4.tnl" );
       
       for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
            cell.getCoordinates().z() >= 0;
@@ -172,7 +172,7 @@ solve( const MeshPointer& mesh,
             }
          }
       }
-      aux.save( "aux-5.tnl" );
+      //aux.save( "aux-5.tnl" );
 
       for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
            cell.getCoordinates().z() >= 0;
@@ -193,7 +193,7 @@ solve( const MeshPointer& mesh,
             }
          }
       }
-      aux.save( "aux-6.tnl" );
+      //aux.save( "aux-6.tnl" );
       
       for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
            cell.getCoordinates().z() >= 0;
@@ -214,7 +214,7 @@ solve( const MeshPointer& mesh,
             }
          }
       }
-      aux.save( "aux-7.tnl" );
+      //aux.save( "aux-7.tnl" );
 
       for( cell.getCoordinates().z() = mesh->getDimensions().z() - 1;
            cell.getCoordinates().z() >= 0;
@@ -235,7 +235,7 @@ solve( const MeshPointer& mesh,
             }
          }
       }
-      aux.save( "aux-8.tnl" );
+      //aux.save( "aux-8.tnl" );
       iteration++;
       
    }
-- 
GitLab


From bb045e76958f3b54a244b993e8267d49cbbee4c1 Mon Sep 17 00:00:00 2001
From: Tomas Oberhuber <tomas.oberhuber@fjfi.cvut.cz>
Date: Thu, 5 Apr 2018 11:48:06 +0200
Subject: [PATCH 10/28] Added CUDA switch.

---
 .../tnlDirectEikonalMethodsBase.h             |  17 +-
 .../tnlFastSweepingMethod2D_impl.h            | 234 +++++++++---------
 2 files changed, 129 insertions(+), 122 deletions(-)

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 4dc8e9228..2dabb5f1a 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -9,6 +9,7 @@
 
 #include <TNL/Meshes/Grid.h>
 #include <TNL/Functions/MeshFunction.h>
+#include <TNL/Devices/Cuda.h>
 
 using namespace TNL;
 
@@ -36,8 +37,8 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >
                           InterfaceMapType& interfaceMap );
       
       template< typename MeshEntity >
-      void updateCell( MeshFunctionType& u,
-                       const MeshEntity& cell );
+      __cuda_callable__ void updateCell( MeshFunctionType& u,
+                                         const MeshEntity& cell );
       
 };
 
@@ -61,9 +62,9 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
                           InterfaceMapType& interfaceMap );
       
       template< typename MeshEntity >
-      void updateCell( MeshFunctionType& u,
-                       const MeshEntity& cell,
-                       const RealType velocity = 1.0 );
+      __cuda_callable__ void updateCell( MeshFunctionType& u,
+                                         const MeshEntity& cell,
+                                         const RealType velocity = 1.0 );
 };
 
 template< typename Real,
@@ -85,9 +86,9 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
                           InterfaceMapType& interfaceMap );
       
       template< typename MeshEntity >
-      void updateCell( MeshFunctionType& u,
-                       const MeshEntity& cell,
-                       const RealType velocity = 1.0);
+      __cuda_callable__ void updateCell( MeshFunctionType& u,
+                                         const MeshEntity& cell,
+                                         const RealType velocity = 1.0);
       
       /*Real sort( Real a, Real b, Real c,
                  const RealType& ha,
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 e90743a80..b727f0cd9 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
@@ -71,137 +71,143 @@ solve( const MeshPointer& mesh,
    IndexType iteration( 0 );
    while( iteration < this->maxIterations )
    {
-
-      for( cell.getCoordinates().y() = 0;
-           cell.getCoordinates().y() < mesh->getDimensions().y();
-           cell.getCoordinates().y()++ )
+      if( std::is_same< DeviceType, Devices::Host >::value )
       {
-         for( cell.getCoordinates().x() = 0;
-              cell.getCoordinates().x() < mesh->getDimensions().x();
-              cell.getCoordinates().x()++ )
+         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()-- )
             {
-               cell.refresh();
-               if( ! interfaceMap( cell ) )
-                  this->updateCell( aux, cell );
+            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-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()-- )		
+         //aux.save( "aux-3.tnl" );
+
+         for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
+              cell.getCoordinates().y() >= 0;
+              cell.getCoordinates().y()-- )
             {
-               //std::cerr << "2 -> ";
-               cell.refresh();
-               if( ! interfaceMap( cell ) )            
-                  this->updateCell( aux, cell );
+            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-2.tnl" );
 
-      for( cell.getCoordinates().y() = mesh->getDimensions().y() - 1;
-           cell.getCoordinates().y() >= 0 ;
-           cell.getCoordinates().y()-- )
+         //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().x();
+              cell.getCoordinates().x() < mesh->getDimensions().y();
               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;
+            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 );
+               }
+         }
+         aux.save( "aux-6.tnl" );
+
+         for( cell.getCoordinates().x() = mesh->getDimensions().y() - 1;
               cell.getCoordinates().x() >= 0 ;
-              cell.getCoordinates().x()-- )		
+              cell.getCoordinates().x()-- )
             {
-               //std::cerr << "4 -> ";
-               cell.refresh();
-               if( ! interfaceMap( cell ) )            
-                  this->updateCell( aux, cell );
+            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-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()++ )
+         aux.save( "aux-7.tnl" );
+
+         for( cell.getCoordinates().x() = mesh->getDimensions().y() - 1;
+              cell.getCoordinates().x() >= 0;
+              cell.getCoordinates().x()-- )
             {
-               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 << "4 -> ";
+                  cell.refresh();
+                  if( ! interfaceMap( cell ) )            
+                     this->updateCell( aux, cell );
+               }
+            }*/
+      }
+      if( std::is_same< DeviceType, Devices::Cuda >::value )
       {
-         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 );
-            }
+         // TODO: CUDA code
       }
-      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 );
-            }
-         }*/
       
       iteration++;
    }
-- 
GitLab


From c04814b8bbdf2cec64ed78828286b3bd2d078492 Mon Sep 17 00:00:00 2001
From: Tomas Oberhuber <tomas.oberhuber@fjfi.cvut.cz>
Date: Thu, 12 Apr 2018 12:11:52 +0200
Subject: [PATCH 11/28] Added CUDA build for eikonal solver.

---
 .../Solvers/hamilton-jacobi/CMakeLists.txt               | 9 +++++++--
 .../Solvers/hamilton-jacobi/tnlFastSweepingMethod.h      | 6 +++---
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/CMakeLists.txt b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/CMakeLists.txt
index 176c9859e..fa8dd56d3 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/CMakeLists.txt
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/CMakeLists.txt
@@ -10,8 +10,13 @@ set( tnl_hamilton_jacobi_SOURCES
 #ADD_EXECUTABLE(tnl-hamilton-jacobi${debugExt} main.cpp)
 #target_link_libraries (tnl-hamilton-jacobi${debugExt} tnl${debugExt}-${tnlVersion} )
 
-ADD_EXECUTABLE(tnl-direct-eikonal-solver${debugExt} tnl-direct-eikonal-solver.cpp )
-target_link_libraries (tnl-direct-eikonal-solver${debugExt} tnl${debugExt}-${tnlVersion} )
+IF( BUILD_CUDA )
+    CUDA_ADD_EXECUTABLE(tnl-direct-eikonal-solver${debugExt} tnl-direct-eikonal-solver.cu )
+    target_link_libraries (tnl-direct-eikonal-solver${debugExt} tnl${debugExt}-${tnlVersion} ${CUSPARSE_LIBRARY} )
+ELSE(  BUILD_CUDA )               
+    ADD_EXECUTABLE(tnl-direct-eikonal-solver${debugExt} tnl-direct-eikonal-solver.cpp )
+    target_link_libraries (tnl-direct-eikonal-solver${debugExt} tnl${debugExt}-${tnlVersion} )
+ENDIF( BUILD_CUDA )
 
 
 INSTALL( TARGETS #tnl-hamilton-jacobi${debugExt} 
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 17d139b0e..61104a143 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
@@ -28,7 +28,7 @@ template< typename Real,
 class FastSweepingMethod< Meshes::Grid< 1, Real, Device, Index >, Anisotropy >
    : 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." );
+   //static_assert(  std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
    
    public:
       
@@ -66,7 +66,7 @@ template< typename Real,
 class FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >
    : 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." );
+   //static_assert(  std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
    
    public:
       
@@ -104,7 +104,7 @@ template< typename Real,
 class FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >
    : 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." );
+   //static_assert(  std::is_same< Device, TNL::Devices::Host >::value, "The fast sweeping method works only on CPU." );
    
    public:
       
-- 
GitLab


From 092c253803f6ba824126e16b48b361452b58701b Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Tue, 17 Apr 2018 15:04:48 +0200
Subject: [PATCH 12/28] Cuda init start

---
 .../tnlDirectEikonalMethodsBase.h             |  15 +-
 .../tnlDirectEikonalMethodsBase_impl.h        | 275 ++++++++++++------
 .../tnlFastSweepingMethod2D_impl.h            |  49 +++-
 3 files changed, 245 insertions(+), 94 deletions(-)

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 2dabb5f1a..e6781a7d6 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -49,7 +49,6 @@ template< typename Real,
 class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
 {
    public:
-      
       typedef Meshes::Grid< 2, Real, Device, Index > MeshType;
       typedef Real RealType;
       typedef Device DevcieType;
@@ -65,6 +64,8 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
       __cuda_callable__ void updateCell( MeshFunctionType& u,
                                          const MeshEntity& cell,
                                          const RealType velocity = 1.0 );
+   protected:
+       
 };
 
 template< typename Real,
@@ -100,6 +101,16 @@ template < typename T1, typename T2 >
 T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double v = 1);
 
 template < typename T1 >
-void sortMinims( T1 pom[] );
+__cuda_callable__ void sortMinims( T1 pom[] );
+
+
+template < typename Real, typename Device, typename Index >
+__global__ void CudaUpdateCellCaller( Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
+                                      Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux );
+
+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 );
 
 #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 caa2f533f..1df41cf80 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
@@ -8,7 +8,9 @@
 #pragma once
 
 #include <TNL/TypeInfo.h>
+
 #include <iostream>
+#include "tnlFastSweepingMethod.h"
 template< typename Real,
           typename Device,
           typename Index >
@@ -67,7 +69,6 @@ updateCell( MeshFunctionType& u,
 {
 }
 
-
 template< typename Real,
           typename Device,
           typename Index >
@@ -75,102 +76,125 @@ void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
 initInterface( const MeshFunctionType& input,
                MeshFunctionType& output,
-               InterfaceMapType& interfaceMap  )
+               InterfaceMapType& interfaceMap )
 {
+    /*
+     doplnit přepočty pro cudu:
+     * overit is_same device
+     * na kazdy bod jedno cuda vlakno
+     */
     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 ? TypeInfo< RealType >::getMaxValue() :
-                                   - TypeInfo< RealType >::getMaxValue();
-                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() ++ )
-      {
-         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 )
-            {
-                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
+    
+    if( std::is_same< Device, Devices::Cuda >::value )
+    {
+#ifdef HAVE_CUDA
+        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< Real, Device, Index ><<< gridSize, blockSize >>>( input, output, interfaceMap );
+        TNL_CHECK_CUDA_DEVICE;
+#endif
+    }
+    if( std::is_same< Device, 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() ++ )
                 {
-                    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 ]);
+                    cell.refresh();
+                    output[ cell.getIndex() ] =
+                    input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
+                                       - TypeInfo< RealType >::getMaxValue();
+                    interfaceMap[ cell.getIndex() ] = false;
                 }
-                interfaceMap[ cell.getIndex() ] = true;
-                interfaceMap[ n ] = true;
-            }
-            if( c * input[ e ] <= 0 )
-            {
-                if( c >= 0 )
+
+       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() ++ )
+          {
+             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 )
                 {
-                    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
+                    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 )
                 {
-                    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;
+                    /*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;
                 }
-                interfaceMap[ cell.getIndex() ] = true;
-                interfaceMap[ e ] = 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;
+                }
+             }
+          }
       }
 }
 
@@ -190,7 +214,7 @@ updateCell( MeshFunctionType& u,
    const RealType& hx = mesh.getSpaceSteps().x();
    const RealType& hy = mesh.getSpaceSteps().y();
    const RealType value = u( cell );
-   RealType a, b, tmp1, tmp = TypeInfo< RealType >::getMaxValue();
+   RealType a, b, tmp = TypeInfo< RealType >::getMaxValue();
    
    if( cell.getCoordinates().x() == 0 )
       a = u[ neighborEntities.template getEntityIndex< 1,  0 >() ];
@@ -538,7 +562,7 @@ T1 meet2DCondition( T1 a, T1 b, const T2 ha, const T2 hb, const T1 value, double
 }
 
 template < typename T1 >
-void sortMinims( T1 pom[])
+__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])){
@@ -571,4 +595,75 @@ void sortMinims( T1 pom[])
     {
         pom[ i ] = tmp[ i ];
     }   
+}
+
+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 ) 
+{
+    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.getMesh();
+    
+    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();
+
+
+        output[ cell.getIndex() ] =
+               input( cell ) >= 0 ? TypeInfo< Real >::getMaxValue() :
+                                    - TypeInfo< Real >::getMaxValue();
+        interfaceMap[ cell.getIndex() ] = 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[ cell.getIndex() ] ) > TNL::abs( pom ) ) 
+                   output[ cell.getIndex() ] = pom;
+
+               interfaceMap[ cell.getIndex() ] = true;
+           }
+           if( c * input[ e ] <= 0 )
+           {
+               pom = TNL::sign( c )*( hx * c )/( c - input[ e ]);
+               if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) )
+                   output[ cell.getIndex() ] = pom;                       
+
+               interfaceMap[ cell.getIndex() ] = true;
+           }
+           if( c * input[ w ] <= 0 )
+           {
+               pom = TNL::sign( c )*( hx * c )/( c - input[ w ]);
+               if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) ) 
+                   output[ cell.getIndex() ] = pom;
+
+               interfaceMap[ cell.getIndex() ] = true;
+           }
+           if( c * input[ s ] <= 0 )
+           {
+               pom = TNL::sign( c )*( hy * c )/( c - input[ s ]);
+               if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) ) 
+                   output[ cell.getIndex() ] = pom;
+
+               interfaceMap[ cell.getIndex() ] = true;
+           }
+        }
+    }
 }
\ No newline at end of file
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 b727f0cd9..0f939d755 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
@@ -64,7 +64,17 @@ solve( const MeshPointer& mesh,
    interfaceMap.setMesh( mesh );
    std::cout << "Initiating the interface cells ..." << std::endl;
    BaseType::initInterface( u, aux, interfaceMap );
-   aux.save( "aux-ini.tnl" );
+   
+   //if( std::is_same< DeviceType, Devices::Cuda >::value )
+   //{
+   //    Functions::MeshFunction< Meshes::Grid< 2, Real, TNL::Devices::Host, Index > > h_aux;
+       //cudaMemcpy( h_aux, aux, sizeof(MeshFunctionType), cudaMemcpyDeviceToHost );
+       //h_aux->save("aux-init-cuda.tnl");
+   //}
+   //if( std::is_same< DeviceType, Devices::Host >::value )
+   {
+       aux.save( "aux-ini.tnl" );
+   }
 
    typename MeshType::Cell cell( *mesh );
    
@@ -207,10 +217,45 @@ solve( const MeshPointer& mesh,
       if( std::is_same< DeviceType, Devices::Cuda >::value )
       {
          // TODO: CUDA code
+          int numBlocks = 2;
+          int threadsPerBlock;
+          if( mesh->getDimensions().x() >= mesh->getDimensions().y() )
+               threadsPerBlock = (int)( mesh->getDimensions().x() );
+          else
+               threadsPerBlock = (int)( mesh->getDimensions().y() );
+          
+          CudaUpdateCellCaller< Real, Device, Index ><<< numBlocks, threadsPerBlock >>>( interfaceMap, aux );
+          cudaDeviceSynchronize(); //copak dela?
       }
-      
       iteration++;
    }
    aux.save("aux-final.tnl");
 }
 
+template < typename Real, typename Device, typename Index >
+__global__ void CudaUpdateCellCaller( Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
+                                      Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux )
+{
+    int i = threadIdx.x + blockDim.x*blockIdx.x;
+    int j = threadIdx.y + blockDim.y*blockIdx.y;
+    const Meshes::Grid< 2, Real, Device, Index >& mesh = aux.getMesh();
+    
+    if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+    {
+        //make cell of aux from index
+        typedef typename Meshes::Grid< 2, Real, Device, Index >::Cell Cell;
+        Cell cell( mesh );
+        cell.getCoordinates().x() = i; cell.getCoordinates().y() = j;
+        cell.refresh();
+
+        //update cell value few times 
+        //for( int i = 0; i < mesh.getDimensions() ; i++ )
+        //{
+            cell.refresh();
+            if( ! interfaceMap( cell ) )
+            {
+        //        tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::updateCell( aux, cell );
+            }
+        //}
+    }
+}
-- 
GitLab


From db36cd42cfeb2e6a6c018b9f9ff6c2186a68cc7e Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Wed, 18 Apr 2018 15:55:25 +0200
Subject: [PATCH 13/28] CUDA init problem not solved

---
 .../tnlDirectEikonalMethodsBase.h             |  8 ++++--
 .../tnlDirectEikonalMethodsBase_impl.h        | 28 ++++++++++++++-----
 .../tnlFastSweepingMethod2D_impl.h            | 20 +++++--------
 3 files changed, 34 insertions(+), 22 deletions(-)

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 e6781a7d6..ed176f08a 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -104,13 +104,17 @@ template < typename T1 >
 __cuda_callable__ void sortMinims( T1 pom[] );
 
 
+#ifdef HAVE_CUDA
 template < typename Real, typename Device, typename Index >
 __global__ void CudaUpdateCellCaller( Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux );
 
-template < typename Real, typename Device, typename Index >
+/*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 >, 2, bool >& interfaceMap );*/
+
+__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, double, TNL::Devices::Cuda, int > >& input );
+#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 1df41cf80..8444b6c37 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
@@ -83,25 +83,30 @@ initInterface( const MeshFunctionType& input,
      * overit is_same device
      * na kazdy bod jedno cuda vlakno
      */
-    const MeshType& mesh = input.getMesh();
-    typedef typename MeshType::Cell Cell;
-    Cell cell( mesh );
     
+        
     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< Real, Device, Index ><<< gridSize, blockSize >>>( input, output, interfaceMap );
+        //CudaInitCaller< Real, Device, Index ><<< gridSize, blockSize >>>( input, output, interfaceMap );
+        CudaInitCaller<<< gridSize, blockSize >>>( input );
+        cudaDeviceSynchronize();
         TNL_CHECK_CUDA_DEVICE;
 #endif
     }
     if( std::is_same< Device, Devices::Host >::value )
     {
+        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() ++ )
@@ -597,7 +602,7 @@ __cuda_callable__ void sortMinims( T1 pom[])
     }   
 }
 
-template < typename Real, typename Device, typename Index >
+/*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 ) 
@@ -606,7 +611,7 @@ __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2,
     int j = blockDim.y*blockIdx.y + threadIdx.y;
     const Meshes::Grid< 2, Real, Device, Index >& mesh = input.getMesh();
     
-    if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+    //if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
     {
         typedef typename Meshes::Grid< 2, Real, Device, Index >::Cell Cell;
         Cell cell( mesh );
@@ -666,4 +671,13 @@ __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2,
            }
         }
     }
-}
\ No newline at end of file
+}*/
+
+
+__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, double, TNL::Devices::Cuda, int > >& input )
+{
+    int i = threadIdx.x + blockDim.x*blockIdx.x;
+    int j = blockDim.y*blockIdx.y + threadIdx.y;
+    //const Meshes::Grid< 2, double, TNL::Devices::Cuda, int >& mesh = input.getMesh();
+    
+}
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 0f939d755..48640e3fc 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
@@ -64,17 +64,9 @@ solve( const MeshPointer& mesh,
    interfaceMap.setMesh( mesh );
    std::cout << "Initiating the interface cells ..." << std::endl;
    BaseType::initInterface( u, aux, interfaceMap );
-   
-   //if( std::is_same< DeviceType, Devices::Cuda >::value )
-   //{
-   //    Functions::MeshFunction< Meshes::Grid< 2, Real, TNL::Devices::Host, Index > > h_aux;
-       //cudaMemcpy( h_aux, aux, sizeof(MeshFunctionType), cudaMemcpyDeviceToHost );
-       //h_aux->save("aux-init-cuda.tnl");
-   //}
-   //if( std::is_same< DeviceType, Devices::Host >::value )
-   {
-       aux.save( "aux-ini.tnl" );
-   }
+   cudaDeviceSynchronize();
+        
+   aux.save( "aux-ini.tnl" );
 
    typename MeshType::Cell cell( *mesh );
    
@@ -217,7 +209,8 @@ solve( const MeshPointer& mesh,
       if( std::is_same< DeviceType, Devices::Cuda >::value )
       {
          // TODO: CUDA code
-          int numBlocks = 2;
+#ifdef HAVE_CUDA
+          /*int numBlocks = 2;
           int threadsPerBlock;
           if( mesh->getDimensions().x() >= mesh->getDimensions().y() )
                threadsPerBlock = (int)( mesh->getDimensions().x() );
@@ -225,7 +218,8 @@ solve( const MeshPointer& mesh,
                threadsPerBlock = (int)( mesh->getDimensions().y() );
           
           CudaUpdateCellCaller< Real, Device, Index ><<< numBlocks, threadsPerBlock >>>( interfaceMap, aux );
-          cudaDeviceSynchronize(); //copak dela?
+          cudaDeviceSynchronize(); //copak dela?*/
+#endif
       }
       iteration++;
    }
-- 
GitLab


From 6d1a8d988ab4960715d100af5c6023f3b719207f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1=C5=A1=20Oberhuber?= <oberhuber.tomas@gmail.com>
Date: Wed, 18 Apr 2018 21:19:42 +0200
Subject: [PATCH 14/28] Fixed FSM interface initiation in CUDA.

---
 .../tnlDirectEikonalMethodsBase.h             | 30 +++++++-----
 .../tnlDirectEikonalMethodsBase_impl.h        | 49 ++++++++++++-------
 .../hamilton-jacobi/tnlDirectEikonalProblem.h |  8 +--
 .../tnlDirectEikonalProblem_impl.h            |  6 +--
 .../hamilton-jacobi/tnlFastSweepingMethod.h   | 30 +++++++-----
 .../tnlFastSweepingMethod1D_impl.h            | 14 +++---
 .../tnlFastSweepingMethod2D_impl.h            | 18 ++++---
 .../tnlFastSweepingMethod3D_impl.h            | 18 ++++---
 8 files changed, 102 insertions(+), 71 deletions(-)

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 ed176f08a..d8770b0e7 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -31,10 +31,12 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >
       typedef Index IndexType;
       typedef Functions::MeshFunction< MeshType > MeshFunctionType;
       typedef Functions::MeshFunction< MeshType, 1, bool > InterfaceMapType;
+      using MeshFunctionPointer = SharedPointer< MeshFunctionType >;
+      using InterfaceMapPointer = SharedPointer< InterfaceMapType >;
       
-      void initInterface( const MeshFunctionType& input,
-                          MeshFunctionType& output,
-                          InterfaceMapType& interfaceMap );
+      void initInterface( const MeshFunctionPointer& input,
+                          MeshFunctionPointer& output,
+                          InterfaceMapPointer& interfaceMap );
       
       template< typename MeshEntity >
       __cuda_callable__ void updateCell( MeshFunctionType& u,
@@ -55,10 +57,12 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
       typedef Index IndexType;
       typedef Functions::MeshFunction< MeshType > MeshFunctionType;
       typedef Functions::MeshFunction< MeshType, 2, bool > InterfaceMapType;
+      using MeshFunctionPointer = SharedPointer< MeshFunctionType >;
+      using InterfaceMapPointer = SharedPointer< InterfaceMapType >;      
 
-      void initInterface( const MeshFunctionType& input,
-                          MeshFunctionType& output,
-                          InterfaceMapType& interfaceMap );
+      void initInterface( const MeshFunctionPointer& input,
+                          MeshFunctionPointer& output,
+                          InterfaceMapPointer& interfaceMap );
       
       template< typename MeshEntity >
       __cuda_callable__ void updateCell( MeshFunctionType& u,
@@ -81,10 +85,12 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
       typedef Index IndexType;
       typedef Functions::MeshFunction< MeshType > MeshFunctionType;
       typedef Functions::MeshFunction< MeshType, 3, bool > InterfaceMapType;
+      using MeshFunctionPointer = SharedPointer< MeshFunctionType >;
+      using InterfaceMapPointer = SharedPointer< InterfaceMapType >;      
 
-      void initInterface( const MeshFunctionType& input,
-                          MeshFunctionType& output,
-                          InterfaceMapType& interfaceMap );
+      void initInterface( const MeshFunctionPointer& input,
+                          MeshFunctionPointer& output,
+                          InterfaceMapPointer& interfaceMap );
       
       template< typename MeshEntity >
       __cuda_callable__ void updateCell( MeshFunctionType& u,
@@ -109,12 +115,12 @@ template < typename Real, typename Device, typename Index >
 __global__ void CudaUpdateCellCaller( Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux );
 
-/*template < typename Real, typename Device, typename Index >
+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 >, 2, bool >& interfaceMap );
 
-__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, double, TNL::Devices::Cuda, int > >& input );
+//__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, double, TNL::Devices::Cuda, int > >& input );
 #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 8444b6c37..37fd24512 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
@@ -16,12 +16,15 @@ template< typename Real,
           typename Index >
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >::
-initInterface( const MeshFunctionType& input,
-               MeshFunctionType& output,
-               InterfaceMapType& interfaceMap  )
+initInterface( const MeshFunctionPointer& _input,
+               MeshFunctionPointer& _output,
+               InterfaceMapPointer& _interfaceMap  )
 {
-   const MeshType& mesh = input.getMesh();
+   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() = 1;
         cell.getCoordinates().x() < mesh.getDimensions().x() - 1;
@@ -74,9 +77,9 @@ template< typename Real,
           typename Index >
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
-initInterface( const MeshFunctionType& input,
-               MeshFunctionType& output,
-               InterfaceMapType& interfaceMap )
+initInterface( const MeshFunctionPointer& _input,
+               MeshFunctionPointer& _output,
+               InterfaceMapPointer& _interfaceMap )
 {
     /*
      doplnit přepočty pro cudu:
@@ -88,7 +91,7 @@ initInterface( const MeshFunctionType& input,
     if( std::is_same< Device, Devices::Cuda >::value )
     {
 #ifdef HAVE_CUDA
-        const MeshType& mesh = input.getMesh();
+        const MeshType& mesh = _input->getMesh();
         
         const int cudaBlockSize( 16 );
         int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh.getDimensions().x(), cudaBlockSize );
@@ -96,14 +99,19 @@ initInterface( const MeshFunctionType& input,
         dim3 blockSize( cudaBlockSize, cudaBlockSize );
         dim3 gridSize( numBlocksX, numBlocksY );
         Devices::Cuda::synchronizeDevice();
-        //CudaInitCaller< Real, Device, Index ><<< gridSize, blockSize >>>( input, output, interfaceMap );
-        CudaInitCaller<<< gridSize, blockSize >>>( input );
+        CudaInitCaller<<< gridSize, blockSize >>>( _input.template getData< Device >(),
+                                                   _output.template modifyData< Device >(),
+                                                   _interfaceMap.template modifyData< Device >() );
+        //CudaInitCaller<<< gridSize, blockSize >>>( input );
         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 );
@@ -289,10 +297,13 @@ template< typename Real,
           typename Index >
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
-initInterface( const MeshFunctionType& input,
-               MeshFunctionType& output,
-               InterfaceMapType& interfaceMap  )
+initInterface( const MeshFunctionPointer& _input,
+               MeshFunctionPointer& _output,
+               InterfaceMapPointer& _interfaceMap  )
 {
+   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 );
@@ -602,16 +613,16 @@ __cuda_callable__ void sortMinims( T1 pom[])
     }   
 }
 
-/*template < typename Real, typename Device, typename Index >
+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 ) 
 {
     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.getMesh();
+    const Meshes::Grid< 2, Real, Device, Index >& mesh = input.template getMesh< Devices::Cuda >();
     
-    //if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+    if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
     {
         typedef typename Meshes::Grid< 2, Real, Device, Index >::Cell Cell;
         Cell cell( mesh );
@@ -671,13 +682,13 @@ __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2,
            }
         }
     }
-}*/
+}
 
 
-__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, double, TNL::Devices::Cuda, int > >& input )
+/*__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, double, TNL::Devices::Cuda, int > >& input )
 {
     int i = threadIdx.x + blockDim.x*blockIdx.x;
     int j = blockDim.y*blockIdx.y + threadIdx.y;
     //const Meshes::Grid< 2, double, TNL::Devices::Cuda, int >& mesh = input.getMesh();
     
-}
+}*/
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem.h
index ca3b8368c..66513d76f 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem.h
@@ -36,6 +36,8 @@ class tnlDirectEikonalProblem
       typedef Functions::MeshFunction< Mesh > MeshFunctionType;
       typedef Problems::PDEProblem< Mesh, RealType, DeviceType, IndexType > BaseType;
       using AnisotropyType = Anisotropy;
+      using AnisotropyPointer = SharedPointer< AnisotropyType, DeviceType >;
+      using MeshFunctionPointer = SharedPointer< MeshFunctionType >;
 
       using typename BaseType::MeshType;
       using typename BaseType::DofVectorType;
@@ -76,11 +78,11 @@ class tnlDirectEikonalProblem
 
       protected:
          
-         MeshFunctionType u;
+         MeshFunctionPointer u;
          
-         MeshFunctionType initialData;
+         MeshFunctionPointer initialData;
          
-         AnisotropyType anisotropy;
+         AnisotropyPointer anisotropy;
 
 };
 
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem_impl.h
index b10c6949e..57ac7d689 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem_impl.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem_impl.h
@@ -95,7 +95,7 @@ tnlDirectEikonalProblem< Mesh, Anisotropy, Real, Index >::
 bindDofs( const MeshPointer& mesh,
           const DofVectorPointer& dofs )
 {
-   this->u.bind( mesh, dofs );
+   this->u->bind( mesh, dofs );
 }
 
 template< typename Mesh,
@@ -110,8 +110,8 @@ setInitialCondition( const Config::ParameterContainer& parameters,
                      MeshDependentDataPointer& meshdependentData )
 {
    String inputFile = parameters.getParameter< String >( "input-file" );
-   this->initialData.setMesh( mesh );
-   if( !this->initialData.boundLoad( inputFile ) )
+   this->initialData->setMesh( mesh );
+   if( !this->initialData->boundLoad( inputFile ) )
       return false;
    return true;
 }
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 61104a143..48f330132 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
@@ -34,14 +34,18 @@ class FastSweepingMethod< Meshes::Grid< 1, Real, Device, Index >, Anisotropy >
       
       typedef Meshes::Grid< 1, Real, Device, Index > MeshType;
       typedef Real RealType;
-      typedef TNL::Devices::Host DeviceType;
+      typedef Device DeviceType;
       typedef Index IndexType;
       typedef Anisotropy AnisotropyType;
       typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > > BaseType;
       using MeshPointer = SharedPointer< MeshType >;
+      using AnisotropyPointer = SharedPointer< AnisotropyType, DeviceType >;
       
       using typename BaseType::InterfaceMapType;
       using typename BaseType::MeshFunctionType;
+      using typename BaseType::InterfaceMapPointer;
+      using typename BaseType::MeshFunctionPointer;
+      
       
       FastSweepingMethod();
       
@@ -50,8 +54,8 @@ class FastSweepingMethod< Meshes::Grid< 1, Real, Device, Index >, Anisotropy >
       void setMaxIterations( const IndexType& maxIterations );
       
       void solve( const MeshPointer& mesh,
-                  const AnisotropyType& anisotropy,
-                  MeshFunctionType& u );
+                  const AnisotropyPointer& anisotropy,
+                  MeshFunctionPointer& u );
       
       
    protected:
@@ -72,14 +76,17 @@ class FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >
       
       typedef Meshes::Grid< 2, Real, Device, Index > MeshType;
       typedef Real RealType;
-      typedef TNL::Devices::Host DeviceType;
+      typedef Device DeviceType;
       typedef Index IndexType;
       typedef Anisotropy AnisotropyType;
       typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > BaseType;
       using MeshPointer = SharedPointer< MeshType >;
+      using AnisotropyPointer = SharedPointer< AnisotropyType, DeviceType >;
 
       using typename BaseType::InterfaceMapType;
       using typename BaseType::MeshFunctionType;
+      using typename BaseType::InterfaceMapPointer;
+      using typename BaseType::MeshFunctionPointer;      
 
       FastSweepingMethod();
       
@@ -88,8 +95,8 @@ class FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >
       void setMaxIterations( const IndexType& maxIterations );
       
       void solve( const MeshPointer& mesh,
-                  const AnisotropyType& anisotropy,
-                  MeshFunctionType& u );
+                  const AnisotropyPointer& anisotropy,
+                  MeshFunctionPointer& u );
       
       
    protected:
@@ -110,14 +117,17 @@ class FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >
       
       typedef Meshes::Grid< 3, Real, Device, Index > MeshType;
       typedef Real RealType;
-      typedef TNL::Devices::Host DeviceType;
+      typedef Device DeviceType;
       typedef Index IndexType;
       typedef Anisotropy AnisotropyType;
       typedef tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > BaseType;
       using MeshPointer = SharedPointer< MeshType >;
+      using AnisotropyPointer = SharedPointer< AnisotropyType, DeviceType >;
       
       using typename BaseType::InterfaceMapType;
       using typename BaseType::MeshFunctionType;
+      using typename BaseType::InterfaceMapPointer;
+      using typename BaseType::MeshFunctionPointer;      
       
       FastSweepingMethod();
       
@@ -126,8 +136,8 @@ class FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >
       void setMaxIterations( const IndexType& maxIterations );
       
       void solve( const MeshPointer& mesh,
-                  const AnisotropyType& anisotropy,
-                  MeshFunctionType& u );
+                  const AnisotropyPointer& anisotropy,
+                  MeshFunctionPointer& u );
       
       
    protected:
@@ -135,8 +145,6 @@ class FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >
       const IndexType maxIterations;
 };
 
-
-
 #include "tnlFastSweepingMethod1D_impl.h"
 #include "tnlFastSweepingMethod2D_impl.h"
 #include "tnlFastSweepingMethod3D_impl.h"
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
index 874945eaf..105f1ceda 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
@@ -55,16 +55,16 @@ template< typename Real,
 void
 FastSweepingMethod< Meshes::Grid< 1, Real, Device, Index >, Anisotropy >::
 solve( const MeshPointer& mesh,
-       const AnisotropyType& anisotropy,
-       MeshFunctionType& u )
+       const AnisotropyPointer& anisotropy,
+       MeshFunctionPointer& u )
 {
-   MeshFunctionType aux;
-   InterfaceMapType interfaceMap;
-   aux.setMesh( mesh );
-   interfaceMap.setMesh( mesh );
+   MeshFunctionPointer aux;
+   InterfaceMapPointer interfaceMap;
+   aux->setMesh( mesh );
+   interfaceMap->setMesh( mesh );
    std::cout << "Initiating the interface cells ..." << std::endl;
    BaseType::initInterface( u, aux, interfaceMap );
-   aux.save( "aux-ini.tnl" );
+   aux->save( "aux-ini.tnl" );
 
 }
 
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 48640e3fc..e36224d07 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
@@ -55,22 +55,24 @@ template< typename Real,
 void
 FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >::
 solve( const MeshPointer& mesh,
-       const AnisotropyType& anisotropy,
-       MeshFunctionType& u )
+       const AnisotropyPointer& anisotropy,
+       MeshFunctionPointer& u )
 {
-   MeshFunctionType aux;
-   InterfaceMapType interfaceMap;
-   aux.setMesh( mesh );
-   interfaceMap.setMesh( mesh );
+   MeshFunctionPointer auxPtr;
+   InterfaceMapPointer interfaceMapPtr;
+   auxPtr->setMesh( mesh );
+   interfaceMapPtr->setMesh( mesh );
    std::cout << "Initiating the interface cells ..." << std::endl;
-   BaseType::initInterface( u, aux, interfaceMap );
+   BaseType::initInterface( u, auxPtr, interfaceMapPtr );
    cudaDeviceSynchronize();
         
-   aux.save( "aux-ini.tnl" );
+   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 )
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 9844c4fca..1fd8c9f2a 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
@@ -55,20 +55,22 @@ template< typename Real,
 void
 FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >::
 solve( const MeshPointer& mesh,
-       const AnisotropyType& anisotropy,
-       MeshFunctionType& u )
+       const AnisotropyPointer& anisotropy,
+       MeshFunctionPointer& u )
 {
-   MeshFunctionType aux;
-   InterfaceMapType interfaceMap;
-   aux.setMesh( mesh );
-   interfaceMap.setMesh( mesh );
+   MeshFunctionPointer auxPtr;
+   InterfaceMapPointer interfaceMapPtr;
+   auxPtr->setMesh( mesh );
+   interfaceMapPtr->setMesh( mesh );
    std::cout << "Initiating the interface cells ..." << std::endl;
-   BaseType::initInterface( u, aux, interfaceMap );
-   aux.save( "aux-ini.tnl" );   
+   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 )
    {
       for( cell.getCoordinates().z() = 0;
-- 
GitLab


From 5d1fab6e4fa455940aa0edf159e1ca3229738a36 Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Mon, 23 Apr 2018 11:01:29 +0200
Subject: [PATCH 15/28] CUDA implemented for exact number of loops in 2D and 3D

---
 .../tnlDirectEikonalMethodsBase.h             |  23 +-
 .../tnlDirectEikonalMethodsBase_impl.h        | 456 +++++++++++-------
 .../hamilton-jacobi/tnlFastSweepingMethod.h   |   2 +
 .../tnlFastSweepingMethod2D_impl.h            |  49 +-
 .../tnlFastSweepingMethod3D_impl.h            | 380 ++++++++-------
 5 files changed, 542 insertions(+), 368 deletions(-)

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 d8770b0e7..138671a3b 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -67,9 +67,7 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
       template< typename MeshEntity >
       __cuda_callable__ void updateCell( MeshFunctionType& u,
                                          const MeshEntity& cell,
-                                         const RealType velocity = 1.0 );
-   protected:
-       
+                                         const RealType velocity = 1.0 );      
 };
 
 template< typename Real,
@@ -78,7 +76,6 @@ template< typename Real,
 class tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
 {
    public:
-      
       typedef Meshes::Grid< 3, Real, Device, Index > MeshType;
       typedef Real RealType;
       typedef Device DevcieType;
@@ -96,11 +93,6 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
       __cuda_callable__ void updateCell( MeshFunctionType& u,
                                          const MeshEntity& cell,
                                          const RealType velocity = 1.0);
-      
-      /*Real sort( Real a, Real b, Real c,
-                 const RealType& ha,
-                 const RealType& hb,
-                 const RealType& hc ); */
 };
 
 template < typename T1, typename T2 >
@@ -112,7 +104,8 @@ __cuda_callable__ void sortMinims( T1 pom[] );
 
 #ifdef HAVE_CUDA
 template < typename Real, typename Device, typename Index >
-__global__ void CudaUpdateCellCaller( Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
+__global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr,
+                                      const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux );
 
 template < typename Real, typename Device, typename Index >
@@ -120,7 +113,15 @@ __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2,
                                 Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& output,
                                 Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap );
 
-//__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, double, TNL::Devices::Cuda, int > >& input );
+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 );
+
+template < 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 );
 #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 37fd24512..bf78a02a4 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
@@ -11,6 +11,7 @@
 
 #include <iostream>
 #include "tnlFastSweepingMethod.h"
+
 template< typename Real,
           typename Device,
           typename Index >
@@ -81,13 +82,7 @@ initInterface( const MeshFunctionPointer& _input,
                MeshFunctionPointer& _output,
                InterfaceMapPointer& _interfaceMap )
 {
-    /*
-     doplnit přepočty pro cudu:
-     * overit is_same device
-     * na kazdy bod jedno cuda vlakno
-     */
-    
-        
+            
     if( std::is_same< Device, Devices::Cuda >::value )
     {
 #ifdef HAVE_CUDA
@@ -102,7 +97,6 @@ initInterface( const MeshFunctionPointer& _input,
         CudaInitCaller<<< gridSize, blockSize >>>( _input.template getData< Device >(),
                                                    _output.template modifyData< Device >(),
                                                    _interfaceMap.template modifyData< Device >() );
-        //CudaInitCaller<<< gridSize, blockSize >>>( input );
         cudaDeviceSynchronize();
         TNL_CHECK_CUDA_DEVICE;
 #endif
@@ -215,6 +209,7 @@ template< typename Real,
           typename Device,
           typename Index >
    template< typename MeshEntity >
+__cuda_callable__
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
 updateCell( MeshFunctionType& u,
@@ -301,162 +296,187 @@ initInterface( const MeshFunctionPointer& _input,
                MeshFunctionPointer& _output,
                InterfaceMapPointer& _interfaceMap  )
 {
-   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 ? TypeInfo< RealType >::getMaxValue() :
-                                   - TypeInfo< RealType >::getMaxValue();
-                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 )
-                   {
-                   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
-         }
+    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;
+#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 ? TypeInfo< RealType >::getMaxValue() :
+                                        - TypeInfo< RealType >::getMaxValue();
+                     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 )
+                        {
+                        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 >
+__cuda_callable__
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >::
 updateCell( MeshFunctionType& u,
@@ -628,12 +648,13 @@ __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2,
         Cell cell( mesh );
         cell.getCoordinates().x() = i; cell.getCoordinates().y() = j;
         cell.refresh();
+        const Index cind = cell.getIndex();
 
 
-        output[ cell.getIndex() ] =
+        output[ cind ] =
                input( cell ) >= 0 ? TypeInfo< Real >::getMaxValue() :
                                     - TypeInfo< Real >::getMaxValue();
-        interfaceMap[ cell.getIndex() ] = false; 
+        interfaceMap[ cind ] = false; 
 
         const Real& hx = mesh.getSpaceSteps().x();
         const Real& hy = mesh.getSpaceSteps().y();
@@ -647,48 +668,131 @@ __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2,
            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[ cell.getIndex() ] ) > TNL::abs( pom ) ) 
-                   output[ cell.getIndex() ] = pom;
+               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[ cell.getIndex() ] ) > TNL::abs( pom ) )
-                   output[ cell.getIndex() ] = pom;                       
+               if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) )
+                   output[ cind ] = pom;                       
 
-               interfaceMap[ cell.getIndex() ] = true;
+               interfaceMap[ cind ] = true;
            }
            if( c * input[ w ] <= 0 )
            {
                pom = TNL::sign( c )*( hx * c )/( c - input[ w ]);
-               if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) ) 
-                   output[ cell.getIndex() ] = pom;
+               if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) ) 
+                   output[ cind ] = pom;
 
-               interfaceMap[ cell.getIndex() ] = true;
+               interfaceMap[ cind ] = true;
            }
            if( c * input[ s ] <= 0 )
            {
                pom = TNL::sign( c )*( hy * c )/( c - input[ s ]);
-               if( TNL::abs( output[ cell.getIndex() ] ) > TNL::abs( pom ) ) 
-                   output[ cell.getIndex() ] = pom;
+               if( TNL::abs( output[ cind ] ) > TNL::abs( pom ) ) 
+                   output[ cind ] = pom;
 
-               interfaceMap[ cell.getIndex() ] = true;
+               interfaceMap[ cind ] = true;
            }
         }
     }
 }
 
-
-/*__global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, double, TNL::Devices::Cuda, int > >& input )
+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 )
 {
     int i = threadIdx.x + blockDim.x*blockIdx.x;
     int j = blockDim.y*blockIdx.y + threadIdx.y;
-    //const Meshes::Grid< 2, double, TNL::Devices::Cuda, int >& mesh = input.getMesh();
+    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 ? TypeInfo< Real >::getMaxValue() :
+                                    - TypeInfo< Real >::getMaxValue();
+        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;
+           }
+        }
+    }
+}
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 48f330132..e44aaca87 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
@@ -15,6 +15,7 @@
 #include <TNL/SharedPointer.h>
 #include "tnlDirectEikonalMethodsBase.h"
 
+
 template< typename Mesh,
           typename Anisotropy = Functions::Analytic::Constant< Mesh::getMeshDimension(), typename Mesh::RealType > >
 class FastSweepingMethod
@@ -145,6 +146,7 @@ class FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >
       const IndexType maxIterations;
 };
 
+
 #include "tnlFastSweepingMethod1D_impl.h"
 #include "tnlFastSweepingMethod2D_impl.h"
 #include "tnlFastSweepingMethod3D_impl.h"
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 e36224d07..73e170895 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
@@ -14,6 +14,9 @@
 #pragma once
 
 #include "tnlFastSweepingMethod.h"
+#include <TNL/TypeInfo.h>
+#include <TNL/Devices/Cuda.h>
+
 
 template< typename Real,
           typename Device,
@@ -212,15 +215,23 @@ solve( const MeshPointer& mesh,
       {
          // TODO: CUDA code
 #ifdef HAVE_CUDA
-          /*int numBlocks = 2;
-          int threadsPerBlock;
-          if( mesh->getDimensions().x() >= mesh->getDimensions().y() )
-               threadsPerBlock = (int)( mesh->getDimensions().x() );
-          else
-               threadsPerBlock = (int)( mesh->getDimensions().y() );
+          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();
+          int DIM = mesh->getDimensions().x();
           
-          CudaUpdateCellCaller< Real, Device, Index ><<< numBlocks, threadsPerBlock >>>( interfaceMap, aux );
-          cudaDeviceSynchronize(); //copak dela?*/
+          tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
+          for( int k = 0; k < numBlocksX; k++)
+            CudaUpdateCellCaller< Real, Device, Index ><<< gridSize, blockSize >>>( ptr,
+                                                                                    interfaceMapPtr.template getData< Device >(),
+                                                                                    auxPtr.template modifyData< Device>() );
+          cudaDeviceSynchronize();
+          TNL_CHECK_CUDA_DEVICE;
+          aux = *auxPtr;
+          interfaceMap = *interfaceMapPtr;
 #endif
       }
       iteration++;
@@ -228,30 +239,30 @@ solve( const MeshPointer& mesh,
    aux.save("aux-final.tnl");
 }
 
+//#ifdef HAVE_CUDA
 template < typename Real, typename Device, typename Index >
-__global__ void CudaUpdateCellCaller( Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
+__global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr,
+                                      const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux )
 {
     int i = threadIdx.x + blockDim.x*blockIdx.x;
-    int j = threadIdx.y + blockDim.y*blockIdx.y;
-    const Meshes::Grid< 2, Real, Device, Index >& mesh = aux.getMesh();
+    int j = blockDim.y*blockIdx.y + threadIdx.y;
+    const Meshes::Grid< 2, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
     
     if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
     {
-        //make cell of aux from index
         typedef typename Meshes::Grid< 2, Real, Device, Index >::Cell Cell;
         Cell cell( mesh );
         cell.getCoordinates().x() = i; cell.getCoordinates().y() = j;
         cell.refresh();
-
-        //update cell value few times 
-        //for( int i = 0; i < mesh.getDimensions() ; i++ )
-        //{
-            cell.refresh();
+        //tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
+        for( int k = 0; k < 16; k++ )
+        {
             if( ! interfaceMap( cell ) )
             {
-        //        tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::updateCell( aux, cell );
+               ptr.updateCell( aux, cell );
             }
-        //}
+        }
     }
 }
+//#endif
\ No newline at end of file
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 1fd8c9f2a..3e8cb7bb1 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
@@ -21,7 +21,7 @@ template< typename Real,
           typename Anisotropy >
 FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >::
 FastSweepingMethod()
-: maxIterations( 1 )
+: maxIterations( 2 )
 {
    
 }
@@ -64,6 +64,7 @@ solve( const MeshPointer& mesh,
    interfaceMapPtr->setMesh( mesh );
    std::cout << "Initiating the interface cells ..." << std::endl;
    BaseType::initInterface( u, auxPtr, interfaceMapPtr );
+   cudaDeviceSynchronize();
    auxPtr->save( "aux-ini.tnl" );   
    
    typename MeshType::Cell cell( *mesh );
@@ -71,172 +72,201 @@ solve( const MeshPointer& mesh,
    IndexType iteration( 0 );
    MeshFunctionType aux = *auxPtr;
    InterfaceMapType interfaceMap = * interfaceMapPtr;
-   while( iteration < this->maxIterations )
-   {
-      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" );
+    while( iteration < this->maxIterations )
+    {
+        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() = 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() = 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() = 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().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 )
       {
-         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 );
-            }
-         }
+         // 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 );
+          Devices::Cuda::synchronizeDevice();
+          
+          tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr;
+          for( int k = 0; k < numBlocksX; k++)
+          CudaUpdateCellCaller< Real, Device, Index ><<< gridSize, blockSize >>>( ptr,
+                                                                                  interfaceMapPtr.template getData< Device >(),
+                                                                                  auxPtr.template modifyData< Device>() );
+          cudaDeviceSynchronize();
+          TNL_CHECK_CUDA_DEVICE;
+          aux = *auxPtr;
+          interfaceMap = *interfaceMapPtr;
+#endif
       }
+        
       //aux.save( "aux-8.tnl" );
       iteration++;
       
@@ -244,3 +274,29 @@ solve( const MeshPointer& mesh,
    aux.save("aux-final.tnl");
 }
 
+template < 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 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 = interfaceMap.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();
+        //tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr;
+        for( int l = 0; l < 8; l++ )
+        {
+            if( ! interfaceMap( cell ) )
+            {
+               ptr.updateCell( aux, cell );
+            }
+        }
+    }
+}
\ No newline at end of file
-- 
GitLab


From 9c196197633cd9eeee8fbe3ceeba7aee28a33bf0 Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Wed, 9 May 2018 19:32:45 +0200
Subject: [PATCH 16/28] Implementation of Blocks 18 * 18 using shared memory in
 2D.

---
 .../tnlDirectEikonalMethodsBase.h             |  12 +-
 .../tnlDirectEikonalMethodsBase_impl.h        | 231 +++++++++++++++++-
 .../hamilton-jacobi/tnlFastSweepingMethod.h   |   1 -
 .../tnlFastSweepingMethod2D_impl.h            |  47 +++-
 .../tnlFastSweepingMethod3D_impl.h            |   6 +-
 5 files changed, 277 insertions(+), 20 deletions(-)

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 138671a3b..69fce1467 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -67,7 +67,17 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
       template< typename MeshEntity >
       __cuda_callable__ void updateCell( MeshFunctionType& u,
                                          const MeshEntity& cell,
-                                         const RealType velocity = 1.0 );      
+                                         const RealType velocity = 1.0 );
+      
+      __cuda_callable__ void updateCell( Real sArray[18][18],
+                                         int thri, int thrj, const Real hx, const Real hy,
+                                         const Real velocity = 1.0 );
+      
+      __cuda_callable__ void setsArray( MeshFunctionType& aux, Real sArray[18][18],
+                                            int dimX, int dimY, int i, int j );
+      
+      /*__cuda_callable__ void getsArray( MeshFunctionType& aux, Real sArray[18][18],
+                                            int dimX, int dimY, int i, int j );*/
 };
 
 template< typename Real,
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 bf78a02a4..a89d230f6 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
@@ -218,7 +218,6 @@ updateCell( MeshFunctionType& u,
 {
    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 );
@@ -598,7 +597,7 @@ 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[])
+__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])){
@@ -796,3 +795,231 @@ __global__ void CudaInitCaller3d( const Functions::MeshFunction< Meshes::Grid< 3
         }
     }
 }
+
+template< typename Real,
+          typename Device,
+          typename Index >
+__cuda_callable__ void
+tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
+setsArray( MeshFunctionType& aux, Real sArray[18][18], int dimX, int dimY, int blockIdx, int blockIdy )
+{
+    int numOfBlockx = dimX/16 + ((dimX%16 != 0) ? 1:0);
+    int numOfBlocky = dimY/16 + ((dimY%16 != 0) ? 1:0);
+    int xkolik = 18;
+    int ykolik = 18;
+    int xOd = 0;
+    int yOd = 0;
+    
+    if( blockIdx == 0 )
+        xOd = 1;
+    if( blockIdy == 0 )
+        yOd = 1;
+    
+    if( numOfBlockx - 1 == blockIdx )
+        xkolik = dimX - (numOfBlockx-1)*16+1;
+
+    if( numOfBlocky -1 == blockIdy )
+        ykolik = dimY - (numOfBlocky-1)*16+1;
+    
+    if( dimX > (blockIdx+1) * 16 )
+    {
+        for( int i = yOd; i < ykolik; i++ )
+        {
+            sArray[i][17] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + i*dimX + 17 ];
+        }
+    }
+    if( blockIdx != 0 )
+    {
+        for( int i = yOd; i < ykolik; i++ )
+        {
+            sArray[i][0] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + i*dimX + 0];
+        }
+    } 
+    if( dimY > (blockIdy+1) * 16 )
+    {
+        for( int i = xOd; i < xkolik; i++ )
+        {
+            sArray[17][i] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + 17*dimX + i ];
+        }
+    }
+    if( blockIdy != 0 )
+    {
+        for( int i = xOd; i < xkolik; i++ )
+        {
+            sArray[0][i] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + 0*dimX + i ];
+        }
+    }
+    
+    /*int numOfBlockx = dimX/16 + ((dimX%16 != 0) ? 1:0);
+    int numOfBlocky = dimY/16 + ((dimY%16 != 0) ? 1:0);
+    int xkolik = 18;
+    int ykolik = 18;
+
+    if( numOfBlockx - 1 == blockIdx )
+        xkolik = dimX - (numOfBlockx-1)*16+1;
+
+    if( numOfBlocky -1 == blockIdy )
+        ykolik = dimY - (numOfBlocky-1)*16+1;
+
+
+    if( numOfBlockx == 0 || numOfBlocky == 0 )
+        return;
+    else
+    {
+        if( blockIdx == 0 )
+        {
+            if( blockIdy == 0 )
+            {
+                for( int j = 0; j < ykolik-1; j++ )
+                  for( int i = 0; i < xkolik-1; i++ )
+                    {
+                      sArray[ j+1 ][ i+1 ] = aux[ j*dimX + i ];
+                    }
+
+                //vypis( *sArray );
+            }
+            else
+            {
+                for( int j = 0; j < ykolik; j++ )
+                  for( int i = 0; i < xkolik-1; i++ )
+                    {
+                      sArray[ j ][ i+1 ] = aux[ blockIdy*16*dimX - dimX + j*dimX + i ];
+                    }
+            }
+        }
+        else if( blockIdy == 0 )
+        {
+            for( int j = 0; j < ykolik-1; j++ )
+              for( int i = 0; i < xkolik; i++ )
+                 {
+                   sArray[ j+1 ][ i ] = aux[ blockIdx*16 - 1 + j*dimX + i ];
+                 }
+        }
+        else
+        {
+           for( int j = 0; j < ykolik; j++ )
+              for( int i = 0; i < xkolik; i++ )
+                {
+                  //if( dimX*dimY-1 >  blockIdy*16*dimX - dimX + blockIdx*16 - 1 + j*dimX + i )
+                    sArray[ j ][ i ] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + j*dimX + i ];
+                }
+        }
+    }*/
+}
+
+/*template< typename Real,
+          typename Device,
+          typename Index >
+__cuda_callable__ void
+tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
+getsArray( MeshFunctionType& aux, Real sArray[18][18], int dimX, int dimY, int blockIdx, int blockIdy )
+{
+    int numOfBlockx = dimX/16 + ((dimX%16 != 0) ? 1:0);
+    int numOfBlocky = dimY/16 + ((dimY%16 != 0) ? 1:0);
+    int xkolik = 18;
+    int ykolik = 18;
+    
+    if( numOfBlockx - 1 == blockIdx )
+        xkolik = dimX - (numOfBlockx-1)*16+2;
+
+    if( numOfBlocky -1 == blockIdy )
+        ykolik = dimY - (numOfBlocky-1)*16+2;
+
+    if( numOfBlockx == 0 || numOfBlocky == 0 )
+        return;
+    else
+    {
+        if( blockIdx == 0 )
+        {
+            if( blockIdy == 0 )
+            {
+               for( int j = 0; j < ykolik-2; j++ )
+                  for( int i = 0; i < xkolik-2; i++ )
+                    {
+                      aux[ j*dimX + i ] = sArray[ j+1 ][ i+1 ];
+                    }
+            }
+            else
+            {
+               for( int j = 1; j < ykolik-1; j++ )
+                  for( int i = 0; i < xkolik-2; i++ )
+                    {
+                      aux[ blockIdy*16*dimX - dimX + j*dimX + i ] = sArray[ j ][ i+1 ];
+                    }
+            }
+        }
+        else if( blockIdy == 0 )
+        {
+            for( int j = 0; j < ykolik-2; j++ )
+               for( int i = 1; i < xkolik-1; i++ )
+                 {
+                   aux[ blockIdx*16 - 1 + j*dimX + i ] = sArray[ j+1 ][ i ];
+                 }
+        }
+        else
+        {
+            for( int j = 1; j < ykolik-1; j++ )    
+              for( int i = 1; i < xkolik-1; i++ )
+                {
+                    aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + j*dimX + i ] = sArray[ j ][ i ];
+                }
+        }
+    }
+}*/
+
+template< typename Real,
+          typename Device,
+          typename Index >
+__cuda_callable__
+void
+tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
+updateCell( Real sArray[18][18], int thri, int thrj, const Real hx, const Real hy,
+            const Real v )
+{
+   
+   //const Meshes::Grid< 2, Real, Device, Index >& mesh = u.template getMesh< Devices::Cuda >();
+   /*typedef typename Meshes::Grid< 2, Real, Device, Index >::Cell Cell;
+   Cell cell( mesh );
+   cell.getCoordinates().x() = i; cell.getCoordinates().y() = j;
+   cell.refresh();
+   
+   const auto& neighborEntities = cell.template getNeighborEntities< 2 >();
+   const RealType& hx = mesh.getSpaceSteps().x();
+   const RealType& hy = mesh.getSpaceSteps().y();*/
+   const RealType value = sArray[ thrj ][ thri ];//u( cell );
+   RealType a, b, tmp = TypeInfo< RealType >::getMaxValue();
+   
+   /*if( value != u[ cell.getIndex() ] )
+       return;*/
+   
+   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 ) == TypeInfo< RealType >::getMaxValue() && 
+        fabs( b ) == TypeInfo< RealType >::getMaxValue() )
+       return;
+   
+    RealType pom[6] = { a, b, TypeInfo< Real >::getMaxValue(), (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 );
+        sArray[ thrj ][ thri ] = 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 );
+        sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
+    }
+}
\ No newline at end of file
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 e44aaca87..54e577dac 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod.h
@@ -99,7 +99,6 @@ class FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >
                   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 73e170895..80976772c 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
@@ -221,11 +221,12 @@ solve( const MeshPointer& mesh,
           dim3 blockSize( cudaBlockSize, cudaBlockSize );
           dim3 gridSize( numBlocksX, numBlocksY );
           Devices::Cuda::synchronizeDevice();
-          int DIM = mesh->getDimensions().x();
           
           tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
-          for( int k = 0; k < numBlocksX; k++)
-            CudaUpdateCellCaller< Real, Device, Index ><<< gridSize, blockSize >>>( ptr,
+          int nBlockIter = numBlocksX > numBlocksY ? numBlocksX : numBlocksY;
+          nBlockIter = numBlocksX == numBlocksY ? nBlockIter + 1 : nBlockIter;
+          for( int k = 0; k < nBlockIter; k++)
+            CudaUpdateCellCaller<<< gridSize, blockSize, 18 * 18 * sizeof( Real ) >>>( ptr,
                                                                                     interfaceMapPtr.template getData< Device >(),
                                                                                     auxPtr.template modifyData< Device>() );
           cudaDeviceSynchronize();
@@ -245,24 +246,44 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
                                       const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index >, 2, bool >& interfaceMap,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux )
 {
-    int i = threadIdx.x + blockDim.x*blockIdx.x;
-    int j = blockDim.y*blockIdx.y + threadIdx.y;
+    int thri = threadIdx.x; int thrj = threadIdx.y; // nelze ke stejnym pristupovat znovu pres threadIdx (vzdy da jine hodnoty)
+    int blIdx = blockIdx.x; int blIdy = blockIdx.y;
+    int i = thri + blockDim.x*blIdx;
+    int j = blockDim.y*blIdy + thrj;
     const Meshes::Grid< 2, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
+    const Real hx = mesh.getSpaceSteps().x();
+    const Real hy = mesh.getSpaceSteps().y();
+    
+    __shared__ Real sArray[ 18 ][ 18 ];
+    for( int k = 0; k < 18; k++ )
+        for( int l = 0; l < 18; l++ )
+            sArray[ k ][ l ] = TypeInfo< Real >::getMaxValue();
+    __syncthreads();
+    /*//filling shared array
+    ptr.setsArray( aux, sArray, mesh.getDimensions().x(), mesh.getDimensions().y(), blIdx, blIdy );
+    __syncthreads();*/
     
     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();
-        //tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
-        for( int k = 0; k < 16; k++ )
+        sArray[thrj+1][thri+1] = aux[ j*mesh.getDimensions().x() + i ];
+        ptr.setsArray( aux, sArray, mesh.getDimensions().x(), mesh.getDimensions().y(), blIdx, blIdy ); //fill edges of sArray
+        __syncthreads();
+            
+        
+        if( ! interfaceMap[ j*mesh.getDimensions().x() + i ] ) 
         {
-            if( ! interfaceMap( cell ) )
+            for( int k = 0; k < 17; k++ )
             {
-               ptr.updateCell( aux, cell );
+                ptr.updateCell( sArray, thri+1, thrj+1, hx, hy );
+                __syncthreads();
             }
         }
+        /*for( int k = 0; k < mesh.getDimensions().x(); k++)
+            for( int l = 0; l < mesh.getDimensions().y(); l++)
+                aux[ k*mesh.getDimensions().x() + l ] = TypeInfo< Real >::getMaxValue();*/
+        aux[j*mesh.getDimensions().x() + i] = sArray[thrj+1][thri+1];
+        __syncthreads();
     }
+    //ptr.getsArray( aux, sArray, mesh.getDimensions().x(), mesh.getDimensions().y(), blIdx, blIdy );
 }
 //#endif
\ No newline at end of file
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 3e8cb7bb1..5ddd3bdeb 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
@@ -21,7 +21,7 @@ template< typename Real,
           typename Anisotropy >
 FastSweepingMethod< Meshes::Grid< 3, Real, Device, Index >, Anisotropy >::
 FastSweepingMethod()
-: maxIterations( 2 )
+: maxIterations( 1 )
 {
    
 }
@@ -257,7 +257,7 @@ solve( const MeshPointer& mesh,
           
           tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr;
           for( int k = 0; k < numBlocksX; k++)
-          CudaUpdateCellCaller< Real, Device, Index ><<< gridSize, blockSize >>>( ptr,
+          CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
                                                                                   interfaceMapPtr.template getData< Device >(),
                                                                                   auxPtr.template modifyData< Device>() );
           cudaDeviceSynchronize();
@@ -291,7 +291,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         cell.getCoordinates().x() = i; cell.getCoordinates().y() = j; cell.getCoordinates().z() = k;
         cell.refresh();
         //tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr;
-        for( int l = 0; l < 8; l++ )
+        for( int l = 0; l < 10; l++ )
         {
             if( ! interfaceMap( cell ) )
             {
-- 
GitLab


From 67997f67ce75ea053ace967a3c556c768e08cb45 Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Thu, 10 May 2018 21:26:48 +0200
Subject: [PATCH 17/28] trying git problem

---
 .../tnlDirectEikonalMethodsBase_impl.h        | 334 +++++++++---------
 .../tnlFastSweepingMethod2D_impl.h            |   6 +-
 .../tnlFastSweepingMethod3D_impl.h            |   6 +-
 3 files changed, 178 insertions(+), 168 deletions(-)

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 a89d230f6..e52f2fe1c 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
@@ -632,170 +632,6 @@ __cuda_callable__ void sortMinims( T1 pom[] )
     }   
 }
 
-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 ) 
-{
-    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();
-
-
-        output[ cind ] =
-               input( cell ) >= 0 ? TypeInfo< Real >::getMaxValue() :
-                                    - TypeInfo< Real >::getMaxValue();
-        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;
-           }
-        }
-    }
-}
-
-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 )
-{
-    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 ? TypeInfo< Real >::getMaxValue() :
-                                    - TypeInfo< Real >::getMaxValue();
-        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;
-           }
-        }
-    }
-}
-
 template< typename Real,
           typename Device,
           typename Index >
@@ -967,6 +803,173 @@ getsArray( MeshFunctionType& aux, Real sArray[18][18], int dimX, int dimY, int b
     }
 }*/
 
+
+#ifdef HAVE_CUDA
+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 ) 
+{
+    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();
+
+
+        output[ cind ] =
+               input( cell ) >= 0 ? TypeInfo< Real >::getMaxValue() :
+                                    - TypeInfo< Real >::getMaxValue();
+        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;
+           }
+        }
+    }
+}
+
+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 )
+{
+    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 ? TypeInfo< Real >::getMaxValue() :
+                                    - TypeInfo< Real >::getMaxValue();
+        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;
+           }
+        }
+    }
+}
+
+
 template< typename Real,
           typename Device,
           typename Index >
@@ -1022,4 +1025,5 @@ updateCell( Real sArray[18][18], int thri, int thrj, const Real hx, const Real h
         //u[ cell.getIndex() ]= argAbsMin( value, tmp );
         sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
     }
-}
\ No newline at end of file
+}
+#endif
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 80976772c..1279e3437 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
@@ -67,7 +67,9 @@ solve( const MeshPointer& 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" );
 
@@ -240,7 +242,7 @@ solve( const MeshPointer& mesh,
    aux.save("aux-final.tnl");
 }
 
-//#ifdef HAVE_CUDA
+#ifdef HAVE_CUDA
 template < 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,
@@ -286,4 +288,4 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
     }
     //ptr.getsArray( aux, sArray, mesh.getDimensions().x(), mesh.getDimensions().y(), blIdx, blIdy );
 }
-//#endif
\ No newline at end of file
+#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 5ddd3bdeb..e9fa72e66 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
@@ -64,7 +64,9 @@ solve( const MeshPointer& 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 );
@@ -274,6 +276,7 @@ solve( const MeshPointer& mesh,
    aux.save("aux-final.tnl");
 }
 
+#ifdef HAVE_CUDA
 template < 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,
@@ -299,4 +302,5 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
             }
         }
     }
-}
\ No newline at end of file
+}
+#endif
-- 
GitLab


From 8a8b083e31f59bce56fefdff27375b7c04027ab8 Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Thu, 17 May 2018 08:54:40 +0200
Subject: [PATCH 18/28] pyramid reduction in 2D with loopCounter

---
 .../tnlDirectEikonalMethodsBase.h             |   5 +-
 .../tnlDirectEikonalMethodsBase_impl.h        | 141 ++------------
 .../tnlFastSweepingMethod2D_impl.h            | 184 +++++++++++++++---
 3 files changed, 177 insertions(+), 153 deletions(-)

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 69fce1467..979c70e7b 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -69,7 +69,7 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
                                          const MeshEntity& cell,
                                          const RealType velocity = 1.0 );
       
-      __cuda_callable__ void updateCell( Real sArray[18][18],
+      __cuda_callable__ bool updateCell( volatile Real sArray[18][18],
                                          int thri, int thrj, const Real hx, const Real hy,
                                          const Real velocity = 1.0 );
       
@@ -116,7 +116,8 @@ __cuda_callable__ void sortMinims( T1 pom[] );
 template < 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,
-                                      Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux );
+                                      Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
+                                      bool *BlockIterDevice );
 
 template < typename Real, typename Device, typename Index >
 __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& input, 
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 e52f2fe1c..a3af357c9 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
@@ -684,124 +684,9 @@ setsArray( MeshFunctionType& aux, Real sArray[18][18], int dimX, int dimY, int b
         {
             sArray[0][i] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + 0*dimX + i ];
         }
-    }
-    
-    /*int numOfBlockx = dimX/16 + ((dimX%16 != 0) ? 1:0);
-    int numOfBlocky = dimY/16 + ((dimY%16 != 0) ? 1:0);
-    int xkolik = 18;
-    int ykolik = 18;
-
-    if( numOfBlockx - 1 == blockIdx )
-        xkolik = dimX - (numOfBlockx-1)*16+1;
-
-    if( numOfBlocky -1 == blockIdy )
-        ykolik = dimY - (numOfBlocky-1)*16+1;
-
-
-    if( numOfBlockx == 0 || numOfBlocky == 0 )
-        return;
-    else
-    {
-        if( blockIdx == 0 )
-        {
-            if( blockIdy == 0 )
-            {
-                for( int j = 0; j < ykolik-1; j++ )
-                  for( int i = 0; i < xkolik-1; i++ )
-                    {
-                      sArray[ j+1 ][ i+1 ] = aux[ j*dimX + i ];
-                    }
-
-                //vypis( *sArray );
-            }
-            else
-            {
-                for( int j = 0; j < ykolik; j++ )
-                  for( int i = 0; i < xkolik-1; i++ )
-                    {
-                      sArray[ j ][ i+1 ] = aux[ blockIdy*16*dimX - dimX + j*dimX + i ];
-                    }
-            }
-        }
-        else if( blockIdy == 0 )
-        {
-            for( int j = 0; j < ykolik-1; j++ )
-              for( int i = 0; i < xkolik; i++ )
-                 {
-                   sArray[ j+1 ][ i ] = aux[ blockIdx*16 - 1 + j*dimX + i ];
-                 }
-        }
-        else
-        {
-           for( int j = 0; j < ykolik; j++ )
-              for( int i = 0; i < xkolik; i++ )
-                {
-                  //if( dimX*dimY-1 >  blockIdy*16*dimX - dimX + blockIdx*16 - 1 + j*dimX + i )
-                    sArray[ j ][ i ] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + j*dimX + i ];
-                }
-        }
-    }*/
+    }   
 }
 
-/*template< typename Real,
-          typename Device,
-          typename Index >
-__cuda_callable__ void
-tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
-getsArray( MeshFunctionType& aux, Real sArray[18][18], int dimX, int dimY, int blockIdx, int blockIdy )
-{
-    int numOfBlockx = dimX/16 + ((dimX%16 != 0) ? 1:0);
-    int numOfBlocky = dimY/16 + ((dimY%16 != 0) ? 1:0);
-    int xkolik = 18;
-    int ykolik = 18;
-    
-    if( numOfBlockx - 1 == blockIdx )
-        xkolik = dimX - (numOfBlockx-1)*16+2;
-
-    if( numOfBlocky -1 == blockIdy )
-        ykolik = dimY - (numOfBlocky-1)*16+2;
-
-    if( numOfBlockx == 0 || numOfBlocky == 0 )
-        return;
-    else
-    {
-        if( blockIdx == 0 )
-        {
-            if( blockIdy == 0 )
-            {
-               for( int j = 0; j < ykolik-2; j++ )
-                  for( int i = 0; i < xkolik-2; i++ )
-                    {
-                      aux[ j*dimX + i ] = sArray[ j+1 ][ i+1 ];
-                    }
-            }
-            else
-            {
-               for( int j = 1; j < ykolik-1; j++ )
-                  for( int i = 0; i < xkolik-2; i++ )
-                    {
-                      aux[ blockIdy*16*dimX - dimX + j*dimX + i ] = sArray[ j ][ i+1 ];
-                    }
-            }
-        }
-        else if( blockIdy == 0 )
-        {
-            for( int j = 0; j < ykolik-2; j++ )
-               for( int i = 1; i < xkolik-1; i++ )
-                 {
-                   aux[ blockIdx*16 - 1 + j*dimX + i ] = sArray[ j+1 ][ i ];
-                 }
-        }
-        else
-        {
-            for( int j = 1; j < ykolik-1; j++ )    
-              for( int i = 1; i < xkolik-1; i++ )
-                {
-                    aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + j*dimX + i ] = sArray[ j ][ i ];
-                }
-        }
-    }
-}*/
 
 
 #ifdef HAVE_CUDA
@@ -974,9 +859,9 @@ template< typename Real,
           typename Device,
           typename Index >
 __cuda_callable__
-void
+bool
 tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
-updateCell( Real sArray[18][18], int thri, int thrj, const Real hx, const Real hy,
+updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, const Real hy,
             const Real v )
 {
    
@@ -991,39 +876,39 @@ updateCell( Real sArray[18][18], int thri, int thrj, const Real hx, const Real h
    const RealType& hy = mesh.getSpaceSteps().y();*/
    const RealType value = sArray[ thrj ][ thri ];//u( cell );
    RealType a, b, tmp = TypeInfo< RealType >::getMaxValue();
-   
-   /*if( value != u[ cell.getIndex() ] )
-       return;*/
-   
+      
    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 ) == TypeInfo< RealType >::getMaxValue() && 
         fabs( b ) == TypeInfo< RealType >::getMaxValue() )
-       return;
+       return false;
    
-    RealType pom[6] = { a, b, TypeInfo< Real >::getMaxValue(), (RealType)hx, (RealType)hy, 0.0 };
+    RealType pom[6] = { a, b, TypeInfo< Real >::getMaxValue(), (RealType)hx, (RealType)hy, TypeInfo< Real >::getMaxValue() };
     sortMinims( pom );
     tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]/v;
     
                                 
     if( fabs( tmp ) < fabs( pom[ 1 ] ) ) 
     {
-        //u[ cell.getIndex() ]= argAbsMin( value, tmp );
         sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
+        if ( fabs( value - sArray[ thrj ][ thri ] ) >  0.001 )
+            return true;
+        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 ] );
-        //u[ cell.getIndex() ]= argAbsMin( value, tmp );
         sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
+        if ( fabs( value - sArray[ thrj ][ thri ] ) > 0.001 )
+            return true;
+        return false;
     }
+    return false;
 }
 #endif
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 1279e3437..d4cb1f63b 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
@@ -227,12 +227,47 @@ solve( const MeshPointer& mesh,
           tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
           int nBlockIter = numBlocksX > numBlocksY ? numBlocksX : numBlocksY;
           nBlockIter = numBlocksX == numBlocksY ? nBlockIter + 1 : nBlockIter;
-          for( int k = 0; k < nBlockIter; k++)
+          
+          bool isNotDone = true;
+          bool* BlockIter = (bool*)malloc( ( numBlocksX * numBlocksY ) * sizeof( bool ) );
+          
+          bool *BlockIterDevice;
+          cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ) );
+          
+          while( isNotDone )
+          {
+           for( int i = 0; i < numBlocksX * numBlocksY; i++ )
+                BlockIter[ i ] = false;   
+           
+            isNotDone = false;
+            
             CudaUpdateCellCaller<<< gridSize, blockSize, 18 * 18 * sizeof( Real ) >>>( ptr,
                                                                                     interfaceMapPtr.template getData< Device >(),
-                                                                                    auxPtr.template modifyData< Device>() );
+                                                                                    auxPtr.template modifyData< Device>(),
+                                                                                    BlockIterDevice );
+            cudaMemcpy(BlockIter, BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ), cudaMemcpyDeviceToHost);
+            for( int i = 0; i < numBlocksX; i++ )
+            {    for( int j = 0; j < numBlocksY; j++ )
+                {
+                    if( BlockIter[ j * numBlocksY + i ] )
+                        std::cout << "true." << "\t";
+                    else
+                        std::cout << "false." << "\t";    
+                }
+                std::cout << std::endl;
+            }
+            std::cout << std::endl;
+             
+            for( int i = 0; i < numBlocksX * numBlocksY; i++ )
+                isNotDone = isNotDone || BlockIter[ i ];
+            
+          }
+          delete[] BlockIter;
+          cudaFree( BlockIterDevice );
           cudaDeviceSynchronize();
+          
           TNL_CHECK_CUDA_DEVICE;
+              
           aux = *auxPtr;
           interfaceMap = *interfaceMapPtr;
 #endif
@@ -246,46 +281,149 @@ solve( const MeshPointer& mesh,
 template < 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,
-                                      Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux )
+                                      Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
+                                      bool *BlockIterDevice )
 {
     int thri = threadIdx.x; int thrj = threadIdx.y; // nelze ke stejnym pristupovat znovu pres threadIdx (vzdy da jine hodnoty)
     int blIdx = blockIdx.x; int blIdy = blockIdx.y;
     int i = thri + blockDim.x*blIdx;
     int j = blockDim.y*blIdy + thrj;
+    
+    __shared__ volatile bool changed[256];
+    changed[ thrj * blockDim.x + thri ] = false;
+     __shared__ volatile bool SmallCycleBoolin;
+    if( thrj == 0 && thri == 0 )
+        SmallCycleBoolin = true;
+    
+    
     const Meshes::Grid< 2, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
     const Real hx = mesh.getSpaceSteps().x();
     const Real hy = mesh.getSpaceSteps().y();
     
-    __shared__ Real sArray[ 18 ][ 18 ];
-    for( int k = 0; k < 18; k++ )
-        for( int l = 0; l < 18; l++ )
-            sArray[ k ][ l ] = TypeInfo< Real >::getMaxValue();
-    __syncthreads();
-    /*//filling shared array
-    ptr.setsArray( aux, sArray, mesh.getDimensions().x(), mesh.getDimensions().y(), blIdx, blIdy );
-    __syncthreads();*/
+    __shared__ volatile Real sArray[ 18 ][ 18 ];
+    sArray[thrj][thri] = TypeInfo< Real >::getMaxValue();
     
-    if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+    //filling sArray edges
+    int dimX = mesh.getDimensions().x(); int dimY = mesh.getDimensions().y();
+    int numOfBlockx = dimX/16 + ((dimX%16 != 0) ? 1:0);
+    int numOfBlocky = dimY/16 + ((dimY%16 != 0) ? 1:0);
+    int xkolik = 17;
+    int ykolik = 17;
+    
+    
+    if( numOfBlockx - 1 == blIdx )
+        xkolik = dimX - (blIdx)*16+1;
+
+    if( numOfBlocky -1 == blIdy )
+        ykolik = dimY - (blIdy)*16+1;
+       
+    
+    if( thri == 0 )
+    {        
+        if( dimX > (blIdx+1) * 16  && thrj+1 < ykolik )
+            sArray[thrj+1][xkolik] = aux[ blIdy*16*dimX - dimX + blIdx*16 - 1 + (thrj+1)*dimX + 17 ];
+        else
+            sArray[thrj+1][xkolik] = TypeInfo< Real >::getMaxValue();
+    }
+    
+    if( thri == 1 )
+    {
+        if( blIdx != 0 && thrj+1 < ykolik )
+            sArray[thrj+1][0] = aux[ blIdy*16*dimX - dimX + blIdx*16 - 1 + (thrj+1)*dimX + 0];
+        else
+            sArray[thrj+1][0] = TypeInfo< Real >::getMaxValue();
+    }
+    
+    if( thrj == 0 )
     {
+        if( dimY > (blIdy+1) * 16  && thri+1 < xkolik )
+            sArray[ykolik][thri+1] = aux[ blIdy*16*dimX - dimX + blIdx*16 - 1 + 17*dimX + thri+1 ];
+        else
+           sArray[ykolik][thri+1] = TypeInfo< Real >::getMaxValue();
+    }
+    
+    if( thrj == 1 )
+    {
+        if( blIdy != 0 && thri+1 < xkolik )
+            sArray[0][thri+1] = aux[ blIdy*16*dimX - dimX + blIdx*16 - 1 + 0*dimX + thri+1 ];
+        else
+            sArray[0][thri+1] = TypeInfo< Real >::getMaxValue();
+    }
+    
+    //filling BlockIterDevice
+    BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = false;
+    
+    
+    if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+    {    
         sArray[thrj+1][thri+1] = aux[ j*mesh.getDimensions().x() + i ];
-        ptr.setsArray( aux, sArray, mesh.getDimensions().x(), mesh.getDimensions().y(), blIdx, blIdy ); //fill edges of sArray
         __syncthreads();
-            
+    }    
+    
+    __shared__ int loopcounter;
+    if( thri == 0 && thrj == 0 )
+        loopcounter= 0;
+  //if( blIdx == 5 && blIdy == 4 )
+    while( SmallCycleBoolin )   
+    {
+        if( thri == 1 && thrj == 1 )
+            SmallCycleBoolin = false;
+        
+        changed[ thrj * 16 + thri ] = false;
+        __syncthreads();
         
-        if( ! interfaceMap[ j*mesh.getDimensions().x() + i ] ) 
+    //calculation of update cell
+        if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+        { 
+            if( ! interfaceMap[ j * mesh.getDimensions().x() + i ] )
+            {
+                changed[ thrj * 16 + thri ] = ptr.updateCell( sArray, thri+1, thrj+1, hx, hy );
+                __syncthreads();   
+            }
+        }
+        
+    //pyramid reduction
+        if( blockDim.x*blockDim.y >= 256 )
+        {
+            if( (thrj * 16 + thri) < 128 )
+            {
+                changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 128 ];
+            }
+        __syncthreads();
+        }
+        if( blockDim.x*blockDim.y >= 128 )
         {
-            for( int k = 0; k < 17; k++ )
+            if( (thrj * 16 + thri) < 64 )
             {
-                ptr.updateCell( sArray, thri+1, thrj+1, hx, hy );
-                __syncthreads();
+                changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 64 ];
             }
+        __syncthreads();
+        }
+        if( (thrj * 16 + thri) < 32 )
+        {
+            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 32 ];
+            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 16 ];
+            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 8 ];
+            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 4 ];
+            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 2 ];
+            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 1 ];
+        }
+        __syncthreads();
+
+        if( thrj == 1 && thri == 1 )
+        {
+            loopcounter++;
+            if( loopcounter > 1000 )
+                break;
+            SmallCycleBoolin = changed[ 0 ];
+            //if( SmallCycleBoolin )
+                BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = SmallCycleBoolin;
         }
-        /*for( int k = 0; k < mesh.getDimensions().x(); k++)
-            for( int l = 0; l < mesh.getDimensions().y(); l++)
-                aux[ k*mesh.getDimensions().x() + l ] = TypeInfo< Real >::getMaxValue();*/
-        aux[j*mesh.getDimensions().x() + i] = sArray[thrj+1][thri+1];
         __syncthreads();
     }
-    //ptr.getsArray( aux, sArray, mesh.getDimensions().x(), mesh.getDimensions().y(), blIdx, blIdy );
+        
+    if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
+        aux[ j * mesh.getDimensions().x() + i ] = sArray[ thrj + 1 ][ thri + 1 ];
+        
 }
 #endif
-- 
GitLab


From 02251467ef38f31508762d0019b2c0ba52ba3b69 Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Thu, 7 Jun 2018 08:49:09 +0200
Subject: [PATCH 19/28] debugging kernel for update cell, still not clear what
 is wrong.

---
 .../tnlDirectEikonalMethodsBase.h             |   4 +-
 .../tnlDirectEikonalMethodsBase_impl.h        |  25 +-
 .../tnlFastSweepingMethod2D_impl.h            | 222 +++++++++++-------
 3 files changed, 152 insertions(+), 99 deletions(-)

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 979c70e7b..fe83b61c0 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -69,7 +69,7 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
                                          const MeshEntity& cell,
                                          const RealType velocity = 1.0 );
       
-      __cuda_callable__ bool updateCell( volatile Real sArray[18][18],
+      __cuda_callable__ double updateCell( volatile Real sArray[18][18],
                                          int thri, int thrj, const Real hx, const Real hy,
                                          const Real velocity = 1.0 );
       
@@ -117,7 +117,7 @@ template < 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,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
-                                      bool *BlockIterDevice );
+                                      Real *BlockIterDevice );
 
 template < typename Real, typename Device, typename Index >
 __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& input, 
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 a3af357c9..098dace4b 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
@@ -859,7 +859,7 @@ template< typename Real,
           typename Device,
           typename Index >
 __cuda_callable__
-bool
+double
 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 )
@@ -885,9 +885,9 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
 
     if( fabs( a ) == TypeInfo< RealType >::getMaxValue() && 
         fabs( b ) == TypeInfo< RealType >::getMaxValue() )
-       return false;
+       return 0;
    
-    RealType pom[6] = { a, b, TypeInfo< Real >::getMaxValue(), (RealType)hx, (RealType)hy, TypeInfo< Real >::getMaxValue() };
+    RealType pom[6] = { a, b, TypeInfo< Real >::getMaxValue(), (RealType)hx, (RealType)hy, 0.0 };
     sortMinims( pom );
     tmp = pom[ 0 ] + TNL::sign( value ) * pom[ 3 ]/v;
     
@@ -895,9 +895,11 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
     if( fabs( tmp ) < fabs( pom[ 1 ] ) ) 
     {
         sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
-        if ( fabs( value - sArray[ thrj ][ thri ] ) >  0.001 )
-            return true;
-        return false;
+        tmp = value - sArray[ thrj ][ thri ];
+        if ( fabs( tmp ) >  0.1 )
+            return 10;
+        else
+            return 0;
     }
     else
     {
@@ -905,10 +907,13 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
             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 );
-        if ( fabs( value - sArray[ thrj ][ thri ] ) > 0.001 )
-            return true;
-        return false;
+        tmp = value - sArray[ thrj ][ thri ];
+        if ( fabs( tmp ) > 0.1 )
+            return 10;
+        else
+            return 0;
     }
-    return false;
+    
+    return 0;
 }
 #endif
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 d4cb1f63b..e24201936 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
@@ -225,39 +225,46 @@ solve( const MeshPointer& mesh,
           Devices::Cuda::synchronizeDevice();
           
           tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
-          int nBlockIter = numBlocksX > numBlocksY ? numBlocksX : numBlocksY;
-          nBlockIter = numBlocksX == numBlocksY ? nBlockIter + 1 : nBlockIter;
           
-          bool isNotDone = true;
-          bool* BlockIter = (bool*)malloc( ( numBlocksX * numBlocksY ) * sizeof( bool ) );
           
-          bool *BlockIterDevice;
-          cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ) );
+          /*int nBlockIter = numBlocksX > numBlocksY ? numBlocksX : numBlocksY;
+          nBlockIter = numBlocksX == numBlocksY ? nBlockIter + 1 : nBlockIter;*/
           
-          while( isNotDone )
+          bool isNotDone = true; numBlocksX = 16; numBlocksY = 16;
+          double* BlockIter = (double*)malloc( ( numBlocksX * numBlocksY ) * sizeof( double ) );
+          
+          double *BlockIterDevice;
+          cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( double ) );
+          
+          //while( isNotDone )
           {
            for( int i = 0; i < numBlocksX * numBlocksY; i++ )
-                BlockIter[ i ] = false;   
+                BlockIter[ i ] = 0.0;
+           cudaMemcpy(BlockIterDevice, BlockIter, ( numBlocksX * numBlocksY ) * sizeof( double ), cudaMemcpyHostToDevice);
            
             isNotDone = false;
             
-            CudaUpdateCellCaller<<< gridSize, blockSize, 18 * 18 * sizeof( Real ) >>>( ptr,
-                                                                                    interfaceMapPtr.template getData< Device >(),
-                                                                                    auxPtr.template modifyData< Device>(),
-                                                                                    BlockIterDevice );
-            cudaMemcpy(BlockIter, BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ), cudaMemcpyDeviceToHost);
-            for( int i = 0; i < numBlocksX; i++ )
-            {    for( int j = 0; j < numBlocksY; j++ )
+            CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
+                                                             interfaceMapPtr.template getData< Device >(),
+                                                             auxPtr.template modifyData< Device>(),
+                                                             BlockIterDevice );
+            cudaDeviceSynchronize();         
+            TNL_CHECK_CUDA_DEVICE;
+            cudaMemcpy(BlockIter, BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( double ), cudaMemcpyDeviceToHost);
+            
+            for( int j = numBlocksY-1; j > -1; j-- )
+            {    
+                for( int i = 0; i < numBlocksX; i++ )
                 {
-                    if( BlockIter[ j * numBlocksY + i ] )
-                        std::cout << "true." << "\t";
-                    else
-                        std::cout << "false." << "\t";    
+                    //if( BlockIter[ j * numBlocksX + i ] )
+                        std::cout << BlockIter[ j * numBlocksY + i ] << "\t";
+                    //else
+                    //    std::cout << "0" << " ";    
                 }
                 std::cout << std::endl;
             }
             std::cout << std::endl;
-             
+                         
             for( int i = 0; i < numBlocksX * numBlocksY; i++ )
                 isNotDone = isNotDone || BlockIter[ i ];
             
@@ -282,148 +289,189 @@ template < 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,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
-                                      bool *BlockIterDevice )
+                                      Real *BlockIterDevice )
 {
-    int thri = threadIdx.x; int thrj = threadIdx.y; // nelze ke stejnym pristupovat znovu pres threadIdx (vzdy da jine hodnoty)
+    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 * 16 + thri;
+    
+    __shared__ volatile Real changed[256];
+    changed[ currentIndex ] = 0.0;
+    __syncthreads();
     
-    __shared__ volatile bool changed[256];
-    changed[ thrj * blockDim.x + thri ] = false;
-     __shared__ volatile bool SmallCycleBoolin;
     if( thrj == 0 && thri == 0 )
-        SmallCycleBoolin = true;
+        changed[ 0 ] = 10.0;
+    __syncthreads();
+        //SmallCycleBoolin = true;
     
     
     const Meshes::Grid< 2, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
-    const Real hx = mesh.getSpaceSteps().x();
-    const Real hy = mesh.getSpaceSteps().y();
+    __shared__ Real hx;
+    __shared__ Real hy;
+    if( thrj == 0 && thri == 0 )
+    {
+        hx = mesh.getSpaceSteps().x();
+        hy = mesh.getSpaceSteps().y();
+    }
+    __syncthreads();
     
     __shared__ volatile Real sArray[ 18 ][ 18 ];
     sArray[thrj][thri] = TypeInfo< Real >::getMaxValue();
+    __syncthreads();
     
     //filling sArray edges
     int dimX = mesh.getDimensions().x(); int dimY = mesh.getDimensions().y();
-    int numOfBlockx = dimX/16 + ((dimX%16 != 0) ? 1:0);
-    int numOfBlocky = dimY/16 + ((dimY%16 != 0) ? 1:0);
-    int xkolik = 17;
-    int ykolik = 17;
-    
+    __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 == blIdx )
-        xkolik = dimX - (blIdx)*16+1;
+        if( numOfBlockx - 1 == blIdx )
+            xkolik = dimX - (blIdx)*blockDim.x+1;
 
-    if( numOfBlocky -1 == blIdy )
-        ykolik = dimY - (blIdy)*16+1;
-       
+        if( numOfBlocky -1 == blIdy )
+            ykolik = dimY - (blIdy)*blockDim.y+1;
+    
+    //filling BlockIterDevice
+        //BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = 0.0;
+     }
+    /*if( blIdx == 2 && blIdy == 3 )
+        BlockIterDevice[ currentIndex ] = 0.0;*/
+    __syncthreads();
     
     if( thri == 0 )
     {        
-        if( dimX > (blIdx+1) * 16  && thrj+1 < ykolik )
-            sArray[thrj+1][xkolik] = aux[ blIdy*16*dimX - dimX + blIdx*16 - 1 + (thrj+1)*dimX + 17 ];
+        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] = TypeInfo< Real >::getMaxValue();
     }
+    __syncthreads();
     
     if( thri == 1 )
     {
         if( blIdx != 0 && thrj+1 < ykolik )
-            sArray[thrj+1][0] = aux[ blIdy*16*dimX - dimX + blIdx*16 - 1 + (thrj+1)*dimX + 0];
+            sArray[thrj+1][0] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + (thrj+1)*dimX ];
         else
             sArray[thrj+1][0] = TypeInfo< Real >::getMaxValue();
     }
+    __syncthreads();
     
     if( thrj == 0 )
     {
-        if( dimY > (blIdy+1) * 16  && thri+1 < xkolik )
-            sArray[ykolik][thri+1] = aux[ blIdy*16*dimX - dimX + blIdx*16 - 1 + 17*dimX + thri+1 ];
+        if( dimY > (blIdy+1) * blockDim.y  && thri+1 < xkolik )
+            sArray[ykolik][thri+1] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + ykolik*dimX + thri+1 ];
         else
            sArray[ykolik][thri+1] = TypeInfo< Real >::getMaxValue();
     }
+    __syncthreads();
     
     if( thrj == 1 )
     {
         if( blIdy != 0 && thri+1 < xkolik )
-            sArray[0][thri+1] = aux[ blIdy*16*dimX - dimX + blIdx*16 - 1 + 0*dimX + thri+1 ];
+            sArray[0][thri+1] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + thri+1 ];
         else
             sArray[0][thri+1] = TypeInfo< Real >::getMaxValue();
     }
+    __syncthreads();
     
-    //filling BlockIterDevice
-    BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = false;
-    
-    
+        
     if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
     {    
         sArray[thrj+1][thri+1] = aux[ j*mesh.getDimensions().x() + i ];
         __syncthreads();
-    }    
+    }  
     
-    __shared__ int loopcounter;
-    if( thri == 0 && thrj == 0 )
-        loopcounter= 0;
-  //if( blIdx == 5 && blIdy == 4 )
-    while( SmallCycleBoolin )   
+    int loopcounter;
+    loopcounter = 0;
+    __syncthreads();
+    
+  /*if( ( blIdx == 4 && blIdy == 5 ) || ( blIdx == 2 && blIdy == 2 ) || ( blIdx == 2 && blIdy == 3 ) ||
+      ( blIdx == 5 && blIdy == 5 ) || ( blIdx == 6 && blIdy == 5 ) || ( blIdx == 5 && blIdy == 4 ) ||
+      ( blIdx == 5 && blIdy == 2 ) || ( blIdx == 3 && blIdy == 2 ) || ( blIdx == 4 && blIdy == 2 ) )*/
+    while( changed[ 0 ] > 0.1 )
     {
-        if( thri == 1 && thrj == 1 )
-            SmallCycleBoolin = false;
-        
-        changed[ thrj * 16 + thri ] = false;
+        changed[ currentIndex] = 0.0;
         __syncthreads();
         
     //calculation of update cell
         if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
-        { 
+        {
             if( ! interfaceMap[ j * mesh.getDimensions().x() + i ] )
             {
-                changed[ thrj * 16 + thri ] = ptr.updateCell( sArray, thri+1, thrj+1, hx, hy );
-                __syncthreads();   
+                changed[ currentIndex ] = ptr.updateCell( sArray, thri+1, thrj+1, hx, hy );  
+    __syncthreads();
             }
         }
+        __syncthreads();
         
     //pyramid reduction
         if( blockDim.x*blockDim.y >= 256 )
         {
-            if( (thrj * 16 + thri) < 128 )
+            if( currentIndex < 128 )
             {
-                changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 128 ];
+                changed[ currentIndex ] = changed[ currentIndex ] + changed[ currentIndex + 128 ];
             }
-        __syncthreads();
         }
+        __syncthreads();
         if( blockDim.x*blockDim.y >= 128 )
         {
-            if( (thrj * 16 + thri) < 64 )
+            if( currentIndex < 64 )
             {
-                changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 64 ];
+                changed[ currentIndex ] = changed[ currentIndex ] + changed[ currentIndex + 64 ];
             }
-        __syncthreads();
-        }
-        if( (thrj * 16 + thri) < 32 )
-        {
-            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 32 ];
-            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 16 ];
-            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 8 ];
-            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 4 ];
-            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 2 ];
-            changed[ thrj * 16 + thri ] = changed[ thrj * 16 + thri ] || changed[ thrj * 16 + thri + 1 ];
         }
         __syncthreads();
-
-        if( thrj == 1 && thri == 1 )
+        if( currentIndex < 32 ) //POUZE IF JSOU SINCHRONNI NA JEDNOM WARPU
         {
-            loopcounter++;
-            if( loopcounter > 1000 )
-                break;
-            SmallCycleBoolin = changed[ 0 ];
-            //if( SmallCycleBoolin )
-                BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = SmallCycleBoolin;
+            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( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
-        aux[ j * mesh.getDimensions().x() + i ] = sArray[ thrj + 1 ][ thri + 1 ];
         
+               
+        loopcounter++;
+        __syncthreads();
+        if( currentIndex != 0 )
+            changed[ currentIndex ] = 0.0;
+        __syncthreads();
+        
+        if( loopcounter > 200 )
+            changed[ currentIndex ] = 0.0;
+        __syncthreads();
+    }
+    /*__syncthreads();
+    if( blIdy == 3 && blIdx == 1 )
+    {
+       BlockIterDevice[ currentIndex ] = changed[ currentIndex ];
+       BlockIterDevice[ 0 ] = loopcounter;
+    }*/
+    __syncthreads();
+    if( blIdx == 2 && blIdy == 3 )
+    { 
+        //SmallCycleBoolin = changed[ 0 ];
+        //if( changed[ 0 ] )
+             BlockIterDevice[ currentIndex ] = changed[ currentIndex ];// loopcounter;
+             BlockIterDevice[ 0 ] = loopcounter;
+
+    }
+      
+    
+    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 ];
+    __syncthreads();
 }
 #endif
-- 
GitLab


From b387101c185fb66d078734baa16e46afb5c713f1 Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Tue, 19 Jun 2018 08:49:04 +0200
Subject: [PATCH 20/28] Mistake found and repaied from 2D. Problem with EOC
 appeared.

---
 .../tnlDirectEikonalMethodsBase.h             |   4 +-
 .../tnlDirectEikonalMethodsBase_impl.h        |  30 ++--
 .../tnlFastSweepingMethod2D_impl.h            | 139 ++++++------------
 3 files changed, 59 insertions(+), 114 deletions(-)

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 fe83b61c0..979c70e7b 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -69,7 +69,7 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
                                          const MeshEntity& cell,
                                          const RealType velocity = 1.0 );
       
-      __cuda_callable__ double updateCell( volatile Real sArray[18][18],
+      __cuda_callable__ bool updateCell( volatile Real sArray[18][18],
                                          int thri, int thrj, const Real hx, const Real hy,
                                          const Real velocity = 1.0 );
       
@@ -117,7 +117,7 @@ template < 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,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
-                                      Real *BlockIterDevice );
+                                      bool *BlockIterDevice );
 
 template < typename Real, typename Device, typename Index >
 __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& input, 
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 098dace4b..9ec95c25f 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
@@ -859,22 +859,12 @@ template< typename Real,
           typename Device,
           typename Index >
 __cuda_callable__
-double
+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 Meshes::Grid< 2, Real, Device, Index >& mesh = u.template getMesh< Devices::Cuda >();
-   /*typedef typename Meshes::Grid< 2, Real, Device, Index >::Cell Cell;
-   Cell cell( mesh );
-   cell.getCoordinates().x() = i; cell.getCoordinates().y() = j;
-   cell.refresh();
-   
-   const auto& neighborEntities = cell.template getNeighborEntities< 2 >();
-   const RealType& hx = mesh.getSpaceSteps().x();
-   const RealType& hy = mesh.getSpaceSteps().y();*/
-   const RealType value = sArray[ thrj ][ thri ];//u( cell );
+   const RealType value = sArray[ thrj ][ thri ];
    RealType a, b, tmp = TypeInfo< RealType >::getMaxValue();
       
    b = TNL::argAbsMin( sArray[ thrj+1 ][ thri ],
@@ -885,7 +875,7 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
 
     if( fabs( a ) == TypeInfo< RealType >::getMaxValue() && 
         fabs( b ) == TypeInfo< RealType >::getMaxValue() )
-       return 0;
+       return false;
    
     RealType pom[6] = { a, b, TypeInfo< Real >::getMaxValue(), (RealType)hx, (RealType)hy, 0.0 };
     sortMinims( pom );
@@ -896,10 +886,10 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
     {
         sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
         tmp = value - sArray[ thrj ][ thri ];
-        if ( fabs( tmp ) >  0.1 )
-            return 10;
+        if ( fabs( tmp ) >  0.001 )
+            return true;
         else
-            return 0;
+            return false;
     }
     else
     {
@@ -908,12 +898,12 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
             ( 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 ) > 0.1 )
-            return 10;
+        if ( fabs( tmp ) > 0.01 )
+            return true;
         else
-            return 0;
+            return false;
     }
     
-    return 0;
+    return false;
 }
 #endif
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 e24201936..e72710408 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
@@ -227,46 +227,38 @@ solve( const MeshPointer& mesh,
           tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
           
           
-          /*int nBlockIter = numBlocksX > numBlocksY ? numBlocksX : numBlocksY;
-          nBlockIter = numBlocksX == numBlocksY ? nBlockIter + 1 : nBlockIter;*/
           
-          bool isNotDone = true; numBlocksX = 16; numBlocksY = 16;
-          double* BlockIter = (double*)malloc( ( numBlocksX * numBlocksY ) * sizeof( double ) );
+          bool* BlockIter = (bool*)malloc( ( numBlocksX * numBlocksY ) * sizeof( bool ) );
           
-          double *BlockIterDevice;
-          cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( double ) );
+          bool *BlockIterDevice;
+          cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ) );
           
-          //while( isNotDone )
+          while( BlockIter[ 0 ] )
           {
            for( int i = 0; i < numBlocksX * numBlocksY; i++ )
-                BlockIter[ i ] = 0.0;
-           cudaMemcpy(BlockIterDevice, BlockIter, ( numBlocksX * numBlocksY ) * sizeof( double ), cudaMemcpyHostToDevice);
-           
-            isNotDone = false;
-            
+                BlockIter[ i ] = false;
+           cudaMemcpy(BlockIterDevice, BlockIter, ( numBlocksX * numBlocksY ) * sizeof( bool ), cudaMemcpyHostToDevice);
+                       
             CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
                                                              interfaceMapPtr.template getData< Device >(),
                                                              auxPtr.template modifyData< Device>(),
                                                              BlockIterDevice );
             cudaDeviceSynchronize();         
             TNL_CHECK_CUDA_DEVICE;
-            cudaMemcpy(BlockIter, BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( double ), cudaMemcpyDeviceToHost);
+            cudaMemcpy(BlockIter, BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ), cudaMemcpyDeviceToHost);
             
-            for( int j = numBlocksY-1; j > -1; j-- )
+            /*for( int j = numBlocksY-1; j > -1; j-- )
             {    
                 for( int i = 0; i < numBlocksX; i++ )
                 {
-                    //if( BlockIter[ j * numBlocksX + i ] )
-                        std::cout << BlockIter[ j * numBlocksY + i ] << "\t";
-                    //else
-                    //    std::cout << "0" << " ";    
+                    std::cout << BlockIter[ j * numBlocksY + i ] << "\t";  
                 }
                 std::cout << std::endl;
             }
-            std::cout << std::endl;
+            std::cout << std::endl;*/
                          
-            for( int i = 0; i < numBlocksX * numBlocksY; i++ )
-                isNotDone = isNotDone || BlockIter[ i ];
+            for( int i = 1; i < numBlocksX * numBlocksY; i++ )
+                BlockIter[ 0 ] = BlockIter[ 0 ] || BlockIter[ i ];
             
           }
           delete[] BlockIter;
@@ -289,7 +281,7 @@ template < 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,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
-                                      Real *BlockIterDevice )
+                                      bool *BlockIterDevice )
 {
     int thri = threadIdx.x; int thrj = threadIdx.y;
     int blIdx = blockIdx.x; int blIdy = blockIdx.y;
@@ -297,15 +289,12 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
     int j = blockDim.y*blIdy + thrj;
     int currentIndex = thrj * 16 + thri;
     
-    __shared__ volatile Real changed[256];
-    changed[ currentIndex ] = 0.0;
-    __syncthreads();
+    __shared__ volatile bool changed[256];
+    changed[ currentIndex ] = false;
     
     if( thrj == 0 && thri == 0 )
-        changed[ 0 ] = 10.0;
+        changed[ 0 ] = true;
     __syncthreads();
-        //SmallCycleBoolin = true;
-    
     
     const Meshes::Grid< 2, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
     __shared__ Real hx;
@@ -315,11 +304,9 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         hx = mesh.getSpaceSteps().x();
         hy = mesh.getSpaceSteps().y();
     }
-    __syncthreads();
     
     __shared__ volatile Real sArray[ 18 ][ 18 ];
     sArray[thrj][thri] = TypeInfo< Real >::getMaxValue();
-    __syncthreads();
     
     //filling sArray edges
     int dimX = mesh.getDimensions().x(); int dimY = mesh.getDimensions().y();
@@ -339,12 +326,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
 
         if( numOfBlocky -1 == blIdy )
             ykolik = dimY - (blIdy)*blockDim.y+1;
-    
-    //filling BlockIterDevice
-        //BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = 0.0;
-     }
-    /*if( blIdx == 2 && blIdy == 3 )
-        BlockIterDevice[ currentIndex ] = 0.0;*/
+    }
     __syncthreads();
     
     if( thri == 0 )
@@ -354,7 +336,6 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         else
             sArray[thrj+1][xkolik] = TypeInfo< Real >::getMaxValue();
     }
-    __syncthreads();
     
     if( thri == 1 )
     {
@@ -363,7 +344,6 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         else
             sArray[thrj+1][0] = TypeInfo< Real >::getMaxValue();
     }
-    __syncthreads();
     
     if( thrj == 0 )
     {
@@ -372,7 +352,6 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         else
            sArray[ykolik][thri+1] = TypeInfo< Real >::getMaxValue();
     }
-    __syncthreads();
     
     if( thrj == 1 )
     {
@@ -381,44 +360,47 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         else
             sArray[0][thri+1] = TypeInfo< Real >::getMaxValue();
     }
-    __syncthreads();
     
         
     if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() )
     {    
         sArray[thrj+1][thri+1] = aux[ j*mesh.getDimensions().x() + i ];
-        __syncthreads();
-    }  
-    
-    int loopcounter;
-    loopcounter = 0;
-    __syncthreads();
-    
-  /*if( ( blIdx == 4 && blIdy == 5 ) || ( blIdx == 2 && blIdy == 2 ) || ( blIdx == 2 && blIdy == 3 ) ||
-      ( blIdx == 5 && blIdy == 5 ) || ( blIdx == 6 && blIdy == 5 ) || ( blIdx == 5 && blIdy == 4 ) ||
-      ( blIdx == 5 && blIdy == 2 ) || ( blIdx == 3 && blIdy == 2 ) || ( blIdx == 4 && blIdy == 2 ) )*/
-    while( changed[ 0 ] > 0.1 )
+    }
+    __syncthreads();  
+
+    while( changed[ 0 ] )
     {
-        changed[ currentIndex] = 0.0;
         __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();
+                changed[ currentIndex ] = ptr.updateCell( sArray, thri+1, thrj+1, hx, hy );
             }
         }
         __syncthreads();
         
+        /*if( thri == 0 && thrj == 0 && blIdx == 1 && blIdy == 3 ){
+            for( int h = 15; h > -1; h-- ){
+                for( int g = 0; g < 16; g++ ){
+                    printf( "%d\t", changed[h*16+g] );
+                }
+                printf("\n");
+            }
+            printf("\n");
+        }
+        __syncthreads();*/
+        
     //pyramid reduction
         if( blockDim.x*blockDim.y >= 256 )
         {
             if( currentIndex < 128 )
             {
-                changed[ currentIndex ] = changed[ currentIndex ] + changed[ currentIndex + 128 ];
+                changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 128 ];
             }
         }
         __syncthreads();
@@ -426,52 +408,25 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         {
             if( currentIndex < 64 )
             {
-                changed[ currentIndex ] = changed[ currentIndex ] + changed[ currentIndex + 64 ];
+                changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 64 ];
             }
         }
         __syncthreads();
         if( currentIndex < 32 ) //POUZE IF JSOU SINCHRONNI NA JEDNOM WARPU
         {
-            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( 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( changed[ 0 ] && thri == 0 && thrj == 0 )
+            BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = changed[ 0 ];
         __syncthreads();
-        
-        
-               
-        loopcounter++;
-        __syncthreads();
-        if( currentIndex != 0 )
-            changed[ currentIndex ] = 0.0;
-        __syncthreads();
-        
-        if( loopcounter > 200 )
-            changed[ currentIndex ] = 0.0;
-        __syncthreads();
-    }
-    /*__syncthreads();
-    if( blIdy == 3 && blIdx == 1 )
-    {
-       BlockIterDevice[ currentIndex ] = changed[ currentIndex ];
-       BlockIterDevice[ 0 ] = loopcounter;
-    }*/
-    __syncthreads();
-    if( blIdx == 2 && blIdy == 3 )
-    { 
-        //SmallCycleBoolin = changed[ 0 ];
-        //if( changed[ 0 ] )
-             BlockIterDevice[ currentIndex ] = changed[ currentIndex ];// loopcounter;
-             BlockIterDevice[ 0 ] = loopcounter;
-
     }
-      
-    
+  
     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 ];
-    __syncthreads();
 }
 #endif
-- 
GitLab


From c822988ebf6264b2aa6c438488f028e38db7d8ba Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Sun, 1 Jul 2018 16:22:54 +0200
Subject: [PATCH 21/28] Implementation of 1D solver is done for both CPU and
 GPU. 2D solver is now working fine

---
 .../tnlDirectEikonalMethodsBase.h             |  23 +-
 .../tnlDirectEikonalMethodsBase_impl.h        | 278 ++++++++++++------
 .../tnlDirectEikonalProblem_impl.h            |   2 +-
 .../tnlFastSweepingMethod1D_impl.h            | 191 +++++++++++-
 .../tnlFastSweepingMethod2D_impl.h            |  67 +++--
 5 files changed, 420 insertions(+), 141 deletions(-)

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 979c70e7b..fe7d4b9ba 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -40,8 +40,12 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >
       
       template< typename MeshEntity >
       __cuda_callable__ void updateCell( MeshFunctionType& u,
-                                         const MeshEntity& cell );
+                                         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 );
 };
 
 
@@ -72,12 +76,6 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >
       __cuda_callable__ bool updateCell( volatile Real sArray[18][18],
                                          int thri, int thrj, const Real hx, const Real hy,
                                          const Real velocity = 1.0 );
-      
-      __cuda_callable__ void setsArray( MeshFunctionType& aux, Real sArray[18][18],
-                                            int dimX, int dimY, int i, int j );
-      
-      /*__cuda_callable__ void getsArray( MeshFunctionType& aux, Real sArray[18][18],
-                                            int dimX, int dimY, int i, int j );*/
 };
 
 template< typename Real,
@@ -113,6 +111,17 @@ __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  );
+
+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 );
+
 template < 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,
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 9ec95c25f..c742a11b5 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
@@ -21,45 +21,72 @@ initInterface( const MeshFunctionPointer& _input,
                MeshFunctionPointer& _output,
                InterfaceMapPointer& _interfaceMap  )
 {
-   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() = 1;
-        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();
-         const RealType& h = mesh.getSpaceSteps().x();
-         //const IndexType& c = cell.getIndex();
-         const IndexType e = neighbors.template getEntityIndex<  1 >();
-         const IndexType w = neighbors.template getEntityIndex< -1 >();
-         if( c * input[ e ] <=0 )
-         {
-             output[ cell.getIndex() ] =
-             c >= 0 ? ( h * c )/( c - input[ e ] ) :
-                      - ( h * c )/( c - input[ e ] );
-             interfaceMap[ cell.getIndex() ] = true;
-         }
-         if( c * input[ w ] <=0 )
-         {
-             output[ cell.getIndex() ] =
-             c >= 0 ? ( h * c )/( c - input[ w ] ) :
-                      - ( h * c )/( c - input[ w ] );
-             interfaceMap[ cell.getIndex() ] = true;
-         }
-      }
-      output[ cell.getIndex() ] =
-      c > 0 ? TNL::TypeInfo< RealType >::getMaxValue() :
-             -TypeInfo< RealType >::getMaxValue();
-      interfaceMap[ cell.getIndex() ] = false;
-   }
+    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;
+#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;
+            cell.getCoordinates().x() < mesh.getDimensions().x();
+            cell.getCoordinates().x() ++ )
+           {
+               cell.refresh();
+               output[ cell.getIndex() ] =
+               input( cell ) >= 0 ? TypeInfo< RealType >::getMaxValue() :
+                                  - TypeInfo< RealType >::getMaxValue();
+               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,
@@ -69,8 +96,31 @@ template< typename Real,
 void
 tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > >::
 updateCell( MeshFunctionType& u,
-            const MeshEntity& cell )
+            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 = TypeInfo< RealType >::getMaxValue();
+
+    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 ) == TypeInfo< RealType >::getMaxValue() )
+      return;
+   
+    tmp = a + TNL::sign( value ) * h/v;
+    
+    u[ cell.getIndex() ] = argAbsMin( value, tmp );
 }
 
 template< typename Real,
@@ -103,7 +153,20 @@ initInterface( const MeshFunctionPointer& _input,
     }
     if( std::is_same< Device, Devices::Host >::value )
     {
-         const MeshFunctionType& input = _input.getData();
+        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();
@@ -186,7 +249,7 @@ initInterface( const MeshFunctionPointer& _input,
 
                         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;                         
+                            output[ e ] = pom; 
                     /*}else
                     {
                         pom = - (hx * c)/( c - input[ e ]);
@@ -632,64 +695,60 @@ __cuda_callable__ void sortMinims( T1 pom[] )
     }   
 }
 
-template< typename Real,
-          typename Device,
-          typename Index >
-__cuda_callable__ void
-tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > >::
-setsArray( MeshFunctionType& aux, Real sArray[18][18], int dimX, int dimY, int blockIdx, int blockIdy )
-{
-    int numOfBlockx = dimX/16 + ((dimX%16 != 0) ? 1:0);
-    int numOfBlocky = dimY/16 + ((dimY%16 != 0) ? 1:0);
-    int xkolik = 18;
-    int ykolik = 18;
-    int xOd = 0;
-    int yOd = 0;
-    
-    if( blockIdx == 0 )
-        xOd = 1;
-    if( blockIdy == 0 )
-        yOd = 1;
-    
-    if( numOfBlockx - 1 == blockIdx )
-        xkolik = dimX - (numOfBlockx-1)*16+1;
 
-    if( numOfBlocky -1 == blockIdy )
-        ykolik = dimY - (numOfBlocky-1)*16+1;
+
+#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  )
+{
+    int i = threadIdx.x + blockDim.x*blockIdx.x;
+    const Meshes::Grid< 1, Real, Device, Index >& mesh = input.template getMesh< Devices::Cuda >();
     
-    if( dimX > (blockIdx+1) * 16 )
-    {
-        for( int i = yOd; i < ykolik; i++ )
-        {
-            sArray[i][17] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + i*dimX + 17 ];
-        }
-    }
-    if( blockIdx != 0 )
-    {
-        for( int i = yOd; i < ykolik; i++ )
-        {
-            sArray[i][0] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + i*dimX + 0];
-        }
-    } 
-    if( dimY > (blockIdy+1) * 16 )
+    if( i < mesh.getDimensions().x()  )
     {
-        for( int i = xOd; i < xkolik; i++ )
+        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 ? TypeInfo< Real >::getMaxValue() :
+                                    - TypeInfo< Real >::getMaxValue();
+        interfaceMap[ cind ] = false; 
+
+        const Real& h = mesh.getSpaceSteps().x();
+        cell.refresh();
+        const Real& c = input( cell );
+        if( ! cell.isBoundaryEntity()  )
         {
-            sArray[17][i] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + 17*dimX + i ];
+           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;
+           }
         }
     }
-    if( blockIdy != 0 )
-    {
-        for( int i = xOd; i < xkolik; i++ )
-        {
-            sArray[0][i] = aux[ blockIdy*16*dimX - dimX + blockIdx*16 - 1 + 0*dimX + i ];
-        }
-    }   
+           
 }
-
-
-
-#ifdef HAVE_CUDA
 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,
@@ -886,7 +945,7 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
     {
         sArray[ thrj ][ thri ] = argAbsMin( value, tmp );
         tmp = value - sArray[ thrj ][ thri ];
-        if ( fabs( tmp ) >  0.001 )
+        if ( fabs( tmp ) >  0.01*hx )
             return true;
         else
             return false;
@@ -898,7 +957,7 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
             ( 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 )
+        if ( fabs( tmp ) > 0.01*hx )
             return true;
         else
             return false;
@@ -906,4 +965,33 @@ updateCell( volatile Real sArray[18][18], int thri, int thrj, const Real hx, con
     
     return false;
 }
+
+template< typename Real,
+          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 = TypeInfo< RealType >::getMaxValue();
+      
+   a = TNL::argAbsMin( sArray[ thri+1 ],
+                       sArray[ thri-1 ] );
+
+    if( fabs( a ) == TypeInfo< RealType >::getMaxValue() )
+       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;
+}
 #endif
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem_impl.h
index 57ac7d689..d1eb8fe2e 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem_impl.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalProblem_impl.h
@@ -125,7 +125,7 @@ bool
 tnlDirectEikonalProblem< Mesh, Anisotropy, Real, Index >::
 solve( const MeshPointer& mesh,
        DofVectorPointer& dofs )
-{
+{   
    FastSweepingMethod< MeshType, AnisotropyType > fsm;
    fsm.solve( mesh, anisotropy, initialData );
    return true;
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
index 105f1ceda..87974c684 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
@@ -57,14 +57,191 @@ FastSweepingMethod< Meshes::Grid< 1, Real, Device, Index >, Anisotropy >::
 solve( const MeshPointer& mesh,
        const AnisotropyPointer& anisotropy,
        MeshFunctionPointer& u )
-{
-   MeshFunctionPointer aux;
-   InterfaceMapPointer interfaceMap;
-   aux->setMesh( mesh );
-   interfaceMap->setMesh( mesh );
+{   
+   MeshFunctionPointer auxPtr;
+   InterfaceMapPointer interfaceMapPtr;
+   auxPtr->setMesh( mesh );
+   interfaceMapPtr->setMesh( mesh );
    std::cout << "Initiating the interface cells ..." << std::endl;
-   BaseType::initInterface( u, aux, interfaceMap );
-   aux->save( "aux-ini.tnl" );
+   BaseType::initInterface( u, auxPtr, interfaceMapPtr );
+        
+   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().x() = 0;
+               cell.getCoordinates().x() < mesh->getDimensions().x();
+               cell.getCoordinates().x()++ )
+             {
+                cell.refresh();
+                if( ! interfaceMap( cell ) )
+                   this->updateCell( aux, cell );
+             }
+
+         
+        for( cell.getCoordinates().x() = mesh->getDimensions().x() - 1;
+             cell.getCoordinates().x() >= 0 ;
+             cell.getCoordinates().x()-- )		
+           {
+              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( 16 );
+          int numBlocksX = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().x(), cudaBlockSize );
+          dim3 blockSize( cudaBlockSize );
+          dim3 gridSize( numBlocksX );
+          
+          tnlDirectEikonalMethodsBase< Meshes::Grid< 1, Real, Device, Index > > ptr;
+          
+          
+          
+          bool* BlockIter = (bool*)malloc( ( numBlocksX ) * sizeof( bool ) );
+          
+          bool *BlockIterDevice;
+          cudaMalloc(&BlockIterDevice, ( numBlocksX ) * sizeof( bool ) );
+          
+          while( BlockIter[ 0 ] )
+          {
+           for( int i = 0; i < numBlocksX; i++ )
+                BlockIter[ i ] = false;
+           cudaMemcpy(BlockIterDevice, BlockIter, ( numBlocksX ) * sizeof( bool ), cudaMemcpyHostToDevice);
+                       
+            CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
+                                                             interfaceMapPtr.template getData< Device >(),
+                                                             auxPtr.template modifyData< Device>(),
+                                                             BlockIterDevice );
+            cudaMemcpy(BlockIter, BlockIterDevice, ( numBlocksX ) * sizeof( bool ), cudaMemcpyDeviceToHost);
+                                   
+            for( int i = 1; i < numBlocksX; i++ )
+                BlockIter[ 0 ] = BlockIter[ 0 ] || BlockIter[ i ];
+            
+          }
+          delete[] BlockIter;
+          cudaFree( BlockIterDevice );
+          cudaDeviceSynchronize();
+          
+          TNL_CHECK_CUDA_DEVICE;
+              
+          aux = *auxPtr;
+          interfaceMap = *interfaceMapPtr;
+#endif
+      }
+      iteration++;
+   }
+   
+   aux.save("aux-final.tnl");
+}
+
+#ifdef HAVE_CUDA
+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 )
+{
+    int thri = threadIdx.x;
+    int blIdx = blockIdx.x;
+    int i = thri + blockDim.x*blIdx;
+    
+    __shared__ volatile bool changed[16];
+    changed[ thri ] = false;
+    
+    if( thri == 0 )
+        changed[ 0 ] = true;
+    
+    const Meshes::Grid< 1, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
+    __shared__ Real h;
+    if( thri == 1 )
+    {
+        h = mesh.getSpaceSteps().x();
+    }
+    
+    __shared__ volatile Real sArray[ 18 ];
+    sArray[thri] = TypeInfo< Real >::getMaxValue();
+    
+    //filling sArray edges
+    int dimX = mesh.getDimensions().x(); 
+    __shared__ volatile int numOfBlockx;
+    __shared__ int xkolik;
+    if( thri == 0 )
+    {
+        xkolik = blockDim.x + 1;
+        numOfBlockx = dimX/blockDim.x + ((dimX%blockDim.x != 0) ? 1:0);
+    
+        if( numOfBlockx - 1 == blIdx )
+            xkolik = dimX - (blIdx)*blockDim.x+1;
+    }
+    __syncthreads();
+    
+    if( thri == 0 )
+    {        
+        if( dimX > (blIdx+1) * blockDim.x )
+            sArray[xkolik] = aux[ blIdx*blockDim.x - 1 + xkolik ];
+        else
+            sArray[xkolik] = TypeInfo< Real >::getMaxValue();
+    }
+    
+    if( thri == 1 )
+    {
+        if( blIdx != 0 )
+            sArray[0] = aux[ blIdx*blockDim.x - 1 ];
+        else
+            sArray[0] = TypeInfo< Real >::getMaxValue();
+    }
+    
+        
+    if( i < mesh.getDimensions().x() )
+    {    
+        sArray[thri+1] = aux[ i ];
+    }
+    __syncthreads();  
 
+    while( changed[ 0 ] )
+    {
+        __syncthreads();
+        
+        changed[ thri ] = false;
+        
+    //calculation of update cell
+        if( i < mesh.getDimensions().x() )
+        {
+            if( ! interfaceMap[ i ] )
+            {
+                changed[ thri ] = ptr.updateCell( sArray, thri+1, h );
+            }
+        }
+        __syncthreads();
+        
+        
+        
+    //pyramid reduction
+        if( thri < 8 ) changed[ thri ] = changed[ thri ] || changed[ thri + 8 ];
+        if( thri < 4 ) changed[ thri ] = changed[ thri ] || changed[ thri + 4 ];
+        if( thri < 2 ) changed[ thri ] = changed[ thri ] || changed[ thri + 2 ];
+        if( thri < 1 ) changed[ thri ] = changed[ thri ] || changed[ thri + 1 ];
+        __syncthreads();
+
+        if( changed[ 0 ] && thri == 0 )
+            BlockIterDevice[ blIdx ] = true;
+        __syncthreads();
+    }
+  
+    if( i < mesh.getDimensions().x()  && (!interfaceMap[ i ]) )
+        aux[ i ] = sArray[ thri + 1 ];
 }
+#endif
+
 
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 e72710408..e5a9c39f3 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
@@ -18,13 +18,16 @@
 #include <TNL/Devices/Cuda.h>
 
 
+#include <iostream>
+#include <fstream>
+
 template< typename Real,
           typename Device,
           typename Index,
           typename Anisotropy >
 FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >::
 FastSweepingMethod()
-: maxIterations( 1 )
+: maxIterations( 2 )
 {
    
 }
@@ -61,15 +64,31 @@ 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 );
-#ifdef HAVE_CUDA
-   cudaDeviceSynchronize();
-#endif
         
    auxPtr->save( "aux-ini.tnl" );
 
@@ -92,7 +111,7 @@ solve( const MeshPointer& mesh,
                {
                   cell.refresh();
                   if( ! interfaceMap( cell ) )
-                     this->updateCell( aux, cell );
+                     this->updateCell( aux, cell, v( cell ) );
                }
          }
 
@@ -109,7 +128,7 @@ solve( const MeshPointer& mesh,
                   //std::cerr << "2 -> ";
                   cell.refresh();
                   if( ! interfaceMap( cell ) )            
-                     this->updateCell( aux, cell );
+                     this->updateCell( aux, cell, v( cell ) );
                }
          }
 
@@ -126,7 +145,7 @@ solve( const MeshPointer& mesh,
                   //std::cerr << "3 -> ";
                   cell.refresh();
                   if( ! interfaceMap( cell ) )            
-                     this->updateCell( aux, cell );
+                     this->updateCell( aux, cell, v( cell ) );
                }
             }
 
@@ -143,7 +162,7 @@ solve( const MeshPointer& mesh,
                   //std::cerr << "4 -> ";
                   cell.refresh();
                   if( ! interfaceMap( cell ) )            
-                     this->updateCell( aux, cell );
+                     this->updateCell( aux, cell, v( cell ) );
                }
             }
 
@@ -222,7 +241,6 @@ solve( const MeshPointer& mesh,
           int numBlocksY = Devices::Cuda::getNumberOfBlocks( mesh->getDimensions().y(), cudaBlockSize );
           dim3 blockSize( cudaBlockSize, cudaBlockSize );
           dim3 gridSize( numBlocksX, numBlocksY );
-          Devices::Cuda::synchronizeDevice();
           
           tnlDirectEikonalMethodsBase< Meshes::Grid< 2, Real, Device, Index > > ptr;
           
@@ -243,20 +261,8 @@ solve( const MeshPointer& mesh,
                                                              interfaceMapPtr.template getData< Device >(),
                                                              auxPtr.template modifyData< Device>(),
                                                              BlockIterDevice );
-            cudaDeviceSynchronize();         
-            TNL_CHECK_CUDA_DEVICE;
             cudaMemcpy(BlockIter, BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ), cudaMemcpyDeviceToHost);
-            
-            /*for( int j = numBlocksY-1; j > -1; j-- )
-            {    
-                for( int i = 0; i < numBlocksX; i++ )
-                {
-                    std::cout << BlockIter[ j * numBlocksY + i ] << "\t";  
-                }
-                std::cout << std::endl;
-            }
-            std::cout << std::endl;*/
-                         
+                                   
             for( int i = 1; i < numBlocksX * numBlocksY; i++ )
                 BlockIter[ 0 ] = BlockIter[ 0 ] || BlockIter[ i ];
             
@@ -294,12 +300,11 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
     
     if( thrj == 0 && thri == 0 )
         changed[ 0 ] = true;
-    __syncthreads();
     
     const Meshes::Grid< 2, Real, Device, Index >& mesh = interfaceMap.template getMesh< Devices::Cuda >();
     __shared__ Real hx;
     __shared__ Real hy;
-    if( thrj == 0 && thri == 0 )
+    if( thrj == 1 && thri == 1 )
     {
         hx = mesh.getSpaceSteps().x();
         hy = mesh.getSpaceSteps().y();
@@ -345,20 +350,20 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
             sArray[thrj+1][0] = TypeInfo< Real >::getMaxValue();
     }
     
-    if( thrj == 0 )
+    if( thri == 2 )
     {
         if( dimY > (blIdy+1) * blockDim.y  && thri+1 < xkolik )
-            sArray[ykolik][thri+1] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + ykolik*dimX + thri+1 ];
+            sArray[ykolik][thrj+1] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + ykolik*dimX + thrj+1 ];
         else
-           sArray[ykolik][thri+1] = TypeInfo< Real >::getMaxValue();
+           sArray[ykolik][thrj+1] = TypeInfo< Real >::getMaxValue();
     }
     
-    if( thrj == 1 )
+    if( thri == 3 )
     {
         if( blIdy != 0 && thri+1 < xkolik )
-            sArray[0][thri+1] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + thri+1 ];
+            sArray[0][thrj+1] = aux[ blIdy*blockDim.y*dimX - dimX + blIdx*blockDim.x - 1 + thrj+1 ];
         else
-            sArray[0][thri+1] = TypeInfo< Real >::getMaxValue();
+            sArray[0][thrj+1] = TypeInfo< Real >::getMaxValue();
     }
     
         
@@ -422,7 +427,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
             if( currentIndex < 1 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 1 ];
         }
         if( changed[ 0 ] && thri == 0 && thrj == 0 )
-            BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = changed[ 0 ];
+            BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = true;
         __syncthreads();
     }
   
-- 
GitLab


From 2d97cf97b0a50451791a0f48a47f9962fdb1dce6 Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Wed, 4 Jul 2018 09:35:29 +0200
Subject: [PATCH 22/28] Forgot to comment wildfire computation

---
 .../tnlDirectEikonalMethodsBase_impl.h               |  4 ++--
 .../hamilton-jacobi/tnlFastSweepingMethod2D_impl.h   | 12 ++++++------
 2 files changed, 8 insertions(+), 8 deletions(-)

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 c742a11b5..cd4eb9e24 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
@@ -155,7 +155,7 @@ initInterface( const MeshFunctionPointer& _input,
     {
         MeshFunctionType input = _input.getData();
         
-        double A[320][320];
+        /*double A[320][320];
         std::ifstream fileInit("/home/maty/Downloads/initData.txt");
 
         for (int i = 0; i < 320; i++)
@@ -164,7 +164,7 @@ initInterface( const MeshFunctionPointer& _input,
         fileInit.close();
         for (int i = 0; i < 320; i++)
             for (int j = 0; j < 320; j++)
-                input[i*320 + j] = A[i][j];
+                input[i*320 + j] = A[i][j];*/
         
         
          MeshFunctionType& output = _output.modifyData();
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 e5a9c39f3..83333b190 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
@@ -64,7 +64,7 @@ solve( const MeshPointer& mesh,
        const AnisotropyPointer& anisotropy,
        MeshFunctionPointer& u )
 {
-   MeshFunctionType v;
+   /*MeshFunctionType v;
    v.setMesh(mesh);
    double A[320][320];
     for (int i = 0; i < 320; i++)
@@ -80,7 +80,7 @@ solve( const MeshPointer& mesh,
     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");
+   v.save("mapa.tnl");*/
    
        
    MeshFunctionPointer auxPtr;
@@ -111,7 +111,7 @@ solve( const MeshPointer& mesh,
                {
                   cell.refresh();
                   if( ! interfaceMap( cell ) )
-                     this->updateCell( aux, cell, v( cell ) );
+                     this->updateCell( aux, cell );
                }
          }
 
@@ -128,7 +128,7 @@ solve( const MeshPointer& mesh,
                   //std::cerr << "2 -> ";
                   cell.refresh();
                   if( ! interfaceMap( cell ) )            
-                     this->updateCell( aux, cell, v( cell ) );
+                     this->updateCell( aux, cell );
                }
          }
 
@@ -145,7 +145,7 @@ solve( const MeshPointer& mesh,
                   //std::cerr << "3 -> ";
                   cell.refresh();
                   if( ! interfaceMap( cell ) )            
-                     this->updateCell( aux, cell, v( cell ) );
+                     this->updateCell( aux, cell );
                }
             }
 
@@ -162,7 +162,7 @@ solve( const MeshPointer& mesh,
                   //std::cerr << "4 -> ";
                   cell.refresh();
                   if( ! interfaceMap( cell ) )            
-                     this->updateCell( aux, cell, v( cell ) );
+                     this->updateCell( aux, cell );
                }
             }
 
-- 
GitLab


From b4d605d125cdae2e0065f747f0e9c2334fc41a8f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1=C5=A1=20Oberhuber?= <oberhuber.tomas@gmail.com>
Date: Wed, 4 Jul 2018 21:26:59 +0200
Subject: [PATCH 23/28] Added compilation of tnl-direct-eikonal-solver with
 float.

---
 .../Solvers/hamilton-jacobi/MainBuildConfig.h | 58 +++++++++----------
 .../tnl-direct-eikonal-solver.h               |  4 +-
 2 files changed, 31 insertions(+), 31 deletions(-)

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 1b904c0c8..41cd2e5d0 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/MainBuildConfig.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/MainBuildConfig.h
@@ -1,64 +1,62 @@
 /***************************************************************************
-                          MainBuildConfig.h  -  description
+                          HamiltonJacobiBuildConfigTag.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.                                   *
- *                                                                         *
- ***************************************************************************/
+/* See Copyright Notice in tnl/Copyright */
+
+#pragma once
+
+#include <TNL/Solvers/BuildConfigTags.h>
 
-#ifndef MAINBUILDCONFIG_H_
-#define MAINBUILDCONFIG_H_
+//namespace TNL {
+//namespace Solvers {   
 
-#include <solvers/tnlBuildConfigTags.h>
+using namespace TNL::Solvers;
 
-class MainBuildConfig
+class HamiltonJacobiBuildConfig
 {
    public:
 
-      static void print() { cerr << "MainBuildConfig" << endl; }
+      static void print() { std::cerr << "HamiltonJacobiBuildConfig" << 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 }; };
+//template<> struct ConfigTagReal< HamiltonJacobiBuildConfig, float > { enum { enabled = false }; };
+template<> struct ConfigTagReal< HamiltonJacobiBuildConfig, 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 }; };
+template<> struct ConfigTagIndex< HamiltonJacobiBuildConfig, short int >{ enum { enabled = false }; };
+template<> struct ConfigTagIndex< HamiltonJacobiBuildConfig, long int >{ enum { enabled = false }; };
 
 /****
- * Use of tnlGrid is enabled for allowed dimensions and Real, Device and Index types.
+ * Use of Grid 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 }; };
+/*template< int Dimension, typename Real, typename Device, typename Index >
+   struct ConfigTagMesh< HamiltonJacobiBuildConfig, Meshes::Grid< Dimension, Real, Device, Index > >
+      { enum { enabled = ConfigTagDimension< HamiltonJacobiBuildConfig, Dimension >::enabled  &&
+                         ConfigTagReal< HamiltonJacobiBuildConfig, Real >::enabled &&
+                         ConfigTagDevice< HamiltonJacobiBuildConfig, Device >::enabled &&
+                         ConfigTagIndex< HamiltonJacobiBuildConfig, 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 }; };
+template<> struct ConfigTagTimeDiscretisation< HamiltonJacobiBuildConfig, ExplicitTimeDiscretisationTag >{ enum { enabled = true }; };
+template<> struct ConfigTagTimeDiscretisation< HamiltonJacobiBuildConfig, SemiImplicitTimeDiscretisationTag >{ enum { enabled = true }; };
+template<> struct ConfigTagTimeDiscretisation< HamiltonJacobiBuildConfig, ImplicitTimeDiscretisationTag >{ enum { enabled = false }; };
 
 /****
  * Only the Runge-Kutta-Merson solver is enabled by default.
  */
-template<> struct tnlConfigTagExplicitSolver< MainBuildConfig, tnlExplicitEulerSolverTag >{ enum { enabled = false }; };
+//template<> struct ConfigTagExplicitSolver< HamiltonJacobiBuildConfig, ExplicitEulerSolverTag >{ enum { enabled = false }; };
 
-#endif /* MAINBUILDCONFIG_H_ */
+//} // namespace Solvers
+//} // namespace TNL
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-direct-eikonal-solver.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-direct-eikonal-solver.h
index 225ce04b9..29c89cae8 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-direct-eikonal-solver.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnl-direct-eikonal-solver.h
@@ -20,11 +20,13 @@
 #include <TNL/Functions/MeshFunction.h>
 #include <TNL/Meshes/Grid.h>
 #include "tnlDirectEikonalProblem.h"
+#include "MainBuildConfig.h"
 
 using namespace TNL;
 
 //typedef tnlDefaultBuildMeshConfig BuildConfig;
-typedef Solvers::FastBuildConfig BuildConfig;
+//typedef Solvers::FastBuildConfig BuildConfig;
+typedef HamiltonJacobiBuildConfig BuildConfig;
 
 template< typename MeshConfig >
 class tnlDirectEikonalSolverConfig
-- 
GitLab


From b4611f421a5aa2001a06f2fa10bbad0dd680bad8 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1=C5=A1=20Oberhuber?= <oberhuber.tomas@gmail.com>
Date: Wed, 4 Jul 2018 23:26:23 +0200
Subject: [PATCH 24/28] Added grids with float to tnl-diff.

---
 src/Tools/tnl-diff.cpp | 52 ++++++++++++++++++++++++++++++++++--------
 1 file changed, 42 insertions(+), 10 deletions(-)

diff --git a/src/Tools/tnl-diff.cpp b/src/Tools/tnl-diff.cpp
index 013f7bf21..48e25156f 100644
--- a/src/Tools/tnl-diff.cpp
+++ b/src/Tools/tnl-diff.cpp
@@ -68,23 +68,55 @@ int main( int argc, char* argv[] )
       int dimensions = atoi( parsedMeshType[ 1 ].getString() );
       if( dimensions == 1 )
       {
-         typedef Meshes::Grid< 1, double, Devices::Host, int > MeshType;
-         if( ! processFiles< MeshType >( parameters ) )
-            return EXIT_FAILURE;
+         if( parsedMeshType[ 1 ] == "float" )
+         {
+            typedef Meshes::Grid< 1, float, Devices::Host, int > MeshType;
+            if( ! processFiles< MeshType >( parameters ) )
+               return EXIT_FAILURE;
+            return EXIT_SUCCESS;
+         }
+         if( parsedMeshType[ 1 ] == "double" )
+         {
+            typedef Meshes::Grid< 1, double, Devices::Host, int > MeshType;
+            if( ! processFiles< MeshType >( parameters ) )
+               return EXIT_FAILURE;
+            return EXIT_SUCCESS;
+         }
       }
       if( dimensions == 2 )
       {
-         typedef Meshes::Grid< 2, double, Devices::Host, int > MeshType;
-         if( ! processFiles< MeshType >( parameters ) )
-            return EXIT_FAILURE;
+         if( parsedMeshType[ 1 ] == "float" )
+         {
+            typedef Meshes::Grid< 2, float, Devices::Host, int > MeshType;
+            if( ! processFiles< MeshType >( parameters ) )
+               return EXIT_FAILURE;
+            return EXIT_SUCCESS;
+         }
+         if( parsedMeshType[ 1 ] == "double" )
+         {
+            typedef Meshes::Grid< 2, double, Devices::Host, int > MeshType;
+            if( ! processFiles< MeshType >( parameters ) )
+               return EXIT_FAILURE;
+            return EXIT_SUCCESS;
+         }
       }
       if( dimensions == 3 )
       {
-         typedef Meshes::Grid< 3, double, Devices::Host, int > MeshType;
-         if( ! processFiles< MeshType >( parameters ) )
-            return EXIT_FAILURE;
+         if( parsedMeshType[ 1 ] == "float" )
+         {
+            typedef Meshes::Grid< 3, float, Devices::Host, int > MeshType;
+            if( ! processFiles< MeshType >( parameters ) )
+               return EXIT_FAILURE;
+            return EXIT_SUCCESS;
+         }
+         if( parsedMeshType[ 1 ] == "double" )
+         {
+            typedef Meshes::Grid< 3, double, Devices::Host, int > MeshType;
+            if( ! processFiles< MeshType >( parameters ) )
+               return EXIT_FAILURE;
+            return EXIT_SUCCESS;
+         }
       }
-      return EXIT_SUCCESS;
    }
    return EXIT_FAILURE;
 }
-- 
GitLab


From ccb30dce5f4f52b058e95afbbaefea06707f4e1d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1=C5=A1=20Oberhuber?= <oberhuber.tomas@gmail.com>
Date: Thu, 5 Jul 2018 09:20:21 +0200
Subject: [PATCH 25/28] Fixed tnl-diff.

---
 src/Tools/tnl-diff.cpp | 12 ++++++------
 src/Tools/tnl-diff.h   |  1 +
 2 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/src/Tools/tnl-diff.cpp b/src/Tools/tnl-diff.cpp
index 48e25156f..df87a57b7 100644
--- a/src/Tools/tnl-diff.cpp
+++ b/src/Tools/tnl-diff.cpp
@@ -68,14 +68,14 @@ int main( int argc, char* argv[] )
       int dimensions = atoi( parsedMeshType[ 1 ].getString() );
       if( dimensions == 1 )
       {
-         if( parsedMeshType[ 1 ] == "float" )
+         if( parsedMeshType[ 2 ] == "float" )
          {
             typedef Meshes::Grid< 1, float, Devices::Host, int > MeshType;
             if( ! processFiles< MeshType >( parameters ) )
                return EXIT_FAILURE;
             return EXIT_SUCCESS;
          }
-         if( parsedMeshType[ 1 ] == "double" )
+         if( parsedMeshType[ 2 ] == "double" )
          {
             typedef Meshes::Grid< 1, double, Devices::Host, int > MeshType;
             if( ! processFiles< MeshType >( parameters ) )
@@ -85,14 +85,14 @@ int main( int argc, char* argv[] )
       }
       if( dimensions == 2 )
       {
-         if( parsedMeshType[ 1 ] == "float" )
+         if( parsedMeshType[ 2 ] == "float" )
          {
             typedef Meshes::Grid< 2, float, Devices::Host, int > MeshType;
             if( ! processFiles< MeshType >( parameters ) )
                return EXIT_FAILURE;
             return EXIT_SUCCESS;
          }
-         if( parsedMeshType[ 1 ] == "double" )
+         if( parsedMeshType[ 2 ] == "double" )
          {
             typedef Meshes::Grid< 2, double, Devices::Host, int > MeshType;
             if( ! processFiles< MeshType >( parameters ) )
@@ -102,14 +102,14 @@ int main( int argc, char* argv[] )
       }
       if( dimensions == 3 )
       {
-         if( parsedMeshType[ 1 ] == "float" )
+         if( parsedMeshType[ 2 ] == "float" )
          {
             typedef Meshes::Grid< 3, float, Devices::Host, int > MeshType;
             if( ! processFiles< MeshType >( parameters ) )
                return EXIT_FAILURE;
             return EXIT_SUCCESS;
          }
-         if( parsedMeshType[ 1 ] == "double" )
+         if( parsedMeshType[ 2 ] == "double" )
          {
             typedef Meshes::Grid< 3, double, Devices::Host, int > MeshType;
             if( ! processFiles< MeshType >( parameters ) )
diff --git a/src/Tools/tnl-diff.h b/src/Tools/tnl-diff.h
index 66f707593..cb7fcfff5 100644
--- a/src/Tools/tnl-diff.h
+++ b/src/Tools/tnl-diff.h
@@ -300,6 +300,7 @@ bool computeDifference( const MeshPointer& meshPointer, const String& objectType
    if( objectType == "Containers::Vector" ||
        objectType == "tnlVector" || objectType == "tnlSharedVector" )   // TODO: remove deprecated type name
       return computeDifferenceOfVectors< MeshPointer, Element, Real, Index >( meshPointer, parameters );
+   std::cerr << "Unknown object type " << objectType << "." << std::endl;
    return false;
 }
 
-- 
GitLab


From 68f3276ebc650ddf8f6b1259583bfd0a8e3a2caf Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Thu, 5 Jul 2018 19:27:34 +0200
Subject: [PATCH 26/28] Parallel reduction grid level.

---
 .../tnlDirectEikonalMethodsBase.h             |  3 +-
 .../tnlFastSweepingMethod2D_impl.h            | 86 +++++++++++++++----
 2 files changed, 73 insertions(+), 16 deletions(-)

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 fe7d4b9ba..74393f1fd 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -126,7 +126,8 @@ template < 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,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
-                                      bool *BlockIterDevice );
+                                      int *BlockIterDevice );
+__global__ void CudaParallelReduc( int *BlockIterDevice, int *dBlock, int nBlocks );
 
 template < typename Real, typename Device, typename Index >
 __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& input, 
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 83333b190..efdb13da8 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
@@ -27,7 +27,7 @@ template< typename Real,
           typename Anisotropy >
 FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >::
 FastSweepingMethod()
-: maxIterations( 2 )
+: maxIterations( 1 )
 {
    
 }
@@ -246,29 +246,37 @@ solve( const MeshPointer& mesh,
           
           
           
-          bool* BlockIter = (bool*)malloc( ( numBlocksX * numBlocksY ) * sizeof( bool ) );
+          //int BlockIter = 1;// = (bool*)malloc( ( numBlocksX * numBlocksY ) * sizeof( bool ) );
+
+          int *BlockIterDevice;
+          int BlockIterD = 1;
           
-          bool *BlockIterDevice;
-          cudaMalloc(&BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ) );
+          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( BlockIter[ 0 ] )
+          while( BlockIterD )
           {
-           for( int i = 0; i < numBlocksX * numBlocksY; i++ )
-                BlockIter[ i ] = false;
-           cudaMemcpy(BlockIterDevice, BlockIter, ( numBlocksX * numBlocksY ) * sizeof( bool ), cudaMemcpyHostToDevice);
+           /*for( int i = 0; i < numBlocksX * numBlocksY; i++ )
+                BlockIter[ i ] = false;*/
                        
             CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
                                                              interfaceMapPtr.template getData< Device >(),
                                                              auxPtr.template modifyData< Device>(),
                                                              BlockIterDevice );
-            cudaMemcpy(BlockIter, BlockIterDevice, ( numBlocksX * numBlocksY ) * sizeof( bool ), cudaMemcpyDeviceToHost);
+            
+            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 ];
+            /*for( int i = 1; i < numBlocksX * numBlocksY; i++ )
+                BlockIter[ 0 ] = BlockIter[ 0 ] || BlockIter[ i ];*/
             
           }
-          delete[] BlockIter;
           cudaFree( BlockIterDevice );
+          cudaFree( dBlock );
           cudaDeviceSynchronize();
           
           TNL_CHECK_CUDA_DEVICE;
@@ -283,11 +291,58 @@ solve( const MeshPointer& mesh,
 }
 
 #ifdef HAVE_CUDA
+__global__ void CudaParallelReduc( int *BlockIterDevice, int *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 ];
+        }
+    }
+    __syncthreads();
+    if (blockDim.x * blockDim.y >= 128) {
+        if (i < 64) {
+            sArray[ i ] += sArray[ i + 64 ];
+        }
+    }
+    __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 ];
+    }
+    
+    if( i == 0 )
+        dBlock[ blId ] = sArray[ 0 ];
+}
+
+
+
 template < 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,
                                       Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
-                                      bool *BlockIterDevice )
+                                      int *BlockIterDevice )
 {
     int thri = threadIdx.x; int thrj = threadIdx.y;
     int blIdx = blockIdx.x; int blIdy = blockIdx.y;
@@ -331,6 +386,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
 
         if( numOfBlocky -1 == blIdy )
             ykolik = dimY - (blIdy)*blockDim.y+1;
+        BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = 0;
     }
     __syncthreads();
     
@@ -360,7 +416,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
     
     if( thri == 3 )
     {
-        if( blIdy != 0 && thri+1 < xkolik )
+        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] = TypeInfo< Real >::getMaxValue();
@@ -427,7 +483,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
             if( currentIndex < 1 ) changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 1 ];
         }
         if( changed[ 0 ] && thri == 0 && thrj == 0 )
-            BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = true;
+            BlockIterDevice[ blIdy * numOfBlockx + blIdx ] = 1;
         __syncthreads();
     }
   
-- 
GitLab


From 223c69a75ed82b7a8adabb8713f97ee716fcd3ae Mon Sep 17 00:00:00 2001
From: fencl <fenclmat@fjfi.cvut.cz>
Date: Thu, 23 Aug 2018 11:10:16 +0200
Subject: [PATCH 27/28] Parallel version for CUDA of FSM is implemented.

---
 .../tnlDirectEikonalMethodsBase.h             |  14 +-
 .../tnlDirectEikonalMethodsBase_impl.h        |  74 ++++++
 .../tnlFastSweepingMethod2D_impl.h            |  73 ++++--
 .../tnlFastSweepingMethod3D_impl.h            | 237 ++++++++++++++++--
 4 files changed, 359 insertions(+), 39 deletions(-)

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 74393f1fd..a406a5b17 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlDirectEikonalMethodsBase.h
@@ -101,6 +101,10 @@ class tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > >
       __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 );
 };
 
 template < typename T1, typename T2 >
@@ -125,10 +129,13 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
 template < 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,
-                                      Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
-                                      int *BlockIterDevice );
+                                      Real *aux,
+                                      int *BlockIterDevice);
 __global__ void CudaParallelReduc( int *BlockIterDevice, int *dBlock, int nBlocks );
 
+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 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,
@@ -142,7 +149,8 @@ __global__ void CudaInitCaller3d( const Functions::MeshFunction< Meshes::Grid< 3
 template < 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 );
+                                      Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& aux,
+                                      int *BlockIterDevice );
 #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 cd4eb9e24..31bc95df5 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
@@ -994,4 +994,78 @@ updateCell( volatile Real sArray[18], int thri, const Real h, const Real v )
     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 = TypeInfo< RealType >::getMaxValue();
+   
+   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 ) == TypeInfo< RealType >::getMaxValue() && 
+       fabs( b ) == TypeInfo< RealType >::getMaxValue() &&
+       fabs( c ) == TypeInfo< RealType >::getMaxValue() )
+      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;
+        }
+    }
+    
+    return false;
+}
 #endif
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 efdb13da8..7d5f8cbaa 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
@@ -27,7 +27,7 @@ template< typename Real,
           typename Anisotropy >
 FastSweepingMethod< Meshes::Grid< 2, Real, Device, Index >, Anisotropy >::
 FastSweepingMethod()
-: maxIterations( 1 )
+: maxIterations( 100 )
 {
    
 }
@@ -236,6 +236,13 @@ solve( const MeshPointer& mesh,
       {
          // 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 );
@@ -244,7 +251,7 @@ solve( const MeshPointer& mesh,
           
           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 ) );
 
@@ -263,7 +270,7 @@ solve( const MeshPointer& mesh,
                        
             CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
                                                              interfaceMapPtr.template getData< Device >(),
-                                                             auxPtr.template modifyData< Device>(),
+                                                             dAux,
                                                              BlockIterDevice );
             
             CudaParallelReduc<<< nBlocks , 512 >>>( BlockIterDevice, dBlock, ( numBlocksX * numBlocksY ) );
@@ -275,14 +282,16 @@ solve( const MeshPointer& mesh,
                 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;
+          //aux = *auxPtr;
+          //interfaceMap = *interfaceMapPtr;
 #endif
       }
       iteration++;
@@ -291,6 +300,23 @@ solve( const MeshPointer& mesh,
 }
 
 #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 )
+{
+    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 ];
+    }
+    
+}
+
 __global__ void CudaParallelReduc( int *BlockIterDevice, int *dBlock, int nBlocks )
 {
     int i = threadIdx.x;
@@ -341,16 +367,17 @@ __global__ void CudaParallelReduc( int *BlockIterDevice, int *dBlock, int nBlock
 template < 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,
-                                      Functions::MeshFunction< Meshes::Grid< 2, Real, Device, Index > >& aux,
+                                      Real *aux,
                                       int *BlockIterDevice )
 {
     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 * 16 + thri;
+    int currentIndex = thrj * blockDim.x + thri;
     
-    __shared__ volatile bool changed[256];
+    //__shared__ volatile bool changed[ blockDim.x*blockDim.y ];
+    __shared__ volatile bool changed[16*16];
     changed[ currentIndex ] = false;
     
     if( thrj == 0 && thri == 0 )
@@ -365,7 +392,8 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         hy = mesh.getSpaceSteps().y();
     }
     
-    __shared__ volatile Real sArray[ 18 ][ 18 ];
+    //__shared__ volatile Real sArray[ blockDim.y+2 ][ blockDim.x+2 ];
+    __shared__ volatile Real sArray[18][18];
     sArray[thrj][thri] = TypeInfo< Real >::getMaxValue();
     
     //filling sArray edges
@@ -440,23 +468,28 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         {
             if( ! interfaceMap[ j * mesh.getDimensions().x() + i ] )
             {
-                changed[ currentIndex ] = ptr.updateCell( sArray, thri+1, thrj+1, hx, hy );
+                changed[ currentIndex ] = ptr.updateCell( sArray, thri+1, thrj+1, hx,hy);
             }
         }
         __syncthreads();
         
-        /*if( thri == 0 && thrj == 0 && blIdx == 1 && blIdy == 3 ){
-            for( int h = 15; h > -1; h-- ){
-                for( int g = 0; g < 16; g++ ){
-                    printf( "%d\t", changed[h*16+g] );
-                }
-                printf("\n");
+    //pyramid reduction
+        if( blockDim.x*blockDim.y == 1024 )
+        {
+            if( currentIndex < 512 )
+            {
+                changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 512 ];
             }
-            printf("\n");
         }
-        __syncthreads();*/
-        
-    //pyramid reduction
+        __syncthreads();
+        if( blockDim.x*blockDim.y >= 512 )
+        {
+            if( currentIndex < 256 )
+            {
+                changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 256 ];
+            }
+        }
+        __syncthreads();
         if( blockDim.x*blockDim.y >= 256 )
         {
             if( currentIndex < 128 )
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 e9fa72e66..c20d9a2fa 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
@@ -255,13 +255,34 @@ solve( const MeshPointer& mesh,
               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();
-          
+                 
           tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr;
-          for( int k = 0; k < numBlocksX; k++)
-          CudaUpdateCellCaller<<< gridSize, blockSize >>>( ptr,
-                                                                                  interfaceMapPtr.template getData< Device >(),
-                                                                                  auxPtr.template modifyData< Device>() );
+          
+          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 ];*/
+            
+          }
+          cudaFree( BlockIterDevice );
+          cudaFree( dBlock );
           cudaDeviceSynchronize();
           TNL_CHECK_CUDA_DEVICE;
           aux = *auxPtr;
@@ -280,27 +301,211 @@ solve( const MeshPointer& mesh,
 template < 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 )
+                                      Functions::MeshFunction< Meshes::Grid< 3, Real, Device, Index > >& aux,
+                                      int *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;
+    
+    __shared__ volatile bool changed[8*8*8];
+    changed[ currentIndex ] = false;
+    
+    if( thrj == 0 && thri == 0 && thrk == 0 )
+        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;
+    if( thrj == 1 && thri == 1 && thrk == 1 )
+    {
+        hx = mesh.getSpaceSteps().x();
+        hy = mesh.getSpaceSteps().y();
+        hz = mesh.getSpaceSteps().z();
+    }
+    __shared__ volatile Real sArray[10][10][10];
+    sArray[thrk][thrj][thri] = TypeInfo< Real >::getMaxValue();
+    if(thri == 0 )
+    {
+        sArray[8][thrj+1][thrk+1] = TypeInfo< Real >::getMaxValue();
+        sArray[9][thrj+1][thrk+1] = TypeInfo< Real >::getMaxValue();
+        sArray[thrk+1][thrj+1][8] = TypeInfo< Real >::getMaxValue();
+        sArray[thrk+1][thrj+1][9] = TypeInfo< Real >::getMaxValue();
+        sArray[thrj+1][8][thrk+1] = TypeInfo< Real >::getMaxValue();
+        sArray[thrj+1][9][thrk+1] = TypeInfo< Real >::getMaxValue();
+    }
+            
+    //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;
+    }
+    __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] = TypeInfo< Real >::getMaxValue();
+    }
+    
+    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] = TypeInfo< Real >::getMaxValue();
+    }
+    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] = TypeInfo< Real >::getMaxValue();
+    }
+    
+    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] = TypeInfo< Real >::getMaxValue();
+    }
+    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] = TypeInfo< Real >::getMaxValue();
+    }
+    
+    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] = TypeInfo< Real >::getMaxValue();
+    }
     
     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();
-        //tnlDirectEikonalMethodsBase< Meshes::Grid< 3, Real, Device, Index > > ptr;
-        for( int l = 0; l < 10; l++ )
+        sArray[thrk+1][thrj+1][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( cell ) )
+            if( ! interfaceMap[ k*dimX*dimY + j * mesh.getDimensions().x() + i ] )
             {
-               ptr.updateCell( aux, cell );
+                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 )
+        {
+            if( currentIndex < 256 )
+            {
+                changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 256 ];
+            }
+        }
+        __syncthreads();
+        if( blockDim.x*blockDim.y*blockDim.z >= 256 )
+        {
+            if( currentIndex < 128 )
+            {
+                changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 128 ];
+            }
+        }
+        __syncthreads();
+        if( blockDim.x*blockDim.y*blockDim.z >= 128 )
+        {
+            if( currentIndex < 64 )
+            {
+                changed[ currentIndex ] = changed[ currentIndex ] || changed[ currentIndex + 64 ];
+            }
+        }
+        __syncthreads();
+        if( currentIndex < 32 ) //POUZE IF JSOU SINCHRONNI NA JEDNOM WARPU
+        {
+            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++){
+                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 )
+        {
+            //loopcounter++;
+            BlockIterDevice[ blIdz * numOfBlockx * numOfBlocky + blIdy * numOfBlockx + blIdx ] = 1;
+        }
+        __syncthreads();
+        /*if(thri == 0 && thrj==0 && thrk==0)
+            printf("%i \n",loopcounter);
+        if(loopcounter == 500)
+            break;*/
     }
-}
+  
+    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 ];
+}   
 #endif
-- 
GitLab


From 9609830d5d6c7859a3cd249e61c7413e53aa6287 Mon Sep 17 00:00:00 2001
From: Tomas Oberhuber <tomas.oberhuber@fjfi.cvut.cz>
Date: Tue, 4 Sep 2018 16:56:27 +0200
Subject: [PATCH 28/28] Fixes after merge.

---
 .../tnlDirectEikonalMethodsBase_impl.h        | 12 ++++-----
 .../tnlFastSweepingMethod1D_impl.h            |  6 ++---
 .../tnlFastSweepingMethod2D_impl.h            | 10 +++----
 .../tnlFastSweepingMethod3D_impl.h            | 26 +++++++++----------
 src/TNL/Math.h                                |  7 ++++-
 5 files changed, 33 insertions(+), 28 deletions(-)

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 626282b02..649a5ad43 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
@@ -716,8 +716,8 @@ __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 1,
 
 
         output[ cind ] =
-               input( cell ) >= 0 ? std::numeric_limits< RealType >::max() :
-                                    - std::numeric_limits< RealType >::max();
+               input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
+                                    - std::numeric_limits< Real >::max();
         interfaceMap[ cind ] = false; 
 
         const Real& h = mesh.getSpaceSteps().x();
@@ -768,8 +768,8 @@ __global__ void CudaInitCaller( const Functions::MeshFunction< Meshes::Grid< 2,
 
 
         output[ cind ] =
-               input( cell ) >= 0 ? std::numeric_limits< RealType >::max() :
-                                    - std::numeric_limits< RealType >::max();
+               input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
+                                    - std::numeric_limits< Real >::max();
         interfaceMap[ cind ] = false; 
 
         const Real& hx = mesh.getSpaceSteps().x();
@@ -841,8 +841,8 @@ __global__ void CudaInitCaller3d( const Functions::MeshFunction< Meshes::Grid< 3
 
 
         output[ cind ] =
-               input( cell ) >= 0 ? std::numeric_limits< RealType >::max() :
-                                    - std::numeric_limits< RealType >::max();
+               input( cell ) >= 0 ? std::numeric_limits< Real >::max() :
+                                    - std::numeric_limits< Real >::max();
         interfaceMap[ cind ] = false; 
         cell.refresh();
 
diff --git a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
index 87974c684..890c6cb4c 100644
--- a/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
+++ b/src/TNL/Experimental/Hamilton-Jacobi/Solvers/hamilton-jacobi/tnlFastSweepingMethod1D_impl.h
@@ -170,7 +170,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
     }
     
     __shared__ volatile Real sArray[ 18 ];
-    sArray[thri] = TypeInfo< Real >::getMaxValue();
+    sArray[thri] = std::numeric_limits<  Real >::max();
     
     //filling sArray edges
     int dimX = mesh.getDimensions().x(); 
@@ -191,7 +191,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         if( dimX > (blIdx+1) * blockDim.x )
             sArray[xkolik] = aux[ blIdx*blockDim.x - 1 + xkolik ];
         else
-            sArray[xkolik] = TypeInfo< Real >::getMaxValue();
+            sArray[xkolik] = std::numeric_limits< Real >::max();
     }
     
     if( thri == 1 )
@@ -199,7 +199,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         if( blIdx != 0 )
             sArray[0] = aux[ blIdx*blockDim.x - 1 ];
         else
-            sArray[0] = TypeInfo< Real >::getMaxValue();
+            sArray[0] = std::numeric_limits< Real >::max();
     }
     
         
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 c390bde69..6703843c1 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
@@ -393,7 +393,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
     
     //__shared__ volatile Real sArray[ blockDim.y+2 ][ blockDim.x+2 ];
     __shared__ volatile Real sArray[18][18];
-    sArray[thrj][thri] = TypeInfo< Real >::getMaxValue();
+    sArray[thrj][thri] = std::numeric_limits< Real >::max();
     
     //filling sArray edges
     int dimX = mesh.getDimensions().x(); int dimY = mesh.getDimensions().y();
@@ -422,7 +422,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         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] = TypeInfo< Real >::getMaxValue();
+            sArray[thrj+1][xkolik] = std::numeric_limits< Real >::max();
     }
     
     if( thri == 1 )
@@ -430,7 +430,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         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] = TypeInfo< Real >::getMaxValue();
+            sArray[thrj+1][0] = std::numeric_limits< Real >::max();
     }
     
     if( thri == 2 )
@@ -438,7 +438,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         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] = TypeInfo< Real >::getMaxValue();
+           sArray[ykolik][thrj+1] = std::numeric_limits< Real >::max();
     }
     
     if( thri == 3 )
@@ -446,7 +446,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         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] = TypeInfo< Real >::getMaxValue();
+            sArray[0][thrj+1] = std::numeric_limits< Real >::max();
     }
     
         
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 c20d9a2fa..b024979cc 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
@@ -328,15 +328,15 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         hz = mesh.getSpaceSteps().z();
     }
     __shared__ volatile Real sArray[10][10][10];
-    sArray[thrk][thrj][thri] = TypeInfo< Real >::getMaxValue();
+    sArray[thrk][thrj][thri] = std::numeric_limits< Real >::max();
     if(thri == 0 )
     {
-        sArray[8][thrj+1][thrk+1] = TypeInfo< Real >::getMaxValue();
-        sArray[9][thrj+1][thrk+1] = TypeInfo< Real >::getMaxValue();
-        sArray[thrk+1][thrj+1][8] = TypeInfo< Real >::getMaxValue();
-        sArray[thrk+1][thrj+1][9] = TypeInfo< Real >::getMaxValue();
-        sArray[thrj+1][8][thrk+1] = TypeInfo< Real >::getMaxValue();
-        sArray[thrj+1][9][thrk+1] = TypeInfo< Real >::getMaxValue();
+        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();
     }
             
     //filling sArray edges
@@ -374,7 +374,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         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] = TypeInfo< Real >::getMaxValue();
+            sArray[thrk+1][thrj+1][0] = std::numeric_limits< Real >::max();
     }
     
     if( thri == 1 )
@@ -382,14 +382,14 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         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] = TypeInfo< Real >::getMaxValue();
+            sArray[thrk+1][thrj+1][9] = 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] = TypeInfo< Real >::getMaxValue();
+            sArray[thrk+1][0][thrj+1] = std::numeric_limits< Real >::max();
     }
     
     if( thri == 3 )
@@ -397,14 +397,14 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         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] = TypeInfo< Real >::getMaxValue();
+            sArray[thrk+1][9][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] = TypeInfo< Real >::getMaxValue();
+            sArray[0][thrj+1][thrk+1] = std::numeric_limits< Real >::max();
     }
     
     if( thri == 5 )
@@ -412,7 +412,7 @@ __global__ void CudaUpdateCellCaller( tnlDirectEikonalMethodsBase< Meshes::Grid<
         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] = TypeInfo< Real >::getMaxValue();
+            sArray[9][thrj+1][thrk+1] = std::numeric_limits< Real >::max();
     }
     
     if( i < mesh.getDimensions().x() && j < mesh.getDimensions().y() && k < mesh.getDimensions().z() )
diff --git a/src/TNL/Math.h b/src/TNL/Math.h
index 1f754bcec..67e9f5e08 100644
--- a/src/TNL/Math.h
+++ b/src/TNL/Math.h
@@ -72,7 +72,12 @@ template< class T >
 __cuda_callable__ inline
 T abs( const T& n )
 {
-#if defined(__MIC__)
+#if defined(__CUDA_ARCH__)
+   if( std::is_integral< T >::value )
+      return ::abs( n );
+   else
+      return ::fabs( n );
+#elif defined(__MIC__)
    if( n < ( T ) 0 )
       return -n;
    return n;
-- 
GitLab