X-Git-Url: http://git.salome-platform.org/gitweb/?a=blobdiff_plain;f=src%2FSMESH%2FSMESH_Gen.cxx;h=0d59428af2f569df698a11901a4806b462c633fd;hb=df57f987dd5df72834dd8d1b8647a275123f7465;hp=f2e4e0d9f6e9a3b493f37cc481698a85925490f0;hpb=d5617f4801e52e4347bae8677611a4517cc3b69e;p=modules%2Fsmesh.git diff --git a/src/SMESH/SMESH_Gen.cxx b/src/SMESH/SMESH_Gen.cxx index f2e4e0d9f..0d59428af 100644 --- a/src/SMESH/SMESH_Gen.cxx +++ b/src/SMESH/SMESH_Gen.cxx @@ -28,6 +28,7 @@ //#define CHRONODEF +#include #include "SMESH_Gen.hxx" #include "SMESH_DriverMesh.hxx" @@ -58,7 +59,6 @@ using namespace std; #include -#include namespace fs = boost::filesystem; // Environment variable separator @@ -160,21 +160,87 @@ SMESH_Mesh* SMESH_Gen::CreateMesh(bool theIsEmbeddedMode) return aMesh; } + +//============================================================================= +/*! + * Algo to run the computation of all the submeshes of a mesh in sequentila + */ +//============================================================================= + +bool SMESH_Gen::sequentialComputeSubMeshes( + SMESH_Mesh & aMesh, + const TopoDS_Shape & aShape, + const ::MeshDimension aDim, + TSetOfInt* aShapesId /*=0*/, + TopTools_IndexedMapOfShape* allowedSubShapes, + SMESH_subMesh::compute_event &computeEvent, + const bool includeSelf, + const bool complexShapeFirst, + const bool aShapeOnly) +{ + MESSAGE("Compute submeshes sequentialy"); + + bool ret = true; + + SMESH_subMeshIteratorPtr smIt; + SMESH_subMesh *shapeSM = aMesh.GetSubMesh(aShape); + + smIt = shapeSM->getDependsOnIterator(includeSelf, !complexShapeFirst); + while ( smIt->more() ) + { + SMESH_subMesh* smToCompute = smIt->next(); + + // do not mesh vertices of a pseudo shape + const TopoDS_Shape& shape = smToCompute->GetSubShape(); + const TopAbs_ShapeEnum shapeType = shape.ShapeType(); + if ( !aMesh.HasShapeToMesh() && shapeType == TopAbs_VERTEX ) + continue; + + // check for preview dimension limitations + if ( aShapesId && GetShapeDim( shapeType ) > (int)aDim ) + { + // clear compute state not to show previous compute errors + // if preview invoked less dimension less than previous + smToCompute->ComputeStateEngine( SMESH_subMesh::CHECK_COMPUTE_STATE ); + continue; + } + + if (smToCompute->GetComputeState() == SMESH_subMesh::READY_TO_COMPUTE) + { + if (_compute_canceled) + return false; + smToCompute->SetAllowedSubShapes( fillAllowed( shapeSM, aShapeOnly, allowedSubShapes )); + setCurrentSubMesh( smToCompute ); + smToCompute->ComputeStateEngine( computeEvent ); + setCurrentSubMesh( nullptr ); + smToCompute->SetAllowedSubShapes( nullptr ); + } + + // we check all the sub-meshes here and detect if any of them failed to compute + if (smToCompute->GetComputeState() == SMESH_subMesh::FAILED_TO_COMPUTE && + ( shapeType != TopAbs_EDGE || !SMESH_Algo::isDegenerated( TopoDS::Edge( shape )))) + ret = false; + else if ( aShapesId ) + aShapesId->insert( smToCompute->GetId() ); + } + //aMesh.GetMeshDS()->Modified(); + return ret; + +}; + //============================================================================= /* - * Parallel compute of a submesh - * This function is used to pass to thread_pool + * compute of a submesh + * This function is passed to the thread pool */ //============================================================================= -const std::function - compute_function([&] (int id, - SMESH_subMesh* sm, + compute_function([] (SMESH_subMesh* sm, SMESH_subMesh::compute_event event, SMESH_subMesh *shapeSM, bool aShapeOnly, @@ -184,9 +250,9 @@ const std::functionGetComputeState() == SMESH_subMesh::READY_TO_COMPUTE) { sm->SetAllowedSubShapes( fillAllowed( shapeSM, aShapeOnly, allowedSubShapes )); - setCurrentSubMesh( sm ); + //setCurrentSubMesh( sm ); sm->ComputeStateEngine(event); - setCurrentSubMesh( nullptr ); + //setCurrentSubMesh( nullptr ); sm->SetAllowedSubShapes( nullptr ); } @@ -195,6 +261,107 @@ const std::functiongetDependsOnIterator(includeSelf, !complexShapeFirst); + while ( smIt->more() ) + { + SMESH_subMesh* smToCompute = smIt->next(); + + // do not mesh vertices of a pseudo shape + const TopoDS_Shape& shape = smToCompute->GetSubShape(); + const TopAbs_ShapeEnum shapeType = shape.ShapeType(); + // Not doing in parallel 1D and 2D meshes + if ( !aMesh.HasShapeToMesh() && shapeType == TopAbs_VERTEX ) + continue; + if(shapeType==TopAbs_FACE||shapeType==TopAbs_EDGE) + aMesh.SetNbThreads(0); + else + aMesh.SetNbThreads(nbThreads); + + + if (shapeType != previousShapeType) { + // Waiting for all threads for the previous type to end + aMesh.wait(); + + std::string file_name; + switch(previousShapeType){ + case TopAbs_FACE: + file_name = "Mesh2D.med"; + break; + case TopAbs_EDGE: + file_name = "Mesh1D.med"; + break; + case TopAbs_VERTEX: + file_name = "Mesh0D.med"; + break; + case TopAbs_SOLID: + default: + file_name = ""; + break; + } + if(file_name != "") + { + fs::path mesh_file = fs::path(aMesh.tmp_folder) / fs::path(file_name); + exportMesh(mesh_file.string(), aMesh, "MESH"); + + } + //Resetting threaded pool info + previousShapeType = shapeType; + } + + // check for preview dimension limitations + if ( aShapesId && GetShapeDim( shapeType ) > (int)aDim ) + { + // clear compute state not to show previous compute errors + // if preview invoked less dimension less than previous + smToCompute->ComputeStateEngine( SMESH_subMesh::CHECK_COMPUTE_STATE ); + continue; + } + boost::asio::post(*(aMesh._pool), std::bind(compute_function, smToCompute, computeEvent, + shapeSM, aShapeOnly, allowedSubShapes, + aShapesId)); + } + + // Waiting for the thread for Solids to finish + aMesh.wait(); + + aMesh.GetMeshDS()->Modified(); + + return ret; +}; + + //============================================================================= /* * Compute a mesh @@ -222,10 +389,6 @@ bool SMESH_Gen::Compute(SMESH_Mesh & aMesh, const bool complexShapeFirst = true; const int globalAlgoDim = 100; - // Pool of thread for computation - if(aMesh.IsParallel()) - aMesh.InitPoolThreads(); - SMESH_subMeshIteratorPtr smIt; // Fix of Issue 22150. Due to !BLSURF->OnlyUnaryInput(), BLSURF computes edges @@ -245,89 +408,22 @@ bool SMESH_Gen::Compute(SMESH_Mesh & aMesh, // =============================================== // Mesh all the sub-shapes starting from vertices // =============================================== - - TopAbs_ShapeEnum previousShapeType = TopAbs_VERTEX; - int nbThreads = aMesh.GetNbThreads(); - MESSAGE("Running mesh with threads: " << nbThreads << " mesher: " << aMesh.GetMesherNbThreads()); - - - smIt = shapeSM->getDependsOnIterator(includeSelf, !complexShapeFirst); - while ( smIt->more() ) - { - SMESH_subMesh* smToCompute = smIt->next(); - - // do not mesh vertices of a pseudo shape - const TopoDS_Shape& shape = smToCompute->GetSubShape(); - const TopAbs_ShapeEnum shapeType = shape.ShapeType(); - if ( !aMesh.HasShapeToMesh() && shapeType == TopAbs_VERTEX ) - continue; - if(shapeType==TopAbs_FACE||shapeType==TopAbs_EDGE) - aMesh.SetNbThreads(0); - else - aMesh.SetNbThreads(nbThreads); - if ((aMesh.IsParallel()||nbThreads!=0) && shapeType != previousShapeType) { - // Waiting for all threads for the previous type to end - aMesh.wait(); - - std::string file_name; - switch(previousShapeType){ - case TopAbs_FACE: - file_name = "Mesh2D.med"; - break; - case TopAbs_EDGE: - file_name = "Mesh1D.med"; - break; - case TopAbs_VERTEX: - file_name = "Mesh0D.med"; - break; - case TopAbs_SOLID: - default: - file_name = ""; - break; - } - if(file_name != "") - { - fs::path mesh_file = fs::path(aMesh.tmp_folder) / fs::path(file_name); - exportMesh(mesh_file.string(), aMesh, "MESH"); - - } - //Resetting threaded pool info - previousShapeType = shapeType; - } - - // check for preview dimension limitations - if ( aShapesId && GetShapeDim( shapeType ) > (int)aDim ) - { - // clear compute state not to show previous compute errors - // if preview invoked less dimension less than previous - smToCompute->ComputeStateEngine( SMESH_subMesh::CHECK_COMPUTE_STATE ); - continue; - } - if(aMesh.IsParallel()) - { - boost::asio::post(*(aMesh._pool), std::bind(compute_function, 1, smToCompute, computeEvent, - shapeSM, aShapeOnly, allowedSubShapes, - aShapesId)); - } else { - compute_function(1 ,smToCompute, computeEvent, - shapeSM, aShapeOnly, allowedSubShapes, - aShapesId); - - if (smToCompute->GetComputeState() == SMESH_subMesh::FAILED_TO_COMPUTE && - ( shapeType != TopAbs_EDGE || !SMESH_Algo::isDegenerated( TopoDS::Edge( shape )))) - ret = false; - else if ( aShapesId ) - aShapesId->insert( smToCompute->GetId() ); - } - } - - // TODO: Check error handling in parallel mode - if(aMesh.IsParallel()){ - // Waiting for the thread for Solids to finish - aMesh.wait(); - } - - aMesh.GetMeshDS()->Modified(); + if (aMesh.IsParallel()) + ret = parallelComputeSubMeshes( + aMesh, aShape, aDim, + aShapesId, allowedSubShapes, + computeEvent, + includeSelf, + complexShapeFirst, + aShapeOnly); + else + ret = sequentialComputeSubMeshes( + aMesh, aShape, aDim, + aShapesId, allowedSubShapes, + computeEvent, + includeSelf, + complexShapeFirst, + aShapeOnly); return ret; }