Contiguous storage of CPoint(s) #966
Conversation
pcarruscag
force-pushed
the
contiguous_cpoint
branch
from
9f4d95b to
0cbc58e
Compare
pcarruscag
changed the title
| CPrimalGrid** face; /*!< \brief Face vector (primal grid information). */ | ||
| CPrimalGrid*** bound; /*!< \brief Boundary vector (primal grid information). */ | ||
| CPoint** node; /*!< \brief Node vector (dual grid information). */ | ||
| CPoint* nodes; /*!< \brief Node vector (dual grid information). */ |
| void SetWallDistance(su2double val) override { | ||
| for (unsigned long iPoint = 0; iPoint < GetnPoint(); iPoint++){ | ||
| node[iPoint]->SetWall_Distance(val); | ||
| nodes->SetWall_Distance(iPoint, val); |
| CCompressedSparsePatternUL Point; /*!< \brief Points surrounding the central node of the control volume. */ | ||
| CCompressedSparsePatternL Edge; /*!< \brief Edges that set up a control volume (same sparse structure as Point). */ | ||
| CCompressedSparsePatternL Elem; /*!< \brief Elements that set up a control volume around a node. */ |
There was a problem hiding this comment.
Significant change no 1a: Adjacency information is stored in compressed format (instead of vector of vectors).
| node[jPoint] = new CPoint(Local_Coords[iPoint*nDim+0], | ||
| Local_Coords[iPoint*nDim+1], | ||
| Local_Coords[iPoint*nDim+2], | ||
| Local_to_Global_Point[jPoint], config); | ||
| nodes->SetCoord(jPoint, &Local_Coords[iPoint*nDim]); | ||
| nodes->SetGlobalIndex(jPoint, Local_to_Global_Point[jPoint]); |
There was a problem hiding this comment.
Significant change 2: After allocating the nodes we set data into each instead of instantiating a CPoint object per node.
| for (iPoint = 0; iPoint < nPoint; iPoint++) | ||
| node[iPoint]->SetnNeighbor(node[iPoint]->GetnPoint()); | ||
| SU2_OMP_MASTER | ||
| nodes->SetPoints(points); |
There was a problem hiding this comment.
Significant change 1b: We set the adjacency for all points in one call, the input is a temporary vector<vector> or similar which the function SetPoints compresses, this is only done in 3 places.
| /// TODO: No comms needed for this gradient? The Rmatrix should be allocated somewhere. | ||
|
|
||
| /*--- Deallocate memory ---*/ | ||
| for (iVar = 0; iVar < nDim; iVar++) | ||
| delete [] Cvector[iVar]; | ||
| delete [] Cvector; | ||
| const auto& gridVel = geometry->nodes->GetGridVel(); | ||
| auto& gridVelGrad = geometry->nodes->GetGridVel_Grad(); | ||
| auto rmatrix = CVectorOfMatrix(nPoint,nDim,nDim); | ||
|
|
||
| computeGradientsLeastSquares(nullptr, GRID_VELOCITY, PERIODIC_NONE, *geometry, *config, | ||
| true, gridVel, 0, nDim, gridVelGrad, rmatrix); |
There was a problem hiding this comment.
Usual change 4: Grid velocity gradients are no longer duplicated.
| su2double Coord_3D[3]; | ||
| point_line >> Coord_3D[0]; point_line >> Coord_3D[1]; point_line >> Coord_3D[2]; | ||
| node[ii] = new CPoint(Coord_3D[0], Coord_3D[1], Coord_3D[2], i, config); | ||
| ii++; | ||
| break; | ||
| } | ||
| } | ||
| su2double Coord[3] = {0.0}; | ||
| point_line >> Coord[0]; | ||
| point_line >> Coord[1]; | ||
| if (nDim==3) point_line >> Coord[2]; | ||
| nodes->SetCoord(ii, Coord); | ||
| nodes->SetGlobalIndex(ii, i); |
There was a problem hiding this comment.
@vdweide do the DG regressions cover all the types of mesh formats? I think the changes in this PR only affect the DG solver during mesh reading.
| @@ -42,211 +42,211 @@ CGridAdaptation::CGridAdaptation(CGeometry *geometry, CConfig *config) { | |||
| size = SU2_MPI::GetSize(); | |||
| rank = SU2_MPI::GetRank(); | |||
|
|
|||
| unsigned long iPoint; | |||
There was a problem hiding this comment.
40% of the changes are in this file, I had to replace tabs because of Codefactor
|
|
There was a problem hiding this comment.
Another 20% of the changes are here were I accidentally stripped the spaces...
| CPoint::CPoint(unsigned long npoint, unsigned long ndim) : nDim(ndim) { | ||
|
|
There was a problem hiding this comment.
Significant change 3: There is a "minimal constructor" that only allocates the fields needed to read the mesh file (when geometry_aux is built).
|
Can we get this going folks? |
Proposed Changes
Along the lines of #753 and #959
If this causes someone lots of conflicts when merging, or you have a lot of new code using the old CPoint (that will not compile anymore) it may be helpful to run this python script (refactor_cpoint.txt) before or after respectively, it was used for 80% of the refactoring done here.
PR Checklist