Contiguous storage of CEdge(s) #959
Conversation
pcarruscag
left a comment
pcarruscag
left a comment
There was a problem hiding this comment.
The gist of the PR:
| CPrimalGrid*** bound; /*!< \brief Boundary vector (primal grid information). */ | ||
| CPoint** node; /*!< \brief Node vector (dual grid information). */ | ||
| CEdge** edge; /*!< \brief Edge vector (dual grid information). */ | ||
| CEdge* edges; /*!< \brief Edge vector (dual grid information). */ |
There was a problem hiding this comment.
Significant change 1.
| Index_t iPoint = geometry.edge[iEdge]->GetNode(0); | ||
| Index_t jPoint = geometry.edge[iEdge]->GetNode(1); | ||
| Index_t iPoint = geometry.edges->GetNode(iEdge,0); | ||
| Index_t jPoint = geometry.edges->GetNode(iEdge,1); |
There was a problem hiding this comment.
Significant change 2.
…we must be doing...)
…ld not track root cause with valgrind)
| /*--- Menter's first blending function ---*/ | ||
|
|
||
| if(sst) visc_numerics->SetF1blending(nodes->GetF1blending(iPoint), nodes->GetF1blending(iPoint)); |
There was a problem hiding this comment.
The F1 blending factor was missing in BC_Fluid_Interface of the SST solver (classic copy paste bug...), which is what was causing a bunch of valgrind warnings.
I moved the entire method to CTurbSolver, I rather have one if(sst) than 1 copy pasted method.
By the way, the way this method includes the viscous residual is very fishy... But that is not something I will fix.
talbring
left a comment
talbring
left a comment
There was a problem hiding this comment.
Thanks @pcarruscag! All of the changes make sense. Just two very small questions below.
| for (unsigned short iDim = 0; iDim < nDim; iDim++) | ||
| Normal[iDim]=val_face_normal[iDim]; | ||
| template<class T> | ||
| void SetNormal(unsigned long iEdge, const T& normal) { |
There was a problem hiding this comment.
Why are they templated? To pass different array types? 👍
There was a problem hiding this comment.
That was the idea yes, in case we need to pass std::vector or std::array or even just an object with operator []
|
I'll merge this one to make the diff on my other PR's cleaner, but feel free to comment if you see something that does not look right. |
Proposed Changes
Make CEdge a class of arrays, instead of using it as an array of classes.
Ended up having to add some const-correctness to CNumerics too.
Forgot to turn off "auto blank space stripping" so a few files will show more changes than this actually introduces.
Related Work
Needed for #789 (to eventually support the SIMD part).
Similar changes to #753.
Fixes a bug in the SST fluid-fluid interface.
PR Checklist