TODO.txt

 :
0) ()        ,   data()
1) ()  erase  array  ,  resize'
2) ()      ReferenceArray
3) ()     char,     short,    -   ,   - 
4) ()     
5) (  )         
6) ()      
7) ()     chunk_bulk_array -       chunk_array -  ,   
8) ()        
8.1)         
10) ()        
10.-1) ()     Element    getNodes,getEdges,getFaces,getCells
             
10.0) (, )     array,          
10.1) ()     adjacent      (     ID ,    )
10.1.0) (,   , 10.6)       - array ( reference_array)  dynarray (   )
10.1.1) ()        
10.2) ()  ElementSet -     , LowConn  ,    , 
	HighConn   ,  1-  -  ,  -  ,  -    (   INMOST_experimental)
	 tag_set_size,     ,   LowConn  
	    ?
              ,   
10.2.0) ()  ,  
          :
             -    ,     ,  ,  ,  insert
             -   array,  ,  swap    array
          ,   
10.2.1) ()    - , , ,       10.1
               -    ,   
            
10.2.2) ()  ApplyModification
10.2.3) ()   VTK
10.2) ()         ID
10.3) ()    Cell, Face, Edge, Node, ElementSet -      ,   ID    
10.4) () Cell,Face,Edge,Node,ElementSet    ID
10.5) () reference_array     
10.6) ()  createCell, CreateFace.. -   TieElement  new,       10.3
11) () Markers, HighConn, LowConn -    TagManager, ,  ,      
26) ()   
26.1) (     )        radix (  INMOST_experimental)
       25*n    8- ,        16  
     17*n   11- ,     131  
27) ()   Read*   pmf,   
28) ()  std::vector     array
29) ()  std::vector  array   
31) () TagManager & TagManager::assign(Mesh * m, TagManager const & other)          
33) ()  ElementNum,    ElementType   6 
34) () Storage      ,   
36) ( ) ,      Storage     Mesh
50) ()   container.hpp enumerator  size_type
51) ()         container.hpp (  iterator_category?)
41) ( )  HighConn, LowConn  ,        
44) ( )  PackElementsData         tag
46) ( )   .       ,  
19) ()      (   ,     )
20) ()         
63) ()      expr

=====================================================================================
0)     ,  

14)   ReferenceArray?

16)  parallel_storage  
17)   ?


18)      

19.)         incident_matrix.hpp incident_matrix::compute_measure  geometry.cpp Mesh::GetGeometricData

21)  /-      (kd    examples/OldDrawGrid)
22) ResolveShared       
23)       ,    
24)      
25) ,     
30)  - sort  ReorderApply   ,   chunk_array    
    ReorderApply  radix-?
32)      ,   ,   Cell, Face,  Edge  Node 
35)  Bridge* 
37)      container.hpp,        
38)    GatherBoundaryFaces/GatherInteriorFaces   
39) CastRay    ,     
40)     ?
42)    Inside  discr_common.cpp
43)    RestoreCellNodes?
45)           ?
                 . (    )
47)     ,       ,       ?
49)     -    
           
         
      ,        
     ResolveShared
     ModificationResolve
52) ,  radix    big-endian 
53)     SortByGlobalID,   globalid  
54)      ( unit tests)
55)  mpi_send/mpi_recv           -       
56) Enumerate with mask should accept MarkerType select to select elements
57) GlobalID service class with different operation options to help pick new global ids when mesh is modified locally
    options:
     a) global id continuous among processors
     b) global id may have gaps but local interval for each processor is bounded
     c) global id is just a unique number

58)     vtk-   "VTK_WRITE_SETS_TO_FILES" = "YES"

59) Openmp   
60) ResolveShared   MarkerType,        (  ResolveModification)
61) ResolveModification             .

62)    DATA_UNKNOWN   unknown  



64) Automatizator    

65)            Bulk   MarkerType.

66)   

65)      (???)

66)   Map  RowMerger,              (     OrderInfo     )

67)   Graph,      Partitioner    Solver

68) Redistribute   tag 

69)      ghost-  Partitioner::Evaluate,  ReduceData.




====================================================================================


INMOST Mesh
0) TagDenseFixed, TagSparseFixed, TagDenseVariable, TagSparseVariable
	dynamic_variable, static_variable may be templatized with respect to type of the Tag
	
1) rewrite parallel_storage
	(?) RecomputeParallelStorage detect element types in ExchangeMarked
	Either organise separation per element type in ElementSet, or store 4 element sets.
	
2) volume for nonconvex elements
	implement in incedence_matrix in modify.cpp
3) (ok) New mesh format reader, XML
	search tag positions in file
	parallel reader
4) octree structure for intersection of remote nodes
	in ResolveShared
	in UnpackElementsData
	in file reading procedures
5) algorithm for dynamic mesh adaptation in parallel
	ResolveShared should depend on marker
	Algorithm of shortest path to close the gap in shared skin
	OR remap processors numbers from old cells onto new ones using bijection of cell centers
7) Introduce INMOST DataBase, that should be outside
	(?) ElementSet should have data for it's elements, Mesh should become ElementSet
	Move MGetLink and all the mechanism to access data into TagManager
	Move inline functions for Storage from inmost_mesh to inmost_data
9) (test) add DATA_REMOTE_REFERENCE, RemoteHandleType data type and structure that will allow to link another mesh
	(ok) static Mesh::GetMesh(std::string name), structure to collect meshes by name.
        Save multiple meshes to file
10) (test,ok) algorithm in PackTagData for DATA_VARIABLE
	(ok) GetData,SetData,SetDataSize - add DATA_VARIABLE
	(ok) GetDataSize - returns number of entries in array
	(ok) GetDataCapacity - returns space capacity occupied by all the data (specific for DATA_VARIABLE)
	(ok) GetData - puts all the data into allocated array of size GetDataSpace
	(ok) SetData - given the number of entries is known, knows how to fill the internal data 
	(ok) MPI type for DATA_VARIABLE
	Destroy type on Finalize
	Test!
	(ok) Should get capacity of unpacked data, provide function ComputeCapacity
		See next
		(ok) Organize GetDataCapacity with parameter that represents storage and size information
	(ok) Provide tools to retrive data from bytes arrays in Unpack functions
		(ok) variable::RecordEntry - write into array of Row::entry
		(ok) variable::RetriveEntry - get out of array of pairs
		(ok) variable::RetriveSizeEntry - how many entries used to store
		(ok) variable::RequiredSizeEntry - how many entries are needed to store
11) DATA_REFERENCE, DATA_REMOTE_REFERENCE can use PackElements in PackTagData
	Have to make rounds of communication as in ExchangeMarked
	Presence of Tag of type DATA_REFERENCE and DATA_REMOTE_REFERENCE should internally mark the elements, invoke ExhchangeMarked
12) Orientation of edges as directed, order faces with right hand side rule, each face's normal should follow right hand side rule
	Insertion of faces into edge
	Orientation of faces
	Optimized algorithms:
		all cells of the edge (traverse faces, get back cell)
		all nodes of the face (traverse edges, always get first node)
		all faces of the node (traverse only edges that are directed outside or inside, get their faces)
		

INMOST Solvers:
	-1) (ok) Move annotation into matrix as outstanding structure that can be allocated on demand
	0) (no) BasicRow class for autodiff or above
	1) Rewrite BCGS as with Anderson Acceleration
	3) Internal CSRMatrix format
		3.1) Reorderings for CSRMatrix with PQ (RCM,ColAMD,MetisND)
		3.2) Reorderings with execution tree (Mondriaan)
		3.3) CSRMatrix functions - transpose, swap row/column, compute schur with LU, multiply vector, additive schwartz...
		3.4) types for CSRMatrix
		3.5) fast row/col exchange for MPTILUC
	4) Fast laplacian matrix factorization graphs: Horst Simon, C:\Users\KIRILL\Documents\Read\solver\laplacian_graph
	5) condition estimation for ilu2
	6) Purely algebraic CPR (see galerkin principle)
	7) PrepareMatrix should be able to merge rows in parallel, just sum the row on lowest-rank processor, have to return the solution on every processor.

INMOST Autodiff
	0) Gateaux derivatives
	1) Clean old staff
	2) Superproblem definition??
	3) Read about Fenics
	4) (ok) Automatizator::MakeCurrent, Automatizator::GetCurrent, Automatizator::HasCurrent
		(ok) variable should detect presence of current automatizator and access it's structures
	5) Abstraction of KeyValueTable, 
		add cubic spline, 
		add bc treatment
	6) condition_etype_variable, condition_marker_variable - needed in THREEPHASE in interpolation
	7) support std::swap
	8) expression foreach that will iterate over elements or elements handles
	


INMOST Nonlinear
	1. Newton
	2. Line search
	3. Anderson
	4. http://fenicsproject.org/documentation/tutorial/nonlinear.html
	nonlinear solvers: Hans De Sterck

	Nonlinear solvers for INMOST L-BFGS 
		1) Rewrite onephase into class with ComputeFunction, ComputeGradient, ComputeHessian(?)
		2) Create abstract problem class for nonlinear solvers within INMOST
		3) reimplement Newton/LS/AA into INMOST
		4) implement L-BFGS https://github.com/PatWie/CppNumericalSolvers/blob/master/src/LbfgsSolver.cpp
		5) hessian https://en.wikipedia.org/wiki/Hessian_automatic_differentiation


====================================================================================

7.MSPP: () : , , 
8.MSPP: 
9.MSPP: ()    visual studio
10.MSPP:        (    )
11. ()      
12.    (,   )
12. () MSPP:        Mesh
13. MSPP:   Redistribute!!! ExchangeGhost

13. INMOST / MMTK = Mathematical Modelling ToolKit / NSTK

() last_created_element -  


 
1)      -   ,
         ( ),  
    -        
        ,     
  -      
2) small_hash -> dynarray
3)    
4)           
5)   parent()           
6)      
7)   ,  []         
8)        
9)            , 
     (        )
10)    ,      
11)       LLVM  asmjit
12)  openmp 
13)  opencl ,  openmp       
14)        

 
1) -

 
1)     
2)  ILUC2 
3)    /,     
4)    E,F   
5)       EU^-1 L^-1F
6)        
7)      (openmp) 
8)  (openmp)  EU^-1, L^-1F     
9)    nicole spillane
10)    ,       
11)        amls
12)       

13)     Matrix  Row:
Row::MergeRows(linked_list_storage, row b 
Matrix::MergeRow(alpha, row & a, real beta, row &b)
14)  CSRMatrix, CSRRow
15)  Matrix -> CSRMatrix
16) OrderInfo    CSRMatrix
17) 


solver:
read mc64
read nested dissection
try to postpone ReorderEF to the end of run

5.MSPP:     
6.MSPP:    
 6.0  bicgs(l)
6.0 OrderInfo -> ASM_preconditioner?
6.0.1 OrderInfo.PrepareMatrix(A,levels) -> ASM_preconditioner(A,levels) : Method
6.0.3 OrderInfo.PrepareVector(vector) -> Matrix.PrepareVector(vector)
6.0.4 OrderInfo.Update(vector) -> Matrix.Update(vector)
6.0.5 orderInfo.Accumulate(vector) -> Matrix.Accumulate(vector)
6.1     (,  MPI_Waitall->MPI_Waitsome)
6.2 Update(vector) -> UpdateBegin, UpdateEnd,     ,     (INIT,PREC,MATVEC)
6.3    (?)    ilu2 ( superlu?)
6.3.1     
info->PrepareMatrix(A,0)

6.3.2    
6.4    tau 
6.5        
6.6   superlu   
6.7    -    ..



//FOR 2D
// Node -> Vertex
// Edge -> Vertex
// Face -> Line, Curve
// Cell -> Tri, Quad, Polygon, MultiLine

//FOR 3D
// Node -> Vertex
// Edge -> Line, Curve
// Face -> Tri, Quad, Polygon, MultiLine
// Cell -> Tet, Hex, Prism, Pyramid, Polyhedron, MultiPolygon



//TODO:


1D???


//serial
//1.0 read/write formats
//	1.1 read gmv
//	1.2 read different kinds of vtk besides unstructured
//	1.3	read/write xml vtk formats

//  1.4 pmf: print error when markers cannot be represented in current type and data was lost
//  1.5 pmf: save hide_marker, new_marker values
//  1.6 pmf: save topology checking information


//2. enchance iterators
// iterators should return pointers or references?

//3. geometry.cpp
// 3.0 decide how FixNormalOrientation should update geometric data
// 3.1 replace calculation of normale by hyperplane equation for any dimension??
// 3.2 correct Inside function to check weather point inside Segment, Polygon
// 3.3 dynamically balanced obtree or octree or kd-tree to query elements by position


//4. (done, need check) add some low-level mesh modification procedures to elements
//	5.1 add some high-level mesh modification procedures: split by plane, 
//  5.2 csg: difference, union, intersection
//  5.3 calculate volume for concave elements, choose elements by minimum volume, not minimum number of elements in incident_matrix class (modify.cpp)

//5. reduce size used by sparse data

//6. new ElementSet class, derived from Element class
// 6.1 support ierarhy

//7. algorithm that checks topology on element creation
// 7.1 complete unimplemented tests for star-shaped objects, concavity

//parallel:
//1. algorithm in EndModification or ResolveModification to keep the mesh in parallel consistent state after element creation/deletion
//	1.1 restore contiguous global id

//2. test algorithm that checks for halo in ExchangeMarked, if performance increases

//3. decide how to exchange ElementSets between processors

//4. exchange data of type DATA_REFERENCE, by GlobalID and ElementType

//5. mesh_file.cpp
//  5.1 on parallel load, when there are more processors then parts in mesh - duplicate mesh on other processors??
//  5.2 when mesh is duplicated over many processors use local clusterization algorithm in ResolveShared

// shared parallel
//1. avoid markers in nbAdjElements, getAdjElements, getNodes, getEdges, getFaces, getCells, BridgeAdjacency

// partitioner:
//1. implement K-means clustering

//solver:
//0. workaround for overlapping vertices and matrices

//1. read TODO on top of solver_*.hpp

//autodiff:
//1. read TODO on top of INMOST_autodiff.h


//TODO:


// Recheck computation of derivatives!!!!
// Detect similar parts of the tree in Automatizator
// Restructure tree expressions
// Intorduce multivariate tables
// Generate opencl code

// Make so that stencil may be represented by tags, by set or by callback_function

//RegisterTable and table implementation should account for boundary conditions beyond data




//SOLVER todo:

//TODO:
// how to calculate diagonal perturbation?
// how to calculate schur complement faster
// 
// done! implement crout-ILU from 
//   Documents/Read/solver/crout/crout.pdf to ILUC_preconditioner
// done! implement condition estimation for |L^{-1}| and |U^{-1}| and adaptive tau for ILUC_preconditioner from 
//   Documents\Read\solver\Read_now\bollhofer_siam.pdf
// done! try to make ILUC2_preconditioner - second order ILU from ILUC_preconditioner
// done! implement diagonal pivoting for ILUC - maybe need to update diagonal at every step
//   goto references [7],[10]-(data structure!) from
//   Documents\Read\solver\crout\crout.pdf
// return dropped values to diagonal if control vector is provided from
//   Documents\Read\solver\stabilization\milut.pdf
// try to apply dropping while forming linked list,should correctly calculate condition estimates 

// Calculate schur complement faster:
//   Documents\Read\solver\sparse_matmul\sparse.pdf

// in ILUC_preconditioner, replace matrix structures by CSR, estimate number of nonzeros in rows/cols
// before filling, if necessery (how?)