Merge from V6_main 19/03/2013

[tools/medcoupling.git] / doc / doxygen / medmem.dox

/*!
\page medmem MEDMEM library

\section medmem_introduction Introduction
\subsection medmem_rationale Rationale for Med Memory

The Med data exchange model (DEM in English) is the format used in the Salome platform for communicating data between different components. It manipulates objects that describe the meshes underlying scientific computations and the value fields lying on these meshes. This data exchange can be achieved either through files using the Med-file formalism or directly through memory with the Med Memory (\c %MEDMEM) library.

The Med libraries are oganized in multiple layers:
- The MED file layer : C and Fortran API to implement mesh and field persistency.
- The MED Memory level C++ API to create and manipulate mesh and field objects in memory.
- Python API generated using SWIG which wraps the complete C++ API of the MED Memory
- CORBA API to simplify distributed computation inside SALOME (Server Side).
- MED Client classes to simplify and optimize interaction of distant objects within the local solver.

Thanks to Med Memory, any component can access a distant mesh or field
object. Two codes running on different machines can thus exchange
meshes and fields. These meshes and fields can easily be read/written in a Med file
format, enabling access to the whole Salome suite of tools
(CAD, meshing, Visualization, other components).

\subsection medmem_outline Outline

In this document, we describe the API of the Med Memory library (available in C++ and in Python). This document is intended for developers who are in charge of integrating existing applications in the Salome platform.

As will be seen in section \ref medmem_api, the API consists of very few classes:

- a MED-file browser : \ref MEDFILEBROWSER_class,
- meshes : \ref mesh ,
- structured meshes : \ref grid ,
- supports and derived classes : \ref support ,
- mesh generation tool : \ref meshing ,
- fields : \ref field ,
- \ref medmem_drivers "drivers for reading and writing" in MED, GIBI,
VTK, EnSight and Porflow files.

All these are detailed in the following sections. The C++ 
formalism will be used for the description in these sections.
 Python syntax is very similar and is given in appendix \ref medmem_sec_python.

\subsection medmem_naming Naming conventions

The naming conventions are rather straightforward, but the user familiar with the Med-File semantics may find that there are a few noticeable differences (see the following section).

- \b cell entity of dimension equal to the mesh dimension (1, 2 or 3).
- \b component in a field, represents a value that is available for each element of the support (for instance : \f$ T \f$, \f$ v_x \f$, \f$ \sigma_{xy} \f$)).
- \b connectivity \b (descending) connectivity table expressing connectivity of dimension \a d elements in terms of list of dimension \a d-1 elements.
- \b connectivity \b (nodal) connectivity table expressing connectivity of dimension \a d elements in terms of list of nodes.
- \b constituent \b entity entity having a dimension smaller than that of the mesh.
- \b coordinates in a mesh, coordinates can be described by strings giving the names of the coordinates, the units of the coordinates, and the type of coordinates ('MED_CART', 'MED_SPHER' or 'MED_CYL').
- \b description string of characters used to describ an object without giving any access to a query method.
- \b dimension Med Memory discriminates the mesh dimension from the space dimension (a surface shape in 3D will have 2 as a mesh dimension).
- \b driver object attached to a mesh or a field to read (resp. write) data from (resp. to) a Med-file.
- \b edge entity of dimension 1 in a 2D mesh.
- \b element  elementary component of a mesh (0D, 1D, 2D or 3D).
- \b entity category giving information on the dimension of elementary components of meshes : node, edge, face (only in 3D) or cell.
- \b face for 3D meshes, faces are the 2D entities.
- \b family support which is composed of a set of groups, which do not intersect each other, and which gives access to those groups.
- \b field array of integer, integer array, real or real array lying on a support (the dimension of the array of values for each element of the support is called the number of components). A field is uniquely defined by its name, its support, its iteration number and its order number. -1 is the default value of those two numbers.
- \b group support with additional access to parent families.
- \b iteration number] information attached to a field that expresses the number of the time step in the computation (-1 is its default value).
- \b name information attached to a mesh, support or field to name it and access to it.
- \b node entity of dimension 0.
- \b order \b number information attached to a field that expresses the number of an internal iteration inside a time step in the computation (-1 is its default value).
- \b support list of elements of the same entity.
- \b type category of an entity (triangle, segment, quadrangle, tetrahedron, hexahedron, etc...).


\subsection medmem_diff Differences with Med-File concepts
Though the %MEDMEM library can recompute a descending connectivity
from a nodal connectivity, %MEDMEM drivers can only read MED files containing the nodal
connectivities of the entities. 
In %MEDMEM, constituent entities are stored as \c MED_FACE
or \c MED_EDGE, whereas in %MED File, they should be stored as \c
  MED_MAILLE. 

The field notion in %MED File and %MEDMEM is quite different. In %MEDMEM
a field is of course defined by its name, but also by its iteration
number and its order number. 
In %MED File a field is only flagged by its name. For instance,
a temperature at times \a t=0.0 s, \a t=1.0 s, \a t=2.0 s will be considered as  a single field in Med File terminology, while it will be considered as three distinct fields in the Med Memory sense.

\subsection medmem_drivers Drivers for reading and writing

MEDMEM supports data exchange in following formats:
- \b GIBI - reading and writing the mesh and the fields in ASCII format.
- \b VTK - writing the mesh and the fields in ASCII and binary formats.
- \b EnSight - reading and writing the mesh and the fields in EnSigth6 and EnSigth GOLD formats (ASCII and binary).
- \b PORFLOW - reading the mesh in ASCII format.

Limitation of length of names in GIBI format is overcome by storing names in the specific string pile of GIBI file.
This pile is used to map actual long names of the data objects to their shortened representation.

\section medmem_api Med Memory API

\subsection medmem_conventions Conventions

- In this document, one refers to the main user documentation
\ref RefManualMedMemory where the variable \c $MED_ROOT_DIR (resp.
\c $MED_SRC_DIR) is the Med Memory directory installation (resp. sources
directory).
- All numberings start at one (take care of array index !).
- When one gets a C (resp. C++) type array (resp. STL container) using a \c {get...} method, one should not modify the array. Access is in read only. To modify a such array (resp. STL container) use a \c {set...} method.
- There are many couple of methods that have similar syntaxes (one singular and one 
plural). The plural method returns an array and the singular one returns one 
particular value in this array (see \c double \c getCoordinate(int i)  and 
\c double* \c getCoordinates() for example). Generally, only the plural version
of the methods are documented in this report.
- Difference between local and global number in mesh element connectivity list : when one talks about an 
element number, one could see \f$ i^{th} \f$ quadrangle (\f$ i^{th} \f$ in quadrangles array : local numbering) or \f$ j^{th} \f$ element (\f$ j^{th} \f$ in all elements array : 
global numbering). These two numberings are equivalent only if one has only one 
geometric type.


\subsection namespaces  Namespaces

Med Memory uses two namespaces : \c MEDMEM which is the general 
namespace where the main classes are defined and \c MED_EN
which defines enums that can be used by an English-speaking 
programer.

\subsection classes Classes
At a basic usage level, the API consists in few classes which are located in
the \c MEDMEM C++ namespace (consult figure \ref fig_UML_light which gives
an UML diagram view of the main Med Memory classes):

- \b MEDFILEBROWSER the class provinding information on meshes and fields conatained in a MED file;
- \b MESH the class containing 2D or 3D mesh objects;
- \b SUPPORT the class containing mainly a list of mesh elements;
- \b FIELD the class template containing list of values lying on a particular support.

\anchor fig_UML_light
\image html UML_light.png " UML diagram of basic Med Memory API classes"
\image latex UML_light.eps " UML diagram of basic Med Memory API classes"


The API of those classes is quite sufficient for most of the component
integrations in the Salome platform. The use of the Med Memory libraries may
make easier the code coupling in the Salome framework. With these classes, it
is possible to:

- read/write meshes and fields from MED-files;
- create fields containing scalar or vectorial values on list of elements
of the mesh;
- communicate these fields between different components;
- read/write such fields.

Note that on the figure \ref fig_UML_light as well as on figure
\ref fig_UML the life cycle of mesh and field objects are independent. Destroying a field will have no effect on the mesh. But the user has to maintain the link: a mesh aggregates a support which aggregates a field. If the user has to delete Med Memory objects, the field has to be deleted first, then the support and finally the mesh.

A more advanced usage of the Med Memory is possible through other classes.
Figure \ref fig_UML gives a complete view of the Med Memory API. It includes :

- \b GROUP a class inherited from the SUPPORT class used to create supports linked to mesh groups. It stores restricted list of elements used to set boundary conditions, initial values.
- \b FAMILY which is used to manipulate a certain kind of support which does not intersect each other.
- \b MESHING which builds meshes from scratch, it can be used to transform meshes from a specific format to the MED format or to integrate a mesher within Salome platform (note that class does not add element or node to a mesh).
- \b GRID which enables the user to manipulate specific functions for structured grid.

\anchor fig_UML
\image html UML_small.png "UML diagram of Med Memory API classes"
\image latex UML_small.eps  "UML diagram of Med Memory API classes"


\subsection medmem_enums Enums
A few enums are defined in the \c MED_EN namespace :

- an enum which describes the way node coordinates or field values are stored,
  - \c MED_FULL_INTERLACE for arrays such that \f$ x_1,y_1,z_1,x_2,y_2,z_2,\ldots,x_n,y_n,z_n \f$;
  - \c MED_NO_INTERLACE for arrays such that \f$ x_1,x_2,\ldots,x_n,y_1,y_2,\ldots,y_n,z_1,z_2,\ldots,z_n \f$;
  - \c MED_UNDEFINED_INTERLACE, the undefined interlacing mode.
  .
- an enum which describes the type of connectivity
  - \c MED_NODAL for nodal connectivity;
  - \c MED_DESCENDING for descending connectivity.
  .

The user has to be aware of the fact that the Med Memory considers only meshes defined by their nodal connectivity. Nevertheless, the user may, after loading a file in memory, ask to the mesh object to calculate the descending connectivity.

- an enum which contains the different mesh entities, \c medEntityMesh, the entries of which being :
 - \c MED_CELL
 - \c MED_FACE
 - \c MED_EDGE
 - \c MED_NODE
 - \c MED_ALL_ENTITIES
 .

In 3D (resp. 2D), the user has to be aware of the fact that only mesh
entities \c MED_CELL and  \c MED_FACE (resp. \c MED_EDGE) are
considered. In 1D, of course only mesh entities \c MED_CELL+ are considered. Using our naming convention (consult \ref medmem_naming), in $1$ D mesh
only \b node and \b cell are considered. In 2D mesh, only \b node,
\b cell and \b edge are considered. Finally in 3D mesh only
\b node, \b cell and \b face are considered.

- The \c medGeometryElement enum which defines geometric types. The
available types are linear and quadratic elements (consult
\ref RefManualMedMemory). The entries of this enum are quite
self-explanatory :
  - \c MED_NONE
  - \c MED_POINT1
  - \c MED_SEG2
  - \c MED_SEG3
  - \c MED_TRIA3
  - \c MED_QUAD4
  - \c MED_TRIA6
  - \c MED_QUAD8
  - \c MED_TETRA4
  - \c MED_PYRA5
  - \c MED_PENTA6
  - \c MED_HEXA8
  - \c MED_TETRA10
  - \c MED_PYRA13
  - \c MED_PENTA15
  - \c MED_HEXA20
  - \c MED_POLYGON
  - \c MED_POLYHEDRA
  - \c MED_ALL_ELEMENTS
  .
The connectivity of all these elements is defined in MED project Web page
http://hammi.extra.cea.fr/static/MED/web_med/logiciels/med-2.3.1/doc/ .


*/

/*!
\page paramedmem ParaMEDMEM library


The ParaMEDMEM library is based on several classes that 
describe the coupling between two MPI-based parallel codes.

The classes that make up the API of the library are :
- communication interface : \ref comm_interface,
- definition of processor groups : \ref processor_group,
- Data Exchange Channel \ref dec, and its implementations :
 - \ref interpkerneldec for a \ref InterpKerRemapGlobal based on intersecting elements volume computation,
 - \ref overlapdec based on intersecting elems volume
 computation when source and target meshes are on same process id

Given two groups of processors groupA (source) and groupB (target), the following code excerpt gives a typical use of the InterpKernelDEC class.

    \code
    ...
    InterpKernelDEC dec(groupA, groupB);//creates the data exchange channel
    dec.attachLocalField(field);//associate the local field
    dec.synchronize();//builds the interpolation matrix
    if (groupA.containsMyRank())
    dec.recvData();//receive the target field
    else if (groupB.containsMyRank())
    dec.sendData();//send the source field
    ...
    \endcode
The generation of the \ref interpolationmatrix "interpolation matrix" and the corresponding matrix-vector products are performed either on the source, either on the target side. 

*/


/*!
\page medmem_install Configuring and Installing MEDMEM from sources

The MEDMEM library can be configured in several manners so that it can run inside or outside the Salome platform. Also, partitioning and parallel functionalities are optional.

The sources of the library are located in the \a MED_SRC directory. 
The first step consists in preparing the configuration of the library :
\verbatim
cd ${MED_SRC}
./build_configure
\endverbatim

This will create a MEDMEM library with link to the SALOME Kernel. If it is desirable to have a standalone version of the library to be used independently from SALOME, use :
\verbatim
cd ${MED_SRC}
./build_configure --without-kernel
\endverbatim

The library can then be configured :
\verbatim
mkdir ../MED_BUILD
cd ../MED_BUILD
../MED_SRC/configure --prefix=`pwd`/../MED_INSTALL
\endverbatim

This will configure the library without splitting functionalities. ParaMEDMEM will be compiled if an MPI version has been found.

The following options can be useful to configure MEDMEM :
- \a --enable-splitter=yes will trigger the compilation of the MEDSPLITTER tool,
- \a --with-metis=${METISDIR} will specify a location for the METIS library,
- \a --with-scotch=${SCOTCHDIR} will specify a location for the SCOTCH library,
- \a --with-med3=${MED3DIR} specifies a location for MED-file library,
- \a --with-hdf5=${HDF5DIR} specifies a location for the HDF5 library (must be the same as that used for the MED-file library)
- \a --with-lam=${LAMDIR} specifies an install path for a LAM MPI library,
- \a --with-mpich=${MPICHDIR} specifies an install path for a MPICH-1 library.
*/