diff --git a/utils/converters/matlab/load_gmsh2.m b/utils/converters/matlab/load_gmsh2.m index 09db4bb37e60bf685750bc2a9a75314fe3b22b22..722620428538521ee91ead38c8037d1de43cd9cb 100644 --- a/utils/converters/matlab/load_gmsh2.m +++ b/utils/converters/matlab/load_gmsh2.m @@ -1,13 +1,26 @@ -function msh = load_gmsh2(filename) +function msh = load_gmsh4(filename, which) %% Reads a mesh in msh format, version 1 or 2 +% Usage: +% To define all variables m.LINES, M.TRIANGLES, etc +% (Warning: This creates a very large structure. Do you really need it?) +% m = load_gmsh4('a.msh') +% +% To define only certain variables (for example TETS and HEXS) +% m = load_gmsh4('a.msh', [ 5 6]) +% +% To define no variables (i.e. if you prefer to use m.ELE_INFOS(i,2)) +% m = load_gmsh4('a.msh', -1) +% m = load_gmsh4('a.msh', []) +% % Copyright (C) 2007 JP Moitinho de Almeida (moitinho@civil.ist.utl.pt) +% and R Lorphevre(r(point)lorphevre(at)ulg(point)ac(point)be) +% % based on load_gmsh.m supplied with gmsh-2.0 -% Structure msh has the following elements: - -% msh.NODES_PER_TYPE_OF_ELEMENT - Number of nodes for each type of element +% Structure msh always has the following elements: +% % msh.MIN, msh.MAX - Bounding box % msh.nbNod - Number of nodes % msh.nbElm - Total number of elements @@ -20,23 +33,62 @@ function msh = load_gmsh2(filename) % msh.ELE_TAGS(i,j) - Tags of element i (j in 1:msh.ELE_INFOS(i,3)) % msh.ELE_NODES(i,j) - Nodes of element i (j in 1:k, with % k = msh.NODES_PER_TYPE_OF_ELEMENT(msh.ELE_INFOS(i,2))) - -msh.MIN = [ inf; inf; inf ]; -msh.MAX = [ -inf; -inf; -inf ]; - % +% These elements are created if requested: +% +% msh.nbLines - number of 2 node lines +% msh.LINES(i,1:2) - nodes of line i +% msh.LINES(i,3) - tag (WHICH ?????) of line i +% +% msh.nbTriangles - number of 2 node triangles +% msh.TRIANGLES(i,1:3) - nodes of triangle i +% msh.TRIANGLES(i,4) - tag (WHICH ?????) of triangle i +% +% Etc + % These definitions need to be updated when new elemens types are added to gmsh % -% 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 -% 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 -msh.NODES_PER_TYPE_OF_ELEMENT = [ - 2 3 4 4 8 6 5 3 6 9 10 27 18 14 1 8 20 15 13 ]; +% msh.Types{i}{1} Number of an element of type i +% msh.Types{i}{2} Dimension (0D, 1D, 2D or 3D) of element of type i +% msh.Types{i}{3} Name to add to the structure with elements of type i +% msh.Types{i}{4} Name to add to the structure with the number of elements of type i % -% 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 -% 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 -Dim = [ 1 2 2 3 3 3 3 1 2 2 3 3 3 3 0 3 3 3 3 ]; -ntypes = length(msh.NODES_PER_TYPE_OF_ELEMENT); +nargchk(1, 2, nargin); + +msh.Types = { ... + { 2, 1, 'LINES', 'nbLines'}, ... % 1 + { 3, 2, 'TRIANGLES', 'nbTriangles'}, ... + { 4, 2, 'QUADS', 'nbQuads'}, ... + { 4, 3, 'TETS', 'nbTets'}, ... + { 8, 3, 'HEXAS', 'nbHexas'}, ... %5 + { 6, 3, 'PRISMS', 'nbPrisms'}, ... + { 5, 3, 'PYRAMIDS', 'nbPyramids'}, ... + { 3, 1, 'LINES3', 'nbLines3'}, ... + { 6, 2, 'TRIANGLES6', 'nbTriangles6'}, ... + { 9, 2, 'QUADS9', 'nbQuads9'}, ... % 10 + { 10, 3, 'TETS10', 'nbTets10'}, ... + { 27, 3, 'HEXAS27', 'nbHexas27'}, ... + { 18, 3, 'PRISMS18', 'nbPrisms18'}, ... + { 14, 3, 'PYRAMIDS14', 'nbPyramids14'}, ... + { 1, 0, 'POINTS', 'nbPoints'}, ... % 15 + { 8, 3, 'QUADS8', 'nbQuads8'}, ... + { 20, 3, 'HEXAS20', 'nbHexas20'}, ... + { 15, 3, 'PRISMS15', 'nbPrisms15'}, ... + { 13, 3, 'PYRAMIDS13', 'nbPyramids13'}, ... +}; + +ntypes = length(msh.Types); + +if (nargin==1) + which = 1:ntypes; +else + if isscalar(which) && which == -1 + which = []; + end +end + +% Could check that "which" is properlly defined.... fid = fopen(filename, 'r'); fileformat = 0; % Assume wrong file @@ -88,55 +140,66 @@ end if strcmp(tline, '$NOD') || strcmp(tline, '$Nodes') msh.nbNod = fscanf(fid, '%d', 1); - msh.POS = zeros(msh.nbNod, 3); - for I=1:msh.nbNod - % iNod = - fscanf(fid, '%d', 1); - X = fscanf(fid, '%g', 3); - % IDS(iNod) = I; % Is it safe to assume that the nodes are sequentially presented? - msh.MAX = max(msh.MAX, X); - msh.MIN = min(msh.MIN, X); - msh.POS(I,:) = X; - end + aux = fscanf(fid, '%g', [4 msh.nbNod]); + msh.POS = aux(2:4,:)'; + numids = max(aux(1,:)); + IDS = zeros(1, numids); + IDS(aux(1,:)) = 1:msh.nbNod; % This may not be an identity + msh.MAX = max(msh.POS); + msh.MIN = min(msh.POS); fgetl(fid); % End previous line fgetl(fid); % Has to be $ENDNOD $EndNodes +else + disp (sprintf('Syntax error (no $Nodes/$NOD) in: %s', filename)); + fileformat = 0; end %% Read elements tline = fgetl(fid); if strcmp(tline,'$ELM') || strcmp(tline, '$Elements') msh.nbElm = fscanf(fid, '%d', 1); - msh.ELE_INFOS = zeros(msh.nbElm, 4); % 1 - id, 2 - type, 3 -tags, 4 - Dimension - msh.ELE_NODES = zeros(msh.nbElm,1); % i - Element, j - ElNodes - if eq(fileformat, 1) + % read all info about elements into aux (it is faster!) + aux = fscanf(fid, '%g', inf); + start = 1; + msh.ELE_INFOS = zeros(msh.nbElm, 4); % 1 - id, 2 - type, 3 - ntags, 4 - Dimension + msh.ELE_NODES = zeros(msh.nbElm,6); % i - Element, j - ElNodes + if (fileformat == 1) ntags = 2; else ntags = 3; % just a prediction end - msh.ELE_TAGS = zeros(msh.nbElm, ntags); + msh.ELE_TAGS = zeros(msh.nbElm, ntags); % format 1: 1 - physical number, 2 - geometrical number + % format 2: 1 - physical number, 2 - geometrical number, 3 - mesh partition number msh.nbType = zeros(ntypes,1); for I = 1:msh.nbElm if (fileformat == 2) - msh.ELE_INFOS(I, 1:3) = fscanf(fid, '%d', 3); - ntags = msh.ELE_INFOS(I,3); - msh.ELE_TAGS(I, 1:ntags) = fscanf(fid, '%d', ntags); + finnish = start + 2; + msh.ELE_INFOS(I, 1:3) = aux(start:finnish); + ntags = aux(finnish); + start = finnish + 1; + finnish = start + ntags -1; + msh.ELE_TAGS(I, 1:ntags) = aux(start:finnish); + start = finnish + 1; else - aux = fscanf(fid, '%d', 5); - msh.ELE_INFOS(I, 1:2) = aux(1:2); - msh.ELE_INFOS(I, 3) = 2; - msh.ELE_TAGS(I, :) = aux(3:4); + finnish = start + 1; + msh.ELE_INFOS(I, 1:2) = aux(start:finnish); + start = finnish + 1; % the third element is nnodes, which we get from the type + msh.ELE_INFOS(I, 3) = 2; + finnish = start + 1; + msh.ELE_TAGS(I, 1:2) = aux(start:finnish); + start = finnish + 2; end type = msh.ELE_INFOS(I, 2); msh.nbType(type) = msh.nbType(type) + 1; - msh.ELE_INFOS(I, 4) = Dim(type); - nnodes = msh.NODES_PER_TYPE_OF_ELEMENT(type); - NODES_ELEM = fscanf(fid, '%d', nnodes); - msh.ELE_NODES(I, 1:nnodes) = NODES_ELEM; + msh.ELE_INFOS(I, 4) = msh.Types{type}{2}; + nnodes = msh.Types{type}{1}; + finnish = start + nnodes - 1; + msh.ELE_NODES(I, 1:nnodes) = IDS(aux(start:finnish)); + start=finnish + 1; end - fgetl(fid); % End previous line fgetl(fid); % Has to be $ENDELM or $EndElements else - disp('Error reading elements'); + disp (sprintf('Syntax error (no $Elements/$ELM) in: %s', filename)); fileformat = 0; end @@ -145,4 +208,31 @@ if (fileformat == 0) end fclose(fid); + +%% This is used to create the explicit lists for types of elements + +for i = which + if (~isempty(msh.Types{i}{3})) + cmd = sprintf('msh.%s=msh.nbType(%d);', msh.Types{i}{4}, i); + eval(cmd); + % Dimension + cmd = sprintf('msh.%s=zeros(%d,%d);', msh.Types{i}{3}, msh.nbType(i), msh.Types{i}{1}+1); + eval(cmd); + % Clear nbType for counting, next loop will recompute it + msh.nbType(i) = 0; + end +end + +for i = 1:msh.nbElm + type = msh.ELE_INFOS(i,2); + if (find(which == type)) + if (~isempty(msh.Types{type}{3})) + msh.nbType(type) = msh.nbType(type)+1; + aux=[ msh.ELE_NODES(i,1:msh.Types{type}{1}), msh.ELE_TAGS(i,1) ]; + cmd = sprintf('msh.%s(%d,:)=aux;', msh.Types{type}{3}, msh.nbType(type)); + eval(cmd); + end + end +end + return;