| 1 | /*
|
|---|
| 2 | * CreateElementsNodesAndMaterialsPrognostic.c:
|
|---|
| 3 | */
|
|---|
| 4 |
|
|---|
| 5 | #undef __FUNCT__
|
|---|
| 6 | #define __FUNCT__ "CreateElementsNodesAndMaterialsPrognostic"
|
|---|
| 7 |
|
|---|
| 8 | #include "../../DataSet/DataSet.h"
|
|---|
| 9 | #include "../../toolkits/toolkits.h"
|
|---|
| 10 | #include "../../EnumDefinitions/EnumDefinitions.h"
|
|---|
| 11 | #include "../../objects/objects.h"
|
|---|
| 12 | #include "../../shared/shared.h"
|
|---|
| 13 | #include "../../MeshPartitionx/MeshPartitionx.h"
|
|---|
| 14 | #include "../IoModel.h"
|
|---|
| 15 |
|
|---|
| 16 |
|
|---|
| 17 | void CreateElementsNodesAndMaterialsPrognostic(DataSet** pelements,DataSet** pnodes, DataSet** pmaterials, IoModel* iomodel,ConstDataHandle iomodel_handle){
|
|---|
| 18 |
|
|---|
| 19 |
|
|---|
| 20 | /*output: int* epart, int* my_grids, double* my_bordergrids*/
|
|---|
| 21 |
|
|---|
| 22 |
|
|---|
| 23 | int i,j,k,n;
|
|---|
| 24 | extern int my_rank;
|
|---|
| 25 | extern int num_procs;
|
|---|
| 26 |
|
|---|
| 27 | /*DataSets: */
|
|---|
| 28 | DataSet* elements = NULL;
|
|---|
| 29 | DataSet* nodes = NULL;
|
|---|
| 30 | DataSet* materials = NULL;
|
|---|
| 31 |
|
|---|
| 32 | /*Objects: */
|
|---|
| 33 | Node* node = NULL;
|
|---|
| 34 | Tria* tria = NULL;
|
|---|
| 35 | Penta* penta = NULL;
|
|---|
| 36 | Matice* matice = NULL;
|
|---|
| 37 | Matpar* matpar = NULL;
|
|---|
| 38 |
|
|---|
| 39 | /*output: */
|
|---|
| 40 | int* epart=NULL; //element partitioning.
|
|---|
| 41 | int* npart=NULL; //node partitioning.
|
|---|
| 42 | int* my_grids=NULL;
|
|---|
| 43 | double* my_bordergrids=NULL;
|
|---|
| 44 |
|
|---|
| 45 |
|
|---|
| 46 | /*intermediary: */
|
|---|
| 47 | int elements_width; //size of elements
|
|---|
| 48 | double B_avg;
|
|---|
| 49 |
|
|---|
| 50 | /*tria constructor input: */
|
|---|
| 51 | int tria_id;
|
|---|
| 52 | int tria_mid;
|
|---|
| 53 | int tria_mparid;
|
|---|
| 54 | int tria_g[3];
|
|---|
| 55 | double tria_h[3];
|
|---|
| 56 | double tria_s[3];
|
|---|
| 57 | double tria_b[3];
|
|---|
| 58 | double tria_k[3];
|
|---|
| 59 | double tria_melting[3];
|
|---|
| 60 | double tria_accumulation[3];
|
|---|
| 61 | double tria_geothermalflux[3];
|
|---|
| 62 | int tria_friction_type;
|
|---|
| 63 | double tria_p;
|
|---|
| 64 | double tria_q;
|
|---|
| 65 | int tria_shelf;
|
|---|
| 66 | double tria_meanvel;/*!scaling ratio for velocities*/
|
|---|
| 67 | double tria_epsvel; /*!minimum velocity to avoid infinite velocity ratios*/
|
|---|
| 68 | double tria_viscosity_overshoot;
|
|---|
| 69 | int tria_artdiff;
|
|---|
| 70 | bool tria_onwater;
|
|---|
| 71 |
|
|---|
| 72 | /*matice constructor input: */
|
|---|
| 73 | int matice_mid;
|
|---|
| 74 | double matice_B;
|
|---|
| 75 | double matice_n;
|
|---|
| 76 |
|
|---|
| 77 | /*penta constructor input: */
|
|---|
| 78 |
|
|---|
| 79 | int penta_id;
|
|---|
| 80 | int penta_mid;
|
|---|
| 81 | int penta_mparid;
|
|---|
| 82 | int penta_g[6];
|
|---|
| 83 | double penta_h[6];
|
|---|
| 84 | double penta_s[6];
|
|---|
| 85 | double penta_b[6];
|
|---|
| 86 | double penta_k[6];
|
|---|
| 87 | int penta_friction_type;
|
|---|
| 88 | double penta_p;
|
|---|
| 89 | double penta_q;
|
|---|
| 90 | int penta_shelf;
|
|---|
| 91 | int penta_onbed;
|
|---|
| 92 | int penta_onsurface;
|
|---|
| 93 | double penta_meanvel;/*!scaling ratio for velocities*/
|
|---|
| 94 | double penta_epsvel; /*!minimum velocity to avoid infinite velocity ratios*/
|
|---|
| 95 | int penta_collapse;
|
|---|
| 96 | double penta_melting[6];
|
|---|
| 97 | double penta_accumulation[6];
|
|---|
| 98 | double penta_geothermalflux[6];
|
|---|
| 99 | int penta_artdiff;
|
|---|
| 100 | int penta_thermal_steadystate;
|
|---|
| 101 | double penta_viscosity_overshoot;
|
|---|
| 102 | double penta_stokesreconditioning;
|
|---|
| 103 | bool penta_onwater;
|
|---|
| 104 |
|
|---|
| 105 | /*matpar constructor input: */
|
|---|
| 106 | int matpar_mid;
|
|---|
| 107 | double matpar_rho_ice;
|
|---|
| 108 | double matpar_rho_water;
|
|---|
| 109 | double matpar_heatcapacity;
|
|---|
| 110 | double matpar_thermalconductivity;
|
|---|
| 111 | double matpar_latentheat;
|
|---|
| 112 | double matpar_beta;
|
|---|
| 113 | double matpar_meltingpoint;
|
|---|
| 114 | double matpar_mixed_layer_capacity;
|
|---|
| 115 | double matpar_thermal_exchange_velocity;
|
|---|
| 116 | double matpar_g;
|
|---|
| 117 |
|
|---|
| 118 | /* node constructor input: */
|
|---|
| 119 | int node_id;
|
|---|
| 120 | int node_partitionborder=0;
|
|---|
| 121 | double node_x[3];
|
|---|
| 122 | double node_sigma;
|
|---|
| 123 | int node_onbed;
|
|---|
| 124 | int node_onsurface;
|
|---|
| 125 | int node_onshelf;
|
|---|
| 126 | int node_onsheet;
|
|---|
| 127 | int node_upper_node_id;
|
|---|
| 128 | int node_numdofs;
|
|---|
| 129 |
|
|---|
| 130 |
|
|---|
| 131 | #ifdef _PARALLEL_
|
|---|
| 132 | /*Metis partitioning: */
|
|---|
| 133 | int range;
|
|---|
| 134 | Vec gridborder=NULL;
|
|---|
| 135 | int my_numgrids;
|
|---|
| 136 | int* all_numgrids=NULL;
|
|---|
| 137 | int gridcount;
|
|---|
| 138 | int count;
|
|---|
| 139 | #endif
|
|---|
| 140 | int first_grid_index;
|
|---|
| 141 |
|
|---|
| 142 | /*First create the elements, nodes and material properties: */
|
|---|
| 143 | elements = new DataSet(ElementsEnum());
|
|---|
| 144 | nodes = new DataSet(NodesEnum());
|
|---|
| 145 | materials = new DataSet(MaterialsEnum());
|
|---|
| 146 |
|
|---|
| 147 | /*Width of elements: */
|
|---|
| 148 | if(strcmp(iomodel->meshtype,"2d")==0){
|
|---|
| 149 | elements_width=3; //tria elements
|
|---|
| 150 | }
|
|---|
| 151 | else{
|
|---|
| 152 | elements_width=6; //penta elements
|
|---|
| 153 | }
|
|---|
| 154 |
|
|---|
| 155 | #ifdef _PARALLEL_
|
|---|
| 156 | /*Determine parallel partitioning of elements: we use Metis for now. First load the data, then partition*/
|
|---|
| 157 | if(strcmp(iomodel->meshtype,"2d")==0){
|
|---|
| 158 | /*load elements: */
|
|---|
| 159 | IoModelFetchData((void**)&iomodel->elements,NULL,NULL,iomodel_handle,"elements","Matrix","Mat");
|
|---|
| 160 | }
|
|---|
| 161 | else{
|
|---|
| 162 | /*load elements2d: */
|
|---|
| 163 | IoModelFetchData((void**)&iomodel->elements2d,NULL,NULL,iomodel_handle,"elements2d","Matrix","Mat");
|
|---|
| 164 | }
|
|---|
| 165 |
|
|---|
| 166 |
|
|---|
| 167 | MeshPartitionx(&epart, &npart,iomodel->numberofelements,iomodel->numberofnodes,iomodel->elements, iomodel->numberofelements2d,iomodel->numberofnodes2d,iomodel->elements2d,iomodel->numlayers,elements_width, iomodel->meshtype,num_procs);
|
|---|
| 168 |
|
|---|
| 169 | /*Free elements and elements2d: */
|
|---|
| 170 | xfree((void**)&iomodel->elements);
|
|---|
| 171 | xfree((void**)&iomodel->elements2d);
|
|---|
| 172 |
|
|---|
| 173 | /*Used later on: */
|
|---|
| 174 | my_grids=(int*)xcalloc(iomodel->numberofnodes,sizeof(int));
|
|---|
| 175 | #endif
|
|---|
| 176 |
|
|---|
| 177 |
|
|---|
| 178 |
|
|---|
| 179 | /*elements created vary if we are dealing with a 2d mesh, or a 3d mesh: */
|
|---|
| 180 |
|
|---|
| 181 | /*2d mesh: */
|
|---|
| 182 | if (strcmp(iomodel->meshtype,"2d")==0){
|
|---|
| 183 |
|
|---|
| 184 | /*Fetch data needed: */
|
|---|
| 185 | IoModelFetchData((void**)&iomodel->elements,NULL,NULL,iomodel_handle,"elements","Matrix","Mat");
|
|---|
| 186 | IoModelFetchData((void**)&iomodel->thickness,NULL,NULL,iomodel_handle,"thickness","Matrix","Mat");
|
|---|
| 187 | IoModelFetchData((void**)&iomodel->surface,NULL,NULL,iomodel_handle,"surface","Matrix","Mat");
|
|---|
| 188 | IoModelFetchData((void**)&iomodel->bed,NULL,NULL,iomodel_handle,"bed","Matrix","Mat");
|
|---|
| 189 | IoModelFetchData((void**)&iomodel->elementoniceshelf,NULL,NULL,iomodel_handle,"elementoniceshelf","Matrix","Mat");
|
|---|
| 190 | IoModelFetchData((void**)&iomodel->elementonwater,NULL,NULL,iomodel_handle,"elementonwater","Matrix","Mat");
|
|---|
| 191 |
|
|---|
| 192 | for (i=0;i<iomodel->numberofelements;i++){
|
|---|
| 193 |
|
|---|
| 194 | #ifdef _PARALLEL_
|
|---|
| 195 | /*!All elements have been partitioned above, only create elements for this CPU: */
|
|---|
| 196 | if(my_rank==epart[i]){
|
|---|
| 197 | #endif
|
|---|
| 198 |
|
|---|
| 199 |
|
|---|
| 200 | /*ids: */
|
|---|
| 201 | tria_id=i+1; //matlab indexing.
|
|---|
| 202 | tria_mid=-1; //no need for materials
|
|---|
| 203 | tria_mparid=-1; //no need for materials
|
|---|
| 204 |
|
|---|
| 205 | /*vertices offsets: */
|
|---|
| 206 | tria_g[0]=(int)*(iomodel->elements+elements_width*i+0);
|
|---|
| 207 | tria_g[1]=(int)*(iomodel->elements+elements_width*i+1);
|
|---|
| 208 | tria_g[2]=(int)*(iomodel->elements+elements_width*i+2);
|
|---|
| 209 |
|
|---|
| 210 | /*thickness,surface and bed:*/
|
|---|
| 211 | tria_h[0]= *(iomodel->thickness+ ((int)*(iomodel->elements+elements_width*i+0)-1)); //remember, elements is an index of vertices offsets, in matlab indexing.
|
|---|
| 212 | tria_h[1]=*(iomodel->thickness+ ((int)*(iomodel->elements+elements_width*i+1)-1));
|
|---|
| 213 | tria_h[2]=*(iomodel->thickness+ ((int)*(iomodel->elements+elements_width*i+2)-1)) ;
|
|---|
| 214 |
|
|---|
| 215 | tria_s[0]=*(iomodel->surface+ ((int)*(iomodel->elements+elements_width*i+0)-1));
|
|---|
| 216 | tria_s[1]=*(iomodel->surface+ ((int)*(iomodel->elements+elements_width*i+1)-1));
|
|---|
| 217 | tria_s[2]=*(iomodel->surface+ ((int)*(iomodel->elements+elements_width*i+2)-1));
|
|---|
| 218 |
|
|---|
| 219 | tria_b[0]=*(iomodel->bed+ ((int)*(iomodel->elements+elements_width*i+0)-1));
|
|---|
| 220 | tria_b[1]=*(iomodel->bed+ ((int)*(iomodel->elements+elements_width*i+1)-1));
|
|---|
| 221 | tria_b[2]=*(iomodel->bed+ ((int)*(iomodel->elements+elements_width*i+2)-1));
|
|---|
| 222 |
|
|---|
| 223 | /*element on iceshelf?:*/
|
|---|
| 224 | tria_shelf=(int)*(iomodel->elementoniceshelf+i);
|
|---|
| 225 | tria_onwater=(bool)*(iomodel->elementonwater+i);
|
|---|
| 226 | tria_artdiff=iomodel->artificial_diffusivity;
|
|---|
| 227 |
|
|---|
| 228 | /*Create tria element using its constructor:*/
|
|---|
| 229 | tria=new Tria(tria_id, tria_mid, tria_mparid, tria_g, tria_h, tria_s, tria_b, tria_k, tria_melting,tria_accumulation,tria_geothermalflux,tria_friction_type, tria_p, tria_q, tria_shelf, tria_meanvel, tria_epsvel, tria_viscosity_overshoot,tria_artdiff,tria_onwater);
|
|---|
| 230 |
|
|---|
| 231 | /*Add tria element to elements dataset: */
|
|---|
| 232 | elements->AddObject(tria);
|
|---|
| 233 |
|
|---|
| 234 | #ifdef _PARALLEL_
|
|---|
| 235 | /*Now that we are here, we can also start building the list of grids belonging to this node partition: we use
|
|---|
| 236 | *the element index to do this. For each element n, we know index[n][0:2] holds the indices (matlab indexing)
|
|---|
| 237 | into the grid coordinates. If we start plugging 1 into my_grids for each index[n][i] (i=0:2), then my_grids
|
|---|
| 238 | will hold which grids belong to this partition*/
|
|---|
| 239 | my_grids[(int)*(iomodel->elements+elements_width*i+0)-1]=1;
|
|---|
| 240 | my_grids[(int)*(iomodel->elements+elements_width*i+1)-1]=1;
|
|---|
| 241 | my_grids[(int)*(iomodel->elements+elements_width*i+2)-1]=1;
|
|---|
| 242 | #endif
|
|---|
| 243 |
|
|---|
| 244 | #ifdef _PARALLEL_
|
|---|
| 245 | }//if(my_rank==epart[i])
|
|---|
| 246 | #endif
|
|---|
| 247 |
|
|---|
| 248 | }//for (i=0;i<numberofelements;i++)
|
|---|
| 249 |
|
|---|
| 250 |
|
|---|
| 251 | /*Free data : */
|
|---|
| 252 | xfree((void**)&iomodel->elements);
|
|---|
| 253 | xfree((void**)&iomodel->thickness);
|
|---|
| 254 | xfree((void**)&iomodel->surface);
|
|---|
| 255 | xfree((void**)&iomodel->bed);
|
|---|
| 256 | xfree((void**)&iomodel->elementoniceshelf);
|
|---|
| 257 | xfree((void**)&iomodel->elementonwater);
|
|---|
| 258 |
|
|---|
| 259 | }
|
|---|
| 260 | else{ // if (strcmp(meshtype,"2d")==0)
|
|---|
| 261 |
|
|---|
| 262 | /*Fetch data needed: */
|
|---|
| 263 | IoModelFetchData((void**)&iomodel->elements,NULL,NULL,iomodel_handle,"elements","Matrix","Mat");
|
|---|
| 264 | IoModelFetchData((void**)&iomodel->thickness,NULL,NULL,iomodel_handle,"thickness","Matrix","Mat");
|
|---|
| 265 | IoModelFetchData((void**)&iomodel->surface,NULL,NULL,iomodel_handle,"surface","Matrix","Mat");
|
|---|
| 266 | IoModelFetchData((void**)&iomodel->bed,NULL,NULL,iomodel_handle,"bed","Matrix","Mat");
|
|---|
| 267 | IoModelFetchData((void**)&iomodel->elementoniceshelf,NULL,NULL,iomodel_handle,"elementoniceshelf","Matrix","Mat");
|
|---|
| 268 | IoModelFetchData((void**)&iomodel->elementonbed,NULL,NULL,iomodel_handle,"elementonbed","Matrix","Mat");
|
|---|
| 269 | IoModelFetchData((void**)&iomodel->elementonsurface,NULL,NULL,iomodel_handle,"elementonsurface","Matrix","Mat");
|
|---|
| 270 | IoModelFetchData((void**)&iomodel->elementonwater,NULL,NULL,iomodel_handle,"elementonwater","Matrix","Mat");
|
|---|
| 271 |
|
|---|
| 272 | for (i=0;i<iomodel->numberofelements;i++){
|
|---|
| 273 | #ifdef _PARALLEL_
|
|---|
| 274 | /*We are using our element partition to decide which elements will be created on this node: */
|
|---|
| 275 | if(my_rank==epart[i]){
|
|---|
| 276 | #endif
|
|---|
| 277 |
|
|---|
| 278 |
|
|---|
| 279 | /*name and id: */
|
|---|
| 280 | penta_id=i+1; //matlab indexing.
|
|---|
| 281 | penta_mid=-1;
|
|---|
| 282 | penta_mparid=-1; //no need for materials
|
|---|
| 283 |
|
|---|
| 284 | /*vertices,thickness,surface,bed and drag: */
|
|---|
| 285 | for(j=0;j<6;j++){
|
|---|
| 286 | penta_g[j]=(int)*(iomodel->elements+elements_width*i+j);
|
|---|
| 287 | penta_h[j]=*(iomodel->thickness+ ((int)*(iomodel->elements+elements_width*i+j)-1));
|
|---|
| 288 | penta_s[j]=*(iomodel->surface+ ((int)*(iomodel->elements+elements_width*i+j)-1));
|
|---|
| 289 | penta_b[j]=*(iomodel->bed+ ((int)*(iomodel->elements+elements_width*i+j)-1));
|
|---|
| 290 | }
|
|---|
| 291 |
|
|---|
| 292 | /*diverse: */
|
|---|
| 293 | penta_shelf=(int)*(iomodel->elementoniceshelf+i);
|
|---|
| 294 | penta_onbed=(int)*(iomodel->elementonbed+i);
|
|---|
| 295 | penta_onsurface=(int)*(iomodel->elementonsurface+i);
|
|---|
| 296 | penta_collapse=1;
|
|---|
| 297 | penta_artdiff=iomodel->artificial_diffusivity;
|
|---|
| 298 | penta_onwater=(bool)*(iomodel->elementonwater+i);
|
|---|
| 299 |
|
|---|
| 300 |
|
|---|
| 301 | /*Create Penta using its constructor:*/
|
|---|
| 302 | penta= new Penta( penta_id,penta_mid,penta_mparid,penta_g,penta_h,penta_s,penta_b,penta_k,penta_friction_type,
|
|---|
| 303 | penta_p,penta_q,penta_shelf,penta_onbed,penta_onsurface,penta_meanvel,penta_epsvel,
|
|---|
| 304 | penta_collapse,penta_melting,penta_accumulation,penta_geothermalflux,penta_artdiff,
|
|---|
| 305 | penta_thermal_steadystate,penta_viscosity_overshoot,penta_stokesreconditioning,penta_onwater);
|
|---|
| 306 |
|
|---|
| 307 | /*Add penta element to elements dataset: */
|
|---|
| 308 | elements->AddObject(penta);
|
|---|
| 309 |
|
|---|
| 310 | #ifdef _PARALLEL_
|
|---|
| 311 | /*Now that we are here, we can also start building the list of grids belonging to this node partition: we use
|
|---|
| 312 | *the element index to do this. For each element n, we know index[n][0:2] holds the indices (matlab indexing)
|
|---|
| 313 | into the grid coordinates. If we start plugging 1 into my_grids for each index[n][i] (i=0:2), then my_grids
|
|---|
| 314 | will hold which grids belong to this partition*/
|
|---|
| 315 | my_grids[(int)*(iomodel->elements+elements_width*i+0)-1]=1;
|
|---|
| 316 | my_grids[(int)*(iomodel->elements+elements_width*i+1)-1]=1;
|
|---|
| 317 | my_grids[(int)*(iomodel->elements+elements_width*i+2)-1]=1;
|
|---|
| 318 | my_grids[(int)*(iomodel->elements+elements_width*i+3)-1]=1;
|
|---|
| 319 | my_grids[(int)*(iomodel->elements+elements_width*i+4)-1]=1;
|
|---|
| 320 | my_grids[(int)*(iomodel->elements+elements_width*i+5)-1]=1;
|
|---|
| 321 | #endif
|
|---|
| 322 |
|
|---|
| 323 | #ifdef _PARALLEL_
|
|---|
| 324 | }//if(my_rank==epart[i])
|
|---|
| 325 | #endif
|
|---|
| 326 |
|
|---|
| 327 | }//for (i=0;i<numberofelements;i++)
|
|---|
| 328 |
|
|---|
| 329 | /*Free data: */
|
|---|
| 330 | xfree((void**)&iomodel->elements);
|
|---|
| 331 | xfree((void**)&iomodel->thickness);
|
|---|
| 332 | xfree((void**)&iomodel->surface);
|
|---|
| 333 | xfree((void**)&iomodel->bed);
|
|---|
| 334 | xfree((void**)&iomodel->elementoniceshelf);
|
|---|
| 335 | xfree((void**)&iomodel->elementonbed);
|
|---|
| 336 | xfree((void**)&iomodel->elementonsurface);
|
|---|
| 337 | xfree((void**)&iomodel->elementonwater);
|
|---|
| 338 |
|
|---|
| 339 | } //if (strcmp(meshtype,"2d")==0)
|
|---|
| 340 |
|
|---|
| 341 | #ifdef _PARALLEL_
|
|---|
| 342 | /*From the element partitioning, we can determine which grids are on the inside of this cpu's
|
|---|
| 343 | *element partition, and which are on its border with other nodes:*/
|
|---|
| 344 | gridborder=NewVec(iomodel->numberofnodes);
|
|---|
| 345 |
|
|---|
| 346 | for (i=0;i<iomodel->numberofnodes;i++){
|
|---|
| 347 | if(my_grids[i])VecSetValue(gridborder,i,1,ADD_VALUES);
|
|---|
| 348 | }
|
|---|
| 349 | VecAssemblyBegin(gridborder);
|
|---|
| 350 | VecAssemblyEnd(gridborder);
|
|---|
| 351 |
|
|---|
| 352 | #ifdef _ISSM_DEBUG_
|
|---|
| 353 | VecView(gridborder,PETSC_VIEWER_STDOUT_WORLD);
|
|---|
| 354 | #endif
|
|---|
| 355 |
|
|---|
| 356 | VecToMPISerial(&my_bordergrids,gridborder);
|
|---|
| 357 |
|
|---|
| 358 | #ifdef _ISSM_DEBUG_
|
|---|
| 359 | if(my_rank==0){
|
|---|
| 360 | for (i=0;i<iomodel->numberofnodes;i++){
|
|---|
| 361 | printf("Grid id %i Border grid %lf\n",i+1,my_bordergrids[i]);
|
|---|
| 362 | }
|
|---|
| 363 | }
|
|---|
| 364 | #endif
|
|---|
| 365 | #endif
|
|---|
| 366 |
|
|---|
| 367 | /*Add one constant material property to materials: */
|
|---|
| 368 | matpar_mid=1; //put it at the end of the materials
|
|---|
| 369 | matpar_g=iomodel->g;
|
|---|
| 370 | matpar_rho_ice=iomodel->rho_ice;
|
|---|
| 371 | matpar_rho_water=iomodel->rho_water;
|
|---|
| 372 | matpar_thermalconductivity=iomodel->thermalconductivity;
|
|---|
| 373 | matpar_heatcapacity=iomodel->heatcapacity;
|
|---|
| 374 | matpar_latentheat=iomodel->latentheat;
|
|---|
| 375 | matpar_beta=iomodel->beta;
|
|---|
| 376 | matpar_meltingpoint=iomodel->meltingpoint;
|
|---|
| 377 | matpar_mixed_layer_capacity=iomodel->mixed_layer_capacity;
|
|---|
| 378 | matpar_thermal_exchange_velocity=iomodel->thermal_exchange_velocity;
|
|---|
| 379 |
|
|---|
| 380 | /*Create matpar object using its constructor: */
|
|---|
| 381 | matpar=new Matpar(matpar_mid,matpar_rho_ice,matpar_rho_water,matpar_heatcapacity,matpar_thermalconductivity,
|
|---|
| 382 | matpar_latentheat,matpar_beta,matpar_meltingpoint,matpar_mixed_layer_capacity,
|
|---|
| 383 | matpar_thermal_exchange_velocity,matpar_g);
|
|---|
| 384 |
|
|---|
| 385 | /*Add to materials datset: */
|
|---|
| 386 | materials->AddObject(matpar);
|
|---|
| 387 |
|
|---|
| 388 | /*Partition penalties in 3d: */
|
|---|
| 389 | if(strcmp(iomodel->meshtype,"3d")==0){
|
|---|
| 390 |
|
|---|
| 391 | /*Get penalties: */
|
|---|
| 392 | IoModelFetchData((void**)&iomodel->penalties,&iomodel->numpenalties,NULL,iomodel_handle,"penalties","Matrix","Mat");
|
|---|
| 393 |
|
|---|
| 394 | if(iomodel->numpenalties){
|
|---|
| 395 |
|
|---|
| 396 | iomodel->penaltypartitioning=(int*)xmalloc(iomodel->numpenalties*sizeof(int));
|
|---|
| 397 | #ifdef _SERIAL_
|
|---|
| 398 | for(i=0;i<iomodel->numpenalties;i++)iomodel->penaltypartitioning[i]=1;
|
|---|
| 399 | #else
|
|---|
| 400 | for(i=0;i<iomodel->numpenalties;i++)iomodel->penaltypartitioning[i]=-1;
|
|---|
| 401 |
|
|---|
| 402 | for(i=0;i<iomodel->numpenalties;i++){
|
|---|
| 403 | first_grid_index=(int)(*(iomodel->penalties+i*iomodel->numlayers+0)-1);
|
|---|
| 404 | if((my_grids[first_grid_index]==1) && (my_bordergrids[first_grid_index]<=1.0) ) { //this grid belongs to this node's internal partition grids
|
|---|
| 405 | /*All grids that are being penalised belong to this node's internal grid partition.:*/
|
|---|
| 406 | iomodel->penaltypartitioning[i]=1;
|
|---|
| 407 | }
|
|---|
| 408 | if(my_bordergrids[first_grid_index]>1.0) { //this grid belongs to a partition border
|
|---|
| 409 | iomodel->penaltypartitioning[i]=0;
|
|---|
| 410 | }
|
|---|
| 411 | }
|
|---|
| 412 | #endif
|
|---|
| 413 | }
|
|---|
| 414 |
|
|---|
| 415 | /*Free penalties: */
|
|---|
| 416 | xfree((void**)&iomodel->penalties);
|
|---|
| 417 | }
|
|---|
| 418 |
|
|---|
| 419 | /*Ok, let's summarise. Now, every CPU has the following two arrays: my_grids, and my_bordergrids.
|
|---|
| 420 | We can therefore determine which grids are internal to this node's partition
|
|---|
| 421 | and which ones are shared with other nodes because they are on the border of this node's partition. Knowing
|
|---|
| 422 | that, go and create the grids*/
|
|---|
| 423 |
|
|---|
| 424 | /*Create nodes from x,y,z, as well as the spc values on those grids: */
|
|---|
| 425 |
|
|---|
| 426 | /*First fetch data: */
|
|---|
| 427 | if (strcmp(iomodel->meshtype,"3d")==0){
|
|---|
| 428 | IoModelFetchData((void**)&iomodel->deadgrids,NULL,NULL,iomodel_handle,"deadgrids","Matrix","Mat");
|
|---|
| 429 | IoModelFetchData((void**)&iomodel->uppernodes,NULL,NULL,iomodel_handle,"uppergrids","Matrix","Mat");
|
|---|
| 430 | }
|
|---|
| 431 | IoModelFetchData((void**)&iomodel->x,NULL,NULL,iomodel_handle,"x","Matrix","Mat");
|
|---|
| 432 | IoModelFetchData((void**)&iomodel->y,NULL,NULL,iomodel_handle,"y","Matrix","Mat");
|
|---|
| 433 | IoModelFetchData((void**)&iomodel->z,NULL,NULL,iomodel_handle,"z","Matrix","Mat");
|
|---|
| 434 | IoModelFetchData((void**)&iomodel->thickness,NULL,NULL,iomodel_handle,"thickness","Matrix","Mat");
|
|---|
| 435 | IoModelFetchData((void**)&iomodel->bed,NULL,NULL,iomodel_handle,"bed","Matrix","Mat");
|
|---|
| 436 | IoModelFetchData((void**)&iomodel->gridonbed,NULL,NULL,iomodel_handle,"gridonbed","Matrix","Mat");
|
|---|
| 437 | IoModelFetchData((void**)&iomodel->gridonsurface,NULL,NULL,iomodel_handle,"gridonsurface","Matrix","Mat");
|
|---|
| 438 | IoModelFetchData((void**)&iomodel->gridonicesheet,NULL,NULL,iomodel_handle,"gridonicesheet","Matrix","Mat");
|
|---|
| 439 | IoModelFetchData((void**)&iomodel->gridoniceshelf,NULL,NULL,iomodel_handle,"gridoniceshelf","Matrix","Mat");
|
|---|
| 440 |
|
|---|
| 441 |
|
|---|
| 442 | /*Get number of dofs per node: */
|
|---|
| 443 | DistributeNumDofs(&node_numdofs,iomodel->analysis_type,iomodel->sub_analysis_type);
|
|---|
| 444 |
|
|---|
| 445 | for (i=0;i<iomodel->numberofnodes;i++){
|
|---|
| 446 | #ifdef _PARALLEL_
|
|---|
| 447 | /*keep only this partition's nodes:*/
|
|---|
| 448 | if((my_grids[i]==1)){
|
|---|
| 449 | #endif
|
|---|
| 450 |
|
|---|
| 451 | node_id=i+1; //matlab indexing
|
|---|
| 452 |
|
|---|
| 453 |
|
|---|
| 454 |
|
|---|
| 455 | #ifdef _PARALLEL_
|
|---|
| 456 | if(my_bordergrids[i]>1.0) { //this grid belongs to a partition border
|
|---|
| 457 | node_partitionborder=1;
|
|---|
| 458 | }
|
|---|
| 459 | else{
|
|---|
| 460 | node_partitionborder=0;
|
|---|
| 461 | }
|
|---|
| 462 | #else
|
|---|
| 463 | node_partitionborder=0;
|
|---|
| 464 | #endif
|
|---|
| 465 |
|
|---|
| 466 | node_x[0]=iomodel->x[i];
|
|---|
| 467 | node_x[1]=iomodel->y[i];
|
|---|
| 468 | node_x[2]=iomodel->z[i];
|
|---|
| 469 | node_sigma=(iomodel->z[i]-iomodel->bed[i])/(iomodel->thickness[i]);
|
|---|
| 470 |
|
|---|
| 471 | node_onbed=(int)iomodel->gridonbed[i];
|
|---|
| 472 | node_onsurface=(int)iomodel->gridonsurface[i];
|
|---|
| 473 | node_onshelf=(int)iomodel->gridoniceshelf[i];
|
|---|
| 474 | node_onsheet=(int)iomodel->gridonicesheet[i];
|
|---|
| 475 |
|
|---|
| 476 | if (strcmp(iomodel->meshtype,"3d")==0){
|
|---|
| 477 | if (isnan(iomodel->uppernodes[i])){
|
|---|
| 478 | node_upper_node_id=node_id; //nodes on surface do not have upper nodes, only themselves.
|
|---|
| 479 | }
|
|---|
| 480 | else{
|
|---|
| 481 | node_upper_node_id=(int)iomodel->uppernodes[i];
|
|---|
| 482 | }
|
|---|
| 483 | }
|
|---|
| 484 | else{
|
|---|
| 485 | /*If we are running 2d, upper_node does not mean much. Just point towards itself!:*/
|
|---|
| 486 | node_upper_node_id=node_id;
|
|---|
| 487 | }
|
|---|
| 488 |
|
|---|
| 489 | /*Create node using its constructor: */
|
|---|
| 490 | node=new Node(node_id,node_partitionborder,node_numdofs,node_x,node_sigma,node_onbed,node_onsurface,node_upper_node_id,node_onshelf,node_onsheet);
|
|---|
| 491 |
|
|---|
| 492 | /*set single point constraints.: */
|
|---|
| 493 | if (strcmp(iomodel->meshtype,"3d")==0){
|
|---|
| 494 | /*On a 3d mesh, we may have collapsed elements, hence dead grids. Freeze them out: */
|
|---|
| 495 | if (!iomodel->gridonbed[i]){
|
|---|
| 496 | for(k=1;k<=node_numdofs;k++){
|
|---|
| 497 | node->FreezeDof(k);
|
|---|
| 498 | }
|
|---|
| 499 | }
|
|---|
| 500 | }
|
|---|
| 501 | /*Add node to nodes dataset: */
|
|---|
| 502 | nodes->AddObject(node);
|
|---|
| 503 |
|
|---|
| 504 | #ifdef _PARALLEL_
|
|---|
| 505 | } //if((my_grids[i]==1))
|
|---|
| 506 | #endif
|
|---|
| 507 | }
|
|---|
| 508 |
|
|---|
| 509 | /*All our datasets are already order by ids. Set presort flag so that later on, when sorting is requested on these
|
|---|
| 510 | * datasets, it will not be redone: */
|
|---|
| 511 | elements->Presort();
|
|---|
| 512 | nodes->Presort();
|
|---|
| 513 | materials->Presort();
|
|---|
| 514 |
|
|---|
| 515 | /*Clean fetched data: */
|
|---|
| 516 | xfree((void**)&iomodel->deadgrids);
|
|---|
| 517 | xfree((void**)&iomodel->x);
|
|---|
| 518 | xfree((void**)&iomodel->y);
|
|---|
| 519 | xfree((void**)&iomodel->z);
|
|---|
| 520 | xfree((void**)&iomodel->thickness);
|
|---|
| 521 | xfree((void**)&iomodel->bed);
|
|---|
| 522 | xfree((void**)&iomodel->gridonbed);
|
|---|
| 523 | xfree((void**)&iomodel->gridonsurface);
|
|---|
| 524 | xfree((void**)&iomodel->uppernodes);
|
|---|
| 525 | xfree((void**)&iomodel->gridonicesheet);
|
|---|
| 526 | xfree((void**)&iomodel->gridoniceshelf);
|
|---|
| 527 |
|
|---|
| 528 |
|
|---|
| 529 | /*Keep partitioning information into iomodel*/
|
|---|
| 530 | iomodel->epart=epart;
|
|---|
| 531 | iomodel->my_grids=my_grids;
|
|---|
| 532 | iomodel->my_bordergrids=my_bordergrids;
|
|---|
| 533 |
|
|---|
| 534 | /*Free ressources:*/
|
|---|
| 535 | #ifdef _PARALLEL_
|
|---|
| 536 | xfree((void**)&all_numgrids);
|
|---|
| 537 | xfree((void**)&npart);
|
|---|
| 538 | VecFree(&gridborder);
|
|---|
| 539 | #endif
|
|---|
| 540 |
|
|---|
| 541 | cleanup_and_return:
|
|---|
| 542 |
|
|---|
| 543 | /*Assign output pointer: */
|
|---|
| 544 | *pelements=elements;
|
|---|
| 545 | *pnodes=nodes;
|
|---|
| 546 | *pmaterials=materials;
|
|---|
| 547 |
|
|---|
| 548 | }
|
|---|