Actual source code: ex92.c
2: static char help[] = "Tests MatIncreaseOverlap(), MatGetSubMatrices() for parallel MatSBAIJ format.\n";
4: #include petscmat.h
8: int main(int argc,char **args)
9: {
10: Mat A,Atrans,sA,*submatA,*submatsA;
12: PetscMPIInt size,rank;
13: PetscInt bs=1,mbs=11,ov=1,i,j,k,*rows,*cols,nd=5,*idx,rstart,rend,sz,mm,M,N,Mbs;
14: PetscScalar *vals,rval,one=1.0;
15: IS *is1,*is2;
16: PetscRandom rand;
17: Vec xx,s1,s2;
18: PetscReal s1norm,s2norm,rnorm,tol = 1.e-10;
19: PetscTruth flg;
20: int stages[2];
22: PetscInitialize(&argc,&args,(char *)0,help);
23: MPI_Comm_size(PETSC_COMM_WORLD,&size);
24: MPI_Comm_rank(PETSC_COMM_WORLD,&rank);
26: PetscOptionsGetInt(PETSC_NULL,"-mat_block_size",&bs,PETSC_NULL);
27: PetscOptionsGetInt(PETSC_NULL,"-mat_size",&mbs,PETSC_NULL);
28: PetscOptionsGetInt(PETSC_NULL,"-ov",&ov,PETSC_NULL);
29: PetscOptionsGetInt(PETSC_NULL,"-nd",&nd,PETSC_NULL);
31: MatCreateMPIBAIJ(PETSC_COMM_WORLD,bs,mbs*bs,mbs*bs,PETSC_DECIDE,PETSC_DECIDE,
32: PETSC_DEFAULT,PETSC_NULL,PETSC_DEFAULT,PETSC_NULL,&A);
33: PetscRandomCreate(PETSC_COMM_WORLD,&rand);
34: PetscRandomSetFromOptions(rand);
36: MatGetOwnershipRange(A,&rstart,&rend);
37: MatGetSize(A,&M,&N);
38: Mbs = M/bs;
40: PetscMalloc(bs*sizeof(PetscInt),&rows);
41: PetscMalloc(bs*sizeof(PetscInt),&cols);
42: PetscMalloc(bs*bs*sizeof(PetscScalar),&vals);
43: PetscMalloc(M*sizeof(PetscScalar),&idx);
45: /* Now set blocks of values */
46: for (j=0; j<bs*bs; j++) vals[j] = 0.0;
47: for (i=0; i<Mbs; i++){
48: cols[0] = i*bs; rows[0] = i*bs;
49: for (j=1; j<bs; j++) {
50: rows[j] = rows[j-1]+1;
51: cols[j] = cols[j-1]+1;
52: }
53: MatSetValues(A,bs,rows,bs,cols,vals,ADD_VALUES);
54: }
55: /* second, add random blocks */
56: for (i=0; i<20*bs; i++) {
57: PetscRandomGetValue(rand,&rval);
58: cols[0] = bs*(PetscInt)(PetscRealPart(rval)*Mbs);
59: PetscRandomGetValue(rand,&rval);
60: rows[0] = rstart + bs*(PetscInt)(PetscRealPart(rval)*mbs);
61: for (j=1; j<bs; j++) {
62: rows[j] = rows[j-1]+1;
63: cols[j] = cols[j-1]+1;
64: }
66: for (j=0; j<bs*bs; j++) {
67: PetscRandomGetValue(rand,&rval);
68: vals[j] = rval;
69: }
70: MatSetValues(A,bs,rows,bs,cols,vals,ADD_VALUES);
71: }
73: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
74: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
75:
76: /* make A a symmetric matrix: A <- A^T + A */
77: MatTranspose(A, &Atrans);
78: MatAXPY(A,one,Atrans,DIFFERENT_NONZERO_PATTERN);
79: MatDestroy(Atrans);
80: MatTranspose(A, &Atrans);
81: MatEqual(A, Atrans, &flg);
82: if (!flg) {
83: SETERRQ(1,"A+A^T is non-symmetric");
84: }
85: MatDestroy(Atrans);
86: /* MatView(A,PETSC_VIEWER_STDOUT_WORLD); */
88: /* create a SeqSBAIJ matrix sA (= A) */
89: MatConvert(A,MATMPISBAIJ,MAT_INITIAL_MATRIX,&sA);
90: /* MatView(sA,PETSC_VIEWER_STDOUT_WORLD); */
92: /* Test sA==A through MatMult() */
93: for (i=0; i<nd; i++) {
94: MatGetLocalSize(A, &mm,PETSC_NULL);
95: VecCreate(PETSC_COMM_WORLD,&xx);
96: VecSetSizes(xx,mm,PETSC_DECIDE);
97: VecSetFromOptions(xx);
98: VecDuplicate(xx,&s1);
99: VecDuplicate(xx,&s2);
100: for (j=0; j<3; j++) {
101: VecSetRandom(xx,rand);
102: MatMult(A,xx,s1);
103: MatMult(sA,xx,s2);
104: VecNorm(s1,NORM_2,&s1norm);
105: VecNorm(s2,NORM_2,&s2norm);
106: rnorm = s2norm-s1norm;
107: if (rnorm<-tol || rnorm>tol) {
108: PetscPrintf(PETSC_COMM_SELF,"Error:MatMult - Norm1=%16.14e Norm2=%16.14e\n",s1norm,s2norm);
109: }
110: }
111: VecDestroy(xx);
112: VecDestroy(s1);
113: VecDestroy(s2);
114: }
115:
116: /* Test MatIncreaseOverlap() */
117: PetscMalloc(nd*sizeof(IS **),&is1);
118: PetscMalloc(nd*sizeof(IS **),&is2);
120: for (i=0; i<nd; i++) {
121: PetscRandomGetValue(rand,&rval);
122: sz = (PetscInt)((0.5 + 0.2*PetscRealPart(rval))*mbs); /* 0.5*mbs < sz < 0.7*mbs */
123: sz /= (size*nd*10);
124: /*
125: if (!rank){
126: PetscPrintf(PETSC_COMM_SELF," [%d] IS sz[%d]: %d\n",rank,i,sz);
127: }
128: */
129: for (j=0; j<sz; j++) {
130: PetscRandomGetValue(rand,&rval);
131: idx[j*bs] = bs*(PetscInt)(PetscRealPart(rval)*Mbs);
132: for (k=1; k<bs; k++) idx[j*bs+k] = idx[j*bs]+k;
133: }
134: ISCreateGeneral(PETSC_COMM_SELF,sz*bs,idx,is1+i);
135: ISCreateGeneral(PETSC_COMM_SELF,sz*bs,idx,is2+i);
136: }
138: PetscLogStageRegister(&stages[0],"MatOv_SBAIJ");
139: PetscLogStageRegister(&stages[1],"MatOv_ BAIJ");
141: PetscLogStagePush(stages[0]);
142: MatIncreaseOverlap(sA,nd,is2,ov);
143: PetscLogStagePop();
145: PetscLogStagePush(stages[1]);
146: MatIncreaseOverlap(A,nd,is1,ov);
147: PetscLogStagePop();
149: for (i=0; i<nd; ++i) {
150: ISEqual(is1[i],is2[i],&flg);
151: if (!flg ){
152: if (rank == size){
153: ISSort(is1[i]);
154: ISView(is1[i],PETSC_VIEWER_STDOUT_SELF);
155: ISSort(is2[i]);
156: ISView(is2[i],PETSC_VIEWER_STDOUT_SELF);
157: }
158: SETERRQ1(1,"i=%D, is1 != is2",i);
159: }
160: }
161:
162: for (i=0; i<nd; ++i) {
163: ISSort(is1[i]);
164: ISSort(is2[i]);
165: }
166: MatGetSubMatrices(sA,nd,is2,is2,MAT_INITIAL_MATRIX,&submatsA);
167: MatGetSubMatrices(A,nd,is1,is1,MAT_INITIAL_MATRIX,&submatA);
169: /* Test MatMult() */
170: for (i=0; i<nd; i++) {
171: MatGetSize(submatA[i],&mm,PETSC_NULL);
172: VecCreateSeq(PETSC_COMM_SELF,mm,&xx);
173: VecDuplicate(xx,&s1);
174: VecDuplicate(xx,&s2);
175: for (j=0; j<3; j++) {
176: VecSetRandom(xx,rand);
177: MatMult(submatA[i],xx,s1);
178: MatMult(submatsA[i],xx,s2);
179: VecNorm(s1,NORM_2,&s1norm);
180: VecNorm(s2,NORM_2,&s2norm);
181: rnorm = s2norm-s1norm;
182: if (rnorm<-tol || rnorm>tol) {
183: PetscPrintf(PETSC_COMM_SELF,"[%d]Error:MatMult - Norm1=%16.14e Norm2=%16.14e\n",rank,s1norm,s2norm);
184: }
185: }
186: VecDestroy(xx);
187: VecDestroy(s1);
188: VecDestroy(s2);
189: }
191: /* Now test MatGetSubmatrices with MAT_REUSE_MATRIX option */
192: MatGetSubMatrices(A,nd,is1,is1,MAT_REUSE_MATRIX,&submatA);
193: MatGetSubMatrices(sA,nd,is2,is2,MAT_REUSE_MATRIX,&submatsA);
195: /* Test MatMult() */
196: for (i=0; i<nd; i++) {
197: MatGetSize(submatA[i],&mm,PETSC_NULL);
198: VecCreateSeq(PETSC_COMM_SELF,mm,&xx);
199: VecDuplicate(xx,&s1);
200: VecDuplicate(xx,&s2);
201: for (j=0; j<3; j++) {
202: VecSetRandom(xx,rand);
203: MatMult(submatA[i],xx,s1);
204: MatMult(submatsA[i],xx,s2);
205: VecNorm(s1,NORM_2,&s1norm);
206: VecNorm(s2,NORM_2,&s2norm);
207: rnorm = s2norm-s1norm;
208: if (rnorm<-tol || rnorm>tol) {
209: PetscPrintf(PETSC_COMM_SELF,"[%d]Error:MatMult - Norm1=%16.14e Norm2=%16.14e\n",rank,s1norm,s2norm);
210: }
211: }
212: VecDestroy(xx);
213: VecDestroy(s1);
214: VecDestroy(s2);
215: }
217: /* Free allocated memory */
218: for (i=0; i<nd; ++i) {
219: ISDestroy(is1[i]);
220: ISDestroy(is2[i]);
221: MatDestroy(submatA[i]);
222: MatDestroy(submatsA[i]);
223: }
224: PetscFree(submatA);
225: PetscFree(submatsA);
226: PetscFree(is1);
227: PetscFree(is2);
228: PetscFree(idx);
229: PetscFree(rows);
230: PetscFree(cols);
231: PetscFree(vals);
232: MatDestroy(A);
233: MatDestroy(sA);
234: PetscRandomDestroy(rand);
235: PetscFinalize();
236: return 0;
237: }