Actual source code: ex146.c
petsc-3.7.3 2016-08-01
1: /* This program illustrates use of paralllel real FFT*/
2: static char help[]="This program illustrates the use of parallel real 3D fftw (without PETSc interface)";
3: #include <petscmat.h>
4: #include <fftw3.h>
5: #include <fftw3-mpi.h>
9: PetscInt main(PetscInt argc,char **args)
10: {
11: ptrdiff_t N0=256,N1=256,N2=256,N3=2,dim[4];
12: fftw_plan bplan,fplan;
13: fftw_complex *out;
14: double *in1,*in2;
15: ptrdiff_t alloc_local,local_n0,local_0_start;
16: ptrdiff_t local_n1,local_1_start;
17: PetscInt i,j,indx,n1;
18: PetscInt size,rank,n,N,*in,N_factor,NM;
19: PetscScalar *data_fin,value1,one=1.57,zero=0.0;
20: PetscScalar a,*x_arr,*y_arr,*z_arr,enorm;
21: Vec fin,fout,fout1,ini,final;
22: PetscRandom rnd;
24: VecScatter vecscat,vecscat1;
25: IS indx1,indx2;
26: PetscInt *indx3,k,l,*indx4;
27: PetscInt low,tempindx,tempindx1;
30: PetscInitialize(&argc,&args,(char*)0,help);
31: #if defined(PETSC_USE_COMPLEX)
32: SETERRQ(PETSC_COMM_WORLD,PETSC_ERR_SUP, "This example requires real numbers. Your current scalar type is complex");
33: #endif
34: MPI_Comm_size(PETSC_COMM_WORLD, &size);
35: MPI_Comm_rank(PETSC_COMM_WORLD, &rank);
37: PetscRandomCreate(PETSC_COMM_WORLD,&rnd);
40: alloc_local = fftw_mpi_local_size_3d_transposed(N0,N1,N2/2+1,PETSC_COMM_WORLD,&local_n0,&local_0_start,&local_n1,&local_1_start);
42: /* printf("The value alloc_local is %ld from process %d\n",alloc_local,rank); */
43: printf("The value local_n0 is %ld from process %d\n",local_n0,rank);
44: /* printf("The value local_0_start is %ld from process %d\n",local_0_start,rank);*/
45: /* printf("The value local_n1 is %ld from process %d\n",local_n1,rank); */
46: /* printf("The value local_1_start is %ld from process %d\n",local_1_start,rank);*/
48: /* Allocate space for input and output arrays */
50: in1=(double*)fftw_malloc(sizeof(double)*alloc_local*2);
51: in2=(double*)fftw_malloc(sizeof(double)*alloc_local*2);
52: out=(fftw_complex*)fftw_malloc(sizeof(fftw_complex)*alloc_local);
55: N=2*N0*N1*(N2/2+1);N_factor=N0*N1*N2;
56: n=2*local_n0*N1*(N2/2+1);n1=local_n1*N0*2*N1;
58: /* printf("The value N is %d from process %d\n",N,rank); */
59: /* printf("The value n is %d from process %d\n",n,rank); */
60: /* printf("The value n1 is %d from process %d\n",n1,rank); */
61: /* Creating data vector and accompanying array with VeccreateMPIWithArray */
62: VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)in1,&fin);
63: VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)out,&fout);
64: VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)in2,&fout1);
66: /* VecGetSize(fin,&size); */
67: /* printf("The size is %d\n",size); */
69: VecSet(fin,one);
70: VecSet(fout,zero);
71: VecSet(fout1,zero);
73: VecAssemblyBegin(fin);
74: VecAssemblyEnd(fin);
75: /* VecView(fin,PETSC_VIEWER_STDOUT_WORLD); */
78: VecGetArray(fin,&x_arr);
79: VecGetArray(fout1,&z_arr);
80: VecGetArray(fout,&y_arr);
82: fplan=fftw_mpi_plan_dft_r2c_3d(N0,N1,N2,(double*)x_arr,(fftw_complex*)y_arr,PETSC_COMM_WORLD,FFTW_ESTIMATE);
83: bplan=fftw_mpi_plan_dft_c2r_3d(N0,N1,N2,(fftw_complex*)y_arr,(double*)z_arr,PETSC_COMM_WORLD,FFTW_ESTIMATE);
85: fftw_execute(fplan);
86: fftw_execute(bplan);
88: VecRestoreArray(fin,&x_arr);
89: VecRestoreArray(fout1,&z_arr);
90: VecRestoreArray(fout,&y_arr);
93: /* a = 1.0/(PetscReal)N_factor; */
94: /* VecScale(fout1,a); */
95: VecCreate(PETSC_COMM_WORLD,&ini);
96: VecCreate(PETSC_COMM_WORLD,&final);
97: VecSetSizes(ini,local_n0*N1*N2,N_factor);
98: VecSetSizes(final,local_n0*N1*N2,N_factor);
99: /* VecSetSizes(ini,PETSC_DECIDE,N_factor); */
100: /* VecSetSizes(final,PETSC_DECIDE,N_factor); */
101: VecSetFromOptions(ini);
102: VecSetFromOptions(final);
104: if (N2%2==0) NM=N2+2;
105: else NM=N2+1;
107: VecGetOwnershipRange(fin,&low,NULL);
108: printf("The local index is %d from %d\n",low,rank);
109: PetscMalloc1(local_n0*N1*N2,&indx3);
110: PetscMalloc1(local_n0*N1*N2,&indx4);
111: for (i=0; i<local_n0; i++) {
112: for (j=0;j<N1;j++) {
113: for (k=0;k<N2;k++) {
114: tempindx = i*N1*N2 + j*N2 + k;
115: tempindx1 = i*N1*NM + j*NM + k;
117: indx3[tempindx]=local_0_start*N1*N2+tempindx;
118: indx4[tempindx]=low+tempindx1;
119: }
120: /* printf("index3 %d from proc %d is \n",indx3[tempindx],rank); */
121: /* printf("index4 %d from proc %d is \n",indx4[tempindx],rank); */
122: }
123: }
124: VecGetValues(fin,local_n0*N1*N2,indx4,x_arr);
125: VecSetValues(ini,local_n0*N1*N2,indx3,x_arr,INSERT_VALUES);
126: VecAssemblyBegin(ini);
127: VecAssemblyEnd(ini);
129: VecGetValues(fout1,local_n0*N1*N2,indx4,y_arr);
130: VecSetValues(final,local_n0*N1*N2,indx3,y_arr,INSERT_VALUES);
131: VecAssemblyBegin(final);
132: VecAssemblyEnd(final);
134: printf("The local index value is %ld from %d",local_n0*N1*N2,rank);
135: /*
136: for (i=0;i<N0;i++) {
137: for (j=0;j<N1;j++) {
138: indx=i*N1*NM+j*NM;
139: ISCreateStride(PETSC_COMM_WORLD,N2,indx,1,&indx1);
140: indx=i*N1*N2+j*N2;
141: ISCreateStride(PETSC_COMM_WORLD,N2,indx,1,&indx2);
142: VecScatterCreate(fin,indx1,ini,indx2,&vecscat);
143: VecScatterBegin(vecscat,fin,ini,INSERT_VALUES,SCATTER_FORWARD);
144: VecScatterEnd(vecscat,fin,ini,INSERT_VALUES,SCATTER_FORWARD);
145: VecScatterCreate(fout1,indx1,final,indx2,&vecscat1);
146: VecScatterBegin(vecscat1,fout1,final,INSERT_VALUES,SCATTER_FORWARD);
147: VecScatterEnd(vecscat1,fout1,final,INSERT_VALUES,SCATTER_FORWARD);
148: }
149: }
150: */
151: a = 1.0/(PetscReal)N_factor;
152: VecScale(fout1,a);
153: VecScale(final,a);
155: VecAssemblyBegin(ini);
156: VecAssemblyEnd(ini);
158: VecAssemblyBegin(final);
159: VecAssemblyEnd(final);
161: /* VecView(final,PETSC_VIEWER_STDOUT_WORLD); */
162: VecAXPY(final,-1.0,ini);
163: VecNorm(final,NORM_1,&enorm);
164: PetscPrintf(PETSC_COMM_WORLD," Error norm of |x - z| = %e\n",enorm);
165: fftw_destroy_plan(fplan);
166: fftw_destroy_plan(bplan);
167: fftw_free(in1); VecDestroy(&fin);
168: fftw_free(out); VecDestroy(&fout);
169: fftw_free(in2); VecDestroy(&fout1);
171: PetscFinalize();
172: return 0;
173: }