Actual source code: ex121.c
petsc-3.5.4 2015-05-23
1: static char help[] = "Test sequential FFTW convolution\n\n";
3: /*
4: Compiling the code:
5: This code uses the complex numbers, so configure must be given --with-scalar-type=complex to enable this
6: */
7: #include <petscmat.h>
11: PetscInt main(PetscInt argc,char **args)
12: {
13: typedef enum {RANDOM, CONSTANT, TANH, NUM_FUNCS} FuncType;
14: const char *funcNames[NUM_FUNCS] = {"random", "constant", "tanh"};
15: Mat A;
16: PetscMPIInt size;
17: PetscInt n = 10,N,ndim=4,dim[4],DIM,i,j;
18: Vec w,x,y1,y2,z1,z2;
19: PetscScalar *a, *a2, *a3;
20: PetscScalar s;
21: PetscRandom rdm;
22: PetscReal enorm;
23: PetscInt func = 0;
24: FuncType function = RANDOM;
25: PetscBool view = PETSC_FALSE;
28: PetscInitialize(&argc,&args,(char*)0,help);
29: #if !defined(PETSC_USE_COMPLEX)
30: SETERRQ(PETSC_COMM_WORLD,PETSC_ERR_SUP, "This example requires complex numbers");
31: #endif
32: MPI_Comm_size(PETSC_COMM_WORLD, &size);
33: if (size != 1) SETERRQ(PETSC_COMM_WORLD,PETSC_ERR_SUP, "This is a uniprocessor example only!");
34: PetscOptionsBegin(PETSC_COMM_WORLD, NULL, "FFTW Options", "ex112");
35: PetscOptionsEList("-function", "Function type", "ex121", funcNames, NUM_FUNCS, funcNames[function], &func, NULL);
36: PetscOptionsBool("-vec_view draw", "View the functions", "ex112", view, &view, NULL);
37: function = (FuncType) func;
38: PetscOptionsEnd();
40: for (DIM = 0; DIM < ndim; DIM++) {
41: dim[DIM] = n; /* size of transformation in DIM-dimension */
42: }
43: PetscRandomCreate(PETSC_COMM_SELF, &rdm);
44: PetscRandomSetFromOptions(rdm);
46: for (DIM = 1; DIM < 5; DIM++) {
47: /* create vectors of length N=n^DIM */
48: for (i = 0, N = 1; i < DIM; i++) N *= dim[i];
49: PetscPrintf(PETSC_COMM_SELF, "\n %d-D: FFTW on vector of size %d \n",DIM,N);
50: VecCreateSeq(PETSC_COMM_SELF,N,&x);
51: PetscObjectSetName((PetscObject) x, "Real space vector");
52: VecDuplicate(x,&w);
53: PetscObjectSetName((PetscObject) w, "Window vector");
54: VecDuplicate(x,&y1);
55: PetscObjectSetName((PetscObject) y1, "Frequency space vector");
56: VecDuplicate(x,&y2);
57: PetscObjectSetName((PetscObject) y2, "Frequency space window vector");
58: VecDuplicate(x,&z1);
59: PetscObjectSetName((PetscObject) z1, "Reconstructed convolution");
60: VecDuplicate(x,&z2);
61: PetscObjectSetName((PetscObject) z2, "Real space convolution");
63: if (function == RANDOM) {
64: VecSetRandom(x, rdm);
65: } else if (function == CONSTANT) {
66: VecSet(x, 1.0);
67: } else if (function == TANH) {
68: VecGetArray(x, &a);
69: for (i = 0; i < N; ++i) {
70: a[i] = tanh((i - N/2.0)*(10.0/N));
71: }
72: VecRestoreArray(x, &a);
73: }
74: if (view) {VecView(x, PETSC_VIEWER_DRAW_WORLD);}
76: /* Create window function */
77: VecGetArray(w, &a);
78: for (i = 0; i < N; ++i) {
79: /* Step Function */
80: a[i] = (i > N/4 && i < 3*N/4) ? 1.0 : 0.0;
81: /* Delta Function */
82: /*a[i] = (i == N/2)? 1.0: 0.0; */
83: }
84: VecRestoreArray(w, &a);
85: if (view) {VecView(w, PETSC_VIEWER_DRAW_WORLD);}
87: /* create FFTW object */
88: MatCreateFFT(PETSC_COMM_SELF,DIM,dim,MATFFTW,&A);
90: /* Convolve x with w*/
91: MatMult(A,x,y1);
92: MatMult(A,w,y2);
93: VecPointwiseMult(y1, y1, y2);
94: if (view && i == 0) {VecView(y1, PETSC_VIEWER_DRAW_WORLD);}
95: MatMultTranspose(A,y1,z1);
97: /* Compute the real space convolution */
98: VecGetArray(x, &a);
99: VecGetArray(w, &a2);
100: VecGetArray(z2, &a3);
101: for (i = 0; i < N; ++i) {
102: /* PetscInt checkInd = (i > N/2-1)? i-N/2: i+N/2;*/
104: /*if (!(i%100)) PetscPrintf(PETSC_COMM_WORLD, "Finished convolution row %d\n", i);*/
105: a3[i] = 0.0;
106: for (j = -N/2+1; j < N/2; ++j) {
107: PetscInt xpInd = (j < 0) ? N+j : j;
108: PetscInt diffInd = (i-j < 0) ? N-(j-i) : (i-j > N-1) ? i-j-N : i-j;
110: a3[i] += a[xpInd]*a2[diffInd];
111: }
112: /*if (PetscAbsScalar(a3[i]) > PetscAbsScalar(a[checkInd])+0.1) PetscPrintf(PETSC_COMM_WORLD, "Invalid convolution at row %d\n", i);*/
113: }
114: VecRestoreArray(x, &a);
115: VecRestoreArray(w, &a2);
116: VecRestoreArray(z2, &a3);
118: /* compare z1 and z2. FFTW computes an unnormalized DFT, thus z1 = N*z2 */
119: s = 1.0/(PetscReal)N;
120: VecScale(z1,s);
121: if (view) {VecView(z1, PETSC_VIEWER_DRAW_WORLD);}
122: if (view) {VecView(z2, PETSC_VIEWER_DRAW_WORLD);}
123: VecAXPY(z1,-1.0,z2);
124: VecNorm(z1,NORM_1,&enorm);
125: if (enorm > 1.e-11) {
126: PetscPrintf(PETSC_COMM_SELF," Error norm of |z1 - z2| %g\n",(double)enorm);
127: }
129: /* free spaces */
130: VecDestroy(&x);
131: VecDestroy(&y1);
132: VecDestroy(&y2);
133: VecDestroy(&z1);
134: VecDestroy(&z2);
135: VecDestroy(&w);
136: MatDestroy(&A);
137: }
138: PetscRandomDestroy(&rdm);
139: PetscFinalize();
140: return 0;
141: }