Actual source code: relax.h
petsc-3.4.5 2014-06-29
2: /*
3: This is included by sbaij.c to generate unsigned short and regular versions of these two functions
4: */
6: #if defined(USESHORT)
8: PetscErrorCode MatMult_SeqSBAIJ_1_Hermitian_ushort(Mat A,Vec xx,Vec zz)
9: #else
11: PetscErrorCode MatMult_SeqSBAIJ_1_Hermitian(Mat A,Vec xx,Vec zz)
12: #endif
13: {
14: Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data;
15: const PetscScalar *x;
16: PetscScalar *z,x1,sum;
17: const MatScalar *v;
18: MatScalar vj;
19: PetscErrorCode ierr;
20: PetscInt mbs=a->mbs,i,j,nz;
21: const PetscInt *ai=a->i;
22: #if defined(USESHORT)
23: const unsigned short *ib=a->jshort;
24: unsigned short ibt;
25: #else
26: const PetscInt *ib=a->j;
27: PetscInt ibt;
28: #endif
29: PetscInt nonzerorow = 0,jmin;
32: VecSet(zz,0.0);
33: VecGetArrayRead(xx,&x);
34: VecGetArray(zz,&z);
36: v = a->a;
37: for (i=0; i<mbs; i++) {
38: nz = ai[i+1] - ai[i]; /* length of i_th row of A */
39: if (!nz) continue; /* Move to the next row if the current row is empty */
40: nonzerorow++;
41: x1 = x[i];
42: sum = 0.0;
43: jmin = 0;
44: if (ib[0] == i) {
45: sum = v[0]*x1; /* diagonal term */
46: jmin++;
47: }
48: for (j=jmin; j<nz; j++) {
49: ibt = ib[j];
50: vj = v[j];
51: sum += vj * x[ibt]; /* (strict upper triangular part of A)*x */
52: z[ibt] += PetscConj(v[j]) * x1; /* (strict lower triangular part of A)*x */
53: }
54: z[i] += sum;
55: v += nz;
56: ib += nz;
57: }
59: VecRestoreArrayRead(xx,&x);
60: VecRestoreArray(zz,&z);
61: PetscLogFlops(2.0*(2.0*a->nz - nonzerorow) - nonzerorow);
62: return(0);
63: }
66: #if defined(USESHORT)
68: PetscErrorCode MatMult_SeqSBAIJ_1_ushort(Mat A,Vec xx,Vec zz)
69: #else
71: PetscErrorCode MatMult_SeqSBAIJ_1(Mat A,Vec xx,Vec zz)
72: #endif
73: {
74: Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data;
75: const PetscScalar *x;
76: PetscScalar *z,x1,sum;
77: const MatScalar *v;
78: MatScalar vj;
79: PetscErrorCode ierr;
80: PetscInt mbs=a->mbs,i,j,nz;
81: const PetscInt *ai=a->i;
82: #if defined(USESHORT)
83: const unsigned short *ib=a->jshort;
84: unsigned short ibt;
85: #else
86: const PetscInt *ib=a->j;
87: PetscInt ibt;
88: #endif
89: PetscInt nonzerorow=0,jmin;
92: VecSet(zz,0.0);
93: VecGetArrayRead(xx,&x);
94: VecGetArray(zz,&z);
96: v = a->a;
97: for (i=0; i<mbs; i++) {
98: nz = ai[i+1] - ai[i]; /* length of i_th row of A */
99: if (!nz) continue; /* Move to the next row if the current row is empty */
100: nonzerorow++;
101: sum = 0.0;
102: jmin = 0;
103: x1 = x[i];
104: if (ib[0] == i) {
105: sum = v[0]*x1; /* diagonal term */
106: jmin++;
107: }
108: PetscPrefetchBlock(ib+nz,nz,0,PETSC_PREFETCH_HINT_NTA); /* Indices for the next row (assumes same size as this one) */
109: PetscPrefetchBlock(v+nz,nz,0,PETSC_PREFETCH_HINT_NTA); /* Entries for the next row */
110: for (j=jmin; j<nz; j++) {
111: ibt = ib[j];
112: vj = v[j];
113: z[ibt] += vj * x1; /* (strict lower triangular part of A)*x */
114: sum += vj * x[ibt]; /* (strict upper triangular part of A)*x */
115: }
116: z[i] += sum;
117: v += nz;
118: ib += nz;
119: }
121: VecRestoreArrayRead(xx,&x);
122: VecRestoreArray(zz,&z);
123: PetscLogFlops(2.0*(2.0*a->nz - nonzerorow) - nonzerorow);
124: return(0);
125: }
128: #if defined(USESHORT)
130: PetscErrorCode MatSOR_SeqSBAIJ_ushort(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
131: #else
133: PetscErrorCode MatSOR_SeqSBAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
134: #endif
135: {
136: Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data;
137: const MatScalar *aa=a->a,*v,*v1,*aidiag;
138: PetscScalar *x,*t,sum;
139: const PetscScalar *b;
140: MatScalar tmp;
141: PetscErrorCode ierr;
142: PetscInt m =a->mbs,bs=A->rmap->bs,j;
143: const PetscInt *ai=a->i;
144: #if defined(USESHORT)
145: const unsigned short *aj=a->jshort,*vj,*vj1;
146: #else
147: const PetscInt *aj=a->j,*vj,*vj1;
148: #endif
149: PetscInt nz,nz1,i;
152: if (flag & SOR_EISENSTAT) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support yet for Eisenstat");
154: its = its*lits;
155: if (its <= 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
157: if (bs > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"SSOR for block size > 1 is not yet implemented");
159: VecGetArray(xx,&x);
160: VecGetArrayRead(bb,&b);
162: if (!a->idiagvalid) {
163: if (!a->idiag) {
164: PetscMalloc(m*sizeof(PetscScalar),&a->idiag);
165: }
166: for (i=0; i<a->mbs; i++) a->idiag[i] = 1.0/a->a[a->i[i]];
167: a->idiagvalid = PETSC_TRUE;
168: }
170: if (!a->sor_work) {
171: PetscMalloc(m*sizeof(PetscScalar),&a->sor_work);
172: }
173: t = a->sor_work;
175: aidiag = a->idiag;
177: if (flag == SOR_APPLY_UPPER) {
178: /* apply (U + D/omega) to the vector */
179: PetscScalar d;
180: for (i=0; i<m; i++) {
181: d = fshift + aa[ai[i]];
182: nz = ai[i+1] - ai[i] - 1;
183: vj = aj + ai[i] + 1;
184: v = aa + ai[i] + 1;
185: sum = b[i]*d/omega;
186: PetscSparseDensePlusDot(sum,b,v,vj,nz);
187: x[i] = sum;
188: }
189: PetscLogFlops(a->nz);
190: }
192: if (flag & SOR_ZERO_INITIAL_GUESS) {
193: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
194: PetscMemcpy(t,b,m*sizeof(PetscScalar));
196: v = aa + 1;
197: vj = aj + 1;
198: for (i=0; i<m; i++) {
199: nz = ai[i+1] - ai[i] - 1;
200: tmp = -(x[i] = omega*t[i]*aidiag[i]);
201: for (j=0; j<nz; j++) t[vj[j]] += tmp*v[j];
202: v += nz + 1;
203: vj += nz + 1;
204: }
205: PetscLogFlops(2*a->nz);
206: }
208: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
209: int nz2;
210: if (!(flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP)) {
211: #if defined(PETSC_USE_BACKWARD_LOOP)
212: v = aa + ai[m] - 1;
213: vj = aj + ai[m] - 1;
214: for (i=m-1; i>=0; i--) {
215: sum = b[i];
216: nz = ai[i+1] - ai[i] - 1;
217: {PetscInt __i;for (__i=0; __i<nz; __i++) sum -= v[-__i] * x[vj[-__i]];}
218: #else
219: v = aa + ai[m-1] + 1;
220: vj = aj + ai[m-1] + 1;
221: nz = 0;
222: for (i=m-1; i>=0; i--) {
223: sum = b[i];
224: nz2 = ai[i] - ai[PetscMax(i-1,0)] - 1; /* avoid referencing ai[-1], nonsense nz2 is okay on last iteration */
225: PETSC_Prefetch(v-nz2-1,0,PETSC_PREFETCH_HINT_NTA);
226: PETSC_Prefetch(vj-nz2-1,0,PETSC_PREFETCH_HINT_NTA);
227: PetscSparseDenseMinusDot(sum,x,v,vj,nz);
228: nz = nz2;
229: #endif
230: x[i] = omega*sum*aidiag[i];
231: v -= nz + 1;
232: vj -= nz + 1;
233: }
234: PetscLogFlops(2*a->nz);
235: } else {
236: v = aa + ai[m-1] + 1;
237: vj = aj + ai[m-1] + 1;
238: nz = 0;
239: for (i=m-1; i>=0; i--) {
240: sum = t[i];
241: nz2 = ai[i] - ai[PetscMax(i-1,0)] - 1; /* avoid referencing ai[-1], nonsense nz2 is okay on last iteration */
242: PETSC_Prefetch(v-nz2-1,0,PETSC_PREFETCH_HINT_NTA);
243: PETSC_Prefetch(vj-nz2-1,0,PETSC_PREFETCH_HINT_NTA);
244: PetscSparseDenseMinusDot(sum,x,v,vj,nz);
245: x[i] = (1-omega)*x[i] + omega*sum*aidiag[i];
246: nz = nz2;
247: v -= nz + 1;
248: vj -= nz + 1;
249: }
250: PetscLogFlops(2*a->nz);
251: }
252: }
253: its--;
254: }
256: while (its--) {
257: /*
258: forward sweep:
259: for i=0,...,m-1:
260: sum[i] = (b[i] - U(i,:)x)/d[i];
261: x[i] = (1-omega)x[i] + omega*sum[i];
262: b = b - x[i]*U^T(i,:);
264: */
265: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
266: PetscMemcpy(t,b,m*sizeof(PetscScalar));
268: for (i=0; i<m; i++) {
269: v = aa + ai[i] + 1; v1=v;
270: vj = aj + ai[i] + 1; vj1=vj;
271: nz = ai[i+1] - ai[i] - 1; nz1=nz;
272: sum = t[i];
273: while (nz1--) sum -= (*v1++)*x[*vj1++];
274: x[i] = (1-omega)*x[i] + omega*sum*aidiag[i];
275: while (nz--) t[*vj++] -= x[i]*(*v++);
276: }
277: PetscLogFlops(4.0*a->nz);
278: }
280: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
281: /*
282: backward sweep:
283: b = b - x[i]*U^T(i,:), i=0,...,n-2
284: for i=m-1,...,0:
285: sum[i] = (b[i] - U(i,:)x)/d[i];
286: x[i] = (1-omega)x[i] + omega*sum[i];
287: */
288: /* if there was a forward sweep done above then I thing the next two for loops are not needed */
289: PetscMemcpy(t,b,m*sizeof(PetscScalar));
291: for (i=0; i<m-1; i++) { /* update rhs */
292: v = aa + ai[i] + 1;
293: vj = aj + ai[i] + 1;
294: nz = ai[i+1] - ai[i] - 1;
295: while (nz--) t[*vj++] -= x[i]*(*v++);
296: }
297: PetscLogFlops(2.0*(a->nz - m));
298: for (i=m-1; i>=0; i--) {
299: v = aa + ai[i] + 1;
300: vj = aj + ai[i] + 1;
301: nz = ai[i+1] - ai[i] - 1;
302: sum = t[i];
303: while (nz--) sum -= x[*vj++]*(*v++);
304: x[i] = (1-omega)*x[i] + omega*sum*aidiag[i];
305: }
306: PetscLogFlops(2.0*(a->nz + m));
307: }
308: }
310: VecRestoreArray(xx,&x);
311: VecRestoreArrayRead(bb,&b);
312: return(0);
313: }