/[escript]/trunk/paso/src/BiCGStab.c
ViewVC logotype

Diff of /trunk/paso/src/BiCGStab.c

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 939 by gross, Wed Jan 17 07:41:13 2007 UTC revision 940 by gross, Mon Jan 29 00:09:08 2007 UTC
# Line 154  err_t Paso_Solver_BiCGStab( Line 154  err_t Paso_Solver_BiCGStab(
154      omegaDenumtr = 0.0;      omegaDenumtr = 0.0;
155        }        }
156        #pragma omp barrier        #pragma omp barrier
       #pragma ivdep  
157        #pragma omp for private(i0) reduction(+:sum_1) schedule(static)        #pragma omp for private(i0) reduction(+:sum_1) schedule(static)
158        for (i0 = 0; i0 < n; i0++) sum_1 += rtld[i0] * r[i0];        for (i0 = 0; i0 < n; i0++) sum_1 += rtld[i0] * r[i0];
159        rho = sum_1;        rho = sum_1;
# Line 164  err_t Paso_Solver_BiCGStab( Line 163  err_t Paso_Solver_BiCGStab(
163                
164      if (num_iter > 1) {      if (num_iter > 1) {
165        beta = rho / rho1 * (alpha / omega);        beta = rho / rho1 * (alpha / omega);
           #pragma ivdep  
166            #pragma omp for private(i0) schedule(static)            #pragma omp for private(i0) schedule(static)
167        for (i0 = 0; i0 < n; i0++) p[i0] = r[i0] + beta * (p[i0] - omega * v[i0]);        for (i0 = 0; i0 < n; i0++) p[i0] = r[i0] + beta * (p[i0] - omega * v[i0]);
168      } else {      } else {
           #pragma ivdep  
169            #pragma omp for private(i0) schedule(static)            #pragma omp for private(i0) schedule(static)
170        for (i0 = 0; i0 < n; i0++) p[i0] = r[i0];        for (i0 = 0; i0 < n; i0++) p[i0] = r[i0];
171      }      }
# Line 178  err_t Paso_Solver_BiCGStab( Line 175  err_t Paso_Solver_BiCGStab(
175          Paso_Solver_solvePreconditioner(A,&phat[0], &p[0]);          Paso_Solver_solvePreconditioner(A,&phat[0], &p[0]);
176      Paso_SystemMatrix_MatrixVector_CSR_OFFSET0(ONE, A, &phat[0],ZERO, &v[0]);      Paso_SystemMatrix_MatrixVector_CSR_OFFSET0(ONE, A, &phat[0],ZERO, &v[0]);
177        
         // #pragma ivdep  
178          #pragma omp for private(i0) reduction(+:sum_2) schedule(static)          #pragma omp for private(i0) reduction(+:sum_2) schedule(static)
179      for (i0 = 0; i0 < n; i0++) sum_2 += rtld[i0] * v[i0];      for (i0 = 0; i0 < n; i0++) sum_2 += rtld[i0] * v[i0];
180          if (! (breakFlag = (ABS(sum_2) <= TOLERANCE_FOR_SCALARS))) {          if (! (breakFlag = (ABS(sum_2) <= TOLERANCE_FOR_SCALARS))) {
181         alpha = rho / sum_2;         alpha = rho / sum_2;
182    
            // #pragma ivdep  
183             #pragma omp for private(i0) reduction(+:sum_3) schedule(static)             #pragma omp for private(i0) reduction(+:sum_3) schedule(static)
184         for (i0 = 0; i0 < n; i0++) {         for (i0 = 0; i0 < n; i0++) {
185           r[i0] -= alpha * v[i0];           r[i0] -= alpha * v[i0];
# Line 195  err_t Paso_Solver_BiCGStab( Line 190  err_t Paso_Solver_BiCGStab(
190                    
191         /*        Early check for tolerance. */         /*        Early check for tolerance. */
192         if ( (convergeFlag = (norm_of_residual <= tol)) ) {         if ( (convergeFlag = (norm_of_residual <= tol)) ) {
              // #pragma ivdep  
193               #pragma omp for  private(i0) schedule(static)               #pragma omp for  private(i0) schedule(static)
194           for (i0 = 0; i0 < n; i0++) x[i0] += alpha * phat[i0];           for (i0 = 0; i0 < n; i0++) x[i0] += alpha * phat[i0];
195           maxIterFlag = FALSE;           maxIterFlag = FALSE;
# Line 205  err_t Paso_Solver_BiCGStab( Line 199  err_t Paso_Solver_BiCGStab(
199               Paso_Solver_solvePreconditioner(A,&shat[0], &s[0]);               Paso_Solver_solvePreconditioner(A,&shat[0], &s[0]);
200           Paso_SystemMatrix_MatrixVector_CSR_OFFSET0(ONE, A, &shat[0],ZERO,&t[0]);           Paso_SystemMatrix_MatrixVector_CSR_OFFSET0(ONE, A, &shat[0],ZERO,&t[0]);
201        
              // #pragma ivdep  
202               #pragma omp for private(i0) reduction(+:omegaNumtr,omegaDenumtr) schedule(static)               #pragma omp for private(i0) reduction(+:omegaNumtr,omegaDenumtr) schedule(static)
203           for (i0 = 0; i0 < n; i0++) {           for (i0 = 0; i0 < n; i0++) {
204             omegaNumtr +=t[i0] * s[i0];             omegaNumtr +=t[i0] * s[i0];
# Line 214  err_t Paso_Solver_BiCGStab( Line 207  err_t Paso_Solver_BiCGStab(
207               if (! (breakFlag = (ABS(omegaDenumtr) <= TOLERANCE_FOR_SCALARS))) {               if (! (breakFlag = (ABS(omegaDenumtr) <= TOLERANCE_FOR_SCALARS))) {
208              omega = omegaNumtr / omegaDenumtr;              omega = omegaNumtr / omegaDenumtr;
209        
                 // #pragma ivdep  
210                  #pragma omp for private(i0) reduction(+:sum_4) schedule(static)                  #pragma omp for private(i0) reduction(+:sum_4) schedule(static)
211              for (i0 = 0; i0 < n; i0++) {              for (i0 = 0; i0 < n; i0++) {
212                x[i0] += alpha * phat[i0] + omega * shat[i0];                x[i0] += alpha * phat[i0] + omega * shat[i0];
Legend:
Removed from v.939  
changed lines
  Added in v.940
  ViewVC Help
Powered by ViewVC 1.1.26