| 1 |
/* $Id$ */ |
| 2 |
|
| 3 |
/**************************************************************/ |
| 4 |
|
| 5 |
/* Paso: ILU preconditioner with reordering */ |
| 6 |
|
| 7 |
/**************************************************************/ |
| 8 |
|
| 9 |
/* Copyrights by ACcESS Australia 2003,2004,2005 */ |
| 10 |
/* Author: gross@access.edu.au */ |
| 11 |
|
| 12 |
/**************************************************************/ |
| 13 |
|
| 14 |
#include "Paso.h" |
| 15 |
#include "Solver.h" |
| 16 |
#include "PasoUtil.h" |
| 17 |
|
| 18 |
/**************************************************************/ |
| 19 |
|
| 20 |
/* free all memory used by ILU */ |
| 21 |
|
| 22 |
void Paso_Solver_ILU_free(Paso_Solver_ILU * in) { |
| 23 |
if (in!=NULL) { |
| 24 |
MEMFREE(in->colorOf); |
| 25 |
MEMFREE(in->factors); |
| 26 |
MEMFREE(in->main_iptr); |
| 27 |
Paso_SystemMatrixPattern_dealloc(in->pattern); |
| 28 |
MEMFREE(in); |
| 29 |
} |
| 30 |
} |
| 31 |
|
| 32 |
/**************************************************************/ |
| 33 |
|
| 34 |
/* constructs the incomplete block factorization of |
| 35 |
|
| 36 |
*/ |
| 37 |
Paso_Solver_ILU* Paso_Solver_getILU(Paso_SystemMatrix * A,bool_t verbose) { |
| 38 |
dim_t n=A->num_rows; |
| 39 |
dim_t n_block=A->row_block_size; |
| 40 |
index_t num_colors=0; |
| 41 |
register double A11,A12,A13,A21,A22,A23,A31,A32,A33,D; |
| 42 |
register double mainA11,mainA12,mainA13,mainA21,mainA22,mainA23,mainA31,mainA32,mainA33; |
| 43 |
register double S11,S12,S13,S21,S22,S23,S31,S32,S33; |
| 44 |
register index_t i,iptr_main,iptr,iptr_ik,k,iptr_kj,j,iptr_ij,color,color2; |
| 45 |
double time0,time_color,time_fac; |
| 46 |
/* allocations: */ |
| 47 |
Paso_Solver_ILU* out=MEMALLOC(1,Paso_Solver_ILU); |
| 48 |
if (Paso_checkPtr(out)) return NULL; |
| 49 |
index_t* mis_marker=TMPMEMALLOC(n,index_t); |
| 50 |
out->colorOf=MEMALLOC(n,index_t); |
| 51 |
out->factors=MEMALLOC(A->len,double); |
| 52 |
out->main_iptr=MEMALLOC(n,index_t); |
| 53 |
out->pattern=Paso_SystemMatrixPattern_reference(A->pattern); |
| 54 |
out->n_block=n_block; |
| 55 |
out->n=n; |
| 56 |
time0=Paso_timer(); |
| 57 |
if ( !(Paso_checkPtr(mis_marker) || |
| 58 |
Paso_checkPtr(out->colorOf) || Paso_checkPtr(out->main_iptr) || Paso_checkPtr(out->factors)) ) { |
| 59 |
/* get coloring */ |
| 60 |
index_t num_colors=0; |
| 61 |
#pragma omp parallel for private(i) schedule(static) |
| 62 |
for (i = 0; i < n; ++i) out->colorOf[i]=-1; |
| 63 |
while (Paso_Util_isAny(n,out->colorOf,-1) && Paso_noError()) { |
| 64 |
#pragma omp parallel for private(i) schedule(static) |
| 65 |
for (i = 0; i < n; ++i) mis_marker[i]=out->colorOf[i]; |
| 66 |
Paso_SystemMatrixPattern_mis(A->pattern,mis_marker); |
| 67 |
|
| 68 |
#pragma omp parallel for private(i) schedule(static) |
| 69 |
for (i = 0; i < n; ++i) if (mis_marker[i]) out->colorOf[i]=num_colors; |
| 70 |
++num_colors; |
| 71 |
} |
| 72 |
out->num_colors=num_colors; |
| 73 |
time_color=Paso_timer()-time0; |
| 74 |
time0=Paso_timer(); |
| 75 |
/* find main diagonal and copy matrix values */ |
| 76 |
#pragma omp parallel for schedule(static) private(i,iptr,iptr_main,k) |
| 77 |
for (i = 0; i < n; ++i) { |
| 78 |
for (iptr=A->pattern->ptr[i];iptr<A->pattern->ptr[i+1]; ++iptr) { |
| 79 |
iptr_main=A->pattern->ptr[0]-1; |
| 80 |
for (iptr=A->pattern->ptr[i];iptr<A->pattern->ptr[i+1]; iptr++) { |
| 81 |
if (A->pattern->index[iptr]==i) iptr_main=iptr; |
| 82 |
for (k=0;k<n_block*n_block;++k) out->factors[n_block*n_block*iptr+k]=A->val[n_block*n_block*iptr+k]; |
| 83 |
} |
| 84 |
out->main_iptr[i]=iptr_main; |
| 85 |
if (iptr_main==A->pattern->ptr[0]-1) |
| 86 |
Paso_setError(VALUE_ERROR, "Paso_Solver_getILU: no main diagonal"); |
| 87 |
} |
| 88 |
} |
| 89 |
/* start factorization */ |
| 90 |
|
| 91 |
#pragma omp barrier |
| 92 |
for (color=0;color<out->num_colors && Paso_noError();++color) { |
| 93 |
if (n_block==1) { |
| 94 |
#pragma omp parallel for schedule(static) private(i,color2,iptr_ik,k,iptr_kj,S11,j,iptr_ij,A11,iptr_main,D) |
| 95 |
for (i = 0; i < n; ++i) { |
| 96 |
if (out->colorOf[i]==color) { |
| 97 |
for (color2=0;color2<color;++color2) { |
| 98 |
for (iptr_ik=A->pattern->ptr[i];iptr_ik<A->pattern->ptr[i+1]; ++iptr_ik) { |
| 99 |
k=A->pattern->index[iptr_ik]; |
| 100 |
if (out->colorOf[k]==color2) { |
| 101 |
A11=out->factors[iptr_ik]; |
| 102 |
/* a_ij=a_ij-a_ik*a_kj */ |
| 103 |
for (iptr_kj=A->pattern->ptr[k];iptr_kj<A->pattern->ptr[k+1]; iptr_kj++) { |
| 104 |
j=A->pattern->index[iptr_kj]; |
| 105 |
if (out->colorOf[j]>color2) { |
| 106 |
S11=out->factors[iptr_kj]; |
| 107 |
for (iptr_ij=A->pattern->ptr[i];iptr_ij<A->pattern->ptr[i+1]; iptr_ij++) { |
| 108 |
if (j==A->pattern->index[iptr_ij]) { |
| 109 |
out->factors[iptr_ij]-=A11*S11; |
| 110 |
break; |
| 111 |
} |
| 112 |
} |
| 113 |
} |
| 114 |
} |
| 115 |
} |
| 116 |
} |
| 117 |
} |
| 118 |
iptr_main=out->main_iptr[i]; |
| 119 |
D=out->factors[iptr_main]; |
| 120 |
if (ABS(D)>0.) { |
| 121 |
D=1./D; |
| 122 |
out->factors[iptr_main]=D; |
| 123 |
/* a_ik=a_ii^{-1}*a_ik */ |
| 124 |
for (iptr_ik=A->pattern->ptr[i];iptr_ik<A->pattern->ptr[i+1]; ++iptr_ik) { |
| 125 |
k=A->pattern->index[iptr_ik]; |
| 126 |
if (out->colorOf[k]>color) { |
| 127 |
A11=out->factors[iptr_ik]; |
| 128 |
out->factors[iptr_ik]=A11*D; |
| 129 |
} |
| 130 |
} |
| 131 |
} else { |
| 132 |
Paso_setError(ZERO_DIVISION_ERROR, "Paso_Solver_getILU: non-regular main diagonal block."); |
| 133 |
} |
| 134 |
} |
| 135 |
} |
| 136 |
} else if (n_block==2) { |
| 137 |
#pragma omp parallel for schedule(static) private(i,color2,iptr_ik,k,iptr_kj,S11,S21,S12,S22,j,iptr_ij,A11,A21,A12,A22,iptr_main,D) |
| 138 |
for (i = 0; i < n; ++i) { |
| 139 |
if (out->colorOf[i]==color) { |
| 140 |
for (color2=0;color2<color;++color2) { |
| 141 |
for (iptr_ik=A->pattern->ptr[i];iptr_ik<A->pattern->ptr[i+1]; ++iptr_ik) { |
| 142 |
k=A->pattern->index[iptr_ik]; |
| 143 |
if (out->colorOf[k]==color2) { |
| 144 |
A11=out->factors[iptr_ik*4 ]; |
| 145 |
A21=out->factors[iptr_ik*4+1]; |
| 146 |
A12=out->factors[iptr_ik*4+2]; |
| 147 |
A22=out->factors[iptr_ik*4+3]; |
| 148 |
/* a_ij=a_ij-a_ik*a_kj */ |
| 149 |
for (iptr_kj=A->pattern->ptr[k];iptr_kj<A->pattern->ptr[k+1]; iptr_kj++) { |
| 150 |
j=A->pattern->index[iptr_kj]; |
| 151 |
if (out->colorOf[j]>color2) { |
| 152 |
S11=out->factors[iptr_kj*4]; |
| 153 |
S21=out->factors[iptr_kj*4+1]; |
| 154 |
S12=out->factors[iptr_kj*4+2]; |
| 155 |
S22=out->factors[iptr_kj*4+3]; |
| 156 |
for (iptr_ij=A->pattern->ptr[i];iptr_ij<A->pattern->ptr[i+1]; iptr_ij++) { |
| 157 |
if (j==A->pattern->index[iptr_ij]) { |
| 158 |
out->factors[4*iptr_ij ]-=A11*S11+A12*S21; |
| 159 |
out->factors[4*iptr_ij+1]-=A21*S11+A22*S21; |
| 160 |
out->factors[4*iptr_ij+2]-=A11*S12+A12*S22; |
| 161 |
out->factors[4*iptr_ij+3]-=A21*S12+A22*S22; |
| 162 |
break; |
| 163 |
} |
| 164 |
} |
| 165 |
} |
| 166 |
} |
| 167 |
} |
| 168 |
} |
| 169 |
} |
| 170 |
iptr_main=out->main_iptr[i]; |
| 171 |
A11=out->factors[iptr_main*4]; |
| 172 |
A21=out->factors[iptr_main*4+1]; |
| 173 |
A12=out->factors[iptr_main*4+2]; |
| 174 |
A22=out->factors[iptr_main*4+3]; |
| 175 |
D = A11*A22-A12*A21; |
| 176 |
if (ABS(D)>0.) { |
| 177 |
D=1./D; |
| 178 |
S11= A22*D; |
| 179 |
S21=-A21*D; |
| 180 |
S12=-A12*D; |
| 181 |
S22= A11*D; |
| 182 |
out->factors[iptr_main*4] = S11; |
| 183 |
out->factors[iptr_main*4+1]= S21; |
| 184 |
out->factors[iptr_main*4+2]= S12; |
| 185 |
out->factors[iptr_main*4+3]= S22; |
| 186 |
/* a_ik=a_ii^{-1}*a_ik */ |
| 187 |
for (iptr_ik=A->pattern->ptr[i];iptr_ik<A->pattern->ptr[i+1]; ++iptr_ik) { |
| 188 |
k=A->pattern->index[iptr_ik]; |
| 189 |
if (out->colorOf[k]>color) { |
| 190 |
A11=out->factors[iptr_ik*4 ]; |
| 191 |
A21=out->factors[iptr_ik*4+1]; |
| 192 |
A12=out->factors[iptr_ik*4+2]; |
| 193 |
A22=out->factors[iptr_ik*4+3]; |
| 194 |
out->factors[4*iptr_ik ]=S11*A11+S12*A21; |
| 195 |
out->factors[4*iptr_ik+1]=S21*A11+S22*A21; |
| 196 |
out->factors[4*iptr_ik+2]=S11*A12+S12*A22; |
| 197 |
out->factors[4*iptr_ik+3]=S21*A12+S22*A22; |
| 198 |
} |
| 199 |
} |
| 200 |
} else { |
| 201 |
Paso_setError(ZERO_DIVISION_ERROR, "Paso_Solver_getILU: non-regular main diagonal block."); |
| 202 |
} |
| 203 |
} |
| 204 |
} |
| 205 |
} else if (n_block==3) { |
| 206 |
#pragma omp parallel for schedule(static) private(i,color2,iptr_ik,k,iptr_kj,S11,S21,S31,S12,S22,S32,S13,S23,S33,j,iptr_ij,A11,A21,A31,A12,A22,A32,A13,A23,A33,iptr_main,D) |
| 207 |
for (i = 0; i < n; ++i) { |
| 208 |
if (out->colorOf[i]==color) { |
| 209 |
for (color2=0;color2<color;++color2) { |
| 210 |
for (iptr_ik=A->pattern->ptr[i];iptr_ik<A->pattern->ptr[i+1]; ++iptr_ik) { |
| 211 |
k=A->pattern->index[iptr_ik]; |
| 212 |
if (out->colorOf[k]==color2) { |
| 213 |
A11=out->factors[iptr_ik*9 ]; |
| 214 |
A21=out->factors[iptr_ik*9+1]; |
| 215 |
A31=out->factors[iptr_ik*9+2]; |
| 216 |
A12=out->factors[iptr_ik*9+3]; |
| 217 |
A22=out->factors[iptr_ik*9+4]; |
| 218 |
A32=out->factors[iptr_ik*9+5]; |
| 219 |
A13=out->factors[iptr_ik*9+6]; |
| 220 |
A23=out->factors[iptr_ik*9+7]; |
| 221 |
A33=out->factors[iptr_ik*9+8]; |
| 222 |
/* a_ij=a_ij-a_ik*a_kj */ |
| 223 |
for (iptr_kj=A->pattern->ptr[k];iptr_kj<A->pattern->ptr[k+1]; iptr_kj++) { |
| 224 |
j=A->pattern->index[iptr_kj]; |
| 225 |
if (out->colorOf[j]>color2) { |
| 226 |
S11=out->factors[iptr_kj*9 ]; |
| 227 |
S21=out->factors[iptr_kj*9+1]; |
| 228 |
S31=out->factors[iptr_kj*9+2]; |
| 229 |
S12=out->factors[iptr_kj*9+3]; |
| 230 |
S22=out->factors[iptr_kj*9+4]; |
| 231 |
S32=out->factors[iptr_kj*9+5]; |
| 232 |
S13=out->factors[iptr_kj*9+6]; |
| 233 |
S23=out->factors[iptr_kj*9+7]; |
| 234 |
S33=out->factors[iptr_kj*9+8]; |
| 235 |
for (iptr_ij=A->pattern->ptr[i];iptr_ij<A->pattern->ptr[i+1]; iptr_ij++) { |
| 236 |
if (j==A->pattern->index[iptr_ij]) { |
| 237 |
out->factors[iptr_ij*9 ]-=A11*S11+A12*S21+A13*S31; |
| 238 |
out->factors[iptr_ij*9+1]-=A21*S11+A22*S21+A23*S31; |
| 239 |
out->factors[iptr_ij*9+2]-=A31*S11+A32*S21+A33*S31; |
| 240 |
out->factors[iptr_ij*9+3]-=A11*S12+A12*S22+A13*S32; |
| 241 |
out->factors[iptr_ij*9+4]-=A21*S12+A22*S22+A23*S32; |
| 242 |
out->factors[iptr_ij*9+5]-=A31*S12+A32*S22+A33*S32; |
| 243 |
out->factors[iptr_ij*9+6]-=A11*S13+A12*S23+A13*S33; |
| 244 |
out->factors[iptr_ij*9+7]-=A21*S13+A22*S23+A23*S33; |
| 245 |
out->factors[iptr_ij*9+8]-=A31*S13+A32*S23+A33*S33; |
| 246 |
break; |
| 247 |
} |
| 248 |
} |
| 249 |
} |
| 250 |
} |
| 251 |
} |
| 252 |
} |
| 253 |
} |
| 254 |
iptr_main=out->main_iptr[i]; |
| 255 |
A11=out->factors[iptr_main*9 ]; |
| 256 |
A21=out->factors[iptr_main*9+1]; |
| 257 |
A31=out->factors[iptr_main*9+2]; |
| 258 |
A12=out->factors[iptr_main*9+3]; |
| 259 |
A22=out->factors[iptr_main*9+4]; |
| 260 |
A32=out->factors[iptr_main*9+5]; |
| 261 |
A13=out->factors[iptr_main*9+6]; |
| 262 |
A23=out->factors[iptr_main*9+7]; |
| 263 |
A33=out->factors[iptr_main*9+8]; |
| 264 |
D = A11*(A22*A33-A23*A32)+ A12*(A31*A23-A21*A33)+A13*(A21*A32-A31*A22); |
| 265 |
if (ABS(D)>0.) { |
| 266 |
D=1./D; |
| 267 |
S11=(A22*A33-A23*A32)*D; |
| 268 |
S21=(A31*A23-A21*A33)*D; |
| 269 |
S31=(A21*A32-A31*A22)*D; |
| 270 |
S12=(A13*A32-A12*A33)*D; |
| 271 |
S22=(A11*A33-A31*A13)*D; |
| 272 |
S32=(A12*A31-A11*A32)*D; |
| 273 |
S13=(A12*A23-A13*A22)*D; |
| 274 |
S23=(A13*A21-A11*A23)*D; |
| 275 |
S33=(A11*A22-A12*A21)*D; |
| 276 |
|
| 277 |
out->factors[iptr_main*9 ]=S11; |
| 278 |
out->factors[iptr_main*9+1]=S21; |
| 279 |
out->factors[iptr_main*9+2]=S31; |
| 280 |
out->factors[iptr_main*9+3]=S12; |
| 281 |
out->factors[iptr_main*9+4]=S22; |
| 282 |
out->factors[iptr_main*9+5]=S32; |
| 283 |
out->factors[iptr_main*9+6]=S13; |
| 284 |
out->factors[iptr_main*9+7]=S23; |
| 285 |
out->factors[iptr_main*9+8]=S33; |
| 286 |
|
| 287 |
/* a_ik=a_ii^{-1}*a_ik */ |
| 288 |
for (iptr_ik=A->pattern->ptr[i];iptr_ik<A->pattern->ptr[i+1]; ++iptr_ik) { |
| 289 |
k=A->pattern->index[iptr_ik]; |
| 290 |
if (out->colorOf[k]>color) { |
| 291 |
A11=out->factors[iptr_ik*9 ]; |
| 292 |
A21=out->factors[iptr_ik*9+1]; |
| 293 |
A31=out->factors[iptr_ik*9+2]; |
| 294 |
A12=out->factors[iptr_ik*9+3]; |
| 295 |
A22=out->factors[iptr_ik*9+4]; |
| 296 |
A32=out->factors[iptr_ik*9+5]; |
| 297 |
A13=out->factors[iptr_ik*9+6]; |
| 298 |
A23=out->factors[iptr_ik*9+7]; |
| 299 |
A33=out->factors[iptr_ik*9+8]; |
| 300 |
out->factors[iptr_ik*9 ]=S11*A11+S12*A21+S13*A31; |
| 301 |
out->factors[iptr_ik*9+1]=S21*A11+S22*A21+S23*A31; |
| 302 |
out->factors[iptr_ik*9+2]=S31*A11+S32*A21+S33*A31; |
| 303 |
out->factors[iptr_ik*9+3]=S11*A12+S12*A22+S13*A32; |
| 304 |
out->factors[iptr_ik*9+4]=S21*A12+S22*A22+S23*A32; |
| 305 |
out->factors[iptr_ik*9+5]=S31*A12+S32*A22+S33*A32; |
| 306 |
out->factors[iptr_ik*9+6]=S11*A13+S12*A23+S13*A33; |
| 307 |
out->factors[iptr_ik*9+7]=S21*A13+S22*A23+S23*A33; |
| 308 |
out->factors[iptr_ik*9+8]=S31*A13+S32*A23+S33*A33; |
| 309 |
} |
| 310 |
} |
| 311 |
} else { |
| 312 |
Paso_setError(ZERO_DIVISION_ERROR, "Paso_Solver_getILU: non-regular main diagonal block."); |
| 313 |
} |
| 314 |
} |
| 315 |
} |
| 316 |
} else { |
| 317 |
Paso_setError(VALUE_ERROR, "Paso_Solver_getILU: block size greater than 3 is not supported."); |
| 318 |
} |
| 319 |
#pragma omp barrier |
| 320 |
} |
| 321 |
time_fac=Paso_timer()-time0; |
| 322 |
} |
| 323 |
|
| 324 |
TMPMEMFREE(mis_marker); |
| 325 |
if (Paso_noError()) { |
| 326 |
if (verbose) { |
| 327 |
printf("ILU: %d color used \n",out->num_colors); |
| 328 |
printf("timing: ILU: coloring/elemination : %e/%e\n",time_color,time_fac); |
| 329 |
} |
| 330 |
return out; |
| 331 |
} else { |
| 332 |
Paso_Solver_ILU_free(out); |
| 333 |
return NULL; |
| 334 |
} |
| 335 |
} |
| 336 |
|
| 337 |
/**************************************************************/ |
| 338 |
|
| 339 |
/* apply ILU precondition b-> x |
| 340 |
|
| 341 |
in fact it solves LUx=b in the form x= U^{-1} L^{-1}b |
| 342 |
|
| 343 |
should be called within a parallel region |
| 344 |
barrier synconization should be performed to make sure that the input vector available |
| 345 |
|
| 346 |
*/ |
| 347 |
|
| 348 |
void Paso_Solver_solveILU(Paso_Solver_ILU * ilu, double * x, double * b) { |
| 349 |
register dim_t i,k; |
| 350 |
register index_t color,iptr_ik,iptr_main; |
| 351 |
register double S1,S2,S3,R1,R2,R3; |
| 352 |
dim_t n_block=ilu->n_block; |
| 353 |
dim_t n=ilu->n; |
| 354 |
|
| 355 |
|
| 356 |
/* copy x into b*/ |
| 357 |
#pragma omp for private(i) schedule(static) |
| 358 |
for (i=0;i<n*n_block;++i) x[i]=b[i]; |
| 359 |
/* forward substitution */ |
| 360 |
for (color=0;color<ilu->num_colors;++color) { |
| 361 |
if (n_block==1) { |
| 362 |
#pragma omp for schedule(static) private(i,iptr_ik,k,S1,R1,iptr_main) |
| 363 |
for (i = 0; i < n; ++i) { |
| 364 |
if (ilu->colorOf[i]==color) { |
| 365 |
/* x_i=x_i-a_ik*x_k */ |
| 366 |
S1=x[i]; |
| 367 |
for (iptr_ik=ilu->pattern->ptr[i];iptr_ik<ilu->pattern->ptr[i+1]; ++iptr_ik) { |
| 368 |
k=ilu->pattern->index[iptr_ik]; |
| 369 |
if (ilu->colorOf[k]<color) { |
| 370 |
R1=x[k]; |
| 371 |
S1-=ilu->factors[iptr_ik]*R1; |
| 372 |
} |
| 373 |
} |
| 374 |
iptr_main=ilu->main_iptr[i]; |
| 375 |
x[i]=ilu->factors[iptr_main]*S1; |
| 376 |
} |
| 377 |
} |
| 378 |
} else if (n_block==2) { |
| 379 |
#pragma omp for schedule(static) private(i,iptr_ik,k,iptr_main,S1,S2,R1,R2) |
| 380 |
for (i = 0; i < n; ++i) { |
| 381 |
if (ilu->colorOf[i]==color) { |
| 382 |
/* x_i=x_i-a_ik*x_k */ |
| 383 |
S1=x[2*i]; |
| 384 |
S2=x[2*i+1]; |
| 385 |
for (iptr_ik=ilu->pattern->ptr[i];iptr_ik<ilu->pattern->ptr[i+1]; ++iptr_ik) { |
| 386 |
k=ilu->pattern->index[iptr_ik]; |
| 387 |
if (ilu->colorOf[k]<color) { |
| 388 |
R1=x[2*k]; |
| 389 |
R2=x[2*k+1]; |
| 390 |
S1-=ilu->factors[4*iptr_ik ]*R1+ilu->factors[4*iptr_ik+2]*R2; |
| 391 |
S2-=ilu->factors[4*iptr_ik+1]*R1+ilu->factors[4*iptr_ik+3]*R2; |
| 392 |
} |
| 393 |
} |
| 394 |
iptr_main=ilu->main_iptr[i]; |
| 395 |
x[2*i ]=ilu->factors[4*iptr_main ]*S1+ilu->factors[4*iptr_main+2]*S2; |
| 396 |
x[2*i+1]=ilu->factors[4*iptr_main+1]*S1+ilu->factors[4*iptr_main+3]*S2; |
| 397 |
} |
| 398 |
|
| 399 |
} |
| 400 |
} else if (n_block==3) { |
| 401 |
#pragma omp for schedule(static) private(i,iptr_ik,iptr_main,k,S1,S2,S3,R1,R2,R3) |
| 402 |
for (i = 0; i < n; ++i) { |
| 403 |
if (ilu->colorOf[i]==color) { |
| 404 |
/* x_i=x_i-a_ik*x_k */ |
| 405 |
S1=x[3*i]; |
| 406 |
S2=x[3*i+1]; |
| 407 |
S3=x[3*i+2]; |
| 408 |
for (iptr_ik=ilu->pattern->ptr[i];iptr_ik<ilu->pattern->ptr[i+1]; ++iptr_ik) { |
| 409 |
k=ilu->pattern->index[iptr_ik]; |
| 410 |
if (ilu->colorOf[k]<color) { |
| 411 |
R1=x[3*k]; |
| 412 |
R2=x[3*k+1]; |
| 413 |
R3=x[3*k+2]; |
| 414 |
S1-=ilu->factors[9*iptr_ik ]*R1+ilu->factors[9*iptr_ik+3]*R2+ilu->factors[9*iptr_ik+6]*R3; |
| 415 |
S2-=ilu->factors[9*iptr_ik+1]*R1+ilu->factors[9*iptr_ik+4]*R2+ilu->factors[9*iptr_ik+7]*R3; |
| 416 |
S3-=ilu->factors[9*iptr_ik+2]*R1+ilu->factors[9*iptr_ik+5]*R2+ilu->factors[9*iptr_ik+8]*R3; |
| 417 |
} |
| 418 |
} |
| 419 |
iptr_main=ilu->main_iptr[i]; |
| 420 |
x[3*i ]=ilu->factors[9*iptr_main ]*S1+ilu->factors[9*iptr_main+3]*S2+ilu->factors[9*iptr_main+6]*S3; |
| 421 |
x[3*i+1]=ilu->factors[9*iptr_main+1]*S1+ilu->factors[9*iptr_main+4]*S2+ilu->factors[9*iptr_main+7]*S3; |
| 422 |
x[3*i+2]=ilu->factors[9*iptr_main+2]*S1+ilu->factors[9*iptr_main+5]*S2+ilu->factors[9*iptr_main+8]*S3; |
| 423 |
} |
| 424 |
} |
| 425 |
} |
| 426 |
#pragma omp barrier |
| 427 |
} |
| 428 |
/* backward substitution */ |
| 429 |
for (color=(ilu->num_colors)-1;color>-1;--color) { |
| 430 |
if (n_block==1) { |
| 431 |
#pragma omp for schedule(static) private(i,iptr_ik,k,S1,R1) |
| 432 |
for (i = 0; i < n; ++i) { |
| 433 |
if (ilu->colorOf[i]==color) { |
| 434 |
/* x_i=x_i-a_ik*x_k */ |
| 435 |
S1=x[i]; |
| 436 |
for (iptr_ik=ilu->pattern->ptr[i];iptr_ik<ilu->pattern->ptr[i+1]; ++iptr_ik) { |
| 437 |
k=ilu->pattern->index[iptr_ik]; |
| 438 |
if (ilu->colorOf[k]>color) { |
| 439 |
R1=x[k]; |
| 440 |
S1-=ilu->factors[iptr_ik]*R1; |
| 441 |
} |
| 442 |
} |
| 443 |
x[i]=S1; |
| 444 |
} |
| 445 |
} |
| 446 |
} else if (n_block==2) { |
| 447 |
#pragma omp for schedule(static) private(i,iptr_ik,k,S1,S2,R1,R2) |
| 448 |
for (i = 0; i < n; ++i) { |
| 449 |
if (ilu->colorOf[i]==color) { |
| 450 |
/* x_i=x_i-a_ik*x_k */ |
| 451 |
S1=x[2*i]; |
| 452 |
S2=x[2*i+1]; |
| 453 |
for (iptr_ik=ilu->pattern->ptr[i];iptr_ik<ilu->pattern->ptr[i+1]; ++iptr_ik) { |
| 454 |
k=ilu->pattern->index[iptr_ik]; |
| 455 |
if (ilu->colorOf[k]>color) { |
| 456 |
R1=x[2*k]; |
| 457 |
R2=x[2*k+1]; |
| 458 |
S1-=ilu->factors[4*iptr_ik ]*R1+ilu->factors[4*iptr_ik+2]*R2; |
| 459 |
S2-=ilu->factors[4*iptr_ik+1]*R1+ilu->factors[4*iptr_ik+3]*R2; |
| 460 |
} |
| 461 |
} |
| 462 |
x[2*i]=S1; |
| 463 |
x[2*i+1]=S2; |
| 464 |
} |
| 465 |
} |
| 466 |
} else if (n_block==3) { |
| 467 |
#pragma omp for schedule(static) private(i,iptr_ik,k,S1,S2,S3,R1,R2,R3) |
| 468 |
for (i = 0; i < n; ++i) { |
| 469 |
if (ilu->colorOf[i]==color) { |
| 470 |
/* x_i=x_i-a_ik*x_k */ |
| 471 |
S1=x[3*i ]; |
| 472 |
S2=x[3*i+1]; |
| 473 |
S3=x[3*i+2]; |
| 474 |
for (iptr_ik=ilu->pattern->ptr[i];iptr_ik<ilu->pattern->ptr[i+1]; ++iptr_ik) { |
| 475 |
k=ilu->pattern->index[iptr_ik]; |
| 476 |
if (ilu->colorOf[k]>color) { |
| 477 |
R1=x[3*k]; |
| 478 |
R2=x[3*k+1]; |
| 479 |
R3=x[3*k+2]; |
| 480 |
S1-=ilu->factors[9*iptr_ik ]*R1+ilu->factors[9*iptr_ik+3]*R2+ilu->factors[9*iptr_ik+6]*R3; |
| 481 |
S2-=ilu->factors[9*iptr_ik+1]*R1+ilu->factors[9*iptr_ik+4]*R2+ilu->factors[9*iptr_ik+7]*R3; |
| 482 |
S3-=ilu->factors[9*iptr_ik+2]*R1+ilu->factors[9*iptr_ik+5]*R2+ilu->factors[9*iptr_ik+8]*R3; |
| 483 |
} |
| 484 |
} |
| 485 |
x[3*i]=S1; |
| 486 |
x[3*i+1]=S2; |
| 487 |
x[3*i+2]=S3; |
| 488 |
} |
| 489 |
} |
| 490 |
} |
| 491 |
#pragma omp barrier |
| 492 |
} |
| 493 |
return; |
| 494 |
} |
| 495 |
/* |
| 496 |
* $Log$ |
| 497 |
* Revision 1.2 2005/09/15 03:44:40 jgs |
| 498 |
* Merge of development branch dev-02 back to main trunk on 2005-09-15 |
| 499 |
* |
| 500 |
* Revision 1.1.2.1 2005/09/05 06:29:50 gross |
| 501 |
* These files have been extracted from finley to define a stand alone libray for iterative |
| 502 |
* linear solvers on the ALTIX. main entry through Paso_solve. this version compiles but |
| 503 |
* has not been tested yet. |
| 504 |
* |
| 505 |
* |
| 506 |
*/ |
Parent Directory
| 
Revision Log