| 1 |
SUBROUTINE DSPTRS( UPLO, N, NRHS, AP, IPIV, B, LDB, INFO ) |
| 2 |
* |
| 3 |
* -- LAPACK routine (version 2.0) -- |
| 4 |
* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., |
| 5 |
* Courant Institute, Argonne National Lab, and Rice University |
| 6 |
* March 31, 1993 |
| 7 |
* |
| 8 |
* .. Scalar Arguments .. |
| 9 |
CHARACTER UPLO |
| 10 |
INTEGER INFO, LDB, N, NRHS |
| 11 |
* .. |
| 12 |
* .. Array Arguments .. |
| 13 |
INTEGER IPIV( * ) |
| 14 |
DOUBLE PRECISION AP( * ), B( LDB, * ) |
| 15 |
* .. |
| 16 |
* |
| 17 |
* Purpose |
| 18 |
* ======= |
| 19 |
* |
| 20 |
* DSPTRS solves a system of linear equations A*X = B with a real |
| 21 |
* symmetric matrix A stored in packed format using the factorization |
| 22 |
* A = U*D*U**T or A = L*D*L**T computed by DSPTRF. |
| 23 |
* |
| 24 |
* Arguments |
| 25 |
* ========= |
| 26 |
* |
| 27 |
* UPLO (input) CHARACTER*1 |
| 28 |
* Specifies whether the details of the factorization are stored |
| 29 |
* as an upper or lower triangular matrix. |
| 30 |
* = 'U': Upper triangular, form is A = U*D*U**T; |
| 31 |
* = 'L': Lower triangular, form is A = L*D*L**T. |
| 32 |
* |
| 33 |
* N (input) INTEGER |
| 34 |
* The order of the matrix A. N >= 0. |
| 35 |
* |
| 36 |
* NRHS (input) INTEGER |
| 37 |
* The number of right hand sides, i.e., the number of columns |
| 38 |
* of the matrix B. NRHS >= 0. |
| 39 |
* |
| 40 |
* AP (input) DOUBLE PRECISION array, dimension (N*(N+1)/2) |
| 41 |
* The block diagonal matrix D and the multipliers used to |
| 42 |
* obtain the factor U or L as computed by DSPTRF, stored as a |
| 43 |
* packed triangular matrix. |
| 44 |
* |
| 45 |
* IPIV (input) INTEGER array, dimension (N) |
| 46 |
* Details of the interchanges and the block structure of D |
| 47 |
* as determined by DSPTRF. |
| 48 |
* |
| 49 |
* B (input/output) DOUBLE PRECISION array, dimension (LDB,NRHS) |
| 50 |
* On entry, the right hand side matrix B. |
| 51 |
* On exit, the solution matrix X. |
| 52 |
* |
| 53 |
* LDB (input) INTEGER |
| 54 |
* The leading dimension of the array B. LDB >= max(1,N). |
| 55 |
* |
| 56 |
* INFO (output) INTEGER |
| 57 |
* = 0: successful exit |
| 58 |
* < 0: if INFO = -i, the i-th argument had an illegal value |
| 59 |
* |
| 60 |
* ===================================================================== |
| 61 |
* |
| 62 |
* .. Parameters .. |
| 63 |
DOUBLE PRECISION ONE |
| 64 |
PARAMETER ( ONE = 1.0D+0 ) |
| 65 |
* .. |
| 66 |
* .. Local Scalars .. |
| 67 |
LOGICAL UPPER |
| 68 |
INTEGER J, K, KC, KP |
| 69 |
DOUBLE PRECISION AK, AKM1, AKM1K, BK, BKM1, DENOM |
| 70 |
* .. |
| 71 |
* .. External Functions .. |
| 72 |
LOGICAL LSAME |
| 73 |
EXTERNAL LSAME |
| 74 |
* .. |
| 75 |
* .. External Subroutines .. |
| 76 |
EXTERNAL DGEMV, DGER, DSCAL, DSWAP, XERBLA |
| 77 |
* .. |
| 78 |
* .. Intrinsic Functions .. |
| 79 |
INTRINSIC MAX |
| 80 |
* .. |
| 81 |
* .. Executable Statements .. |
| 82 |
* |
| 83 |
INFO = 0 |
| 84 |
UPPER = LSAME( UPLO, 'U' ) |
| 85 |
IF( .NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN |
| 86 |
INFO = -1 |
| 87 |
ELSE IF( N.LT.0 ) THEN |
| 88 |
INFO = -2 |
| 89 |
ELSE IF( NRHS.LT.0 ) THEN |
| 90 |
INFO = -3 |
| 91 |
ELSE IF( LDB.LT.MAX( 1, N ) ) THEN |
| 92 |
INFO = -7 |
| 93 |
END IF |
| 94 |
IF( INFO.NE.0 ) THEN |
| 95 |
CALL XERBLA( 'DSPTRS', -INFO ) |
| 96 |
RETURN |
| 97 |
END IF |
| 98 |
* |
| 99 |
* Quick return if possible |
| 100 |
* |
| 101 |
IF( N.EQ.0 .OR. NRHS.EQ.0 ) |
| 102 |
$ RETURN |
| 103 |
* |
| 104 |
IF( UPPER ) THEN |
| 105 |
* |
| 106 |
* Solve A*X = B, where A = U*D*U'. |
| 107 |
* |
| 108 |
* First solve U*D*X = B, overwriting B with X. |
| 109 |
* |
| 110 |
* K is the main loop index, decreasing from N to 1 in steps of |
| 111 |
* 1 or 2, depending on the size of the diagonal blocks. |
| 112 |
* |
| 113 |
K = N |
| 114 |
KC = N*( N+1 ) / 2 + 1 |
| 115 |
10 CONTINUE |
| 116 |
* |
| 117 |
* If K < 1, exit from loop. |
| 118 |
* |
| 119 |
IF( K.LT.1 ) |
| 120 |
$ GO TO 30 |
| 121 |
* |
| 122 |
KC = KC - K |
| 123 |
IF( IPIV( K ).GT.0 ) THEN |
| 124 |
* |
| 125 |
* 1 x 1 diagonal block |
| 126 |
* |
| 127 |
* Interchange rows K and IPIV(K). |
| 128 |
* |
| 129 |
KP = IPIV( K ) |
| 130 |
IF( KP.NE.K ) |
| 131 |
$ CALL DSWAP( NRHS, B( K, 1 ), LDB, B( KP, 1 ), LDB ) |
| 132 |
* |
| 133 |
* Multiply by inv(U(K)), where U(K) is the transformation |
| 134 |
* stored in column K of A. |
| 135 |
* |
| 136 |
CALL DGER( K-1, NRHS, -ONE, AP( KC ), 1, B( K, 1 ), LDB, |
| 137 |
$ B( 1, 1 ), LDB ) |
| 138 |
* |
| 139 |
* Multiply by the inverse of the diagonal block. |
| 140 |
* |
| 141 |
CALL DSCAL( NRHS, ONE / AP( KC+K-1 ), B( K, 1 ), LDB ) |
| 142 |
K = K - 1 |
| 143 |
ELSE |
| 144 |
* |
| 145 |
* 2 x 2 diagonal block |
| 146 |
* |
| 147 |
* Interchange rows K-1 and -IPIV(K). |
| 148 |
* |
| 149 |
KP = -IPIV( K ) |
| 150 |
IF( KP.NE.K-1 ) |
| 151 |
$ CALL DSWAP( NRHS, B( K-1, 1 ), LDB, B( KP, 1 ), LDB ) |
| 152 |
* |
| 153 |
* Multiply by inv(U(K)), where U(K) is the transformation |
| 154 |
* stored in columns K-1 and K of A. |
| 155 |
* |
| 156 |
CALL DGER( K-2, NRHS, -ONE, AP( KC ), 1, B( K, 1 ), LDB, |
| 157 |
$ B( 1, 1 ), LDB ) |
| 158 |
CALL DGER( K-2, NRHS, -ONE, AP( KC-( K-1 ) ), 1, |
| 159 |
$ B( K-1, 1 ), LDB, B( 1, 1 ), LDB ) |
| 160 |
* |
| 161 |
* Multiply by the inverse of the diagonal block. |
| 162 |
* |
| 163 |
AKM1K = AP( KC+K-2 ) |
| 164 |
AKM1 = AP( KC-1 ) / AKM1K |
| 165 |
AK = AP( KC+K-1 ) / AKM1K |
| 166 |
DENOM = AKM1*AK - ONE |
| 167 |
DO 20 J = 1, NRHS |
| 168 |
BKM1 = B( K-1, J ) / AKM1K |
| 169 |
BK = B( K, J ) / AKM1K |
| 170 |
B( K-1, J ) = ( AK*BKM1-BK ) / DENOM |
| 171 |
B( K, J ) = ( AKM1*BK-BKM1 ) / DENOM |
| 172 |
20 CONTINUE |
| 173 |
KC = KC - K + 1 |
| 174 |
K = K - 2 |
| 175 |
END IF |
| 176 |
* |
| 177 |
GO TO 10 |
| 178 |
30 CONTINUE |
| 179 |
* |
| 180 |
* Next solve U'*X = B, overwriting B with X. |
| 181 |
* |
| 182 |
* K is the main loop index, increasing from 1 to N in steps of |
| 183 |
* 1 or 2, depending on the size of the diagonal blocks. |
| 184 |
* |
| 185 |
K = 1 |
| 186 |
KC = 1 |
| 187 |
40 CONTINUE |
| 188 |
* |
| 189 |
* If K > N, exit from loop. |
| 190 |
* |
| 191 |
IF( K.GT.N ) |
| 192 |
$ GO TO 50 |
| 193 |
* |
| 194 |
IF( IPIV( K ).GT.0 ) THEN |
| 195 |
* |
| 196 |
* 1 x 1 diagonal block |
| 197 |
* |
| 198 |
* Multiply by inv(U'(K)), where U(K) is the transformation |
| 199 |
* stored in column K of A. |
| 200 |
* |
| 201 |
CALL DGEMV( 'Transpose', K-1, NRHS, -ONE, B, LDB, AP( KC ), |
| 202 |
$ 1, ONE, B( K, 1 ), LDB ) |
| 203 |
* |
| 204 |
* Interchange rows K and IPIV(K). |
| 205 |
* |
| 206 |
KP = IPIV( K ) |
| 207 |
IF( KP.NE.K ) |
| 208 |
$ CALL DSWAP( NRHS, B( K, 1 ), LDB, B( KP, 1 ), LDB ) |
| 209 |
KC = KC + K |
| 210 |
K = K + 1 |
| 211 |
ELSE |
| 212 |
* |
| 213 |
* 2 x 2 diagonal block |
| 214 |
* |
| 215 |
* Multiply by inv(U'(K+1)), where U(K+1) is the transformation |
| 216 |
* stored in columns K and K+1 of A. |
| 217 |
* |
| 218 |
CALL DGEMV( 'Transpose', K-1, NRHS, -ONE, B, LDB, AP( KC ), |
| 219 |
$ 1, ONE, B( K, 1 ), LDB ) |
| 220 |
CALL DGEMV( 'Transpose', K-1, NRHS, -ONE, B, LDB, |
| 221 |
$ AP( KC+K ), 1, ONE, B( K+1, 1 ), LDB ) |
| 222 |
* |
| 223 |
* Interchange rows K and -IPIV(K). |
| 224 |
* |
| 225 |
KP = -IPIV( K ) |
| 226 |
IF( KP.NE.K ) |
| 227 |
$ CALL DSWAP( NRHS, B( K, 1 ), LDB, B( KP, 1 ), LDB ) |
| 228 |
KC = KC + 2*K + 1 |
| 229 |
K = K + 2 |
| 230 |
END IF |
| 231 |
* |
| 232 |
GO TO 40 |
| 233 |
50 CONTINUE |
| 234 |
* |
| 235 |
ELSE |
| 236 |
* |
| 237 |
* Solve A*X = B, where A = L*D*L'. |
| 238 |
* |
| 239 |
* First solve L*D*X = B, overwriting B with X. |
| 240 |
* |
| 241 |
* K is the main loop index, increasing from 1 to N in steps of |
| 242 |
* 1 or 2, depending on the size of the diagonal blocks. |
| 243 |
* |
| 244 |
K = 1 |
| 245 |
KC = 1 |
| 246 |
60 CONTINUE |
| 247 |
* |
| 248 |
* If K > N, exit from loop. |
| 249 |
* |
| 250 |
IF( K.GT.N ) |
| 251 |
$ GO TO 80 |
| 252 |
* |
| 253 |
IF( IPIV( K ).GT.0 ) THEN |
| 254 |
* |
| 255 |
* 1 x 1 diagonal block |
| 256 |
* |
| 257 |
* Interchange rows K and IPIV(K). |
| 258 |
* |
| 259 |
KP = IPIV( K ) |
| 260 |
IF( KP.NE.K ) |
| 261 |
$ CALL DSWAP( NRHS, B( K, 1 ), LDB, B( KP, 1 ), LDB ) |
| 262 |
* |
| 263 |
* Multiply by inv(L(K)), where L(K) is the transformation |
| 264 |
* stored in column K of A. |
| 265 |
* |
| 266 |
IF( K.LT.N ) |
| 267 |
$ CALL DGER( N-K, NRHS, -ONE, AP( KC+1 ), 1, B( K, 1 ), |
| 268 |
$ LDB, B( K+1, 1 ), LDB ) |
| 269 |
* |
| 270 |
* Multiply by the inverse of the diagonal block. |
| 271 |
* |
| 272 |
CALL DSCAL( NRHS, ONE / AP( KC ), B( K, 1 ), LDB ) |
| 273 |
KC = KC + N - K + 1 |
| 274 |
K = K + 1 |
| 275 |
ELSE |
| 276 |
* |
| 277 |
* 2 x 2 diagonal block |
| 278 |
* |
| 279 |
* Interchange rows K+1 and -IPIV(K). |
| 280 |
* |
| 281 |
KP = -IPIV( K ) |
| 282 |
IF( KP.NE.K+1 ) |
| 283 |
$ CALL DSWAP( NRHS, B( K+1, 1 ), LDB, B( KP, 1 ), LDB ) |
| 284 |
* |
| 285 |
* Multiply by inv(L(K)), where L(K) is the transformation |
| 286 |
* stored in columns K and K+1 of A. |
| 287 |
* |
| 288 |
IF( K.LT.N-1 ) THEN |
| 289 |
CALL DGER( N-K-1, NRHS, -ONE, AP( KC+2 ), 1, B( K, 1 ), |
| 290 |
$ LDB, B( K+2, 1 ), LDB ) |
| 291 |
CALL DGER( N-K-1, NRHS, -ONE, AP( KC+N-K+2 ), 1, |
| 292 |
$ B( K+1, 1 ), LDB, B( K+2, 1 ), LDB ) |
| 293 |
END IF |
| 294 |
* |
| 295 |
* Multiply by the inverse of the diagonal block. |
| 296 |
* |
| 297 |
AKM1K = AP( KC+1 ) |
| 298 |
AKM1 = AP( KC ) / AKM1K |
| 299 |
AK = AP( KC+N-K+1 ) / AKM1K |
| 300 |
DENOM = AKM1*AK - ONE |
| 301 |
DO 70 J = 1, NRHS |
| 302 |
BKM1 = B( K, J ) / AKM1K |
| 303 |
BK = B( K+1, J ) / AKM1K |
| 304 |
B( K, J ) = ( AK*BKM1-BK ) / DENOM |
| 305 |
B( K+1, J ) = ( AKM1*BK-BKM1 ) / DENOM |
| 306 |
70 CONTINUE |
| 307 |
KC = KC + 2*( N-K ) + 1 |
| 308 |
K = K + 2 |
| 309 |
END IF |
| 310 |
* |
| 311 |
GO TO 60 |
| 312 |
80 CONTINUE |
| 313 |
* |
| 314 |
* Next solve L'*X = B, overwriting B with X. |
| 315 |
* |
| 316 |
* K is the main loop index, decreasing from N to 1 in steps of |
| 317 |
* 1 or 2, depending on the size of the diagonal blocks. |
| 318 |
* |
| 319 |
K = N |
| 320 |
KC = N*( N+1 ) / 2 + 1 |
| 321 |
90 CONTINUE |
| 322 |
* |
| 323 |
* If K < 1, exit from loop. |
| 324 |
* |
| 325 |
IF( K.LT.1 ) |
| 326 |
$ GO TO 100 |
| 327 |
* |
| 328 |
KC = KC - ( N-K+1 ) |
| 329 |
IF( IPIV( K ).GT.0 ) THEN |
| 330 |
* |
| 331 |
* 1 x 1 diagonal block |
| 332 |
* |
| 333 |
* Multiply by inv(L'(K)), where L(K) is the transformation |
| 334 |
* stored in column K of A. |
| 335 |
* |
| 336 |
IF( K.LT.N ) |
| 337 |
$ CALL DGEMV( 'Transpose', N-K, NRHS, -ONE, B( K+1, 1 ), |
| 338 |
$ LDB, AP( KC+1 ), 1, ONE, B( K, 1 ), LDB ) |
| 339 |
* |
| 340 |
* Interchange rows K and IPIV(K). |
| 341 |
* |
| 342 |
KP = IPIV( K ) |
| 343 |
IF( KP.NE.K ) |
| 344 |
$ CALL DSWAP( NRHS, B( K, 1 ), LDB, B( KP, 1 ), LDB ) |
| 345 |
K = K - 1 |
| 346 |
ELSE |
| 347 |
* |
| 348 |
* 2 x 2 diagonal block |
| 349 |
* |
| 350 |
* Multiply by inv(L'(K-1)), where L(K-1) is the transformation |
| 351 |
* stored in columns K-1 and K of A. |
| 352 |
* |
| 353 |
IF( K.LT.N ) THEN |
| 354 |
CALL DGEMV( 'Transpose', N-K, NRHS, -ONE, B( K+1, 1 ), |
| 355 |
$ LDB, AP( KC+1 ), 1, ONE, B( K, 1 ), LDB ) |
| 356 |
CALL DGEMV( 'Transpose', N-K, NRHS, -ONE, B( K+1, 1 ), |
| 357 |
$ LDB, AP( KC-( N-K ) ), 1, ONE, B( K-1, 1 ), |
| 358 |
$ LDB ) |
| 359 |
END IF |
| 360 |
* |
| 361 |
* Interchange rows K and -IPIV(K). |
| 362 |
* |
| 363 |
KP = -IPIV( K ) |
| 364 |
IF( KP.NE.K ) |
| 365 |
$ CALL DSWAP( NRHS, B( K, 1 ), LDB, B( KP, 1 ), LDB ) |
| 366 |
KC = KC - ( N-K+2 ) |
| 367 |
K = K - 2 |
| 368 |
END IF |
| 369 |
* |
| 370 |
GO TO 90 |
| 371 |
100 CONTINUE |
| 372 |
END IF |
| 373 |
* |
| 374 |
RETURN |
| 375 |
* |
| 376 |
* End of DSPTRS |
| 377 |
* |
| 378 |
END |