1*> \brief \b DSYTRD_SB2ST reduces a real symmetric band matrix A to real symmetric tridiagonal form T 2* 3* =========== DOCUMENTATION =========== 4* 5* Online html documentation available at 6* http://www.netlib.org/lapack/explore-html/ 7* 8*> \htmlonly 9*> Download DSYTRD_SB2ST + dependencies 10*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dsytrd_sb2st.f"> 11*> [TGZ]</a> 12*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dsytrd_sb2st.f"> 13*> [ZIP]</a> 14*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dsytrd_sb2st.f"> 15*> [TXT]</a> 16*> \endhtmlonly 17* 18* Definition: 19* =========== 20* 21* SUBROUTINE DSYTRD_SB2ST( STAGE1, VECT, UPLO, N, KD, AB, LDAB, 22* D, E, HOUS, LHOUS, WORK, LWORK, INFO ) 23* 24* #if defined(_OPENMP) 25* use omp_lib 26* #endif 27* 28* IMPLICIT NONE 29* 30* .. Scalar Arguments .. 31* CHARACTER STAGE1, UPLO, VECT 32* INTEGER N, KD, IB, LDAB, LHOUS, LWORK, INFO 33* .. 34* .. Array Arguments .. 35* DOUBLE PRECISION D( * ), E( * ) 36* DOUBLE PRECISION AB( LDAB, * ), HOUS( * ), WORK( * ) 37* .. 38* 39* 40*> \par Purpose: 41* ============= 42*> 43*> \verbatim 44*> 45*> DSYTRD_SB2ST reduces a real symmetric band matrix A to real symmetric 46*> tridiagonal form T by a orthogonal similarity transformation: 47*> Q**T * A * Q = T. 48*> \endverbatim 49* 50* Arguments: 51* ========== 52* 53*> \param[in] STAGE1 54*> \verbatim 55*> STAGE1 is CHARACTER*1 56*> = 'N': "No": to mention that the stage 1 of the reduction 57*> from dense to band using the dsytrd_sy2sb routine 58*> was not called before this routine to reproduce AB. 59*> In other term this routine is called as standalone. 60*> = 'Y': "Yes": to mention that the stage 1 of the 61*> reduction from dense to band using the dsytrd_sy2sb 62*> routine has been called to produce AB (e.g., AB is 63*> the output of dsytrd_sy2sb. 64*> \endverbatim 65*> 66*> \param[in] VECT 67*> \verbatim 68*> VECT is CHARACTER*1 69*> = 'N': No need for the Housholder representation, 70*> and thus LHOUS is of size max(1, 4*N); 71*> = 'V': the Householder representation is needed to 72*> either generate or to apply Q later on, 73*> then LHOUS is to be queried and computed. 74*> (NOT AVAILABLE IN THIS RELEASE). 75*> \endverbatim 76*> 77*> \param[in] UPLO 78*> \verbatim 79*> UPLO is CHARACTER*1 80*> = 'U': Upper triangle of A is stored; 81*> = 'L': Lower triangle of A is stored. 82*> \endverbatim 83*> 84*> \param[in] N 85*> \verbatim 86*> N is INTEGER 87*> The order of the matrix A. N >= 0. 88*> \endverbatim 89*> 90*> \param[in] KD 91*> \verbatim 92*> KD is INTEGER 93*> The number of superdiagonals of the matrix A if UPLO = 'U', 94*> or the number of subdiagonals if UPLO = 'L'. KD >= 0. 95*> \endverbatim 96*> 97*> \param[in,out] AB 98*> \verbatim 99*> AB is DOUBLE PRECISION array, dimension (LDAB,N) 100*> On entry, the upper or lower triangle of the symmetric band 101*> matrix A, stored in the first KD+1 rows of the array. The 102*> j-th column of A is stored in the j-th column of the array AB 103*> as follows: 104*> if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for max(1,j-kd)<=i<=j; 105*> if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=min(n,j+kd). 106*> On exit, the diagonal elements of AB are overwritten by the 107*> diagonal elements of the tridiagonal matrix T; if KD > 0, the 108*> elements on the first superdiagonal (if UPLO = 'U') or the 109*> first subdiagonal (if UPLO = 'L') are overwritten by the 110*> off-diagonal elements of T; the rest of AB is overwritten by 111*> values generated during the reduction. 112*> \endverbatim 113*> 114*> \param[in] LDAB 115*> \verbatim 116*> LDAB is INTEGER 117*> The leading dimension of the array AB. LDAB >= KD+1. 118*> \endverbatim 119*> 120*> \param[out] D 121*> \verbatim 122*> D is DOUBLE PRECISION array, dimension (N) 123*> The diagonal elements of the tridiagonal matrix T. 124*> \endverbatim 125*> 126*> \param[out] E 127*> \verbatim 128*> E is DOUBLE PRECISION array, dimension (N-1) 129*> The off-diagonal elements of the tridiagonal matrix T: 130*> E(i) = T(i,i+1) if UPLO = 'U'; E(i) = T(i+1,i) if UPLO = 'L'. 131*> \endverbatim 132*> 133*> \param[out] HOUS 134*> \verbatim 135*> HOUS is DOUBLE PRECISION array, dimension LHOUS, that 136*> store the Householder representation. 137*> \endverbatim 138*> 139*> \param[in] LHOUS 140*> \verbatim 141*> LHOUS is INTEGER 142*> The dimension of the array HOUS. LHOUS = MAX(1, dimension) 143*> If LWORK = -1, or LHOUS=-1, 144*> then a query is assumed; the routine 145*> only calculates the optimal size of the HOUS array, returns 146*> this value as the first entry of the HOUS array, and no error 147*> message related to LHOUS is issued by XERBLA. 148*> LHOUS = MAX(1, dimension) where 149*> dimension = 4*N if VECT='N' 150*> not available now if VECT='H' 151*> \endverbatim 152*> 153*> \param[out] WORK 154*> \verbatim 155*> WORK is DOUBLE PRECISION array, dimension LWORK. 156*> \endverbatim 157*> 158*> \param[in] LWORK 159*> \verbatim 160*> LWORK is INTEGER 161*> The dimension of the array WORK. LWORK = MAX(1, dimension) 162*> If LWORK = -1, or LHOUS=-1, 163*> then a workspace query is assumed; the routine 164*> only calculates the optimal size of the WORK array, returns 165*> this value as the first entry of the WORK array, and no error 166*> message related to LWORK is issued by XERBLA. 167*> LWORK = MAX(1, dimension) where 168*> dimension = (2KD+1)*N + KD*NTHREADS 169*> where KD is the blocking size of the reduction, 170*> FACTOPTNB is the blocking used by the QR or LQ 171*> algorithm, usually FACTOPTNB=128 is a good choice 172*> NTHREADS is the number of threads used when 173*> openMP compilation is enabled, otherwise =1. 174*> \endverbatim 175*> 176*> \param[out] INFO 177*> \verbatim 178*> INFO is INTEGER 179*> = 0: successful exit 180*> < 0: if INFO = -i, the i-th argument had an illegal value 181*> \endverbatim 182* 183* Authors: 184* ======== 185* 186*> \author Univ. of Tennessee 187*> \author Univ. of California Berkeley 188*> \author Univ. of Colorado Denver 189*> \author NAG Ltd. 190* 191*> \ingroup real16OTHERcomputational 192* 193*> \par Further Details: 194* ===================== 195*> 196*> \verbatim 197*> 198*> Implemented by Azzam Haidar. 199*> 200*> All details are available on technical report, SC11, SC13 papers. 201*> 202*> Azzam Haidar, Hatem Ltaief, and Jack Dongarra. 203*> Parallel reduction to condensed forms for symmetric eigenvalue problems 204*> using aggregated fine-grained and memory-aware kernels. In Proceedings 205*> of 2011 International Conference for High Performance Computing, 206*> Networking, Storage and Analysis (SC '11), New York, NY, USA, 207*> Article 8 , 11 pages. 208*> http://doi.acm.org/10.1145/2063384.2063394 209*> 210*> A. Haidar, J. Kurzak, P. Luszczek, 2013. 211*> An improved parallel singular value algorithm and its implementation 212*> for multicore hardware, In Proceedings of 2013 International Conference 213*> for High Performance Computing, Networking, Storage and Analysis (SC '13). 214*> Denver, Colorado, USA, 2013. 215*> Article 90, 12 pages. 216*> http://doi.acm.org/10.1145/2503210.2503292 217*> 218*> A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. 219*> A novel hybrid CPU-GPU generalized eigensolver for electronic structure 220*> calculations based on fine-grained memory aware tasks. 221*> International Journal of High Performance Computing Applications. 222*> Volume 28 Issue 2, Pages 196-209, May 2014. 223*> http://hpc.sagepub.com/content/28/2/196 224*> 225*> \endverbatim 226*> 227* ===================================================================== 228 SUBROUTINE DSYTRD_SB2ST( STAGE1, VECT, UPLO, N, KD, AB, LDAB, 229 $ D, E, HOUS, LHOUS, WORK, LWORK, INFO ) 230* 231#if defined(_OPENMP) 232 use omp_lib 233#endif 234* 235 IMPLICIT NONE 236* 237* -- LAPACK computational routine -- 238* -- LAPACK is a software package provided by Univ. of Tennessee, -- 239* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- 240* 241* .. Scalar Arguments .. 242 CHARACTER STAGE1, UPLO, VECT 243 INTEGER N, KD, LDAB, LHOUS, LWORK, INFO 244* .. 245* .. Array Arguments .. 246 DOUBLE PRECISION D( * ), E( * ) 247 DOUBLE PRECISION AB( LDAB, * ), HOUS( * ), WORK( * ) 248* .. 249* 250* ===================================================================== 251* 252* .. Parameters .. 253 DOUBLE PRECISION RZERO 254 DOUBLE PRECISION ZERO, ONE 255 PARAMETER ( RZERO = 0.0D+0, 256 $ ZERO = 0.0D+0, 257 $ ONE = 1.0D+0 ) 258* .. 259* .. Local Scalars .. 260 LOGICAL LQUERY, WANTQ, UPPER, AFTERS1 261 INTEGER I, M, K, IB, SWEEPID, MYID, SHIFT, STT, ST, 262 $ ED, STIND, EDIND, BLKLASTIND, COLPT, THED, 263 $ STEPERCOL, GRSIZ, THGRSIZ, THGRNB, THGRID, 264 $ NBTILES, TTYPE, TID, NTHREADS, DEBUG, 265 $ ABDPOS, ABOFDPOS, DPOS, OFDPOS, AWPOS, 266 $ INDA, INDW, APOS, SIZEA, LDA, INDV, INDTAU, 267 $ SIDEV, SIZETAU, LDV, LHMIN, LWMIN 268* .. 269* .. External Subroutines .. 270 EXTERNAL DSB2ST_KERNELS, DLACPY, DLASET, XERBLA 271* .. 272* .. Intrinsic Functions .. 273 INTRINSIC MIN, MAX, CEILING, REAL 274* .. 275* .. External Functions .. 276 LOGICAL LSAME 277 INTEGER ILAENV2STAGE 278 EXTERNAL LSAME, ILAENV2STAGE 279* .. 280* .. Executable Statements .. 281* 282* Determine the minimal workspace size required. 283* Test the input parameters 284* 285 DEBUG = 0 286 INFO = 0 287 AFTERS1 = LSAME( STAGE1, 'Y' ) 288 WANTQ = LSAME( VECT, 'V' ) 289 UPPER = LSAME( UPLO, 'U' ) 290 LQUERY = ( LWORK.EQ.-1 ) .OR. ( LHOUS.EQ.-1 ) 291* 292* Determine the block size, the workspace size and the hous size. 293* 294 IB = ILAENV2STAGE( 2, 'DSYTRD_SB2ST', VECT, N, KD, -1, -1 ) 295 LHMIN = ILAENV2STAGE( 3, 'DSYTRD_SB2ST', VECT, N, KD, IB, -1 ) 296 LWMIN = ILAENV2STAGE( 4, 'DSYTRD_SB2ST', VECT, N, KD, IB, -1 ) 297* 298 IF( .NOT.AFTERS1 .AND. .NOT.LSAME( STAGE1, 'N' ) ) THEN 299 INFO = -1 300 ELSE IF( .NOT.LSAME( VECT, 'N' ) ) THEN 301 INFO = -2 302 ELSE IF( .NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN 303 INFO = -3 304 ELSE IF( N.LT.0 ) THEN 305 INFO = -4 306 ELSE IF( KD.LT.0 ) THEN 307 INFO = -5 308 ELSE IF( LDAB.LT.(KD+1) ) THEN 309 INFO = -7 310 ELSE IF( LHOUS.LT.LHMIN .AND. .NOT.LQUERY ) THEN 311 INFO = -11 312 ELSE IF( LWORK.LT.LWMIN .AND. .NOT.LQUERY ) THEN 313 INFO = -13 314 END IF 315* 316 IF( INFO.EQ.0 ) THEN 317 HOUS( 1 ) = LHMIN 318 WORK( 1 ) = LWMIN 319 END IF 320* 321 IF( INFO.NE.0 ) THEN 322 CALL XERBLA( 'DSYTRD_SB2ST', -INFO ) 323 RETURN 324 ELSE IF( LQUERY ) THEN 325 RETURN 326 END IF 327* 328* Quick return if possible 329* 330 IF( N.EQ.0 ) THEN 331 HOUS( 1 ) = 1 332 WORK( 1 ) = 1 333 RETURN 334 END IF 335* 336* Determine pointer position 337* 338 LDV = KD + IB 339 SIZETAU = 2 * N 340 SIDEV = 2 * N 341 INDTAU = 1 342 INDV = INDTAU + SIZETAU 343 LDA = 2 * KD + 1 344 SIZEA = LDA * N 345 INDA = 1 346 INDW = INDA + SIZEA 347 NTHREADS = 1 348 TID = 0 349* 350 IF( UPPER ) THEN 351 APOS = INDA + KD 352 AWPOS = INDA 353 DPOS = APOS + KD 354 OFDPOS = DPOS - 1 355 ABDPOS = KD + 1 356 ABOFDPOS = KD 357 ELSE 358 APOS = INDA 359 AWPOS = INDA + KD + 1 360 DPOS = APOS 361 OFDPOS = DPOS + 1 362 ABDPOS = 1 363 ABOFDPOS = 2 364 365 ENDIF 366* 367* Case KD=0: 368* The matrix is diagonal. We just copy it (convert to "real" for 369* real because D is double and the imaginary part should be 0) 370* and store it in D. A sequential code here is better or 371* in a parallel environment it might need two cores for D and E 372* 373 IF( KD.EQ.0 ) THEN 374 DO 30 I = 1, N 375 D( I ) = ( AB( ABDPOS, I ) ) 376 30 CONTINUE 377 DO 40 I = 1, N-1 378 E( I ) = RZERO 379 40 CONTINUE 380* 381 HOUS( 1 ) = 1 382 WORK( 1 ) = 1 383 RETURN 384 END IF 385* 386* Case KD=1: 387* The matrix is already Tridiagonal. We have to make diagonal 388* and offdiagonal elements real, and store them in D and E. 389* For that, for real precision just copy the diag and offdiag 390* to D and E while for the COMPLEX case the bulge chasing is 391* performed to convert the hermetian tridiagonal to symmetric 392* tridiagonal. A simpler conversion formula might be used, but then 393* updating the Q matrix will be required and based if Q is generated 394* or not this might complicate the story. 395* 396 IF( KD.EQ.1 ) THEN 397 DO 50 I = 1, N 398 D( I ) = ( AB( ABDPOS, I ) ) 399 50 CONTINUE 400* 401 IF( UPPER ) THEN 402 DO 60 I = 1, N-1 403 E( I ) = ( AB( ABOFDPOS, I+1 ) ) 404 60 CONTINUE 405 ELSE 406 DO 70 I = 1, N-1 407 E( I ) = ( AB( ABOFDPOS, I ) ) 408 70 CONTINUE 409 ENDIF 410* 411 HOUS( 1 ) = 1 412 WORK( 1 ) = 1 413 RETURN 414 END IF 415* 416* Main code start here. 417* Reduce the symmetric band of A to a tridiagonal matrix. 418* 419 THGRSIZ = N 420 GRSIZ = 1 421 SHIFT = 3 422 NBTILES = CEILING( REAL(N)/REAL(KD) ) 423 STEPERCOL = CEILING( REAL(SHIFT)/REAL(GRSIZ) ) 424 THGRNB = CEILING( REAL(N-1)/REAL(THGRSIZ) ) 425* 426 CALL DLACPY( "A", KD+1, N, AB, LDAB, WORK( APOS ), LDA ) 427 CALL DLASET( "A", KD, N, ZERO, ZERO, WORK( AWPOS ), LDA ) 428* 429* 430* openMP parallelisation start here 431* 432#if defined(_OPENMP) 433!$OMP PARALLEL PRIVATE( TID, THGRID, BLKLASTIND ) 434!$OMP$ PRIVATE( THED, I, M, K, ST, ED, STT, SWEEPID ) 435!$OMP$ PRIVATE( MYID, TTYPE, COLPT, STIND, EDIND ) 436!$OMP$ SHARED ( UPLO, WANTQ, INDV, INDTAU, HOUS, WORK) 437!$OMP$ SHARED ( N, KD, IB, NBTILES, LDA, LDV, INDA ) 438!$OMP$ SHARED ( STEPERCOL, THGRNB, THGRSIZ, GRSIZ, SHIFT ) 439!$OMP MASTER 440#endif 441* 442* main bulge chasing loop 443* 444 DO 100 THGRID = 1, THGRNB 445 STT = (THGRID-1)*THGRSIZ+1 446 THED = MIN( (STT + THGRSIZ -1), (N-1)) 447 DO 110 I = STT, N-1 448 ED = MIN( I, THED ) 449 IF( STT.GT.ED ) EXIT 450 DO 120 M = 1, STEPERCOL 451 ST = STT 452 DO 130 SWEEPID = ST, ED 453 DO 140 K = 1, GRSIZ 454 MYID = (I-SWEEPID)*(STEPERCOL*GRSIZ) 455 $ + (M-1)*GRSIZ + K 456 IF ( MYID.EQ.1 ) THEN 457 TTYPE = 1 458 ELSE 459 TTYPE = MOD( MYID, 2 ) + 2 460 ENDIF 461 462 IF( TTYPE.EQ.2 ) THEN 463 COLPT = (MYID/2)*KD + SWEEPID 464 STIND = COLPT-KD+1 465 EDIND = MIN(COLPT,N) 466 BLKLASTIND = COLPT 467 ELSE 468 COLPT = ((MYID+1)/2)*KD + SWEEPID 469 STIND = COLPT-KD+1 470 EDIND = MIN(COLPT,N) 471 IF( ( STIND.GE.EDIND-1 ).AND. 472 $ ( EDIND.EQ.N ) ) THEN 473 BLKLASTIND = N 474 ELSE 475 BLKLASTIND = 0 476 ENDIF 477 ENDIF 478* 479* Call the kernel 480* 481#if defined(_OPENMP) && _OPENMP >= 201307 482 IF( TTYPE.NE.1 ) THEN 483!$OMP TASK DEPEND(in:WORK(MYID+SHIFT-1)) 484!$OMP$ DEPEND(in:WORK(MYID-1)) 485!$OMP$ DEPEND(out:WORK(MYID)) 486 TID = OMP_GET_THREAD_NUM() 487 CALL DSB2ST_KERNELS( UPLO, WANTQ, TTYPE, 488 $ STIND, EDIND, SWEEPID, N, KD, IB, 489 $ WORK ( INDA ), LDA, 490 $ HOUS( INDV ), HOUS( INDTAU ), LDV, 491 $ WORK( INDW + TID*KD ) ) 492!$OMP END TASK 493 ELSE 494!$OMP TASK DEPEND(in:WORK(MYID+SHIFT-1)) 495!$OMP$ DEPEND(out:WORK(MYID)) 496 TID = OMP_GET_THREAD_NUM() 497 CALL DSB2ST_KERNELS( UPLO, WANTQ, TTYPE, 498 $ STIND, EDIND, SWEEPID, N, KD, IB, 499 $ WORK ( INDA ), LDA, 500 $ HOUS( INDV ), HOUS( INDTAU ), LDV, 501 $ WORK( INDW + TID*KD ) ) 502!$OMP END TASK 503 ENDIF 504#else 505 CALL DSB2ST_KERNELS( UPLO, WANTQ, TTYPE, 506 $ STIND, EDIND, SWEEPID, N, KD, IB, 507 $ WORK ( INDA ), LDA, 508 $ HOUS( INDV ), HOUS( INDTAU ), LDV, 509 $ WORK( INDW + TID*KD ) ) 510#endif 511 IF ( BLKLASTIND.GE.(N-1) ) THEN 512 STT = STT + 1 513 EXIT 514 ENDIF 515 140 CONTINUE 516 130 CONTINUE 517 120 CONTINUE 518 110 CONTINUE 519 100 CONTINUE 520* 521#if defined(_OPENMP) 522!$OMP END MASTER 523!$OMP END PARALLEL 524#endif 525* 526* Copy the diagonal from A to D. Note that D is REAL thus only 527* the Real part is needed, the imaginary part should be zero. 528* 529 DO 150 I = 1, N 530 D( I ) = ( WORK( DPOS+(I-1)*LDA ) ) 531 150 CONTINUE 532* 533* Copy the off diagonal from A to E. Note that E is REAL thus only 534* the Real part is needed, the imaginary part should be zero. 535* 536 IF( UPPER ) THEN 537 DO 160 I = 1, N-1 538 E( I ) = ( WORK( OFDPOS+I*LDA ) ) 539 160 CONTINUE 540 ELSE 541 DO 170 I = 1, N-1 542 E( I ) = ( WORK( OFDPOS+(I-1)*LDA ) ) 543 170 CONTINUE 544 ENDIF 545* 546 HOUS( 1 ) = LHMIN 547 WORK( 1 ) = LWMIN 548 RETURN 549* 550* End of DSYTRD_SB2ST 551* 552 END 553 554