test_check_correctness_template.F90 25.9 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
!    This file is part of ELPA.
!
!    The ELPA library was originally created by the ELPA consortium,
!    consisting of the following organizations:
!
!    - Max Planck Computing and Data Facility (MPCDF), formerly known as
!      Rechenzentrum Garching der Max-Planck-Gesellschaft (RZG),
!    - Bergische Universität Wuppertal, Lehrstuhl für angewandte
!      Informatik,
!    - Technische Universität München, Lehrstuhl für Informatik mit
!      Schwerpunkt Wissenschaftliches Rechnen ,
!    - Fritz-Haber-Institut, Berlin, Abt. Theorie,
!    - Max-Plack-Institut für Mathematik in den Naturwissenschaften,
!      Leipzig, Abt. Komplexe Strukutren in Biologie und Kognition,
!      and
!    - IBM Deutschland GmbH
!
!
!    More information can be found here:
!    http://elpa.mpcdf.mpg.de/
!
!    ELPA is free software: you can redistribute it and/or modify
!    it under the terms of the version 3 of the license of the
!    GNU Lesser General Public License as published by the Free
!    Software Foundation.
!
!    ELPA is distributed in the hope that it will be useful,
!    but WITHOUT ANY WARRANTY; without even the implied warranty of
!    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
!    GNU Lesser General Public License for more details.
!
!    You should have received a copy of the GNU Lesser General Public License
!    along with ELPA.  If not, see <http://www.gnu.org/licenses/>
!
!    ELPA reflects a substantial effort on the part of the original
!    ELPA consortium, and we ask you to respect the spirit of the
!    license that we chose: i.e., please contribute any changes you
!    may have back to the original ELPA library distribution, and keep
!    any derivatives of ELPA under the same license that we chose for
!    the original distribution, the GNU Lesser General Public License.
!
! Author: A. Marek, MPCDF

44
    function check_correctness_evp_numeric_residuals_&
45
46
47
    &MATH_DATATYPE&
    &_&
    &PRECISION&
Pavel Kus's avatar
Pavel Kus committed
48
    & (na, nev, as, z, ev, sc_desc, nblk, myid, np_rows, np_cols, my_prow, my_pcol) result(status)
49
      implicit none
50
#include "../../src/general/precision_kinds.F90"
51
      integer(kind=ik)                 :: status
Pavel Kus's avatar
Pavel Kus committed
52
      integer(kind=ik), intent(in)     :: na, nev, nblk, myid, np_rows, np_cols, my_prow, my_pcol
53
54
55
56
      MATH_DATATYPE(kind=rck), intent(in)     :: as(:,:), z(:,:)
      real(kind=rk)                 :: ev(:)
      MATH_DATATYPE(kind=rck), dimension(size(as,dim=1),size(as,dim=2)) :: tmp1, tmp2
      MATH_DATATYPE(kind=rck)                :: xc
57

58
#ifndef WITH_MPI
59
60
#if REALCASE == 1
      real(kind=rck)                   :: dnrm2, snrm2
61
62
#endif
#if COMPLEXCASE == 1
63
      complex(kind=rck)                :: zdotc, cdotc
64
#endif /* COMPLEXCASE */
65
#endif
66
67
68

      integer(kind=ik)                 :: sc_desc(:)

Pavel Kus's avatar
Pavel Kus committed
69
      integer(kind=ik)                 :: i, rowLocal, colLocal
70
      real(kind=rck)                   :: err, errmax
71
72
73
74
75
76
77
78
79
80

      integer :: mpierr

      status = 0

      ! 1. Residual (maximum of || A*Zi - Zi*EVi ||)
      ! tmp1 =  A * Z
      ! as is original stored matrix, Z are the EVs

#ifdef WITH_MPI
81
82
      call scal_PRECISION_GEMM('N', 'N', na, nev, na, ONE, as, 1, 1, sc_desc, &
                  z, 1, 1, sc_desc, ZERO, tmp1, 1, 1, sc_desc)
83
#else /* WITH_MPI */
84
      call PRECISION_GEMM('N','N',na,nev,na,ONE,as,na,z,na,ZERO,tmp1,na)
85
86
87
88
89
90
#endif /* WITH_MPI */


      ! tmp2 = Zi*EVi
      tmp2(:,:) = z(:,:)
      do i=1,nev
Pavel Kus's avatar
Pavel Kus committed
91
        xc = ev(i)
92
93
94
95
#if REALCASE == 1
#ifdef WITH_MPI

#ifdef DOUBLE_PRECISION_REAL
Pavel Kus's avatar
Pavel Kus committed
96
        call pdscal(na, xc, tmp2, 1, i, sc_desc, 1)
97
#else
Pavel Kus's avatar
Pavel Kus committed
98
        call psscal(na, xc, tmp2, 1, i, sc_desc, 1)
99
100
101
102
103
#endif

#else /* WITH_MPI */

#ifdef DOUBLE_PRECISION_REAL
Pavel Kus's avatar
Pavel Kus committed
104
        call dscal(na,xc,tmp2(:,i),1)
105
#else
Pavel Kus's avatar
Pavel Kus committed
106
        call sscal(na,xc,tmp2(:,i),1)
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
#endif

#endif /* WITH_MPI */
#endif /* REALCASE */

#if COMPLEXCASE == 1
#ifdef WITH_MPI

#ifdef DOUBLE_PRECISION_COMPLEX
        call pzscal(na, xc, tmp2, 1, i, sc_desc, 1)
#else
        call pcscal(na, xc, tmp2, 1, i, sc_desc, 1)
#endif

#else /* WITH_MPI */

#ifdef DOUBLE_PRECISION_COMPLEX
        call zscal(na,xc,tmp2(1,i),1)
#else
        call cscal(na,xc,tmp2(1,i),1)
#endif

#endif /* WITH_MPI */
#endif /* COMPLEXCASE */
      enddo

      !  tmp1 = A*Zi - Zi*EVi
      tmp1(:,:) =  tmp1(:,:) - tmp2(:,:)

      ! Get maximum norm of columns of tmp1
137
      errmax = 0.0_rk
138
139
140

      do i=1,nev
#if REALCASE == 1
141
        err = 0.0_rk
142
#ifdef WITH_MPI
Pavel Kus's avatar
Pavel Kus committed
143
        call scal_PRECISION_NRM2(na, err, tmp1, 1, i, sc_desc, 1)
144
#else /* WITH_MPI */
Pavel Kus's avatar
Pavel Kus committed
145
        err = PRECISION_NRM2(na,tmp1(1,i),1)
146
147
148
149
150
151
152
#endif /* WITH_MPI */
        errmax = max(errmax, err)
#endif /* REALCASE */

#if COMPLEXCASE == 1
        xc = 0
#ifdef WITH_MPI
Pavel Kus's avatar
Pavel Kus committed
153
        call scal_PRECISION_DOTC(na, xc, tmp1, 1, i, sc_desc, 1, tmp1, 1, i, sc_desc, 1)
154
#else /* WITH_MPI */
Pavel Kus's avatar
Pavel Kus committed
155
        xc = PRECISION_DOTC(na,tmp1,1,tmp1,1)
156
#endif /* WITH_MPI */
157
        errmax = max(errmax, sqrt(real(xc,kind=REAL_DATATYPE)))
158
159
160
161
162
163
#endif /* COMPLEXCASE */
      enddo

      ! Get maximum error norm over all processors
      err = errmax
#ifdef WITH_MPI
164
      call mpi_allreduce(err, errmax, 1, MPI_REAL_PRECISION, MPI_MAX, MPI_COMM_WORLD, mpierr)
165
166
167
#else /* WITH_MPI */
      errmax = err
#endif /* WITH_MPI */
168
      if (myid==0) print *,'Results of numerical residual checks:'
169
170
      if (myid==0) print *,'Error Residual     :',errmax
#if REALCASE == 1
171
      if (nev .ge. 2) then
172
#ifdef DOUBLE_PRECISION_REAL
173
        if (errmax .gt. 5e-12_rk8 .or. errmax .eq. 0.0_rk8) then
174
#else
175
        if (errmax .gt. 3e-3_rk4 .or. errmax .eq. 0.0_rk4) then
176
#endif
177
178
179
180
181
182
183
184
185
186
187
          status = 1
        endif
      else
#ifdef DOUBLE_PRECISION_REAL
        if (errmax .gt. 5e-12_rk8) then
#else
        if (errmax .gt. 3e-3_rk4) then
#endif
          status = 1
        endif
      endif
188
189
#endif
#if COMPLEXCASE == 1
190
      if (nev .gt. 2) then
191
#ifdef DOUBLE_PRECISION_COMPLEX
192
        if (errmax .gt. 5e-12_rk8 .or. errmax .eq. 0.0_rk8) then
193
#else
194
        if (errmax .gt. 3e-3_rk4 .or. errmax .eq. 0.0_rk4) then
195
#endif
196
197
198
199
200
201
202
          status =1
        endif
      else
#ifdef DOUBLE_PRECISION_COMPLEX
        if (errmax .gt. 5e-12_rk8) then
#else
        if (errmax .gt. 3e-3_rk4) then
203
#endif
204
205
          status = 1
        endif
206
      endif
207
#endif
208
209
210
211
212
213

      ! 2. Eigenvector orthogonality

      ! tmp1 = Z**T * Z
      tmp1 = 0
#ifdef WITH_MPI
214
215
      call scal_PRECISION_GEMM(BLAS_TRANS_OR_CONJ, 'N', nev, nev, na, ONE, z, 1, 1, &
                        sc_desc, z, 1, 1, sc_desc, ZERO, tmp1, 1, 1, sc_desc)
216
#else /* WITH_MPI */
217
      call PRECISION_GEMM(BLAS_TRANS_OR_CONJ,'N',nev,nev,na,ONE,z,na,z,na,ZERO,tmp1,na)
218
#endif /* WITH_MPI */
Pavel Kus's avatar
Pavel Kus committed
219
220
      !TODO for the C interface, not all information is passed (zeros instead)
      !TODO than this part of the test cannot be done
221
      !TODO either we will not have this part of test at all, or it will be improved
Pavel Kus's avatar
Pavel Kus committed
222
223
      if(nblk > 0) then
        ! First check, whether the elements on diagonal are 1 .. "normality" of the vectors
224
        err = 0.0_rk
Pavel Kus's avatar
Pavel Kus committed
225
226
        do i=1, nev
          if (map_global_array_index_to_local_index(i, i, rowLocal, colLocal, nblk, np_rows, np_cols, my_prow, my_pcol)) then
227
             err = max(err, abs(tmp1(rowLocal,colLocal) - 1.0_rk))
Pavel Kus's avatar
Pavel Kus committed
228
229
230
231
232
233
234
235
236
           endif
        end do
#ifdef WITH_MPI
        call mpi_allreduce(err, errmax, 1, MPI_REAL_PRECISION, MPI_MAX, MPI_COMM_WORLD, mpierr)
#else /* WITH_MPI */
        errmax = err
#endif /* WITH_MPI */
        if (myid==0) print *,'Maximal error in eigenvector lengths:',errmax
      end if
237

Pavel Kus's avatar
Pavel Kus committed
238
      ! Second, find the maximal error in the whole Z**T * Z matrix (its diference from identity matrix)
239
240
241
      ! Initialize tmp2 to unit matrix
      tmp2 = 0
#ifdef WITH_MPI
Pavel Kus's avatar
Pavel Kus committed
242
      call scal_PRECISION_LASET('A', nev, nev, ZERO, ONE, tmp2, 1, 1, sc_desc)
243
#else /* WITH_MPI */
Pavel Kus's avatar
Pavel Kus committed
244
      call PRECISION_LASET('A',nev,nev,ZERO,ONE,tmp2,na)
245
246
247
248
249
250
251
252
#endif /* WITH_MPI */

      !      ! tmp1 = Z**T * Z - Unit Matrix
      tmp1(:,:) =  tmp1(:,:) - tmp2(:,:)

      ! Get maximum error (max abs value in tmp1)
      err = maxval(abs(tmp1))
#ifdef WITH_MPI
253
      call mpi_allreduce(err, errmax, 1, MPI_REAL_PRECISION, MPI_MAX, MPI_COMM_WORLD, mpierr)
254
255
256
257
258
259
#else /* WITH_MPI */
      errmax = err
#endif /* WITH_MPI */

      if (myid==0) print *,'Error Orthogonality:',errmax
#if REALCASE == 1
260
      if (nev .ge. 2) then
261
#ifdef DOUBLE_PRECISION_REAL
262
        if (errmax .gt. 5e-12_rk8 .or. errmax .eq. 0.0_rk8) then
263
#else
264
        if (errmax .gt. 9e-4_rk4 .or. errmax .eq. 0.0_rk4) then
265
#endif
266
267
268
269
270
271
272
273
274
275
276
          status = 1
        endif
      else
#ifdef DOUBLE_PRECISION_REAL
        if (errmax .gt. 5e-12_rk8) then
#else
        if (errmax .gt. 9e-4_rk4) then
#endif
          status = 1
        endif
      endif
277
278
#endif
#if COMPLEXCASE == 1
279
280
281
282
283
284
285
286
287
      if (nev .ge. 2) then
#ifdef DOUBLE_PRECISION_COMPLEX
        if (errmax .gt. 5e-12_rk8 .or. errmax .eq. 0.0_rk8) then
#else
        if (errmax .gt. 9e-4_rk4 .or. errmax .eq. 0.0_rk4) then
#endif
          status = 1
        endif
      else
288
#ifdef DOUBLE_PRECISION_COMPLEX
289
        if (errmax .gt. 5e-12_rk8) then
290
#else
291
        if (errmax .gt. 9e-4_rk4) then
292
#endif
293
294
295
          status = 1
        endif
      endif
296
297
298
299
300
301
302
303
#endif

    end function


#if REALCASE == 1

#ifdef DOUBLE_PRECISION_REAL
304
    !c> int check_correctness_evp_numeric_residuals_real_double_f(int na, int nev, int na_rows, int na_cols,
305
306
307
    !c>                                         double *as, double *z, double *ev,
    !c>                                         int sc_desc[9], int myid);
#else
308
    !c> int check_correctness_evp_numeric_residuals_real_single_f(int na, int nev, int na_rows, int na_cols,
309
310
311
312
313
314
315
316
    !c>                                         float *as, float *z, float *ev,
    !c>                                         int sc_desc[9], int myid);
#endif

#endif /* REALCASE */

#if COMPLEXCASE == 1
#ifdef DOUBLE_PRECISION_COMPLEX
317
    !c> int check_correctness_evp_numeric_residuals_complex_double_f(int na, int nev, int na_rows, int na_cols,
318
319
320
    !c>                                         complex double *as, complex double *z, double *ev,
    !c>                                         int sc_desc[9], int myid);
#else
321
    !c> int check_correctness_evp_numeric_residuals_complex_single_f(int na, int nev, int na_rows, int na_cols,
Andreas Marek's avatar
Andreas Marek committed
322
    !c>                                         complex float *as, complex float *z, float *ev,
323
324
325
326
    !c>                                         int sc_desc[9], int myid);
#endif
#endif /* COMPLEXCASE */

327
function check_correctness_evp_numeric_residuals_&
328
329
330
331
&MATH_DATATYPE&
&_&
&PRECISION&
&_f (na, nev, na_rows, na_cols, as, z, ev, sc_desc, myid) result(status) &
332
      bind(C,name="check_correctness_evp_numeric_residuals_&
333
334
335
336
337
338
339
340
      &MATH_DATATYPE&
      &_&
      &PRECISION&
      &_f")

      use iso_c_binding

      implicit none
341
#include "../../src/general/precision_kinds.F90"
342
343
344

      integer(kind=c_int)            :: status
      integer(kind=c_int), value     :: na, nev, myid, na_rows, na_cols
345
346
      MATH_DATATYPE(kind=rck)     :: as(1:na_rows,1:na_cols), z(1:na_rows,1:na_cols)
      real(kind=rck)    :: ev(1:na)
347
348
      integer(kind=c_int)            :: sc_desc(1:9)

Pavel Kus's avatar
Pavel Kus committed
349
350
351
      ! TODO: I did not want to add all the variables to the C interface as well
      ! TODO: I think that we should find a better way to pass this information
      ! TODO: to all the functions anyway (get it from sc_desc, pass elpa_t, etc..)
352
      status = check_correctness_evp_numeric_residuals_&
353
354
355
      &MATH_DATATYPE&
      &_&
      &PRECISION&
Pavel Kus's avatar
Pavel Kus committed
356
      & (na, nev, as, z, ev, sc_desc, 0, myid, 0, 0, 0, 0)
357
358
359

    end function

Andreas Marek's avatar
Andreas Marek committed
360
361
362
363
364
365
366
    function check_correctness_eigenvalues_toeplitz_&
    &MATH_DATATYPE&
    &_&
    &PRECISION&
    & (na, diagonalElement, subdiagonalElement, ev, z, myid) result(status)
      use iso_c_binding
      implicit none
367
#include "../../src/general/precision_kinds.F90"
Andreas Marek's avatar
Andreas Marek committed
368
369
370

      integer               :: status, ii, j, myid
      integer, intent(in)   :: na
371
372
373
      real(kind=rck) :: diagonalElement, subdiagonalElement
      real(kind=rck) :: ev_analytic(na), ev(na)
      MATH_DATATYPE(kind=rck) :: z(:,:)
Andreas Marek's avatar
Andreas Marek committed
374
375

#if defined(DOUBLE_PRECISION_REAL) || defined(DOUBLE_PRECISION_COMPLEX)
376
      real(kind=rck), parameter   :: pi = 3.141592653589793238462643383279_c_double
Andreas Marek's avatar
Andreas Marek committed
377
#else
378
      real(kind=rck), parameter   :: pi = 3.1415926535897932_c_float
Andreas Marek's avatar
Andreas Marek committed
379
#endif
380
      real(kind=rck)              :: tmp, maxerr
381
      integer                     :: loctmp
Andreas Marek's avatar
Andreas Marek committed
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
      status = 0

     ! analytic solution
     do ii=1, na
#if defined(DOUBLE_PRECISION_REAL) || defined(DOUBLE_PRECISION_COMPLEX)
       ev_analytic(ii) = diagonalElement + 2.0_c_double * &
                         subdiagonalElement *cos( pi*real(ii,kind=c_double)/ &
			 real(na+1,kind=c_double) )
#else
       ev_analytic(ii) = diagonalElement + 2.0_c_float * &
                         subdiagonalElement *cos( pi*real(ii,kind=c_float)/ &
                         real(na+1,kind=c_float) )
#endif
     enddo

     ! sort analytic solution:

     ! this hack is neither elegant, nor optimized: for huge matrixes it might be expensive
     ! a proper sorting algorithmus might be implemented here

     tmp    = minval(ev_analytic)
     loctmp = minloc(ev_analytic, 1)

     ev_analytic(loctmp) = ev_analytic(1)
     ev_analytic(1) = tmp
     do ii=2, na
       tmp = ev_analytic(ii)
       do j= ii, na
         if (ev_analytic(j) .lt. tmp) then
           tmp    = ev_analytic(j)
           loctmp = j
         endif
       enddo
       ev_analytic(loctmp) = ev_analytic(ii)
       ev_analytic(ii) = tmp
     enddo

     ! compute a simple error max of eigenvalues
     maxerr = 0.0
     maxerr = maxval( (ev(:) - ev_analytic(:))/ev_analytic(:) , 1)

#if defined(DOUBLE_PRECISION_REAL) || defined(DOUBLE_PRECISION_COMPLEX)
424
     if (maxerr .gt. 8.e-13_c_double .or. maxerr .eq. 0.0_c_double) then
Andreas Marek's avatar
Andreas Marek committed
425
#else
426
     if (maxerr .gt. 8.e-4_c_float .or. maxerr .eq. 0.0_c_float) then
Andreas Marek's avatar
Andreas Marek committed
427
428
429
430
431
432
433
#endif
       status = 1
       if (myid .eq. 0) then
         print *,"Eigenvalues differ from analytic solution: maxerr = ",maxerr
       endif
     endif
    end function
434

435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
    function check_correctness_cholesky_&
    &MATH_DATATYPE&
    &_&
    &PRECISION&
    & (na, a, as, na_rows, sc_desc, myid) result(status)
      implicit none
#include "../../src/general/precision_kinds.F90"
      integer(kind=ik)                 :: status
      integer(kind=ik), intent(in)     :: na, myid, na_rows
#if REALCASE == 1
      real(kind=rck), intent(in)       :: a(:,:), as(:,:)
      real(kind=rck), dimension(size(as,dim=1),size(as,dim=2)) :: tmp1, tmp2
      real(kind=rck)                   :: norm, normmax

#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL
      real(kind=rck)                   :: pdlange
#else
      real(kind=rck)                   :: pslange
#endif

#else /* WITH_MPI */

#ifdef DOUBLE_PRECISION_REAL
      real(kind=rck)                   :: dlange
#else
      real(kind=rck)                   :: slange
#endif

#endif /* WITH_MPI */

#endif /* REALCASE */

#if COMPLEXCASE == 1
      complex(kind=rck), intent(in)    :: a(:,:), as(:,:)
      complex(kind=rck), dimension(size(as,dim=1),size(as,dim=2)) :: tmp1, tmp2
      real(kind=rck)                :: norm, normmax
#ifdef DOUBLE_PRECISION_COMPLEX
      complex(kind=ck8), parameter   :: CZERO = (0.0_rk8,0.0_rk8), CONE = (1.0_rk8,0.0_rk8)
#else
      complex(kind=ck4), parameter   :: CZERO = (0.0_rk4,0.0_rk4), CONE = (1.0_rk4,0.0_rk8)
#endif

#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX
Andreas Marek's avatar
Andreas Marek committed
480
      real(kind=rck)                   :: pzlange
481
#else
Andreas Marek's avatar
Andreas Marek committed
482
      real(kind=rck)                   :: pclange
483
484
485
486
487
#endif

#else /* WITH_MPI */

#ifdef DOUBLE_PRECISION_COMPLEX
Andreas Marek's avatar
Andreas Marek committed
488
      real(kind=rck)                   :: zlange
489
#else
Andreas Marek's avatar
Andreas Marek committed
490
      real(kind=rck)                   :: clange
491
492
493
494
495
496
497
498
499
500
501
502
503
504
#endif

#endif /* WITH_MPI */
#endif /* COMPLEXCASE */

      integer(kind=ik)                 :: sc_desc(:)

      real(kind=rck)                   :: err, errmax

      integer :: mpierr

      status = 0

#if REALCASE == 1
505
#ifdef DOUBLE_PRECISION_REAL
506
      tmp1(:,:) = 0.0_rk8
507
508
#else
      tmp1(:,:) = 0.0_rk4
509
#endif
510
511
#endif /* REALCASE */

512
#if COMPLEXCASE == 1
513
#ifdef DOUBLE_PRECISION_COMPLEX
514
      tmp1(:,:) = 0.0_ck8
515
516
#else
      tmp1(:,:) = 0.0_ck4
517
#endif
518
#endif /* COMPLEXCASE */
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596


#if REALCASE == 1
      ! tmp1 = a**T
#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL
      call pdtran(na, na, 1.0_rk8, a, 1, 1, sc_desc, 0.0_rk8, tmp1, 1, 1, sc_desc)
#else

      call pstran(na, na, 1.0_rk4, a, 1, 1, sc_desc, 0.0_rk4, tmp1, 1, 1, sc_desc)
#endif

#else /* WITH_MPI */
      tmp1 = transpose(a)
#endif /* WITH_MPI */

      ! tmp2 = a * a**T
#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL
      call pdgemm("N","N", na, na, na, 1.0_rk8, a, 1, 1, sc_desc, tmp1, 1, 1, &
               sc_desc, 0.0_rk8, tmp2, 1, 1, sc_desc)
#else
      call psgemm("N","N", na, na, na, 1.0_rk4, a, 1, 1, sc_desc, tmp1, 1, 1, &
               sc_desc, 0.0_rk4, tmp2, 1, 1, sc_desc)

#endif

#else /* WITH_MPI */

#ifdef DOUBLE_PRECISION_REAL
      call dgemm("N","N", na, na, na, 1.0_rk8, a, na, tmp1, na, 0.0_rk8, tmp2, na)
#else
      call sgemm("N","N", na, na, na, 1.0_rk4, a, na, tmp1, na, 0.0_rk4, tmp2, na)
#endif
#endif /* WITH_MPI */


#endif /* REALCASE == 1 */

#if COMPLEXCASE == 1
      ! tmp1 = a**H
#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX
      call pztranc(na, na, CONE, a, 1, 1, sc_desc, CZERO, tmp1, 1, 1, sc_desc)
#else

      call pctranc(na, na, CONE, a, 1, 1, sc_desc, CZERO, tmp1, 1, 1, sc_desc)
#endif

#else /* WITH_MPI */
      tmp1 = transpose(conjg(a))
#endif /* WITH_MPI */

      ! tmp2 = a * a**T
#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX
      call pzgemm("N","N", na, na, na, CONE, a, 1, 1, sc_desc, tmp1, 1, 1, &
               sc_desc, CZERO, tmp2, 1, 1, sc_desc)
#else
      call pcgemm("N","N", na, na, na, CONE, a, 1, 1, sc_desc, tmp1, 1, 1, &
               sc_desc, CZERO, tmp2, 1, 1, sc_desc)

#endif

#else /* WITH_MPI */

#ifdef DOUBLE_PRECISION_COMPLEX
      call zgemm("N","N", na, na, na, CONE, a, na, tmp1, na, CZERO, tmp2, na)
#else
      call cgemm("N","N", na, na, na, CONE, a, na, tmp1, na, CZERO, tmp2, na)
#endif
#endif /* WITH_MPI */


#endif /* COMPLEXCASE == 1 */

      ! compare tmp2 with original matrix
      tmp2(:,:) = tmp2(:,:) - as(:,:)
597

598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
#if REALCASE == 1

#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL

      norm = pdlange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
#else
      norm = pslange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
#endif

#else /* WITH_MPI */
#ifdef DOUBLE_PRECISION_REAL
      norm = dlange("M", na, na, tmp2, na_rows, tmp1)
#else
      norm = slange("M", na, na, tmp2, na_rows, tmp1)

#endif
#endif /* WITH_MPI */


#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL
      call mpi_allreduce(norm,normmax,1,MPI_REAL8,MPI_MAX,MPI_COMM_WORLD,mpierr)
#else
      call mpi_allreduce(norm,normmax,1,MPI_REAL4,MPI_MAX,MPI_COMM_WORLD,mpierr)
#endif
#else /* WITH_MPI */
      normmax = norm
#endif /* WITH_MPI */


#endif /* REALCASE == 1 */

#if COMPLEXCASE == 1

#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX

      norm = pzlange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
#else
      norm = pclange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
#endif

#else /* WITH_MPI */
#ifdef DOUBLE_PRECISION_COMPLEX
      norm = zlange("M", na, na, tmp2, na_rows, tmp1)
#else
      norm = clange("M", na, na, tmp2, na_rows, tmp1)

#endif
#endif /* WITH_MPI */


#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX
      call mpi_allreduce(norm,normmax,1,MPI_REAL8,MPI_MAX,MPI_COMM_WORLD,mpierr)
#else
      call mpi_allreduce(norm,normmax,1,MPI_REAL4,MPI_MAX,MPI_COMM_WORLD,mpierr)
#endif
#else /* WITH_MPI */
      normmax = norm
#endif /* WITH_MPI */


662
#endif /* COMPLEXCASE == 1 */
663
664
665
666
667

      if (myid .eq. 0) then
        print *," Maximum error of result: ", normmax
      endif

668
#if REALCASE == 1
669
#ifdef DOUBLE_PRECISION_REAL
670
!      if (normmax .gt. 5e-12_rk8 .or. normmax .eq. 0.0_rk8) then
671
672
673
674
      if (normmax .gt. 5e-12_rk8) then
        status = 1
      endif
#else
675
676
!      if (normmax .gt. 5e-4_rk4 .or. normmax .eq. 0.0_rk4) then
      if (normmax .gt. 5e-4_rk4 ) then
677
678
679
        status = 1
      endif
#endif
680
#endif
681

682
683
#if COMPLEXCASE == 1
#ifdef DOUBLE_PRECISION_COMPLEX
684
685
!      if (normmax .gt. 5e-11_rk8 .or. normmax .eq. 0.0_rk8) then
      if (normmax .gt. 5e-11_rk8 ) then
686
687
688
        status = 1
      endif
#else
689
!      if (normmax .gt. 5e-3_rk4 .or. normmax .eq. 0.0_rk4) then
690
691
692
693
694
      if (normmax .gt. 5e-3_rk4) then
        status = 1
      endif
#endif
#endif
695
696
    end function

697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
    function check_correctness_hermitian_multiply_&
    &MATH_DATATYPE&
    &_&
    &PRECISION&
    & (na, a, b, c, na_rows, sc_desc, myid) result(status)
      implicit none
#include "../../src/general/precision_kinds.F90"
      integer(kind=ik)                 :: status
      integer(kind=ik), intent(in)     :: na, myid, na_rows
#if REALCASE == 1
      real(kind=rck), intent(in)       :: a(:,:), b(:,:), c(:,:)
      real(kind=rck), dimension(size(a,dim=1),size(a,dim=2)) :: tmp1, tmp2
      real(kind=rck)                   :: norm, normmax

#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL
      real(kind=rck)                   :: pdlange
#else
      real(kind=rck)                   :: pslange
#endif

#else /* WITH_MPI */

#ifdef DOUBLE_PRECISION_REAL
      real(kind=rck)                   :: dlange
#else
      real(kind=rck)                   :: slange
#endif

#endif /* WITH_MPI */

#endif /* REALCASE */

#if COMPLEXCASE == 1
      complex(kind=rck), intent(in)    :: a(:,:), b(:,:), c(:,:)
      complex(kind=rck), dimension(size(a,dim=1),size(a,dim=2)) :: tmp1, tmp2
      real(kind=rck)                :: norm, normmax
#ifdef DOUBLE_PRECISION_COMPLEX
      complex(kind=ck8), parameter   :: CZERO = (0.0_rk8,0.0_rk8), CONE = (1.0_rk8,0.0_rk8)
#else
      complex(kind=ck4), parameter   :: CZERO = (0.0_rk4,0.0_rk4), CONE = (1.0_rk4,0.0_rk8)
#endif

#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX
Andreas Marek's avatar
Andreas Marek committed
742
      real(kind=rck)                   :: pzlange
743
#else
Andreas Marek's avatar
Andreas Marek committed
744
      real(kind=rck)                   :: pclange
745
746
747
748
749
#endif

#else /* WITH_MPI */

#ifdef DOUBLE_PRECISION_COMPLEX
Andreas Marek's avatar
Andreas Marek committed
750
      real(kind=rck)                   :: zlange
751
#else
Andreas Marek's avatar
Andreas Marek committed
752
      real(kind=rck)                   :: clange
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
#endif

#endif /* WITH_MPI */
#endif /* COMPLEXCASE */

      integer(kind=ik)                 :: sc_desc(:)

      real(kind=rck)                   :: err, errmax

      integer :: mpierr

      status = 0

#if REALCASE == 1

#ifdef DOUBLE_PRECISION_REAL
      tmp1(:,:) = 0.0_rk8
#else
      tmp1(:,:) = 0.0_rk4
#endif

#endif /* REALCASE */
775

776
777
#if COMPLEXCASE == 1
#ifdef DOUBLE_PRECISION_COMPLEX
778
      tmp1(:,:) = (0.0_c_double, 0.0_c_double)
779
#else
780
      tmp1(:,:) = (0.0_c_float, 0.0_c_float)
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
#endif
#endif /* COMPLEXCASE */


#if REALCASE == 1
      ! tmp1 = a**T
#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL
      call pdtran(na, na, 1.0_rk8, a, 1, 1, sc_desc, 0.0_rk8, tmp1, 1, 1, sc_desc)
#else

      call pstran(na, na, 1.0_rk4, a, 1, 1, sc_desc, 0.0_rk4, tmp1, 1, 1, sc_desc)
#endif

#else /* WITH_MPI */
      tmp1 = transpose(a)
#endif /* WITH_MPI */

   ! tmp2 = tmp1 * b
#ifdef DOUBLE_PRECISION_REAL
#ifdef WITH_MPI
   call pdgemm("N","N", na, na, na, 1.0_rk8, tmp1, 1, 1, sc_desc, b, 1, 1, &
               sc_desc, 0.0_rk8, tmp2, 1, 1, sc_desc)
#else
   call dgemm("N","N", na, na, na, 1.0_rk8, tmp1, na, b, na, 0.0_rk8, tmp2, na)
#endif

#else /* DOUBLE_PRECISION_REAL */
#ifdef WITH_MPI
   call psgemm("N","N", na, na, na, 1.0_rk4, tmp1, 1, 1, sc_desc, b, 1, 1, &
               sc_desc, 0.0_rk4, tmp2, 1, 1, sc_desc)
#else
   call sgemm("N","N", na, na, na, 1.0_rk4, tmp1, na, b, na, 0.0_rk4, tmp2, na)
#endif

#endif /* DOUBLE_PRECISION_REAL */

#endif /* REALCASE == 1 */

#if COMPLEXCASE == 1
      ! tmp1 = a**H
#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX
      call pztranc(na, na, CONE, a, 1, 1, sc_desc, CZERO, tmp1, 1, 1, sc_desc)
#else

      call pctranc(na, na, CONE, a, 1, 1, sc_desc, CZERO, tmp1, 1, 1, sc_desc)
#endif

#else /* WITH_MPI */
      tmp1 = transpose(conjg(a))
#endif /* WITH_MPI */

   ! tmp2 = tmp1 * b
#ifdef DOUBLE_PRECISION_COMPLEX
#ifdef WITH_MPI
   call pzgemm("N","N", na, na, na, CONE, tmp1, 1, 1, sc_desc, b, 1, 1, &
               sc_desc, CZERO, tmp2, 1, 1, sc_desc)
#else
   call zgemm("N","N", na, na, na, CONE, tmp1, na, b, na, CZERO, tmp2, na)
#endif

#else /* DOUBLE_PRECISION_COMPLEX */

#ifdef WITH_MPI
   call pcgemm("N","N", na, na, na, CONE, tmp1, 1, 1, sc_desc, b, 1, 1, &
               sc_desc, CZERO, tmp2, 1, 1, sc_desc)
#else
   call cgemm("N","N", na, na, na, CONE, tmp1, na, b, na, CZERO, tmp2, na)
#endif
#endif /* DOUBLE_PRECISION_COMPLEX */

#endif /* COMPLEXCASE == 1 */

      ! compare tmp2 with c
      tmp2(:,:) = tmp2(:,:) - c(:,:)
#if REALCASE == 1

#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL

      norm = pdlange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
#else
      norm = pslange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
#endif

#else /* WITH_MPI */
#ifdef DOUBLE_PRECISION_REAL
      norm = dlange("M", na, na, tmp2, na_rows, tmp1)
#else
      norm = slange("M", na, na, tmp2, na_rows, tmp1)

#endif
#endif /* WITH_MPI */


#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_REAL
      call mpi_allreduce(norm,normmax,1,MPI_REAL8,MPI_MAX,MPI_COMM_WORLD,mpierr)
#else
      call mpi_allreduce(norm,normmax,1,MPI_REAL4,MPI_MAX,MPI_COMM_WORLD,mpierr)
#endif
#else /* WITH_MPI */
      normmax = norm
#endif /* WITH_MPI */


#endif /* REALCASE == 1 */

#if COMPLEXCASE == 1

#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX
      norm = pzlange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
#else
      norm = pclange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
#endif

#else /* WITH_MPI */
#ifdef DOUBLE_PRECISION_COMPLEX
      norm = zlange("M", na, na, tmp2, na_rows, tmp1)
#else
      norm = clange("M", na, na, tmp2, na_rows, tmp1)

#endif
#endif /* WITH_MPI */


#ifdef WITH_MPI
#ifdef DOUBLE_PRECISION_COMPLEX
      call mpi_allreduce(norm,normmax,1,MPI_REAL8,MPI_MAX,MPI_COMM_WORLD,mpierr)
#else
      call mpi_allreduce(norm,normmax,1,MPI_REAL4,MPI_MAX,MPI_COMM_WORLD,mpierr)
#endif
#else /* WITH_MPI */
      normmax = norm
#endif /* WITH_MPI */


#endif /* REALCASE == 1 */

      if (myid .eq. 0) then
        print *," Maximum error of result: ", normmax
      endif

#ifdef DOUBLE_PRECISION_REAL
927
      if (normmax .gt. 5e-11_rk8 ) then
928
929
930
        status = 1
      endif
#else
931
      if (normmax .gt. 5e-3_rk4 ) then
932
933
934
935
        status = 1
      endif
#endif

936
937
938
939
940
941
942
943
944
#ifdef DOUBLE_PRECISION_COMPLEX
      if (normmax .gt. 5e-11_rk8 ) then
        status = 1
      endif
#else
      if (normmax .gt. 5e-3_rk4 ) then
        status = 1
      endif
#endif
945
946
947
    end function


948
949


950
! vim: syntax=fortran