elpa_impl.F90 72.5 KB
Newer Older
1
2
3
!
!    Copyright 2017, L. Hüdepohl and A. Marek, MPCDF
!
Andreas Marek's avatar
Andreas Marek committed
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
44
45
46
47
!    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
!
!    This particular source code file contains additions, changes and
!    enhancements authored by Intel Corporation which is not part of
!    the ELPA consortium.
!
!    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.
!
48
#include "config-f90.h"
49

50
!> \brief Fortran module which provides the actual implementation of the API. Do not use directly! Use the module "elpa"
51
module elpa_impl
52
  use elpa_abstract_impl
53
  use, intrinsic :: iso_c_binding
54
  implicit none
55

56
57
  private
  public :: elpa_impl_allocate
58

59
!> \brief Definition of the extended elpa_impl_t type
60
  type, extends(elpa_abstract_impl_t) :: elpa_impl_t
Andreas Marek's avatar
Andreas Marek committed
61
   private
62

63
   !> \brief methods available with the elpa_impl_t type
64
   contains
65
     !> \brief the puplic methods
66
     ! con-/destructor
67
68
     procedure, public :: setup => elpa_setup                   !< a setup method: implemented in elpa_setup
     procedure, public :: destroy => elpa_destroy               !< a destroy method: implemented in elpa_destroy
69

70
     ! KV store
71
72
73
74
     procedure, public :: is_set => elpa_is_set                 !< a method to check whether a key/value pair has been set : implemented
                                                                !< in elpa_is_set
     procedure, public :: can_set => elpa_can_set               !< a method to check whether a key/value pair can be set : implemented
                                                                !< in elpa_can_set
75

76
77
78
79
80
81

     ! timer
     procedure, public :: get_time => elpa_get_time
     procedure, public :: print_times => elpa_print_times


82
     !> \brief the private methods
83

84
     procedure, private :: elpa_eigenvectors_d                  !< private methods to implement the solve step for real/complex
85
                                                                !< double/single matrices
86
87
88
     procedure, private :: elpa_eigenvectors_f
     procedure, private :: elpa_eigenvectors_dc
     procedure, private :: elpa_eigenvectors_fc
89

90
91
     procedure, private :: elpa_hermitian_multiply_d            !< private methods to implement a "hermitian" multiplication of matrices a and b
     procedure, private :: elpa_hermitian_multiply_f            !< for real valued matrices:   a**T * b
Andreas Marek's avatar
Andreas Marek committed
92
     procedure, private :: elpa_hermitian_multiply_dc           !< for complex valued matrices:   a**H * b
93
     procedure, private :: elpa_hermitian_multiply_fc
94

Andreas Marek's avatar
Andreas Marek committed
95
     procedure, private :: elpa_cholesky_d                      !< private methods to implement the cholesky factorisation of
96
                                                                !< real/complex double/single matrices
97
98
99
     procedure, private :: elpa_cholesky_f
     procedure, private :: elpa_cholesky_dc
     procedure, private :: elpa_cholesky_fc
100

Andreas Marek's avatar
Andreas Marek committed
101
     procedure, private :: elpa_invert_trm_d                    !< private methods to implement the inversion of a triangular
102
                                                                !< real/complex double/single matrix
103
104
105
     procedure, private :: elpa_invert_trm_f
     procedure, private :: elpa_invert_trm_dc
     procedure, private :: elpa_invert_trm_fc
106

Andreas Marek's avatar
Andreas Marek committed
107
108
     procedure, private :: elpa_solve_tridi_d                   !< private methods to implement the solve step for a real valued
     procedure, private :: elpa_solve_tridi_f                   !< double/single tridiagonal matrix
109

110
     procedure, private :: associate_int => elpa_associate_int  !< private method to set some pointers
111

112
  end type elpa_impl_t
113

114
  !> \brief the implementation of the private methods
115
  contains
116
117
118
119
    !> \brief function to allocate an ELPA object
    !> Parameters
    !> \param   error      integer, optional to get an error code
    !> \result  obj        class(elpa_impl_t) allocated ELPA object
120
    function elpa_impl_allocate(error) result(obj)
Andreas Marek's avatar
Andreas Marek committed
121
122
      use precision
      use elpa_utilities, only : error_unit
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
123
      use elpa_generated_fortran_interfaces
Andreas Marek's avatar
Andreas Marek committed
124

125
126
127
128
      type(elpa_impl_t), pointer   :: obj
      integer, optional            :: error

      allocate(obj)
Andreas Marek's avatar
Andreas Marek committed
129

Andreas Marek's avatar
Andreas Marek committed
130
      ! check whether init has ever been called
131
      if ( elpa_initialized() .ne. ELPA_OK) then
132
        write(error_unit, *) "elpa_allocate(): you must call elpa_init() once before creating instances of ELPA"
133
134
        if(present(error)) then
          error = ELPA_ERROR
135
        endif
Andreas Marek's avatar
Andreas Marek committed
136
137
        return
      endif
Andreas Marek's avatar
Andreas Marek committed
138

139
      obj%index = elpa_index_instance_c()
140
141

      ! Associate some important integer pointers for convenience
142
143
144
145
146
147
148
149
      obj%na => obj%associate_int("na")
      obj%nev => obj%associate_int("nev")
      obj%local_nrows => obj%associate_int("local_nrows")
      obj%local_ncols => obj%associate_int("local_ncols")
      obj%nblk => obj%associate_int("nblk")

      if(present(error)) then
        error = ELPA_OK
150
151
      endif
    end function
Andreas Marek's avatar
Andreas Marek committed
152

153
154
155
156
157
    !c> /*! \brief C interface for the implementation of the elpa_allocate method
    !c> *
    !c> *  \param  none
    !c> *  \result elpa_t handle
    !c> */
158
    !c> elpa_t elpa_allocate();
159
    function elpa_impl_allocate_c(error) result(ptr) bind(C, name="elpa_allocate")
160
161
162
163
164
165
166
167
      integer(kind=c_int) :: error
      type(c_ptr) :: ptr
      type(elpa_impl_t), pointer :: obj

      obj => elpa_impl_allocate(error)
      ptr = c_loc(obj)
    end function

168
169
170
171
172
    !c> /*! \brief C interface for the implementation of the elpa_deallocate method
    !c> *
    !c> *  \param  elpa_t  handle of ELPA object to be deallocated
    !c> *  \result void
    !c> */
173
    !c> void elpa_deallocate(elpa_t handle);
174
    subroutine elpa_impl_deallocate_c(handle) bind(C, name="elpa_deallocate")
175
176
177
178
179
180
181
182
183
      type(c_ptr), value :: handle
      type(elpa_impl_t), pointer :: self

      call c_f_pointer(handle, self)
      call self%destroy()
      deallocate(self)
    end subroutine


184
185
186
187
    !> \brief function to setup an ELPA object and to store the MPI communicators internally
    !> Parameters
    !> \param   self       class(elpa_impl_t), the allocated ELPA object
    !> \result  error      integer, the error code
188
    function elpa_setup(self) result(error)
189
      use elpa1_impl, only : elpa_get_communicators_impl
190
      class(elpa_impl_t), intent(inout) :: self
191
192
193
      integer                           :: error
      integer                           :: mpi_comm_parent, mpi_comm_rows, mpi_comm_cols, &
                                           mpierr, process_row, process_col, timings
194

195
#ifdef WITH_MPI
196
197
198
199
      error = ELPA_ERROR
      if (self%is_set("mpi_comm_parent") == 1 .and. &
          self%is_set("process_row") == 1 .and. &
          self%is_set("process_col") == 1) then
200

201
202
203
        call self%get("mpi_comm_parent", mpi_comm_parent)
        call self%get("process_row", process_row)
        call self%get("process_col", process_col)
204
        mpierr = elpa_get_communicators_impl(&
205
206
207
                        mpi_comm_parent, &
                        process_row, &
                        process_col, &
208
209
                        mpi_comm_rows, &
                        mpi_comm_cols)
210

211
212
213
        call self%set("mpi_comm_rows", mpi_comm_rows)
        call self%set("mpi_comm_cols", mpi_comm_cols)

214
        error = ELPA_OK
215
      endif
216

217
218
      if (self%is_set("mpi_comm_rows") == 1 .and. self%is_set("mpi_comm_cols") == 1) then
        error = ELPA_OK
219
      endif
220
221
222
#else
      error = ELPA_OK
#endif
223

224
#ifdef HAVE_DETAILED_TIMINGS
225
226
      call self%get("timings",timings)
      if (timings == 1) then
227
228
        call self%timer%enable()
      endif
229
#endif
230

231
    end function
232

233
234
235
236
237
238
    !c> /*! \brief C interface for the implementation of the elpa_setup method
    !c> *
    !c> *  \param  elpa_t  handle of the ELPA object which describes the problem to
    !c> *                  be set up
    !c> *  \result int     error code, which can be queried with elpa_strerr
    !c> */
239
    !c> int elpa_setup(elpa_t handle);
240
    function elpa_setup_c(handle) result(error) bind(C, name="elpa_setup")
241
242
243
244
245
246
247
248
249
      type(c_ptr), intent(in), value :: handle
      type(elpa_impl_t), pointer :: self
      integer(kind=c_int) :: error

      call c_f_pointer(handle, self)
      error = self%setup()
    end function


250
251
252
253
254
255
256
257
258
    !c> /*! \brief C interface for the implementation of the elpa_set_integer method
    !c> *  This method is available to the user as C generic elpa_set method
    !c> *
    !c> *  \param  handle  handle of the ELPA object for which a key/value pair should be set
    !c> *  \param  name    the name of the key
    !c> *  \param  value   the value to be set for the key
    !c> *  \param  error   on return the error code, which can be queried with elpa_strerr()
    !c> *  \result void
    !c> */
259
    !c> void elpa_set_integer(elpa_t handle, const char *name, int value, int *error);
260
    subroutine elpa_set_integer_c(handle, name_p, value, error) bind(C, name="elpa_set_integer")
261
262
263
264
265
266
267
268
269
270
271
272
273
      type(c_ptr), intent(in), value :: handle
      type(elpa_impl_t), pointer :: self
      type(c_ptr), intent(in), value :: name_p
      character(len=elpa_strlen_c(name_p)), pointer :: name
      integer(kind=c_int), intent(in), value :: value
      integer(kind=c_int), optional, intent(in) :: error

      call c_f_pointer(handle, self)
      call c_f_pointer(name_p, name)
      call elpa_set_integer(self, name, value, error)
    end subroutine


274
275
276
277
278
279
280
281
282
    !c> /*! \brief C interface for the implementation of the elpa_get_integer method
    !c> *  This method is available to the user as C generic elpa_get method
    !c> *
    !c> *  \param  handle  handle of the ELPA object for which a key/value pair should be queried
    !c> *  \param  name    the name of the key
    !c> *  \param  value   the value to be obtain for the key
    !c> *  \param  error   on return the error code, which can be queried with elpa_strerr()
    !c> *  \result void
    !c> */
283
284
    !c> void elpa_get_integer(elpa_t handle, const char *name, int *value, int *error);
    subroutine elpa_get_integer_c(handle, name_p, value, error) bind(C, name="elpa_get_integer")
Andreas Marek's avatar
Andreas Marek committed
285
286
287
288
289
290
291
292
293
      type(c_ptr), intent(in), value :: handle
      type(elpa_impl_t), pointer :: self
      type(c_ptr), intent(in), value :: name_p
      character(len=elpa_strlen_c(name_p)), pointer :: name
      integer(kind=c_int)  :: value
      integer(kind=c_int), optional, intent(inout) :: error

      call c_f_pointer(handle, self)
      call c_f_pointer(name_p, name)
294
295
      call elpa_get_integer(self, name, value, error)
    end subroutine
Andreas Marek's avatar
Andreas Marek committed
296
297


298
299
300
301
302
    !> \brief function to check whether a key/value pair is set
    !> Parameters
    !> \param   self       class(elpa_impl_t) the allocated ELPA object
    !> \param   name       string, the key
    !> \result  state      integer, the state of the key/value pair
303
    function elpa_is_set(self, name) result(state)
304
305
      use iso_c_binding
      use elpa_generated_fortran_interfaces
306
      class(elpa_impl_t)       :: self
307
      character(*), intent(in) :: name
308
      integer                  :: state
309

310
      state = elpa_index_value_is_set_c(self%index, name // c_null_char)
311
312
    end function

313
314
315
316
317
318
    !> \brief function to check whether a key/value pair can be set
    !> Parameters
    !> \param   self       class(elpa_impl_t) the allocated ELPA object
    !> \param   name       string, the key
    !> \param   value      integer, value
    !> \result  error      integer, error code
319
320
321
322
323
324
325
326
327
328
329
330
331
    function elpa_can_set(self, name, value) result(error)
      use iso_c_binding
      use elpa_generated_fortran_interfaces
      class(elpa_impl_t)       :: self
      character(*), intent(in) :: name
      integer(kind=c_int), intent(in) :: value
      integer                  :: error

      error = elpa_index_int_is_valid_c(self%index, name // c_null_char, value)
    end function


    function elpa_value_to_string(self, option_name, error) result(string)
332
333
334
      use elpa_generated_fortran_interfaces
      class(elpa_impl_t), intent(in) :: self
      character(kind=c_char, len=*), intent(in) :: option_name
335
336
337
338
      type(c_ptr) :: ptr
      integer, intent(out), optional :: error
      integer :: val, actual_error
      character(kind=c_char, len=elpa_index_int_value_to_strlen_c(self%index, option_name // C_NULL_CHAR)), pointer :: string
339

340
341
      nullify(string)

342
      call self%get(option_name, val, actual_error)
343
344
345
346
347
      if (actual_error /= ELPA_OK) then
        if (present(error)) then
          error = actual_error
        endif
        return
348
349
      endif

350
351
352
353
      actual_error = elpa_int_value_to_string_c(option_name // C_NULL_CHAR, val, ptr)
      if (c_associated(ptr)) then
        call c_f_pointer(ptr, string)
      endif
354

355
356
357
358
      if (present(error)) then
        error = actual_error
      endif
    end function
359

Andreas Marek's avatar
Andreas Marek committed
360

361
362
363
364
365
366
367
368
369
    !c> /*! \brief C interface for the implementation of the elpa_set_double method
    !c> *  This method is available to the user as C generic elpa_set method
    !c> *
    !c> *  \param  handle  handle of the ELPA object for which a key/value pair should be set
    !c> *  \param  name    the name of the key
    !c> *  \param  value   the value to be set for the key
    !c> *  \param  error   on return the error code, which can be queried with elpa_strerr()
    !c> *  \result void
    !c> */
370
    !c> void elpa_set_double(elpa_t handle, const char *name, double value, int *error);
371
    subroutine elpa_set_double_c(handle, name_p, value, error) bind(C, name="elpa_set_double")
372
373
374
375
376
377
378
379
380
381
382
383
      type(c_ptr), intent(in), value :: handle
      type(elpa_impl_t), pointer :: self
      type(c_ptr), intent(in), value :: name_p
      character(len=elpa_strlen_c(name_p)), pointer :: name
      real(kind=c_double), intent(in), value :: value
      integer(kind=c_int), optional, intent(in) :: error

      call c_f_pointer(handle, self)
      call c_f_pointer(name_p, name)
      call elpa_set_double(self, name, value, error)
    end subroutine

384

385
    !c> /*! \brief C interface for the implementation of the elpa_get_double method
386
387
388
389
390
391
392
393
    !c> *  This method is available to the user as C generic elpa_get method
    !c> *
    !c> *  \param  handle  handle of the ELPA object for which a key/value pair should be queried
    !c> *  \param  name    the name of the key
    !c> *  \param  value   the value to be obtain for the key
    !c> *  \param  error   on return the error code, which can be queried with elpa_strerr()
    !c> *  \result void
    !c> */
394
395
    !c> void elpa_get_double(elpa_t handle, const char *name, double *value, int *error);
    subroutine elpa_get_double_c(handle, name_p, value, error) bind(C, name="elpa_get_double")
Andreas Marek's avatar
Andreas Marek committed
396
397
398
399
400
401
402
403
404
      type(c_ptr), intent(in), value :: handle
      type(elpa_impl_t), pointer :: self
      type(c_ptr), intent(in), value :: name_p
      character(len=elpa_strlen_c(name_p)), pointer :: name
      real(kind=c_double)  :: value
      integer(kind=c_int), optional, intent(inout) :: error

      call c_f_pointer(handle, self)
      call c_f_pointer(name_p, name)
405
406
      call elpa_get_double(self, name, value, error)
    end subroutine
Andreas Marek's avatar
Andreas Marek committed
407
408


409
    function elpa_associate_int(self, name) result(value)
Andreas Marek's avatar
Andreas Marek committed
410
      use iso_c_binding
411
      use elpa_generated_fortran_interfaces
412
413
      use elpa_utilities, only : error_unit
      class(elpa_impl_t)             :: self
414
415
      character(*), intent(in)       :: name
      integer(kind=c_int), pointer   :: value
Andreas Marek's avatar
Andreas Marek committed
416

417
418
      type(c_ptr)                    :: value_p

419
      value_p = elpa_index_get_int_loc_c(self%index, name // c_null_char)
420
421
422
      if (.not. c_associated(value_p)) then
        write(error_unit, '(a,a,a)') "ELPA: Warning, received NULL pointer for entry '", name, "'"
      endif
423
424
      call c_f_pointer(value_p, value)
    end function
Andreas Marek's avatar
Andreas Marek committed
425

426

427
428
429
430
431
432
    function elpa_get_time(self, name1, name2, name3, name4, name5, name6) result(s)
      class(elpa_impl_t), intent(in) :: self
      ! this is clunky, but what can you do..
      character(len=*), intent(in), optional :: name1, name2, name3, name4, name5, name6
      real(kind=c_double) :: s

433
#ifdef HAVE_DETAILED_TIMINGS
434
      s = self%timer%get(name1, name2, name3, name4, name5, name6)
435
436
437
#else
      s = -1.0
#endif
438
439
440
441
442
    end function


    subroutine elpa_print_times(self)
      class(elpa_impl_t), intent(in) :: self
443
#ifdef HAVE_DETAILED_TIMINGS
444
      call self%timer%print()
445
#endif
446
447
    end subroutine

448
    !>  \brief elpa_eigenvectors_d: class method to solve the eigenvalue problem for double real matrices
Andreas Marek's avatar
Andreas Marek committed
449
    !>
450
451
    !>  The dimensions of the matrix a (locally ditributed and global), the block-cyclic distribution
    !>  blocksize, the number of eigenvectors
Andreas Marek's avatar
Andreas Marek committed
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
    !>  to be computed and the MPI communicators are already known to the object and MUST be set BEFORE
    !>  with the class method "setup"
    !>
    !>  It is possible to change the behaviour of the method by setting tunable parameters with the
    !>  class method "set"
    !>
    !>  Parameters
    !>
    !>  \param a                                    Distributed matrix for which eigenvalues are to be computed.
    !>                                              Distribution is like in Scalapack.
    !>                                              The full matrix must be set (not only one half like in scalapack).
    !>                                              Destroyed on exit (upper and lower half).
    !>
    !>  \param ev                                   On output: eigenvalues of a, every processor gets the complete set
    !>
    !>  \param q                                    On output: Eigenvectors of a
    !>                                              Distribution is like in Scalapack.
    !>                                              Must be always dimensioned to the full size (corresponding to (na,na))
    !>                                              even if only a part of the eigenvalues is needed.
    !>
    !>  \param error                                integer, optional: returns an error code, which can be queried with elpa_strerr
473
    subroutine elpa_eigenvectors_d(self, a, ev, q, error)
474
475
      use elpa2_impl
      use elpa1_impl
476
      use elpa_utilities, only : error_unit
Andreas Marek's avatar
Andreas Marek committed
477
      use iso_c_binding
478
      class(elpa_impl_t)  :: self
Andreas Marek's avatar
Andreas Marek committed
479

480
481
482
#ifdef USE_ASSUMED_SIZE
      real(kind=c_double) :: a(self%local_nrows, *), q(self%local_nrows, *)
#else
483
      real(kind=c_double) :: a(self%local_nrows, self%local_ncols), q(self%local_nrows, self%local_ncols)
484
#endif
485
      real(kind=c_double) :: ev(self%na)
486

487
      integer, optional   :: error
488
      integer(kind=c_int) :: solver
489
      logical             :: success_l
490

491

492
493
      call self%get("solver", solver)
      if (solver .eq. ELPA_SOLVER_1STAGE) then
494
        success_l = elpa_solve_evp_real_1stage_double_impl(self, a, ev, q)
495

496
      else if (solver .eq. ELPA_SOLVER_2STAGE) then
497
        success_l = elpa_solve_evp_real_2stage_double_impl(self, a, ev, q)
498
499
500
501
      else
        print *,"unknown solver"
        stop
      endif
502

503
      if (present(error)) then
504
        if (success_l) then
505
          error = ELPA_OK
506
        else
507
          error = ELPA_ERROR
508
509
510
511
512
513
        endif
      else if (.not. success_l) then
        write(error_unit,'(a)') "ELPA: Error in solve() and you did not check for errors!"
      endif
    end subroutine

514
515
    !c> void elpa_eigenvectors_d(elpa_t handle, double *a, double *ev, double *q, int *error);
    subroutine elpa_eigenvectors_d_c(handle, a_p, ev_p, q_p, error) bind(C, name="elpa_eigenvectors_d")
516
517
518
519
520
521
522
523
524
525
526
      type(c_ptr), intent(in), value :: handle, a_p, ev_p, q_p
      integer(kind=c_int), optional, intent(in) :: error

      real(kind=c_double), pointer :: a(:, :), q(:, :), ev(:)
      type(elpa_impl_t), pointer  :: self

      call c_f_pointer(handle, self)
      call c_f_pointer(a_p, a, [self%local_nrows, self%local_ncols])
      call c_f_pointer(ev_p, ev, [self%na])
      call c_f_pointer(q_p, q, [self%local_nrows, self%local_ncols])

527
      call elpa_eigenvectors_d(self, a, ev, q, error)
528
529
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
530

531
    !>  \brief elpa_eigenvectors_f: class method to solve the eigenvalue problem for float real matrices
Andreas Marek's avatar
Andreas Marek committed
532
    !>
533
534
    !>  The dimensions of the matrix a (locally ditributed and global), the block-cyclic distribution
    !>  blocksize, the number of eigenvectors
Andreas Marek's avatar
Andreas Marek committed
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
    !>  to be computed and the MPI communicators are already known to the object and MUST be set BEFORE
    !>  with the class method "setup"
    !>
    !>  It is possible to change the behaviour of the method by setting tunable parameters with the
    !>  class method "set"
    !>
    !>  Parameters
    !>
    !>  \param a                                    Distributed matrix for which eigenvalues are to be computed.
    !>                                              Distribution is like in Scalapack.
    !>                                              The full matrix must be set (not only one half like in scalapack).
    !>                                              Destroyed on exit (upper and lower half).
    !>
    !>  \param ev                                   On output: eigenvalues of a, every processor gets the complete set
    !>
    !>  \param q                                    On output: Eigenvectors of a
    !>                                              Distribution is like in Scalapack.
    !>                                              Must be always dimensioned to the full size (corresponding to (na,na))
    !>                                              even if only a part of the eigenvalues is needed.
    !>
    !>  \param error                                integer, optional: returns an error code, which can be queried with elpa_strerr
556
    subroutine elpa_eigenvectors_f(self, a, ev, q, error)
557
558
      use elpa2_impl
      use elpa1_impl
559
560
      use elpa_utilities, only : error_unit
      use iso_c_binding
561
      class(elpa_impl_t)  :: self
562
563
564
#ifdef USE_ASSUMED_SIZE
      real(kind=c_float)  :: a(self%local_nrows, *), q(self%local_nrows, *)
#else
565
      real(kind=c_float)  :: a(self%local_nrows, self%local_ncols), q(self%local_nrows, self%local_ncols)
566
#endif
567
      real(kind=c_float)  :: ev(self%na)
568

569
      integer, optional   :: error
570
      integer(kind=c_int) :: solver
571
      logical             :: success_l
572

573
#ifdef WANT_SINGLE_PRECISION_REAL
574

575
576
      call self%get("solver",solver)
      if (solver .eq. ELPA_SOLVER_1STAGE) then
577
        success_l = elpa_solve_evp_real_1stage_single_impl(self, a, ev, q)
578

579
      else if (solver .eq. ELPA_SOLVER_2STAGE) then
580
        success_l = elpa_solve_evp_real_2stage_single_impl(self, a, ev, q)
581
582
583
584
      else
        print *,"unknown solver"
        stop
      endif
585

586
      if (present(error)) then
587
        if (success_l) then
588
          error = ELPA_OK
589
        else
590
          error = ELPA_ERROR
591
592
593
594
595
        endif
      else if (.not. success_l) then
        write(error_unit,'(a)') "ELPA: Error in solve() and you did not check for errors!"
      endif
#else
596
      print *,"This installation of the ELPA library has not been build with single-precision support"
597
      error = ELPA_ERROR
598
599
600
#endif
    end subroutine

601

602
603
    !c> void elpa_eigenvectors_f(elpa_t handle, float *a, float *ev, float *q, int *error);
    subroutine elpa_eigenvectors_f_c(handle, a_p, ev_p, q_p, error) bind(C, name="elpa_eigenvectors_f")
604
605
606
607
608
609
610
611
612
613
614
      type(c_ptr), intent(in), value :: handle, a_p, ev_p, q_p
      integer(kind=c_int), optional, intent(in) :: error

      real(kind=c_float), pointer :: a(:, :), q(:, :), ev(:)
      type(elpa_impl_t), pointer  :: self

      call c_f_pointer(handle, self)
      call c_f_pointer(a_p, a, [self%local_nrows, self%local_ncols])
      call c_f_pointer(ev_p, ev, [self%na])
      call c_f_pointer(q_p, q, [self%local_nrows, self%local_ncols])

615
      call elpa_eigenvectors_f(self, a, ev, q, error)
616
617
618
    end subroutine


619
    !>  \brief elpa_eigenvectors_dc: class method to solve the eigenvalue problem for double complex matrices
Andreas Marek's avatar
Andreas Marek committed
620
    !>
621
622
    !>  The dimensions of the matrix a (locally ditributed and global), the block-cyclic distribution
    !>  blocksize, the number of eigenvectors
Andreas Marek's avatar
Andreas Marek committed
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
    !>  to be computed and the MPI communicators are already known to the object and MUST be set BEFORE
    !>  with the class method "setup"
    !>
    !>  It is possible to change the behaviour of the method by setting tunable parameters with the
    !>  class method "set"
    !>
    !>  Parameters
    !>
    !>  \param a                                    Distributed matrix for which eigenvalues are to be computed.
    !>                                              Distribution is like in Scalapack.
    !>                                              The full matrix must be set (not only one half like in scalapack).
    !>                                              Destroyed on exit (upper and lower half).
    !>
    !>  \param ev                                   On output: eigenvalues of a, every processor gets the complete set
    !>
    !>  \param q                                    On output: Eigenvectors of a
    !>                                              Distribution is like in Scalapack.
    !>                                              Must be always dimensioned to the full size (corresponding to (na,na))
    !>                                              even if only a part of the eigenvalues is needed.
    !>
    !>  \param error                                integer, optional: returns an error code, which can be queried with elpa_strerr
644
    subroutine elpa_eigenvectors_dc(self, a, ev, q, error)
645
646
      use elpa2_impl
      use elpa1_impl
647
648
      use elpa_utilities, only : error_unit
      use iso_c_binding
649
      class(elpa_impl_t)             :: self
650

651
652
653
#ifdef USE_ASSUMED_SIZE
      complex(kind=c_double_complex) :: a(self%local_nrows, *), q(self%local_nrows, *)
#else
654
      complex(kind=c_double_complex) :: a(self%local_nrows, self%local_ncols), q(self%local_nrows, self%local_ncols)
655
#endif
656
      real(kind=c_double)            :: ev(self%na)
657

658
      integer, optional              :: error
659
      integer(kind=c_int)            :: solver
660
      logical                        :: success_l
661

662
663
      call self%get("solver", solver)
      if (solver .eq. ELPA_SOLVER_1STAGE) then
664
        success_l = elpa_solve_evp_complex_1stage_double_impl(self, a, ev, q)
665

666
      else if (solver .eq. ELPA_SOLVER_2STAGE) then
667
        success_l = elpa_solve_evp_complex_2stage_double_impl(self,  a, ev, q)
668
669
670
671
      else
        print *,"unknown solver"
        stop
      endif
672

673
      if (present(error)) then
674
        if (success_l) then
675
          error = ELPA_OK
676
        else
677
          error = ELPA_ERROR
678
679
680
681
682
683
684
        endif
      else if (.not. success_l) then
        write(error_unit,'(a)') "ELPA: Error in solve() and you did not check for errors!"
      endif
    end subroutine


685
686
    !c> void elpa_eigenvectors_dc(elpa_t handle, double complex *a, double *ev, double complex *q, int *error);
    subroutine elpa_eigenvectors_dc_c(handle, a_p, ev_p, q_p, error) bind(C, name="elpa_eigenvectors_dc")
687
688
689
690
691
692
693
694
695
696
697
698
      type(c_ptr), intent(in), value :: handle, a_p, ev_p, q_p
      integer(kind=c_int), optional, intent(in) :: error

      complex(kind=c_double_complex), pointer :: a(:, :), q(:, :)
      real(kind=c_double), pointer :: ev(:)
      type(elpa_impl_t), pointer  :: self

      call c_f_pointer(handle, self)
      call c_f_pointer(a_p, a, [self%local_nrows, self%local_ncols])
      call c_f_pointer(ev_p, ev, [self%na])
      call c_f_pointer(q_p, q, [self%local_nrows, self%local_ncols])

699
      call elpa_eigenvectors_dc(self, a, ev, q, error)
700
701
702
    end subroutine


703
    !>  \brief elpa_eigenvectors_fc: class method to solve the eigenvalue problem for float complex matrices
Andreas Marek's avatar
Andreas Marek committed
704
    !>
705
706
    !>  The dimensions of the matrix a (locally ditributed and global), the block-cyclic distribution
    !>  blocksize, the number of eigenvectors
Andreas Marek's avatar
Andreas Marek committed
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
    !>  to be computed and the MPI communicators are already known to the object and MUST be set BEFORE
    !>  with the class method "setup"
    !>
    !>  It is possible to change the behaviour of the method by setting tunable parameters with the
    !>  class method "set"
    !>
    !>  Parameters
    !>
    !>  \param a                                    Distributed matrix for which eigenvalues are to be computed.
    !>                                              Distribution is like in Scalapack.
    !>                                              The full matrix must be set (not only one half like in scalapack).
    !>                                              Destroyed on exit (upper and lower half).
    !>
    !>  \param ev                                   On output: eigenvalues of a, every processor gets the complete set
    !>
    !>  \param q                                    On output: Eigenvectors of a
    !>                                              Distribution is like in Scalapack.
    !>                                              Must be always dimensioned to the full size (corresponding to (na,na))
    !>                                              even if only a part of the eigenvalues is needed.
    !>
    !>  \param error                                integer, optional: returns an error code, which can be queried with elpa_strerr
728
    subroutine elpa_eigenvectors_fc(self, a, ev, q, error)
729
730
      use elpa2_impl
      use elpa1_impl
731
732
733
      use elpa_utilities, only : error_unit

      use iso_c_binding
734
      class(elpa_impl_t)            :: self
735
#ifdef USE_ASSUMED_SIZE
Andreas Marek's avatar
Andreas Marek committed
736
      complex(kind=c_float_complex) :: a(self%local_nrows, *), q(self%local_nrows, *)
737
#else
Andreas Marek's avatar
Andreas Marek committed
738
      complex(kind=c_float_complex) :: a(self%local_nrows, self%local_ncols), q(self%local_nrows, self%local_ncols)
739
#endif
Andreas Marek's avatar
Andreas Marek committed
740
      real(kind=c_float)            :: ev(self%na)
741

742
      integer, optional             :: error
743
      integer(kind=c_int)           :: solver
744
      logical                       :: success_l
745
746

#ifdef WANT_SINGLE_PRECISION_COMPLEX
747

748
749
      call self%get("solver", solver)
      if (solver .eq. ELPA_SOLVER_1STAGE) then
750
        success_l = elpa_solve_evp_complex_1stage_single_impl(self, a, ev, q)
751

752
      else if (solver .eq. ELPA_SOLVER_2STAGE) then
753
        success_l = elpa_solve_evp_complex_2stage_single_impl(self,  a, ev, q)
754
755
756
757
      else
        print *,"unknown solver"
        stop
      endif
758

759
      if (present(error)) then
760
        if (success_l) then
761
          error = ELPA_OK
762
        else
763
          error = ELPA_ERROR
764
765
766
767
768
        endif
      else if (.not. success_l) then
        write(error_unit,'(a)') "ELPA: Error in solve() and you did not check for errors!"
      endif
#else
769
      print *,"This installation of the ELPA library has not been build with single-precision support"
770
      error = ELPA_ERROR
771
772
773
#endif
    end subroutine

774

775
776
    !c> void elpa_eigenvectors_fc(elpa_t handle, float complex *a, float *ev, float complex *q, int *error);
    subroutine elpa_eigenvectors_fc_c(handle, a_p, ev_p, q_p, error) bind(C, name="elpa_eigenvectors_fc")
777
778
779
780
781
782
783
784
785
786
787
788
      type(c_ptr), intent(in), value :: handle, a_p, ev_p, q_p
      integer(kind=c_int), optional, intent(in) :: error

      complex(kind=c_float_complex), pointer :: a(:, :), q(:, :)
      real(kind=c_float), pointer :: ev(:)
      type(elpa_impl_t), pointer  :: self

      call c_f_pointer(handle, self)
      call c_f_pointer(a_p, a, [self%local_nrows, self%local_ncols])
      call c_f_pointer(ev_p, ev, [self%na])
      call c_f_pointer(q_p, q, [self%local_nrows, self%local_ncols])

789
      call elpa_eigenvectors_fc(self, a, ev, q, error)
790
791
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
792
    !> \brief  elpa_hermitian_multiply_d: class method to perform C : = A**T * B for double real matrices
793
794
795
    !>         where   A is a square matrix (self%na,self%na) which is optionally upper or lower triangular
    !>                 B is a (self%na,ncb) matrix
    !>                 C is a (self%na,ncb) matrix where optionally only the upper or lower
Andreas Marek's avatar
Andreas Marek committed
796
797
798
799
800
801
802
    !>                   triangle may be computed
    !>
    !> the MPI commicators and the block-cyclic distribution block size are already known to the type.
    !> Thus the class method "setup" must be called BEFORE this method is used
    !>
    !> \details
    !>
803
    !> \param  self                 class(elpa_t), the ELPA object
Andreas Marek's avatar
Andreas Marek committed
804
805
806
807
808
809
810
811
812
813
814
815
816
817
    !> \param  uplo_a               'U' if A is upper triangular
    !>                              'L' if A is lower triangular
    !>                              anything else if A is a full matrix
    !>                              Please note: This pertains to the original A (as set in the calling program)
    !>                                           whereas the transpose of A is used for calculations
    !>                              If uplo_a is 'U' or 'L', the other triangle is not used at all,
    !>                              i.e. it may contain arbitrary numbers
    !> \param uplo_c                'U' if only the upper diagonal part of C is needed
    !>                              'L' if only the upper diagonal part of C is needed
    !>                              anything else if the full matrix C is needed
    !>                              Please note: Even when uplo_c is 'U' or 'L', the other triangle may be
    !>                                            written to a certain extent, i.e. one shouldn't rely on the content there!
    !> \param ncb                   Number of columns  of global matrices B and C
    !> \param a                     matrix a
818
819
    !> \param local_nrows           number of rows of local (sub) matrix a, set with class method set("local_nrows",value)
    !> \param local_ncols           number of columns of local (sub) matrix a, set with class method set("local_ncols",value)
Andreas Marek's avatar
Andreas Marek committed
820
821
822
823
824
825
826
    !> \param b                     matrix b
    !> \param nrows_b               number of rows of local (sub) matrix b
    !> \param ncols_b               number of columns of local (sub) matrix b
    !> \param c                     matrix c
    !> \param nrows_c               number of rows of local (sub) matrix c
    !> \param ncols_c               number of columns of local (sub) matrix c
    !> \param error                 optional argument, error code which can be queried with elpa_strerr
827
    subroutine elpa_hermitian_multiply_d (self,uplo_a, uplo_c, ncb, a, b, nrows_b, ncols_b, &
Andreas Marek's avatar
Andreas Marek committed
828
                                          c, nrows_c, ncols_c, error)
829
      use iso_c_binding
830
      use elpa1_auxiliary_impl
831
      class(elpa_impl_t)              :: self
832
      character*1                     :: uplo_a, uplo_c
833
      integer(kind=c_int), intent(in) :: nrows_b, ncols_b, nrows_c, ncols_c, ncb
834
#ifdef USE_ASSUMED_SIZE
835
      real(kind=c_double)             :: a(self%local_nrows,*), b(nrows_b,*), c(nrows_c,*)
836
#else
837
      real(kind=c_double)             :: a(self%local_nrows,self%local_ncols), b(nrows_b,ncols_b), c(nrows_c,ncols_c)
838
#endif
839
      integer, optional               :: error
840
841
      logical                         :: success_l

842
      success_l = elpa_mult_at_b_real_double_impl(self, uplo_a, uplo_c, ncb, a, b, nrows_b, ncols_b, &
Andreas Marek's avatar
Andreas Marek committed
843
                                                  c, nrows_c, ncols_c)
844
      if (present(error)) then
845
        if (success_l) then
846
          error = ELPA_OK
847
        else
848
          error = ELPA_ERROR
849
850
        endif
      else if (.not. success_l) then
851
        write(error_unit,'(a)') "ELPA: Error in hermitian_multiply() and you did not check for errors!"
852
853
854
      endif
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
855
    !> \brief  elpa_hermitian_multiply_f: class method to perform C : = A**T * B for float real matrices
856
857
858
    !>         where   A is a square matrix (self%na,self%na) which is optionally upper or lower triangular
    !>                 B is a (self%na,ncb) matrix
    !>                 C is a (self%na,ncb) matrix where optionally only the upper or lower
Andreas Marek's avatar
Andreas Marek committed
859
860
861
862
863
864
865
    !>                   triangle may be computed
    !>
    !> the MPI commicators and the block-cyclic distribution block size are already known to the type.
    !> Thus the class method "setup" must be called BEFORE this method is used
    !>
    !> \details
    !>
866
    !> \param  self                 class(elpa_t), the ELPA object
Andreas Marek's avatar
Andreas Marek committed
867
868
869
870
871
872
873
874
875
876
877
878
879
880
    !> \param  uplo_a               'U' if A is upper triangular
    !>                              'L' if A is lower triangular
    !>                              anything else if A is a full matrix
    !>                              Please note: This pertains to the original A (as set in the calling program)
    !>                                           whereas the transpose of A is used for calculations
    !>                              If uplo_a is 'U' or 'L', the other triangle is not used at all,
    !>                              i.e. it may contain arbitrary numbers
    !> \param uplo_c                'U' if only the upper diagonal part of C is needed
    !>                              'L' if only the upper diagonal part of C is needed
    !>                              anything else if the full matrix C is needed
    !>                              Please note: Even when uplo_c is 'U' or 'L', the other triangle may be
    !>                                            written to a certain extent, i.e. one shouldn't rely on the content there!
    !> \param ncb                   Number of columns  of global matrices B and C
    !> \param a                     matrix a
881
882
    !> \param self%local_nrows      number of rows of local (sub) matrix a, set with class method set("local_nrows",value)
    !> \param self%local_ncols      number of columns of local (sub) matrix a, set with class method set("local_ncols",value)
Andreas Marek's avatar
Andreas Marek committed
883
884
885
886
887
888
    !> \param b                     matrix b
    !> \param nrows_b               number of rows of local (sub) matrix b
    !> \param ncols_b               number of columns of local (sub) matrix b
    !> \param c                     matrix c
    !> \param nrows_c               number of rows of local (sub) matrix c
    !> \param ncols_c               number of columns of local (sub) matrix c
889
    !> \param error                 optional argument, returns an error code, which can be queried with elpa_strerr
890
    subroutine elpa_hermitian_multiply_f (self,uplo_a, uplo_c, ncb, a, b, nrows_b, ncols_b, &
Andreas Marek's avatar
Andreas Marek committed
891
                                          c, nrows_c, ncols_c, error)
892
      use iso_c_binding
893
      use elpa1_auxiliary_impl
894
      class(elpa_impl_t)              :: self
895
      character*1                     :: uplo_a, uplo_c
896
      integer(kind=c_int), intent(in) :: nrows_b, ncols_b, nrows_c, ncols_c, ncb
897
#ifdef USE_ASSUMED_SIZE
898
      real(kind=c_float)              :: a(self%local_nrows,*), b(self%local_nrows,*), c(nrows_c,*)
899
#else
Andreas Marek's avatar
Andreas Marek committed
900
      real(kind=c_float)              :: a(self%local_nrows,self%local_ncols), b(nrows_b,ncols_b), c(nrows_c,ncols_c)
901
#endif
902
      integer, optional               :: error
903
904
      logical                         :: success_l
#ifdef WANT_SINGLE_PRECISION_REAL
905
      success_l = elpa_mult_at_b_real_single_impl(self, uplo_a, uplo_c, ncb, a, b, nrows_b, ncols_b, &
Andreas Marek's avatar
Andreas Marek committed
906
                                                  c, nrows_c, ncols_c)
907
      if (present(error)) then
908
        if (success_l) then
909
          error = ELPA_OK
910
        else
911
          error = ELPA_ERROR
912
913
        endif
      else if (.not. success_l) then
914
        write(error_unit,'(a)') "ELPA: Error in hermitian_multiply() and you did not check for errors!"
915
      endif
916
917
#else
      print *,"This installation of the ELPA library has not been build with single-precision support"
918
      error = ELPA_ERROR
919
920
921
#endif
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
922
    !> \brief  elpa_hermitian_multiply_dc: class method to perform C : = A**H * B for double complex matrices
923
924
925
    !>         where   A is a square matrix (self%na,self%na) which is optionally upper or lower triangular
    !>                 B is a (self%na,ncb) matrix
    !>                 C is a (self%na,ncb) matrix where optionally only the upper or lower
Andreas Marek's avatar
Andreas Marek committed
926
927
928
929
930
931
932
    !>                   triangle may be computed
    !>
    !> the MPI commicators and the block-cyclic distribution block size are already known to the type.
    !> Thus the class method "setup" must be called BEFORE this method is used
    !>
    !> \details
    !>
933
    !> \param  self                 class(elpa_t), the ELPA object
Andreas Marek's avatar
Andreas Marek committed
934
935
936
937
938
939
940
941
942
943
944
945
946
947
    !> \param  uplo_a               'U' if A is upper triangular
    !>                              'L' if A is lower triangular
    !>                              anything else if A is a full matrix
    !>                              Please note: This pertains to the original A (as set in the calling program)
    !>                                           whereas the transpose of A is used for calculations
    !>                              If uplo_a is 'U' or 'L', the other triangle is not used at all,
    !>                              i.e. it may contain arbitrary numbers
    !> \param uplo_c                'U' if only the upper diagonal part of C is needed
    !>                              'L' if only the upper diagonal part of C is needed
    !>                              anything else if the full matrix C is needed
    !>                              Please note: Even when uplo_c is 'U' or 'L', the other triangle may be
    !>                                            written to a certain extent, i.e. one shouldn't rely on the content there!
    !> \param ncb                   Number of columns  of global matrices B and C
    !> \param a                     matrix a
948
949
    !> \param self%local_nrows      number of rows of local (sub) matrix a, set with class method set("local_nows",value)
    !> \param self%local_ncols      number of columns of local (sub) matrix a, set with class method set("local_ncols",value)
Andreas Marek's avatar
Andreas Marek committed
950
951
952
953
954
955
    !> \param b                     matrix b
    !> \param nrows_b               number of rows of local (sub) matrix b
    !> \param ncols_b               number of columns of local (sub) matrix b
    !> \param c                     matrix c
    !> \param nrows_c               number of rows of local (sub) matrix c
    !> \param ncols_c               number of columns of local (sub) matrix c
956
    !> \param error                 optional argument, returns an error code, which can be queried with elpa_strerr
957
    subroutine elpa_hermitian_multiply_dc (self,uplo_a, uplo_c, ncb, a, b, nrows_b, ncols_b, &
Andreas Marek's avatar
Andreas Marek committed
958
                                          c, nrows_c, ncols_c, error)
959
      use iso_c_binding
960
      use elpa1_auxiliary_impl
961
      class(elpa_impl_t)              :: self
962
      character*1                     :: uplo_a, uplo_c
963
      integer(kind=c_int), intent(in) :: nrows_b, ncols_b, nrows_c,