elpa_impl.F90 74.3 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
   type(c_ptr)         :: index = C_NULL_PTR
63

64
   !> \brief methods available with the elpa_impl_t type
65
   contains
66
     !> \brief the puplic methods
67
     ! con-/destructor
68
69
     procedure, public :: setup => elpa_setup                   !< a setup method: implemented in elpa_setup
     procedure, public :: destroy => elpa_destroy               !< a destroy method: implemented in elpa_destroy
70

71
     ! KV store
72
73
74
75
     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
76

77
78
79
80
81
82

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


83
     !> \brief the private methods
84

85
     procedure, private :: elpa_set_integer                     !< private methods to implement the setting of an integer/double key/value pair
86
     procedure, private :: elpa_set_double
87

88
89
90
     procedure, private :: elpa_get_integer                     !< private methods to implement the querry of an integer/double key/value pair
     procedure, private :: elpa_get_double

Andreas Marek's avatar
Andreas Marek committed
91
     procedure, private :: elpa_solve_d                         !< private methods to implement the solve step for real/complex
92
                                                                !< double/single matrices
93
94
95
     procedure, private :: elpa_solve_f
     procedure, private :: elpa_solve_dc
     procedure, private :: elpa_solve_fc
96

97
98
     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
99
     procedure, private :: elpa_hermitian_multiply_dc           !< for complex valued matrices:   a**H * b
100
     procedure, private :: elpa_hermitian_multiply_fc
101

Andreas Marek's avatar
Andreas Marek committed
102
     procedure, private :: elpa_cholesky_d                      !< private methods to implement the cholesky factorisation of
103
                                                                !< real/complex double/single matrices
104
105
106
     procedure, private :: elpa_cholesky_f
     procedure, private :: elpa_cholesky_dc
     procedure, private :: elpa_cholesky_fc
107

Andreas Marek's avatar
Andreas Marek committed
108
     procedure, private :: elpa_invert_trm_d                    !< private methods to implement the inversion of a triangular
109
                                                                !< real/complex double/single matrix
110
111
112
     procedure, private :: elpa_invert_trm_f
     procedure, private :: elpa_invert_trm_dc
     procedure, private :: elpa_invert_trm_fc
113

Andreas Marek's avatar
Andreas Marek committed
114
115
     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
116

117
     procedure, private :: associate_int => elpa_associate_int  !< private method to set some pointers
118

119
  end type elpa_impl_t
120

121
  !> \brief the implementation of the private methods
122
  contains
123
124
125
126
    !> \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
127
    function elpa_impl_allocate(error) result(obj)
Andreas Marek's avatar
Andreas Marek committed
128
129
      use precision
      use elpa_utilities, only : error_unit
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
130
      use elpa_generated_fortran_interfaces
Andreas Marek's avatar
Andreas Marek committed
131

132
133
134
135
      type(elpa_impl_t), pointer   :: obj
      integer, optional            :: error

      allocate(obj)
Andreas Marek's avatar
Andreas Marek committed
136

Andreas Marek's avatar
Andreas Marek committed
137
      ! check whether init has ever been called
138
      if ( elpa_initialized() .ne. ELPA_OK) then
139
        write(error_unit, *) "elpa_allocate(): you must call elpa_init() once before creating instances of ELPA"
140
141
        if(present(error)) then
          error = ELPA_ERROR
142
        endif
Andreas Marek's avatar
Andreas Marek committed
143
144
        return
      endif
Andreas Marek's avatar
Andreas Marek committed
145

146
      obj%index = elpa_index_instance_c()
147
148

      ! Associate some important integer pointers for convenience
149
150
151
152
153
154
155
156
      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
157
158
      endif
    end function
Andreas Marek's avatar
Andreas Marek committed
159

160
161

    !c> elpa_t elpa_allocate();
162
    function elpa_impl_allocate_c(error) result(ptr) bind(C, name="elpa_allocate")
163
164
165
166
167
168
169
170
171
172
      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


    !c> void elpa_deallocate(elpa_t handle);
173
    subroutine elpa_impl_deallocate_c(handle) bind(C, name="elpa_deallocate")
174
175
176
177
178
179
180
181
182
      type(c_ptr), value :: handle
      type(elpa_impl_t), pointer :: self

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


183
184
185
186
    !> \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
187
    function elpa_setup(self) result(error)
188
      use elpa1_impl, only : elpa_get_communicators_impl
189
      class(elpa_impl_t), intent(inout) :: self
190
      integer :: error, error2
191
      integer :: mpi_comm_parent, mpi_comm_rows, mpi_comm_cols, mpierr, process_row, process_col, timings
192

193
#ifdef WITH_MPI
194
195
196
197
      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
198

199
200
201
        call self%get("mpi_comm_parent", mpi_comm_parent)
        call self%get("process_row", process_row)
        call self%get("process_col", process_col)
202
        mpierr = elpa_get_communicators_impl(&
203
204
205
                        mpi_comm_parent, &
                        process_row, &
                        process_col, &
206
207
                        mpi_comm_rows, &
                        mpi_comm_cols)
208

209
210
211
        call self%set("mpi_comm_rows", mpi_comm_rows)
        call self%set("mpi_comm_cols", mpi_comm_cols)

212
        error = ELPA_OK
213
      endif
214

215
216
      if (self%is_set("mpi_comm_rows") == 1 .and. self%is_set("mpi_comm_cols") == 1) then
        error = ELPA_OK
217
      endif
218
219
220
#else
      error = ELPA_OK
#endif
221

222
223
      call self%get("timings",timings)
      if (timings == 1) then
224
225
226
        call self%timer%enable()
      endif

227
    end function
228

229
230

    !c> int elpa_setup(elpa_t handle);
231
    function elpa_setup_c(handle) result(error) bind(C, name="elpa_setup")
232
233
234
235
236
237
238
239
240
      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


241
242
243
244
245
246
    !> \brief subroutine to set an integer key/value pair
    !> Parameters
    !> \param   self       class(elpa_impl_t) the allocated ELPA object
    !> \param   name       string, the key
    !> \param   value      integer, the value to be set
    !> \result  error      integer, the error code
247
    subroutine elpa_set_integer(self, name, value, error)
248
      use iso_c_binding
249
250
      use elpa_generated_fortran_interfaces
      use elpa_utilities, only : error_unit
251
      class(elpa_impl_t)              :: self
252
253
      character(*), intent(in)        :: name
      integer(kind=c_int), intent(in) :: value
254
255
      integer, optional               :: error
      integer                         :: actual_error
256

257
      actual_error = elpa_index_set_int_value_c(self%index, name // c_null_char, value, 0)
258

259
260
      if (present(error)) then
        error = actual_error
261

262
      else if (actual_error /= ELPA_OK) then
263
264
        write(error_unit,'(a,i0,a)') "ELPA: Error setting option '" // name // "' to value ", value, &
                " (got: " // elpa_strerr(actual_error) // ") and you did not check for errors!"
265
      end if
266
267
    end subroutine

268
269

    !c> void elpa_set_integer(elpa_t handle, const char *name, int value, int *error);
270
    subroutine elpa_set_integer_c(handle, name_p, value, error) bind(C, name="elpa_set_integer")
271
272
273
274
275
276
277
278
279
280
281
282
283
      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


284
285
286
287
    !> \brief function to get an integer key/value pair
    !> Parameters
    !> \param   self       class(elpa_impl_t) the allocated ELPA object
    !> \param   name       string, the key
288
    !> \param   value      integer, the value of the key/vaue pair
289
    !> \param   error      integer, optional, to store an error code
290
    subroutine elpa_get_integer(self, name, value, error)
291
      use iso_c_binding
292
      use elpa_generated_fortran_interfaces
293
      use elpa_utilities, only : error_unit
294
      class(elpa_impl_t)             :: self
295
296
      character(*), intent(in)       :: name
      integer(kind=c_int)            :: value
297
      integer, intent(out), optional :: error
298
      integer                        :: actual_error
299

300
301
302
303
304
305
306
      value = elpa_index_get_int_value_c(self%index, name // c_null_char, actual_error)
      if (present(error)) then
        error = actual_error
      else if (actual_error /= ELPA_OK) then
        write(error_unit,'(a)') "ELPA: Error getting option '" // name // "'" // &
                " (got: " // elpa_strerr(actual_error) // ") and you did not check for errors!"
      end if
307
    end subroutine
Andreas Marek's avatar
Andreas Marek committed
308

Andreas Marek's avatar
Andreas Marek committed
309

310
311
    !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
312
313
314
315
316
317
318
319
320
      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)
321
322
      call elpa_get_integer(self, name, value, error)
    end subroutine
Andreas Marek's avatar
Andreas Marek committed
323
324


325
326
327
328
329
    !> \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
330
    function elpa_is_set(self, name) result(state)
331
332
      use iso_c_binding
      use elpa_generated_fortran_interfaces
333
      class(elpa_impl_t)       :: self
334
      character(*), intent(in) :: name
335
      integer                  :: state
336

337
      state = elpa_index_value_is_set_c(self%index, name // c_null_char)
338
339
    end function

340
341
342
343
344
345
    !> \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
346
347
348
349
350
351
352
353
354
355
356
357
358
    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)
359
360
361
      use elpa_generated_fortran_interfaces
      class(elpa_impl_t), intent(in) :: self
      character(kind=c_char, len=*), intent(in) :: option_name
362
363
364
365
      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
366

367
368
      nullify(string)

369
      call self%get(option_name, val, actual_error)
370
371
372
373
374
      if (actual_error /= ELPA_OK) then
        if (present(error)) then
          error = actual_error
        endif
        return
375
376
      endif

377
378
379
380
      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
381

382
383
384
385
      if (present(error)) then
        error = actual_error
      endif
    end function
386

387
388

    subroutine elpa_set_double(self, name, value, error)
Andreas Marek's avatar
Andreas Marek committed
389
      use iso_c_binding
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
390
      use elpa_generated_fortran_interfaces
391
      use elpa_utilities, only : error_unit
392
      class(elpa_impl_t)              :: self
393
      character(*), intent(in)        :: name
394
      real(kind=c_double), intent(in) :: value
395
396
      integer, optional               :: error
      integer                         :: actual_error
Andreas Marek's avatar
Andreas Marek committed
397

398
      actual_error = elpa_index_set_double_value_c(self%index, name // c_null_char, value, 0)
Andreas Marek's avatar
Andreas Marek committed
399

400
401
402
      if (present(error)) then
        error = actual_error
      else if (actual_error /= ELPA_OK) then
403
404
        write(error_unit,'(a,es12.5,a)') "ELPA: Error setting option '" // name // "' to value ", value, &
                " (got: " // elpa_strerr(actual_error) // ") and you did not check for errors!"
405
406
      end if
    end subroutine
Andreas Marek's avatar
Andreas Marek committed
407
408


409
    !c> void elpa_set_double(elpa_t handle, const char *name, double value, int *error);
410
    subroutine elpa_set_double_c(handle, name_p, value, error) bind(C, name="elpa_set_double")
411
412
413
414
415
416
417
418
419
420
421
422
423
      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


424
    subroutine elpa_get_double(self, name, value, error)
Andreas Marek's avatar
Andreas Marek committed
425
      use iso_c_binding
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
426
      use elpa_generated_fortran_interfaces
427
      use elpa_utilities, only : error_unit
428
      class(elpa_impl_t)             :: self
429
      character(*), intent(in)       :: name
430
      real(kind=c_double)            :: value
431
      integer, intent(out), optional :: error
432
      integer                        :: actual_error
433

434
435
436
437
438
439
440
      value = elpa_index_get_double_value_c(self%index, name // c_null_char, actual_error)
      if (present(error)) then
        error = actual_error
      else if (actual_error /= ELPA_OK) then
        write(error_unit,'(a)') "ELPA: Error getting option '" // name // "'" // &
                " (got: " // elpa_strerr(actual_error) // ") and you did not check for errors!"
      end if
441
    end subroutine
Andreas Marek's avatar
Andreas Marek committed
442

443
444
    !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
445
446
447
448
449
450
451
452
453
      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)
454
455
      call elpa_get_double(self, name, value, error)
    end subroutine
Andreas Marek's avatar
Andreas Marek committed
456
457


458
    function elpa_associate_int(self, name) result(value)
Andreas Marek's avatar
Andreas Marek committed
459
      use iso_c_binding
460
      use elpa_generated_fortran_interfaces
461
462
      use elpa_utilities, only : error_unit
      class(elpa_impl_t)             :: self
463
464
      character(*), intent(in)       :: name
      integer(kind=c_int), pointer   :: value
Andreas Marek's avatar
Andreas Marek committed
465

466
467
      type(c_ptr)                    :: value_p

468
      value_p = elpa_index_get_int_loc_c(self%index, name // c_null_char)
469
470
471
      if (.not. c_associated(value_p)) then
        write(error_unit, '(a,a,a)') "ELPA: Warning, received NULL pointer for entry '", name, "'"
      endif
472
473
      call c_f_pointer(value_p, value)
    end function
Andreas Marek's avatar
Andreas Marek committed
474

475

476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
    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

      s = self%timer%get(name1, name2, name3, name4, name5, name6)
    end function


    subroutine elpa_print_times(self)
      class(elpa_impl_t), intent(in) :: self
      call self%timer%print()
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
491
492
    !>  \brief elpa_solve_d: class method to solve the eigenvalue problem for double real matrices
    !>
493
494
    !>  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
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
    !>  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
516
    subroutine elpa_solve_d(self, a, ev, q, error)
517
518
      use elpa2_impl
      use elpa1_impl
519
      use elpa_utilities, only : error_unit
Andreas Marek's avatar
Andreas Marek committed
520
      use iso_c_binding
521
      class(elpa_impl_t)  :: self
Andreas Marek's avatar
Andreas Marek committed
522

523
524
525
#ifdef USE_ASSUMED_SIZE
      real(kind=c_double) :: a(self%local_nrows, *), q(self%local_nrows, *)
#else
526
      real(kind=c_double) :: a(self%local_nrows, self%local_ncols), q(self%local_nrows, self%local_ncols)
527
#endif
528
      real(kind=c_double) :: ev(self%na)
529

530
      integer, optional   :: error
531
      integer(kind=c_int) :: error_actual, solver
532
      logical             :: success_l
533

534

535
536
      call self%get("solver", solver)
      if (solver .eq. ELPA_SOLVER_1STAGE) then
537
        success_l = elpa_solve_evp_real_1stage_double_impl(self, a, ev, q)
538

539
      else if (solver .eq. ELPA_SOLVER_2STAGE) then
540
        success_l = elpa_solve_evp_real_2stage_double_impl(self, a, ev, q)
541
542
543
544
      else
        print *,"unknown solver"
        stop
      endif
545

546
      if (present(error)) then
547
        if (success_l) then
548
          error = ELPA_OK
549
        else
550
          error = ELPA_ERROR
551
552
553
554
555
556
        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

557
558
    !c> void elpa_solve_d(elpa_t handle, double *a, double *ev, double *q, int *error);
    subroutine elpa_solve_d_c(handle, a_p, ev_p, q_p, error) bind(C, name="elpa_solve_d")
559
560
561
562
563
564
565
566
567
568
569
      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])

570
      call elpa_solve_d(self, a, ev, q, error)
571
572
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
573
574
575

    !>  \brief elpa_solve_f: class method to solve the eigenvalue problem for float real matrices
    !>
576
577
    !>  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
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
    !>  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
599
    subroutine elpa_solve_f(self, a, ev, q, error)
600
601
      use elpa2_impl
      use elpa1_impl
602
603
      use elpa_utilities, only : error_unit
      use iso_c_binding
604
      class(elpa_impl_t)  :: self
605
606
607
#ifdef USE_ASSUMED_SIZE
      real(kind=c_float)  :: a(self%local_nrows, *), q(self%local_nrows, *)
#else
608
      real(kind=c_float)  :: a(self%local_nrows, self%local_ncols), q(self%local_nrows, self%local_ncols)
609
#endif
610
      real(kind=c_float)  :: ev(self%na)
611

612
      integer, optional   :: error
613
      integer(kind=c_int) :: error_actual, solver
614
      logical             :: success_l
615

616
#ifdef WANT_SINGLE_PRECISION_REAL
617

618
619
      call self%get("solver",solver)
      if (solver .eq. ELPA_SOLVER_1STAGE) then
620
        success_l = elpa_solve_evp_real_1stage_single_impl(self, a, ev, q)
621

622
      else if (solver .eq. ELPA_SOLVER_2STAGE) then
623
        success_l = elpa_solve_evp_real_2stage_single_impl(self, a, ev, q)
624
625
626
627
      else
        print *,"unknown solver"
        stop
      endif
628

629
      if (present(error)) then
630
        if (success_l) then
631
          error = ELPA_OK
632
        else
633
          error = ELPA_ERROR
634
635
636
637
638
        endif
      else if (.not. success_l) then
        write(error_unit,'(a)') "ELPA: Error in solve() and you did not check for errors!"
      endif
#else
639
      print *,"This installation of the ELPA library has not been build with single-precision support"
640
      error = ELPA_ERROR
641
642
643
#endif
    end subroutine

644

645
646
    !c> void elpa_solve_f(elpa_t handle, float *a, float *ev, float *q, int *error);
    subroutine elpa_solve_f_c(handle, a_p, ev_p, q_p, error) bind(C, name="elpa_solve_f")
647
648
649
650
651
652
653
654
655
656
657
      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])

658
      call elpa_solve_f(self, a, ev, q, error)
659
660
661
    end subroutine


Andreas Marek's avatar
Andreas Marek committed
662
663
    !>  \brief elpa_solve_dc: class method to solve the eigenvalue problem for double complex matrices
    !>
664
665
    !>  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
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
    !>  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
687
    subroutine elpa_solve_dc(self, a, ev, q, error)
688
689
      use elpa2_impl
      use elpa1_impl
690
691
      use elpa_utilities, only : error_unit
      use iso_c_binding
692
      class(elpa_impl_t)             :: self
693

694
695
696
#ifdef USE_ASSUMED_SIZE
      complex(kind=c_double_complex) :: a(self%local_nrows, *), q(self%local_nrows, *)
#else
697
      complex(kind=c_double_complex) :: a(self%local_nrows, self%local_ncols), q(self%local_nrows, self%local_ncols)
698
#endif
699
      real(kind=c_double)            :: ev(self%na)
700

701
      integer, optional              :: error
702
      integer(kind=c_int)            :: error_actual, solver
703
      logical                        :: success_l
704

705
706
      call self%get("solver", solver)
      if (solver .eq. ELPA_SOLVER_1STAGE) then
707
        success_l = elpa_solve_evp_complex_1stage_double_impl(self, a, ev, q)
708

709
      else if (solver .eq. ELPA_SOLVER_2STAGE) then
710
        success_l = elpa_solve_evp_complex_2stage_double_impl(self,  a, ev, q)
711
712
713
714
      else
        print *,"unknown solver"
        stop
      endif
715

716
      if (present(error)) then
717
        if (success_l) then
718
          error = ELPA_OK
719
        else
720
          error = ELPA_ERROR
721
722
723
724
725
726
727
        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


728
729
    !c> void elpa_solve_dc(elpa_t handle, double complex *a, double *ev, double complex *q, int *error);
    subroutine elpa_solve_dc_c(handle, a_p, ev_p, q_p, error) bind(C, name="elpa_solve_dc")
730
731
732
733
734
735
736
737
738
739
740
741
      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])

742
      call elpa_solve_dc(self, a, ev, q, error)
743
744
745
    end subroutine


Andreas Marek's avatar
Andreas Marek committed
746
747
    !>  \brief elpa_solve_fc: class method to solve the eigenvalue problem for float complex matrices
    !>
748
749
    !>  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
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
    !>  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
771
    subroutine elpa_solve_fc(self, a, ev, q, error)
772
773
      use elpa2_impl
      use elpa1_impl
774
775
776
      use elpa_utilities, only : error_unit

      use iso_c_binding
777
      class(elpa_impl_t)            :: self
778
#ifdef USE_ASSUMED_SIZE
Andreas Marek's avatar
Andreas Marek committed
779
      complex(kind=c_float_complex) :: a(self%local_nrows, *), q(self%local_nrows, *)
780
#else
Andreas Marek's avatar
Andreas Marek committed
781
      complex(kind=c_float_complex) :: a(self%local_nrows, self%local_ncols), q(self%local_nrows, self%local_ncols)
782
#endif
Andreas Marek's avatar
Andreas Marek committed
783
      real(kind=c_float)            :: ev(self%na)
784

785
      integer, optional             :: error
786
      integer(kind=c_int)           :: error_actual, solver
787
      logical                       :: success_l
788
789

#ifdef WANT_SINGLE_PRECISION_COMPLEX
790

791
792
      call self%get("solver", solver)
      if (solver .eq. ELPA_SOLVER_1STAGE) then
793
        success_l = elpa_solve_evp_complex_1stage_single_impl(self, a, ev, q)
794

795
      else if (solver .eq. ELPA_SOLVER_2STAGE) then
796
        success_l = elpa_solve_evp_complex_2stage_single_impl(self,  a, ev, q)
797
798
799
800
      else
        print *,"unknown solver"
        stop
      endif
801

802
      if (present(error)) then
803
        if (success_l) then
804
          error = ELPA_OK
805
        else
806
          error = ELPA_ERROR
807
808
809
810
811
        endif
      else if (.not. success_l) then
        write(error_unit,'(a)') "ELPA: Error in solve() and you did not check for errors!"
      endif
#else
812
      print *,"This installation of the ELPA library has not been build with single-precision support"
813
      error = ELPA_ERROR
814
815
816
#endif
    end subroutine

817

818
819
    !c> void elpa_solve_fc(elpa_t handle, float complex *a, float *ev, float complex *q, int *error);
    subroutine elpa_solve_fc_c(handle, a_p, ev_p, q_p, error) bind(C, name="elpa_solve_fc")
820
821
822
823
824
825
826
827
828
829
830
831
      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])

832
      call elpa_solve_fc(self, a, ev, q, error)
833
834
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
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
    !> \brief  elpa_hermitian_multiply_d: class method to perform C : = A**T * B for double real matrices
    !>         where   A is a square matrix (na,na) which is optionally upper or lower triangular
    !>                 B is a (na,ncb) matrix
    !>                 C is a (na,ncb) matrix where optionally only the upper or lower
    !>                   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
    !>
    !> \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 na                    Number of rows/columns of global matrix A, number of rows of global matrices B and C
    !> \param ncb                   Number of columns  of global matrices B and C
    !> \param a                     matrix a
    !> \param nrows_a               number of rows of local (sub) matrix a
    !> \param ncols_a               number of columns of local (sub) matrix a
    !> \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
    subroutine elpa_hermitian_multiply_d (self,uplo_a, uplo_c, na, ncb, a, nrows_a, ncols_a, b, nrows_b, ncols_b, &
                                          c, nrows_c, ncols_c, error)
872
      use iso_c_binding
873
      use elpa1_auxiliary_impl
874
      class(elpa_impl_t)              :: self
875
      character*1                     :: uplo_a, uplo_c
Andreas Marek's avatar
Andreas Marek committed
876
      integer(kind=c_int), intent(in) :: na, nrows_a, ncols_a, nrows_b, ncols_b, nrows_c, ncols_c, ncb
877
#ifdef USE_ASSUMED_SIZE
Andreas Marek's avatar
Andreas Marek committed
878
      real(kind=c_double)             :: a(nrows_a,*), b(nrows_b,*), c(nrows_c,*)
879
#else
Andreas Marek's avatar
Andreas Marek committed
880
      real(kind=c_double)             :: a(nrows_a,ncols_a), b(nrows_b,ncols_b), c(nrows_c,ncols_c)
881
#endif
882
      integer, optional               :: error
883
884
      logical                         :: success_l

Andreas Marek's avatar
Andreas Marek committed
885
886
      success_l = elpa_mult_at_b_real_double_impl(self, uplo_a, uplo_c, na, ncb, a, nrows_a, ncols_a, b, nrows_b, ncols_b, &
                                                  c, nrows_c, ncols_c)
887
      if (present(error)) then
888
        if (success_l) then
889
          error = ELPA_OK
890
        else
891
          error = ELPA_ERROR
892
893
        endif
      else if (.not. success_l) then
894
        write(error_unit,'(a)') "ELPA: Error in hermitian_multiply() and you did not check for errors!"
895
896
897
      endif
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
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
927
928
929
930
931
    !> \brief  elpa_hermitian_multiply_f: class method to perform C : = A**T * B for float real matrices
    !>         where   A is a square matrix (na,na) which is optionally upper or lower triangular
    !>                 B is a (na,ncb) matrix
    !>                 C is a (na,ncb) matrix where optionally only the upper or lower
    !>                   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
    !>
    !> \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 na                    Number of rows/columns of global matrix A, number of rows of global matrices B and C
    !> \param ncb                   Number of columns  of global matrices B and C
    !> \param a                     matrix a
    !> \param nrows_a               number of rows of local (sub) matrix a
    !> \param ncols_a               number of columns of local (sub) matrix a
    !> \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
932
    !> \param error                 optional argument, returns an error code, which can be queried with elpa_strerr
Andreas Marek's avatar
Andreas Marek committed
933
934
    subroutine elpa_hermitian_multiply_f (self,uplo_a, uplo_c, na, ncb, a, nrows_a, ncols_a, b, nrows_b, ncols_b, &
                                          c, nrows_c, ncols_c, error)
935
      use iso_c_binding
936
      use elpa1_auxiliary_impl
937
      class(elpa_impl_t)              :: self
938
      character*1                     :: uplo_a, uplo_c
Andreas Marek's avatar
Andreas Marek committed
939
      integer(kind=c_int), intent(in) :: na, nrows_a, ncols_a, nrows_b, ncols_b, nrows_c, ncols_c, ncb
940
#ifdef USE_ASSUMED_SIZE
Andreas Marek's avatar
Andreas Marek committed
941
      real(kind=c_float)              :: a(nrows_a,*), b(nrows_b,*), c(nrows_c,*)
942
#else
Andreas Marek's avatar
Andreas Marek committed
943
      real(kind=c_float)              :: a(nrows_a,ncols_a), b(nrows_b,ncols_b), c(nrows_c,ncols_c)
944
#endif
945
      integer, optional               :: error
946
947
      logical                         :: success_l
#ifdef WANT_SINGLE_PRECISION_REAL
Andreas Marek's avatar
Andreas Marek committed
948
949
      success_l = elpa_mult_at_b_real_single_impl(self, uplo_a, uplo_c, na, ncb, a, nrows_a, ncols_a, b, nrows_a, ncols_b, &
                                                  c, nrows_c, ncols_c)
950
      if (present(error)) then
951
        if (success_l) then
952
          error = ELPA_OK
953
        else
954
          error = ELPA_ERROR
955
956
        endif
      else if (.not. success_l) then
957
        write(error_unit,'(a)') "ELPA: Error in hermitian_multiply() and you did not check for errors!"
958
      endif
959
960
#else
      print *,"This installation of the ELPA library has not been build with single-precision support"
961
      error = ELPA_ERROR
962
963
964
#endif
    end subroutine

Andreas Marek's avatar
Andreas Marek committed
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997