test_real2.F90 12.5 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
!    This file is part of ELPA.
!
!    The ELPA library was originally created by the ELPA consortium, 
!    consisting of the following organizations:
!
!    - 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 Naturwissenschaftrn, 
!      Leipzig, Abt. Komplexe Strukutren in Biologie und Kognition, 
!      and  
!    - IBM Deutschland GmbH
!
!
!    More information can be found here:
!    http://elpa.rzg.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.
!
!
42
#include "config-f90.h"
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
program test_real2

!-------------------------------------------------------------------------------
! Standard eigenvalue problem - REAL version
!
! This program demonstrates the use of the ELPA module
! together with standard scalapack routines
! 
! Copyright of the original code rests with the authors inside the ELPA
! consortium. The copyright of any additional modifications shall rest
! with their original authors, but shall adhere to the licensing terms
! distributed along with the original code in the file "COPYING".
!
!-------------------------------------------------------------------------------

   use ELPA1
   use ELPA2

   implicit none
   include 'mpif.h'

   !-------------------------------------------------------------------------------
   ! Please set system size parameters below!
   ! na:   System size
   ! nev:  Number of eigenvectors to be calculated
   ! nblk: Blocking factor in block cyclic distribution
   !-------------------------------------------------------------------------------

   integer :: nblk
   integer na, nev

   !-------------------------------------------------------------------------------
   !  Local Variables

   integer np_rows, np_cols, na_rows, na_cols

   integer myid, nprocs, my_prow, my_pcol, mpi_comm_rows, mpi_comm_cols
   integer i, mpierr, my_blacs_ctxt, sc_desc(9), info, nprow, npcol

   integer, external :: numroc

   real*8 err, errmax
   real*8, allocatable :: a(:,:), z(:,:), tmp1(:,:), tmp2(:,:), as(:,:), ev(:)

   integer :: iseed(4096) ! Random seed, size should be sufficient for every generator
   integer :: STATUS
89
#ifdef WITH_OPENMP
90
   integer :: omp_get_max_threads,  required_mpi_thread_level, provided_mpi_thread_level
91
#endif
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
   logical :: write_to_file
   !-------------------------------------------------------------------------------
   !  Parse command line argumnents, if given
   character*16 arg1
   character*16 arg2
   character*16 arg3
   character*16 arg4

   write_to_file = .false.

   nblk = 16
   na = 4000
   nev = 1500

   if (iargc() == 3) then
      call getarg(1, arg1)
      call getarg(2, arg2)
      call getarg(3, arg3)
      read(arg1, *) na
      read(arg2, *) nev
      read(arg3, *) nblk
   endif

   if (iargc() == 4) then
      call getarg(1, arg1)
      call getarg(2, arg2)
      call getarg(3, arg3)
      call getarg(4, arg4)
      read(arg1, *) na
      read(arg2, *) nev
      read(arg3, *) nblk
   endif
   !-------------------------------------------------------------------------------
   !  MPI Initialization

127
#ifndef WITH_OPENMP
128
   call mpi_init(mpierr)
129
130
131
132
133
134
135
136
#else
   required_mpi_thread_level = MPI_THREAD_MULTIPLE
   call mpi_init_thread(required_mpi_thread_level,     &
                        provided_mpi_thread_level, mpierr)

   if (required_mpi_thread_level .ne. provided_mpi_thread_level) then
      print *,"MPI ERROR: MPI_THREAD_MULTIPLE is not provided on this system"
      print *,"           ", provided_mpi_thread_level, " is available"
137
138
      call EXIT(1)
      stop 1
139
140
141
   endif

#endif  
142
143
144
145
   call mpi_comm_rank(mpi_comm_world,myid,mpierr)
   call mpi_comm_size(mpi_comm_world,nprocs,mpierr)

   STATUS = 0
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
#ifdef WITH_OPENMP
   if (myid .eq. 0) then
      print *,"Threaded version of test program"
      print *,"Using ",omp_get_max_threads()," threads"
      print *," "
   endif
#endif

   if (myid .eq. 0) then
      print *," "
      print *,"This ELPA2 is build with"
#ifdef WITH_AVX_REAL_BLOCK2
      print *,"AVX optimized kernel (2 blocking) for real matrices"
#endif
#ifdef WITH_AVX_REAL_BLOCK4
      print *,"AVX optimized kernel (4 blocking) for real matrices"
#endif
#ifdef WITH_AVX_REAL_BLOCK6
      print *,"AVX optimized kernel (6 blocking) for real matrices"
#endif
#ifdef WITH_AVX_SANDYBRIDGE
     print *,"AVX SANDYBRIDGE optimized kernel for real matrices"
#endif
#ifdef WITH_GENERIC
     print *,"GENERIC kernel for real matrices"
#endif
#ifdef WITH_GENERIC_SIMPLE
     print *,"GENERIC SIMPLE kernel for real matrices"
#endif
#ifdef WITH_SSE_AS
     print *,"SSE ASSEMBLER kernel for real matrices"
#endif
#ifdef WITH_BGP
     print *,"BGP kernel for real matrices"
#endif
#ifdef WITH_BGQ
     print *,"BGQ kernel for real matrices"
#endif
   endif
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
   if (arg4 .eq. "output") then 
      write_to_file = .true.
      if (myid .eq. 0) print *,"Writing output files"
   endif
   !-------------------------------------------------------------------------------
   ! Selection of number of processor rows/columns
   ! We try to set up the grid square-like, i.e. start the search for possible
   ! divisors of nprocs with a number next to the square root of nprocs
   ! and decrement it until a divisor is found.

   do np_cols = NINT(SQRT(REAL(nprocs))),2,-1
      if(mod(nprocs,np_cols) == 0 ) exit
   enddo
   ! at the end of the above loop, nprocs is always divisible by np_cols

   np_rows = nprocs/np_cols

   if(myid==0) then
      print *
      print '(a)','Standard eigenvalue problem - REAL version'
      print *
      print '(3(a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
      print '(3(a,i0))','Number of processor rows=',np_rows,', cols=',np_cols,', total=',nprocs
      print *
   endif

   !-------------------------------------------------------------------------------
   ! Set up BLACS context and MPI communicators
   !
   ! The BLACS context is only necessary for using Scalapack.
   !
   ! For ELPA, the MPI communicators along rows/cols are sufficient,
   ! and the grid setup may be done in an arbitrary way as long as it is
   ! consistent (i.e. 0<=my_prow<np_rows, 0<=my_pcol<np_cols and every
   ! process has a unique (my_prow,my_pcol) pair).

   my_blacs_ctxt = mpi_comm_world
   call BLACS_Gridinit( my_blacs_ctxt, 'C', np_rows, np_cols )
   call BLACS_Gridinfo( my_blacs_ctxt, nprow, npcol, my_prow, my_pcol )

   if (myid==0) then
     print '(a)','| Past BLACS_Gridinfo.'
   end if

   ! All ELPA routines need MPI communicators for communicating within
   ! rows or columns of processes, these are set in get_elpa_row_col_comms.

   call get_elpa_row_col_comms(mpi_comm_world, my_prow, my_pcol, &
                               mpi_comm_rows, mpi_comm_cols)

   if (myid==0) then
     print '(a)','| Past split communicator setup for rows and columns.'
   end if

   ! Determine the necessary size of the distributed matrices,
   ! we use the Scalapack tools routine NUMROC for that.

   na_rows = numroc(na, nblk, my_prow, 0, np_rows)
   na_cols = numroc(na, nblk, my_pcol, 0, np_cols)

   ! Set up a scalapack descriptor for the checks below.
   ! For ELPA the following restrictions hold:
   ! - block sizes in both directions must be identical (args 4+5)
   ! - first row and column of the distributed matrix must be on row/col 0/0 (args 6+7)

   call descinit( sc_desc, na, na, nblk, nblk, 0, 0, my_blacs_ctxt, na_rows, info )

   if (myid==0) then
     print '(a)','| Past scalapack descriptor setup.'
   end if

   !-------------------------------------------------------------------------------
   ! Allocate matrices and set up a test matrix for the eigenvalue problem

   allocate(a (na_rows,na_cols))
   allocate(z (na_rows,na_cols))
   allocate(as(na_rows,na_cols))

   allocate(ev(na))

   ! For getting a symmetric test matrix A we get a random matrix Z
   ! and calculate A = Z + Z**T

   ! We want different random numbers on every process
   ! (otherways A might get rank deficient):

   iseed(:) = myid
   call RANDOM_SEED(put=iseed)

   call RANDOM_NUMBER(z)

   a(:,:) = z(:,:)

   if (myid==0) then
     print '(a)','| Random matrix block has been set up. (only processor 0 confirms this step)'
   end if

   call pdtran(na, na,  1.d0, z, 1, 1, sc_desc, 1.d0, a, 1, 1, sc_desc) ! A = A + Z**T

   if (myid==0) then
     print '(a)','| Random matrix has been symmetrized.'
   end if

   ! Save original matrix A for later accuracy checks

   as = a

   ! set print flag in elpa1
   elpa_print_times = .true.

   !-------------------------------------------------------------------------------
   ! Calculate eigenvalues/eigenvectors

   if (myid==0) then
     print '(a)','| Entering two-stage ELPA solver ... '
     print *
   end if

   call mpi_barrier(mpi_comm_world, mpierr) ! for correct timings only
   call solve_evp_real_2stage(na, nev, a, na_rows, ev, z, na_rows, nblk, &
                              mpi_comm_rows, mpi_comm_cols, mpi_comm_world)

   if (myid==0) then
     print '(a)','| Two-step ELPA solver complete.'
     print *
   end if

   if(myid == 0) print *,'Time transform to tridi :',time_evp_fwd
   if(myid == 0) print *,'Time solve tridi        :',time_evp_solve
   if(myid == 0) print *,'Time transform back EVs :',time_evp_back
315
316
317
   if(myid == 0) print *,'Total time (sum above)  :',time_evp_back+time_evp_solve+time_evp_fwd
      

318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395

   if(write_to_file) then
      if (myid == 0) then
         open(17,file="EVs_real2_out.txt",form='formatted',status='new')
         do i=1,na
            write(17,*) i,ev(i)
         enddo
         close(17)
      endif
   endif
   !-------------------------------------------------------------------------------
   ! Test correctness of result (using plain scalapack routines)

   deallocate(a)
   allocate(tmp1(na_rows,na_cols))

   ! 1. Residual (maximum of || A*Zi - Zi*EVi ||)

   ! tmp1 =  A * Z
   call pdgemm('N','N',na,nev,na,1.d0,as,1,1,sc_desc, &
           z,1,1,sc_desc,0.d0,tmp1,1,1,sc_desc)

   deallocate(as)
   allocate(tmp2(na_rows,na_cols))

   ! tmp2 = Zi*EVi
   tmp2(:,:) = z(:,:)
   do i=1,nev
      call pdscal(na,ev(i),tmp2,1,i,sc_desc,1)
   enddo

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

   ! Get maximum norm of columns of tmp1
   errmax = 0
   do i=1,nev
      err = 0
      call pdnrm2(na,err,tmp1,1,i,sc_desc,1)
      errmax = max(errmax, err)
   enddo

   ! Get maximum error norm over all processors
   err = errmax
   call mpi_allreduce(err,errmax,1,MPI_REAL8,MPI_MAX,MPI_COMM_WORLD,mpierr)
   if(myid==0) print *
   if(myid==0) print *,'Error Residual     :',errmax

   if (errmax .gt. 5e-12) then
      status = 1
   endif

   ! 2. Eigenvector orthogonality

   ! tmp1 = Z**T * Z
   tmp1 = 0
   call pdgemm('T','N',nev,nev,na,1.d0,z,1,1,sc_desc, &
           z,1,1,sc_desc,0.d0,tmp1,1,1,sc_desc)
   ! Initialize tmp2 to unit matrix
   tmp2 = 0
   call pdlaset('A',nev,nev,0.d0,1.d0,tmp2,1,1,sc_desc)

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

   ! Get maximum error (max abs value in tmp1)
   err = maxval(abs(tmp1))
   call mpi_allreduce(err,errmax,1,MPI_REAL8,MPI_MAX,MPI_COMM_WORLD,mpierr)
   if(myid==0) print *,'Error Orthogonality:',errmax
   
   if (errmax .gt. 5e-12) then
      status = 1
   endif

   deallocate(z)
   deallocate(tmp1)
   deallocate(tmp2)
   deallocate(ev)
396
   call blacs_gridexit(my_blacs_ctxt)
397
398
399
400
401
   call mpi_finalize(mpierr)
   call EXIT(STATUS)
end

!-------------------------------------------------------------------------------