Line data Source code
1 : ! This file is part of ELPA.
2 : !
3 : ! The ELPA library was originally created by the ELPA consortium,
4 : ! consisting of the following organizations:
5 : !
6 : ! - Max Planck Computing and Data Facility (MPCDF), formerly known as
7 : ! Rechenzentrum Garching der Max-Planck-Gesellschaft (RZG),
8 : ! - Bergische Universität Wuppertal, Lehrstuhl für angewandte
9 : ! Informatik,
10 : ! - Technische Universität München, Lehrstuhl für Informatik mit
11 : ! Schwerpunkt Wissenschaftliches Rechnen ,
12 : ! - Fritz-Haber-Institut, Berlin, Abt. Theorie,
13 : ! - Max-Plack-Institut für Mathematik in den Naturwissenschaften,
14 : ! Leipzig, Abt. Komplexe Strukutren in Biologie und Kognition,
15 : ! and
16 : ! - IBM Deutschland GmbH
17 : !
18 : !
19 : ! More information can be found here:
20 : ! http://elpa.mpcdf.mpg.de/
21 : !
22 : ! ELPA is free software: you can redistribute it and/or modify
23 : ! it under the terms of the version 3 of the license of the
24 : ! GNU Lesser General Public License as published by the Free
25 : ! Software Foundation.
26 : !
27 : ! ELPA is distributed in the hope that it will be useful,
28 : ! but WITHOUT ANY WARRANTY; without even the implied warranty of
29 : ! MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
30 : ! GNU Lesser General Public License for more details.
31 : !
32 : ! You should have received a copy of the GNU Lesser General Public License
33 : ! along with ELPA. If not, see <http://www.gnu.org/licenses/>
34 : !
35 : ! ELPA reflects a substantial effort on the part of the original
36 : ! ELPA consortium, and we ask you to respect the spirit of the
37 : ! license that we chose: i.e., please contribute any changes you
38 : ! may have back to the original ELPA library distribution, and keep
39 : ! any derivatives of ELPA under the same license that we chose for
40 : ! the original distribution, the GNU Lesser General Public License.
41 : !
42 : !
43 : #include "config-f90.h"
44 : !>
45 : !> Fortran test programm to demonstrates the use of
46 : !> ELPA 1 real case library.
47 : !> If "HAVE_REDIRECT" was defined at build time
48 : !> the stdout and stderr output of each MPI task
49 : !> can be redirected to files if the environment
50 : !> variable "REDIRECT_ELPA_TEST_OUTPUT" is set
51 : !> to "true".
52 : !>
53 : !> By calling executable [arg1] [arg2] [arg3] [arg4]
54 : !> one can define the size (arg1), the number of
55 : !> Eigenvectors to compute (arg2), and the blocking (arg3).
56 : !> If these values are not set default values (500, 150, 16)
57 : !> are choosen.
58 : !> If these values are set the 4th argument can be
59 : !> "output", which specifies that the EV's are written to
60 : !> an ascii file.
61 : !>
62 96 : program test_real_single_precision
63 :
64 : !-------------------------------------------------------------------------------
65 : ! Standard eigenvalue problem - REAL version
66 : !
67 : ! This program demonstrates the use of the ELPA module
68 : ! together with standard scalapack routines
69 : !
70 : ! Copyright of the original code rests with the authors inside the ELPA
71 : ! consortium. The copyright of any additional modifications shall rest
72 : ! with their original authors, but shall adhere to the licensing terms
73 : ! distributed along with the original code in the file "COPYING".
74 : !
75 : !-------------------------------------------------------------------------------
76 96 : use elpa1
77 : use elpa_utilities, only : error_unit
78 : use test_util
79 :
80 : use test_read_input_parameters
81 : use test_check_correctness
82 : use test_setup_mpi
83 : use test_blacs_infrastructure
84 : use test_prepare_matrix
85 :
86 : #ifdef HAVE_REDIRECT
87 : use test_redirect
88 : #endif
89 : use test_output_type
90 :
91 : implicit none
92 :
93 : !-------------------------------------------------------------------------------
94 : ! Please set system size parameters below!
95 : ! na: System size
96 : ! nev: Number of eigenvectors to be calculated
97 : ! nblk: Blocking factor in block cyclic distribution
98 : !-------------------------------------------------------------------------------
99 : integer(kind=ik) :: nblk
100 : integer(kind=ik) :: na, nev
101 :
102 : integer(kind=ik) :: np_rows, np_cols, na_rows, na_cols
103 :
104 : integer(kind=ik) :: myid, nprocs, my_prow, my_pcol, mpi_comm_rows, mpi_comm_cols
105 : integer(kind=ik) :: i, mpierr, my_blacs_ctxt, sc_desc(9), info, nprow, npcol
106 :
107 : integer(kind=ik), external :: numroc
108 :
109 96 : real(kind=rk4), allocatable :: a(:,:), z(:,:), as(:,:), ev(:)
110 :
111 : integer(kind=ik) :: STATUS
112 : #ifdef WITH_OPENMP
113 : integer(kind=ik) :: omp_get_max_threads, required_mpi_thread_level, &
114 : provided_mpi_thread_level
115 : #endif
116 : type(output_t) :: write_to_file
117 : logical :: success
118 : character(len=8) :: task_suffix
119 : integer(kind=ik) :: j
120 : !-------------------------------------------------------------------------------
121 :
122 : #undef DOUBLE_PRECISION_REAL
123 :
124 96 : success = .true.
125 :
126 96 : call read_input_parameters(na, nev, nblk, write_to_file)
127 :
128 : !-------------------------------------------------------------------------------
129 : ! MPI Initialization
130 96 : call setup_mpi(myid, nprocs)
131 :
132 96 : STATUS = 0
133 :
134 : #define REALCASE
135 : #define ELPA1
136 : #include "../../elpa_print_headers.F90"
137 :
138 96 : do np_cols = NINT(SQRT(REAL(nprocs))),2,-1
139 0 : if(mod(nprocs,np_cols) == 0 ) exit
140 : enddo
141 :
142 : ! at the end of the above loop, nprocs is always divisible by np_cols
143 :
144 96 : np_rows = nprocs/np_cols
145 :
146 96 : if(myid==0) then
147 64 : print *
148 64 : print '(a)','Standard eigenvalue problem - ELPA1, REAL version'
149 64 : print *
150 64 : print '((a,i0))', 'Matrix size: ', na
151 64 : print '((a,i0))', 'Num eigenvectors: ', nev
152 64 : print '((a,i0))', 'Blocksize: ', nblk
153 64 : print '((a,i0))', 'Num MPI proc: ', nprocs
154 64 : print '((a))', 'Using gpu: NO'
155 64 : print '((a,i0))', 'Num gpu devices: ', 0
156 64 : print '((a))', 'Number type: real'
157 64 : print '((a))', 'Number precision: single'
158 64 : print *
159 64 : print '(3(a,i0))','Number of processor rows=',np_rows,', cols=',np_cols,', total=',nprocs
160 64 : print *
161 : endif
162 :
163 : !-------------------------------------------------------------------------------
164 : ! Set up BLACS context and MPI communicators
165 : !
166 : ! The BLACS context is only necessary for using Scalapack.
167 : !
168 : ! For ELPA, the MPI communicators along rows/cols are sufficient,
169 : ! and the grid setup may be done in an arbitrary way as long as it is
170 : ! consistent (i.e. 0<=my_prow<np_rows, 0<=my_pcol<np_cols and every
171 : ! process has a unique (my_prow,my_pcol) pair).
172 :
173 : call set_up_blacsgrid(mpi_comm_world, np_rows, np_cols, 'C', &
174 96 : my_blacs_ctxt, my_prow, my_pcol)
175 :
176 96 : if (myid==0) then
177 64 : print '(a)','| Past BLACS_Gridinfo.'
178 : end if
179 :
180 : ! All ELPA routines need MPI communicators for communicating within
181 : ! rows or columns of processes, these are set in elpa_get_communicators.
182 :
183 : mpierr = elpa_get_communicators(mpi_comm_world, my_prow, my_pcol, &
184 96 : mpi_comm_rows, mpi_comm_cols)
185 :
186 96 : if (myid==0) then
187 64 : print '(a)','| Past split communicator setup for rows and columns.'
188 : end if
189 :
190 : call set_up_blacs_descriptor(na ,nblk, my_prow, my_pcol, np_rows, np_cols, &
191 96 : na_rows, na_cols, sc_desc, my_blacs_ctxt, info)
192 :
193 96 : if (myid==0) then
194 64 : print '(a)','| Past scalapack descriptor setup.'
195 : end if
196 :
197 : !-------------------------------------------------------------------------------
198 : ! Allocate matrices and set up a test matrix for the eigenvalue problem
199 96 : allocate(a (na_rows,na_cols))
200 96 : allocate(z (na_rows,na_cols))
201 96 : allocate(as(na_rows,na_cols))
202 :
203 96 : allocate(ev(na))
204 :
205 96 : call prepare_matrix_random(na, myid, sc_desc, a, z, as)
206 :
207 : !-------------------------------------------------------------------------------
208 : ! Calculate eigenvalues/eigenvectors
209 96 : elpa_print_times = .true.
210 96 : if (myid==0) then
211 64 : print '(a)','| Entering one-step ELPA solver ... '
212 64 : print *
213 : end if
214 : #ifdef WITH_MPI
215 64 : call mpi_barrier(mpi_comm_world, mpierr) ! for correct timings only
216 : #endif
217 : success = elpa_solve_evp_real_1stage_single(na, nev, a, na_rows, ev, z, na_rows, nblk, &
218 96 : na_cols, mpi_comm_rows, mpi_comm_cols, mpi_comm_world)
219 :
220 96 : if (.not.(success)) then
221 0 : write(error_unit,*) "solve_evp_real_1stage produced an error! Aborting..."
222 : #ifdef WITH_MPI
223 0 : call MPI_ABORT(mpi_comm_world, 1, mpierr)
224 : #else
225 0 : call exit(1)
226 : #endif
227 : endif
228 :
229 :
230 96 : if (myid==0) then
231 64 : print '(a)','| One-step ELPA solver complete.'
232 64 : print *
233 : end if
234 :
235 96 : if(myid == 0) print *,'Time tridiag_real :',time_evp_fwd
236 96 : if(myid == 0) print *,'Time solve_tridi :',time_evp_solve
237 96 : if(myid == 0) print *,'Time trans_ev_real :',time_evp_back
238 96 : if(myid == 0) print *,'Total time (sum above):',time_evp_back+time_evp_solve+time_evp_fwd
239 :
240 96 : if(write_to_file%eigenvectors) then
241 0 : write(unit = task_suffix, fmt = '(i8.8)') myid
242 0 : open(17,file="EVs_real_out_task_"//task_suffix(1:8)//".txt",form='formatted',status='new')
243 0 : write(17,*) "Part of eigenvectors: na_rows=",na_rows,"of na=",na," na_cols=",na_cols," of na=",na
244 :
245 0 : do i=1,na_rows
246 0 : do j=1,na_cols
247 0 : write(17,*) "row=",i," col=",j," element of eigenvector=",z(i,j)
248 : enddo
249 : enddo
250 0 : close(17)
251 : endif
252 :
253 96 : if(write_to_file%eigenvalues) then
254 0 : if (myid == 0) then
255 0 : open(17,file="Eigenvalues_real_out.txt",form='formatted',status='new')
256 0 : do i=1,na
257 0 : write(17,*) i,ev(i)
258 : enddo
259 0 : close(17)
260 : endif
261 : endif
262 :
263 :
264 : !-------------------------------------------------------------------------------
265 : ! Test correctness of result (using plain scalapack routines)
266 96 : status = check_correctness_evp_numeric_residuals(na, nev, as, z, ev, sc_desc, nblk, myid, np_rows, np_cols, my_prow, my_pcol)
267 :
268 96 : deallocate(a)
269 96 : deallocate(as)
270 96 : deallocate(z)
271 96 : deallocate(ev)
272 :
273 : #ifdef WITH_MPI
274 64 : call blacs_gridexit(my_blacs_ctxt)
275 64 : call mpi_finalize(mpierr)
276 : #endif
277 :
278 96 : call EXIT(STATUS)
279 :
280 :
281 : end
282 :
283 : !-------------------------------------------------------------------------------
|
