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 2 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 : !> The real ELPA 2 kernel is set as the default kernel.
63 : !> However, this can be overriden by setting
64 : !> the environment variable "REAL_ELPA_KERNEL" to an
65 : !> appropiate value.
66 : !>
67 144 : program test_real2_single_precision
68 :
69 : !-------------------------------------------------------------------------------
70 : ! Standard eigenvalue problem - REAL version
71 : !
72 : ! This program demonstrates the use of the ELPA module
73 : ! together with standard scalapack routines
74 : !
75 : ! Copyright of the original code rests with the authors inside the ELPA
76 : ! consortium. The copyright of any additional modifications shall rest
77 : ! with their original authors, but shall adhere to the licensing terms
78 : ! distributed along with the original code in the file "COPYING".
79 : !
80 : !-------------------------------------------------------------------------------
81 144 : use elpa1
82 : use elpa2
83 :
84 : use elpa_utilities, only : error_unit
85 : use test_util
86 :
87 : use test_read_input_parameters
88 : use test_check_correctness
89 : use test_setup_mpi
90 : use test_blacs_infrastructure
91 : use test_prepare_matrix
92 : #ifdef HAVE_REDIRECT
93 : use test_redirect
94 : #endif
95 : use test_output_type
96 : implicit none
97 :
98 : !-------------------------------------------------------------------------------
99 : ! Please set system size parameters below!
100 : ! na: System size
101 : ! nev: Number of eigenvectors to be calculated
102 : ! nblk: Blocking factor in block cyclic distribution
103 : !-------------------------------------------------------------------------------
104 :
105 : integer(kind=ik) :: nblk
106 : integer(kind=ik) :: na, nev
107 :
108 : integer(kind=ik) :: np_rows, np_cols, na_rows, na_cols
109 :
110 : integer(kind=ik) :: myid, nprocs, my_prow, my_pcol, mpi_comm_rows, mpi_comm_cols
111 : integer(kind=ik) :: i, mpierr, my_blacs_ctxt, sc_desc(9), info, nprow, npcol
112 :
113 : integer(kind=ik), external :: numroc
114 :
115 144 : real(kind=rk4), allocatable :: a(:,:), z(:,:), as(:,:), ev(:)
116 :
117 : integer(kind=ik) :: STATUS
118 : #ifdef WITH_OPENMP
119 : integer(kind=ik) :: omp_get_max_threads, required_mpi_thread_level, provided_mpi_thread_level
120 : #endif
121 : logical :: successELPA, success
122 : logical :: gpuAvailable
123 : type(output_t) :: write_to_file
124 : character(len=8) :: task_suffix
125 : integer(kind=ik) :: j
126 :
127 : #undef DOUBLE_PRECISION_REAL
128 :
129 144 : successELPA = .true.
130 144 : gpuAvailable = .false.
131 :
132 144 : call read_input_parameters(na, nev, nblk, write_to_file)
133 :
134 : !-------------------------------------------------------------------------------
135 : ! MPI Initialization
136 144 : call setup_mpi(myid, nprocs)
137 :
138 :
139 144 : STATUS = 0
140 :
141 : #define REALCASE
142 : #include "../../elpa_print_headers.F90"
143 :
144 : !-------------------------------------------------------------------------------
145 : ! Selection of number of processor rows/columns
146 : ! We try to set up the grid square-like, i.e. start the search for possible
147 : ! divisors of nprocs with a number next to the square root of nprocs
148 : ! and decrement it until a divisor is found.
149 :
150 144 : do np_cols = NINT(SQRT(REAL(nprocs))),2,-1
151 0 : if(mod(nprocs,np_cols) == 0 ) exit
152 : enddo
153 : ! at the end of the above loop, nprocs is always divisible by np_cols
154 :
155 144 : np_rows = nprocs/np_cols
156 :
157 144 : if(myid==0) then
158 96 : print *
159 96 : print '(a)','Standard eigenvalue problem - REAL version'
160 96 : print *
161 96 : print '(3(a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
162 96 : print '(3(a,i0))','Number of processor rows=',np_rows,', cols=',np_cols,', total=',nprocs
163 96 : print *
164 : endif
165 :
166 : !-------------------------------------------------------------------------------
167 : ! Set up BLACS context and MPI communicators
168 : !
169 : ! The BLACS context is only necessary for using Scalapack.
170 : !
171 : ! For ELPA, the MPI communicators along rows/cols are sufficient,
172 : ! and the grid setup may be done in an arbitrary way as long as it is
173 : ! consistent (i.e. 0<=my_prow<np_rows, 0<=my_pcol<np_cols and every
174 : ! process has a unique (my_prow,my_pcol) pair).
175 :
176 : call set_up_blacsgrid(mpi_comm_world, np_rows, np_cols, 'C', &
177 144 : my_blacs_ctxt, my_prow, my_pcol)
178 :
179 144 : if (myid==0) then
180 96 : print '(a)','| Past BLACS_Gridinfo.'
181 : end if
182 :
183 : ! All ELPA routines need MPI communicators for communicating within
184 : ! rows or columns of processes, these are set in elpa_get_communicators.
185 :
186 : mpierr = elpa_get_communicators(mpi_comm_world, my_prow, my_pcol, &
187 144 : mpi_comm_rows, mpi_comm_cols)
188 :
189 144 : if (myid==0) then
190 96 : print '(a)','| Past split communicator setup for rows and columns.'
191 : end if
192 :
193 : call set_up_blacs_descriptor(na ,nblk, my_prow, my_pcol, np_rows, np_cols, &
194 144 : na_rows, na_cols, sc_desc, my_blacs_ctxt, info)
195 :
196 144 : if (myid==0) then
197 96 : print '(a)','| Past scalapack descriptor setup.'
198 : end if
199 :
200 : !-------------------------------------------------------------------------------
201 : ! Allocate matrices and set up a test matrix for the eigenvalue problem
202 144 : allocate(a (na_rows,na_cols))
203 144 : allocate(z (na_rows,na_cols))
204 144 : allocate(as(na_rows,na_cols))
205 :
206 144 : allocate(ev(na))
207 :
208 144 : call prepare_matrix_random(na, myid, sc_desc, a, z, as)
209 :
210 : ! set print flag in elpa1
211 144 : elpa_print_times = .true.
212 :
213 : !-------------------------------------------------------------------------------
214 : ! Calculate eigenvalues/eigenvectors
215 :
216 144 : if (myid==0) then
217 96 : print '(a)','| Entering two-stage ELPA solver ... '
218 96 : print *
219 : end if
220 : #ifdef WITH_MPI
221 96 : call mpi_barrier(mpi_comm_world, mpierr) ! for correct timings only
222 : #endif
223 : successELPA = elpa_solve_evp_real_2stage_single(na, nev, a, na_rows, ev, z, na_rows, nblk, na_cols, &
224 144 : mpi_comm_rows, mpi_comm_cols, mpi_comm_world)
225 :
226 144 : if (.not.(successELPA)) then
227 0 : write(error_unit,*) "solve_evp_real_2stage produced an error! Aborting..."
228 : #ifdef WITH_MPI
229 0 : call MPI_ABORT(mpi_comm_world, 1, mpierr)
230 : #else
231 0 : call exit(1)
232 : #endif
233 : endif
234 :
235 144 : if (myid==0) then
236 96 : print '(a)','| Two-step ELPA solver complete.'
237 96 : print *
238 : end if
239 :
240 144 : if(myid == 0) print *,'Time transform to tridi :',time_evp_fwd
241 144 : if(myid == 0) print *,'Time solve tridi :',time_evp_solve
242 144 : if(myid == 0) print *,'Time transform back EVs :',time_evp_back
243 144 : if(myid == 0) print *,'Total time (sum above) :',time_evp_back+time_evp_solve+time_evp_fwd
244 :
245 :
246 144 : if(write_to_file%eigenvectors) then
247 0 : write(unit = task_suffix, fmt = '(i8.8)') myid
248 0 : open(17,file="EVs_real2_out_task_"//task_suffix(1:8)//".txt",form='formatted',status='new')
249 0 : write(17,*) "Part of eigenvectors: na_rows=",na_rows,"of na=",na," na_cols=",na_cols," of na=",na
250 :
251 0 : do i=1,na_rows
252 0 : do j=1,na_cols
253 0 : write(17,*) "row=",i," col=",j," element of eigenvector=",z(i,j)
254 : enddo
255 : enddo
256 0 : close(17)
257 : endif
258 :
259 144 : if(write_to_file%eigenvalues) then
260 0 : if (myid == 0) then
261 0 : open(17,file="Eigenvalues_real2_out.txt",form='formatted',status='new')
262 0 : do i=1,na
263 0 : write(17,*) i,ev(i)
264 : enddo
265 0 : close(17)
266 : endif
267 : endif
268 :
269 :
270 : !-------------------------------------------------------------------------------
271 : ! Test correctness of result (using plain scalapack routines)
272 144 : status = check_correctness_evp_numeric_residuals(na, nev, as, z, ev, sc_desc, nblk, myid, np_rows, np_cols, my_prow, my_pcol)
273 :
274 144 : deallocate(a)
275 144 : deallocate(as)
276 144 : deallocate(z)
277 144 : deallocate(ev)
278 :
279 : #ifdef WITH_MPI
280 96 : call blacs_gridexit(my_blacs_ctxt)
281 96 : call mpi_finalize(mpierr)
282 : #endif
283 144 : call EXIT(STATUS)
284 : end
285 :
286 : !-------------------------------------------------------------------------------
|