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 : !>
67 288 : program test_real2_default_kernel_double_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 : !-------------------------------------------------------------------------------
81 288 : use elpa1
82 : use elpa2
83 :
84 : use elpa_utilities, only : error_unit
85 : use elpa2_utilities
86 : use test_read_input_parameters
87 : use test_check_correctness
88 : use test_setup_mpi
89 : use test_blacs_infrastructure
90 : use test_prepare_matrix
91 : use test_util
92 :
93 : #ifdef HAVE_REDIRECT
94 : use test_redirect
95 : #endif
96 :
97 : use test_output_type
98 : implicit none
99 :
100 : !-------------------------------------------------------------------------------
101 : ! Please set system size parameters below!
102 : ! na: System size
103 : ! nev: Number of eigenvectors to be calculated
104 : ! nblk: Blocking factor in block cyclic distribution
105 : !-------------------------------------------------------------------------------
106 :
107 : integer(kind=ik) :: nblk
108 : integer(kind=ik) :: na, nev
109 :
110 : integer(kind=ik) :: np_rows, np_cols, na_rows, na_cols
111 :
112 : integer(kind=ik) :: myid, nprocs, my_prow, my_pcol, mpi_comm_rows, mpi_comm_cols
113 : integer(kind=ik) :: i, mpierr, my_blacs_ctxt, sc_desc(9), info, nprow, npcol
114 :
115 : integer(kind=ik), external :: numroc
116 :
117 288 : real(kind=rk8), allocatable :: a(:,:), z(:,:), as(:,:), ev(:)
118 :
119 : integer(kind=ik) :: STATUS
120 : #ifdef WITH_OPENMP
121 : integer(kind=ik) :: omp_get_max_threads, required_mpi_thread_level, provided_mpi_thread_level
122 : #endif
123 : logical :: successELPA, success
124 : logical :: gpuAvailable
125 : type(output_t) :: write_to_file
126 : character(len=8) :: task_suffix
127 : integer(kind=ik) :: j
128 :
129 : #define DOUBLE_PRECISION_REAL 1
130 :
131 288 : successELPA = .true.
132 288 : gpuAvailable = .false.
133 :
134 288 : call read_input_parameters(na, nev, nblk, write_to_file)
135 : !-------------------------------------------------------------------------------
136 : ! MPI Initialization
137 288 : call setup_mpi(myid, nprocs)
138 :
139 :
140 288 : STATUS = 0
141 :
142 : #define REALCASE
143 : #include "../../elpa_print_headers.F90"
144 :
145 : !-------------------------------------------------------------------------------
146 : ! Selection of number of processor rows/columns
147 : ! We try to set up the grid square-like, i.e. start the search for possible
148 : ! divisors of nprocs with a number next to the square root of nprocs
149 : ! and decrement it until a divisor is found.
150 :
151 288 : do np_cols = NINT(SQRT(REAL(nprocs))),2,-1
152 0 : if(mod(nprocs,np_cols) == 0 ) exit
153 : enddo
154 : ! at the end of the above loop, nprocs is always divisible by np_cols
155 :
156 288 : np_rows = nprocs/np_cols
157 :
158 288 : if(myid==0) then
159 192 : print *
160 192 : print '(a)','Standard eigenvalue problem - REAL version'
161 : #ifdef WITH_GPU_VERSION
162 : print *,"with GPU version"
163 : #endif
164 192 : print *
165 192 : print '(3(a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
166 192 : print '(3(a,i0))','Number of processor rows=',np_rows,', cols=',np_cols,', total=',nprocs
167 192 : print *
168 192 : print *, "This is an example how ELPA2 chooses a default kernel,"
169 : #ifdef HAVE_ENVIRONMENT_CHECKING
170 192 : print *, "or takes the kernel defined in the environment variable,"
171 : #endif
172 192 : print *, "since the ELPA API call does not contain any kernel specification"
173 192 : print *
174 192 : print *," "
175 : #ifndef HAVE_ENVIRONMENT_CHECKING
176 : print *, " Notice that it is not possible with this build to set the "
177 : print *, " kernel via an environment variable! To change this re-install"
178 : print *, " the library and have a look at the log files"
179 : #endif
180 : endif
181 :
182 : !-------------------------------------------------------------------------------
183 : ! Set up BLACS context and MPI communicators
184 : !
185 : ! The BLACS context is only necessary for using Scalapack.
186 : !
187 : ! For ELPA, the MPI communicators along rows/cols are sufficient,
188 : ! and the grid setup may be done in an arbitrary way as long as it is
189 : ! consistent (i.e. 0<=my_prow<np_rows, 0<=my_pcol<np_cols and every
190 : ! process has a unique (my_prow,my_pcol) pair).
191 :
192 : call set_up_blacsgrid(mpi_comm_world, np_rows, np_cols, 'C', &
193 288 : my_blacs_ctxt, my_prow, my_pcol)
194 :
195 288 : if (myid==0) then
196 192 : print '(a)','| Past BLACS_Gridinfo.'
197 : end if
198 :
199 : ! All ELPA routines need MPI communicators for communicating within
200 : ! rows or columns of processes, these are set in elpa_get_communicators.
201 :
202 : mpierr = elpa_get_communicators(mpi_comm_world, my_prow, my_pcol, &
203 288 : mpi_comm_rows, mpi_comm_cols)
204 :
205 288 : if (myid==0) then
206 192 : print '(a)','| Past split communicator setup for rows and columns.'
207 : end if
208 :
209 : call set_up_blacs_descriptor(na ,nblk, my_prow, my_pcol, np_rows, np_cols, &
210 288 : na_rows, na_cols, sc_desc, my_blacs_ctxt, info)
211 :
212 288 : if (myid==0) then
213 192 : print '(a)','| Past scalapack descriptor setup.'
214 : end if
215 :
216 : !-------------------------------------------------------------------------------
217 : ! Allocate matrices and set up a test matrix for the eigenvalue problem
218 288 : allocate(a (na_rows,na_cols))
219 288 : allocate(z (na_rows,na_cols))
220 288 : allocate(as(na_rows,na_cols))
221 :
222 288 : allocate(ev(na))
223 :
224 288 : call prepare_matrix_random(na, myid, sc_desc, a, z, as)
225 :
226 : ! set print flag in elpa1
227 288 : elpa_print_times = .true.
228 :
229 : !-------------------------------------------------------------------------------
230 : ! Calculate eigenvalues/eigenvectors
231 :
232 288 : if (myid==0) then
233 192 : print '(a)','| Entering two-stage ELPA solver ... '
234 192 : print *
235 : end if
236 :
237 :
238 : ! ELPA is called without any kernel specification in the API,
239 : ! furthermore, if the environment variable is not set, the
240 : ! default kernel is called. Otherwise, the kernel defined in the
241 : ! environment variable
242 : #ifdef WITH_MPI
243 192 : call mpi_barrier(mpi_comm_world, mpierr) ! for correct timings only
244 : #endif
245 : successELPA = elpa_solve_evp_real_2stage_double(na, nev, a, na_rows, ev, z, na_rows, nblk, &
246 288 : na_cols, mpi_comm_rows, mpi_comm_cols, mpi_comm_world)
247 :
248 288 : if (.not.(successELPA)) then
249 0 : write(error_unit,*) "solve_evp_real_2stage produced an error! Aborting..."
250 : #ifdef WITH_MPI
251 0 : call MPI_ABORT(mpi_comm_world, 1, mpierr)
252 : #else
253 0 : call exit(1)
254 : #endif
255 : endif
256 :
257 288 : if (myid==0) then
258 192 : print '(a)','| Two-step ELPA solver complete.'
259 192 : print *
260 : end if
261 :
262 288 : if(myid == 0) print *,'Time transform to tridi :',time_evp_fwd
263 288 : if(myid == 0) print *,'Time solve tridi :',time_evp_solve
264 288 : if(myid == 0) print *,'Time transform back EVs :',time_evp_back
265 288 : if(myid == 0) print *,'Total time (sum above) :',time_evp_back+time_evp_solve+time_evp_fwd
266 :
267 288 : if(write_to_file%eigenvectors) then
268 0 : write(unit = task_suffix, fmt = '(i8.8)') myid
269 0 : open(17,file="EVs_real2_out_task_"//task_suffix(1:8)//".txt",form='formatted',status='new')
270 0 : write(17,*) "Part of eigenvectors: na_rows=",na_rows,"of na=",na," na_cols=",na_cols," of na=",na
271 :
272 0 : do i=1,na_rows
273 0 : do j=1,na_cols
274 0 : write(17,*) "row=",i," col=",j," element of eigenvector=",z(i,j)
275 : enddo
276 : enddo
277 0 : close(17)
278 : endif
279 288 : if(write_to_file%eigenvalues) then
280 0 : if (myid == 0) then
281 0 : open(17,file="Eigenvalues_real2_out.txt",form='formatted',status='new')
282 0 : do i=1,na
283 0 : write(17,*) i,ev(i)
284 : enddo
285 0 : close(17)
286 : endif
287 : endif
288 :
289 :
290 : !-------------------------------------------------------------------------------
291 : ! Test correctness of result (using plain scalapack routines)
292 :
293 288 : status = check_correctness_evp_numeric_residuals(na, nev, as, z, ev, sc_desc, nblk, myid, np_rows, np_cols, my_prow, my_pcol)
294 :
295 288 : deallocate(a)
296 288 : deallocate(as)
297 :
298 288 : deallocate(z)
299 288 : deallocate(ev)
300 :
301 : #ifdef WITH_MPI
302 192 : call blacs_gridexit(my_blacs_ctxt)
303 192 : call mpi_finalize(mpierr)
304 : #endif
305 288 : call EXIT(STATUS)
306 : end
307 :
308 : !-------------------------------------------------------------------------------
|