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 complex ELPA 2 kernel is set in this program via
63 : !> the API call. 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_choose_kernel_with_api_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 :
85 : use elpa_utilities, only : error_unit
86 : use elpa2_utilities
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 : use test_util
93 :
94 : #ifdef HAVE_REDIRECT
95 : use test_redirect
96 : #endif
97 :
98 : use test_output_type
99 : implicit none
100 :
101 : !-------------------------------------------------------------------------------
102 : ! Please set system size parameters below!
103 : ! na: System size
104 : ! nev: Number of eigenvectors to be calculated
105 : ! nblk: Blocking factor in block cyclic distribution
106 : !-------------------------------------------------------------------------------
107 :
108 : integer(kind=ik) :: nblk
109 : integer(kind=ik) :: na, nev
110 :
111 : integer(kind=ik) :: np_rows, np_cols, na_rows, na_cols
112 :
113 : integer(kind=ik) :: myid, nprocs, my_prow, my_pcol, mpi_comm_rows, mpi_comm_cols
114 : integer(kind=ik) :: i, mpierr, my_blacs_ctxt, sc_desc(9), info, nprow, npcol
115 :
116 : integer(kind=ik), external :: numroc
117 :
118 144 : real(kind=rk4), allocatable :: a(:,:), z(:,:), as(:,:), ev(:)
119 :
120 : integer(kind=ik) :: STATUS
121 : #ifdef WITH_OPENMP
122 : integer(kind=ik) :: omp_get_max_threads, required_mpi_thread_level, provided_mpi_thread_level
123 : #endif
124 : logical :: successELPA, success
125 : logical :: gpuAvailable
126 : type(output_t) :: write_to_file
127 : character(len=8) :: task_suffix
128 : integer(kind=ik) :: j
129 :
130 : #undef DOUBLE_PRECISION_REAL
131 :
132 144 : successELPA = .true.
133 144 : gpuAvailable = .false.
134 :
135 144 : call read_input_parameters(na, nev, nblk, write_to_file)
136 :
137 : !-------------------------------------------------------------------------------
138 : ! MPI Initialization
139 144 : call setup_mpi(myid, nprocs)
140 :
141 :
142 144 : STATUS = 0
143 :
144 : #define REALCASE
145 : #include "../../elpa_print_headers.F90"
146 :
147 : !-------------------------------------------------------------------------------
148 : ! Selection of number of processor rows/columns
149 : ! We try to set up the grid square-like, i.e. start the search for possible
150 : ! divisors of nprocs with a number next to the square root of nprocs
151 : ! and decrement it until a divisor is found.
152 :
153 144 : do np_cols = NINT(SQRT(REAL(nprocs))),2,-1
154 0 : if(mod(nprocs,np_cols) == 0 ) exit
155 : enddo
156 : ! at the end of the above loop, nprocs is always divisible by np_cols
157 :
158 144 : np_rows = nprocs/np_cols
159 :
160 144 : if(myid==0) then
161 96 : print *
162 96 : print '(a)','Standard eigenvalue problem - REAL version'
163 : #ifdef WITH_GPU_VERSION
164 : print *,"with GPU Version"
165 : #endif
166 96 : print *
167 96 : print '(3(a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
168 96 : print '(3(a,i0))','Number of processor rows=',np_rows,', cols=',np_cols,', total=',nprocs
169 96 : print *
170 96 : print *, "This is an example how to determine the ELPA2 kernel with"
171 96 : print *, "an api call. Note, however, that setting the kernel via"
172 96 : print *, "an environment variable will always take precedence over"
173 96 : print *, "everything else! "
174 96 : print *
175 96 : print *," "
176 : #ifndef HAVE_ENVIRONMENT_CHECKING
177 : print *, " Notice that it is not possible with this build to set the "
178 : print *, " kernel via an environment variable! To change this re-install"
179 : print *, " the library and have a look at the log files"
180 : #endif
181 :
182 : #ifndef WITH_FIXED_REAL_KERNEL
183 96 : print *, " The settings are: REAL_ELPA_KERNEL_GENERIC_SIMPLE"
184 : #else /* WITH_FIXED_REAL_KERNEL */
185 :
186 : #ifdef WITH_REAL_GENERIC_KERNEL
187 : print *, " The settings are: REAL_ELPA_KERNEL_GENERIC"
188 : #endif
189 : #ifdef WITH_REAL_GENERIC_SIMPLE_KERNEL
190 : print *, " The settings are: REAL_ELPA_KERNEL_GENERIC_SIMPLE"
191 : #endif
192 : #ifdef WITH_REAL_GENERIC_SSE_ASSEMBLY_KERNEL
193 : print *, " The settings are: REAL_ELPA_KERNEL_SSE"
194 : #endif
195 : #ifdef WITH_REAL_SSE_BLOCK2_KERNEL
196 : print *, " The settings are: REAL_ELPA_KERNEL_SSE_BLOCK2"
197 : #endif
198 : #ifdef WITH_REAL_SSE_BLOCK4_KERNEL
199 : print *, " The settings are: REAL_ELPA_KERNEL_SSE_BLOCK4"
200 : #endif
201 : #ifdef WITH_REAL_SSE_BLOCK6_KERNEL
202 : print *, " The settings are: REAL_ELPA_KERNEL_SSE_BLOCK6"
203 : #endif
204 : #ifdef WITH_REAL_AVX_BLOCK2_KERNEL
205 : print *, " The settings are: REAL_ELPA_KERNEL_AVX_BLOCK2"
206 : #endif
207 : #ifdef WITH_REAL_AVX_BLOCK4_KERNEL
208 : print *, " The settings are: REAL_ELPA_KERNEL_AVX_BLOCK4"
209 : #endif
210 : #ifdef WITH_REAL_AVX_BLOCK6_KERNEL
211 : print *, " The settings are: REAL_ELPA_KERNEL_AVX_BLOCK6"
212 : #endif
213 : #ifdef WITH_REAL_AVX2_BLOCK2_KERNEL
214 : print *, " The settings are: REAL_ELPA_KERNEL_AVX2_BLOCK2"
215 : #endif
216 : #ifdef WITH_REAL_AVX2_BLOCK4_KERNEL
217 : print *, " The settings are: REAL_ELPA_KERNEL_AVX2_BLOCK4"
218 : #endif
219 : #ifdef WITH_REAL_AVX2_BLOCK6_KERNEL
220 : print *, " The settings are: REAL_ELPA_KERNEL_AVX2_BLOCK6"
221 : #endif
222 : #ifdef WITH_REAL_AVX512_BLOCK2_KERNEL
223 : print *, " The settings are: REAL_ELPA_KERNEL_AVX512_BLOCK2"
224 : #endif
225 : #ifdef WITH_REAL_AVX512_BLOCK4_KERNEL
226 : print *, " The settings are: REAL_ELPA_KERNEL_AVX512_BLOCK4"
227 : #endif
228 : #ifdef WITH_REAL_AVX512_BLOCK6_KERNEL
229 : print *, " The settings are: REAL_ELPA_KERNEL_AVX512_BLOCK6"
230 : #endif
231 : #ifdef WITH_REAL_BGP_KERNEL
232 : print *, " The settings are: REAL_ELPA_KERNEL_BGP"
233 : #endif
234 : #ifdef WITH_REAL_BGQ_KERNEL
235 : print *, " The settings are: REAL_ELPA_KERNEL_BGQ"
236 : #endif
237 : #ifdef WITH_GPU_VERSION
238 : print *, " The settings are: REAL_ELPA_GPU"
239 : #endif
240 :
241 : #endif /* WITH_FIXED_REAL_KERNEL */
242 :
243 96 : print *
244 :
245 :
246 : endif
247 :
248 : !-------------------------------------------------------------------------------
249 : ! Set up BLACS context and MPI communicators
250 : !
251 : ! The BLACS context is only necessary for using Scalapack.
252 : !
253 : ! For ELPA, the MPI communicators along rows/cols are sufficient,
254 : ! and the grid setup may be done in an arbitrary way as long as it is
255 : ! consistent (i.e. 0<=my_prow<np_rows, 0<=my_pcol<np_cols and every
256 : ! process has a unique (my_prow,my_pcol) pair).
257 :
258 : call set_up_blacsgrid(mpi_comm_world, np_rows, np_cols, 'C', &
259 144 : my_blacs_ctxt, my_prow, my_pcol)
260 :
261 144 : if (myid==0) then
262 96 : print '(a)','| Past BLACS_Gridinfo.'
263 : end if
264 :
265 : ! All ELPA routines need MPI communicators for communicating within
266 : ! rows or columns of processes, these are set in elpa_get_communicators.
267 :
268 : mpierr = elpa_get_communicators(mpi_comm_world, my_prow, my_pcol, &
269 144 : mpi_comm_rows, mpi_comm_cols)
270 :
271 144 : if (myid==0) then
272 96 : print '(a)','| Past split communicator setup for rows and columns.'
273 : end if
274 :
275 : call set_up_blacs_descriptor(na ,nblk, my_prow, my_pcol, np_rows, np_cols, &
276 144 : na_rows, na_cols, sc_desc, my_blacs_ctxt, info)
277 :
278 144 : if (myid==0) then
279 96 : print '(a)','| Past scalapack descriptor setup.'
280 : end if
281 :
282 : !-------------------------------------------------------------------------------
283 : ! Allocate matrices and set up a test matrix for the eigenvalue problem
284 144 : allocate(a (na_rows,na_cols))
285 144 : allocate(z (na_rows,na_cols))
286 144 : allocate(as(na_rows,na_cols))
287 :
288 144 : allocate(ev(na))
289 :
290 144 : call prepare_matrix_random(na, myid, sc_desc, a, z, as)
291 :
292 : ! set print flag in elpa1
293 144 : elpa_print_times = .true.
294 :
295 : !-------------------------------------------------------------------------------
296 : ! Calculate eigenvalues/eigenvectors
297 :
298 144 : if (myid==0) then
299 96 : print '(a)','| Entering two-stage ELPA solver ... '
300 96 : print *
301 : end if
302 :
303 :
304 : ! ELPA is called with a kernel specification in the API
305 : #ifdef WITH_MPI
306 96 : call mpi_barrier(mpi_comm_world, mpierr) ! for correct timings only
307 : #endif
308 : successELPA = elpa_solve_evp_real_2stage_single(na, nev, a, na_rows, ev, z, na_rows, nblk, &
309 : na_cols, mpi_comm_rows, mpi_comm_cols, mpi_comm_world, &
310 : #ifndef WITH_FIXED_REAL_KERNEL
311 144 : REAL_ELPA_KERNEL_GENERIC_SIMPLE)
312 : #else /* WITH_FIXED_REAL_KERNEL */
313 :
314 : #ifdef WITH_REAL_GENERIC_KERNEL
315 : REAL_ELPA_KERNEL_GENERIC)
316 : #endif
317 :
318 : #ifdef WITH_REAL_GENERIC_SIMPLE_KERNEL
319 : REAL_ELPA_KERNEL_GENERIC_SIMPLE)
320 : #endif
321 :
322 : #ifdef WITH_REAL_SSE_ASSEMBLY_KERNEL
323 : REAL_ELPA_KERNEL_SSE)
324 : #endif
325 :
326 : #ifdef WITH_REAL_SSE_BLOCK6_KERNEL
327 : REAL_ELPA_KERNEL_SSE_BLOCK6)
328 : #else
329 : #ifdef WITH_REAL_SSE_BLOCK4_KERNEL
330 : REAL_ELPA_KERNEL_SSE_BLOCK4)
331 : #else
332 : #ifdef WITH_REAL_SSE_BLOCK2_KERNEL
333 : REAL_ELPA_KERNEL_SSE_BLOCK2)
334 : #endif
335 : #endif
336 : #endif
337 :
338 : #ifdef WITH_REAL_AVX_BLOCK6_KERNEL
339 : REAL_ELPA_KERNEL_AVX_BLOCK6)
340 : #else
341 : #ifdef WITH_REAL_AVX_BLOCK4_KERNEL
342 : REAL_ELPA_KERNEL_AVX_BLOCK4)
343 : #else
344 : #ifdef WITH_REAL_AVX_BLOCK2_KERNEL
345 : REAL_ELPA_KERNEL_AVX_BLOCK2)
346 : #endif
347 : #endif
348 : #endif
349 :
350 : #ifdef WITH_REAL_AVX2_BLOCK6_KERNEL
351 : REAL_ELPA_KERNEL_AVX2_BLOCK6)
352 : #else
353 : #ifdef WITH_REAL_AVX2_BLOCK4_KERNEL
354 : REAL_ELPA_KERNEL_AVX2_BLOCK4)
355 : #else
356 : #ifdef WITH_REAL_AVX2_BLOCK2_KERNEL
357 : REAL_ELPA_KERNEL_AVX2_BLOCK2)
358 : #endif
359 : #endif
360 : #endif
361 :
362 : #ifdef WITH_REAL_AVX512_BLOCK6_KERNEL
363 : REAL_ELPA_KERNEL_AVX512_BLOCK6)
364 : #else
365 : #ifdef WITH_REAL_AVX512_BLOCK4_KERNEL
366 : REAL_ELPA_KERNEL_AVX512_BLOCK4)
367 : #else
368 : #ifdef WITH_REAL_AVX512_BLOCK2_KERNEL
369 : REAL_ELPA_KERNEL_AVX512_BLOCK2)
370 : #endif
371 : #endif
372 : #endif
373 : #ifdef WITH_REAL_BGP_KERNEL
374 : REAL_ELPA_KERNEL_BGP)
375 : #endif
376 :
377 : #ifdef WITH_REAL_BGQ_KERNEL
378 : REAL_ELPA_KERNEL_BGQ)
379 : #endif
380 :
381 : #ifdef WITH_GPU_VERSION
382 : REAL_ELPA_KERNEL_GPU)
383 : #endif
384 :
385 : #endif /* WITH_FIXED_REAL_KERNEL */
386 :
387 144 : if (.not.(successELPA)) then
388 0 : write(error_unit,*) "solve_evp_real_2stage produced an error! Aborting..."
389 : #ifdef WITH_MPI
390 0 : call MPI_ABORT(mpi_comm_world, 1, mpierr)
391 : #else
392 0 : call exit(1)
393 : #endif
394 : endif
395 :
396 144 : if (myid==0) then
397 96 : print '(a)','| Two-step ELPA solver complete.'
398 96 : print *
399 : end if
400 :
401 144 : if(myid == 0) print *,'Time transform to tridi :',time_evp_fwd
402 144 : if(myid == 0) print *,'Time solve tridi :',time_evp_solve
403 144 : if(myid == 0) print *,'Time transform back EVs :',time_evp_back
404 144 : if(myid == 0) print *,'Total time (sum above) :',time_evp_back+time_evp_solve+time_evp_fwd
405 :
406 144 : if(write_to_file%eigenvectors) then
407 0 : write(unit = task_suffix, fmt = '(i8.8)') myid
408 0 : open(17,file="EVs_real2_out_task_"//task_suffix(1:8)//".txt",form='formatted',status='new')
409 0 : write(17,*) "Part of eigenvectors: na_rows=",na_rows,"of na=",na," na_cols=",na_cols," of na=",na
410 :
411 0 : do i=1,na_rows
412 0 : do j=1,na_cols
413 0 : write(17,*) "row=",i," col=",j," element of eigenvector=",z(i,j)
414 : enddo
415 : enddo
416 0 : close(17)
417 : endif
418 :
419 144 : if(write_to_file%eigenvalues) then
420 0 : if (myid == 0) then
421 0 : open(17,file="Eigenvalues_real2_out.txt",form='formatted',status='new')
422 0 : do i=1,na
423 0 : write(17,*) i,ev(i)
424 : enddo
425 0 : close(17)
426 : endif
427 : endif
428 :
429 : !-------------------------------------------------------------------------------
430 : ! Test correctness of result (using plain scalapack routines)
431 144 : status = check_correctness_evp_numeric_residuals(na, nev, as, z, ev, sc_desc, nblk, myid, np_rows, np_cols, my_prow, my_pcol)
432 :
433 144 : deallocate(a)
434 144 : deallocate(as)
435 144 : deallocate(z)
436 144 : deallocate(ev)
437 :
438 : #ifdef WITH_MPI
439 96 : call blacs_gridexit(my_blacs_ctxt)
440 96 : call mpi_finalize(mpierr)
441 : #endif
442 144 : call EXIT(STATUS)
443 : end
444 :
445 : !-------------------------------------------------------------------------------
|