Line data Source code
1 : !--------------------------------------------------------------------------------------------------!
2 : ! CP2K: A general program to perform molecular dynamics simulations !
3 : ! Copyright 2000-2026 CP2K developers group <https://cp2k.org> !
4 : ! !
5 : ! SPDX-License-Identifier: GPL-2.0-or-later !
6 : !--------------------------------------------------------------------------------------------------!
7 :
8 : ! **************************************************************************************************
9 : MODULE cp2k_runs
10 : USE atom, ONLY: atom_code
11 : USE bibliography, ONLY: Iannuzzi2026,&
12 : cite_reference,&
13 : cp2kqs2020
14 : USE bsse, ONLY: do_bsse_calculation
15 : USE cell_opt, ONLY: cp_cell_opt
16 : USE cp2k_debug, ONLY: cp2k_debug_energy_and_forces
17 : USE cp2k_info, ONLY: compile_date,&
18 : compile_revision,&
19 : cp2k_version,&
20 : cp2k_year
21 : USE cp_control_types, ONLY: dft_control_type
22 : USE cp_dbcsr_api, ONLY: dbcsr_finalize_lib,&
23 : dbcsr_print_config,&
24 : dbcsr_print_statistics
25 : USE cp_dbcsr_cp2k_link, ONLY: cp_dbcsr_config
26 : USE cp_files, ONLY: close_file,&
27 : open_file
28 : USE cp_log_handling, ONLY: cp_get_default_logger,&
29 : cp_logger_get_default_io_unit,&
30 : cp_logger_type,&
31 : cp_logger_would_log,&
32 : cp_note_level
33 : USE cp_output_handling, ONLY: cp_add_iter_level,&
34 : cp_print_key_finished_output,&
35 : cp_print_key_unit_nr,&
36 : cp_rm_iter_level
37 : USE cp_parser_methods, ONLY: parser_search_string
38 : USE cp_parser_types, ONLY: cp_parser_type,&
39 : parser_create,&
40 : parser_release
41 : USE cp_units, ONLY: cp_unit_set_create,&
42 : cp_unit_set_release,&
43 : cp_unit_set_type,&
44 : export_units_as_xml
45 : USE dbm_api, ONLY: dbm_library_print_stats
46 : USE dbt_api, ONLY: dbt_library_print_stats
47 : USE environment, ONLY: cp2k_finalize,&
48 : cp2k_init,&
49 : cp2k_read,&
50 : cp2k_setup
51 : USE f77_interface, ONLY: configure_dbcsr,&
52 : create_force_env,&
53 : destroy_force_env,&
54 : f77_default_para_env => default_para_env,&
55 : f_env_add_defaults,&
56 : f_env_rm_defaults,&
57 : f_env_type
58 : USE farming_methods, ONLY: do_deadlock,&
59 : do_nothing,&
60 : do_wait,&
61 : farming_parse_input,&
62 : get_next_job
63 : USE farming_types, ONLY: deallocate_farming_env,&
64 : farming_env_type,&
65 : init_farming_env,&
66 : job_finished,&
67 : job_running
68 : USE force_env_methods, ONLY: force_env_calc_energy_force
69 : USE force_env_types, ONLY: force_env_get,&
70 : force_env_type
71 : USE geo_opt, ONLY: cp_geo_opt
72 : USE global_types, ONLY: global_environment_type,&
73 : globenv_create,&
74 : globenv_release
75 : USE grid_api, ONLY: grid_library_print_stats,&
76 : grid_library_set_config
77 : USE input_constants, ONLY: &
78 : bsse_run, cell_opt_run, debug_run, do_atom, do_band, do_cp2k, do_embed, do_farming, &
79 : do_fist, do_ipi, do_mixed, do_nnp, do_opt_basis, do_optimize_input, do_qmmm, do_qs, &
80 : do_sirius, do_swarm, do_tamc, do_test, do_tree_mc, do_tree_mc_ana, driver_run, ehrenfest, &
81 : energy_force_run, energy_run, geo_opt_run, history_guess, linear_response_run, mimic_run, &
82 : mol_dyn_run, mon_car_run, mtlr_run, negf_run, none_run, pint_run, real_time_propagation, &
83 : rtp_method_bse, rtp_method_bse_linearized, tree_mc_run, vib_anal
84 : USE input_cp2k, ONLY: create_cp2k_root_section
85 : USE input_cp2k_check, ONLY: check_cp2k_input
86 : USE input_cp2k_global, ONLY: create_global_section
87 : USE input_cp2k_read, ONLY: read_input
88 : USE input_keyword_types, ONLY: keyword_release
89 : USE input_parsing, ONLY: section_vals_parse
90 : USE input_section_types, ONLY: &
91 : section_release, section_type, section_vals_create, section_vals_get, &
92 : section_vals_get_subs_vals, section_vals_release, section_vals_retain, section_vals_type, &
93 : section_vals_val_get, section_vals_write, write_section_xml
94 : USE ipi_driver, ONLY: run_driver
95 : USE kinds, ONLY: default_path_length,&
96 : default_string_length,&
97 : dp,&
98 : int_8
99 : USE library_tests, ONLY: lib_test
100 : USE machine, ONLY: default_output_unit,&
101 : m_chdir,&
102 : m_flush,&
103 : m_getcwd,&
104 : m_memory,&
105 : m_memory_max,&
106 : m_walltime
107 : USE mc_run, ONLY: do_mon_car
108 : USE md_run, ONLY: qs_mol_dyn
109 : USE message_passing, ONLY: mp_any_source,&
110 : mp_comm_type,&
111 : mp_para_env_release,&
112 : mp_para_env_type
113 : USE mimic_loop, ONLY: do_mimic_loop
114 : USE mscfg_methods, ONLY: do_mol_loop,&
115 : loop_over_molecules
116 : USE mtlr_u_j_methods, ONLY: do_mtlr_u_j
117 : USE neb_methods, ONLY: neb
118 : USE negf_methods, ONLY: do_negf
119 : USE offload_api, ONLY: offload_mempool_stats_print
120 : USE optimize_basis, ONLY: run_optimize_basis
121 : USE optimize_input, ONLY: run_optimize_input
122 : USE pint_methods, ONLY: do_pint_run
123 : USE qs_environment_types, ONLY: get_qs_env
124 : USE qs_linres_module, ONLY: linres_calculation
125 : USE reference_manager, ONLY: export_references_as_xml
126 : USE rt_bse, ONLY: run_propagation_bse
127 : USE rt_bse_linearized, ONLY: run_propagation_linearized_bse
128 : USE rt_propagation, ONLY: rt_prop_setup
129 : USE swarm, ONLY: run_swarm
130 : USE tamc_run, ONLY: qs_tamc
131 : USE tmc_setup, ONLY: do_analyze_files,&
132 : do_tmc
133 : USE vibrational_analysis, ONLY: vb_anal
134 : #include "../base/base_uses.f90"
135 :
136 : IMPLICIT NONE
137 :
138 : PRIVATE
139 :
140 : PUBLIC :: write_xml_file, run_input
141 :
142 : CHARACTER(len=*), PARAMETER, PRIVATE :: moduleN = 'cp2k_runs'
143 :
144 : CONTAINS
145 :
146 : ! **************************************************************************************************
147 : !> \brief performs an instance of a cp2k run
148 : !> \param input_declaration ...
149 : !> \param input_file_name name of the file to be opened for input
150 : !> \param output_unit unit to which output should be written
151 : !> \param mpi_comm ...
152 : !> \param initial_variables key-value list of initial preprocessor variables
153 : !> \author Joost VandeVondele
154 : !> \note
155 : !> para_env should be a valid communicator
156 : !> output_unit should be writeable by at least the lowest rank of the mpi group
157 : !>
158 : !> recursive because a given run_type might need to be able to perform
159 : !> another cp2k_run as part of its job (e.g. farming, classical equilibration, ...)
160 : !>
161 : !> the idea is that a cp2k instance should be able to run with just three
162 : !> arguments, i.e. a given input file, output unit, mpi communicator.
163 : !> giving these three to cp2k_run should produce a valid run.
164 : !> the only task of the PROGRAM cp2k is to create valid instances of the
165 : !> above arguments. Ideally, anything that is called afterwards should be
166 : !> able to run simultaneously / multithreaded / sequential / parallel / ...
167 : !> and able to fail safe
168 : ! **************************************************************************************************
169 11304 : RECURSIVE SUBROUTINE cp2k_run(input_declaration, input_file_name, output_unit, mpi_comm, initial_variables)
170 : TYPE(section_type), POINTER :: input_declaration
171 : CHARACTER(LEN=*), INTENT(IN) :: input_file_name
172 : INTEGER, INTENT(IN) :: output_unit
173 :
174 : CLASS(mp_comm_type) :: mpi_comm
175 : CHARACTER(len=default_path_length), &
176 : DIMENSION(:, :), INTENT(IN) :: initial_variables
177 :
178 : INTEGER :: f_env_handle, grid_backend, ierr, &
179 : iter_level, method_name_id, &
180 : new_env_id, prog_name_id, run_type_id
181 : INTEGER(KIND=int_8) :: m_memory_max_mpi
182 : LOGICAL :: echo_input, grid_apply_cutoff, &
183 : grid_validate, has_excited_states, &
184 : has_properties, I_was_ionode
185 : TYPE(cp_logger_type), POINTER :: logger, sublogger
186 : TYPE(mp_para_env_type), POINTER :: para_env
187 : TYPE(dft_control_type), POINTER :: dft_control
188 : TYPE(f_env_type), POINTER :: f_env
189 : TYPE(force_env_type), POINTER :: force_env
190 : TYPE(global_environment_type), POINTER :: globenv
191 : TYPE(section_vals_type), POINTER :: glob_section, input_file, root_section, &
192 : section
193 :
194 11304 : NULLIFY (para_env, f_env, dft_control)
195 11304 : ALLOCATE (para_env)
196 11304 : para_env = mpi_comm
197 :
198 : ! The surrounding CP2K runtime owns DBCSR. Only redirect its diagnostics
199 : ! to this run's output; the DBCSR API leaves an active library initialized.
200 11304 : CALL configure_dbcsr(mpi_comm, output_unit)
201 :
202 11304 : NULLIFY (globenv, force_env)
203 :
204 11304 : CALL cite_reference(cp2kqs2020)
205 11304 : CALL cite_reference(Iannuzzi2026)
206 :
207 : ! Parse the input
208 : input_file => read_input(input_declaration, input_file_name, &
209 : initial_variables=initial_variables, &
210 11304 : para_env=para_env)
211 :
212 11304 : CALL para_env%sync()
213 :
214 11304 : logger => cp_get_default_logger()
215 :
216 11304 : glob_section => section_vals_get_subs_vals(input_file, "GLOBAL")
217 11304 : CALL section_vals_val_get(glob_section, "ECHO_INPUT", l_val=echo_input)
218 11304 : IF (echo_input .AND. (output_unit > 0)) THEN
219 : CALL section_vals_write(input_file, &
220 : unit_nr=output_unit, &
221 : hide_root=.TRUE., &
222 16 : hide_defaults=.FALSE.)
223 : END IF
224 :
225 11304 : CALL check_cp2k_input(input_declaration, input_file, para_env=para_env, output_unit=output_unit)
226 11304 : root_section => input_file
227 : CALL section_vals_val_get(input_file, "GLOBAL%PROGRAM_NAME", &
228 11304 : i_val=prog_name_id)
229 : CALL section_vals_val_get(input_file, "GLOBAL%RUN_TYPE", &
230 11304 : i_val=run_type_id)
231 11304 : CALL section_vals_val_get(root_section, "FORCE_EVAL%METHOD", i_val=method_name_id)
232 :
233 11304 : IF (prog_name_id /= do_cp2k) THEN
234 : ! initial setup (cp2k does it in the creation of the force_env)
235 528 : CALL globenv_create(globenv)
236 528 : CALL section_vals_retain(input_file)
237 528 : CALL cp2k_init(para_env, output_unit, globenv, input_file_name=input_file_name)
238 528 : CALL cp2k_read(root_section, para_env, globenv)
239 528 : CALL cp2k_setup(root_section, para_env, globenv)
240 : END IF
241 :
242 11304 : CALL cp_dbcsr_config(root_section)
243 11304 : IF (output_unit > 0 .AND. &
244 : cp_logger_would_log(logger, cp_note_level)) THEN
245 5680 : CALL dbcsr_print_config(unit_nr=output_unit)
246 5680 : WRITE (UNIT=output_unit, FMT='()')
247 : END IF
248 :
249 : ! Configure the grid library.
250 11304 : CALL section_vals_val_get(root_section, "GLOBAL%GRID%BACKEND", i_val=grid_backend)
251 11304 : CALL section_vals_val_get(root_section, "GLOBAL%GRID%VALIDATE", l_val=grid_validate)
252 11304 : CALL section_vals_val_get(root_section, "GLOBAL%GRID%APPLY_CUTOFF", l_val=grid_apply_cutoff)
253 :
254 : CALL grid_library_set_config(backend=grid_backend, &
255 : validate=grid_validate, &
256 11304 : apply_cutoff=grid_apply_cutoff)
257 :
258 364 : SELECT CASE (prog_name_id)
259 : CASE (do_atom)
260 364 : globenv%run_type_id = none_run
261 364 : CALL atom_code(root_section)
262 : CASE (do_optimize_input)
263 6 : CALL run_optimize_input(input_declaration, root_section, para_env)
264 : CASE (do_swarm)
265 6 : CALL run_swarm(input_declaration, root_section, para_env, globenv, input_file_name)
266 : CASE (do_farming)
267 : ! FARMING temporarily replaces the runtime communicator with disjoint
268 : ! job communicators. Each job owns DBCSR until its communicator is freed.
269 24 : CALL dbcsr_finalize_lib()
270 24 : CALL farming_run(input_declaration, root_section, para_env, initial_variables)
271 24 : CALL configure_dbcsr(mpi_comm, output_unit)
272 : CASE (do_opt_basis)
273 8 : CALL run_optimize_basis(input_declaration, root_section, para_env, globenv)
274 8 : globenv%run_type_id = none_run
275 : CASE (do_cp2k)
276 : CALL create_force_env(new_env_id, &
277 : input_declaration=input_declaration, &
278 : input_path=input_file_name, &
279 : output_path="__STD_OUT__", mpi_comm=para_env, &
280 : output_unit=output_unit, &
281 : owns_out_unit=.FALSE., &
282 10776 : input=input_file, ierr=ierr)
283 10776 : CPASSERT(ierr == 0)
284 10776 : CALL f_env_add_defaults(new_env_id, f_env, handle=f_env_handle)
285 10776 : force_env => f_env%force_env
286 10776 : CALL force_env_get(force_env, globenv=globenv)
287 : CASE (do_test)
288 80 : CALL lib_test(root_section, para_env, globenv)
289 : CASE (do_tree_mc) ! TMC entry point
290 28 : CALL do_tmc(input_declaration, root_section, para_env, globenv)
291 : CASE (do_tree_mc_ana)
292 12 : CALL do_analyze_files(input_declaration, root_section, para_env)
293 : CASE default
294 22080 : CPABORT("Unknown program")
295 : END SELECT
296 11304 : CALL section_vals_release(input_file)
297 :
298 11372 : SELECT CASE (globenv%run_type_id)
299 : CASE (pint_run)
300 68 : CALL do_pint_run(para_env, root_section, input_declaration, globenv)
301 : CASE (none_run, tree_mc_run)
302 : ! do nothing
303 : CASE (driver_run)
304 0 : CALL run_driver(force_env, globenv)
305 : CASE (energy_run, energy_force_run)
306 : IF (method_name_id /= do_qs .AND. &
307 : method_name_id /= do_sirius .AND. &
308 : method_name_id /= do_qmmm .AND. &
309 : method_name_id /= do_mixed .AND. &
310 : method_name_id /= do_nnp .AND. &
311 : method_name_id /= do_embed .AND. &
312 6434 : method_name_id /= do_fist .AND. &
313 : method_name_id /= do_ipi) THEN
314 0 : CPABORT("Energy/Force run not available for all methods ")
315 : END IF
316 :
317 6434 : sublogger => cp_get_default_logger()
318 6434 : CALL cp_add_iter_level(sublogger, "JUST_ENERGY", n_rlevel_new=iter_level)
319 :
320 : ! loop over molecules to generate a molecular guess
321 : ! this procedure is initiated here to avoid passing globenv deep down
322 : ! the subroutine stack
323 6434 : IF (do_mol_loop(force_env=force_env)) THEN
324 16 : CALL loop_over_molecules(globenv, force_env)
325 : END IF
326 :
327 : ! Only this driver knows that no later geometry step will be requested.
328 : ! Keep history for nested/repeated SCF consumers and for library/shell use.
329 6434 : IF (method_name_id == do_qs) THEN
330 5734 : CALL get_qs_env(force_env%qs_env, dft_control=dft_control)
331 5734 : section => section_vals_get_subs_vals(force_env%qs_env%input, "PROPERTIES")
332 5734 : CALL section_vals_get(section, explicit=has_properties)
333 5734 : section => section_vals_get_subs_vals(force_env%qs_env%input, "DFT%EXCITED_STATES")
334 5734 : CALL section_vals_get(section, explicit=has_excited_states)
335 : force_env%qs_env%skip_wf_history = .NOT. (has_properties .OR. has_excited_states .OR. &
336 : dft_control%qs_control%cdft .OR. &
337 : dft_control%qs_control%et_coupling_calc .OR. &
338 : dft_control%do_xas_calculation .OR. &
339 : dft_control%do_xas_tdp_calculation .OR. &
340 6618 : force_env%qs_env%scf_control%density_guess == history_guess)
341 : END IF
342 :
343 11598 : SELECT CASE (globenv%run_type_id)
344 : CASE (energy_run)
345 5164 : CALL force_env_calc_energy_force(force_env, calc_force=.FALSE.)
346 : CASE (energy_force_run)
347 1270 : CALL force_env_calc_energy_force(force_env, calc_force=.TRUE.)
348 : CASE default
349 6434 : CPABORT("Unknown run type")
350 : END SELECT
351 6434 : IF (method_name_id == do_qs) force_env%qs_env%skip_wf_history = .FALSE.
352 6434 : CALL cp_rm_iter_level(sublogger, "JUST_ENERGY", n_rlevel_att=iter_level)
353 : CASE (mol_dyn_run)
354 1644 : CALL qs_mol_dyn(force_env, globenv)
355 : CASE (geo_opt_run)
356 812 : CALL cp_geo_opt(force_env, globenv)
357 : CASE (cell_opt_run)
358 234 : CALL cp_cell_opt(force_env, globenv)
359 : CASE (mon_car_run)
360 20 : CALL do_mon_car(force_env, globenv, input_declaration, input_file_name)
361 : CASE (do_tamc)
362 2 : CALL qs_tamc(force_env, globenv)
363 : CASE (real_time_propagation)
364 210 : IF (method_name_id /= do_qs) THEN
365 0 : CPABORT("Real time propagation needs METHOD QS. ")
366 : END IF
367 210 : CALL get_qs_env(force_env%qs_env, dft_control=dft_control)
368 210 : dft_control%rtp_control%fixed_ions = .TRUE.
369 344 : SELECT CASE (dft_control%rtp_control%rtp_method)
370 : CASE (rtp_method_bse_linearized)
371 : ! Run the linearized TD-BSE method
372 58 : CALL run_propagation_linearized_bse(force_env)
373 : CASE (rtp_method_bse)
374 : ! Run the TD-BSE method
375 14 : CALL run_propagation_bse(force_env)
376 : CASE default
377 : ! Run the TDDFT method
378 210 : CALL rt_prop_setup(force_env)
379 : END SELECT
380 : CASE (ehrenfest)
381 76 : IF (method_name_id /= do_qs) THEN
382 0 : CPABORT("Ehrenfest dynamics needs METHOD QS ")
383 : END IF
384 76 : CALL get_qs_env(force_env%qs_env, dft_control=dft_control)
385 76 : dft_control%rtp_control%fixed_ions = .FALSE.
386 76 : CALL qs_mol_dyn(force_env, globenv)
387 : CASE (bsse_run)
388 12 : CALL do_bsse_calculation(force_env, globenv)
389 : CASE (linear_response_run)
390 178 : IF (method_name_id /= do_qs .AND. &
391 : method_name_id /= do_qmmm) THEN
392 0 : CPABORT("Property calculations by Linear Response only within the QS or QMMM program ")
393 : END IF
394 : ! The Ground State is needed, it can be read from Restart
395 178 : CALL force_env_calc_energy_force(force_env, calc_force=.FALSE., linres=.TRUE.)
396 178 : CALL linres_calculation(force_env)
397 : CASE (debug_run)
398 1030 : SELECT CASE (method_name_id)
399 : CASE (do_qs, do_qmmm, do_fist)
400 970 : CALL cp2k_debug_energy_and_forces(force_env)
401 : CASE DEFAULT
402 970 : CPABORT("Debug run available only with QS, FIST, and QMMM program ")
403 : END SELECT
404 : CASE (vib_anal)
405 60 : CALL vb_anal(root_section, input_declaration, para_env, globenv)
406 : CASE (do_band)
407 34 : CALL neb(root_section, input_declaration, para_env, globenv)
408 : CASE (negf_run)
409 6 : CALL do_negf(force_env)
410 : CASE (mimic_run)
411 0 : CALL do_mimic_loop(force_env)
412 : CASE (mtlr_run)
413 16 : IF (method_name_id /= do_qs) THEN
414 0 : CPABORT("RUN_TYPE MTLR is available only for METHOD QS.")
415 : END IF
416 16 : CALL do_mtlr_u_j(force_env)
417 : CASE default
418 17738 : CPABORT("Unknown run type")
419 : END SELECT
420 :
421 : ! Sample peak memory
422 11304 : CALL m_memory()
423 :
424 11304 : CALL dbcsr_print_statistics()
425 11304 : CALL dbm_library_print_stats(mpi_comm=mpi_comm, output_unit=output_unit)
426 11304 : CALL dbt_library_print_stats(mpi_comm=mpi_comm, output_unit=output_unit)
427 11304 : CALL grid_library_print_stats(mpi_comm=mpi_comm, output_unit=output_unit)
428 11304 : CALL offload_mempool_stats_print(mpi_comm=mpi_comm, output_unit=output_unit)
429 :
430 11304 : m_memory_max_mpi = m_memory_max
431 11304 : CALL mpi_comm%max(m_memory_max_mpi)
432 11304 : IF (output_unit > 0) THEN
433 5680 : WRITE (output_unit, *)
434 : WRITE (output_unit, '(T2,"MEMORY| Estimated peak process memory [MiB]",T73,I8)') &
435 5680 : (m_memory_max_mpi + (1024*1024) - 1)/(1024*1024)
436 : END IF
437 :
438 11304 : IF (prog_name_id == do_cp2k) THEN
439 10776 : f_env%force_env => force_env ! for mc
440 10776 : IF (ASSOCIATED(force_env%globenv)) THEN
441 10776 : IF (.NOT. ASSOCIATED(force_env%globenv, globenv)) THEN
442 0 : CALL globenv_release(force_env%globenv) !mc
443 : END IF
444 : END IF
445 10776 : force_env%globenv => globenv !mc
446 : CALL f_env_rm_defaults(f_env, ierr=ierr, &
447 10776 : handle=f_env_handle)
448 10776 : CPASSERT(ierr == 0)
449 10776 : CALL destroy_force_env(new_env_id, ierr=ierr)
450 10776 : CPASSERT(ierr == 0)
451 : ELSE
452 : I_was_ionode = para_env%is_source()
453 528 : CALL cp2k_finalize(root_section, para_env, globenv)
454 528 : CPASSERT(globenv%ref_count == 1)
455 528 : CALL section_vals_release(root_section)
456 528 : CALL globenv_release(globenv)
457 : END IF
458 :
459 11304 : CALL mp_para_env_release(para_env)
460 :
461 11304 : END SUBROUTINE cp2k_run
462 :
463 : ! **************************************************************************************************
464 : !> \brief performs a farming run that performs several independent cp2k_runs
465 : !> \param input_declaration ...
466 : !> \param root_section ...
467 : !> \param para_env ...
468 : !> \param initial_variables ...
469 : !> \author Joost VandeVondele
470 : !> \note
471 : !> needs to be part of this module as the cp2k_run -> farming_run -> cp2k_run
472 : !> calling style creates a hard circular dependency
473 : ! **************************************************************************************************
474 24 : RECURSIVE SUBROUTINE farming_run(input_declaration, root_section, para_env, initial_variables)
475 : TYPE(section_type), POINTER :: input_declaration
476 : TYPE(section_vals_type), POINTER :: root_section
477 : TYPE(mp_para_env_type), POINTER :: para_env
478 : CHARACTER(len=default_path_length), DIMENSION(:, :), INTENT(IN) :: initial_variables
479 :
480 : CHARACTER(len=*), PARAMETER :: routineN = 'farming_run'
481 : INTEGER, PARAMETER :: minion_status_done = -3, &
482 : minion_status_wait = -4
483 :
484 : CHARACTER(len=7) :: label
485 : CHARACTER(LEN=default_path_length) :: output_file
486 : CHARACTER(LEN=default_string_length) :: str
487 : INTEGER :: dest, handle, i, i_job_to_restart, ierr, ijob, ijob_current, &
488 : ijob_end, ijob_start, iunit, n_jobs_to_run, new_output_unit, &
489 : new_rank, ngroups, num_minions, output_unit, primus_minion, &
490 : minion_rank, source, tag, todo
491 24 : INTEGER, DIMENSION(:), POINTER :: group_distribution, &
492 24 : captain_minion_partition, &
493 24 : minion_distribution, &
494 24 : minion_status
495 : LOGICAL :: found, captain, minion
496 : REAL(KIND=dp) :: t1, t2
497 24 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: waittime
498 : TYPE(cp_logger_type), POINTER :: logger
499 : TYPE(cp_parser_type), POINTER :: my_parser
500 : TYPE(cp_unit_set_type) :: default_units
501 : TYPE(farming_env_type), POINTER :: farming_env
502 : TYPE(section_type), POINTER :: g_section
503 : TYPE(section_vals_type), POINTER :: g_data
504 : TYPE(mp_comm_type) :: minion_group, new_group
505 :
506 : ! the primus of all minions, talks to the captain on topics concerning all minions
507 24 : CALL timeset(routineN, handle)
508 24 : NULLIFY (my_parser, g_section, g_data)
509 :
510 24 : logger => cp_get_default_logger()
511 : output_unit = cp_print_key_unit_nr(logger, root_section, "FARMING%PROGRAM_RUN_INFO", &
512 24 : extension=".log")
513 :
514 24 : IF (output_unit > 0) WRITE (output_unit, FMT="(T2,A)") "FARMING| Hi, welcome on this farm!"
515 :
516 24 : ALLOCATE (farming_env)
517 24 : CALL init_farming_env(farming_env)
518 : ! remember where we started
519 24 : CALL m_getcwd(farming_env%cwd)
520 24 : CALL farming_parse_input(farming_env, root_section, para_env)
521 :
522 : ! the full mpi group is first split in a minion group and a captain group, the latter being at most 1 process
523 24 : minion = .TRUE.
524 24 : captain = .FALSE.
525 24 : IF (farming_env%captain_minion) THEN
526 4 : IF (output_unit > 0) WRITE (output_unit, FMT="(T2,A)") "FARMING| Using a Captain-Minion setup"
527 :
528 4 : ALLOCATE (captain_minion_partition(0:1))
529 12 : captain_minion_partition = [1, para_env%num_pe - 1]
530 12 : ALLOCATE (group_distribution(0:para_env%num_pe - 1))
531 :
532 : CALL minion_group%from_split(para_env, ngroups, group_distribution, &
533 4 : n_subgroups=2, group_partition=captain_minion_partition)
534 4 : DEALLOCATE (captain_minion_partition)
535 4 : DEALLOCATE (group_distribution)
536 4 : num_minions = minion_group%num_pe
537 4 : minion_rank = minion_group%mepos
538 :
539 4 : IF (para_env%mepos == 0) THEN
540 2 : minion = .FALSE.
541 2 : captain = .TRUE.
542 : ! on the captain node, num_minions corresponds to the size of the captain group
543 2 : CPASSERT(num_minions == 1)
544 2 : num_minions = para_env%num_pe - 1
545 2 : minion_rank = -1
546 : END IF
547 4 : CPASSERT(num_minions == para_env%num_pe - 1)
548 : ELSE
549 : ! all processes are minions
550 20 : IF (output_unit > 0) WRITE (output_unit, FMT="(T2,A)") "FARMING| Using a Minion-only setup"
551 20 : CALL minion_group%from_dup(para_env)
552 20 : num_minions = minion_group%num_pe
553 20 : minion_rank = minion_group%mepos
554 : END IF
555 24 : IF (output_unit > 0) WRITE (output_unit, FMT="(T2,A,I0)") "FARMING| Number of Minions ", num_minions
556 :
557 : ! keep track of which para_env rank is which minion/captain
558 72 : ALLOCATE (minion_distribution(0:para_env%num_pe - 1))
559 72 : minion_distribution = 0
560 24 : minion_distribution(para_env%mepos) = minion_rank
561 120 : CALL para_env%sum(minion_distribution)
562 : ! we do have a primus inter pares
563 24 : primus_minion = 0
564 48 : DO i = 1, para_env%num_pe - 1
565 48 : IF (minion_distribution(i) == 0) primus_minion = i
566 : END DO
567 :
568 : ! split the current communicator for the minions
569 : ! in a new_group, new_size and new_rank according to the number of groups required according to the input
570 72 : ALLOCATE (group_distribution(0:num_minions - 1))
571 68 : group_distribution = -1
572 24 : IF (minion) THEN
573 22 : IF (farming_env%group_size_wish_set) THEN
574 4 : farming_env%group_size_wish = MIN(farming_env%group_size_wish, para_env%num_pe)
575 : CALL new_group%from_split(minion_group, ngroups, group_distribution, &
576 4 : subgroup_min_size=farming_env%group_size_wish, stride=farming_env%stride)
577 18 : ELSE IF (farming_env%ngroup_wish_set) THEN
578 18 : IF (ASSOCIATED(farming_env%group_partition)) THEN
579 : CALL new_group%from_split(minion_group, ngroups, group_distribution, &
580 : n_subgroups=farming_env%ngroup_wish, &
581 0 : group_partition=farming_env%group_partition, stride=farming_env%stride)
582 : ELSE
583 : CALL new_group%from_split(minion_group, ngroups, group_distribution, &
584 18 : n_subgroups=farming_env%ngroup_wish, stride=farming_env%stride)
585 : END IF
586 : ELSE
587 0 : CPABORT("must set either group_size_wish or ngroup_wish")
588 : END IF
589 22 : new_rank = new_group%mepos
590 : END IF
591 :
592 : ! transfer the info about the minion group distribution to the captain
593 24 : IF (farming_env%captain_minion) THEN
594 4 : IF (para_env%mepos == primus_minion) THEN
595 2 : tag = 1
596 4 : CALL para_env%send(group_distribution, 0, tag)
597 2 : tag = 2
598 2 : CALL para_env%send(ngroups, 0, tag)
599 : END IF
600 4 : IF (para_env%mepos == 0) THEN
601 2 : tag = 1
602 6 : CALL para_env%recv(group_distribution, primus_minion, tag)
603 2 : tag = 2
604 2 : CALL para_env%recv(ngroups, primus_minion, tag)
605 : END IF
606 : END IF
607 :
608 : ! write info on group distribution
609 24 : IF (output_unit > 0) THEN
610 12 : WRITE (output_unit, FMT="(T2,A,T71,I10)") "FARMING| Number of created MPI (Minion) groups:", ngroups
611 12 : WRITE (output_unit, FMT="(T2,A)", ADVANCE="NO") "FARMING| MPI (Minion) process to group correspondence:"
612 34 : DO i = 0, num_minions - 1
613 22 : IF (MODULO(i, 4) == 0) WRITE (output_unit, *)
614 : WRITE (output_unit, FMT='(A3,I6,A3,I6,A1)', ADVANCE="NO") &
615 34 : " (", i, " : ", group_distribution(i), ")"
616 : END DO
617 12 : WRITE (output_unit, *)
618 12 : CALL m_flush(output_unit)
619 : END IF
620 :
621 : ! protect about too many jobs being run in single go. Not more jobs are allowed than the number in the input file
622 : ! and determine the future restart point
623 24 : IF (farming_env%cycle) THEN
624 2 : n_jobs_to_run = farming_env%max_steps*ngroups
625 2 : i_job_to_restart = MODULO(farming_env%restart_n + n_jobs_to_run - 1, farming_env%njobs) + 1
626 : ELSE
627 22 : n_jobs_to_run = MIN(farming_env%njobs, farming_env%max_steps*ngroups)
628 22 : n_jobs_to_run = MIN(n_jobs_to_run, farming_env%njobs - farming_env%restart_n + 1)
629 22 : i_job_to_restart = n_jobs_to_run + farming_env%restart_n
630 : END IF
631 :
632 : ! and write the restart now, that's the point where the next job starts, even if this one is running
633 : iunit = cp_print_key_unit_nr(logger, root_section, "FARMING%RESTART", &
634 24 : extension=".restart")
635 24 : IF (iunit > 0) THEN
636 12 : WRITE (iunit, *) i_job_to_restart
637 : END IF
638 24 : CALL cp_print_key_finished_output(iunit, logger, root_section, "FARMING%RESTART")
639 :
640 : ! this is the job range to be executed.
641 24 : ijob_start = farming_env%restart_n
642 24 : ijob_end = ijob_start + n_jobs_to_run - 1
643 24 : IF (output_unit > 0 .AND. ijob_end - ijob_start < 0) THEN
644 0 : WRITE (output_unit, FMT="(T2,A)") "FARMING| --- WARNING --- NO JOBS NEED EXECUTION ? "
645 0 : WRITE (output_unit, FMT="(T2,A)") "FARMING| is the cycle keyword required ?"
646 0 : WRITE (output_unit, FMT="(T2,A)") "FARMING| or is a stray RESTART file present ?"
647 0 : WRITE (output_unit, FMT="(T2,A)") "FARMING| or is the group_size requested smaller than the number of CPUs?"
648 : END IF
649 :
650 : ! actual executions of the jobs in two different modes
651 24 : IF (farming_env%captain_minion) THEN
652 4 : IF (minion) THEN
653 : ! keep on doing work until captain has decided otherwise
654 2 : todo = do_wait
655 : DO
656 20 : IF (new_rank == 0) THEN
657 : ! the head minion tells the captain he's done or ready to start
658 : ! the message tells what has been done lately
659 20 : tag = 1
660 20 : dest = 0
661 20 : CALL para_env%send(todo, dest, tag)
662 :
663 : ! gets the new todo item
664 20 : tag = 2
665 20 : source = 0
666 20 : CALL para_env%recv(todo, source, tag)
667 :
668 : ! and informs his peer minions
669 20 : CALL new_group%bcast(todo, 0)
670 : ELSE
671 0 : CALL new_group%bcast(todo, 0)
672 : END IF
673 :
674 : ! if the todo is do_nothing we are flagged to quit. Otherwise it is the job number
675 0 : SELECT CASE (todo)
676 : CASE (do_wait, do_deadlock)
677 : ! go for a next round, but we first wait a bit
678 0 : t1 = m_walltime()
679 : DO
680 0 : t2 = m_walltime()
681 0 : IF (t2 - t1 > farming_env%wait_time) EXIT
682 : END DO
683 : CASE (do_nothing)
684 18 : EXIT
685 : CASE (1:)
686 20 : CALL execute_job(todo)
687 : END SELECT
688 : END DO
689 : ELSE ! captain
690 6 : ALLOCATE (minion_status(0:ngroups - 1))
691 4 : minion_status = minion_status_wait
692 2 : ijob_current = ijob_start - 1
693 :
694 20 : DO
695 24 : IF (ALL(minion_status == minion_status_done)) EXIT
696 :
697 : ! who's the next minion waiting for work
698 20 : tag = 1
699 20 : source = mp_any_source
700 20 : CALL para_env%recv(todo, source, tag) ! updates source
701 20 : IF (todo > 0) THEN
702 18 : farming_env%Job(todo)%status = job_finished
703 18 : IF (output_unit > 0) THEN
704 18 : WRITE (output_unit, FMT=*) "Job finished: ", todo
705 18 : CALL m_flush(output_unit)
706 : END IF
707 : END IF
708 :
709 : ! get the next job in line, this could be do_nothing, if we're finished
710 20 : CALL get_next_job(farming_env, ijob_start, ijob_end, ijob_current, todo)
711 20 : dest = source
712 20 : tag = 2
713 20 : CALL para_env%send(todo, dest, tag)
714 :
715 22 : IF (todo > 0) THEN
716 18 : farming_env%Job(todo)%status = job_running
717 18 : IF (output_unit > 0) THEN
718 18 : WRITE (output_unit, FMT=*) "Job: ", todo, " Dir: ", TRIM(farming_env%Job(todo)%cwd), &
719 36 : " assigned to group ", group_distribution(minion_distribution(dest))
720 18 : CALL m_flush(output_unit)
721 : END IF
722 : ELSE
723 2 : IF (todo == do_nothing) THEN
724 2 : minion_status(group_distribution(minion_distribution(dest))) = minion_status_done
725 2 : IF (output_unit > 0) THEN
726 2 : WRITE (output_unit, FMT=*) "group done: ", group_distribution(minion_distribution(dest))
727 2 : CALL m_flush(output_unit)
728 : END IF
729 : END IF
730 2 : IF (todo == do_deadlock) THEN
731 0 : IF (output_unit > 0) THEN
732 0 : WRITE (output_unit, FMT=*) ""
733 0 : WRITE (output_unit, FMT=*) "FARMING JOB DEADLOCKED ... CIRCULAR DEPENDENCIES"
734 0 : WRITE (output_unit, FMT=*) ""
735 0 : CALL m_flush(output_unit)
736 : END IF
737 0 : CPASSERT(todo /= do_deadlock)
738 : END IF
739 : END IF
740 :
741 : END DO
742 :
743 2 : DEALLOCATE (minion_status)
744 :
745 : END IF
746 : ELSE
747 : ! this is the non-captain-minion mode way of executing the jobs
748 : ! the i-th job in the input is always executed by the MODULO(i-1,ngroups)-th group
749 : ! (needed for cyclic runs, we don't want two groups working on the same job)
750 20 : IF (output_unit > 0) THEN
751 10 : IF (ijob_end - ijob_start >= 0) THEN
752 10 : WRITE (output_unit, FMT="(T2,A)") "FARMING| List of jobs : "
753 81 : DO ijob = ijob_start, ijob_end
754 71 : i = MODULO(ijob - 1, farming_env%njobs) + 1
755 71 : WRITE (output_unit, FMT=*) "Job: ", i, " Dir: ", TRIM(farming_env%Job(i)%cwd), " Input: ", &
756 152 : TRIM(farming_env%Job(i)%input), " MPI group:", MODULO(i - 1, ngroups)
757 : END DO
758 : END IF
759 10 : CALL m_flush(output_unit)
760 : END IF
761 :
762 162 : DO ijob = ijob_start, ijob_end
763 142 : i = MODULO(ijob - 1, farming_env%njobs) + 1
764 : ! this farms out the jobs
765 162 : IF (MODULO(i - 1, ngroups) == group_distribution(minion_rank)) THEN
766 104 : IF (output_unit > 0) THEN
767 54 : WRITE (output_unit, FMT="(T2,A,I5.5,A)", ADVANCE="NO") " Running Job ", i, &
768 108 : " in "//TRIM(farming_env%Job(i)%cwd)//"."
769 54 : CALL m_flush(output_unit)
770 : END IF
771 104 : CALL execute_job(i)
772 104 : IF (output_unit > 0) THEN
773 54 : WRITE (output_unit, FMT="(A)") " Done, output in "//TRIM(output_file)
774 54 : CALL m_flush(output_unit)
775 : END IF
776 : END IF
777 : END DO
778 : END IF
779 :
780 : ! keep information about how long each process has to wait
781 : ! i.e. the load imbalance
782 24 : t1 = m_walltime()
783 24 : CALL para_env%sync()
784 24 : t2 = m_walltime()
785 72 : ALLOCATE (waittime(0:para_env%num_pe - 1))
786 24 : waittime = 0.0_dp
787 24 : waittime(para_env%mepos) = t2 - t1
788 24 : CALL para_env%sum(waittime)
789 24 : IF (output_unit > 0) THEN
790 12 : WRITE (output_unit, '(T2,A)') "Process idle times [s] at the end of the run"
791 36 : DO i = 0, para_env%num_pe - 1
792 : WRITE (output_unit, FMT='(A2,I6,A3,F8.3,A1)', ADVANCE="NO") &
793 24 : " (", i, " : ", waittime(i), ")"
794 36 : IF (MOD(i + 1, 4) == 0) WRITE (output_unit, '(A)') ""
795 : END DO
796 12 : CALL m_flush(output_unit)
797 : END IF
798 24 : DEALLOCATE (waittime)
799 :
800 : ! give back the communicators of the split groups
801 24 : IF (minion) CALL new_group%free()
802 24 : CALL minion_group%free()
803 :
804 : ! and message passing deallocate structures
805 24 : DEALLOCATE (group_distribution)
806 24 : DEALLOCATE (minion_distribution)
807 :
808 : ! clean the farming env
809 24 : CALL deallocate_farming_env(farming_env)
810 :
811 : CALL cp_print_key_finished_output(output_unit, logger, root_section, &
812 24 : "FARMING%PROGRAM_RUN_INFO")
813 :
814 336 : CALL timestop(handle)
815 :
816 : CONTAINS
817 : ! **************************************************************************************************
818 : !> \brief ...
819 : !> \param i ...
820 : ! **************************************************************************************************
821 122 : RECURSIVE SUBROUTINE execute_job(i)
822 : INTEGER :: i
823 :
824 : ! change to the new working directory
825 :
826 122 : CALL m_chdir(TRIM(farming_env%Job(i)%cwd), ierr)
827 122 : IF (ierr /= 0) THEN
828 0 : CPABORT("Failed to change dir to: "//TRIM(farming_env%Job(i)%cwd))
829 : END IF
830 :
831 : ! generate a fresh call to cp2k_run
832 122 : IF (new_rank == 0) THEN
833 :
834 89 : IF (farming_env%Job(i)%output == "") THEN
835 : ! generate the output file
836 85 : WRITE (output_file, '(A12,I5.5)') "FARMING_OUT_", i
837 255 : ALLOCATE (my_parser)
838 85 : CALL parser_create(my_parser, file_name=TRIM(farming_env%Job(i)%input))
839 85 : label = "&GLOBAL"
840 85 : CALL parser_search_string(my_parser, label, ignore_case=.TRUE., found=found)
841 170 : IF (found) THEN
842 85 : CALL create_global_section(g_section)
843 85 : CALL section_vals_create(g_data, g_section)
844 : CALL cp_unit_set_create(default_units, "OUTPUT")
845 85 : CALL section_vals_parse(g_data, my_parser, default_units)
846 85 : CALL cp_unit_set_release(default_units)
847 : CALL section_vals_val_get(g_data, "PROJECT", &
848 85 : c_val=str)
849 85 : IF (str /= "") output_file = TRIM(str)//".out"
850 : CALL section_vals_val_get(g_data, "OUTPUT_FILE_NAME", &
851 85 : c_val=str)
852 85 : IF (str /= "") output_file = str
853 85 : CALL section_vals_release(g_data)
854 85 : CALL section_release(g_section)
855 : END IF
856 85 : CALL parser_release(my_parser)
857 85 : DEALLOCATE (my_parser)
858 : ELSE
859 4 : output_file = farming_env%Job(i)%output
860 : END IF
861 :
862 : CALL open_file(file_name=TRIM(output_file), &
863 : file_action="WRITE", &
864 : file_status="UNKNOWN", &
865 : file_position="APPEND", &
866 89 : unit_number=new_output_unit)
867 : ELSE
868 : ! this unit should be negative, otherwise all processors that get a default unit
869 : ! start writing output (to the same file, adding to confusion).
870 : ! error handling should be careful, asking for a local output unit if required
871 33 : new_output_unit = -1
872 : END IF
873 :
874 122 : CALL cp2k_run(input_declaration, TRIM(farming_env%Job(i)%input), new_output_unit, new_group, initial_variables)
875 122 : CALL dbcsr_finalize_lib()
876 :
877 122 : IF (new_rank == 0) CALL close_file(unit_number=new_output_unit)
878 :
879 : ! change to the original working directory
880 122 : CALL m_chdir(TRIM(farming_env%cwd), ierr)
881 122 : CPASSERT(ierr == 0)
882 :
883 122 : END SUBROUTINE execute_job
884 : END SUBROUTINE farming_run
885 :
886 : ! **************************************************************************************************
887 : !> \brief ...
888 : ! **************************************************************************************************
889 0 : SUBROUTINE write_xml_file()
890 :
891 : INTEGER :: i, unit_number
892 : TYPE(section_type), POINTER :: root_section
893 :
894 0 : NULLIFY (root_section)
895 0 : CALL create_cp2k_root_section(root_section)
896 0 : CALL keyword_release(root_section%keywords(0)%keyword)
897 : CALL open_file(unit_number=unit_number, &
898 : file_name="cp2k_input.xml", &
899 : file_action="WRITE", &
900 0 : file_status="REPLACE")
901 :
902 0 : WRITE (UNIT=unit_number, FMT="(A)") '<?xml version="1.0" encoding="utf-8"?>'
903 :
904 : !MK CP2K input structure
905 : WRITE (UNIT=unit_number, FMT="(A)") &
906 0 : "<CP2K_INPUT>", &
907 0 : " <CP2K_VERSION>"//TRIM(cp2k_version)//"</CP2K_VERSION>", &
908 0 : " <CP2K_YEAR>"//TRIM(cp2k_year)//"</CP2K_YEAR>", &
909 0 : " <COMPILE_DATE>"//TRIM(compile_date)//"</COMPILE_DATE>", &
910 0 : " <COMPILE_REVISION>"//TRIM(compile_revision)//"</COMPILE_REVISION>"
911 :
912 0 : CALL export_references_as_xml(unit_number)
913 0 : CALL export_units_as_xml(unit_number)
914 :
915 0 : DO i = 1, root_section%n_subsections
916 0 : CALL write_section_xml(root_section%subsections(i)%section, 1, unit_number)
917 : END DO
918 :
919 0 : WRITE (UNIT=unit_number, FMT="(A)") "</CP2K_INPUT>"
920 0 : CALL close_file(unit_number=unit_number)
921 0 : CALL section_release(root_section)
922 :
923 0 : END SUBROUTINE write_xml_file
924 :
925 : ! **************************************************************************************************
926 : !> \brief runs the given input
927 : !> \param input_declaration ...
928 : !> \param input_file_path the path of the input file
929 : !> \param output_file_path path of the output file (to which it is appended)
930 : !> if it is "__STD_OUT__" the default_output_unit is used
931 : !> \param initial_variables key-value list of initial preprocessor variables
932 : !> \param mpi_comm the mpi communicator to be used for this environment
933 : !> it will not be freed
934 : !> \author fawzi
935 : !> \note
936 : !> moved here because of circular dependencies
937 : ! **************************************************************************************************
938 11182 : SUBROUTINE run_input(input_declaration, input_file_path, output_file_path, initial_variables, mpi_comm)
939 : TYPE(section_type), POINTER :: input_declaration
940 : CHARACTER(len=*), INTENT(in) :: input_file_path, output_file_path
941 : CHARACTER(len=default_path_length), &
942 : DIMENSION(:, :), INTENT(IN) :: initial_variables
943 : TYPE(mp_comm_type), INTENT(in), OPTIONAL :: mpi_comm
944 :
945 : INTEGER :: unit_nr
946 : TYPE(mp_para_env_type), POINTER :: para_env
947 :
948 11182 : IF (PRESENT(mpi_comm)) THEN
949 4 : ALLOCATE (para_env)
950 4 : para_env = mpi_comm
951 : ELSE
952 11178 : para_env => f77_default_para_env
953 11178 : CALL para_env%retain()
954 : END IF
955 11182 : IF (para_env%is_source()) THEN
956 5591 : IF (output_file_path == "__STD_OUT__") THEN
957 5589 : unit_nr = default_output_unit
958 : ELSE
959 2 : INQUIRE (FILE=output_file_path, NUMBER=unit_nr)
960 : END IF
961 : ELSE
962 5591 : unit_nr = -1
963 : END IF
964 11182 : CALL cp2k_run(input_declaration, input_file_path, unit_nr, para_env, initial_variables)
965 : ! Restore DBCSR diagnostics before the caller closes its output unit.
966 11182 : CALL configure_dbcsr(f77_default_para_env, cp_logger_get_default_io_unit())
967 11182 : CALL mp_para_env_release(para_env)
968 11182 : END SUBROUTINE run_input
969 :
970 : END MODULE cp2k_runs
|
