67 #include "./base/base_uses.f90"
72 LOGICAL,
PRIVATE,
PARAMETER :: debug_this_module = .true.
73 CHARACTER(len=*),
PARAMETER,
PRIVATE :: moduleN =
'input_cp2k_global'
85 TYPE(section_type),
POINTER :: section
87 INTEGER :: default_dgemm
88 TYPE(keyword_type),
POINTER :: keyword
89 TYPE(section_type),
POINTER :: print_key, sub_section
92 cpassert(.NOT.
ASSOCIATED(section))
94 description=
"Section with general information regarding which kind "// &
95 "of simulation to perform an parameters for the whole PROGRAM", &
96 n_keywords=7, n_subsections=0, repeats=.false.)
100 description=
"how to distribute the processors on the 2d grid needed "// &
101 "by BLACS (and thus SCALAPACK)", usage=
"BLACS_GRID SQUARE", &
103 enum_desc=s2a(
"Distribution by matrix blocks",
"Distribution by matrix rows", &
104 "Distribution by matrix columns"), &
109 CALL keyword_create(keyword, __location__, name=
"BLACS_REPEATABLE", &
110 description=
"Use a topology for BLACS collectives that is guaranteed to be repeatable "// &
111 "on homogeneous architectures", &
112 usage=
"BLACS_REPEATABLE", &
113 default_l_val=.false., lone_keyword_l_val=.true.)
117 CALL keyword_create(keyword, __location__, name=
"PREFERRED_DIAG_LIBRARY", &
118 description=
"Specifies the diagonalization library to be used. If not available, "// &
119 "the ScaLAPACK library is used", &
120 usage=
"PREFERRED_DIAG_LIBRARY ELPA", &
126 enum_c_vals=s2a(
"ELPA",
"ScaLAPACK",
"SL",
"CUSOLVER",
"DLAF"), &
127 enum_desc=s2a(
"ELPA library", &
128 "ScaLAPACK library", &
129 "ScaLAPACK library (shorthand)", &
130 "cuSOLVER (CUDA GPU library)", &
131 "DLA-Future (CUDA/HIP GPU library)"), &
136 #if defined(__SPLA) && defined(__OFFLOAD_GEMM)
141 CALL keyword_create(keyword, __location__, name=
"PREFERRED_DGEMM_LIBRARY", &
142 description=
"Specifies the DGEMM library to be used. If not available, "// &
143 "the BLAS routine is used. This keyword affects some DGEMM calls in the WFC code and turns on their "// &
144 "acceleration with SpLA. This keyword affects only local DGEMM calls, not the calls to PDGEMM "// &
145 "(see keyword FM%TYPE_OF_MATRIX_MULTIPLICATION).", &
146 usage=
"PREFERRED_DGEMM_LIBRARY SPLA", &
147 default_i_val=default_dgemm, &
149 enum_c_vals=s2a(
"SPLA",
"BLAS"), &
150 enum_desc=s2a(
"SPLA library",
"BLAS library"))
154 CALL keyword_create(keyword, __location__, name=
"EPS_CHECK_DIAG", &
155 description=
"Check that the orthonormality of the eigenvectors after a diagonalization "// &
156 "fulfills the specified numerical accuracy. A negative threshold value disables the check.", &
157 usage=
"EPS_CHECK_DIAG 1.0E-14", &
158 default_r_val=-1.0_dp)
163 description=
"Specifies the kernel to be used when ELPA is in use", &
171 CALL keyword_create(keyword, __location__, name=
"ELPA_NEIGVEC_MIN", &
172 description=
"Minimum number of eigenvectors for the use of the eigensolver from "// &
173 "the ELPA library. The eigensolver from the ScaLAPACK library is used as fallback "// &
174 "for all smaller cases", &
175 usage=
"ELPA_NEIGVEC_MIN 32", &
181 description=
"For ELPA, enable a blocked QR step when reducing the input matrix "// &
182 "to banded form in preparation for the actual diagonalization step. "// &
183 "See implementation paper for more details. Requires ELPA version 201505 or newer, "// &
184 "automatically deactivated otherwise. If true, QR is activated only when the "// &
185 "the size of the diagonalized matrix is suitable. Print key PRINT_ELPA is "// &
186 "useful in determining which matrices are suitable for QR. Might accelerate the "// &
187 "diagonalization of suitable matrices.", &
189 default_l_val=.false., lone_keyword_l_val=.true.)
193 CALL keyword_create(keyword, __location__, name=
"ELPA_QR_UNSAFE", &
194 description=
"For ELPA, disable block size limitations when used together with ELPA_QR. "// &
195 "Keyword relevant only with ELPA versions 201605 or newer. Use keyword with caution, "// &
196 "as it might result in wrong eigenvalues with some matrix orders/block sizes "// &
197 "when the number of MPI processes is varied. If the print key PRINT_ELPA is "// &
198 "active the validity of the eigenvalues is checked against values calculated without "// &
201 default_l_val=.false., lone_keyword_l_val=.true.)
206 description=
"Controls the printing of ELPA diagonalization information. "// &
207 "Useful for testing purposes, especially together with keyword ELPA_QR.", &
208 filename=
"__STD_OUT__")
212 CALL keyword_create(keyword, __location__, name=
"DLAF_NEIGVEC_MIN", &
213 description=
"Minimum number of eigenvectors for the use of the eigensolver from "// &
214 "the DLA-Future library. The eigensolver from the ScaLAPACK library is used as fallback "// &
215 "for all smaller cases", &
216 usage=
"DLAF_NEIGVEC_MIN 512", &
222 keyword, __location__, name=
"PREFERRED_FFT_LIBRARY", &
223 description=
"Specifies the FFT library which should be preferred. "// &
224 "If it is not available, use FFTW3 if this is linked in, if FFTW3 is not available use FFTSG. "// &
225 "Improved performance with FFTW3 can be obtained specifying a proper value for FFTW_PLAN_TYPE. "// &
226 "Contrary to earlier CP2K versions, all libraries will result in the same grids, "// &
227 "i.e. the subset of grids which all FFT libraries can transform. "// &
228 "See EXTENDED_FFT_LENGTHS if larger FFTs or grids that more precisely match a given cutoff are needed, "// &
229 "or older results need to be reproduced. "// &
230 "FFTW3 is often (close to) optimal, and well tested with CP2K.", &
231 usage=
"PREFERRED_FFT_LIBRARY FFTW3", &
235 enum_c_vals=s2a(
"FFTSG",
"FFTW3",
"FFTW"), &
236 enum_desc=s2a(
"Stefan Goedecker's FFT (FFTSG), always available, "// &
237 "will be used in case a FFT library is specified and not available.", &
238 "a fast portable FFT library. Recommended. "// &
239 "See also the FFTW_PLAN_TYPE, and FFTW_WISDOM_FILE_NAME keywords.", &
240 "Same as FFTW3 (for compatibility with CP2K 2.3)"))
244 CALL keyword_create(keyword, __location__, name=
"FFTW_WISDOM_FILE_NAME", &
245 description=
"The name of the file that contains wisdom (pre-planned FFTs) for use with FFTW3. "// &
246 "Using wisdom can significantly speed up the FFTs (see the FFTW homepage for details). "// &
247 "Note that wisdom is not transferable between different computer (architectures). "// &
248 "Wisdom can be generated using the fftw-wisdom tool that is part of the fftw installation. "// &
249 "cp2k/tools/cp2k-wisdom is a script that contains some additional info, and can help "// &
250 "to generate a useful default for /etc/fftw/wisdom or particular values for a given simulation.", &
251 usage=
"FFTW_WISDOM_FILE_NAME wisdom.dat", default_lc_val=
"/etc/fftw/wisdom")
255 CALL keyword_create(keyword, __location__, name=
"FFTW_PLAN_TYPE", &
256 description=
"FFTW can have improved performance if it is allowed to plan with "// &
257 "explicit measurements which strategy is best for a given FFT. "// &
258 "While a plan based on measurements is generally faster, "// &
259 "differences in machine load will lead to different plans for the same input file, "// &
260 "and thus numerics for the FFTs will be slightly different from run to run. "// &
261 "PATIENT planning is recommended for long ab initio MD runs.", &
262 usage=
"FFTW_PLAN_TYPE PATIENT", &
266 enum_c_vals=s2a(
"ESTIMATE", &
270 enum_desc=s2a(
"Quick estimate, no runtime measurements.", &
271 "Quick measurement, somewhat faster FFTs.", &
272 "Measurements trying a wider range of possibilities.", &
273 "Measurements trying all possibilities - use with caution."))
277 CALL keyword_create(keyword, __location__, name=
"EXTENDED_FFT_LENGTHS", &
278 description=
"Use fft library specific values for the allows number of points in FFTs. "// &
279 "The default is to use the internal FFT lengths. For external fft libraries this may "// &
280 "create an error at the external library level, because the length provided by cp2k is "// &
281 "not supported by the external library. In this case switch on this keyword "// &
282 "to obtain, with certain fft libraries, lengths matching the external fft library lengths, or "// &
283 "larger allowed grids, or grids that more precisely match a given cutoff. "// &
284 "IMPORTANT NOTE: in this case, the actual grids used in CP2K depends on the FFT library. "// &
285 "A change of FFT library must therefore be considered equivalent to a change of basis, "// &
286 "which implies a change of total energy. ", &
287 usage=
"EXTENDED_FFT_LENGTHS", &
288 default_l_val=.false., lone_keyword_l_val=.true.)
292 CALL keyword_create(keyword, __location__, name=
"FFT_POOL_SCRATCH_LIMIT", &
293 description=
"Limits the memory usage of the FFT scratch pool, potentially reducing efficiency a bit", &
294 usage=
"FFT_POOL_SCRATCH_LIMIT {INTEGER}", default_i_val=15)
299 description=
"All-to-all communication (FFT) should use single precision", &
300 usage=
"ALLTOALL_SGL YES", &
301 default_l_val=.false., lone_keyword_l_val=.true.)
306 variants=(/
"IOLEVEL"/), &
307 description=
"How much output is written out.", &
308 usage=
"PRINT_LEVEL HIGH", &
310 s2a(
"SILENT",
"LOW",
"MEDIUM",
"HIGH",
"DEBUG"), &
311 enum_desc=s2a(
"Almost no output", &
312 "Little output",
"Quite some output",
"Lots of output", &
313 "Everything is written out, useful for debugging purposes only"), &
320 keyword, __location__, name=
"PROGRAM_NAME", &
321 variants=(/
"PROGRAM"/), &
322 description=
"Which program should be run", &
323 usage=
"PROGRAM_NAME {STRING}", &
324 enum_c_vals=s2a(
"ATOM",
"FARMING",
"TEST",
"CP2K",
"OPTIMIZE_INPUT",
"OPTIMIZE_BASIS",
"TMC",
"MC_ANALYSIS",
"SWARM"), &
325 enum_desc=s2a(
"Runs single atom calculations", &
326 "Runs N independent jobs in a single run", &
327 "Do some benchmarking and testing", &
328 "Runs one of the CP2K package", &
329 "A tool to optimize parameters in a CP2K input", &
330 "A tool to create a MOLOPT or ADMM basis for a given set"// &
331 " of training structures", &
332 "Runs Tree Monte Carlo algorithm using additional input file(s)", &
333 "Runs (Tree) Monte Carlo trajectory file analysis", &
334 "Runs swarm based calculation"), &
342 variants=(/
"PROJECT"/), &
343 description=
"Name of the project (used to build the name of the "// &
344 "trajectory, and other files generated by the program)", &
345 usage=
"PROJECT_NAME {STRING}", &
346 default_c_val=
"PROJECT")
350 CALL keyword_create(keyword, __location__, name=
"OUTPUT_FILE_NAME", &
351 description=
"Name of the output file. "// &
352 "Relevant only if automatically started (through farming for example). "// &
353 "If empty uses the project name as basis for it.", &
354 usage=
"OUTPUT_FILE_NAME {filename}", default_lc_val=
"")
359 keyword, __location__, name=
"RUN_TYPE", &
360 description=
"Type of run that you want to perform Geometry "// &
361 "optimization, md, montecarlo,...", &
362 usage=
"RUN_TYPE MD", &
365 enum_c_vals=s2a(
"NONE",
"ENERGY",
"ENERGY_FORCE",
"MD",
"GEO_OPT", &
366 "MC",
"SPECTRA",
"DEBUG",
"BSSE",
"LR",
"PINT",
"VIBRATIONAL_ANALYSIS", &
367 "BAND",
"CELL_OPT",
"WFN_OPT",
"WAVEFUNCTION_OPTIMIZATION", &
368 "MOLECULAR_DYNAMICS",
"GEOMETRY_OPTIMIZATION",
"MONTECARLO", &
369 "ELECTRONIC_SPECTRA",
"LINEAR_RESPONSE",
"NORMAL_MODES",
"RT_PROPAGATION", &
370 "EHRENFEST_DYN",
"TAMC",
"TMC",
"DRIVER",
"NEGF"), &
377 enum_desc=s2a(
"Perform no tasks",
"Computes energy",
"Computes energy and forces", &
378 "Molecular Dynamics",
"Geometry Optimization",
"Monte Carlo",
"Computes absorption Spectra", &
379 "Performs a Debug analysis",
"Basis set superposition error",
"Linear Response", &
380 "Path integral",
"Vibrational analysis",
"Band methods", &
381 "Cell optimization. Both cell vectors and atomic positions are optimised.", &
382 "Alias for ENERGY",
"Alias for ENERGY",
"Alias for MD",
"Alias for GEO_OPT", &
383 "Alias for MC",
"Alias for SPECTRA",
"Alias for LR",
"Alias for VIBRATIONAL_ANALYSIS", &
384 "Real Time propagation run (fixed ionic positions)", &
385 "Ehrenfest dynamics (using real time propagation of the wavefunction)", &
386 "Temperature Accelerated Monte Carlo (TAMC)", &
387 "Tree Monte Carlo (TMC), a pre-sampling MC algorithm", &
388 "i-PI driver mode", &
389 "Non-equilibrium Green's function method"))
394 variants=(/
"WALLTI"/), &
395 description=
"Maximum execution time for this run. Time in seconds or in HH:MM:SS.", &
396 usage=
"WALLTIME {real} or {HH:MM:SS}", default_lc_val=
"")
401 description=
"If the input should be echoed to the output with all the "// &
402 "defaults made explicit", &
403 usage=
"ECHO_INPUT NO", default_l_val=.false., lone_keyword_l_val=.true.)
407 CALL keyword_create(keyword, __location__, name=
"ECHO_ALL_HOSTS", &
408 description=
"Echo a list of hostname and pid for all MPI processes.", &
409 usage=
"ECHO_ALL_HOSTS NO", default_l_val=.false., lone_keyword_l_val=.true.)
413 CALL keyword_create(keyword, __location__, name=
"ENABLE_MPI_IO", &
414 description=
"Enable MPI parallelization for all supported I/O routines "// &
415 "Currently, only cube file writer/reader routines use MPI I/O. Disabling "// &
416 "this flag might speed up calculations dominated by I/O.", &
417 usage=
"ENABLE_MPI_IO FALSE", default_l_val=.true., lone_keyword_l_val=.true.)
422 description=
"If a debug trace of the execution of the program should be written ", &
424 default_l_val=.false., lone_keyword_l_val=.true.)
429 description=
"For parallel TRACEd runs: only the master node writes output.", &
430 usage=
"TRACE_MASTER", &
431 default_l_val=.true., lone_keyword_l_val=.true.)
436 keyword, __location__, name=
"TRACE_MAX", &
437 description=
"Limit the total number a given subroutine is printed in the trace. Accounting is not influenced.", &
438 usage=
"TRACE_MAX 100", default_i_val=huge(0))
443 keyword, __location__, name=
"TRACE_ROUTINES", &
444 description=
"A list of routines to trace. If left empty all routines are traced. Accounting is not influenced.", &
445 usage=
"TRACE_ROUTINES {routine_name1} {routine_name2} ...", type_of_var=
char_t, &
451 keyword, __location__, name=
"FLUSH_SHOULD_FLUSH", &
452 description=
"Flush output regularly, enabling this option might degrade performance significantly on certain machines.", &
453 usage=
"FLUSH_SHOULD_FLUSH", &
454 default_l_val=.true., lone_keyword_l_val=.true.)
459 description=
"At the end of the run write a callgraph to file, "// &
460 "which contains detailed timing informations. "// &
461 "This callgraph can be viewed e.g. with the open-source program kcachegrind.", &
462 usage=
"CALLGRAPH {NONE|MASTER|ALL}", &
464 enum_c_vals=s2a(
"NONE",
"MASTER",
"ALL"), &
465 enum_desc=s2a(
"No callgraph gets written", &
466 "Only the master process writes his callgraph", &
467 "All processes write their callgraph (into a separate files)."), &
472 CALL keyword_create(keyword, __location__, name=
"CALLGRAPH_FILE_NAME", &
473 description=
"Name of the callgraph file, which is written at the end of the run. "// &
474 "If not specified the project name will be used as filename.", &
475 usage=
"CALLGRAPH_FILE_NAME {filename}", default_lc_val=
"")
480 description=
"Initial seed for the global (pseudo)random number generator "// &
481 "to create a stream of normally Gaussian distributed random numbers. "// &
482 "Exactly 1 or 6 positive integer values are expected. A single value is "// &
483 "replicated to fill up the full seed array with 6 numbers.", &
486 usage=
"SEED {INTEGER} .. {INTEGER}", &
487 default_i_vals=(/2000/))
492 description=
"Some sections of the input structure are deallocated when not needed,"// &
493 " and reallocated only when used. This reduces the required maximum memory ", &
495 default_l_val=.false., lone_keyword_l_val=.true.)
500 "Controls the printing of the timing report at the end of CP2K execution", &
504 description=
"Specify % of CPUTIME above which the contribution will be inserted in the"// &
505 " final timing report (e.g. 0.02 = 2%)", &
506 usage=
"THRESHOLD {REAL}", &
507 default_r_val=0.02_dp)
511 CALL keyword_create(keyword, __location__, name=
"SORT_BY_SELF_TIME", &
512 description=
"Sort the final timing report by the average self (exclusive) time instead of the "// &
513 "total (inclusive) time of a routine", &
514 usage=
"SORT_BY_SELF_TIME on", &
515 default_l_val=.false., lone_keyword_l_val=.true.)
519 CALL keyword_create(keyword, __location__, name=
"REPORT_MAXLOC", &
520 description=
"Report the rank with the slowest maximum self timing."// &
521 " Can be used to debug hard- or software."// &
522 " Also enables ECHO_ALL_HOSTS to link rank to hostname.", &
523 usage=
"REPORT_MAXLOC on", &
524 default_l_val=.false., lone_keyword_l_val=.true.)
529 description=
"Include message_passing calls in the timing report (useful with CALLGRAPH).", &
530 usage=
"TIME_MPI .FALSE.", &
531 default_l_val=.true., lone_keyword_l_val=.true.)
535 CALL keyword_create(keyword, __location__, name=
"TIMINGS_LEVEL", &
536 description=
"Specify the level of timings report. "// &
537 "Possible values are: 0 (report only CP2K root timer), 1 (all timers).", &
538 usage=
"TIMINGS_LEVEL 1", &
547 "Controls the printing of the references relevant to the calculations performed", &
553 description=
"controls the printing of initialization controlled by the global section", &
559 "controls the printing of physical and mathematical constants", &
562 CALL keyword_create(keyword, __location__, name=
"BASIC_DATA_TYPES", &
563 description=
"Controls the printing of the basic data types.", &
564 default_l_val=.false., lone_keyword_l_val=.true.)
568 description=
"if the printkey is active prints the physical constants", &
569 default_l_val=.true., lone_keyword_l_val=.true.)
572 CALL keyword_create(keyword, __location__, name=
"SPHERICAL_HARMONICS", &
573 description=
"if the printkey is active prints the spherical harmonics", &
578 description=
"Prints the transformation matrices used by the random number generator", &
579 default_l_val=.false., &
580 lone_keyword_l_val=.true.)
584 description=
"Performs a check of the global (pseudo)random "// &
585 "number generator (RNG) and prints the result", &
586 default_l_val=.false., &
587 lone_keyword_l_val=.true.)
590 CALL keyword_create(keyword, __location__, name=
"GLOBAL_GAUSSIAN_RNG", &
591 description=
"Prints the initial status of the global Gaussian "// &
592 "(pseudo)random number stream which is mostly used for "// &
593 "the velocity initialization", &
594 default_l_val=.false., &
595 lone_keyword_l_val=.true.)
601 NULLIFY (sub_section)
603 CALL create_fm_section(sub_section)
611 CALL create_fm_diag_rules_section(sub_section)
615 CALL create_grid_section(sub_section)
627 SUBROUTINE create_fm_section(section)
628 TYPE(section_type),
POINTER :: section
630 INTEGER :: default_matmul
631 TYPE(keyword_type),
POINTER :: keyword
633 cpassert(.NOT.
ASSOCIATED(section))
635 description=
"Configuration options for the full matrices.", &
636 n_keywords=1, n_subsections=0, repeats=.false.)
641 description=
"Defines the number of rows per scalapack block in "// &
642 "the creation of block cyclic dense matrices ", &
648 description=
"Defines the number of columns per scalapack block in "// &
649 "the creation of vlock cyclic dense matrices ", &
654 CALL keyword_create(keyword, __location__, name=
"FORCE_BLOCK_SIZE", &
655 description=
"Ensure for small matrices that the layout is compatible "// &
656 "with bigger ones, i.e. no subdivision is performed (can break LAPACK!!!).", &
657 usage=
"FORCE_BLOCK_SIZE", &
658 default_l_val=.false., lone_keyword_l_val=.true.)
668 CALL keyword_create(keyword, __location__, name=
"TYPE_OF_MATRIX_MULTIPLICATION", &
669 description=
"Allows to switch between scalapack pxgemm and COSMA pxgemm. "// &
670 "COSMA reduces the communication costs but increases the memory demands. "// &
671 "The performance of Scalapack's pxgemm on GPU's depends "// &
672 "crucially on the BLOCK_SIZES. Make sure optimized kernels are available.", &
673 default_i_val=default_matmul, &
675 enum_c_vals=s2a(
"SCALAPACK",
"PDGEMM",
"COSMA"), &
676 enum_desc=s2a(
"Standard ScaLAPACK pdgemm", &
677 "Alias for ScaLAPACK", &
678 "COSMA is employed. See <https://github.com/eth-cscs/COSMA>."))
683 END SUBROUTINE create_fm_section
690 SUBROUTINE create_fm_diag_rules_section(section)
691 TYPE(section_type),
POINTER :: section
693 TYPE(keyword_type),
POINTER :: keyword
695 cpassert(.NOT.
ASSOCIATED(section))
696 CALL section_create(section, __location__, name=
"FM_DIAG_SETTINGS", &
697 description=
"This section defines a set of heuristic rules which are "// &
698 "used to calculate the optimal number of CPUs, M, needed to diagonalize a "// &
699 "full matrix distributed on N processors (FM type). If M < N, the matrix "// &
700 "is redistributed onto M processors before it is diagonalized. "// &
701 "The optimal value is calculate according to M = ((K+a*x-1)/(a*x))*a, "// &
702 "where K is the size of the matrix, and {a, x} are integers defined below. "// &
703 "The default values have been selected based on timings on a Cray XE6. "// &
704 "Supports diagonalization libraries SL and ELPA (see keyword ELPA_FORCE_REDISTRIBUTE).", &
705 n_keywords=3, n_subsections=0, repeats=.false.)
710 description=
"Parameter used for defining the rule which determines the optimal "// &
711 "number of CPUs needed to diagonalize a full distributed matrix. The optimal "// &
712 "number of CPUs will be an integer multiple of this variable.", &
713 usage=
"PARAMETER_A 4", type_of_var=
integer_t, &
719 description=
"Parameter used for defining the rule which determines the optimal "// &
720 "number of CPUs needed to diagonalize a full distributed matrix. The optimal "// &
721 "number of CPUs will be roughly proportional to this value.", &
722 usage=
"PARAMETER_X 60", type_of_var=
integer_t, &
727 CALL keyword_create(keyword, __location__, name=
"PRINT_FM_REDISTRIBUTE", &
728 description=
"Controls printing of information related to this section. For each "// &
729 "diagonalized matrix, prints the size of the matrix, the optimal number of CPUs, "// &
730 "as well as notifies if the matrix was redistributed. Useful for testing.", &
731 usage=
"PRINT_FM_REDISTRIBUTE", type_of_var=
logical_t, &
732 default_l_val=.false., lone_keyword_l_val=.true.)
736 CALL keyword_create(keyword, __location__, name=
"ELPA_FORCE_REDISTRIBUTE", &
737 description=
"Controls how to perform redistribution when ELPA is used for diagonalization. "// &
738 "By default, redistribution is always performed using the defined rules. "// &
739 "By turning off this keyword, matrices are redistributed only to prevent crashes in the ELPA "// &
740 "library which happens when the original matrix is distributed over too many processors. ", &
741 usage=
"ELPA_FORCE_REDISTRIBUTE", type_of_var=
logical_t, &
742 default_l_val=.true., lone_keyword_l_val=.true.)
746 END SUBROUTINE create_fm_diag_rules_section
753 SUBROUTINE create_grid_section(section)
754 TYPE(section_type),
POINTER :: section
756 TYPE(keyword_type),
POINTER :: keyword
758 cpassert(.NOT.
ASSOCIATED(section))
760 description=
"Configuration options for the grid library, "// &
761 "which performs e.g. the collocate and integrate of the GPW method.", &
762 n_keywords=1, n_subsections=0, repeats=.false.)
766 description=
"Selects the backed used by the grid library.", &
770 enum_c_vals=s2a(
"AUTO",
"REFERENCE",
"CPU",
"DGEMM",
"GPU",
"HIP"), &
771 enum_desc=s2a(
"Let the grid library pick the backend automatically", &
772 "Reference backend implementation", &
773 "Optimized CPU backend", &
774 "Alternative CPU backend based on DGEMM", &
775 "GPU backend optimized for CUDA that also supports HIP", &
776 "HIP backend optimized for ROCm"))
781 description=
"When enabled the reference backend is run in shadow mode "// &
782 "and its results are compared with those from the selected backend. "// &
783 "If the two results differ by too much then the calculation is aborted.", &
784 default_l_val=.false., lone_keyword_l_val=.true.)
789 description=
"When enabled the cpu backend "// &
790 "apply a spherical cutoff on the top of the cube. "// &
791 "There is a performance penalty using it in "// &
792 "combination with the cpu backend but it is on by "// &
793 "default for the regtests", default_l_val=.true., &
794 lone_keyword_l_val=.true.)
798 END SUBROUTINE create_grid_section
collects all references to literature in CP2K as new algorithms / method are included from literature...
integer, save, public schonherr2014
integer, save, public frigo2005
integer, save, public ceriotti2014
methods related to the blacs parallel environment
integer, parameter, public blacs_grid_row
integer, parameter, public blacs_grid_col
integer, parameter, public blacs_grid_square
Routines that link DBCSR and CP2K concepts together.
subroutine, public create_dbcsr_section(section)
Creates the dbcsr section for configuring DBCSR.
used for collecting some of the diagonalization schemes available for cp_fm_type. cp_fm_power also mo...
integer, parameter, public fm_diag_type_cusolver
integer, parameter, public fm_diag_type_dlaf
integer, parameter, public fm_diag_type_scalapack
integer, parameter, public fm_diag_type_default
integer, parameter, public fm_diag_type_elpa
character(len=14), dimension(1), parameter, public elpa_kernel_names
character(len=44), dimension(1), parameter, public elpa_kernel_descriptions
integer, dimension(1), parameter, public elpa_kernel_ids
routines to handle the output, The idea is to remove the decision of wheter to output and what to out...
integer, parameter, public debug_print_level
integer, parameter, public low_print_level
integer, parameter, public medium_print_level
integer, parameter, public high_print_level
integer, parameter, public add_last_numeric
integer, parameter, public silent_print_level
subroutine, public cp_print_key_section_create(print_key_section, location, name, description, print_level, each_iter_names, each_iter_values, add_last, filename, common_iter_levels, citations, unit_str)
creates a print_key section
Fortran API for the grid package, which is written in C.
integer, parameter, public grid_backend_auto
integer, parameter, public grid_backend_gpu
integer, parameter, public grid_backend_hip
integer, parameter, public grid_backend_dgemm
integer, parameter, public grid_backend_cpu
integer, parameter, public grid_backend_ref
Defines the basic variable types.
integer, parameter, public dp
Utilities for string manipulations.
Timing routines for accounting.
integer, parameter, public default_timings_level