80#include "./base/base_uses.f90"
85 LOGICAL,
PRIVATE,
PARAMETER :: debug_this_module = .true.
86 CHARACTER(len=*),
PARAMETER,
PRIVATE :: moduleN =
'input_cp2k_global'
100 INTEGER :: default_dgemm
105 cpassert(.NOT.
ASSOCIATED(section))
107 description=
"Section with general information on which kind "// &
108 "of simulation to perform and parameters for the whole PROGRAM", &
109 n_keywords=7, n_subsections=1, repeats=.false.)
113 description=
"how to distribute the processors on the 2d grid needed "// &
114 "by BLACS (and thus SCALAPACK)", usage=
"BLACS_GRID SQUARE", &
116 enum_desc=
s2a(
"Distribution by matrix blocks",
"Distribution by matrix rows", &
117 "Distribution by matrix columns"), &
122 CALL keyword_create(keyword, __location__, name=
"BLACS_REPEATABLE", &
123 description=
"Use a topology for BLACS collectives that is guaranteed to be repeatable "// &
124 "on homogeneous architectures", &
125 usage=
"BLACS_REPEATABLE", &
126 default_l_val=.false., lone_keyword_l_val=.true.)
130 CALL keyword_create(keyword, __location__, name=
"PREFERRED_DIAG_LIBRARY", &
131 description=
"Specifies the diagonalization library to be used. If not available, "// &
132 "the ScaLAPACK library is used. Complex (k-point) matrices use ELPA only when it is "// &
133 "requested explicitly", &
134 usage=
"PREFERRED_DIAG_LIBRARY ELPA", &
140 enum_c_vals=
s2a(
"ELPA",
"ScaLAPACK",
"SL",
"CUSOLVER",
"DLAF"), &
141 enum_desc=
s2a(
"ELPA library", &
142 "ScaLAPACK library", &
143 "ScaLAPACK library (shorthand)", &
144 "cuSOLVER (CUDA GPU library)", &
145 "DLA-Future (CUDA/HIP GPU library)"), &
150 CALL keyword_create(keyword, __location__, name=
"DIRECT_GENERALIZED_DIAGONALIZATION", &
151 description=
"Request direct generalized eigenvalue problem diagonalization "// &
152 "without a CP2K-side Cholesky reduction in supported dense matrix paths. "// &
153 "The eigensolver is still selected by PREFERRED_DIAG_LIBRARY.", &
154 usage=
"DIRECT_GENERALIZED_DIAGONALIZATION", &
155 default_l_val=.false., lone_keyword_l_val=.true.)
159 CALL keyword_create(keyword, __location__, name=
"PREFERRED_CHOLESKY_LIBRARY", &
160 description=
"Specifies Cholesky decomposition library to be used. If not available, "// &
161 "the ScaLAPACK library is used", &
162 usage=
"PREFERRED_CHOLESKY_LIBRARY DLAF", &
166 enum_c_vals=
s2a(
"ScaLAPACK",
"SL",
"DLAF"), &
167 enum_desc=
s2a(
"ScaLAPACK library", &
168 "ScaLAPACK library (shorthand)", &
169 "DLA-Future (CUDA/HIP GPU library)"), &
174#if defined(__SPLA) && defined(__OFFLOAD_GEMM)
179 CALL keyword_create(keyword, __location__, name=
"PREFERRED_DGEMM_LIBRARY", &
180 description=
"Specifies the DGEMM library to be used. If not available, "// &
181 "the BLAS routine is used. This keyword affects some DGEMM calls in the WFC code and turns on their "// &
182 "acceleration with SpLA. This keyword affects only local DGEMM calls, not the calls to PDGEMM "// &
183 "(see keyword FM%TYPE_OF_MATRIX_MULTIPLICATION).", &
184 usage=
"PREFERRED_DGEMM_LIBRARY SPLA", &
185 default_i_val=default_dgemm, &
187 enum_c_vals=
s2a(
"SPLA",
"BLAS"), &
188 enum_desc=
s2a(
"SPLA library",
"BLAS library"))
192 CALL keyword_create(keyword, __location__, name=
"EPS_CHECK_DIAG", &
193 description=
"Check that the orthonormality of the eigenvectors after a diagonalization "// &
194 "fulfills the specified numerical accuracy. A negative threshold value disables the check.", &
195 usage=
"EPS_CHECK_DIAG 1.0E-14", &
196 default_r_val=-1.0_dp)
201 description=
"Specifies the kernel to be used when ELPA is in use for real "// &
202 "symmetric matrices", &
210 CALL keyword_create(keyword, __location__, name=
"ELPA_COMPLEX_KERNEL", &
211 description=
"Specifies the kernel to be used when ELPA is in use for complex "// &
212 "Hermitian matrices (k-points)", &
220 CALL keyword_create(keyword, __location__, name=
"ELPA_NEIGVEC_MIN", &
221 description=
"Minimum number of eigenvectors for the use of the eigensolver from "// &
222 "the ELPA library. The eigensolver from the ScaLAPACK library is used as fallback "// &
223 "for all smaller cases", &
224 usage=
"ELPA_NEIGVEC_MIN 32", &
230 description=
"For ELPA, enable a blocked QR step when reducing the input matrix "// &
231 "to banded form before diagonalization. Requires ELPA version 201505 or newer and "// &
232 "is automatically deactivated otherwise. QR is activated only when the matrix size "// &
233 "is suitable. Keyword ELPA_PRINT helps identify suitable cases. Can accelerate "// &
234 "diagonalization for suitable matrices.", &
236 default_l_val=
elpa_qr, lone_keyword_l_val=.true.)
240 CALL keyword_create(keyword, __location__, name=
"ELPA_ONE_STAGE", &
241 description=
"For ELPA, enable the one-stage solver (instead of the two-stage solver). "// &
242 "Please note, ELPA_QR and ELPA_KERNEL settings may be ignored.", &
243 usage=
"ELPA_ONE_STAGE", &
249 description=
"Controls the printing of ELPA diagonalization information. "// &
250 "Useful for testing purposes, especially together with keyword ELPA_QR.", &
251 usage=
"ELPA_PRINT", default_l_val=
elpa_print, lone_keyword_l_val=.true.)
255 CALL keyword_create(keyword, __location__, name=
"DLAF_NEIGVEC_MIN", &
256 description=
"Minimum number of eigenvectors for the use of the eigensolver from "// &
257 "the DLA-Future library. The eigensolver from the ScaLAPACK library is used as fallback "// &
258 "for all smaller cases", &
259 usage=
"DLAF_NEIGVEC_MIN 512", &
264 CALL keyword_create(keyword, __location__, name=
"DLAF_CHOLESKY_N_MIN", &
265 description=
"Minimum matrix size for the use of the Cholesky decomposition from "// &
266 "the DLA-Future library. The Cholesky decomposition from the ScaLAPACK library is used as fallback "// &
267 "for all smaller cases", &
268 usage=
"DLAF_CHOLESKY_N_MIN 512", &
274 keyword, __location__, name=
"PREFERRED_FFT_LIBRARY", &
275 description=
"Specifies the FFT library which should be preferred. "// &
276 "If it is not available, use FFTW3 if this is linked in, if FFTW3 is not available use FFTSG. "// &
277 "Improved performance with FFTW3 can be obtained specifying a proper value for FFTW_PLAN_TYPE. "// &
278 "Contrary to earlier CP2K versions, all libraries will result in the same grids, "// &
279 "i.e. the subset of grids which all FFT libraries can transform. "// &
280 "FFTW3 is often (close to) optimal, and well tested with CP2K.", &
281 usage=
"PREFERRED_FFT_LIBRARY FFTW3", &
285 enum_c_vals=
s2a(
"FFTSG",
"FFTW3",
"FFTW"), &
286 enum_desc=
s2a(
"Stefan Goedecker's FFT (FFTSG), always available, "// &
287 "will be used in case a FFT library is specified and not available.", &
288 "a fast portable FFT library. Recommended. "// &
289 "See also the FFTW_PLAN_TYPE, and FFTW_WISDOM_FILE_NAME keywords.", &
290 "Same as FFTW3 (for compatibility with CP2K 2.3)"))
294 NULLIFY (sub_section)
295 CALL section_create(sub_section, __location__, name=
"preferred_integral_library", &
296 description=
"Selects the implementation of individual Gaussian integral operations.", &
297 n_keywords=2, n_subsections=0, repeats=.false.)
300 description=
"Preferred library for two-center Coulomb integrals (Coulomb_2c). "// &
301 "Libint is used by default when available; otherwise the native implementation is used.", &
302 usage=
"COULOMB_2C LIBINT", &
303 enum_c_vals=
s2a(
"NATIVE",
"LIBINT"), &
305 enum_desc=
s2a(
"Use CP2K's native implementation.", &
306 "Use the Libint implementation."), &
312 description=
"Preferred library for three-center Coulomb integrals (Coulomb_3c). "// &
313 "Libint is used by default when available; otherwise the native implementation is used.", &
314 usage=
"COULOMB_3C LIBINT", &
315 enum_c_vals=
s2a(
"NATIVE",
"LIBINT"), &
317 enum_desc=
s2a(
"Use CP2K's native implementation.", &
318 "Use the Libint implementation."), &
326 CALL keyword_create(keyword, __location__, name=
"FFTW_WISDOM_FILE_NAME", &
327 description=
"The name of the file that contains wisdom (pre-planned FFTs) for use with FFTW3. "// &
328 "Using wisdom can significantly speed up the FFTs (see the FFTW homepage for details). "// &
329 "Note that wisdom is not transferable between different computer (architectures). "// &
330 "Wisdom can be generated using the fftw-wisdom tool that is part of the fftw installation. "// &
331 "cp2k/tools/cp2k-wisdom is a script that contains some additional info, and can help "// &
332 "to generate a useful default for /etc/fftw/wisdom or particular values for a given simulation.", &
333 usage=
"FFTW_WISDOM_FILE_NAME wisdom.dat", default_lc_val=
"/etc/fftw/wisdom")
337 CALL keyword_create(keyword, __location__, name=
"FFTW_PLAN_TYPE", &
338 description=
"FFTW can have improved performance if it is allowed to plan with "// &
339 "explicit measurements which strategy is best for a given FFT. "// &
340 "While a plan based on measurements is generally faster, "// &
341 "differences in machine load will lead to different plans for the same input file, "// &
342 "and thus numerics for the FFTs will be slightly different from run to run. "// &
343 "PATIENT planning is recommended for long ab initio MD runs.", &
344 usage=
"FFTW_PLAN_TYPE PATIENT", &
348 enum_c_vals=
s2a(
"ESTIMATE", &
352 enum_desc=
s2a(
"Quick estimate, no runtime measurements.", &
353 "Quick measurement, somewhat faster FFTs.", &
354 "Measurements trying a wider range of possibilities.", &
355 "Measurements trying all possibilities - use with caution."))
359 CALL keyword_create(keyword, __location__, name=
"FFT_POOL_SCRATCH_LIMIT", &
360 description=
"Limits the memory usage of the FFT scratch pool, potentially reducing efficiency a bit", &
361 usage=
"FFT_POOL_SCRATCH_LIMIT {INTEGER}", default_i_val=15)
366 description=
"All-to-all communication (FFT) should use single precision", &
367 usage=
"ALLTOALL_SGL YES", &
368 default_l_val=.false., lone_keyword_l_val=.true.)
373 variants=[
"IOLEVEL"], &
374 description=
"How much output is written out.", &
375 usage=
"PRINT_LEVEL HIGH", &
377 s2a(
"SILENT",
"LOW",
"MEDIUM",
"HIGH",
"DEBUG"), &
378 enum_desc=
s2a(
"Almost no output", &
379 "Little output",
"Quite some output",
"Lots of output", &
380 "Everything is written out, useful for debugging purposes only"), &
387 keyword, __location__, name=
"PROGRAM_NAME", &
388 variants=[
"PROGRAM"], &
389 description=
"Which program should be run", &
390 usage=
"PROGRAM_NAME {STRING}", &
391 enum_c_vals=
s2a(
"ATOM",
"FARMING",
"TEST",
"CP2K",
"OPTIMIZE_INPUT",
"OPTIMIZE_BASIS",
"TMC",
"MC_ANALYSIS",
"SWARM"), &
392 enum_desc=
s2a(
"Runs single atom calculations", &
393 "Runs N independent jobs in a single run", &
394 "Do some benchmarking and testing", &
395 "Runs one of the CP2K package", &
396 "A tool to optimize parameters in a CP2K input", &
397 "A tool to create a MOLOPT or ADMM basis for a given set"// &
398 " of training structures", &
399 "Runs Tree Monte Carlo algorithm using additional input file(s)", &
400 "Runs (Tree) Monte Carlo trajectory file analysis", &
401 "Runs swarm based calculation"), &
409 variants=[
"PROJECT"], &
410 description=
"Name of the project (used to build the name of the "// &
411 "trajectory, and other files generated by the program)", &
412 usage=
"PROJECT_NAME {STRING}", &
413 default_c_val=
"PROJECT")
417 CALL keyword_create(keyword, __location__, name=
"OUTPUT_FILE_NAME", &
418 description=
"Name of the output file. "// &
419 "Relevant only if automatically started (through farming for example). "// &
420 "If empty uses the project name as basis for it.", &
421 usage=
"OUTPUT_FILE_NAME {filename}", default_lc_val=
"")
426 keyword, __location__, name=
"RUN_TYPE", &
427 description=
"Selects the top-level task CP2K should run, such as an energy, "// &
428 "energy-and-force, molecular dynamics, geometry optimization, or response calculation.", &
429 usage=
"RUN_TYPE MD", &
432 enum_c_vals=
s2a(
"NONE",
"ENERGY",
"ENERGY_FORCE",
"MD",
"GEO_OPT", &
433 "MC",
"DEBUG",
"BSSE",
"LR",
"PINT",
"VIBRATIONAL_ANALYSIS", &
434 "BAND",
"CELL_OPT",
"WFN_OPT",
"WAVEFUNCTION_OPTIMIZATION", &
435 "MOLECULAR_DYNAMICS",
"GEOMETRY_OPTIMIZATION",
"MONTECARLO", &
436 "LINEAR_RESPONSE",
"NORMAL_MODES",
"RT_PROPAGATION", &
437 "EHRENFEST_DYN",
"TAMC",
"TMC",
"DRIVER",
"NEGF",
"MIMIC",
"RTP", &
446 enum_desc=
s2a(
"Perform no tasks",
"Computes energy",
"Computes energy and forces", &
447 "Molecular Dynamics",
"Geometry Optimization",
"Monte Carlo", &
448 "Performs a Debug analysis",
"Basis set superposition error",
"Linear Response", &
449 "Path integral",
"Vibrational analysis",
"Band methods", &
450 "Cell optimization. Both cell vectors and atomic positions are optimised.", &
451 "Alias for ENERGY",
"Alias for ENERGY",
"Alias for MD",
"Alias for GEO_OPT", &
452 "Alias for MC",
"Alias for LR",
"Alias for VIBRATIONAL_ANALYSIS", &
453 "Real Time propagation run (fixed ionic positions)", &
454 "Ehrenfest dynamics (using real time propagation of the wavefunction)", &
455 "Temperature Accelerated Monte Carlo (TAMC)", &
456 "Tree Monte Carlo (TMC), a pre-sampling MC algorithm", &
457 "i-PI driver mode", &
458 "Non-equilibrium Green's function method", &
459 "Run as a client in a simulation through the MiMiC framework", &
460 "Alias for RT_PROPAGATION", &
461 "Minimum tracking linear response calculation for Hubbard U and Hund's J"))
466 variants=[
"WALLTI"], &
467 description=
"Maximum execution time for this run. Time in seconds or in HH:MM:SS.", &
468 usage=
"WALLTIME {real} or {HH:MM:SS}", default_lc_val=
"")
473 description=
"If the input should be echoed to the output with all the "// &
474 "defaults made explicit", &
475 usage=
"ECHO_INPUT NO", default_l_val=.false., lone_keyword_l_val=.true.)
479 CALL keyword_create(keyword, __location__, name=
"ECHO_ALL_HOSTS", &
480 description=
"Echo a list of hostname and pid for all MPI processes.", &
481 usage=
"ECHO_ALL_HOSTS NO", default_l_val=.false., lone_keyword_l_val=.true.)
485 CALL keyword_create(keyword, __location__, name=
"ENABLE_MPI_IO", &
486 description=
"Enable MPI parallelization for all supported I/O routines "// &
487 "Currently, only cube file writer/reader routines use MPI I/O. Disabling "// &
488 "this flag might speed up calculations dominated by I/O.", &
489 usage=
"ENABLE_MPI_IO FALSE", default_l_val=.true., lone_keyword_l_val=.true.)
494 description=
"If a debug trace of the execution of the program should be written", &
496 default_l_val=.false., lone_keyword_l_val=.true.)
501 description=
"For parallel TRACEd runs: only the master node writes output.", &
502 usage=
"TRACE_MASTER", &
503 default_l_val=.true., lone_keyword_l_val=.true.)
508 keyword, __location__, name=
"TRACE_MAX", &
509 description=
"Limit the total number a given subroutine is printed in the trace. Accounting is not influenced.", &
510 usage=
"TRACE_MAX 100", default_i_val=huge(0))
515 keyword, __location__, name=
"TRACE_ROUTINES", &
516 description=
"A list of routines to trace. If left empty all routines are traced. Accounting is not influenced.", &
517 usage=
"TRACE_ROUTINES {routine_name1} {routine_name2} ...", type_of_var=
char_t, &
523 keyword, __location__, name=
"FLUSH_SHOULD_FLUSH", &
524 description=
"Flush output regularly, enabling this option might degrade performance significantly on certain machines.", &
525 usage=
"FLUSH_SHOULD_FLUSH", &
526 default_l_val=.true., lone_keyword_l_val=.true.)
531 description=
"At the end of the run write a callgraph to file, "// &
532 "which contains detailed timing informations. "// &
533 "This callgraph can be viewed e.g. with the open-source program kcachegrind.", &
534 usage=
"CALLGRAPH {NONE|MASTER|ALL}", &
536 enum_c_vals=
s2a(
"NONE",
"MASTER",
"ALL"), &
537 enum_desc=
s2a(
"No callgraph gets written", &
538 "Only the master process writes his callgraph", &
539 "All processes write their callgraph (into a separate files)."), &
544 CALL keyword_create(keyword, __location__, name=
"CALLGRAPH_FILE_NAME", &
545 description=
"Name of the callgraph file, which is written at the end of the run. "// &
546 "If not specified the project name will be used as filename.", &
547 usage=
"CALLGRAPH_FILE_NAME {filename}", default_lc_val=
"")
552 description=
"Initial seed for the global (pseudo)random number generator "// &
553 "to create a stream of normally Gaussian distributed random numbers. "// &
554 "Exactly 1 or 6 positive integer values are expected. A single value is "// &
555 "replicated to fill up the full seed array with 6 numbers.", &
558 usage=
"SEED {INTEGER} .. {INTEGER}", &
559 default_i_vals=[2000])
564 description=
"Some sections of the input structure are deallocated when not needed,"// &
565 " and reallocated only when used. This reduces the required maximum memory.", &
567 default_l_val=.false., lone_keyword_l_val=.true.)
572 "Controls the printing of the timing report at the end of CP2K execution", &
576 description=
"Specify % of CPUTIME above which the contribution will be inserted in the"// &
577 " final timing report (e.g. 0.02 = 2%)", &
578 usage=
"THRESHOLD {REAL}", &
579 default_r_val=0.02_dp)
583 CALL keyword_create(keyword, __location__, name=
"SORT_BY_SELF_TIME", &
584 description=
"Sort the final timing report by the average self (exclusive) time instead of the "// &
585 "total (inclusive) time of a routine", &
586 usage=
"SORT_BY_SELF_TIME on", &
587 default_l_val=.false., lone_keyword_l_val=.true.)
591 CALL keyword_create(keyword, __location__, name=
"REPORT_MAXLOC", &
592 description=
"Report the rank with the slowest maximum self timing."// &
593 " Can be used to debug hard- or software."// &
594 " Also enables ECHO_ALL_HOSTS to link rank to hostname.", &
595 usage=
"REPORT_MAXLOC on", &
596 default_l_val=.false., lone_keyword_l_val=.true.)
601 description=
"Include message_passing calls in the timing report (useful with CALLGRAPH).", &
602 usage=
"TIME_MPI .FALSE.", &
603 default_l_val=.true., lone_keyword_l_val=.true.)
607 CALL keyword_create(keyword, __location__, name=
"TIMINGS_LEVEL", &
608 description=
"Specify the level of timings report. "// &
609 "Possible values are: 0 (report only CP2K root timer), 1 (all timers).", &
610 usage=
"TIMINGS_LEVEL 1", &
619 "Controls the printing of the references relevant to the calculations performed", &
625 description=
"controls the printing of initialization controlled by the global section", &
631 "controls the printing of physical and mathematical constants", &
634 CALL keyword_create(keyword, __location__, name=
"BASIC_DATA_TYPES", &
635 description=
"Controls the printing of the basic data types.", &
636 default_l_val=.false., lone_keyword_l_val=.true.)
640 description=
"if the printkey is active prints the physical constants", &
641 default_l_val=.true., lone_keyword_l_val=.true.)
644 CALL keyword_create(keyword, __location__, name=
"SPHERICAL_HARMONICS", &
645 description=
"if the printkey is active prints the spherical harmonics", &
650 description=
"Prints the transformation matrices used by the random number generator", &
651 default_l_val=.false., &
652 lone_keyword_l_val=.true.)
656 description=
"Performs a check of the global (pseudo)random "// &
657 "number generator (RNG) and prints the result", &
658 default_l_val=.false., &
659 lone_keyword_l_val=.true.)
662 CALL keyword_create(keyword, __location__, name=
"GLOBAL_GAUSSIAN_RNG", &
663 description=
"Prints the initial status of the global Gaussian "// &
664 "(pseudo)random number stream which is mostly used for "// &
665 "the velocity initialization", &
666 default_l_val=.false., &
667 lone_keyword_l_val=.true.)
673 NULLIFY (sub_section)
675 CALL create_fm_section(sub_section)
683 CALL create_fm_diag_rules_section(sub_section)
687 CALL create_grid_section(sub_section)
699 SUBROUTINE create_fm_section(section)
702 INTEGER :: default_matmul
705 cpassert(.NOT.
ASSOCIATED(section))
707 description=
"Configuration options for the full matrices.", &
708 n_keywords=1, n_subsections=0, repeats=.false.)
713 description=
"Defines the number of rows per scalapack block in "// &
714 "the creation of block cyclic dense matrices. "// &
715 "Use an internal default if zero or negative.", &
721 description=
"Defines the number of columns per scalapack block in "// &
722 "the creation of vlock cyclic dense matrices. "// &
723 "Use an internal default if zero or negative.", &
728 CALL keyword_create(keyword, __location__, name=
"FORCE_BLOCK_SIZE", &
729 description=
"Ensure for small matrices that the layout is compatible "// &
730 "with bigger ones, i.e. no subdivision is performed (can break LAPACK).", &
731 usage=
"FORCE_BLOCK_SIZE", &
732 default_l_val=.false., lone_keyword_l_val=.true.)
742 CALL keyword_create(keyword, __location__, name=
"TYPE_OF_MATRIX_MULTIPLICATION", &
743 description=
"Allows to switch between scalapack pxgemm and COSMA pxgemm. "// &
744 "COSMA reduces the communication costs but increases the memory demands. "// &
745 "The performance of Scalapack's pxgemm on GPU's depends "// &
746 "crucially on the BLOCK_SIZES. Make sure optimized kernels are available.", &
747 default_i_val=default_matmul, &
749 enum_c_vals=
s2a(
"SCALAPACK",
"PDGEMM",
"COSMA"), &
750 enum_desc=
s2a(
"Standard ScaLAPACK pdgemm", &
751 "Alias for ScaLAPACK", &
752 "COSMA is employed. See <https://github.com/eth-cscs/COSMA>."))
757 END SUBROUTINE create_fm_section
764 SUBROUTINE create_fm_diag_rules_section(section)
769 cpassert(.NOT.
ASSOCIATED(section))
770 CALL section_create(section, __location__, name=
"FM_DIAG_SETTINGS", &
771 description=
"This section defines a set of heuristic rules which are "// &
772 "used to calculate the optimal number of CPUs, M, needed to diagonalize a "// &
773 "full matrix distributed on N processors (FM type). If M < N, the matrix "// &
774 "is redistributed onto M processors before it is diagonalized. "// &
775 "The optimal value is calculate according to M = ((K+a*x-1)/(a*x))*a, "// &
776 "where K is the size of the matrix, and {a, x} are integers defined below. "// &
777 "The default values have been selected based on timings on a Cray XE6. "// &
778 "Supports diagonalization libraries SL and ELPA (see keyword ELPA_FORCE_REDISTRIBUTE).", &
779 n_keywords=3, n_subsections=0, repeats=.false.)
784 description=
"Parameter used for defining the rule which determines the optimal "// &
785 "number of CPUs needed to diagonalize a full distributed matrix. The optimal "// &
786 "number of CPUs will be an integer multiple of this variable.", &
787 usage=
"PARAMETER_A 4", type_of_var=
integer_t, &
793 description=
"Parameter used for defining the rule which determines the optimal "// &
794 "number of CPUs needed to diagonalize a full distributed matrix. The optimal "// &
795 "number of CPUs will be roughly proportional to this value.", &
796 usage=
"PARAMETER_X 60", type_of_var=
integer_t, &
801 CALL keyword_create(keyword, __location__, name=
"PRINT_FM_REDISTRIBUTE", &
802 description=
"Controls printing of information related to this section. For each "// &
803 "diagonalized matrix, prints the size of the matrix, the optimal number of CPUs, "// &
804 "as well as notifies if the matrix was redistributed. Useful for testing.", &
805 usage=
"PRINT_FM_REDISTRIBUTE", type_of_var=
logical_t, &
806 default_l_val=.false., lone_keyword_l_val=.true.)
810 CALL keyword_create(keyword, __location__, name=
"ELPA_FORCE_REDISTRIBUTE", &
811 description=
"Controls how to perform redistribution when ELPA is used for diagonalization. "// &
812 "By default, redistribution is always performed using the defined rules. "// &
813 "By turning off this keyword, matrices are redistributed only to prevent crashes in the ELPA "// &
814 "library which happens when the original matrix is distributed over too many processors.", &
815 usage=
"ELPA_FORCE_REDISTRIBUTE", type_of_var=
logical_t, &
816 default_l_val=.true., lone_keyword_l_val=.true.)
820 END SUBROUTINE create_fm_diag_rules_section
827 SUBROUTINE create_grid_section(section)
832 cpassert(.NOT.
ASSOCIATED(section))
834 description=
"Configuration options for the grid library, "// &
835 "which performs e.g. the collocate and integrate of the GPW method.", &
836 n_keywords=1, n_subsections=0, repeats=.false.)
840 description=
"Selects the backed used by the grid library.", &
844 enum_c_vals=
s2a(
"AUTO",
"REFERENCE",
"CPU",
"DGEMM",
"GPU"), &
845 enum_desc=
s2a(
"Let the grid library pick the backend automatically", &
846 "Reference backend implementation", &
847 "Optimized CPU backend", &
848 "Alternative CPU backend based on DGEMM", &
849 "GPU backend optimized for NVIDIA and AMD GPU"))
854 description=
"When enabled the reference backend is run in shadow mode "// &
855 "and its results are compared with those from the selected backend. "// &
856 "If the two results differ by too much then the calculation is aborted.", &
857 default_l_val=.false., lone_keyword_l_val=.true.)
862 description=
"When enabled the cpu backend "// &
863 "apply a spherical cutoff on the top of the cube. "// &
864 "There is a performance penalty using it in "// &
865 "combination with the cpu backend but it is on by "// &
866 "default for the regtests", default_l_val=.true., &
867 lone_keyword_l_val=.true.)
871 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
Wrapper for ELPA (complex matrices, i.e. cp_cfm_type)
character(len=44), dimension(1), parameter, public elpa_c_kernel_descriptions
character(len=14), dimension(1), parameter, public elpa_c_kernel_names
integer, dimension(1), parameter, public elpa_c_kernel_ids
Routines that link DBCSR and CP2K concepts together.
subroutine, public create_dbcsr_section(section)
Creates the dbcsr section for configuring DBCSR.
various cholesky decomposition related routines
integer, parameter, public fm_cholesky_type_dlaf
integer, parameter, public fm_cholesky_type_default
integer, parameter, public fm_cholesky_type_scalapack
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
logical, save, public elpa_qr
character(len=14), dimension(1), parameter, public elpa_kernel_names
logical, save, public elpa_print
character(len=44), dimension(1), parameter, public elpa_kernel_descriptions
integer, dimension(1), parameter, public elpa_kernel_ids
logical, save, public elpa_one_stage
represent the structure of a full matrix
integer function, public cp_fm_struct_get_nrow_block()
...
integer function, public cp_fm_struct_get_ncol_block()
...
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_dgemm
integer, parameter, public grid_backend_cpu
integer, parameter, public grid_backend_ref
Library choices for electronic integral APIs.
integer, parameter, public default_coulomb_library
integer, parameter, public library_native
integer, parameter, public library_libint
Defines the basic variable types.
integer, parameter, public dp
Utilities for string manipulations.
Timing routines for accounting.
integer, parameter, public default_timings_level