scnc
/
quantum-espresso
mirror of https://gitlab.com/QEF/q-e.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
quantum-espresso/Modules/mp_bands.f90

179 lines
5.6 KiB
Fortran
Raw Normal View History

Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!
! Copyright (C) 2013 Quantum ESPRESSO group
! This file is distributed under the terms of the
! GNU General Public License. See the file `License'
! in the root directory of the present distribution,
! or http://www.gnu.org/copyleft/gpl.txt .
!
!----------------------------------------------------------------------------
MODULE mp_bands
!----------------------------------------------------------------------------
!
Calls to mpi_comm_split replaced by interface mp_comm_split git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10675 c92efa57-630b-4861-b058-cf58834340f0
2014-01-11 17:14:24 +08:00
USE mp, ONLY : mp_barrier, mp_bcast, mp_size, mp_rank, mp_comm_split
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
USE parallel_include
!
IMPLICIT NONE
SAVE
!
! ... Band groups (processors within a pool of bands)
! ... Subdivision of pool group, used for parallelization over bands
!
INTEGER :: nbgrp = 1 ! number of band groups
INTEGER :: nproc_bgrp = 1 ! number of processors within a band group
INTEGER :: me_bgrp = 0 ! index of the processor within a band group
INTEGER :: root_bgrp = 0 ! index of the root processor within a band group
INTEGER :: my_bgrp_id = 0 ! index of my band group
INTEGER :: inter_bgrp_comm = 0 ! inter band group communicator
INTEGER :: intra_bgrp_comm = 0 ! intra band group communicator
More explicit variable name and comments to clarify how to activate/deactivate band parallelization inside h_psi and s_psi. Variable "use_bgrp_in_hpsi" should be actually moved in a more appropriate module, since it is related to how h_psi behaves, not strictly to the parallelization over bands git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@12052 c92efa57-630b-4861-b058-cf58834340f0
2016-01-23 19:53:56 +08:00
! Next variable is .T. if band parallelization is performed inside H\psi
! and S\psi, .F. otherwise (band parallelization can be performed outside
! H\psi and S\psi, though)
LOGICAL :: use_bgrp_in_hpsi = .FALSE.
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!
Variables for task group (actually a single one) moved to where it belongs (I think): mp_bands.f90 . Many changes but nothing dangerous. Note that codes not in svn may be broken by this change, but the fix is very simple git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10567 c92efa57-630b-4861-b058-cf58834340f0
2013-11-04 03:16:37 +08:00
! ... "task" groups (for band parallelization of FFT)
!
INTEGER :: ntask_groups = 1 ! number of proc. in an orbital "task group"
!
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
CONTAINS
!
!----------------------------------------------------------------------------
Variables for task group (actually a single one) moved to where it belongs (I think): mp_bands.f90 . Many changes but nothing dangerous. Note that codes not in svn may be broken by this change, but the fix is very simple git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10567 c92efa57-630b-4861-b058-cf58834340f0
2013-11-04 03:16:37 +08:00
SUBROUTINE mp_start_bands( nband_, ntg_, parent_comm )
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!---------------------------------------------------------------------------
!
Unprecise routine name in error messages and modified a few comments. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10663 c92efa57-630b-4861-b058-cf58834340f0
2014-01-07 21:54:17 +08:00
! ... Divide processors (of the "parent_comm" group) into nband_ pools
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
! ... Requires: nband_, read from command line
! ... parent_comm, typically processors of a k-point pool
! ... (intra_pool_comm)
!
IMPLICIT NONE
!
Cleanup of new initialization + restored missing in action ntask_group git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9885 c92efa57-630b-4861-b058-cf58834340f0
2013-01-29 18:31:17 +08:00
INTEGER, INTENT(IN) :: nband_, parent_comm
Variables for task group (actually a single one) moved to where it belongs (I think): mp_bands.f90 . Many changes but nothing dangerous. Note that codes not in svn may be broken by this change, but the fix is very simple git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10567 c92efa57-630b-4861-b058-cf58834340f0
2013-11-04 03:16:37 +08:00
INTEGER, INTENT(IN), OPTIONAL :: ntg_
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!
Calls to mpi_comm_split replaced by interface mp_comm_split git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10675 c92efa57-630b-4861-b058-cf58834340f0
2014-01-11 17:14:24 +08:00
INTEGER :: parent_nproc = 1, parent_mype = 0
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!
#if defined (__MPI)
!
parent_nproc = mp_size( parent_comm )
parent_mype = mp_rank( parent_comm )
!
! ... nband_ must have been previously read from command line argument
! ... by a call to routine get_command_line
!
nbgrp = nband_
!
Unprecise routine name in error messages and modified a few comments. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10663 c92efa57-630b-4861-b058-cf58834340f0
2014-01-07 21:54:17 +08:00
IF ( nbgrp < 1 .OR. nbgrp > parent_nproc ) CALL errore( 'mp_start_bands',&
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
'invalid number of band groups, out of range', 1 )
Unprecise routine name in error messages and modified a few comments. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10663 c92efa57-630b-4861-b058-cf58834340f0
2014-01-07 21:54:17 +08:00
IF ( MOD( parent_nproc, nbgrp ) /= 0 ) CALL errore( 'mp_start_bands', &
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
'n. of band groups must be divisor of parent_nproc', 1 )
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
!
More explicit variable name and comments to clarify how to activate/deactivate band parallelization inside h_psi and s_psi. Variable "use_bgrp_in_hpsi" should be actually moved in a more appropriate module, since it is related to how h_psi behaves, not strictly to the parallelization over bands git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@12052 c92efa57-630b-4861-b058-cf58834340f0
2016-01-23 19:53:56 +08:00
! set logical flag so that band parallelization in H\psi is allowed
! (can be disabled before calling H\psi if not desired)
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
!
More explicit variable name and comments to clarify how to activate/deactivate band parallelization inside h_psi and s_psi. Variable "use_bgrp_in_hpsi" should be actually moved in a more appropriate module, since it is related to how h_psi behaves, not strictly to the parallelization over bands git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@12052 c92efa57-630b-4861-b058-cf58834340f0
2016-01-23 19:53:56 +08:00
use_bgrp_in_hpsi = ( nbgrp > 1 )
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!
! ... Set number of processors per band group
!
nproc_bgrp = parent_nproc / nbgrp
!
! ... set index of band group for this processor ( 0 : nbgrp - 1 )
!
my_bgrp_id = parent_mype / nproc_bgrp
!
! ... set index of processor within the image ( 0 : nproc_image - 1 )
!
me_bgrp = MOD( parent_mype, nproc_bgrp )
!
CALL mp_barrier( parent_comm )
!
! ... the intra_bgrp_comm communicator is created
!
Calls to mpi_comm_split replaced by interface mp_comm_split git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10675 c92efa57-630b-4861-b058-cf58834340f0
2014-01-11 17:14:24 +08:00
CALL mp_comm_split( parent_comm, my_bgrp_id, parent_mype, intra_bgrp_comm )
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!
CALL mp_barrier( parent_comm )
!
! ... the inter_bgrp_comm communicator is created
!
Calls to mpi_comm_split replaced by interface mp_comm_split git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10675 c92efa57-630b-4861-b058-cf58834340f0
2014-01-11 17:14:24 +08:00
CALL mp_comm_split( parent_comm, me_bgrp, parent_mype, inter_bgrp_comm )
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!
Variables for task group (actually a single one) moved to where it belongs (I think): mp_bands.f90 . Many changes but nothing dangerous. Note that codes not in svn may be broken by this change, but the fix is very simple git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@10567 c92efa57-630b-4861-b058-cf58834340f0
2013-11-04 03:16:37 +08:00
IF ( PRESENT(ntg_) ) THEN
ntask_groups = ntg_
END IF
!
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
#endif
RETURN
!
END SUBROUTINE mp_start_bands
!
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
SUBROUTINE set_bgrp_indices(nbnd, ib_start, ib_end)
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
!
IMPLICIT NONE
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
INTEGER, INTENT(IN) :: nbnd
INTEGER, INTENT(OUT) :: ib_start, ib_end
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
INTEGER :: rest, nbnd_per_bgrp
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
rest = mod ( nbnd, nbgrp )
nbnd_per_bgrp = int( nbnd / nbgrp )
IF (rest > my_bgrp_id) THEN
ib_start = my_bgrp_id * (nbnd_per_bgrp+1) + 1
ib_end = (my_bgrp_id+1) * (nbnd_per_bgrp+1)
ELSE
ib_start = my_bgrp_id * nbnd_per_bgrp + rest + 1
ib_end = (my_bgrp_id+1) * nbnd_per_bgrp + rest
ENDIF
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
END SUBROUTINE set_bgrp_indices
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
INTEGER FUNCTION bgrp_start(nbnd)
!
IMPLICIT NONE
INTEGER, INTENT(IN) :: nbnd
INTEGER :: rest, nbnd_per_bgrp
rest = mod ( nbnd, nbgrp )
nbnd_per_bgrp = int( nbnd / nbgrp )
IF (rest > my_bgrp_id) THEN
bgrp_start = my_bgrp_id * (nbnd_per_bgrp+1) + 1
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
ELSE
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
bgrp_start = my_bgrp_id * nbnd_per_bgrp + rest + 1
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
ENDIF
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
END FUNCTION bgrp_start
INTEGER FUNCTION bgrp_end(nbnd)
!
IMPLICIT NONE
INTEGER, INTENT(IN) :: nbnd
INTEGER :: rest, nbnd_per_bgrp
rest = mod ( nbnd, nbgrp )
nbnd_per_bgrp = int( nbnd / nbgrp )
IF (rest > my_bgrp_id) THEN
bgrp_end = (my_bgrp_id+1) * (nbnd_per_bgrp+1)
ELSE
bgrp_end = (my_bgrp_id+1) * nbnd_per_bgrp + rest
ENDIF
END FUNCTION bgrp_end
Module mp_global split into more modules, one per parallelization level. Module mp_global is still there for compatibility and still contains the routine mp_startup that performs the intialization, but shouldn't be any longer used in new developments and should be slowly replaced by specific modules. New module containing command-line options added. Command-line options are read at the beginning of the run and broadcast to all processors (there is no guarantee that command-line options cam be accessed by all mpi processes). Nothing should be broken by these changes but please verify if all parallelization levels work (I have limited access to parallel machines right now) The reason for these changes, and for those that will come soon, is to make image parallelization easier (in particular in NEB but also for other cases). Right now it is a pain, in part because the initialization is inadequate, in part because it is never clear who reads/writes what from/to where. It will take a few weeks before everything converges to a stable state. Meanwhile, please be patient and fix/report problems. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@9876 c92efa57-630b-4861-b058-cf58834340f0
2013-01-28 17:21:12 +08:00
END MODULE mp_bands
band group parallelization slightly modified to make it more flexible, and little more efficient. subroutine init_index_over_band ( comm, nbnd ) that set ibnd_start and ibnd_end variables requiring comm=inter_bgrp_comm is removed and replaced by subroutine set_bgrp_indices ( nbnd, ibnd_start, ibnd_end ) implementing the same relationships between its arguments but: - forcing the use of inter_bgrp_comm from the same mp_bands module, - returning ibnd_start and ibnd_end as explicit outputs that are not anymore kept in the module. In this way other quantities can be distributes if needed in any given routine without too many non-local effects. For compatibility with TDDFPT, that uses the bgrp parallelization and loads ibnd_start/ibnd_end trhough mp_global module, these two variables are moved in a dedicated module mp_bands_TDDFPT included in Module/mp_bands.f90. This is done to avoid too much invasive changes in a code i don't know well. In this way the needed changes are very localized and transparent, the code compiles correctly so I think it should work exactly as before. In my opinion the two variables should be moved somewhere inside TDDFPT. Band parallelization is extended to h_psi(lda,n,m,psi,hpsi) and s_psi routines (only when .not.exx_is_active because otherwise it is already used inside vexx) for generic values of m (of course it gives a speedup only when m is not too small compared to nbgrp but it works also if m < nbgrp ). Compatibility with task groups has not be explored but should not be conceptually different from how it works in the exx case. git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@11835 c92efa57-630b-4861-b058-cf58834340f0
2015-11-07 08:06:40 +08:00
!
!
MODULE mp_bands_TDDFPT
!
! NB: These two varialbles used to be in mp_bands and are loaded from mp_global in TDDFPT
! I think they would better stay in a TDDFPT specific module but leave them here not to
! be too invasive on a code I don't know well. SdG
!
INTEGER :: ibnd_start = 0 ! starting band index used in bgrp parallelization
INTEGER :: ibnd_end = 0 ! ending band index used in bgrp parallelization
!
END MODULE mp_bands_TDDFPT
!