dbcsr_mpiwrap.F Source File


Contents

Source Code


Source Code

# 1 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F" 1
!--------------------------------------------------------------------------------------------------!
! Copyright (C) by the DBCSR developers group - All rights reserved                                !
! This file is part of the DBCSR library.                                                          !
!                                                                                                  !
! For information on the license, see the LICENSE file.                                            !
! For further information please visit https://dbcsr.cp2k.org                                      !
! SPDX-License-Identifier: GPL-2.0+                                                                !
!--------------------------------------------------------------------------------------------------!

MODULE dbcsr_mpiwrap
   !! Interface to the message passing library MPI
   USE ISO_C_BINDING, ONLY: C_F_POINTER, &
                            C_PTR
   USE dbcsr_kinds, ONLY: &
      dp, int_4, int_4_size, int_8, int_8_size, real_4, real_4_size, real_8, &
      real_8_size
   USE dbcsr_machine, ONLY: m_abort, m_hostnm

#include "base/dbcsr_base_uses.f90"
# 1 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.fypp" 1
# 9 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.fypp"

# 31 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.fypp"
# 21 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F" 2

#if defined(__parallel)
   USE mpi
! subroutines: unfortunately, mpi implementations do not provide interfaces for all subroutines (problems with types and ranks explosion),
!              we do not quite know what is in the module, so we can not include any....
!              to nevertheless get checking for what is included, we use the mpi module without use clause, getting all there is
! USE mpi, ONLY: mpi_allgather, mpi_allgatherv, mpi_alloc_mem, mpi_allreduce, mpi_alltoall, mpi_alltoallv, mpi_bcast,&
!                mpi_cart_coords, mpi_cart_create, mpi_cart_get, mpi_cart_rank, mpi_cart_sub, mpi_dims_create, mpi_file_close,&
!                mpi_file_get_size, mpi_file_open, mpi_file_read_at_all, mpi_file_read_at, mpi_file_write_at_all,&
!                mpi_file_write_at, mpi_free_mem, mpi_gather, mpi_gatherv, mpi_get_address, mpi_group_translate_ranks, mpi_irecv,&
!                mpi_isend, mpi_recv, mpi_reduce, mpi_reduce_scatter, mpi_rget, mpi_scatter, mpi_send,&
!                mpi_sendrecv, mpi_sendrecv_replace, mpi_testany, mpi_waitall, mpi_waitany, mpi_win_create
! functions
! USE mpi, ONLY: mpi_wtime
! constants
! USE mpi, ONLY: MPI_DOUBLE_PRECISION, MPI_DOUBLE_COMPLEX, MPI_REAL, MPI_COMPLEX, MPI_ANY_TAG,&
!                MPI_ANY_SOURCE, MPI_COMM_NULL, MPI_REQUEST_NULL, MPI_WIN_NULL, MPI_STATUS_SIZE, MPI_STATUS_IGNORE, MPI_STATUSES_IGNORE, &
!                MPI_ADDRESS_KIND, MPI_OFFSET_KIND, MPI_MODE_CREATE, MPI_MODE_RDONLY, MPI_MODE_WRONLY,&
!                MPI_MODE_RDWR, MPI_MODE_EXCL, MPI_COMM_SELF, MPI_COMM_WORLD, MPI_THREAD_FUNNELED,&
!                MPI_ERRORS_RETURN, MPI_SUCCESS, MPI_MAX_PROCESSOR_NAME, MPI_MAX_ERROR_STRING, MPI_IDENT,&
!                MPI_UNEQUAL, MPI_MAX, MPI_SUM, MPI_INFO_NULL, MPI_IN_PLACE, MPI_CONGRUENT, MPI_SIMILAR, MPI_MIN, MPI_SOURCE,&
!                MPI_TAG, MPI_INTEGER8, MPI_INTEGER, MPI_MAXLOC, MPI_2INTEGER, MPI_MINLOC, MPI_LOGICAL, MPI_2DOUBLE_PRECISION,&
!                MPI_LOR, MPI_CHARACTER, MPI_BOTTOM, MPI_MODE_NOCHECK, MPI_2REAL
#endif

   IMPLICIT NONE
   PRIVATE

   ! parameters that might be needed
#if defined(__parallel)
   INTEGER, PARAMETER     :: MP_STD_REAL = MPI_DOUBLE_PRECISION
   INTEGER, PARAMETER     :: MP_STD_COMPLEX = MPI_DOUBLE_COMPLEX
   INTEGER, PARAMETER     :: MP_STD_HALF_REAL = MPI_REAL
   INTEGER, PARAMETER     :: MP_STD_HALF_COMPLEX = MPI_COMPLEX

   LOGICAL, PARAMETER :: dbcsr_is_parallel = .TRUE.
   INTEGER, PARAMETER, PUBLIC :: mp_any_tag = MPI_ANY_TAG
   INTEGER, PARAMETER, PUBLIC :: mp_any_source = MPI_ANY_SOURCE
   INTEGER, PARAMETER, PUBLIC :: mp_comm_null = MPI_COMM_NULL
   INTEGER, PARAMETER, PUBLIC :: mp_comm_self = MPI_COMM_SELF
   INTEGER, PARAMETER, PUBLIC :: mp_comm_world = MPI_COMM_WORLD
   INTEGER, PARAMETER, PUBLIC :: mp_request_null = MPI_REQUEST_NULL
   INTEGER, PARAMETER, PUBLIC :: mp_win_null = MPI_WIN_NULL
   INTEGER, PARAMETER, PUBLIC :: mp_status_size = MPI_STATUS_SIZE
   INTEGER, PARAMETER, PUBLIC :: mp_proc_null = MPI_PROC_NULL
   ! Set max allocatable memory by MPI to 2 GiByte
   INTEGER(KIND=MPI_ADDRESS_KIND), PARAMETER, PRIVATE :: mp_max_memory_size = HUGE(INT(1, KIND=int_4))

   INTEGER, PARAMETER, PUBLIC :: mp_max_library_version_string = MPI_MAX_LIBRARY_VERSION_STRING
   INTEGER, PARAMETER, PUBLIC :: mp_max_processor_name = MPI_MAX_PROCESSOR_NAME

   INTEGER, PARAMETER, PUBLIC :: file_offset = MPI_OFFSET_KIND
   INTEGER, PARAMETER, PUBLIC :: address_kind = MPI_ADDRESS_KIND
   INTEGER, PARAMETER, PUBLIC :: file_amode_create = MPI_MODE_CREATE
   INTEGER, PARAMETER, PUBLIC :: file_amode_rdonly = MPI_MODE_RDONLY
   INTEGER, PARAMETER, PUBLIC :: file_amode_wronly = MPI_MODE_WRONLY
   INTEGER, PARAMETER, PUBLIC :: file_amode_rdwr = MPI_MODE_RDWR
   INTEGER, PARAMETER, PUBLIC :: file_amode_excl = MPI_MODE_EXCL
   INTEGER, PARAMETER, PUBLIC :: file_amode_append = MPI_MODE_APPEND
#else
   LOGICAL, PARAMETER :: dbcsr_is_parallel = .FALSE.
   INTEGER, PARAMETER, PUBLIC :: mp_any_tag = -1
   INTEGER, PARAMETER, PUBLIC :: mp_any_source = -2
   INTEGER, PARAMETER, PUBLIC :: mp_comm_null = -3
   INTEGER, PARAMETER, PUBLIC :: mp_comm_self = -11
   INTEGER, PARAMETER, PUBLIC :: mp_comm_world = -12
   INTEGER, PARAMETER, PUBLIC :: mp_request_null = -4
   INTEGER, PARAMETER, PUBLIC :: mp_win_null = -5
   INTEGER, PARAMETER, PUBLIC :: mp_status_size = -6
   INTEGER, PARAMETER, PUBLIC :: mp_proc_null = -7
   INTEGER, PARAMETER, PUBLIC :: mp_max_library_version_string = 1
   INTEGER, PARAMETER, PUBLIC :: mp_max_processor_name = 1

   INTEGER, PARAMETER, PUBLIC :: file_offset = int_8
   INTEGER, PARAMETER, PUBLIC :: address_kind = int_8
   INTEGER, PARAMETER, PUBLIC :: file_amode_create = 1
   INTEGER, PARAMETER, PUBLIC :: file_amode_rdonly = 2
   INTEGER, PARAMETER, PUBLIC :: file_amode_wronly = 4
   INTEGER, PARAMETER, PUBLIC :: file_amode_rdwr = 8
   INTEGER, PARAMETER, PUBLIC :: file_amode_excl = 64
   INTEGER, PARAMETER, PUBLIC :: file_amode_append = 128
#endif

   ! we need to fix this to a given number (crossing fingers)
   ! so that the serial code using Fortran stream IO and the MPI have the same sizes.
   INTEGER, PARAMETER, PUBLIC :: mpi_character_size = 1
   INTEGER, PARAMETER, PUBLIC :: mpi_integer_size = 4

   CHARACTER(LEN=*), PARAMETER, PRIVATE :: moduleN = 'dbcsr_mpiwrap'

#if defined(__parallel)
   ! internal reference counter used to debug communicator leaks
   INTEGER, PRIVATE, SAVE :: debug_comm_count = 0
#endif

   ! init and error
   PUBLIC :: mp_world_init, mp_world_finalize
   PUBLIC :: mp_get_comm_count
   PUBLIC :: mp_abort

   ! performance gathering
   PUBLIC :: mp_perf_env_type
   PUBLIC :: mp_perf_env_retain, mp_perf_env_release
   PUBLIC :: add_mp_perf_env, rm_mp_perf_env, get_mp_perf_env, describe_mp_perf_env
   PUBLIC :: has_mp_perf_env

   ! informational / generation of sub comms
   PUBLIC :: mp_environ, mp_comm_compare, mp_cart_coords, mp_rank_compare
   PUBLIC :: mp_cart_create, mp_dims_create, mp_cart_rank, mp_cart_sub, mp_comm_free
   PUBLIC :: mp_comm_dup, mp_comm_split, mp_comm_split_direct
   PUBLIC :: dbcsr_is_parallel
   PUBLIC :: mp_probe

   ! message passing
   PUBLIC :: mp_bcast, mp_sum, mp_sum_partial, mp_max, mp_maxloc, mp_minloc, mp_min, mp_prod, mp_sync
   PUBLIC :: mp_isync, mp_isum
   PUBLIC :: mp_gather, mp_alltoall, mp_sendrecv, mp_allgather, mp_iallgather
   PUBLIC :: mp_isend, mp_irecv, mp_ibcast
   PUBLIC :: mp_isendrecv, mp_wait, mp_waitall, mp_waitany, mp_testany
   PUBLIC :: mp_testall, mp_iscatter, mp_test
   PUBLIC :: mp_gatherv
   PUBLIC :: mp_send, mp_recv

   ! Memory management
   PUBLIC :: mp_allocate, mp_deallocate

   ! MPI re-ordering
   PUBLIC :: mp_reordering

   ! I/O
   PUBLIC :: mp_file_open, mp_file_close
   PUBLIC :: mp_file_delete
   PUBLIC :: mp_file_write_at
   PUBLIC :: mp_file_write_at_all, mp_file_read_at_all
   PUBLIC :: mp_file_get_size
   PUBLIC :: mp_file_get_position

   ! some 'advanced types' currently only used for dbcsr
   PUBLIC :: mp_type_descriptor_type
   PUBLIC :: mp_type_make
   PUBLIC :: mp_type_size

   ! one-sided communication
   PUBLIC :: mp_win_create, mp_win_free, mp_win_lock_all, &
             mp_win_unlock_all, mp_rget, mp_win_flush_all

   ! vector types
   PUBLIC :: mp_type_indexed_make_r, mp_type_indexed_make_d, &
             mp_type_indexed_make_c, mp_type_indexed_make_z, &
             mp_type_indexed_make_i, mp_type_indexed_make_l

   ! misc
   PUBLIC :: mp_get_library_version, mp_get_processor_name

   ! assumed to be private

! Interface declarations for non-data-oriented subroutines.

   INTERFACE mp_environ
      MODULE PROCEDURE mp_environ_l, mp_environ_c, mp_environ_c2
   END INTERFACE

   INTERFACE mp_waitall
      MODULE PROCEDURE mp_waitall_1, mp_waitall_2
   END INTERFACE

   INTERFACE mp_testall
      MODULE PROCEDURE mp_testall_tv
   END INTERFACE

   INTERFACE mp_test
      MODULE PROCEDURE mp_test_1
   END INTERFACE

   INTERFACE mp_testany
      MODULE PROCEDURE mp_testany_1, mp_testany_2
   END INTERFACE

   !
   ! interfaces to deal easily with scalars / vectors / matrices / ...
   ! of the different types (integers, doubles, logicals, characters)
   !
   INTERFACE mp_minloc
      MODULE PROCEDURE mp_minloc_dv
   END INTERFACE

   INTERFACE mp_maxloc
      MODULE PROCEDURE mp_maxloc_dv
   END INTERFACE

   INTERFACE mp_bcast
# 211 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_bcast_i, mp_bcast_iv, mp_bcast_im, mp_bcast_i3, mp_bcast_l, mp_bcast_lv, mp_bcast_lm, mp_bcast_l3,&
# 211 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_bcast_d, mp_bcast_dv, mp_bcast_dm, mp_bcast_d3, mp_bcast_r, mp_bcast_rv, mp_bcast_rm, mp_bcast_r3, mp_bcast_z,&
# 211 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_bcast_zv, mp_bcast_zm, mp_bcast_z3, mp_bcast_c, mp_bcast_cv, mp_bcast_cm, mp_bcast_c3, mp_bcast_b, mp_bcast_bv,&
# 211 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_bcast_av, mp_bcast_am
# 211 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_ibcast
# 212 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_ibcast_i, mp_ibcast_iv, mp_ibcast_l, mp_ibcast_lv, mp_ibcast_d, mp_ibcast_dv, mp_ibcast_r, mp_ibcast_rv,&
# 212 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_ibcast_z, mp_ibcast_zv, mp_ibcast_c, mp_ibcast_cv
# 212 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE

   INTERFACE mp_sum
      MODULE PROCEDURE mp_sum_i, mp_sum_iv, mp_sum_im, mp_sum_im3, mp_sum_im4, &
         mp_sum_l, mp_sum_lv, mp_sum_lm, mp_sum_lm3, mp_sum_lm4, &
         mp_sum_r, mp_sum_rv, mp_sum_rm, mp_sum_rm3, mp_sum_rm4, &
         mp_sum_d, mp_sum_dv, mp_sum_dm, mp_sum_dm3, mp_sum_dm4, &
         mp_sum_c, mp_sum_cv, mp_sum_cm, mp_sum_cm3, mp_sum_cm4, &
         mp_sum_z, mp_sum_zv, mp_sum_zm, mp_sum_zm3, mp_sum_zm4, &
         mp_sum_root_iv, mp_sum_root_im, &
         mp_sum_root_lv, mp_sum_root_lm, &
         mp_sum_root_rv, mp_sum_root_rm, &
         mp_sum_root_dv, mp_sum_root_dm, &
         mp_sum_root_cv, mp_sum_root_cm, &
         mp_sum_root_zv, mp_sum_root_zm
      MODULE PROCEDURE mp_sum_b, mp_sum_bv
   END INTERFACE

   INTERFACE mp_isum
# 230 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_isum_iv, mp_isum_lv, mp_isum_dv, mp_isum_rv, mp_isum_zv, mp_isum_cv, mp_isum_bv
# 230 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_sum_partial
# 231 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_sum_partial_im, mp_sum_partial_lm, mp_sum_partial_dm, mp_sum_partial_rm, mp_sum_partial_zm,&
# 231 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_sum_partial_cm
# 231 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_max
# 232 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_max_i, mp_max_iv, mp_max_l, mp_max_lv, mp_max_d, mp_max_dv, mp_max_r, mp_max_rv, mp_max_z, mp_max_zv,&
# 232 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_max_c, mp_max_cv
# 232 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_min
# 233 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_min_i, mp_min_iv, mp_min_l, mp_min_lv, mp_min_d, mp_min_dv, mp_min_r, mp_min_rv, mp_min_z, mp_min_zv,&
# 233 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_min_c, mp_min_cv
# 233 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_prod
# 234 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_prod_i, mp_prod_l, mp_prod_d, mp_prod_r, mp_prod_z, mp_prod_c
# 234 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_gather
# 235 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_gather_i, mp_gather_im, mp_gather_iv, mp_gather_l, mp_gather_lm, mp_gather_lv, mp_gather_d,&
# 235 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_gather_dm, mp_gather_dv, mp_gather_r, mp_gather_rm, mp_gather_rv, mp_gather_z, mp_gather_zm, mp_gather_zv,&
# 235 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_gather_c, mp_gather_cm, mp_gather_cv
# 235 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_gatherv
# 236 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_gatherv_iv, mp_gatherv_lv, mp_gatherv_dv, mp_gatherv_rv, mp_gatherv_zv, mp_gatherv_cv
# 236 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE

   INTERFACE mp_allgather
     !! @todo move allgatherv to a separate declaration
      MODULE PROCEDURE &
         mp_allgather_i, mp_allgather_i2, &
         mp_allgather_i12, mp_allgather_i23, mp_allgather_i34, &
         mp_allgather_i22, &
         mp_allgather_l, mp_allgather_l2, &
         mp_allgather_l12, mp_allgather_l23, mp_allgather_l34, &
         mp_allgather_l22, &
         mp_allgather_r, mp_allgather_r2, &
         mp_allgather_r12, mp_allgather_r23, mp_allgather_r34, &
         mp_allgather_r22, &
         mp_allgather_d, mp_allgather_d2, &
         mp_allgather_d12, mp_allgather_d23, mp_allgather_d34, &
         mp_allgather_d22, &
         mp_allgather_c, mp_allgather_c2, &
         mp_allgather_c12, mp_allgather_c23, mp_allgather_c34, &
         mp_allgather_c22, &
         mp_allgather_z, mp_allgather_z2, &
         mp_allgather_z12, mp_allgather_z23, mp_allgather_z34, &
         mp_allgather_z22, &
         mp_allgatherv_iv, &
         mp_allgatherv_lv, &
         mp_allgatherv_rv, &
         mp_allgatherv_dv, &
         mp_allgatherv_cv, &
         mp_allgatherv_zv
   END INTERFACE

   INTERFACE mp_iallgather
      MODULE PROCEDURE &
         mp_iallgather_i, mp_iallgather_l, &
         mp_iallgather_r, mp_iallgather_d, &
         mp_iallgather_c, mp_iallgather_z, &
         mp_iallgather_i11, mp_iallgather_l11, &
         mp_iallgather_r11, mp_iallgather_d11, &
         mp_iallgather_c11, mp_iallgather_z11, &
         mp_iallgather_i13, mp_iallgather_l13, &
         mp_iallgather_r13, mp_iallgather_d13, &
         mp_iallgather_c13, mp_iallgather_z13, &
         mp_iallgather_i22, mp_iallgather_l22, &
         mp_iallgather_r22, mp_iallgather_d22, &
         mp_iallgather_c22, mp_iallgather_z22, &
         mp_iallgather_i24, mp_iallgather_l24, &
         mp_iallgather_r24, mp_iallgather_d24, &
         mp_iallgather_c24, mp_iallgather_z24, &
         mp_iallgather_i33, mp_iallgather_l33, &
         mp_iallgather_r33, mp_iallgather_d33, &
         mp_iallgather_c33, mp_iallgather_z33, &
         mp_iallgatherv_iv, mp_iallgatherv_iv2, &
         mp_iallgatherv_lv, mp_iallgatherv_lv2, &
         mp_iallgatherv_rv, mp_iallgatherv_rv2, &
         mp_iallgatherv_dv, mp_iallgatherv_dv2, &
         mp_iallgatherv_cv, mp_iallgatherv_cv2, &
         mp_iallgatherv_zv, mp_iallgatherv_zv2
   END INTERFACE

   INTERFACE mp_iscatter
      MODULE PROCEDURE mp_iscatter_i, &
         mp_iscatter_l, &
         mp_iscatter_r, &
         mp_iscatter_d, &
         mp_iscatter_c, &
         mp_iscatter_z, &
         mp_iscatter_iv2, &
         mp_iscatter_lv2, &
         mp_iscatter_rv2, &
         mp_iscatter_dv2, &
         mp_iscatter_cv2, &
         mp_iscatter_zv2, &
         mp_iscatterv_iv, &
         mp_iscatterv_lv, &
         mp_iscatterv_rv, &
         mp_iscatterv_dv, &
         mp_iscatterv_cv, &
         mp_iscatterv_zv
   END INTERFACE

   INTERFACE mp_alltoall
# 316 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_alltoall_i, mp_alltoall_i22, mp_alltoall_i44, mp_alltoall_i11v, mp_alltoall_l, mp_alltoall_l22,&
# 316 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_alltoall_l44, mp_alltoall_l11v, mp_alltoall_d, mp_alltoall_d22, mp_alltoall_d44, mp_alltoall_d11v, mp_alltoall_r,&
# 316 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_alltoall_r22, mp_alltoall_r44, mp_alltoall_r11v, mp_alltoall_z, mp_alltoall_z22, mp_alltoall_z44, mp_alltoall_z11v,&
# 316 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_alltoall_c, mp_alltoall_c22, mp_alltoall_c44, mp_alltoall_c11v
# 316 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE

   INTERFACE mp_send
# 318 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_send_i, mp_send_iv, mp_send_l, mp_send_lv, mp_send_d, mp_send_dv, mp_send_r, mp_send_rv, mp_send_z,&
# 318 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_send_zv, mp_send_c, mp_send_cv
# 318 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_recv
# 319 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_recv_i, mp_recv_iv, mp_recv_l, mp_recv_lv, mp_recv_d, mp_recv_dv, mp_recv_r, mp_recv_rv, mp_recv_z,&
# 319 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_recv_zv, mp_recv_c, mp_recv_cv
# 319 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_sendrecv
# 320 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_sendrecv_iv, mp_sendrecv_lv, mp_sendrecv_dv, mp_sendrecv_rv, mp_sendrecv_zv, mp_sendrecv_cv
# 320 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_isendrecv
# 321 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_isendrecv_i, mp_isendrecv_iv, mp_isendrecv_l, mp_isendrecv_lv, mp_isendrecv_d, mp_isendrecv_dv,&
# 321 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_isendrecv_r, mp_isendrecv_rv, mp_isendrecv_z, mp_isendrecv_zv, mp_isendrecv_c, mp_isendrecv_cv
# 321 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE

   INTERFACE mp_isend
# 323 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_isend_iv, mp_isend_im2, mp_isend_lv, mp_isend_lm2, mp_isend_dv, mp_isend_dm2, mp_isend_rv, mp_isend_rm2,&
# 323 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_isend_zv, mp_isend_zm2, mp_isend_cv, mp_isend_cm2, mp_isend_bv, mp_isend_custom
# 323 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_irecv
# 324 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_irecv_iv, mp_irecv_im2, mp_irecv_lv, mp_irecv_lm2, mp_irecv_dv, mp_irecv_dm2, mp_irecv_rv, mp_irecv_rm2,&
# 324 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_irecv_zv, mp_irecv_zm2, mp_irecv_cv, mp_irecv_cm2, mp_irecv_bv, mp_irecv_custom
# 324 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE

   INTERFACE mp_win_create
# 326 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_win_create_iv, mp_win_create_lv, mp_win_create_dv, mp_win_create_rv, mp_win_create_zv, mp_win_create_cv
# 326 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_rget
# 327 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_rget_iv, mp_rget_lv, mp_rget_dv, mp_rget_rv, mp_rget_zv, mp_rget_cv
# 327 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_allocate
# 328 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_allocate_i, mp_allocate_l, mp_allocate_d, mp_allocate_r, mp_allocate_z, mp_allocate_c
# 328 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_deallocate
# 329 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_deallocate_i, mp_deallocate_l, mp_deallocate_d, mp_deallocate_r, mp_deallocate_z, mp_deallocate_c
# 329 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE

   INTERFACE mp_type_make
# 331 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_type_make_i, mp_type_make_l, mp_type_make_d, mp_type_make_r, mp_type_make_z, mp_type_make_c,&
# 331 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_type_make_struct
# 331 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE

   INTERFACE mp_file_write_at
# 333 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_file_write_at_i, mp_file_write_at_iv, mp_file_write_at_l, mp_file_write_at_lv, mp_file_write_at_d,&
# 333 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_file_write_at_dv, mp_file_write_at_r, mp_file_write_at_rv, mp_file_write_at_z, mp_file_write_at_zv,&
# 333 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_file_write_at_c, mp_file_write_at_cv, mp_file_write_at_ch
# 333 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_file_write_at_all
# 334 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_file_write_at_all_i, mp_file_write_at_all_iv, mp_file_write_at_all_l, mp_file_write_at_all_lv,&
# 334 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_file_write_at_all_d, mp_file_write_at_all_dv, mp_file_write_at_all_r, mp_file_write_at_all_rv,&
# 334 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_file_write_at_all_z, mp_file_write_at_all_zv, mp_file_write_at_all_c, mp_file_write_at_all_cv,&
# 334 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_file_write_at_all_ch
# 334 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_file_read_at_all
# 335 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_file_read_at_all_i, mp_file_read_at_all_iv, mp_file_read_at_all_l, mp_file_read_at_all_lv,&
# 335 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_file_read_at_all_d, mp_file_read_at_all_dv, mp_file_read_at_all_r, mp_file_read_at_all_rv, mp_file_read_at_all_z,&
# 335 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
          & mp_file_read_at_all_zv, mp_file_read_at_all_c, mp_file_read_at_all_cv, mp_file_read_at_all_ch
# 335 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE

#if defined(__parallel)
   INTERFACE mp_alloc_mem
# 338 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_alloc_mem_i, mp_alloc_mem_l, mp_alloc_mem_d, mp_alloc_mem_r, mp_alloc_mem_z, mp_alloc_mem_c
# 338 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
   INTERFACE mp_free_mem
# 339 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      MODULE PROCEDURE mp_free_mem_i, mp_free_mem_l, mp_free_mem_d, mp_free_mem_r, mp_free_mem_z, mp_free_mem_c
# 339 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
   END INTERFACE
#endif

! Type declarations
   TYPE mp_indexing_meta_type
      INTEGER, DIMENSION(:), POINTER :: index, chunks
   END TYPE mp_indexing_meta_type

   TYPE mp_type_descriptor_type
      INTEGER :: type_handle
      INTEGER :: length
#if defined(__parallel)
      INTEGER(kind=mpi_address_kind) :: base
#endif
      INTEGER(kind=int_4), DIMENSION(:), POINTER :: data_i
      INTEGER(kind=int_8), DIMENSION(:), POINTER :: data_l
      REAL(kind=real_4), DIMENSION(:), POINTER :: data_r
      REAL(kind=real_8), DIMENSION(:), POINTER :: data_d
      COMPLEX(kind=real_4), DIMENSION(:), POINTER :: data_c
      COMPLEX(kind=real_8), DIMENSION(:), POINTER :: data_z
      TYPE(mp_type_descriptor_type), DIMENSION(:), POINTER :: subtype
      INTEGER :: vector_descriptor(2)
      LOGICAL :: has_indexing
      TYPE(mp_indexing_meta_type) :: index_descriptor
   END TYPE mp_type_descriptor_type

   TYPE mp_file_indexing_meta_type
      INTEGER, DIMENSION(:), POINTER   :: index
      INTEGER(kind=address_kind), &
         DIMENSION(:), POINTER         :: chunks
   END TYPE mp_file_indexing_meta_type

   ! type internally used to store message passing performance indicators
! **************************************************************************************************
   TYPE mp_perf_type
      CHARACTER(LEN=20) :: name
      INTEGER :: count
      REAL(KIND=dp) :: msg_size
   END TYPE mp_perf_type

   INTEGER, PARAMETER :: MAX_PERF = 28

! **************************************************************************************************
   TYPE mp_perf_env_type
      !private
      INTEGER :: ref_count, id_nr
      TYPE(mp_perf_type), DIMENSION(MAX_PERF) :: mp_perfs
   END TYPE mp_perf_env_type

! **************************************************************************************************
   TYPE mp_perf_env_p_type
      TYPE(mp_perf_env_type), POINTER         :: mp_perf_env => Null()
   END TYPE mp_perf_env_p_type

   ! introduce a stack of mp_perfs, first index is the stack pointer, for convenience is replacing
   INTEGER, PARAMETER :: max_stack_size = 10
   INTEGER            :: stack_pointer = 0
   ! target attribute needed as a hack around ifc 7.1 bug
   TYPE(mp_perf_env_p_type), DIMENSION(max_stack_size), TARGET, SAVE :: mp_perf_stack

   CHARACTER(LEN=20), PARAMETER :: sname(MAX_PERF) = &
                                   (/"MP_Group            ", "MP_Bcast            ", "MP_Allreduce        ", &
                                     "MP_Gather           ", "MP_Sync             ", "MP_Alltoall         ", &
                                     "MP_SendRecv         ", "MP_ISendRecv        ", "MP_Wait             ", &
                                     "MP_comm_split       ", "MP_ISend            ", "MP_IRecv            ", &
                                     "MP_Send             ", "MP_Recv             ", "MP_Memory           ", &
                                     "MP_Put              ", "MP_Get              ", "MP_Fence            ", &
                                     "MP_Win_Lock         ", "MP_Win_Create       ", "MP_Win_Free         ", &
                                     "MP_IBcast           ", "MP_IAllreduce       ", "MP_IScatter         ", &
                                     "MP_RGet             ", "MP_Isync            ", "MP_Read_All         ", &
                                     "MP_Write_All        "/)

   ! we make some assumptions on the length of INTEGERS, REALS and LOGICALS
   INTEGER, PARAMETER :: intlen = BIT_SIZE(0)/8
   INTEGER, PARAMETER :: reallen = 8
   INTEGER, PARAMETER :: loglen = BIT_SIZE(0)/8
   INTEGER, PARAMETER :: charlen = 1
   INTEGER, SAVE, PRIVATE :: last_mp_perf_env_id = 0

CONTAINS

   SUBROUTINE mp_world_init(mp_comm)
      !! initializes the system default communicator
      !! @note
      !! should only be called once

      INTEGER, INTENT(OUT)                     :: mp_comm
         !! [output] : handle of the default communicator
#if defined(__parallel)
      INTEGER                                  :: ierr
!$    INTEGER                                  :: provided_tsl
!$    LOGICAL                                  :: no_threading_support

#if defined(__NO_MPI_THREAD_SUPPORT_CHECK)
      ! Hack that does not request or check MPI thread support level.
      ! User asserts that the MPI library will work correctly with
      ! threads.
!
!$    no_threading_support = .TRUE.
#else
      ! Does the right thing when using OpenMP: requests that the MPI
      ! library supports funneled mode and verifies that the MPI library
      ! provides that support.
      !
      ! Developers: Only the master thread will ever make calls to the
      ! MPI library.
!
!$    no_threading_support = .FALSE.
#endif
!$    IF (no_threading_support) THEN
         CALL mpi_init(ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_init @ mp_world_init")
!$    ELSE
!$OMP MASTER
!$       CALL mpi_init_thread(MPI_THREAD_FUNNELED, provided_tsl, ierr)
!$       IF (ierr /= 0) CALL mp_stop(ierr, "mpi_init_thread @ mp_world_init")
!$       IF (provided_tsl .LT. MPI_THREAD_FUNNELED) THEN
!$          CALL mp_stop(0, "MPI library does not support the requested level of threading (MPI_THREAD_FUNNELED).")
!$       END IF
!$OMP END MASTER
!$    END IF
      CALL mpi_comm_set_errhandler(MPI_COMM_WORLD, MPI_ERRORS_RETURN, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_set_errhandler @ mp_world_init")
      mp_comm = MPI_COMM_WORLD
      debug_comm_count = 1
#else
      mp_comm = 0
#endif
      CALL add_mp_perf_env()
   END SUBROUTINE mp_world_init

   FUNCTION mp_get_comm_count()
     !! Return the current number of communicators
      INTEGER :: mp_get_comm_count

      mp_get_comm_count = 0
#if defined(__parallel)
      mp_get_comm_count = debug_comm_count
#endif
   END FUNCTION mp_get_comm_count

   SUBROUTINE mp_reordering(mp_comm, mp_new_comm, ranks_order)
      !! re-create the system default communicator with a different MPI
      !! rank order
      !! @note
      !! should only be called once, at very beginning of CP2K run

      INTEGER, INTENT(IN)                      :: mp_comm
         !! [output] : handle of the default communicator
      INTEGER, INTENT(out)                     :: mp_new_comm
      INTEGER, DIMENSION(:), CONTIGUOUS        :: ranks_order

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_reordering'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: newcomm, newgroup, oldgroup
#endif

      CALL timeset(routineN, handle)
      ierr = 0
#if defined(__parallel)

      CALL mpi_comm_group(mp_comm, oldgroup, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_group @ "//routineN)
      CALL mpi_group_incl(oldgroup, SIZE(ranks_order), ranks_order, newgroup, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_group_incl @ "//routineN)

      CALL mpi_comm_create(mp_comm, newgroup, newcomm, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_create @ "//routineN)

      CALL mpi_group_free(oldgroup, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_group_free @ "//routineN)
      CALL mpi_group_free(newgroup, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_group_free @ "//routineN)

      ! update the system default communicator
      mp_new_comm = newcomm
      debug_comm_count = debug_comm_count + 1

#else
      MARK_USED(ranks_order)
      mp_new_comm = mp_comm
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_reordering

   SUBROUTINE mp_world_finalize()
      !! finalizes the system default communicator

#if defined(__parallel)
      INTEGER                                  :: ierr
      CALL mpi_barrier(MPI_COMM_WORLD, ierr) ! call mpi directly to avoid 0 stack pointer
      CALL rm_mp_perf_env()
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_barrier @ mp_world_finalize")
      debug_comm_count = debug_comm_count - 1
      IF (debug_comm_count .NE. 0) THEN
         ! A bug, we're leaking or double-freeing communicators. Needs to be fixed where the leak happens.
         ! Memory leak checking might be helpful to locate the culprit
         DBCSR_ABORT("mp_world_finalize: assert failed: leaking communicators")
      END IF
      CALL mpi_finalize(ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_finalize @ mp_world_finalize")
#else
      CALL rm_mp_perf_env()
#endif

   END SUBROUTINE mp_world_finalize

! all the following routines should work for a given communicator, not MPI_WORLD

   SUBROUTINE add_mp_perf_env(perf_env)
      !! start and stop the performance indicators
      !! for every call to start there has to be (exactly) one call to stop
      !! @note
      !! can be used to measure performance of a sub-part of a program.
      !! timings measured here will not show up in the outer start/stops
      !! Doesn't need a fresh communicator

      TYPE(mp_perf_env_type), OPTIONAL, POINTER          :: perf_env

      stack_pointer = stack_pointer + 1
      IF (stack_pointer > max_stack_size) THEN
         DBCSR_ABORT("stack_pointer too large : mpiwrap @ add_mp_perf_env")
      END IF
      NULLIFY (mp_perf_stack(stack_pointer)%mp_perf_env)
      IF (PRESENT(perf_env)) THEN
         mp_perf_stack(stack_pointer)%mp_perf_env => perf_env
         IF (ASSOCIATED(perf_env)) CALL mp_perf_env_retain(perf_env)
      END IF
      IF (.NOT. ASSOCIATED(mp_perf_stack(stack_pointer)%mp_perf_env)) THEN
         CALL mp_perf_env_create(mp_perf_stack(stack_pointer)%mp_perf_env)
      END IF
   END SUBROUTINE add_mp_perf_env

   SUBROUTINE mp_perf_env_create(perf_env)
      TYPE(mp_perf_env_type), OPTIONAL, POINTER          :: perf_env

      INTEGER                                            :: i, stat

      NULLIFY (perf_env)
      ALLOCATE (perf_env, stat=stat)
      IF (stat /= 0) THEN
         DBCSR_ABORT("allocation failed in mp_perf_env_create")
      END IF
      last_mp_perf_env_id = last_mp_perf_env_id + 1
      perf_env%id_nr = last_mp_perf_env_id
      perf_env%ref_count = 1
      DO i = 1, MAX_PERF
         perf_env%mp_perfs(i)%name = sname(i)
         perf_env%mp_perfs(i)%count = 0
         perf_env%mp_perfs(i)%msg_size = 0.0_dp
      END DO

   END SUBROUTINE mp_perf_env_create

   SUBROUTINE mp_perf_env_release(perf_env)
      TYPE(mp_perf_env_type), POINTER                    :: perf_env

      IF (ASSOCIATED(perf_env)) THEN
         IF (perf_env%ref_count < 1) THEN
            DBCSR_ABORT("invalid ref_count: mpiwrap @ mp_perf_env_release")
         END IF
         perf_env%ref_count = perf_env%ref_count - 1
         IF (perf_env%ref_count == 0) THEN
            DEALLOCATE (perf_env)
         END IF
      END IF
      NULLIFY (perf_env)
   END SUBROUTINE mp_perf_env_release

   SUBROUTINE mp_perf_env_retain(perf_env)
      TYPE(mp_perf_env_type), POINTER                    :: perf_env

      IF (.NOT. ASSOCIATED(perf_env)) THEN
         DBCSR_ABORT("unassociated perf_env: mpiwrap @ mp_perf_env_retain")
      END IF
      IF (perf_env%ref_count < 1) THEN
         DBCSR_ABORT("invalid ref_count: mpiwrap @ mp_perf_env_retain")
      END IF
      perf_env%ref_count = perf_env%ref_count + 1
   END SUBROUTINE mp_perf_env_retain

!.. reports the performance counters for the MPI run
   SUBROUTINE mp_perf_env_describe(perf_env, iw)
      TYPE(mp_perf_env_type), POINTER          :: perf_env
      INTEGER, INTENT(IN)                      :: iw

#if defined(__parallel)
      INTEGER                                  :: i
      REAL(KIND=dp)                            :: vol
#endif

      IF (.NOT. ASSOCIATED(perf_env)) THEN
         DBCSR_ABORT("unassociated perf_env : mpiwrap @ mp_perf_env_describe")
      END IF
      IF (perf_env%ref_count < 1) THEN
         DBCSR_ABORT("invalid perf_env%ref_count : mpiwrap @ mp_perf_env_describe")
      END IF
#if defined(__parallel)
      IF (iw > 0) THEN
         WRITE (iw, '( " -", 77X, "-" )')
         WRITE (iw, '( " -", 21X, A, 21X, "-" )') ' DBCSR MESSAGE PASSING PERFORMANCE '
         WRITE (iw, '( " -", 77X, "-" )')
         WRITE (iw, '( 1X, 79("-"))')
         WRITE (iw, '( A, A, A )') ' ROUTINE', '             CALLS ', &
            '     AVE VOLUME [Bytes]'
         DO i = 1, MAX_PERF

            IF (perf_env%mp_perfs(i)%count > 0) THEN
               vol = perf_env%mp_perfs(i)%msg_size/REAL(perf_env%mp_perfs(i)%count, KIND=dp)
               IF (vol < 1.0_dp) THEN
                  WRITE (iw, '(1X,A15,T17,I10)') &
                     ADJUSTL(perf_env%mp_perfs(i)%name), perf_env%mp_perfs(i)%count
               ELSE
                  WRITE (iw, '(1X,A15,T17,I10,T40,F11.0)') &
                     ADJUSTL(perf_env%mp_perfs(i)%name), perf_env%mp_perfs(i)%count, &
                     vol
               END IF
            END IF

         END DO
         WRITE (iw, '( 1X, 79("-"))')
      END IF
#else
      MARK_USED(iw)
#endif
   END SUBROUTINE mp_perf_env_describe

   SUBROUTINE rm_mp_perf_env()
      IF (stack_pointer < 1) THEN
         DBCSR_ABORT("no perf_env in the stack : mpiwrap @ rm_mp_perf_env")
      END IF
      CALL mp_perf_env_release(mp_perf_stack(stack_pointer)%mp_perf_env)
      stack_pointer = stack_pointer - 1
   END SUBROUTINE rm_mp_perf_env

   PURE FUNCTION has_mp_perf_env() RESULT(res)
      LOGICAL :: res

      res = .FALSE.
      IF (stack_pointer < 1) RETURN
      IF (.NOT. ASSOCIATED(mp_perf_stack(stack_pointer)%mp_perf_env)) RETURN
      res = .TRUE.
   END FUNCTION has_mp_perf_env

   FUNCTION get_mp_perf_env() RESULT(res)
      TYPE(mp_perf_env_type), POINTER                    :: res

      IF (stack_pointer < 1) THEN
         DBCSR_ABORT("no perf_env in the stack : mpiwrap @ get_mp_perf_env")
      END IF
      res => mp_perf_stack(stack_pointer)%mp_perf_env
   END FUNCTION get_mp_perf_env

   SUBROUTINE describe_mp_perf_env(scr)
      INTEGER, INTENT(in)                                :: scr

      TYPE(mp_perf_env_type), POINTER                    :: perf_env

      perf_env => get_mp_perf_env()
      CALL mp_perf_env_describe(perf_env, scr)
   END SUBROUTINE describe_mp_perf_env

#if defined(__parallel)
   SUBROUTINE add_perf(perf_id, msg_size)
      !! adds the performance informations of one call
      INTEGER, INTENT(in)                      :: perf_id
      INTEGER, INTENT(in)                      :: msg_size

      TYPE(mp_perf_type), POINTER              :: mp_perf

      IF (stack_pointer < 1) return
      IF (.NOT. ASSOCIATED(mp_perf_stack(stack_pointer)%mp_perf_env)) return

      mp_perf => mp_perf_stack(stack_pointer)%mp_perf_env%mp_perfs(perf_id)
      mp_perf%count = mp_perf%count + 1
      mp_perf%msg_size = mp_perf%msg_size + REAL(msg_size, dp)
   END SUBROUTINE add_perf
#endif

   SUBROUTINE mp_abort()
      !! globally stops all tasks
      !! this is intended to be low level, most of CP2K should call dbcsr_abort()

      INTEGER                                            :: ierr

      ierr = 0

#if !defined(__NO_ABORT)
#if defined(__parallel)
      CALL mpi_abort(MPI_COMM_WORLD, 1, ierr)
#else
      CALL m_abort()
#endif
#endif
      ! this routine never returns and levels with non-zero exit code
      STOP 1
   END SUBROUTINE mp_abort

   SUBROUTINE mp_stop(ierr, prg_code)
      !! stops *after an mpi error* translating the error code
      !! @note
      !! this function is private to mpiwrap.F

      INTEGER, INTENT(IN)                       :: ierr
         !! an error code * returned by an mpi call *
      CHARACTER(LEN=*)                          :: prg_code

#if defined(__parallel)
      INTEGER                                   :: istat, len
      CHARACTER(LEN=MPI_MAX_ERROR_STRING)       :: error_string
      CHARACTER(LEN=MPI_MAX_ERROR_STRING + 512) :: full_error
#else
      CHARACTER(LEN=512)                        :: full_error
#endif

#if defined(__parallel)
      CALL mpi_error_string(ierr, error_string, len, istat)
      WRITE (full_error, '(A,I0,A)') ' MPI error ', ierr, ' in '//TRIM(prg_code)//' : '//error_string(1:len)
#else
      WRITE (full_error, '(A,I0,A)') ' MPI error (!?) ', ierr, ' in '//TRIM(prg_code)
#endif

      DBCSR_ABORT(full_error)

   END SUBROUTINE mp_stop

   SUBROUTINE mp_sync(group)
      !! synchronizes with a barrier a given group of mpi tasks

      INTEGER, INTENT(IN)                                :: group
         !! mpi communicator

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sync'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      CALL mpi_barrier(group, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_barrier @ "//routineN)
#else
      MARK_USED(group)
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_sync

   SUBROUTINE mp_isync(group, request)
      !! synchronizes with a barrier a given group of mpi tasks

      INTEGER, INTENT(IN)                                :: group
         !! mpi communicator
      INTEGER, INTENT(OUT)                               :: request

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_isync'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      CALL mpi_ibarrier(group, request, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_ibarrier @ "//routineN)
#else
      MARK_USED(group)
      request = mp_request_null
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_isync

   RECURSIVE SUBROUTINE mp_environ_l(numtask, taskid, groupid)
      !! returns number of tasks and task id for a given mpi group
      !! simple and cartesian version.. recursive needed in case of failing mpi_comm_rank.
      !! @note
      !! ..mp_world_setup is gone, use mp_environ instead (i.e. give a groupid explicitly)

      INTEGER, OPTIONAL, INTENT(OUT)                     :: numtask, taskid
      INTEGER, INTENT(IN)                                :: groupid
         !! mpi communicator

      INTEGER                                            :: ierr

      ierr = 0

      IF (PRESENT(numtask)) numtask = 1
      IF (PRESENT(taskid)) taskid = 0
#if defined(__parallel)
      IF (PRESENT(taskid)) THEN
         CALL mpi_comm_rank(groupid, taskid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_rank @ mp_environ_l")
      END IF

      IF (PRESENT(numtask)) THEN
         CALL mpi_comm_size(groupid, numtask, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_size @ mp_environ_l")
      END IF
#else
      MARK_USED(groupid)
#endif

   END SUBROUTINE mp_environ_l

   SUBROUTINE mp_environ_c(numtask, dims, task_coor, groupid)

      INTEGER, INTENT(OUT)                     :: numtask, dims(2), &
                                                  task_coor(2)
      INTEGER, INTENT(IN)                      :: groupid

      INTEGER                                  :: ierr
#if defined(__parallel)
      LOGICAL, DIMENSION(2)                    :: periods
#endif

      ierr = 0
      numtask = 1
      task_coor = 0
      dims = 1
#if defined(__parallel)
      CALL mpi_comm_size(groupid, numtask, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_size @ mp_environ_c")

      CALL mpi_cart_get(groupid, 2, dims, periods, task_coor, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_cart_get @ mp_environ_c")
#else
      MARK_USED(groupid)
#endif

   END SUBROUTINE mp_environ_c

   SUBROUTINE mp_environ_c2(comm, ndims, dims, task_coor, periods)

      INTEGER, INTENT(IN)                                :: comm, ndims
      INTEGER, INTENT(OUT)                               :: dims(ndims), task_coor(ndims)
      LOGICAL, INTENT(out)                               :: periods(ndims)

      INTEGER                                            :: ierr

      ierr = 0

      task_coor = 0
      dims = 1
      periods = .FALSE.
#if defined(__parallel)
      CALL mpi_cart_get(comm, ndims, dims, periods, task_coor, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_cart_get @ mp_environ_c")
#else
      MARK_USED(comm)
#endif

   END SUBROUTINE mp_environ_c2

!..mp_cart_create
   SUBROUTINE mp_cart_create(comm_old, ndims, dims, pos, comm_cart)

      INTEGER, INTENT(IN)                      :: comm_old, ndims
      INTEGER, CONTIGUOUS, INTENT(INOUT)       :: dims(:)
      INTEGER, CONTIGUOUS, INTENT(OUT)         :: pos(:)
      INTEGER, INTENT(OUT)                     :: comm_cart

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_cart_create'

      INTEGER                                  :: handle, ierr, nodes
#if defined(__parallel)
      LOGICAL, DIMENSION(1:ndims)              :: period
      LOGICAL                                  :: reorder
#endif

      ierr = 0
      CALL timeset(routineN, handle)

      nodes = 0
      pos(1:ndims) = -1
      comm_cart = comm_old
#if defined(__parallel)

      CALL mpi_comm_size(comm_old, nodes, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_size @ "//routineN)

      IF (ANY(dims == 0)) CALL mpi_dims_create(nodes, ndims, dims, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_dims_create @ "//routineN)

      ! FIX ME.  Quick hack to avoid problems with realspace grids for compilers
      ! like IBM that actually reorder the processors when creating the new
      ! communicator
      reorder = .FALSE.
      period = .TRUE.
      CALL mpi_cart_create(comm_old, ndims, dims, period, reorder, comm_cart, &
                           ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_cart_create @ "//routineN)

      IF (comm_cart /= MPI_COMM_NULL) THEN
         debug_comm_count = debug_comm_count + 1
         CALL mpi_cart_get(comm_cart, ndims, dims, period, pos, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_cart_get @ "//routineN)
      END IF
#else
      pos(1:ndims) = 0
      dims = 1
      comm_cart = 0
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_cart_create

!..mp_cart_coords
   SUBROUTINE mp_cart_coords(comm, rank, coords)

      INTEGER, INTENT(IN)                                :: comm, rank
      INTEGER, DIMENSION(:), CONTIGUOUS, INTENT(OUT)     :: coords

      INTEGER                                            :: ierr, m

      ierr = 0

      m = SIZE(coords)
#if defined(__parallel)
      CALL mpi_cart_coords(comm, rank, m, coords, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_cart_coords @ mp_cart_coords")
#else
      coords = 0
      MARK_USED(rank)
      MARK_USED(comm)
#endif

   END SUBROUTINE mp_cart_coords

!..mp_comm_compare
   SUBROUTINE mp_comm_compare(comm1, comm2, res)

      INTEGER, INTENT(IN)                                :: comm1, comm2
      INTEGER, INTENT(OUT)                               :: res

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_comm_compare'

      INTEGER                                            :: handle, ierr, iout

      ierr = 0
      CALL timeset(routineN, handle)

      iout = 0
      res = 0
#if defined(__parallel)
      CALL mpi_comm_compare(comm1, comm2, iout, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_compare @ "//routineN)
      SELECT CASE (iout)
      CASE (MPI_IDENT)
         res = 0
      CASE (MPI_CONGRUENT)
         res = 1
      CASE (MPI_SIMILAR)
         res = 2
      CASE (MPI_UNEQUAL)
         res = 3
      CASE default
         res = 4
      END SELECT
#else
      MARK_USED(comm1)
      MARK_USED(comm2)
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_comm_compare

!..mp_cart_sub
   SUBROUTINE mp_cart_sub(comm, rdim, sub_comm)

      INTEGER, INTENT(IN)                                :: comm
      LOGICAL, DIMENSION(:), CONTIGUOUS, INTENT(IN)      :: rdim
      INTEGER, INTENT(OUT)                               :: sub_comm

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_cart_sub'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

      sub_comm = 0
#if defined(__parallel)
      CALL mpi_cart_sub(comm, rdim, sub_comm, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_cart_sub @ "//routineN)
      debug_comm_count = debug_comm_count + 1
#else
      MARK_USED(comm)
      MARK_USED(rdim)
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_cart_sub

!..mp_comm_free
   SUBROUTINE mp_comm_free(comm)

      INTEGER, INTENT(INOUT)                             :: comm

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_comm_free'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      CALL mpi_comm_free(comm, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_free @ "//routineN)
      debug_comm_count = debug_comm_count - 1
#else
      MARK_USED(comm)
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_comm_free

!..mp_comm_dup
   SUBROUTINE mp_comm_dup(comm1, comm2)

      INTEGER, INTENT(IN)                                :: comm1
      INTEGER, INTENT(OUT)                               :: comm2

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_comm_dup'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      CALL mpi_comm_dup(comm1, comm2, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_dup @ "//routineN)
      debug_comm_count = debug_comm_count + 1
#else
      comm2 = comm1
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_comm_dup

!..mp_rank_compare
   SUBROUTINE mp_rank_compare(comm1, comm2, rank)

      INTEGER, INTENT(IN)                      :: comm1, comm2
      INTEGER, DIMENSION(:), CONTIGUOUS, INTENT(OUT) :: rank

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_rank_compare'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: g1, g2, i, n, n1, n2
      INTEGER, ALLOCATABLE, DIMENSION(:)       :: rin
#endif

      ierr = 0
      CALL timeset(routineN, handle)

      rank = 0
#if defined(__parallel)
      CALL mpi_comm_size(comm1, n1, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_size @ "//routineN)
      CALL mpi_comm_size(comm2, n2, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_size @ "//routineN)
      n = MAX(n1, n2)
      CALL mpi_comm_group(comm1, g1, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_group @ "//routineN)
      CALL mpi_comm_group(comm2, g2, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_group @ "//routineN)
      ALLOCATE (rin(0:n - 1), STAT=ierr)
      IF (ierr /= 0) &
         DBCSR_ABORT("allocate @ "//routineN)
      DO i = 0, n - 1
         rin(i) = i
      END DO
      CALL mpi_group_translate_ranks(g1, n, rin, g2, rank, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, &
                                  "mpi_group_translate_rank @ "//routineN)
      CALL mpi_group_free(g1, ierr)
      IF (ierr /= 0) &
         DBCSR_ABORT("group_free @ "//routineN)
      CALL mpi_group_free(g2, ierr)
      IF (ierr /= 0) &
         DBCSR_ABORT("group_free @ "//routineN)
      DEALLOCATE (rin)
#else
      MARK_USED(comm1)
      MARK_USED(comm2)
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_rank_compare

!..mp_dims_create
   SUBROUTINE mp_dims_create(nodes, dims)

      INTEGER, INTENT(IN)                                :: nodes
      INTEGER, DIMENSION(:), CONTIGUOUS, INTENT(INOUT)   :: dims

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_dims_create'

      INTEGER                                            :: handle, ierr, ndim

      ierr = 0
      CALL timeset(routineN, handle)

      ndim = SIZE(dims)
#if defined(__parallel)
      IF (ANY(dims == 0)) CALL mpi_dims_create(nodes, ndim, dims, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_dims_create @ "//routineN)
#else
      dims = 1
      MARK_USED(nodes)
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_dims_create

!..mp_cart_rank
   SUBROUTINE mp_cart_rank(group, pos, rank)
      INTEGER, INTENT(IN)                                :: group
      INTEGER, DIMENSION(:), CONTIGUOUS, INTENT(IN)      :: pos
      INTEGER, INTENT(OUT)                               :: rank

      INTEGER                                            :: ierr

      ierr = 0

#if defined(__parallel)
      CALL mpi_cart_rank(group, pos, rank, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_cart_rank @ mp_cart_rank")
#else
      rank = 0
      MARK_USED(group)
      MARK_USED(pos)
#endif

   END SUBROUTINE mp_cart_rank

   SUBROUTINE mp_wait(request)
      !! waits for completion of the given request
      !! @note
      !! see isendrecv

      INTEGER, INTENT(inout)                             :: request

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_wait'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      CALL mpi_wait(request, MPI_STATUS_IGNORE, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_wait @ "//routineN)
#else
      MARK_USED(request)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_wait

   SUBROUTINE mp_waitall_1(requests)
      !! waits for completion of the given requests
      !! @note
      !! see isendrecv

      INTEGER, DIMENSION(:), CONTIGUOUS, INTENT(inout) :: requests

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_waitall_1'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: count
      INTEGER, ALLOCATABLE, DIMENSION(:, :)    :: status
#endif

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      count = SIZE(requests)
      ALLOCATE (status(MPI_STATUS_SIZE, count))
      CALL mpi_waitall_internal(count, requests, status, ierr) ! MPI_STATUSES_IGNORE openmpi workaround
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_waitall @ "//routineN)
      DEALLOCATE (status)
#else
      MARK_USED(requests)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_waitall_1

   SUBROUTINE mp_waitall_2(requests)
      !! waits for completion of the given requests
      INTEGER, DIMENSION(:, :), CONTIGUOUS, INTENT(inout)  :: requests

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_waitall_2'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: count
      INTEGER, ALLOCATABLE, DIMENSION(:, :)    :: status
#endif

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      count = SIZE(requests)
      ALLOCATE (status(MPI_STATUS_SIZE, count))

      CALL mpi_waitall_internal(count, requests, status, ierr) ! MPI_STATUSES_IGNORE openmpi workaround
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_waitall @ "//routineN)
      DEALLOCATE (status)
#else
      MARK_USED(requests)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_waitall_2

#if defined(__parallel)
   SUBROUTINE mpi_waitall_internal(count, array_of_requests, array_of_statuses, ierr)
      !! wrapper needed to deal with interfaces as present in openmpi 1.8.1
      !! the issue is with the rank or requests

      INTEGER, INTENT(in)                                :: count
      INTEGER, DIMENSION(count), INTENT(inout)           :: array_of_requests
      INTEGER, DIMENSION(MPI_STATUS_SIZE, *), &
         INTENT(out)                                     :: array_of_statuses
      INTEGER, INTENT(out)                               :: ierr

      CALL mpi_waitall(count, array_of_requests, array_of_statuses, ierr)

   END SUBROUTINE mpi_waitall_internal
#endif

   SUBROUTINE mp_waitany(requests, completed)
      !! waits for completion of any of the given requests
      INTEGER, DIMENSION(:), CONTIGUOUS, INTENT(inout) :: requests
      INTEGER, INTENT(out)                     :: completed

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_waitany'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: count
#endif

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      count = SIZE(requests)

      CALL mpi_waitany(count, requests, completed, MPI_STATUS_IGNORE, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_waitany @ "//routineN)
#else
      MARK_USED(requests)
      completed = 1
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_waitany

   FUNCTION mp_testall_tv(requests) RESULT(flag)
      !! Tests for completion of the given requests.
      !! We use mpi_test so that we can use a single status.

      INTEGER, DIMENSION(:)                 :: requests
         !! the list of requests to test
      LOGICAL                               :: flag
         !! logical which determines if requests are complete

      INTEGER                               :: ierr

#if defined(__parallel)
      INTEGER                               :: i
      LOGICAL, DIMENSION(:), POINTER        :: flags
#endif

      ierr = 0
      flag = .TRUE.

#if defined(__parallel)
      ALLOCATE (flags(SIZE(requests)))
      DO i = 1, SIZE(requests)
         CALL mpi_test(requests(i), flags(i), MPI_STATUS_IGNORE, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_test @ mp_testall_tv")
         flag = flag .AND. flags(i)
      END DO
      DEALLOCATE (flags)
#else
      requests = mp_request_null
#endif
   END FUNCTION mp_testall_tv

   SUBROUTINE mp_test_1(request, flag)
      !! Tests for completion of the given request.

      INTEGER, INTENT(inout)                             :: request
         !! the request
      LOGICAL, INTENT(out)                               :: flag
         !! logical which determines if the request is completed

      INTEGER                                            :: ierr

      ierr = 0

#if defined(__parallel)
      CALL mpi_test(request, flag, MPI_STATUS_IGNORE, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_test @ mp_test_1")
#else
      MARK_USED(request)
      flag = .TRUE.
#endif
   END SUBROUTINE mp_test_1

   SUBROUTINE mp_testany_1(requests, completed, flag)
      !! tests for completion of the given requests
      INTEGER, DIMENSION(:), CONTIGUOUS, INTENT(inout) :: requests
      INTEGER, INTENT(out), OPTIONAL           :: completed
      LOGICAL, INTENT(out), OPTIONAL           :: flag

      INTEGER                                  :: ierr
#if defined(__parallel)
      INTEGER                                  :: completed_l, count
      LOGICAL                                  :: flag_l
#endif

      ierr = 0

#if defined(__parallel)
      count = SIZE(requests)

      CALL mpi_testany_internal(count, requests, completed_l, flag_l, MPI_STATUS_IGNORE, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_testany_1 @ mp_testany")

      IF (PRESENT(completed)) completed = completed_l
      IF (PRESENT(flag)) flag = flag_l
#else
      MARK_USED(requests)
      IF (PRESENT(completed)) completed = 1
      IF (PRESENT(flag)) flag = .TRUE.
#endif
   END SUBROUTINE mp_testany_1

   SUBROUTINE mp_testany_2(requests, completed, flag)
      !! tests for completion of the given requests
      INTEGER, DIMENSION(:, :), CONTIGUOUS, INTENT(inout)   :: requests
      INTEGER, INTENT(out), OPTIONAL           :: completed
      LOGICAL, INTENT(out), OPTIONAL           :: flag

      INTEGER                                  :: ierr
#if defined(__parallel)
      INTEGER                                  :: completed_l, count
      LOGICAL                                  :: flag_l
#endif

      ierr = 0

#if defined(__parallel)
      count = SIZE(requests)

      CALL mpi_testany_internal(count, requests, completed_l, flag_l, MPI_STATUS_IGNORE, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_testany_2 @ mp_testany")

      IF (PRESENT(completed)) completed = completed_l
      IF (PRESENT(flag)) flag = flag_l
#else
      MARK_USED(requests)
      IF (PRESENT(completed)) completed = 1
      IF (PRESENT(flag)) flag = .TRUE.
#endif
   END SUBROUTINE mp_testany_2

#if defined(__parallel)
   SUBROUTINE mpi_testany_internal(count, array_of_requests, index, flag, status, ierr)
      !! wrapper needed to deal with interfaces as present in openmpi 1.8.1
      !! the issue is with the rank or requests

      INTEGER, INTENT(in)                                :: count
      INTEGER, DIMENSION(count), INTENT(inout)           :: array_of_requests
      INTEGER, INTENT(out)                               :: index
      LOGICAL, INTENT(out)                               :: flag
      INTEGER, DIMENSION(MPI_STATUS_SIZE), INTENT(out)   :: status
      INTEGER, INTENT(out)                               :: ierr

      CALL mpi_testany(count, array_of_requests, index, flag, status, ierr)

   END SUBROUTINE mpi_testany_internal
#endif

   SUBROUTINE mp_comm_split_direct(comm, sub_comm, color, key)
      !! the direct way to split a communicator each color is a sub_comm,
      !! the rank order is accoring to the order in the orig comm

      INTEGER, INTENT(in)                                :: comm
      INTEGER, INTENT(OUT)                               :: sub_comm
      INTEGER, INTENT(in)                                :: color
      INTEGER, INTENT(in), OPTIONAL                      :: key

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_comm_split_direct'

      INTEGER                                            :: handle, ierr, my_key

      ierr = 0
      CALL timeset(routineN, handle)

      my_key = 0
#if defined(__parallel)
      IF (PRESENT(key)) my_key = key
      CALL mpi_comm_split(comm, color, my_key, sub_comm, ierr)
      debug_comm_count = debug_comm_count + 1
      IF (ierr /= mpi_success) CALL mp_stop(ierr, routineN)
#else
      CALL mp_comm_dup(comm, sub_comm)
      MARK_USED(color)
      MARK_USED(key)
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_comm_split_direct
   SUBROUTINE mp_comm_split(comm, sub_comm, ngroups, group_distribution, &
                            subgroup_min_size, n_subgroups, group_partition, stride)
      !! splits the given communicator in group in subgroups trying to organize
      !! them in a way that the communication within each subgroup is
      !! efficient (but not necessarily the communication between subgroups)
      !! @note
      !! at least one of subgroup_min_size and n_subgroups is needed,
      !! the other default to the value needed to use most processors.
      !! if less cpus are present than needed for subgroup min size, n_subgroups,
      !! just one comm is created that contains all cpus

      INTEGER, INTENT(in)                      :: comm
         !! the mpi communicator that you want to split
      INTEGER, INTENT(out)                     :: sub_comm, ngroups
         !! the communicator for the subgroup (created, needs to be freed later)
         !! actual number of groups
      INTEGER, DIMENSION(0:)                   :: group_distribution
         !! input  : allocated with array with the nprocs entries (0 .. nprocs-1)
      INTEGER, INTENT(in), OPTIONAL            :: subgroup_min_size, n_subgroups
         !! the minimum size of the subgroup
         !! the number of subgroups wanted
      INTEGER, DIMENSION(0:), OPTIONAL         :: group_partition
         !! n_subgroups sized array containing the number of cpus wanted per group. should match the total number of cpus (only used
         !! if present and associated) (0..ngroups-1)
      INTEGER, OPTIONAL                        :: stride
         !! create groups using a stride (default=1) through the ranks of the comm to be split.

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_comm_split', routineP = moduleN//':'//routineN

      INTEGER                                  :: handle, ierr, mepos, nnodes
#if defined(__parallel)
      INTEGER                                  :: color, i, j, k, &
                                                  my_subgroup_min_size, &
                                                  istride, local_stride, irank
      INTEGER, DIMENSION(:), ALLOCATABLE       :: rank_permutation
#endif

      ierr = 0
      CALL timeset(routineN, handle)

      ! actual number of groups

      IF (.NOT. PRESENT(subgroup_min_size) .AND. .NOT. PRESENT(n_subgroups)) THEN
         DBCSR_ABORT(routineP//" missing arguments")
      END IF
      IF (PRESENT(subgroup_min_size) .AND. PRESENT(n_subgroups)) THEN
         DBCSR_ABORT(routineP//" too many arguments")
      END IF

      CALL mp_environ(nnodes, mepos, comm)

      IF (UBOUND(group_distribution, 1) .NE. nnodes - 1) THEN
         DBCSR_ABORT(routineP//" group_distribution wrong bounds")
      END IF

#if defined(__parallel)
      IF (PRESENT(subgroup_min_size)) THEN
         IF (subgroup_min_size < 0 .OR. subgroup_min_size > nnodes) THEN
            DBCSR_ABORT(routineP//" subgroup_min_size too small or too large")
         END IF
         ngroups = nnodes/subgroup_min_size
         my_subgroup_min_size = subgroup_min_size
      ELSE ! n_subgroups
         IF (n_subgroups <= 0) THEN
            DBCSR_ABORT(routineP//" n_subgroups too small")
         END IF
         IF (nnodes/n_subgroups > 0) THEN ! we have a least one cpu per group
            ngroups = n_subgroups
         ELSE ! well, only one group then
            ngroups = 1
         END IF
         my_subgroup_min_size = nnodes/ngroups
      END IF

      ! rank_permutation: is a permutation of ranks, so that groups are not necessarily continuous in rank of the master group
      ! while the order is not critical (we only color ranks), it can e.g. be used to make groups that have just 1 rank per node
      ! (by setting stride equal to the number of mpi ranks per node), or by sharing  a node between two groups (stride 2).
      ALLOCATE (rank_permutation(0:nnodes - 1))
      local_stride = 1
      IF (PRESENT(stride)) local_stride = stride
      k = 0
      DO istride = 1, local_stride
         DO irank = istride - 1, nnodes - 1, local_stride
            rank_permutation(k) = irank
            k = k + 1
         END DO
      END DO

      DO i = 0, nnodes - 1
         group_distribution(rank_permutation(i)) = MIN(i/my_subgroup_min_size, ngroups - 1)
      END DO
      ! even the user gave a partition, see if we can use it to overwrite this choice
      IF (PRESENT(group_partition)) THEN
         IF (ALL(group_partition > 0) .AND. (SUM(group_partition) .EQ. nnodes) .AND. (ngroups == SIZE(group_partition))) THEN
            k = 0
            DO i = 0, SIZE(group_partition) - 1
               DO j = 1, group_partition(i)
                  group_distribution(rank_permutation(k)) = i
                  k = k + 1
               END DO
            END DO
         ELSE
            ! just ignore silently as we have reasonable defaults. Probably a warning would not be to bad
         END IF
      END IF
      color = group_distribution(mepos)
      CALL mpi_comm_split(comm, color, 0, sub_comm, ierr)
      debug_comm_count = debug_comm_count + 1
      IF (ierr /= mpi_success) CALL mp_stop(ierr, "in "//routineP//" split")
#else
      CALL mp_comm_dup(comm, sub_comm)
      group_distribution(0) = 0
      ngroups = 1
      MARK_USED(stride)
      MARK_USED(group_partition)
#endif
      CALL timestop(handle)

   END SUBROUTINE mp_comm_split

   SUBROUTINE mp_probe(source, comm, tag)
      !! probes for an incoming message with any tag

      INTEGER                                  :: source
         !! the source of the possible incoming message, if MP_ANY_SOURCE it is a blocking one and return value is the source of the
         !! next incoming message if source is a different value it is a non-blocking probe retuning MP_ANY_SOURCE if there is no
         !! incoming message
      INTEGER, INTENT(IN)                      :: comm
         !! the communicator
      INTEGER, INTENT(OUT)                     :: tag
         !! the tag of the incoming message

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_probe'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER, DIMENSION(mp_status_size)       :: status_single
      LOGICAL                                  :: flag
#endif

!   ---------------------------------------------------------------------------

      CALL timeset(routineN, handle)

      ierr = 0
#if defined(__parallel)
      IF (source .EQ. mp_any_source) THEN
         CALL mpi_probe(mp_any_source, mp_any_tag, comm, status_single, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_probe @ "//routineN)
         source = status_single(MPI_SOURCE)
         tag = status_single(MPI_TAG)
      ELSE
         flag = .FALSE.
         CALL mpi_iprobe(source, mp_any_tag, comm, flag, status_single, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_iprobe @ "//routineN)
         IF (flag .EQV. .FALSE.) THEN
            source = mp_any_source
            tag = -1 !status_single(MPI_TAG) ! in case of flag==false status is undefined
         ELSE
            tag = status_single(MPI_TAG)
         END IF
      END IF
#else
      tag = -1
      MARK_USED(comm)
      MARK_USED(source)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_probe

! **************************************************************************************************
! Here come the data routines with none of the standard data types.
! **************************************************************************************************

   SUBROUTINE mp_bcast_b(msg, source, gid)
      LOGICAL                                            :: msg
      INTEGER                                            :: source, gid

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_bcast_b'

      INTEGER                                            :: handle, ierr, msglen

      ierr = 0
      CALL timeset(routineN, handle)

      msglen = 1
#if defined(__parallel)
      CALL mpi_bcast(msg, msglen, MPI_LOGICAL, source, gid, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_bcast @ "//routineN)
      CALL add_perf(perf_id=2, msg_size=msglen*loglen)
#else
      MARK_USED(msg)
      MARK_USED(source)
      MARK_USED(gid)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_bcast_b

   SUBROUTINE mp_bcast_bv(msg, source, gid)
      LOGICAL, CONTIGUOUS                                :: msg(:)
      INTEGER                                            :: source, gid

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_bcast_bv'

      INTEGER                                            :: handle, ierr, msglen

      ierr = 0
      CALL timeset(routineN, handle)

      msglen = SIZE(msg)
#if defined(__parallel)
      CALL mpi_bcast(msg, msglen, MPI_LOGICAL, source, gid, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_bcast @ "//routineN)
      CALL add_perf(perf_id=2, msg_size=msglen*loglen)
#else
      MARK_USED(source)
      MARK_USED(gid)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_bcast_bv

   SUBROUTINE mp_isend_bv(msgin, dest, comm, request, tag)
      !! Non-blocking send of logical vector data
      !! @note see mp_irecv_iv
      !! @endnote
      !! @note
      !! arrays can be pointers or assumed shape, but they must be contiguous!

      LOGICAL, DIMENSION(:), CONTIGUOUS        :: msgin
         !! the input message
      INTEGER, INTENT(IN)                      :: dest, comm
         !! the destination processor
         !! the communicator object
      INTEGER, INTENT(out)                     :: request
         !! communication request index
      INTEGER, INTENT(in), OPTIONAL            :: tag
         !! message tag

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_isend_bv'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: msglen, my_tag
      LOGICAL                                  :: foo(1)
#endif

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      my_tag = 0
      IF (PRESENT(tag)) my_tag = tag

      msglen = SIZE(msgin, 1)
      IF (msglen > 0) THEN
         CALL mpi_isend(msgin(1), msglen, MPI_LOGICAL, dest, my_tag, &
                        comm, request, ierr)
      ELSE
         CALL mpi_isend(foo, msglen, MPI_LOGICAL, dest, my_tag, &
                        comm, request, ierr)
      END IF
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_isend @ "//routineN)

      CALL add_perf(perf_id=11, msg_size=msglen*loglen)
#else
      DBCSR_ABORT("mp_isend called in non parallel case")
      MARK_USED(msgin)
      MARK_USED(dest)
      MARK_USED(comm)
      MARK_USED(request)
      MARK_USED(tag)
      request = 0
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_isend_bv

   SUBROUTINE mp_irecv_bv(msgout, source, comm, request, tag)
      !! Non-blocking receive of logical vector data
      !! @note see mp_irecv_iv
      !! @endnote
      !! @note
      !! arrays can be pointers or assumed shape, but they must be contiguous!

      LOGICAL, DIMENSION(:), CONTIGUOUS        :: msgout
         !! the received message
      INTEGER, INTENT(IN)                      :: source, comm
         !! the source processor
         !! the communicator object
      INTEGER, INTENT(out)                     :: request
         !! communication request index
      INTEGER, INTENT(in), OPTIONAL            :: tag
         !! message tag

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_irecv_bv'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: msglen, my_tag
      LOGICAL                                  :: foo(1)
#endif

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      my_tag = 0
      IF (PRESENT(tag)) my_tag = tag

      msglen = SIZE(msgout, 1)
      IF (msglen > 0) THEN
         CALL mpi_irecv(msgout(1), msglen, MPI_LOGICAL, source, my_tag, &
                        comm, request, ierr)
      ELSE
         CALL mpi_irecv(foo, msglen, MPI_LOGICAL, source, my_tag, &
                        comm, request, ierr)
      END IF
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_ircv @ "//routineN)

      CALL add_perf(perf_id=12, msg_size=msglen*loglen)
#else
      DBCSR_ABORT("mp_irecv called in non parallel case")
      MARK_USED(msgout)
      MARK_USED(source)
      MARK_USED(comm)
      MARK_USED(request)
      MARK_USED(tag)
      request = 0
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_irecv_bv

   SUBROUTINE mp_bcast_av(msg, source, gid)
      CHARACTER(LEN=*)                         :: msg
      INTEGER                                  :: source, gid

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_bcast_av'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: i, msglen, numtask, taskid
      INTEGER, DIMENSION(:), ALLOCATABLE       :: imsg
#endif

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)

      CALL mp_environ(numtask, taskid, gid)
      IF (taskid == source) msglen = LEN_TRIM(msg)

      CALL mp_bcast(msglen, source, gid)
      ! this is a workaround to avoid problems on the T3E
      ! at the moment we have a data alignment error when trying to
      ! broadcast characters on the T3E (not always!)
      ! JH 19/3/99 on galileo
      ! CALL mpi_bcast(msg,msglen,MPI_CHARACTER,source,gid,ierr)
      ALLOCATE (imsg(1:msglen))
      DO i = 1, msglen
         imsg(i) = ICHAR(msg(i:i))
      END DO
      CALL mpi_bcast(imsg, msglen, MPI_INTEGER, source, gid, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_bcast @ "//routineN)
      msg = ""
      DO i = 1, msglen
         msg(i:i) = CHAR(imsg(i))
      END DO
      DEALLOCATE (imsg)
      CALL add_perf(perf_id=2, msg_size=msglen*charlen)
#else
      MARK_USED(msg)
      MARK_USED(source)
      MARK_USED(gid)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_bcast_av

   SUBROUTINE mp_bcast_am(msg, source, gid)
      CHARACTER(LEN=*)                         :: msg(:)
      INTEGER                                  :: source, gid

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_bcast_am'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: i, j, k, msglen, msgsiz, &
                                                  numtask, taskid
      INTEGER, ALLOCATABLE                     :: imsg(:), imsglen(:)
#endif

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      CALL mp_environ(numtask, taskid, gid)
      msgsiz = SIZE(msg)
      ! Determine size of the minimum array of integers to broadcast the string
      ALLOCATE (imsglen(1:msgsiz))
      DO j = 1, msgsiz
         IF (taskid == source) imsglen(j) = LEN_TRIM(msg(j))
      END DO
      CALL mp_bcast(imsglen, source, gid)
      msglen = SUM(imsglen)
      ! this is a workaround to avoid problems on the T3E
      ! at the moment we have a data alignment error when trying to
      ! broadcast characters on the T3E (not always!)
      ! JH 19/3/99 on galileo
      ! CALL mpi_bcast(msg,msglen,MPI_CHARACTER,source,gid,ierr)
      ALLOCATE (imsg(1:msglen))
      k = 0
      DO j = 1, msgsiz
         DO i = 1, imsglen(j)
            k = k + 1
            imsg(k) = ICHAR(msg(j) (i:i))
         END DO
      END DO
      CALL mpi_bcast(imsg, msglen, MPI_INTEGER, source, gid, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_bcast @ "//routineN)
      msg = ""
      k = 0
      DO j = 1, msgsiz
         DO i = 1, imsglen(j)
            k = k + 1
            msg(j) (i:i) = CHAR(imsg(k))
         END DO
      END DO
      DEALLOCATE (imsg)
      DEALLOCATE (imsglen)
      CALL add_perf(perf_id=2, msg_size=msglen*charlen*msgsiz)
#else
      MARK_USED(msg)
      MARK_USED(source)
      MARK_USED(gid)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_bcast_am

   SUBROUTINE mp_minloc_dv(msg, gid)
      !! Finds the location of the minimal element in a vector.
      !!
      !! MPI mapping
      !! mpi_allreduce with the MPI_MINLOC reduction function identifier
      !!
      !! Invalid data types
      !! This routine is invalid for (int_8) data!

      REAL(kind=real_8), CONTIGUOUS, INTENT(INOUT)         :: msg(:)
         !! Find location of maximum element among these data (input).
      INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_minloc_dv'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: msglen
      REAL(kind=real_8), ALLOCATABLE           :: res(:)
#endif

      ierr = 0
      IF ("d" .EQ. "l" .AND. real_8 .EQ. int_8) THEN
         DBCSR_ABORT("Minimal location not available with long integers @ "//routineN)
      END IF
      CALL timeset(routineN, handle)

#if defined(__parallel)
      msglen = SIZE(msg)
      ALLOCATE (res(1:msglen), STAT=ierr)
      IF (ierr /= 0) &
         DBCSR_ABORT("allocate @ "//routineN)
      CALL mpi_allreduce(msg, res, msglen/2, MPI_2DOUBLE_PRECISION, MPI_MINLOC, gid, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
      msg = res
      DEALLOCATE (res)
      CALL add_perf(perf_id=3, msg_size=msglen*real_8_size)
#else
      MARK_USED(msg)
      MARK_USED(gid)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_minloc_dv

   SUBROUTINE mp_maxloc_dv(msg, gid)
      !! Finds the location of the maximal element in a vector.
      !!
      !! MPI mapping
      !! mpi_allreduce with the MPI_MAXLOC reduction function identifier
      !!
      !! Invalid data types
      !! This routine is invalid for (int_8) data!

      REAL(kind=real_8), CONTIGUOUS, INTENT(INOUT)         :: msg(:)
         !! Find location of maximum element among these data (input).
      INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_maxloc_dv'

      INTEGER                                  :: handle, ierr
#if defined(__parallel)
      INTEGER                                  :: msglen
      REAL(kind=real_8), ALLOCATABLE           :: res(:)
#endif

      ierr = 0
      IF ("d" .EQ. "l" .AND. real_8 .EQ. int_8) THEN
         DBCSR_ABORT("Maximal location not available with long integers @ "//routineN)
      END IF
      CALL timeset(routineN, handle)

#if defined(__parallel)
      msglen = SIZE(msg)
      ALLOCATE (res(1:msglen))
      CALL mpi_allreduce(msg, res, msglen/2, MPI_2DOUBLE_PRECISION, MPI_MAXLOC, gid, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
      msg = res
      DEALLOCATE (res)
      CALL add_perf(perf_id=3, msg_size=msglen*real_8_size)
#else
      MARK_USED(msg)
      MARK_USED(gid)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_maxloc_dv

   SUBROUTINE mp_sum_b(msg, gid)
      !! Logical OR reduction
      !!
      !! MPI mapping
      !! mpi_allreduce

      LOGICAL, INTENT(INOUT)                             :: msg
         !! Datum to perform inclusive disjunction (input) and resultant inclusive disjunction (output)
      INTEGER, INTENT(IN)                                :: gid
         !! Message passing environment identifier

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_b'

      INTEGER                                            :: handle, ierr, msglen

      CALL timeset(routineN, handle)
      ierr = 0
      msglen = 1
#if defined(__parallel)
      CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_LOGICAL, MPI_LOR, gid, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
#else
      MARK_USED(msg)
      MARK_USED(gid)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_sum_b

   SUBROUTINE mp_sum_bv(msg, gid)
      !! Logical OR reduction
      !!
      !! MPI mapping
      !! mpi_allreduce

      LOGICAL, DIMENSION(:), CONTIGUOUS, INTENT(INOUT)   :: msg
         !! Datum to perform inclusive disjunction (input) and resultant inclusive disjunction (output)
      INTEGER, INTENT(IN)                                :: gid
         !! Message passing environment identifier

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_bv'

      INTEGER                                            :: handle, ierr, msglen

      CALL timeset(routineN, handle)
      ierr = 0
      msglen = SIZE(msg)
#if defined(__parallel)
      IF (msglen .GT. 0) THEN
         CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_LOGICAL, MPI_LOR, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
      END IF
#else
      MARK_USED(msg)
      MARK_USED(gid)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_sum_bv

   SUBROUTINE mp_isum_bv(msg, gid, request)
      !! Logical OR reduction
      !!
      !! MPI mapping
      !! mpi_allreduce

      LOGICAL, DIMENSION(:), CONTIGUOUS, INTENT(INOUT)   :: msg
         !! Datum to perform inclusive disjunction (input) and resultant inclusive disjunction (output)
      INTEGER, INTENT(IN)                                :: gid
         !! Message passing environment identifier
      INTEGER, INTENT(INOUT)                             :: request

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_isum_bv'

      INTEGER                                            :: handle, ierr, msglen

      CALL timeset(routineN, handle)
      ierr = 0
      msglen = SIZE(msg)
#if defined(__parallel)
      IF (msglen .GT. 0) THEN
         CALL mpi_iallreduce(MPI_IN_PLACE, msg, msglen, MPI_LOGICAL, MPI_LOR, gid, request, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_iallreduce @ "//routineN)
      ELSE
         request = mp_request_null
      END IF
#else
      MARK_USED(msg)
      MARK_USED(gid)
      MARK_USED(request)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_isum_bv

   SUBROUTINE mp_get_library_version(version, resultlen)
      !! Get Version of the MPI Library (MPI 3)

      CHARACTER(LEN=*), INTENT(OUT)                      :: version
         !! Version of the library, declared as CHARACTER(LEN=mp_max_library_version_string)
      INTEGER, INTENT(OUT)                               :: resultlen
         !! Length (in printable characters) of the result returned in version (integer)

#if defined(__parallel)
      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_get_library_version'

      INTEGER                                            :: ierr

      ierr = 0
      CALL mpi_get_library_version(version, resultlen, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_get_library_version @ "//routineN)
#else
      version = ''
      resultlen = 0
#endif
   END SUBROUTINE mp_get_library_version

   SUBROUTINE mp_get_processor_name(procname, resultlen)
      !! Get a unique specifier for the actual (as opposed to virtual) node (MPI 2.1)

      CHARACTER(LEN=*), INTENT(OUT)                      :: procname
         !! Name of processor, declared as CHARACTER(LEN=mp_max_processor_name)
      INTEGER, OPTIONAL, INTENT(OUT)                     :: resultlen
         !! Length (in characters) of procname (INTEGER)

#if defined(__parallel)
      INTEGER                                            :: namelen, ierr

      CALL mpi_get_processor_name(procname, namelen, ierr)
      IF (ierr .EQ. 0) THEN
         IF (PRESENT(resultlen)) resultlen = namelen
      ELSE
#endif
         CALL m_hostnm(procname)
         IF (PRESENT(resultlen)) resultlen = LEN_TRIM(procname)
#if defined(__parallel)
      END IF
#endif
   END SUBROUTINE mp_get_processor_name

   SUBROUTINE mp_file_open(groupid, fh, filepath, amode_status, info)
      !! Opens a file
      !!
      !! MPI-I/O mapping  mpi_file_open
      !!
      !! STREAM-I/O mapping  OPEN

      INTEGER, INTENT(IN)                      :: groupid
         !! message passing environment identifier
      INTEGER, INTENT(OUT)                     :: fh
         !! file handle (file storage unit)
      CHARACTER(LEN=*), INTENT(IN)             :: filepath
         !! path to the file
      INTEGER, INTENT(IN)                      :: amode_status
         !! access mode
      INTEGER, INTENT(IN), OPTIONAL            :: info
         !! info object

      INTEGER                                  :: ierr, istat
#if defined(__parallel)
      INTEGER                                  :: my_info
#else
      CHARACTER(LEN=10)                        :: fstatus, fposition
      INTEGER                                  :: amode
      LOGICAL                                  :: exists, is_open
#endif

      ierr = 0
      istat = 0
#if defined(__parallel)
      my_info = mpi_info_null
      IF (PRESENT(info)) my_info = info
      CALL mpi_file_open(groupid, filepath, amode_status, my_info, fh, ierr)
      CALL mpi_file_set_errhandler(fh, MPI_ERRORS_RETURN, ierr)
      IF (ierr .NE. 0) CALL mp_stop(ierr, "mpi_file_set_errhandler @ mp_file_open")
#else
      MARK_USED(groupid)
      MARK_USED(info)
      amode = amode_status
      IF (amode .GT. file_amode_append) THEN
         fposition = "APPEND"
         amode = amode - file_amode_append
      ELSE
         fposition = "REWIND"
      END IF
      IF ((amode .EQ. file_amode_create) .OR. &
          (amode .EQ. file_amode_create + file_amode_wronly) .OR. &
          (amode .EQ. file_amode_create + file_amode_wronly + file_amode_excl)) THEN
         fstatus = "UNKNOWN"
      ELSE
         fstatus = "OLD"
      END IF
      ! Get a new unit number
      DO fh = 1, 999
         INQUIRE (UNIT=fh, EXIST=exists, OPENED=is_open, IOSTAT=istat)
         IF (exists .AND. (.NOT. is_open) .AND. (istat == 0)) EXIT
      END DO
      OPEN (UNIT=fh, FILE=filepath, STATUS=fstatus, ACCESS="STREAM", POSITION=fposition)
#endif
   END SUBROUTINE mp_file_open

   SUBROUTINE mp_file_delete(filepath, info)
      !! Deletes a file. Auxiliary routine to emulate 'replace' action for mp_file_open.
      !! Only the master processor should call this routine.

      CHARACTER(LEN=*), INTENT(IN)             :: filepath
         !! path to the file
      INTEGER, INTENT(IN), OPTIONAL            :: info
         !! info object

#if defined(__parallel)
      INTEGER                                  :: ierr
      INTEGER                                  :: my_info
      LOGICAL                                  :: exists
#endif

#if defined(__parallel)
      ierr = 0
      my_info = mpi_info_null
      IF (PRESENT(info)) my_info = info
      INQUIRE (FILE=filepath, EXIST=exists)
      IF (exists) CALL mpi_file_delete(filepath, my_info, ierr)
      IF (ierr .NE. 0) CALL mp_stop(ierr, "mpi_file_set_errhandler @ mp_file_delete")
#else
      MARK_USED(filepath)
      MARK_USED(info)
      ! Explicit file delete not necessary, handled by subsequent call to open_file with action 'replace'
#endif

   END SUBROUTINE mp_file_delete

   SUBROUTINE mp_file_close(fh)
      !! Closes a file
      !!
      !! MPI-I/O mapping   mpi_file_close
      !!
      !! STREAM-I/O mapping   CLOSE

      INTEGER, INTENT(INOUT)                             :: fh
         !! file handle (file storage unit)

      INTEGER                                            :: ierr

      ierr = 0
#if defined(__parallel)
      CALL mpi_file_set_errhandler(fh, MPI_ERRORS_RETURN, ierr)
      CALL mpi_file_close(fh, ierr)
      IF (ierr .NE. 0) CALL mp_stop(ierr, "mpi_file_set_errhandler @ mp_file_close")
#else
      CLOSE (fh)
#endif
   END SUBROUTINE mp_file_close

   SUBROUTINE mp_file_get_size(fh, file_size)
      !! Returns the file size
      !!
      !! MPI-I/O mapping   mpi_file_get_size
      !!
      !! STREAM-I/O mapping   INQUIRE

      INTEGER, INTENT(IN)                                :: fh
         !! file handle (file storage unit)
      INTEGER(kind=file_offset), INTENT(OUT)             :: file_size
         !! the file size

      INTEGER                                            :: ierr

      ierr = 0
#if defined(__parallel)
      CALL mpi_file_set_errhandler(fh, MPI_ERRORS_RETURN, ierr)
      CALL mpi_file_get_size(fh, file_size, ierr)
      IF (ierr .NE. 0) CALL mp_stop(ierr, "mpi_file_set_errhandler @ mp_file_get_size")
#else
      INQUIRE (UNIT=fh, SIZE=file_size)
#endif
   END SUBROUTINE mp_file_get_size

   SUBROUTINE mp_file_get_position(fh, pos)
      !! Returns the file position
      !!
      !! MPI-I/O mapping   mpi_file_get_position
      !!
      !! STREAM-I/O mapping   INQUIRE

      INTEGER, INTENT(IN)                                :: fh
         !! file handle (file storage unit)
      INTEGER(kind=file_offset), INTENT(OUT)             :: pos
         !! the file position

      INTEGER                                            :: ierr

      ierr = 0
#if defined(__parallel)
      CALL mpi_file_set_errhandler(fh, MPI_ERRORS_RETURN, ierr)
      CALL mpi_file_get_position(fh, pos, ierr)
      IF (ierr .NE. 0) CALL mp_stop(ierr, "mpi_file_set_errhandler @ mp_file_get_position")
#else
      INQUIRE (UNIT=fh, POS=pos)
#endif
   END SUBROUTINE mp_file_get_position

   SUBROUTINE mp_file_write_at_ch(fh, offset, msg)
      CHARACTER(LEN=*), INTENT(IN)               :: msg
      INTEGER, INTENT(IN)                        :: fh
      INTEGER(kind=file_offset), INTENT(IN)      :: offset

#if defined(__parallel)
      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_file_write_at_ch'

      INTEGER                                    :: ierr

      CALL MPI_FILE_WRITE_AT(fh, offset, msg, LEN(msg), MPI_CHARACTER, MPI_STATUS_IGNORE, ierr)
      IF (ierr .NE. 0) &
         DBCSR_ABORT("mpi_file_write_at_ch @ "//routineN)
#else
      WRITE (UNIT=fh, POS=offset + 1) msg
#endif
   END SUBROUTINE mp_file_write_at_ch

   SUBROUTINE mp_file_write_at_all_ch(fh, offset, msg)
      CHARACTER(LEN=*), INTENT(IN)               :: msg
      INTEGER, INTENT(IN)                        :: fh
      INTEGER(kind=file_offset), INTENT(IN)      :: offset

#if defined(__parallel)
      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_file_write_at_all_ch'

      INTEGER                                    :: ierr

      CALL MPI_FILE_WRITE_AT_ALL(fh, offset, msg, LEN(msg), MPI_CHARACTER, MPI_STATUS_IGNORE, ierr)
      IF (ierr .NE. 0) &
         DBCSR_ABORT("mpi_file_write_at_all_ch @ "//routineN)
#else
      WRITE (UNIT=fh, POS=offset + 1) msg
#endif
   END SUBROUTINE mp_file_write_at_all_ch

   SUBROUTINE mp_file_read_at_all_ch(fh, offset, msg)
      CHARACTER(LEN=*), INTENT(OUT)              :: msg
      INTEGER, INTENT(IN)                        :: fh
      INTEGER(kind=file_offset), INTENT(IN)      :: offset

#if defined(__parallel)
      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_file_read_at_all_ch'

      INTEGER                                    :: ierr

      CALL MPI_FILE_READ_AT_ALL(fh, offset, msg, LEN(msg), MPI_CHARACTER, MPI_STATUS_IGNORE, ierr)
      IF (ierr .NE. 0) &
         DBCSR_ABORT("mpi_file_read_at_all_ch @ "//routineN)
#else
      READ (UNIT=fh, POS=offset + 1) msg
#endif
   END SUBROUTINE mp_file_read_at_all_ch

   SUBROUTINE mp_type_size(type_descriptor, type_size)
      !! Returns the size of a data type in bytes
      !!
      !! MPI mapping
      !! mpi_type_size

      TYPE(mp_type_descriptor_type), INTENT(IN)          :: type_descriptor
         !! data type
      INTEGER, INTENT(OUT)                               :: type_size
         !! size of the data type

#if defined(__parallel)
      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_type_size'
      INTEGER                                            :: ierr

      ierr = 0
      CALL MPI_TYPE_SIZE(type_descriptor%type_handle, type_size, ierr)
      IF (ierr .NE. 0) &
         DBCSR_ABORT("mpi_type_size @ "//routineN)
#else
      SELECT CASE (type_descriptor%type_handle)
      CASE (1)
         type_size = real_4_size
      CASE (3)
         type_size = real_8_size
      CASE (5)
         type_size = 2*real_4_size
      CASE (7)
         type_size = 2*real_8_size
      END SELECT
#endif
   END SUBROUTINE mp_type_size

   FUNCTION mp_type_make_struct(subtypes, &
                                vector_descriptor, index_descriptor) &
      RESULT(type_descriptor)
      TYPE(mp_type_descriptor_type), &
         DIMENSION(:), INTENT(IN)               :: subtypes
      INTEGER, DIMENSION(2), INTENT(IN), &
         OPTIONAL                               :: vector_descriptor
      TYPE(mp_indexing_meta_type), &
         INTENT(IN), OPTIONAL                   :: index_descriptor
      TYPE(mp_type_descriptor_type)            :: type_descriptor

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_type_make_struct'

      INTEGER                                  :: i, ierr, n
#if defined(__parallel)
      INTEGER(kind=mpi_address_kind), &
         ALLOCATABLE, DIMENSION(:)              :: displacements
#endif
      INTEGER, ALLOCATABLE, DIMENSION(:)       :: lengths, old_types

      ierr = 0
      n = SIZE(subtypes)
      !type_descriptor%mpi_type_handle = MPI_DATATYPE_NULL
      type_descriptor%length = 1
#if defined(__parallel)
      CALL mpi_get_address(MPI_BOTTOM, type_descriptor%base, ierr)
      IF (ierr /= 0) &
         DBCSR_ABORT("MPI_get_address @ "//routineN)
      ALLOCATE (displacements(n))
#endif
      type_descriptor%vector_descriptor(1:2) = 1
      type_descriptor%has_indexing = .FALSE.
      ALLOCATE (type_descriptor%subtype(n))
      type_descriptor%subtype(:) = subtypes(:)
      ALLOCATE (lengths(n), old_types(n))
      DO i = 1, SIZE(subtypes)
#if defined(__parallel)
         displacements(i) = subtypes(i)%base
#endif
         old_types(i) = subtypes(i)%type_handle
         lengths(i) = subtypes(i)%length
      END DO
#if defined(__parallel)
      CALL MPI_Type_create_struct(n, &
                                  lengths, displacements, old_types, &
                                  type_descriptor%type_handle, ierr)
      IF (ierr /= 0) &
         DBCSR_ABORT("MPI_Type_create_struct @ "//routineN)
      CALL MPI_Type_commit(type_descriptor%type_handle, ierr)
      IF (ierr /= 0) &
         DBCSR_ABORT("MPI_Type_commit @ "//routineN)
#endif
      IF (PRESENT(vector_descriptor) .OR. PRESENT(index_descriptor)) THEN
         DBCSR_ABORT(routineN//" Vectors and indices NYI")
      END IF
   END FUNCTION mp_type_make_struct

   RECURSIVE SUBROUTINE mp_type_free_m(type_descriptor)
      TYPE(mp_type_descriptor_type), INTENT(inout)       :: type_descriptor

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_type_free_m'

      INTEGER                                            :: handle, i, ierr

      CALL timeset(routineN, handle)
      ierr = 0

      ! If the subtype is associated, then it's a user-defined data type.

      IF (ASSOCIATED(type_descriptor%subtype)) THEN
         DO i = 1, SIZE(type_descriptor%subtype)
            CALL mp_type_free_m(type_descriptor%subtype(i))
         END DO
         DEALLOCATE (type_descriptor%subtype)
      END IF
#if defined(__parallel)
      CALL MPI_Type_free(type_descriptor%type_handle, ierr)
      IF (ierr /= 0) &
         DBCSR_ABORT("MPI_Type_free @ "//routineN)
#endif

      CALL timestop(handle)

   END SUBROUTINE mp_type_free_m

   SUBROUTINE mp_isend_custom(msgin, dest, comm, request, tag)
      !! Non-blocking send of custom type
      TYPE(mp_type_descriptor_type), INTENT(IN)          :: msgin
      INTEGER, INTENT(IN)                                :: dest, comm
      INTEGER, INTENT(out)                               :: request
      INTEGER, INTENT(in), OPTIONAL                      :: tag

      INTEGER                                            :: ierr

#if defined(__parallel)
      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_isend_custom'
      INTEGER                                            :: my_tag
      my_tag = 0
      IF (PRESENT(tag)) my_tag = tag

      ierr = 0
      CALL mpi_isend(MPI_BOTTOM, 1, msgin%type_handle, dest, my_tag, &
                     comm, request, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_isend @ "//routineN)
#else
      MARK_USED(msgin)
      MARK_USED(dest)
      MARK_USED(comm)
      MARK_USED(request)
      MARK_USED(tag)
      ierr = 1
      request = 0
      CALL mp_stop(ierr, "mp_isend called in non parallel case")
#endif
   END SUBROUTINE mp_isend_custom

   SUBROUTINE mp_irecv_custom(msgout, source, comm, request, tag)
      !! Non-blocking receive of vector data
      TYPE(mp_type_descriptor_type), INTENT(INOUT)       :: msgout
      INTEGER, INTENT(IN)                                :: source, comm
      INTEGER, INTENT(out)                               :: request
      INTEGER, INTENT(in), OPTIONAL                      :: tag

      INTEGER                                            :: ierr

#if defined(__parallel)
      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_irecv_custom'
      INTEGER                                            :: my_tag
      ierr = 0
      my_tag = 0

      IF (PRESENT(tag)) my_tag = tag

      CALL mpi_irecv(MPI_BOTTOM, 1, msgout%type_handle, source, my_tag, &
                     comm, request, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_irecv @ "//routineN)
#else
      MARK_USED(msgout)
      MARK_USED(source)
      MARK_USED(comm)
      MARK_USED(request)
      MARK_USED(tag)
      ierr = 1
      request = 0
      DBCSR_ABORT("mp_irecv called in non parallel case")
#endif
   END SUBROUTINE mp_irecv_custom

   SUBROUTINE mp_win_free(win)
      !! Window free
      INTEGER, INTENT(INOUT)                             :: win

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_win_free'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)

      CALL mpi_win_free(win, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_win_free @ "//routineN)
#else
      MARK_USED(win)
      win = mp_win_null
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_win_free

   SUBROUTINE mp_win_flush_all(win)
      !! Window flush
      INTEGER, INTENT(IN)                                :: win

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_win_flush_all'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)
      CALL mpi_win_flush_all(win, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_win_flush_all @ "//routineN)
#else
      MARK_USED(win)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_win_flush_all

   SUBROUTINE mp_win_lock_all(win)
      !! Window lock
      INTEGER, INTENT(INOUT)                             :: win

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_win_lock_all'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)

      CALL mpi_win_lock_all(MPI_MODE_NOCHECK, win, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_win_lock_all @ "//routineN)
#else
      MARK_USED(win)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_win_lock_all

   SUBROUTINE mp_win_unlock_all(win)
      !! Window lock
      INTEGER, INTENT(INOUT)                             :: win

      CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_win_unlock_all'

      INTEGER                                            :: handle, ierr

      ierr = 0
      CALL timeset(routineN, handle)

#if defined(__parallel)

      CALL mpi_win_unlock_all(win, ierr)
      IF (ierr /= 0) CALL mp_stop(ierr, "mpi_win_unlock_all @ "//routineN)
#else
      MARK_USED(win)
#endif
      CALL timestop(handle)
   END SUBROUTINE mp_win_unlock_all

# 2606 "/__w/dbcsr/dbcsr/src/mpi/dbcsr_mpiwrap.F"
      SUBROUTINE mp_alltoall_i11v(sb, scount, sdispl, rb, rcount, rdispl, group)
      !! All-to-all data exchange, rank-1 data of different sizes
      !!
      !! MPI mapping
      !! mpi_alltoallv
      !!
      !! Array sizes
      !! The scount, rcount, and the sdispl and rdispl arrays have a
      !! size equal to the number of processes.
      !!
      !! Offsets
      !! Values in sdispl and rdispl start with 0.

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: sb(:)
         !! Data to send
         INTEGER, CONTIGUOUS, INTENT(IN)          :: scount(:), sdispl(:)
         !! Data counts for data sent to other processes
         !! Respective data offsets for data sent to process
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: rb(:)
         !! Buffer into which to receive data
         INTEGER, CONTIGUOUS, INTENT(IN)          :: rcount(:), rdispl(:)
         !! Data counts for data received from other processes
         !! Respective data offsets for data received from other processes
         INTEGER, INTENT(IN)                      :: group
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_alltoall_i11v'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: msglen
#else
         INTEGER                                  :: i
#endif

         CALL timeset(routineN, handle)

         ierr = 0
#if defined(__parallel)
         CALL mpi_alltoallv(sb, scount, sdispl, MPI_INTEGER, &
                            rb, rcount, rdispl, MPI_INTEGER, group, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_alltoallv @ "//routineN)
         msglen = SUM(scount) + SUM(rcount)
         CALL add_perf(perf_id=6, msg_size=msglen*int_4_size)
#else
         MARK_USED(group)
         MARK_USED(scount)
         MARK_USED(sdispl)
!$OMP     PARALLEL DO DEFAULT(NONE) PRIVATE(i) SHARED(rcount,rdispl,sdispl,rb,sb)
         DO i = 1, rcount(1)
            rb(rdispl(1) + i) = sb(sdispl(1) + i)
         END DO
#endif
         CALL timestop(handle)

      END SUBROUTINE mp_alltoall_i11v

      SUBROUTINE mp_alltoall_i (sb, rb, count, group)
      !! All-to-all data exchange, rank 1 arrays, equal sizes
      !!
      !! Index meaning
      !!
      !! The first two indices specify the data while the last index counts
      !! the processes
      !!
      !! Sizes of ranks
      !! All processes have the same data size.
      !!
      !! MPI mapping
      !! mpi_alltoall

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: sb(:)
         !! array with data to send
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: rb(:)
         !! array into which data is received
         INTEGER, INTENT(IN)                      :: count, group
         !! number of elements to send/receive (product of the extents of the first two dimensions)
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_alltoall_i'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: msglen, np
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         CALL mpi_alltoall(sb, count, MPI_INTEGER, &
                           rb, count, MPI_INTEGER, group, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_alltoall @ "//routineN)
         CALL mpi_comm_size(group, np, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_size @ "//routineN)
         msglen = 2*count*np
         CALL add_perf(perf_id=6, msg_size=msglen*int_4_size)
#else
         MARK_USED(count)
         MARK_USED(group)
         rb = sb
#endif
         CALL timestop(handle)

      END SUBROUTINE mp_alltoall_i

      SUBROUTINE mp_alltoall_i22(sb, rb, count, group)
      !! All-to-all data exchange, rank-2 arrays, equal sizes
      !! @note see mp_alltoall_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: sb(:, :)
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: rb(:, :)
         INTEGER, INTENT(IN)                      :: count, group

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_alltoall_i22'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: msglen, np
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         CALL mpi_alltoall(sb, count, MPI_INTEGER, &
                           rb, count, MPI_INTEGER, group, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_alltoall @ "//routineN)
         CALL mpi_comm_size(group, np, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_size @ "//routineN)
         msglen = 2*SIZE(sb)*np
         CALL add_perf(perf_id=6, msg_size=msglen*int_4_size)
#else
         MARK_USED(count)
         MARK_USED(group)
         rb = sb
#endif
         CALL timestop(handle)

      END SUBROUTINE mp_alltoall_i22

      SUBROUTINE mp_alltoall_i44(sb, rb, count, group)
      !! All-to-all data exchange, rank 4 data, equal sizes
      !! @note see mp_alltoall_i

         INTEGER(KIND=int_4), DIMENSION(:, :, :, :), CONTIGUOUS, &
            INTENT(IN)                            :: sb
         INTEGER(KIND=int_4), DIMENSION(:, :, :, :), CONTIGUOUS, &
            INTENT(OUT)                           :: rb
         INTEGER, INTENT(IN)                      :: count, group

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_alltoall_i44'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: msglen, np
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         CALL mpi_alltoall(sb, count, MPI_INTEGER, &
                           rb, count, MPI_INTEGER, group, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_alltoall @ "//routineN)
         CALL mpi_comm_size(group, np, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_size @ "//routineN)
         msglen = 2*count*np
         CALL add_perf(perf_id=6, msg_size=msglen*int_4_size)
#else
         MARK_USED(count)
         MARK_USED(group)
         rb = sb
#endif
         CALL timestop(handle)

      END SUBROUTINE mp_alltoall_i44

      SUBROUTINE mp_send_i (msg, dest, tag, gid)
      !! Send one datum to another process
      !!
      !! MPI mapping
      !! mpi_send

         INTEGER(KIND=int_4)                                :: msg
         !! Scalar to send
         INTEGER                                  :: dest, tag, gid
         !! Destination process
         !! Transfer identifier
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_send_i'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_send(msg, msglen, MPI_INTEGER, dest, tag, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_send @ "//routineN)
         CALL add_perf(perf_id=13, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(dest)
         MARK_USED(tag)
         MARK_USED(gid)
         ! only defined in parallel
         DBCSR_ABORT("not in parallel mode")
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_send_i

      SUBROUTINE mp_send_iv(msg, dest, tag, gid)
      !! Send rank-1 data to another process
      !! @note see mp_send_i

         INTEGER(KIND=int_4), CONTIGUOUS                    :: msg(:)
         !! Rank-1 data to send
         INTEGER                                  :: dest, tag, gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_send_iv'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_send(msg, msglen, MPI_INTEGER, dest, tag, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_send @ "//routineN)
         CALL add_perf(perf_id=13, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(dest)
         MARK_USED(tag)
         MARK_USED(gid)
         ! only defined in parallel
         DBCSR_ABORT("not in parallel mode")
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_send_iv

      SUBROUTINE mp_recv_i (msg, source, tag, gid)
      !! Receive one datum from another process
      !!
      !! MPI mapping
      !! mpi_send

         INTEGER(KIND=int_4), INTENT(INOUT)                 :: msg
         !! Place received data into this variable
         INTEGER, INTENT(INOUT)                   :: source, tag
         !! Process to receive from
         !! Transfer identifier
         INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_recv_i'

         INTEGER                                  :: handle, ierr, msglen
#if defined(__parallel)
         INTEGER, ALLOCATABLE, DIMENSION(:)       :: status
#endif

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         ALLOCATE (status(MPI_STATUS_SIZE))
         CALL mpi_recv(msg, msglen, MPI_INTEGER, source, tag, gid, status, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_recv @ "//routineN)
         CALL add_perf(perf_id=14, msg_size=msglen*int_4_size)
         source = status(MPI_SOURCE)
         tag = status(MPI_TAG)
         DEALLOCATE (status)
#else
         MARK_USED(msg)
         MARK_USED(source)
         MARK_USED(tag)
         MARK_USED(gid)
         ! only defined in parallel
         DBCSR_ABORT("not in parallel mode")
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_recv_i

      SUBROUTINE mp_recv_iv(msg, source, tag, gid)
      !! Receive rank-1 data from another process
      !! @note see mp_recv_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:)
         !! Place received data into this rank-1 array
         INTEGER, INTENT(INOUT)                   :: source, tag
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_recv_iv'

         INTEGER                                  :: handle, ierr, msglen
#if defined(__parallel)
         INTEGER, ALLOCATABLE, DIMENSION(:)       :: status
#endif

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         ALLOCATE (status(MPI_STATUS_SIZE))
         CALL mpi_recv(msg, msglen, MPI_INTEGER, source, tag, gid, status, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_recv @ "//routineN)
         CALL add_perf(perf_id=14, msg_size=msglen*int_4_size)
         source = status(MPI_SOURCE)
         tag = status(MPI_TAG)
         DEALLOCATE (status)
#else
         MARK_USED(msg)
         MARK_USED(source)
         MARK_USED(tag)
         MARK_USED(gid)
         ! only defined in parallel
         DBCSR_ABORT("not in parallel mode")
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_recv_iv

      SUBROUTINE mp_bcast_i (msg, source, gid)
      !! Broadcasts a datum to all processes.
      !!
      !! MPI mapping
      !! mpi_bcast

         INTEGER(KIND=int_4)                                :: msg
         !! Datum to broadcast
         INTEGER                                  :: source, gid
         !! Processes which broadcasts
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_bcast_i'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_bcast(msg, msglen, MPI_INTEGER, source, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_bcast @ "//routineN)
         CALL add_perf(perf_id=2, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(source)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_bcast_i

      SUBROUTINE mp_ibcast_i (msg, source, gid, request)
      !! Broadcasts a datum to all processes.
      !!
      !! MPI mapping
      !! mpi_bcast

         INTEGER(KIND=int_4)                                :: msg
         !! Datum to broadcast
         INTEGER                                  :: source, gid
         !! Processes which broadcasts
         !! Message passing environment identifier
         INTEGER, INTENT(INOUT)                   :: request

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_ibcast_i'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_ibcast(msg, msglen, MPI_INTEGER, source, gid, request, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_ibcast @ "//routineN)
         CALL add_perf(perf_id=22, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(source)
         MARK_USED(gid)
         request = mp_request_null
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_ibcast_i

      SUBROUTINE mp_bcast_iv(msg, source, gid)
      !! Broadcasts rank-1 data to all processes
      !! @note see mp_bcast_i1

         INTEGER(KIND=int_4), CONTIGUOUS                    :: msg(:)
         !! Data to broadcast
         INTEGER                                  :: source, gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_bcast_iv'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_bcast(msg, msglen, MPI_INTEGER, source, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_bcast @ "//routineN)
         CALL add_perf(perf_id=2, msg_size=msglen*int_4_size)
#else
         MARK_USED(source)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_bcast_iv

      SUBROUTINE mp_ibcast_iv(msg, source, gid, request)
      !! Broadcasts rank-1 data to all processes
      !! @note see mp_bcast_i1

         INTEGER(KIND=int_4), CONTIGUOUS                    :: msg(:)
         !! Data to broadcast
         INTEGER                                  :: source, gid
         INTEGER, INTENT(INOUT)                   :: request

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_ibcast_iv'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_ibcast(msg, msglen, MPI_INTEGER, source, gid, request, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_ibcast @ "//routineN)
         CALL add_perf(perf_id=22, msg_size=msglen*int_4_size)
#else
         MARK_USED(source)
         MARK_USED(gid)
         request = mp_request_null
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_ibcast_iv

      SUBROUTINE mp_bcast_im(msg, source, gid)
      !! Broadcasts rank-2 data to all processes
      !! @note see mp_bcast_i1

         INTEGER(KIND=int_4), CONTIGUOUS                    :: msg(:, :)
         !! Data to broadcast
         INTEGER                                  :: source, gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_bcast_im'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_bcast(msg, msglen, MPI_INTEGER, source, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_bcast @ "//routineN)
         CALL add_perf(perf_id=2, msg_size=msglen*int_4_size)
#else
         MARK_USED(source)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_bcast_im

      SUBROUTINE mp_bcast_i3(msg, source, gid)
      !! Broadcasts rank-3 data to all processes
      !! @note see mp_bcast_i1

         INTEGER(KIND=int_4), CONTIGUOUS                    :: msg(:, :, :)
         !! Data to broadcast
         INTEGER                                  :: source, gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_bcast_i3'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_bcast(msg, msglen, MPI_INTEGER, source, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_bcast @ "//routineN)
         CALL add_perf(perf_id=2, msg_size=msglen*int_4_size)
#else
         MARK_USED(source)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_bcast_i3

      SUBROUTINE mp_sum_i (msg, gid)
      !! Sums a datum from all processes with result left on all processes.
      !!
      !! MPI mapping
      !! mpi_allreduce

         INTEGER(KIND=int_4), INTENT(INOUT)    :: msg
         !! Datum to sum (input) and result (output)
         INTEGER, INTENT(IN)         :: gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_i'

         INTEGER                     :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_SUM, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_sum_i

      SUBROUTINE mp_sum_iv(msg, gid)
      !! Element-wise sum of a rank-1 array on all processes.
      !! @note see mp_sum_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:)
         !! Vector to sum and result
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_iv'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: msglen
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         msglen = SIZE(msg)
         IF (msglen > 0) THEN
            CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_SUM, gid, ierr)
            IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         END IF
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_sum_iv

      SUBROUTINE mp_isum_iv(msg, gid, request)
      !! Element-wise sum of a rank-1 array on all processes.
      !! @note see mp_sum_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:)
         !! Vector to sum and result
         INTEGER, INTENT(IN)                      :: gid
         INTEGER, INTENT(INOUT)                   :: request

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_isum_iv'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: msglen
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         msglen = SIZE(msg)
         IF (msglen > 0) THEN
            CALL mpi_iallreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_SUM, gid, request, ierr)
            IF (ierr /= 0) CALL mp_stop(ierr, "mpi_iallreduce @ "//routineN)
         ELSE
            request = mp_request_null
         END IF
         CALL add_perf(perf_id=23, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(gid)
         request = mp_request_null
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_isum_iv

      SUBROUTINE mp_sum_im(msg, gid)
      !! Element-wise sum of a rank-2 array on all processes.
      !! @note see mp_sum_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:, :)
         !! Matrix to sum and result
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_im'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER, PARAMETER :: max_msg = 2**25
         INTEGER                                  :: m1, msglen, step, msglensum
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         ! chunk up the call so that message sizes are limited, to avoid overflows in mpich triggered in large rpa calcs
         step = MAX(1, SIZE(msg, 2)/MAX(1, SIZE(msg)/max_msg))
         msglensum = 0
         DO m1 = LBOUND(msg, 2), UBOUND(msg, 2), step
            msglen = SIZE(msg, 1)*(MIN(UBOUND(msg, 2), m1 + step - 1) - m1 + 1)
            msglensum = msglensum + msglen
            IF (msglen > 0) THEN
               CALL mpi_allreduce(MPI_IN_PLACE, msg(LBOUND(msg, 1), m1), msglen, MPI_INTEGER, MPI_SUM, gid, ierr)
               IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
            END IF
         END DO
         CALL add_perf(perf_id=3, msg_size=msglensum*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_sum_im

      SUBROUTINE mp_sum_im3(msg, gid)
      !! Element-wise sum of a rank-3 array on all processes.
      !! @note see mp_sum_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:, :, :)
         !! Array to sum and result
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_im3'

         INTEGER                                  :: handle, ierr, &
                                                     msglen
         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         IF (msglen > 0) THEN
            CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_SUM, gid, ierr)
            IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         END IF
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_sum_im3

      SUBROUTINE mp_sum_im4(msg, gid)
      !! Element-wise sum of a rank-4 array on all processes.
      !! @note see mp_sum_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:, :, :, :)
         !! Array to sum and result
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_im4'

         INTEGER                                  :: handle, ierr, &
                                                     msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         IF (msglen > 0) THEN
            CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_SUM, gid, ierr)
            IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         END IF
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_sum_im4

      SUBROUTINE mp_sum_root_iv(msg, root, gid)
      !! Element-wise sum of data from all processes with result left only on
      !! one.
      !!
      !! MPI mapping
      !! mpi_reduce

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:)
         !! Vector to sum (input) and (only on process root) result (output)
         INTEGER, INTENT(IN)                      :: root, gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_root_iv'

         INTEGER                                  :: handle, ierr, msglen
#if defined(__parallel)
         INTEGER                                  :: m1, taskid
         INTEGER(KIND=int_4), ALLOCATABLE                     :: res(:)
#endif

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_comm_rank(gid, taskid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_rank @ "//routineN)
         IF (msglen > 0) THEN
            m1 = SIZE(msg, 1)
            ALLOCATE (res(m1))
            CALL mpi_reduce(msg, res, msglen, MPI_INTEGER, MPI_SUM, &
                            root, gid, ierr)
            IF (ierr /= 0) CALL mp_stop(ierr, "mpi_reduce @ "//routineN)
            IF (taskid == root) THEN
               msg = res
            END IF
            DEALLOCATE (res)
         END IF
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(root)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_sum_root_iv

      SUBROUTINE mp_sum_root_im(msg, root, gid)
      !! Element-wise sum of data from all processes with result left only on
      !! one.
      !! @note see mp_sum_root_iv

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:, :)
         !! Matrix to sum (input) and (only on process root) result (output)
         INTEGER, INTENT(IN)                      :: root, gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_root_rm'

         INTEGER                                  :: handle, ierr, msglen
#if defined(__parallel)
         INTEGER                                  :: m1, m2, taskid
         INTEGER(KIND=int_4), ALLOCATABLE                     :: res(:, :)
#endif

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_comm_rank(gid, taskid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_rank @ "//routineN)
         IF (msglen > 0) THEN
            m1 = SIZE(msg, 1)
            m2 = SIZE(msg, 2)
            ALLOCATE (res(m1, m2))
            CALL mpi_reduce(msg, res, msglen, MPI_INTEGER, MPI_SUM, root, gid, ierr)
            IF (ierr /= 0) CALL mp_stop(ierr, "mpi_reduce @ "//routineN)
            IF (taskid == root) THEN
               msg = res
            END IF
            DEALLOCATE (res)
         END IF
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(root)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_sum_root_im

      SUBROUTINE mp_sum_partial_im(msg, res, gid)
      !! Partial sum of data from all processes with result on each process.

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)  :: msg(:, :)
         !! Matrix to sum (input)
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT) :: res(:, :)
         !! Matrix containing result (output)
         INTEGER, INTENT(IN)                :: gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER        :: routineN = 'mp_sum_partial_im'

         INTEGER                            :: handle, ierr, msglen
#if defined(__parallel)
         INTEGER                            :: taskid
#endif

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_comm_rank(gid, taskid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_comm_rank @ "//routineN)
         IF (msglen > 0) THEN
            CALL mpi_scan(msg, res, msglen, MPI_INTEGER, MPI_SUM, gid, ierr)
            IF (ierr /= 0) CALL mp_stop(ierr, "mpi_scan @ "//routineN)
         END IF
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
         ! perf_id is same as for other summation routines
#else
         res = msg
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_sum_partial_im

      SUBROUTINE mp_max_i (msg, gid)
      !! Finds the maximum of a datum with the result left on all processes.
      !!
      !! MPI mapping
      !! mpi_allreduce

         INTEGER(KIND=int_4), INTENT(INOUT)                 :: msg
         !! Find maximum among these data (input) and maximum (output)
         INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_max_i'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_MAX, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_max_i

      SUBROUTINE mp_max_iv(msg, gid)
      !! Finds the element-wise maximum of a vector with the result left on
      !! all processes.
      !! @note see mp_max_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:)
         !! Find maximum among these data (input) and maximum (output)
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_max_iv'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_MAX, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_max_iv

      SUBROUTINE mp_min_i (msg, gid)
      !! Finds the minimum of a datum with the result left on all processes.
      !!
      !! MPI mapping
      !! mpi_allreduce

         INTEGER(KIND=int_4), INTENT(INOUT)                 :: msg
         !! Find minimum among these data (input) and maximum (output)
         INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_min_i'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_MIN, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_min_i

      SUBROUTINE mp_min_iv(msg, gid)
      !! Finds the element-wise minimum of vector with the result left on
      !! all processes.
      !!
      !! MPI mapping
      !! mpi_allreduce
      !! @note see mp_min_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:)
         !! Find minimum among these data (input) and maximum (output)
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_min_iv'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_MIN, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_min_iv

      SUBROUTINE mp_prod_i (msg, gid)
      !! Multiplies a set of numbers scattered across a number of processes,
      !! then replicates the result.
      !!
      !! MPI mapping
      !! mpi_allreduce

         INTEGER(KIND=int_4), INTENT(INOUT)                 :: msg
         !! a number to multiply (input) and result (output)
         INTEGER, INTENT(IN)                      :: gid
         !! message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_sum_i'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_allreduce(MPI_IN_PLACE, msg, msglen, MPI_INTEGER, MPI_PROD, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allreduce @ "//routineN)
         CALL add_perf(perf_id=3, msg_size=msglen*int_4_size)
#else
         MARK_USED(msg)
         MARK_USED(gid)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_prod_i

      SUBROUTINE mp_iscatter_i (msg_scatter, msg, root, gid, request)
      !! Scatters data from one processes to all others
      !!
      !! MPI mapping
      !! mpi_scatter

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: msg_scatter(:)
         !! Data to scatter (for root process)
         INTEGER(KIND=int_4), INTENT(INOUT)                 :: msg
         INTEGER, INTENT(IN)                      :: root, gid
         !! Process which scatters data
         !! Message passing environment identifier
         INTEGER, INTENT(INOUT)                   :: request

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_iscatter_i'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_iscatter(msg_scatter, msglen, MPI_INTEGER, msg, &
                           msglen, MPI_INTEGER, root, gid, request, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_iscatter @ "//routineN)
         CALL add_perf(perf_id=24, msg_size=1*int_4_size)
#else
         MARK_USED(root)
         MARK_USED(gid)
         msg = msg_scatter(1)
         request = mp_request_null
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_iscatter_i

      SUBROUTINE mp_iscatter_iv2(msg_scatter, msg, root, gid, request)
      !! Scatters data from one processes to all others
      !!
      !! MPI mapping
      !! mpi_scatter

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: msg_scatter(:, :)
         !! Data to scatter (for root process)
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:)
         INTEGER, INTENT(IN)                      :: root, gid
         !! Process which scatters data
         !! Message passing environment identifier
         INTEGER, INTENT(INOUT)                   :: request

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_iscatter_iv2'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_iscatter(msg_scatter, msglen, MPI_INTEGER, msg, &
                           msglen, MPI_INTEGER, root, gid, request, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_iscatter @ "//routineN)
         CALL add_perf(perf_id=24, msg_size=1*int_4_size)
#else
         MARK_USED(root)
         MARK_USED(gid)
         msg(:) = msg_scatter(:, 1)
         request = mp_request_null
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_iscatter_iv2

      SUBROUTINE mp_iscatterv_iv(msg_scatter, sendcounts, displs, msg, recvcount, root, gid, request)
      !! Scatters data from one processes to all others
      !!
      !! MPI mapping
      !! mpi_scatter

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: msg_scatter(:)
         !! Data to scatter (for root process)
         INTEGER, CONTIGUOUS, INTENT(IN)          :: sendcounts(:), displs(:)
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(INOUT)     :: msg(:)
         INTEGER, INTENT(IN)                      :: recvcount, root, gid
         !! Process which scatters data
         !! Message passing environment identifier
         INTEGER, INTENT(INOUT)                   :: request

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_iscatterv_iv'

         INTEGER                                  :: handle, ierr

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         CALL mpi_iscatterv(msg_scatter, sendcounts, displs, MPI_INTEGER, msg, &
                            recvcount, MPI_INTEGER, root, gid, request, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_iscatterv @ "//routineN)
         CALL add_perf(perf_id=24, msg_size=1*int_4_size)
#else
         MARK_USED(sendcounts)
         MARK_USED(displs)
         MARK_USED(recvcount)
         MARK_USED(root)
         MARK_USED(gid)
         msg(1:recvcount) = msg_scatter(1 + displs(1):1 + displs(1) + sendcounts(1))
         request = mp_request_null
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_iscatterv_iv

      SUBROUTINE mp_gather_i (msg, msg_gather, root, gid)
      !! Gathers a datum from all processes to one
      !!
      !! MPI mapping
      !! mpi_gather

         INTEGER(KIND=int_4), INTENT(IN)                    :: msg
         !! Datum to send to root
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msg_gather(:)
         !! Received data (on root)
         INTEGER, INTENT(IN)                      :: root, gid
         !! Process which gathers the data
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_gather_i'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = 1
#if defined(__parallel)
         CALL mpi_gather(msg, msglen, MPI_INTEGER, msg_gather, &
                         msglen, MPI_INTEGER, root, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_gather @ "//routineN)
         CALL add_perf(perf_id=4, msg_size=msglen*int_4_size)
#else
         MARK_USED(root)
         MARK_USED(gid)
         msg_gather(1) = msg
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_gather_i

      SUBROUTINE mp_gather_iv(msg, msg_gather, root, gid)
      !! Gathers data from all processes to one
      !!
      !! Data length
      !! All data (msg) is equal-sized
      !!
      !! MPI mapping
      !! mpi_gather
      !! @note see mp_gather_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: msg(:)
         !! Datum to send to root
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msg_gather(:)
         INTEGER, INTENT(IN)                      :: root, gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_gather_iv'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_gather(msg, msglen, MPI_INTEGER, msg_gather, &
                         msglen, MPI_INTEGER, root, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_gather @ "//routineN)
         CALL add_perf(perf_id=4, msg_size=msglen*int_4_size)
#else
         MARK_USED(root)
         MARK_USED(gid)
         msg_gather = msg
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_gather_iv

      SUBROUTINE mp_gather_im(msg, msg_gather, root, gid)
      !! Gathers data from all processes to one
      !!
      !! Data length
      !! All data (msg) is equal-sized
      !!
      !! MPI mapping
      !! mpi_gather
      !! @note see mp_gather_i

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: msg(:, :)
         !! Datum to send to root
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msg_gather(:, :)
         INTEGER, INTENT(IN)                      :: root, gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_gather_im'

         INTEGER                                  :: handle, ierr, msglen

         ierr = 0
         CALL timeset(routineN, handle)

         msglen = SIZE(msg)
#if defined(__parallel)
         CALL mpi_gather(msg, msglen, MPI_INTEGER, msg_gather, &
                         msglen, MPI_INTEGER, root, gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_gather @ "//routineN)
         CALL add_perf(perf_id=4, msg_size=msglen*int_4_size)
#else
         MARK_USED(root)
         MARK_USED(gid)
         msg_gather = msg
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_gather_im

      SUBROUTINE mp_gatherv_iv(sendbuf, recvbuf, recvcounts, displs, root, comm)
      !! Gathers data from all processes to one.
      !!
      !! Data length
      !! Data can have different lengths
      !!
      !! Offsets
      !! Offsets start at 0
      !!
      !! MPI mapping
      !! mpi_gather

         INTEGER(KIND=int_4), DIMENSION(:), CONTIGUOUS, INTENT(IN)      :: sendbuf
         !! Data to send to root
         INTEGER(KIND=int_4), DIMENSION(:), CONTIGUOUS, INTENT(OUT)     :: recvbuf
         !! Received data (on root)
         INTEGER, DIMENSION(:), CONTIGUOUS, INTENT(IN)        :: recvcounts, displs
         !! Sizes of data received from processes
         !! Offsets of data received from processes
         INTEGER, INTENT(IN)                      :: root, comm
         !! Process which gathers the data
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_gatherv_iv'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: sendcount
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         sendcount = SIZE(sendbuf)
         CALL mpi_gatherv(sendbuf, sendcount, MPI_INTEGER, &
                          recvbuf, recvcounts, displs, MPI_INTEGER, &
                          root, comm, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_gatherv @ "//routineN)
         CALL add_perf(perf_id=4, &
                       msg_size=sendcount*int_4_size)
#else
         MARK_USED(recvcounts)
         MARK_USED(root)
         MARK_USED(comm)
         recvbuf(1 + displs(1):) = sendbuf
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_gatherv_iv

      SUBROUTINE mp_allgather_i (msgout, msgin, gid)
      !! Gathers a datum from all processes and all processes receive the
      !! same data
      !!
      !! Data size
      !! All processes send equal-sized data
      !!
      !! MPI mapping
      !! mpi_allgather

         INTEGER(KIND=int_4), INTENT(IN)                    :: msgout
         !! Datum to send
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msgin(:)
         !! Received data
         INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_allgather_i'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: rcount, scount
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         scount = 1
         rcount = 1
         CALL MPI_ALLGATHER(msgout, scount, MPI_INTEGER, &
                            msgin, rcount, MPI_INTEGER, &
                            gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allgather @ "//routineN)
#else
         MARK_USED(gid)
         msgin = msgout
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_allgather_i

      SUBROUTINE mp_allgather_i2(msgout, msgin, gid)
      !! Gathers a datum from all processes and all processes receive the
      !! same data
      !!
      !! Data size
      !! All processes send equal-sized data
      !!
      !! MPI mapping
      !! mpi_allgather

         INTEGER(KIND=int_4), INTENT(IN)                    :: msgout
         !! Datum to send
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msgin(:, :)
         !! Received data
         INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_allgather_i2'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: rcount, scount
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         scount = 1
         rcount = 1
         CALL MPI_ALLGATHER(msgout, scount, MPI_INTEGER, &
                            msgin, rcount, MPI_INTEGER, &
                            gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allgather @ "//routineN)
#else
         MARK_USED(gid)
         msgin = msgout
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_allgather_i2

      SUBROUTINE mp_iallgather_i (msgout, msgin, gid, request)
      !! Gathers a datum from all processes and all processes receive the
      !! same data
      !!
      !! Data size
      !! All processes send equal-sized data
      !!
      !! MPI mapping
      !! mpi_allgather

         INTEGER(KIND=int_4), INTENT(IN)                    :: msgout
         !! Datum to send
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msgin(:)
         !! Received data
         INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier
         INTEGER, INTENT(INOUT)                   :: request

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_iallgather_i'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: rcount, scount
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         scount = 1
         rcount = 1
         CALL MPI_IALLGATHER(msgout, scount, MPI_INTEGER, &
                             msgin, rcount, MPI_INTEGER, &
                             gid, request, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_iallgather @ "//routineN)
#else
         MARK_USED(gid)
         msgin = msgout
         request = mp_request_null
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_iallgather_i

      SUBROUTINE mp_allgather_i12(msgout, msgin, gid)
      !! Gathers vector data from all processes and all processes receive the
      !! same data
      !!
      !! Data size
      !! All processes send equal-sized data
      !!
      !! Ranks
      !! The last rank counts the processes
      !!
      !! MPI mapping
      !! mpi_allgather

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: msgout(:)
         !! Rank-1 data to send
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msgin(:, :)
         !! Received data
         INTEGER, INTENT(IN)                      :: gid
         !! Message passing environment identifier

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_allgather_i12'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: rcount, scount
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         scount = SIZE(msgout(:))
         rcount = scount
         CALL MPI_ALLGATHER(msgout, scount, MPI_INTEGER, &
                            msgin, rcount, MPI_INTEGER, &
                            gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allgather @ "//routineN)
#else
         MARK_USED(gid)
         msgin(:, 1) = msgout(:)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_allgather_i12

      SUBROUTINE mp_allgather_i23(msgout, msgin, gid)
      !! Gathers matrix data from all processes and all processes receive the
      !! same data
      !! @note see mp_allgather_i12

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: msgout(:, :)
         !! Rank-2 data to send
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msgin(:, :, :)
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_allgather_i23'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER                                  :: rcount, scount
#endif

         ierr = 0
         CALL timeset(routineN, handle)

#if defined(__parallel)
         scount = SIZE(msgout(:, :))
         rcount = scount
         CALL MPI_ALLGATHER(msgout, scount, MPI_INTEGER, &
                            msgin, rcount, MPI_INTEGER, &
                            gid, ierr)
         IF (ierr /= 0) CALL mp_stop(ierr, "mpi_allgather @ "//routineN)
#else
         MARK_USED(gid)
         msgin(:, :, 1) = msgout(:, :)
#endif
         CALL timestop(handle)
      END SUBROUTINE mp_allgather_i23

      SUBROUTINE mp_allgather_i34(msgout, msgin, gid)
      !! Gathers rank-3 data from all processes and all processes receive the
      !! same data
      !! @note see mp_allgather_i12

         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(IN)        :: msgout(:, :, :)
         !! Rank-3 data to send
         INTEGER(KIND=int_4), CONTIGUOUS, INTENT(OUT)       :: msgin(:, :, :, :)
         INTEGER, INTENT(IN)                      :: gid

         CHARACTER(LEN=*), PARAMETER :: routineN = 'mp_allgather_i34'

         INTEGER                                  :: handle, ierr
#if defined(__parallel)
         INTEGER