Commit 1427731f authored by Andreas Marek's avatar Andreas Marek

Test case for complex banded matrix

parent 1df480c8
......@@ -313,6 +313,7 @@ dist_files_DATA = \
test/Fortran/test_complex2.F90 \
test/Fortran/test_complex2_default.F90 \
test/Fortran/test_complex2_api.F90 \
test/Fortran/test_complex2_banded.F90 \
test/Fortran/test_complex.F90 \
test/Fortran/test_real2.F90 \
test/Fortran/test_real2_default.F90 \
......@@ -354,6 +355,7 @@ noinst_PROGRAMS = \
elpa2_test_complex_default@SUFFIX@ \
elpa2_test_complex_api@SUFFIX@ \
elpa2_test_complex_api@SUFFIX@ \
elpa2_test_complex_banded@SUFFIX@ \
elpa_driver_real@SUFFIX@ \
elpa_driver_complex@SUFFIX@ \
elpa1_real_toeplitz@SUFFIX@ \
......@@ -563,6 +565,12 @@ elpa2_test_complex_api@SUFFIX@_LDADD = $(build_lib)
elpa2_test_complex_api@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
EXTRA_elpa2_test_complex_api@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa2_test_complex_banded@SUFFIX@_SOURCES = test/Fortran/test_complex2_banded.F90
elpa2_test_complex_banded@SUFFIX@_LDADD = $(build_lib)
elpa2_test_complex_banded@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
EXTRA_elpa2_test_complex_banded@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa_driver_real@SUFFIX@_SOURCES = test/Fortran/test_driver_real.F90
elpa_driver_real@SUFFIX@_LDADD = $(build_lib)
elpa_driver_real@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
......@@ -744,6 +752,7 @@ check_SCRIPTS = \ \ \ \ \ \ \ \
! This file is part of ELPA.
! The ELPA library was originally created by the ELPA consortium,
! consisting of the following organizations:
! - Max Planck Computing and Data Facility (MPCDF), formerly known as
! Rechenzentrum Garching der Max-Planck-Gesellschaft (RZG),
! - Bergische Universität Wuppertal, Lehrstuhl für angewandte
! Informatik,
! - Technische Universität München, Lehrstuhl für Informatik mit
! Schwerpunkt Wissenschaftliches Rechnen ,
! - Fritz-Haber-Institut, Berlin, Abt. Theorie,
! - Max-Plack-Institut für Mathematik in den Naturwissenschaften,
! Leipzig, Abt. Komplexe Strukutren in Biologie und Kognition,
! and
! - IBM Deutschland GmbH
! More information can be found here:
! ELPA is free software: you can redistribute it and/or modify
! it under the terms of the version 3 of the license of the
! GNU Lesser General Public License as published by the Free
! Software Foundation.
! ELPA is distributed in the hope that it will be useful,
! but WITHOUT ANY WARRANTY; without even the implied warranty of
! GNU Lesser General Public License for more details.
! You should have received a copy of the GNU Lesser General Public License
! along with ELPA. If not, see <>
! ELPA reflects a substantial effort on the part of the original
! ELPA consortium, and we ask you to respect the spirit of the
! license that we chose: i.e., please contribute any changes you
! may have back to the original ELPA library distribution, and keep
! any derivatives of ELPA under the same license that we chose for
! the original distribution, the GNU Lesser General Public License.
#include "config-f90.h"
!> Fortran test programm to demonstrates the use of
!> ELPA 2 complex case library.
!> If "HAVE_REDIRECT" was defined at build time
!> the stdout and stderr output of each MPI task
!> can be redirected to files if the environment
!> variable "REDIRECT_ELPA_TEST_OUTPUT" is set
!> to "true".
!> By calling executable [arg1] [arg2] [arg3] [arg4]
!> one can define the size (arg1), the number of
!> Eigenvectors to compute (arg2), and the blocking (arg3).
!> If these values are not set default values (4000, 1500, 16)
!> are choosen.
!> If these values are set the 4th argument can be
!> "output", which specifies that the EV's are written to
!> an ascii file.
!> The complex ELPA 2 kernel is set as the default kernel.
!> However, this can be overriden by setting
!> the environment variable "COMPLEX_ELPA_KERNEL" to an
!> appropiate value.
program test_complex2_double_precision
! Standard eigenvalue problem - COMPLEX version
! This program demonstrates the use of the ELPA module
! together with standard scalapack routines
! Copyright of the original code rests with the authors inside the ELPA
! consortium. The copyright of any additional modifications shall rest
! with their original authors, but shall adhere to the licensing terms
! distributed along with the original code in the file "COPYING".
use precision
use ELPA1
use ELPA2
use mod_check_for_gpu, only : check_for_gpu
use elpa_utilities, only : error_unit
use test_util
use mod_read_input_parameters
use mod_check_correctness
use mod_setup_mpi
use mod_blacs_infrastructure
use mod_prepare_matrix
use elpa_mpi
use redirect
use timings
use output_types
implicit none
! Please set system size parameters below!
! na: System size
! nev: Number of eigenvectors to be calculated
! nblk: Blocking factor in block cyclic distribution
integer(kind=ik) :: nblk
integer(kind=ik) :: na, nev
integer(kind=ik) :: np_rows, np_cols, na_rows, na_cols
integer(kind=ik) :: myid, nprocs, my_prow, my_pcol, mpi_comm_rows, mpi_comm_cols
integer(kind=ik) :: i, mpierr, my_blacs_ctxt, sc_desc(9), info, nprow, npcol
#ifdef WITH_MPI
integer(kind=ik), external :: numroc
complex(kind=ck8), parameter :: CZERO = (0.0_rk8,0.0_rk8), CONE = (1.0_rk8,0.0_rk8)
real(kind=rk8), allocatable :: ev(:), xr(:,:)
complex(kind=ck8), allocatable :: a(:,:), z(:,:), tmp1(:,:), tmp2(:,:), as(:,:)
integer(kind=ik) :: iseed(4096) ! Random seed, size should be sufficient for every generator
integer(kind=ik) :: STATUS
integer(kind=ik) :: omp_get_max_threads, required_mpi_thread_level, provided_mpi_thread_level
type(output_t) :: write_to_file
logical :: success
character(len=8) :: task_suffix
integer(kind=ik) :: j
logical :: successELPA
integer(kind=ik) :: numberOfDevices
logical :: gpuAvailable
integer(kind=ik) :: global_row, global_col, local_row, local_col
integer(kind=ik) :: bandwidth
successELPA = .true.
gpuAvailable = .false.
call read_input_parameters(na, nev, nblk, write_to_file)
! MPI Initialization
call setup_mpi(myid, nprocs)
gpuAvailable = check_for_gpu(myid, numberOfDevices)
#include "elpa_print_headers.X90"
! initialise the timing functionality
call timer%measure_flops(.true.)
call timer%measure_allocated_memory(.true.)
call timer%measure_virtual_memory(.true.)
call timer%measure_max_allocated_memory(.true.)
call timer%set_print_options(&
print_flop_count=.true., &
print_flop_rate=.true., &
print_allocated_memory = .true. , &
print_virtual_memory=.true., &
call timer%enable()
call timer%start("program: test_complex2_double_precision")
! Selection of number of processor rows/columns
! We try to set up the grid square-like, i.e. start the search for possible
! divisors of nprocs with a number next to the square root of nprocs
! and decrement it until a divisor is found.
do np_cols = NINT(SQRT(REAL(nprocs))),2,-1
if(mod(nprocs,np_cols) == 0 ) exit
! at the end of the above loop, nprocs is always divisible by np_cols
np_rows = nprocs/np_cols
if(myid==0) then
print *
print '(a)','Standard eigenvalue problem - COMPLEX version'
print *
print '(3(a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
print '(3(a,i0))','Number of processor rows=',np_rows,', cols=',np_cols,', total=',nprocs
print *
! Set up BLACS context and MPI communicators
! The BLACS context is only necessary for using Scalapack.
! For ELPA, the MPI communicators along rows/cols are sufficient,
! and the grid setup may be done in an arbitrary way as long as it is
! consistent (i.e. 0<=my_prow<np_rows, 0<=my_pcol<np_cols and every
! process has a unique (my_prow,my_pcol) pair).
call set_up_blacsgrid(mpi_comm_world, my_blacs_ctxt, np_rows, np_cols, &
nprow, npcol, my_prow, my_pcol)
if (myid==0) then
print '(a)','| Past BLACS_Gridinfo.'
end if
! All ELPA routines need MPI communicators for communicating within
! rows or columns of processes, these are set in elpa_get_communicators.
mpierr = elpa_get_communicators(mpi_comm_world, my_prow, my_pcol, &
mpi_comm_rows, mpi_comm_cols)
if (myid==0) then
print '(a)','| Past split communicator setup for rows and columns.'
end if
! Determine the necessary size of the distributed matrices,
! we use the Scalapack tools routine NUMROC for that.
call set_up_blacs_descriptor(na ,nblk, my_prow, my_pcol, np_rows, np_cols, &
na_rows, na_cols, sc_desc, my_blacs_ctxt, info)
if (myid==0) then
print '(a)','| Past scalapack descriptor setup.'
end if
! Allocate matrices and set up a test matrix for the eigenvalue problem
call timer%start("set up matrix")
allocate(a (na_rows,na_cols))
allocate(z (na_rows,na_cols))
call prepare_matrix_double(na, myid, sc_desc, iseed, xr, a, z, as)
! set values outside of the bandwidth to zero
bandwidth = nblk
do local_row = 1, na_rows
global_row = index_l2g( local_row, nblk, my_prow, np_rows )
do local_col = 1, na_cols
global_col = index_l2g( local_col, nblk, my_pcol, np_cols )
if (ABS(global_row-global_col) > bandwidth) then
a(local_row, local_col) = 0
as(local_row, local_col) = 0
end if
end do
end do
call timer%stop("set up matrix")
! set print flag in elpa1
elpa_print_times = .true.
! Calculate eigenvalues/eigenvectors
#ifdef WITH_MPI
call mpi_barrier(mpi_comm_world, mpierr) ! for correct timings only
successELPA = elpa_solve_evp_complex_2stage_double(na, nev, a, na_rows, ev, z, na_rows, nblk, &
na_cols, mpi_comm_rows, mpi_comm_cols, mpi_comm_world, bandwidth=bandwidth)
if (.not.(successELPA)) then
write(error_unit,*) "solve_evp_complex_2stage produced an error! Aborting..."
#ifdef WITH_MPI
call MPI_ABORT(mpi_comm_world, 1, mpierr)
if(myid == 0) print *,'Time transform to tridi :',time_evp_fwd
if(myid == 0) print *,'Time solve tridi :',time_evp_solve
if(myid == 0) print *,'Time transform back EVs :',time_evp_back
if(myid == 0) print *,'Total time (sum above) :',time_evp_back+time_evp_solve+time_evp_fwd
if(write_to_file%eigenvectors) then
write(unit = task_suffix, fmt = '(i8.8)') myid
write(17,*) "Part of eigenvectors: na_rows=",na_rows,"of na=",na," na_cols=",na_cols," of na=",na
do i=1,na_rows
do j=1,na_cols
write(17,*) "row=",i," col=",j," element of eigenvector=",z(i,j)
if(write_to_file%eigenvalues) then
if (myid == 0) then
do i=1,na
write(17,*) i,ev(i)
! Test correctness of result (using plain scalapack routines)
status = check_correctness(na, nev, as, z, ev, sc_desc, myid, tmp1, tmp2)
call timer%stop("program: test_complex2_double_precision")
print *," "
print *,"Timings program: test_complex2_double_precision"
call timer%print("program: test_complex2_double_precision")
print *," "
print *,"End timings program: test_complex2_double_precision"
#ifdef WITH_MPI
call blacs_gridexit(my_blacs_ctxt)
call mpi_finalize(mpierr)
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment