Skip to content

GitLab

Commit 9e3f0636 authored by Andreas Marek's avatar Andreas Marek
Browse files

New feature "bandwidth" 

Setting a bandwdith of the input matrix via

elpa%set("bandwidth",bw)

tells ELPA 2stage, that matrix is already a banded matrix
with bandwidth bw. Then some compute steps can be skipped
parent a59bc1fd
Hide whitespace changes
Inline Side-by-side
Showing with 816 additions and 155 deletions
Makefile.am
View file @ 9e3f0636
......@@ -417,6 +417,8 @@ noinst_PROGRAMS = \
test_real_double_2stage \
test_complex_double_1stage \
test_complex_double_2stage \
elpa_test_new_interface_real_2stage_banded@SUFFIX@ \
elpa_test_new_interface_complex_2stage_banded@SUFFIX@ \
elpa_test_new_interface_real_2stage@SUFFIX@ \
elpa_test_new_interface_complex_2stage@SUFFIX@ \
elpa_test_new_interface_real_1stage@SUFFIX@ \
......@@ -426,6 +428,7 @@ if WANT_SINGLE_PRECISION_REAL
noinst_PROGRAMS += \
test_real_single_1stage \
test_real_single_2stage \
elpa_test_new_interface_real_single_2stage_banded@SUFFIX@ \
elpa_test_new_interface_real_single_2stage@SUFFIX@ \
elpa_test_new_interface_real_single_1stage@SUFFIX@
endif
......@@ -434,6 +437,7 @@ if WANT_SINGLE_PRECISION_COMPLEX
noinst_PROGRAMS += \
test_complex_single_1stage \
test_complex_single_2stage \
elpa_test_new_interface_complex_single_2stage_banded@SUFFIX@ \
elpa_test_new_interface_complex_single_2stage@SUFFIX@ \
elpa_test_new_interface_complex_single_1stage@SUFFIX@
endif
......@@ -467,11 +471,9 @@ noinst_PROGRAMS += \
elpa2_test_real_default@SUFFIX@ \
elpa2_test_real_qr@SUFFIX@ \
elpa2_test_real_api@SUFFIX@ \
elpa2_test_real_banded@SUFFIX@ \
elpa2_test_complex@SUFFIX@ \
elpa2_test_complex_default@SUFFIX@ \
elpa2_test_complex_api@SUFFIX@ \
elpa2_test_complex_banded@SUFFIX@ \
elpa_driver_real@SUFFIX@ \
elpa_driver_complex@SUFFIX@ \
elpa1_real_toeplitz@SUFFIX@ \
......@@ -823,10 +825,10 @@ elpa2_test_real_api@SUFFIX@_LDADD = $(build_lib)
elpa2_test_real_api@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
EXTRA_elpa2_test_real_api@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa2_test_real_banded@SUFFIX@_SOURCES = test/Fortran/test_real2_banded.F90
elpa2_test_real_banded@SUFFIX@_LDADD = $(build_lib)
elpa2_test_real_banded@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
EXTRA_elpa2_test_real_banded@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa_test_new_interface_real_2stage_banded@SUFFIX@_SOURCES = test/Fortran/test_new_interface_real_2stage_banded.F90
elpa_test_new_interface_real_2stage_banded@SUFFIX@_LDADD = $(build_lib)
elpa_test_new_interface_real_2stage_banded@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
EXTRA_elpa_test_new_interface_real_2stage_banded@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa1_test_complex@SUFFIX@_SOURCES = test/Fortran/elpa1/complex.F90
elpa1_test_complex@SUFFIX@_LDADD = $(build_lib)
......@@ -848,10 +850,10 @@ 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_test_new_interface_complex_2stage_banded@SUFFIX@_SOURCES = test/Fortran/test_new_interface_complex_2stage_banded.F90
elpa_test_new_interface_complex_2stage_banded@SUFFIX@_LDADD = $(build_lib)
elpa_test_new_interface_complex_2stage_banded@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
EXTRA_elpa_test_new_interface_complex_2stage_banded@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa_driver_real@SUFFIX@_SOURCES = test/Fortran/test_driver_real.F90
......@@ -869,6 +871,11 @@ elpa_tests@SUFFIX@_LDADD = $(build_lib)
elpa_tests@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
if WANT_SINGLE_PRECISION_REAL
elpa_test_new_interface_real_single_2stage_banded@SUFFIX@_SOURCES = test/Fortran/test_new_interface_real_single_2stage_banded.F90
elpa_test_new_interface_real_single_2stage_banded@SUFFIX@_LDADD = $(build_lib)
elpa_test_new_interface_real_single_2stage_banded@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
EXTRA_elpa_test_new_interface_real_single_2stage_banded@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa1_test_real_single_precision@SUFFIX@_SOURCES = test/Fortran/elpa1/single_real.F90
elpa1_test_real_single_precision@SUFFIX@_LDADD = $(build_lib)
elpa1_test_real_single_precision@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
......@@ -927,6 +934,12 @@ EXTRA_elpa2_test_real_api_single_precision@SUFFIX@_DEPENDENCIES = test/Fortran/e
endif
if WANT_SINGLE_PRECISION_COMPLEX
elpa_test_new_interface_complex_single_2stage_banded@SUFFIX@_SOURCES = test/Fortran/test_new_interface_complex_single_2stage_banded.F90
elpa_test_new_interface_complex_single_2stage_banded@SUFFIX@_LDADD = $(build_lib)
elpa_test_new_interface_complex_single_2stage_banded@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
EXTRA_elpa_test_new_interface_complex_single_2stage_banded@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa1_test_complex_single_precision@SUFFIX@_SOURCES = test/Fortran/elpa1/single_complex.F90
elpa1_test_complex_single_precision@SUFFIX@_LDADD = $(build_lib)
elpa1_test_complex_single_precision@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) $(FC_MODOUT)private_modules $(FC_MODINC)private_modules
......@@ -1025,6 +1038,8 @@ check_SCRIPTS = \
test_real_double_2stage.sh \
test_complex_double_1stage.sh \
test_complex_double_2stage.sh \
elpa_test_new_interface_real_2stage_banded@SUFFIX@.sh \
elpa_test_new_interface_complex_2stage_banded@SUFFIX@.sh \
elpa_test_new_interface_real_2stage@SUFFIX@.sh \
elpa_test_new_interface_complex_2stage@SUFFIX@.sh \
elpa_test_new_interface_real_1stage@SUFFIX@.sh \
......@@ -1034,6 +1049,7 @@ if WANT_SINGLE_PRECISION_REAL
check_SCRIPTS += \
test_real_single_1stage.sh \
test_real_single_2stage.sh \
elpa_test_new_interface_real_single_2stage_banded@SUFFIX@.sh \
elpa_test_new_interface_real_single_2stage@SUFFIX@.sh \
elpa_test_new_interface_real_single_1stage@SUFFIX@.sh
endif
......@@ -1042,6 +1058,7 @@ if WANT_SINGLE_PRECISION_COMPLEX
check_SCRIPTS += \
test_complex_single_1stage.sh \
test_complex_single_2stage.sh \
elpa_test_new_interface_complex_single_2stage_banded@SUFFIX@.sh \
elpa_test_new_interface_complex_single_2stage@SUFFIX@.sh \
elpa_test_new_interface_complex_single_1stage@SUFFIX@.sh
endif
......@@ -1076,9 +1093,7 @@ check_SCRIPTS += \
elpa2_test_real_qr@SUFFIX@.sh \
elpa2_test_complex_default@SUFFIX@.sh \
elpa2_test_real_api@SUFFIX@.sh \
elpa2_test_real_banded@SUFFIX@.sh \
elpa2_test_complex_api@SUFFIX@.sh \
elpa2_test_complex_banded@SUFFIX@.sh \
elpa_driver_real@SUFFIX@.sh \
elpa_driver_complex@SUFFIX@.sh \
elpa1_real_toeplitz@SUFFIX@.sh \
......
src/elpa2/elpa2_template.X90
View file @ 9e3f0636
......@@ -255,18 +255,18 @@
endif
endif
bandwidth = -1
if (bandwidth .ne. -1) then
nbw = bandwidth
if (obj%is_set("bandwidth") == 1) then
nbw=obj%get("bandwidth")
if ((nbw == 0) .or. (mod(nbw, nblk) .ne. 0)) then
if (wantDebug) then
write(error_unit,*) "Specified bandwidth has to be a multiple of blocksize"
write(error_unit,*) "Specified bandwidth has to be a multiple of blocksize: ",nbw
endif
print *, "Specified bandwidth has to be a multiple of blocksize"
success = .false.
return
endif
!ttts = MPI_Wtime()
else
! Choose bandwidth, must be a multiple of nblk, set to a value >= 32
......@@ -404,7 +404,7 @@
stop 1
endif
if( bandwidth == -1) then
if(obj%is_set("bandwidth") .ne. 1) then
if ( (do_useGPU) .and. .not.(do_useGPU_trans_ev_tridi) ) then
! copy to device if we want to continue on GPU
successCUDA = cuda_malloc(q_dev, ldq*matrixCols*size_of_datatype)
......
src/elpa_index.c
View file @ 9e3f0636
......@@ -121,7 +121,7 @@ static const elpa_index_int_entry_t int_entries[] = {
INT_PARAMETER_ENTRY("local_ncols", "Number of matrix columns stored on this process"),
INT_PARAMETER_ENTRY("process_row", "Process row number in the 2D domain decomposition"),
INT_PARAMETER_ENTRY("process_col", "Process column number in the 2D domain decomposition"),
INT_ENTRY("bandwidth", "If specified, a band matrix with this bandwidth is expected as input", -1,
INT_ENTRY("bandwidth", "If specified, a band matrix with this bandwidth is expected as input; bandwidth must be multiply of nblk", -1,
NULL, NULL, bw_is_valid, NULL),
INT_ANY_ENTRY("mpi_comm_rows", "Communicator for inter-row communication"),
INT_ANY_ENTRY("mpi_comm_cols", "Communicator for inter-column communication"),
......
test/Fortran/test_complex2_banded.F90 → test/Fortran/test_new_interface_complex_2stage_banded.F90
View file @ 9e3f0636
......@@ -41,6 +41,7 @@
!
!
#include "config-f90.h"
#include "assert.h"
!>
!> Fortran test programm to demonstrates the use of
!> ELPA 2 complex case library.
......@@ -64,7 +65,7 @@
!> the environment variable "COMPLEX_ELPA_KERNEL" to an
!> appropiate value.
!>
program test_complex2_double_precision
program test_complex2_double_banded
!-------------------------------------------------------------------------------
! Standard eigenvalue problem - COMPLEX version
......@@ -78,10 +79,8 @@ program test_complex2_double_precision
! distributed along with the original code in the file "COPYING".
!-------------------------------------------------------------------------------
use precision
use elpa1
use elpa2
use elpa
use mod_check_for_gpu, only : check_for_gpu
use elpa_utilities, only : error_unit
#ifdef WITH_OPENMP
use test_util
......@@ -130,21 +129,20 @@ program test_complex2_double_precision
integer(kind=ik) :: omp_get_max_threads, required_mpi_thread_level, provided_mpi_thread_level
#endif
type(output_t) :: write_to_file
logical :: success
integer(kind=ik) :: 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
class(elpa_t), pointer :: e
#define COMPLEXCASE
#define DOUBLE_PRECISION_COMPLEX 1
successELPA = .true.
gpuAvailable = .false.
call read_input_parameters_traditional(na, nev, nblk, write_to_file)
......@@ -182,7 +180,7 @@ program test_complex2_double_precision
call timer%enable()
call timer%start("program: test_complex2_double_precision")
call timer%start("program: test_complex2_double_banded")
#endif
!-------------------------------------------------------------------------------
......@@ -224,16 +222,6 @@ program test_complex2_double_precision
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.
......@@ -277,50 +265,53 @@ program test_complex2_double_precision
call timer%stop("set up matrix")
#endif
! 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
if (elpa_init(CURRENT_API_VERSION) /= ELPA_OK) then
print *, "ELPA API version not supported"
stop 1
endif
#ifdef HAVE_DETAILED_TIMINGS
call timer%start("prepare_elpa")
#endif
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)
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)
e => elpa_allocate()
call e%set("na", na, success)
assert_elpa_ok(success)
call e%set("nev", nev, success)
assert_elpa_ok(success)
call e%set("local_nrows", na_rows, success)
assert_elpa_ok(success)
call e%set("local_ncols", na_cols, success)
assert_elpa_ok(success)
call e%set("nblk", nblk, success)
assert_elpa_ok(success)
call e%set("mpi_comm_parent", MPI_COMM_WORLD, success)
assert_elpa_ok(success)
call e%set("process_row", my_prow, success)
assert_elpa_ok(success)
call e%set("process_col", my_pcol, success)
assert_elpa_ok(success)
call e%set("bandwidth", bandwidth, success)
assert_elpa_ok(success)
assert(e%setup() .eq. ELPA_OK)
call e%set("solver", ELPA_SOLVER_2STAGE, success)
assert_elpa_ok(success)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("prepare_elpa")
#endif
endif
#ifdef HAVE_DETAILED_TIMINGS
call timer%start("solve")
#endif
call e%solve(a, ev, z, success)
assert_elpa_ok(success)
#ifdef HAVE_DETAILED_TIMINGS
call timer%stop("solve")
#endif
call elpa_deallocate(e)
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
open(17,file="EVs_complex2_out_task_"//task_suffix(1:8)//".txt",form='formatted',status='new')
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)
enddo
enddo
close(17)
endif
if(write_to_file%eigenvalues) then
if (myid == 0) then
open(17,file="Eigenvalues_complex2_out.txt",form='formatted',status='new')
do i=1,na
write(17,*) i,ev(i)
enddo
close(17)
endif
endif
call elpa_uninit()
!-------------------------------------------------------------------------------
! Test correctness of result (using plain scalapack routines)
......
test/Fortran/test_new_interface_complex_single_2stage_banded.F90 0 → 100644
View file @ 9e3f0636
! 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:
! http://elpa.mpcdf.mpg.de/
!
! 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
! MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
! 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 <http://www.gnu.org/licenses/>
!
! 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"
#include "assert.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_single_banded
!-------------------------------------------------------------------------------
! 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 elpa
use mod_check_for_gpu, only : check_for_gpu
#ifdef WITH_OPENMP
use test_util
#endif
use mod_read_input_parameters
use mod_check_correctness
use mod_setup_mpi
use mod_blacs_infrastructure
use mod_prepare_matrix
use elpa_mpi
#ifdef HAVE_REDIRECT
use redirect
#endif
#ifdef HAVE_DETAILED_TIMINGS
use timings
#endif
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
#endif
complex(kind=ck4), parameter :: CZERO = (0.0_rk4,0.0_rk4), CONE = (1.0_rk4,0.0_rk4)
real(kind=rk4), allocatable :: ev(:)
complex(kind=ck4), allocatable :: a(:,:), z(:,:), as(:,:)
integer(kind=ik) :: STATUS
#ifdef WITH_OPENMP
integer(kind=ik) :: omp_get_max_threads, required_mpi_thread_level, provided_mpi_thread_level
#endif
type(output_t) :: write_to_file
integer(kind=ik) :: success
character(len=8) :: task_suffix
integer(kind=ik) :: j
integer(kind=ik) :: numberOfDevices
logical :: gpuAvailable
integer(kind=ik) :: global_row, global_col, local_row, local_col
integer(kind=ik) :: bandwidth
class(elpa_t), pointer :: e
#define COMPLEXCASE
#define DOUBLE_PRECISION_COMPLEX 1
gpuAvailable = .false.
call read_input_parameters_traditional(na, nev, nblk, write_to_file)
!-------------------------------------------------------------------------------
! MPI Initialization
call setup_mpi(myid, nprocs)
gpuAvailable = check_for_gpu(myid, numberOfDevices)
STATUS = 0
#include "elpa_print_headers.X90"
#ifdef HAVE_DETAILED_TIMINGS
! initialise the timing functionality
#ifdef HAVE_LIBPAPI
call timer%measure_flops(.true.)
#endif
call timer%measure_allocated_memory(.true.)
call timer%measure_virtual_memory(.true.)
call timer%measure_max_allocated_memory(.true.)
call timer%set_print_options(&
#ifdef HAVE_LIBPAPI
print_flop_count=.true., &
print_flop_rate=.true., &
#endif
print_allocated_memory = .true. , &
print_virtual_memory=.true., &
print_max_allocated_memory=.true.)
call timer%enable()
call timer%start("program: test_complex2_single_banded")
#endif
!-------------------------------------------------------------------------------
! 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
enddo
! 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 *
endif
!-------------------------------------------------------------------------------
! 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
! 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
#ifdef HAVE_DETAILED_TIMINGS
call timer%start("set up matrix")
#endif
allocate(a (na_rows,na_cols))
allocate(z (na_rows,na_cols))
allocate(as(na_rows,na_cols))
allocate(ev(na))
call prepare_matrix(na, myid, sc_desc, 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