Commit f5e3490a authored by Andreas Marek's avatar Andreas Marek

Test program for elpa_mult_at_b_real

A random matrix is set up, and then result = a**T * 2*b is computed
this is compared with the result of plain scalapack routines and
the maximum norm of "scalapack - elpa" is computed (should be zero on
machine precision)
parent 52a44217
......@@ -196,6 +196,7 @@ dist_files_DATA = \
test/Fortran/test_real.F90 \
test/Fortran/test_real_with_c.F90 \
test/Fortran/test_toeplitz.F90 \
test/Fortran/test_transpose_multiply_real.F90 \
dist_doc_DATA = LICENSE Changelog COPYING/COPYING COPYING/gpl.txt COPYING/lgpl.txt
......@@ -219,6 +220,7 @@ noinst_PROGRAMS = \
elpa2_test_real_api@SUFFIX@ \
elpa2_test_complex_api@SUFFIX@ \
elpa1_real_toeplitz@SUFFIX@ \
elpa1_real_transpose_multiply@SUFFIX@ \
elpa1_test_real_with_c@SUFFIX@ \
elpa1_test_real_c_version@SUFFIX@ \
elpa1_test_complex_c_version@SUFFIX@ \
......@@ -278,6 +280,12 @@ elpa1_real_toeplitz@SUFFIX@_LDADD = $(build_lib)
elpa1_real_toeplitz@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
EXTRA_elpa1_real_toeplitz@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa1_real_transpose_multiply@SUFFIX@_SOURCES = test/Fortran/test_transpose_multiply_real.F90
elpa1_real_transpose_multiply@SUFFIX@_LDADD = $(build_lib)
elpa1_real_transpose_multiply@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
EXTRA_elpa1_real_transpose_multiply@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
elpa1_test_real_with_c@SUFFIX@_SOURCES = test/Fortran/test_real_with_c.F90
elpa1_test_real_with_c@SUFFIX@_LDADD = $(build_lib)
elpa1_test_real_with_c@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
......@@ -341,6 +349,7 @@ check_SCRIPTS = \ \ \ \ \
elpa2_print_kernels@SUFFIX@ \ \ \
! 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"
program test_transpose_multiply
use precision
use ELPA1
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
integer, external :: numroc
real(kind=rk), allocatable :: a(:,:), b(:,:), c(:,:), z(:,:), tmp1(:,:), tmp2(:,:), as(:,:), ev(:)
real(kind=rk) :: norm, normmax, pdlange
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, &
type(output_t) :: write_to_file
logical :: success
character(len=8) :: task_suffix
integer(kind=ik) :: j
success = .true.
call read_input_parameters(na, nev, nblk, write_to_file)
! MPI Initialization
call setup_mpi(myid, nprocs)
! 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")
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 '(3(a,i0))','Matrix size=',na,', 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 get_elpa_communicators.
mpierr = get_elpa_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
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(b (na_rows,na_cols))
allocate(c (na_rows,na_cols))
allocate(z (na_rows,na_cols))
call prepare_matrix(na, myid, sc_desc, iseed, a, z, as)
b(:,:) = 2.0 * a(:,:)
c(:,:) = 0.0
call timer%stop("set up matrix")
! Calculate eigenvalues/eigenvectors
if (myid==0) then
print '(a)','| Compute c= a**T * b ... '
print *
end if
#ifdef WITH_MPI
call mpi_barrier(mpi_comm_world, mpierr) ! for correct timings only
success = elpa_mult_at_b_real("F","F", na, na, a, na_rows, b, na_rows, nblk, &
mpi_comm_rows, mpi_comm_cols, c, na_rows)
if (.not.(success)) then
write(error_unit,*) "solve_evp_real_1stage produced an error! Aborting..."
#ifdef WITH_MPI
call MPI_ABORT(mpi_comm_world, 1, mpierr)
if (myid==0) then
print '(a)','| Solve c = a**T * b complete.'
print *
end if
! Test correctness of result (using plain scalapack routines)
tmp1(:,:) = 0.0_rk
! tmp1 = a**T
call pdtran(na, na, 1.0_rk, a, 1, 1, sc_desc, 0.0_rk, tmp1, 1, 1, sc_desc)
! tmp2 = tmp1 * b
call pdgemm("N","N", na, na, na, 1.0_rk, tmp1, 1, 1, sc_desc, b, 1, 1, &
sc_desc, 0.0_rk, tmp2, 1, 1, sc_desc)
! compare tmp2 with c
tmp2(:,:) = tmp2(:,:) - c(:,:)
norm = pdlange("M",na, na, tmp2, 1, 1, sc_desc, tmp1)
call mpi_allreduce(norm,normmax,1,MPI_REAL8,MPI_MAX,MPI_COMM_WORLD,mpierr)
if (myid .eq. 0) then
print *," Maximum error of result: ", normmax
if (normmax .gt. 5e-12) then
status = 1
call timer%stop("program")
print *," "
print *,"Timings program:"
print *," "
call timer%print("program")
print *," "
print *,"End timings program"
print *," "
#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