Merge remote-tracking branch 'origin/development' into stress-ramp-loadcase

This commit is contained in:
Martin Diehl 2020-09-19 08:46:38 +02:00
commit c6c34ccf7e
119 changed files with 26970 additions and 2164 deletions

View File

@ -141,13 +141,6 @@ Pre_General:
- release
###################################################################################################
Post_ASCIItable:
stage: postprocessing
script: ASCIItable/test.py
except:
- master
- release
Post_General:
stage: postprocessing
script: PostProcessing/test.py

@ -1 +1 @@
Subproject commit 92ca3e83b6093c1af277cfc06a504e4bb09fe8bc
Subproject commit 555f3e01f2b5cf43ade1bd48423b890adca21771

View File

@ -1 +1 @@
v3.0.0-alpha-147-gf0806a9e
v3.0.0-alpha-233-g190f8a82

View File

@ -4,7 +4,7 @@
# Normal exit status is 0.
#
DIR=/nethome/f.roters/temp/msc/marc2018.1
DIR=/tmp/msc/marc2018.1
if test $MARCDIR1
then
DIR=$MARCDIR1

View File

@ -4,7 +4,7 @@
# Normal exit status is 0.
#
DIR=/nethome/storage/raid3/f.roters/Software/MSC/Marc2018-RH7.1/marc2018
DIR=/tmp/msc/marc2018
if test $MARCDIR1
then
DIR=$MARCDIR1

View File

@ -4,7 +4,7 @@
# Normal exit status is 0.
#
DIR=/nethome/f.roters/temp/msc/marc2019.1
DIR=/tmp/msc/marc2019.1
if test $MARCDIR1
then
DIR=$MARCDIR1

View File

@ -4,7 +4,7 @@
# Normal exit status is 0.
#
DIR=/nethome/f.roters/temp/msc/marc2019
DIR=/tmp/msc/marc2019
if test $MARCDIR1
then
DIR=$MARCDIR1

View File

@ -0,0 +1,53 @@
#!/bin/ksh
# 1st arg: $DIR
# 2nd arg: $DIRJOB
# 3rd arg: $user
# 4th arg: $program
DIR=$1
user=$3
program=$4
usernoext=$user
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .f`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .F`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .for`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .f90`
# add BLAS options for linking
BLAS="%BLAS%"
. $DIR/tools/include
DIRJOB=$2
cd $DIRJOB
echo "Compiling and linking user subroutine $user on host `hostname`"
echo "program: $program"
$DFORTHIGHMP $user || \
{
echo "$0: compile failed for $user"
exit 1
}
/bin/rm $program 2>/dev/null
userobj=$usernoext.o
$LOAD ${program} $DIR/lib/main.o\
$DIR/lib/blkdta.o $DIR/lib/comm?.o \
${userobj-} \
$DIR/lib/srclib.a \
$MNFLIBS \
$MDUSER \
../lib/mdsrc.a \
../lib/mcvfit.a \
$STUBS \
${SOLVERLIBS} \
$TKLIBS \
$MRCLIBS \
$METISLIBS \
$BLAS \
$SYSLIBS || \
{
echo "$0: link failed for $usernoext.o on host `hostname`"
exit 1
}
/bin/rm $userobj
/bin/rm $DIRJOB/*.mod
/bin/rm $DIRJOB/*.smod

View File

@ -0,0 +1,53 @@
#!/bin/ksh
# 1st arg: $DIR
# 2nd arg: $DIRJOB
# 3rd arg: $user
# 4th arg: $program
DIR=$1
user=$3
program=$4
usernoext=$user
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .f`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .F`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .for`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .f90`
# add BLAS options for linking
BLAS="%BLAS%"
. $DIR/tools/include
DIRJOB=$2
cd $DIRJOB
echo "Compiling and linking user subroutine $user on host `hostname`"
echo "program: $program"
$DFORTRANLOWMP $user || \
{
echo "$0: compile failed for $user"
exit 1
}
/bin/rm $program 2>/dev/null
userobj=$usernoext.o
$LOAD ${program} $DIR/lib/main.o\
$DIR/lib/blkdta.o $DIR/lib/comm?.o \
${userobj-} \
$DIR/lib/srclib.a \
$MNFLIBS \
$MDUSER \
../lib/mdsrc.a \
../lib/mcvfit.a \
$STUBS \
${SOLVERLIBS} \
$TKLIBS \
$MRCLIBS \
$METISLIBS \
$BLAS \
$SYSLIBS || \
{
echo "$0: link failed for $usernoext.o on host `hostname`"
exit 1
}
/bin/rm $userobj
/bin/rm $DIRJOB/*.mod
/bin/rm $DIRJOB/*.smod

View File

@ -0,0 +1,53 @@
#!/bin/ksh
# 1st arg: $DIR
# 2nd arg: $DIRJOB
# 3rd arg: $user
# 4th arg: $program
DIR=$1
user=$3
program=$4
usernoext=$user
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .f`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .F`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .for`
usernoext=`dirname $usernoext`/`$BASENAME $usernoext .f90`
# add BLAS options for linking
BLAS="%BLAS%"
. $DIR/tools/include
DIRJOB=$2
cd $DIRJOB
echo "Compiling and linking user subroutine $user on host `hostname`"
echo "program: $program"
$DFORTRANMP $user || \
{
echo "$0: compile failed for $user"
exit 1
}
/bin/rm $program 2>/dev/null
userobj=$usernoext.o
$LOAD ${program} $DIR/lib/main.o\
$DIR/lib/blkdta.o $DIR/lib/comm?.o \
${userobj-} \
$DIR/lib/srclib.a \
$MNFLIBS \
$MDUSER \
../lib/mdsrc.a \
../lib/mcvfit.a \
$STUBS \
${SOLVERLIBS} \
$TKLIBS \
$MRCLIBS \
$METISLIBS \
$BLAS \
$SYSLIBS || \
{
echo "$0: link failed for $usernoext.o on host `hostname`"
exit 1
}
/bin/rm $userobj
/bin/rm $DIRJOB/*.mod
/bin/rm $DIRJOB/*.smod

View File

@ -0,0 +1,41 @@
#!/bin/ksh
# 1st arg: $DIR
# 2nd arg: $DIRJOB
# 3rd arg: $user
# 4th arg: $program
DIR=$1
user=$3
program=$4
. $DIR/tools/include
DIRJOB=$2
cd $DIRJOB
echo "Compiling and linking user subroutine $user.f on host `hostname`"
echo "program: $program"
$FORTRAN $user.f || \
{
echo "$0: compile failed for $user.f"
exit 1
}
/bin/rm $program 2>/dev/null
userobj=$user.o
$LOAD ${program} $DIR/lib/main.o\
$DIR/lib/blkdta.o $DIR/lib/comm?.o \
${userobj-} \
$DIR/lib/srclib.a \
$MNFLIBS \
$MDUSER \
../lib/mdsrc.a \
../lib/mcvfit.a \
$STUBS \
${SOLVERLIBS} \
$TKLIBS \
$MRCLIBS \
$METISLIBS \
$SYSLIBS || \
{
echo "$0: link failed for $user.o on host `hostname`"
exit 1
}
/bin/rm $userobj

View File

@ -0,0 +1,814 @@
#
# General definitions for the Marc 2020 version
#
# EM64T
#
# Linux RedHat 7.3 / SuSE 12 SP1
#
# 64 bit MPI version
#
# Intel(R) Fortran Intel(R) 64 Compiler XE for applications
# running on Intel(R) 64, Version 19.0.4.243 Build 20190416
#
# Intel(R) C Intel(R) 64 Compiler XE for applications
# running on Intel(R) 64, Version 19.0.4.243 Build 20190416
#
# To check the O/S level, type:
# uname -a
#
# Distributed parallel MPI libraries:
# Intel MPI 2019 Update 4
# To check the mpi version, type:
# mpiexec.hydra -version
#
# To check the Compiler level, type using the compiler
# installation path:
# ifort -V
# icc -V
#
# REMARKS : This file contains the definitions of variables used during
# compilation loading and use of the MARC programmes . The
# current machine type is identified by means of the variable
# MACHINE , defined below.
#
#
# MPI_ROOT: root directory in which mpi shared libraries, etc. are located
# DIRJOB : directory in which spawned jobs should look for Marc input
# MPI_ARCH: system architecture
# MPI_EPATH: path where executable resides
#
REVISION="VERSION, BUILD"
HOSTNAME=`hostname`
# find available memory in Mbyte on the machine
# can be set explicitly
MEMLIMIT=`free -m | awk '/Mem:/ {print $2}'`
# set _OEM_NASTRAN to 1 for MD Nastran build
# override _OEM_NASTRAN setting with MARC_MD_NASTRAN environment variable
_OEM_NASTRAN="${MARC_MD_NASTRAN:-0}"
# uncomment the following line for an autoforge build
#AUTOFORGE=1
AUTOFORGE=0
export AUTOFORGE
# integer size
if test "$MARC_INTEGER_SIZE" = "" ; then
INTEGER_PATH=
else
INTEGER_PATH=/$MARC_INTEGER_SIZE
fi
FCOMP=ifort
INTELPATH="/opt/intel/compilers_and_libraries_2019/linux"
# find the root directory of the compiler installation:
# - if ifort is found in $PATH, then the root directory is derived
# from the path to ifort
# - if ifort is not found in $PATH, the root directory is assumed
# to be $INTELPATH and the directory in which ifort is found is
# added to $PATH
FCOMPPATH=`which "$FCOMP" 2>/dev/null`
if test -n "$FCOMPPATH"; then
# derive the root directory from $FCOMPPATH
FCOMPROOT="${FCOMPPATH%/bin/intel64/$FCOMP}"
if test "$FCOMPROOT" = "$FCOMPPATH"; then
FCOMPROOT="${FCOMPPATH%/bin/$FCOMP}"
fi
if test "$FCOMPROOT" = "$FCOMPPATH"; then
FCOMPROOT=
fi
elif test -d "$INTELPATH"; then
# check for compiler in $INTELPATH
if test -d "$INTELPATH/bin/intel64" -a \
-x "$INTELPATH/bin/intel64/$FCOMP" ; then
FCOMPROOT="$INTELPATH"
PATH="$INTELPATH/bin/intel64:$PATH"
elif test -d "$INTELPATH/bin" -a \
-x "$INTELPATH/bin/$FCOMP"; then
FCOMPROOT="$INTELPATH"
PATH="$INTELPATH/bin:$PATH"
else
FCOMPROOT=
fi
else
FCOMPROOT=
fi
# DAMASK uses the HDF5 compiler wrapper around the Intel compiler
H5FC="$(h5fc -shlib -show)"
HDF5_LIB=${H5FC//ifort/}
FCOMP="$H5FC -DDAMASK_HDF5"
# AEM
if test "$MARCDLLOUTDIR" = ""; then
DLLOUTDIR="$MARC_LIB"
else
DLLOUTDIR="$MARCDLLOUTDIR"
fi
# settings for MKL
if test "$IMKLDIR" = ""; then
MARC_MKL="$FCOMPROOT/mkl/lib/intel64"
else
MARC_MKL=$IMKLDIR/lib/intel64
fi
#
# settings for Metis
#
METIS="-I$METIS_SOURCE/include"
METISLIBS="$METISLIB_DIR/libmarcddm.a $METISLIB_DIR/libmarcmetis.a "
#
# settings for MPI
#
# RCP and RSH are used for parallel network runs
# replace with similar commands like rsh if needed
RCP=/usr/bin/scp
RSH=/usr/bin/ssh
#
MPI_DEFAULT=intelmpi
MPI_OTHER=
MPITYPE=$MPI_DEFAULT
if test $AUTOFORGE
then
if test $AUTOFORGE = 1
then
MPITYPE=none
fi
fi
# overrule MPITYPE setting with environmental variable MARC_MPITYPE
if test $MARC_MPITYPE
then
MPITYPE=$MARC_MPITYPE
fi
# always set MPITYPE to none for MD Nastran
if test "$_OEM_NASTRAN" -ne 0
then
MPITYPE=none
fi
# Edit following lines to build with GPGPU version of BCS Solver for
# NVIDIA platforms
#BCSGPUSOLVER=NONE
BCSGPUSOLVER=BCSGPU
# Edit following lines to build solver 2 with GPU support
SOLVER2GPU=GPU
#SOLVER2GPU=NONE
# Edit following lines to set the openssl library
if test "$OPENSSL" != "NONE"
then
OPENSSL_LIB="$MARC_LIB/libcrypto.a"
fi
OPENSSL_INCLUDE=-I"$MARC_OPENSSL/include/"
MARCHDF_HDF=HDF
#MARCHDF_HDF=NONE
# activate contact component build if flagged
AEM_DLL=0
if test "$AEM_BUILD" = "ON" ; then
AEM_DLL=1
LINK_MARC_DLL="-shared -fPIC"
EXT_DLL="so"
MPITYPE=none
MPI_OTHER=
MARCHDF_HDF=NONE
BCSGPUSOLVER=NONE
SOLVER2GPU=NONE
MUMPSSOLVER=NONE
CASISOLVER=NONE
fi
if test "$MARCHDF_HDF" = "HDF"; then
HDF_INCLUDE="-I$MARC_HDF/include"
HDF_LIBS="$MARC_LIB/libhdf5_fortran.so.100 $MARC_LIB/libhdf5.so.103"
fi
SOLVERFLAGS=
if test "$BCSGPUSOLVER" = BCSGPU
then
SOLVERFLAGS="$SOLVERFLAGS -DBCSGPU -DCUDA"
BCS_DIR=bcsgpusolver
else
BCS_DIR=bcssolver
fi
#
# settings for MPI
#
DDM=
if test $MPITYPE != none
then
if test $MPITYPE = intelmpi
then
INTELMPI_VERSION=HYDRA
FCOMPMPI=mpiifort
if test -n "$MARC_INTELMPI" ; then
MPI_ROOT=$MARC_INTELMPI
else
MPI_ROOT=/opt/intel/compilers_and_libraries_2019/linux/mpi/intel64
MARC_INTELMPI=$MPI_ROOT
fi
DDM="-I${MPI_ROOT}/include -DDDM"
PATH=$MPI_ROOT/bin:$PATH
export PATH
LD_LIBRARY_PATH=$MPI_ROOT/lib:$MPI_ROOT/lib/release:$MPI_ROOT/libfabric/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH
if test $INTELMPI_VERSION = HYDRA
then
RUN_JOB1="${MPI_ROOT}/bin/mpiexec.hydra -genvall -n "
RUN_JOB2="${MPI_ROOT}/bin/mpiexec.hydra -genvall"
else
RUN_JOB1="${MPI_ROOT}/bin/mpiexec -n "
RUN_JOB2="${MPI_ROOT}/bin/mpiexec -configfile "
fi
RUN_JOB0=
MPI_CLEAN=
MPI_EPATH=$MARC_BIN
MPIR_HOME=$MPI_ROOT
MPICH_F77=$FCOMP
MPICH_F77LINKER=$FCOMP
export MPI_ROOT MPI_EPATH MPIR_HOME MPICH_F77 MPICH_F77LINKER
FI_PROVIDER_PATH=$MPI_ROOT/libfabric/lib/prov
export FI_PROVIDER_PATH
fi
else
MPI_ROOT=$MARC_DUMMYMPI
export MPI_ROOT=$MARC_DUMMYMPI
DDM="-I$MPI_ROOT/include"
fi
#
# variables for the "maintain" script
#
MACHINENAME=LINUX
MACHINE64BIT=yes
MACHINE=Linux_EM64T
DEV=/dev/tape
GETLOG="whoami"
CLEAR="clear"
MY_UNAME=`uname -a`
# Edit following 2 lines to build with VKI Solver
#VKISOLVER=VKI
VKISOLVER=NONE
# Edit following 2 lines to build with CASI Solver
CASISOLVER=CASI
if test "$MARC_CASISOLVER" = "NONE" ; then
CASISOLVER=NONE
fi
#CASISOLVER=NONE
# Edit following 2 lines to build with MF2 Solver
MF2SOLVER=NONE
#MF2SOLVER=SERIAL
#MF2SOLVER=MF2PARALLEL
# Edit following lines to build with Intel(c) Multithreaded solver (PARDISO)
#INTELSOLVER=NONE
INTELSOLVER=PARDISO
# Edit following lines to build with MUMPS
if test "$MARC_INTEGER_SIZE" = "i4" ; then
#MUMPSSOLVER=NONE
MUMPSSOLVER=MUMPS
else
#MUMPSSOLVER=NONE
MUMPSSOLVER=MUMPS
fi
# Edit following 2 lines to build MARC dynamic shared library
MARC_DLL=MARC_DLL
MARC_DLL=NONE
# always set VKISOLVER, CASISOLVER, BCSGPUSOLVER, and MARC_DLL to NONE for MD Nastran
if test "$_OEM_NASTRAN" -ne 0
then
VKISOLVER=NONE
CASISOLVER=NONE
MF2SOLVER=NONE
INTELSOLVER=NONE
MUMPSSOLVER=NONE
BCSGPUSOLVER=NONE
SOLVER2GPU=NONE
MARC_DLL=NONE
fi
if test "$AEM_DLL" -eq 1
then
VKISOLVER=NONE
CASISOLVER=NONE
MF2SOLVER=NONE
INTELSOLVER=NONE
MUMPSSOLVER=NONE
BCSGPUSOLVER=NONE
SOLVER2GPU=NONE
fi
#
# define Fortran and C compile syntax
#
if test "$VKISOLVER" = VKI
then
SOLVERFLAGS="$SOLVERFLAGS -DVKI"
fi
if test "$CASISOLVER" = CASI
then
SOLVERFLAGS="$SOLVERFLAGS -DCASI"
fi
if test "$MF2SOLVER" = MF2PARALLEL
then
SOLVERFLAGS="$SOLVERFLAGS -DMF2PARALLEL"
fi
if test "$MF2SOLVER" = MF2SERIAL
then
SOLVERFLAGS="$SOLVERFLAGS -DMF2SERIAL"
fi
if test "$INTELSOLVER" = PARDISO
then
SOLVERFLAGS="$SOLVERFLAGS -DPARDISO"
fi
if test "$MUMPSSOLVER" = MUMPS
then
SOLVERFLAGS="$SOLVERFLAGS -DMUMPS"
fi
if test "$MARC_DLL" = MARC_DLL
then
SOLVERFLAGS="$SOLVERFLAGS -DMARC_DLL"
fi
if test "$SOLVER2GPU" = GPU
then
SOLVERFLAGS="$SOLVERFLAGS -DSOLVER2GPU"
fi
LINK_OPT=
DEBUG_OPT=
C_DEBUG_OPT=
#Uncomment following line to build Marc in debuggable mode
MARCDEBUG=
#MARCDEBUG="ON"
if test "$MARCDEBUG" = "ON"
then
LINK_OPT="-debug -traceback"
DEBUG_OPT="-debug -traceback"
C_DEBUG_OPT="-debug -traceback"
fi
MARCCHECK=
#MARCCHECK="ON"
if test "$MARCCHECK" = "ON"
then
DEBUG_OPT="$DEBUG_OPT -fpe0 -fp-stack-check -check all -ftrapuv "
C_DEBUG_OPT="$C_DEBUG_OPT -fp-stack-check -check-uninit -Wformat -ftrapuv "
fi
MARCCODECOV=
#MARCCODECOV="ON"
MARCCODEPROF=
#MARCCODEPROF="ON"
if test "$MARC_INTEGER_SIZE" = "i4" ; then
I8FFLAGS=
I8DEFINES=
I8CDEFINES=
else
I8FFLAGS="-i8 -integer-size 64"
I8DEFINES="-DI64 -DINT=8"
I8CDEFINES="-U_DOUBLE -D_SINGLE"
fi
MTHREAD=OPENMP
if test "$MARC_OPENMP" = "NONE" ; then
MTHREAD=NONE
fi
#MTHREAD=NONE
if test "$_OEM_NASTRAN" -ne 0
then
MTHREAD=NONE
fi
if test "$AEM_DLL" -eq 1
then
MTHREAD=NONE
CASISOLVER=NONE
VKISOLVER=NONE
MF2SOLVER=NONE
INTELSOLVER=NONE
BCSGPUSOLVER=NONE
SOLVER2GPU=NONE
OPENSSL_LIB=
MARC_DLL=NONE
METISLIBS=
fi
OMP_COMPAT=NO
OMP_COMPAT=YES
if test "$MTHREAD" = "NONE"
then
OMP_COMPAT=NO
fi
CDEFINES=
FDEFINES=
if test "$_OEM_NASTRAN" -ne 0
then
CDEFINES="$CDEFINES -D_OEM_NASTRAN"
FDEFINES="$FDEFINES -D_OEM_NASTRAN"
fi
FDEFINES="$FDEFINES -D_IMPLICITNONE"
if test "$_OEM_NASTRAN" -eq 0
then
FDEFINES="$FDEFINES -DMKL -DOPENMP"
fi
if test "$OMP_COMPAT" = "YES"
then
FDEFINES="$FDEFINES -DOMP_COMPAT"
fi
# -D_MSCMARC
FDEFINES="$FDEFINES -D_MSCMARC $DEBUG_OPT $MARC_SIMUFACT"
CDEFINES="$CDEFINES -D_MSCMARC $C_DEBUG_OPT $I8CDEFINES"
if test "$AEM_DLL" -eq 1
then
FDEFINES="$FDEFINES -D_AEMNL -DAAA"
CDEFINES="$CDEFINES -D_AEMNL -DAAA"
fi
CINCL="-I$MARC_SOURCE/mdsrc -I$MARC_SOURCE/csource $METIS -I$LAPI_IMPORTS/common/include"
if test "$_OEM_NASTRAN" -ne 0
then
CINCL="$CINCL -I../../include"
fi
CC_OPT=
if test "$MTHREAD" = "OPENMP"
then
CC_OPT=" $CC_OPT -qopenmp"
fi
CC="icc -c $CC_OPT -O1 $I8DEFINES -DLinux -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
CCLOW="icc -c $CC_OPT -O0 $I8DEFINES -DLinux -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
CCHIGH="icc -c $CC_OPT -O3 $I8DEFINES -DLinux -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
if test "$MARCDEBUG" = "ON"
then
CC="icc -c $CC_OPT -DLinux $I8DEFINES -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
CCLOW="icc $CC_OPT -c -DLinux $I8DEFINES -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
CCHIGH="icc $CC_OPT -c -DLinux $I8DEFINES -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
fi
LOAD_CC="icc $CC_OPT -O1 -DLinux -DLINUX -DLinux_intel"
CCT="$CC"
CCTLOW="$CCLOW"
CCTHIGH="$CCHIGH"
#PROFILE="-Mprof=func"
#PROFILE="-Mprof=lines"
#PROFILE="-Mprof=func,mpi"
PROFILE=
#PROFILE="-init=snan,arrays -CB -traceback -fpe0 -fp-stack-check -check all -check uninit -ftrapuv"
if test "$MARCCODECOV" = "ON"
then
PROFILE="-prof-gen=srcpos"
fi
if test "$MARCCODEPROF" = "ON"
then
PROFILE=" $PROFILE -pg"
fi
FORT_OPT="-c -implicitnone -stand f08 -standard-semantics -assume nostd_mod_proc_name -safe_cray_ptr -mp1 -WB -fp-model source"
if test "$MTHREAD" = "OPENMP"
then
FORT_OPT=" $FORT_OPT -qopenmp"
if test "$OMP_COMPAT" = "YES"
then
FORT_OPT=" $FORT_OPT -qopenmp-threadprivate=compat"
fi
else
# FORT_OPT=" $FORT_OPT -auto "
FORT_OPT=" $FORT_OPT -save -zero"
fi
if test "$MARCHDF_HDF" = "HDF"; then
FORT_OPT="$FORT_OPT -DMARCHDF=$MARCHDF_HDF"
fi
FORTLOW="$FCOMP $FORT_OPT $PROFILE -O0 $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTRAN="$FCOMP $FORT_OPT $PROFILE -O1 $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTHIGH="$FCOMP $FORT_OPT $PROFILE -fno-alias -O3 $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTNA="$FCOMP $FORT_OPT -fno-alias -O3 $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM"
# for compiling free form f90 files. high opt, integer(4)
FORTF90="$FCOMP -c -O3"
# determine DAMASK version
if test -n "$DAMASK_USER"; then
DAMASKROOT=`dirname $DAMASK_USER`/..
read DAMASKVERSION < $DAMASKROOT/VERSION
DAMASKVERSION="'"$DAMASKVERSION"'"
else
DAMASKVERSION="'N/A'"
fi
# DAMASK compiler calls: additional flags are in line 2 OpenMP flags in line 3
DFORTLOWMP="$FCOMP -c -O0 -qno-offload -implicitnone -stand f08 -standard-semantics -assume nostd_mod_proc_name -safe_cray_ptr $PROFILE -zero -mp1 -WB $I8FFLAGS -I$MARC_SOURCE/common \
-fpp -ftz -diag-disable 5268 -warn declarations -warn general -warn usage -warn interfaces -warn ignore_loc -warn alignments -DMarc4DAMASK=2019 -DDAMASKVERSION=$DAMASKVERSION \
-qopenmp -qopenmp-threadprivate=compat\
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
DFORTRANMP="$FCOMP -c -O1 -qno-offload -implicitnone -stand f08 -standard-semantics -assume nostd_mod_proc_name -safe_cray_ptr $PROFILE -zero -mp1 -WB $I8FFLAGS -I$MARC_SOURCE/common \
-fpp -ftz -diag-disable 5268 -warn declarations -warn general -warn usage -warn interfaces -warn ignore_loc -warn alignments -DMarc4DAMASK=2019 -DDAMASKVERSION=$DAMASKVERSION \
-qopenmp -qopenmp-threadprivate=compat\
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
DFORTHIGHMP="$FCOMP -c -O3 -qno-offload -implicitnone -stand f08 -standard-semantics -assume nostd_mod_proc_name -safe_cray_ptr $PROFILE -zero -mp1 -WB $I8FFLAGS -I$MARC_SOURCE/common \
-fpp -ftz -diag-disable 5268 -warn declarations -warn general -warn usage -warn interfaces -warn ignore_loc -warn alignments -DMarc4DAMASK=2019 -DDAMASKVERSION=$DAMASKVERSION \
-qopenmp -qopenmp-threadprivate=compat\
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
if test "$MARCDEBUG" = "ON"
then
FORTLOW="$FCOMP $FORT_OPT $PROFILE $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTRAN="$FCOMP $FORT_OPT $PROFILE $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTHIGH="$FCOMP $FORT_OPT $PROFILE -fno-alias $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTNA="$FCOMP $FORT_OPT -fno-alias $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM"
fi
FORTLOWT="$FORTLOW"
FORTRANT="$FORTRAN"
FORTHIGHT="$FORTHIGH"
FORTRANMNF="$FCOMP -c $FDEFINES "
CCMNF="icc -c -O1 -DLinux -DLINUX -DLinux_intel -Dport2egcs -I$MARC_SOURCE/marctoadams/mnf/include -D_LARGEFILE64_SOURCE"
if test $MPITYPE != none
then
if test $MPITYPE = hpmpi
then
LOAD="$MPI_ROOT/bin/$FCOMPMPI ${LOADOPTIONS} -L$MPI_ROOT/lib/$ARCHITECTURE $PROFILE $LINK_OPT -o "
LOADT="$MPI_ROOT/bin/$FCOMPMPI ${LOADOPTIONS} -L$MPI_ROOT/lib/$ARCHITECTURE $PROFILE $LINK_OPT -o "
fi
# Uncomment the following lines to turn on the tracer and commnet out the next 5 lines
# if test $MPITYPE = intelmpi
# then
# INCLUDEMPI="-I$MPI_ROOT/include -I$VT_ROOT/include"
# LOAD="$MPI_ROOT/bin/$FCOMPMPI $PROFILE $INCLUDEMPI -g -t=log $LINK_OPT -o "
# LOADT="$MPI_ROOT/bin/$FCOMPMPI $PROFILE $INCLUDEMPI -g -t=log $LINK_OPT -o "
# fi
if test $MPITYPE = intelmpi
then
LOAD="ifort $PROFILE $LINK_OPT -o "
LOADT="ifort $PROFILE $LINK_OPT -o "
fi
else
LOAD="$FCOMP $LINK_OPT -o "
LOADT="$FCOMP $LINK_OPT -o "
fi
if test "$MARC_DLL" = MARC_DLL
then
FORTLOW="$FORTLOW -fpp -fPIC"
FORTRAN="$FORTRAN -fpp -fPIC"
FORTHIGH="$FORTHIGH -fpp -fPIC"
FORTRANMNF="$FORTRANMNF -fpp -fPIC"
CC="$CC -fPIC"
CCMNF="$CCMNF -fPIC"
LINK_EXE_MARC="-L$MARC_LIB -lmarc -L$MARC_LIB_SHARED -lguide -lpthread"
LINK_MARC_DLL="-shared -fPIC"
LOAD_DLL=$LOAD
LOADT_DLL=$LOADT
EXT_DLL="so"
fi
if test "$AEM_DLL" -eq 1
then
FORTLOW="$FORTLOW -fpp -fPIC"
FORTRAN="$FORTRAN -fpp -fPIC"
FORTHIGH="$FORTHIGH -fpp -fPIC"
FORTRANMNF="$FORTRANMNF -fpp -fPIC"
CC="$CC -fPIC"
CCMNF="$CCMNF -fPIC"
LINK_EXE_MARC="-L$MARC_LIB -lmarc -L$MARC_LIB_SHARED -lguide"
LINK_MARC_DLL="-shared -fPIC"
LOAD_DLL=$LOAD
LOADT_DLL=$LOADT
EXT_DLL="so"
fi
XLIBS="-L/usr/X11/lib -lX11 "
#
# define archive and ranlib syntax
#
ARC="ar rvl"
ARD="ar dvl"
ARX="ar xl"
RAN=""
#
# choose which libraries you want to use ( e.g. blas )
#
if test "$VKISOLVER" = VKI
then
VKISOLVERLIBS="$MARC_LIB/vkisolver.a"
else
VKISOLVERLIBS=
fi
if test "$CASISOLVER" = CASI
then
CASISOLVERLIBS="$CASILIB_DIR/libmarccasi.a $CASILIB_DIR/libcasi.a"
else
CASISOLVERLIBS=
fi
MF2SOLVERLIBS=
if test "$MF2SOLVER" = MF2PARALLEL
then
MF2SOLVERLIBS="$MARC_LIB/mf2parallel/libseq.a \
$MARC_LIB/mf2parallel/libsym.a \
$MARC_LIB/mf2parallel/libmet.a \
$MARC_LIB/mf2parallel/libmf2.a \
$MARC_LIB/mf2parallel/libgauss.a \
$MARC_LIB/mf2parallel/libmf2.a \
$MARC_LIB/mf2parallel/libgauss.a \
$MARC_LIB/mf2parallel/libnum.a \
$MARC_LIB/mf2parallel/libutl.a \
$MARC_LIB/mf2parallel/libr8.a \
$MARC_LIB/mf2parallel/libz.a "
fi
if test "$MUMPSSOLVER" = MUMPS
then
MUMPSSOLVERLIBS="$MUMPSLIB_DIR/libmumps.a"
if test $MPITYPE = none
then
MUMPSSOLVERLIBS2=
echo hello > /dev/null
fi
if test $MPITYPE = intelmpi
then
MUMPSSOLVERLIBS="$MUMPSLIB_DIR/libmumps_intelmpi.a"
if test "$MARC_INTEGER_SIZE" = "i4" ; then
MUMPSSOLVERLIBS2=" $MARC_MKL/libmkl_blacs_intelmpi_lp64.a "
else
MUMPSSOLVERLIBS2=" $MARC_MKL/libmkl_blacs_intelmpi_ilp64.a "
fi
fi
if test $MPITYPE = hpmpi
then
MUMPSSOLVERLIBS="$MUMPSLIB_DIR/libmumps_hpmpi.a"
if test "$MARC_INTEGER_SIZE" = "i4" ; then
MUMPSSOLVERLIBS2=" $MARC_MKL/libmkl_blacs_intelmpi_lp64.a"
else
MUMPSSOLVERLIBS2=" $MARC_MKL/libmkl_blacs_intelmpi_ilp64.a"
fi
fi
else
MUMPSSOLVERLIBS=
MUMPSSOLVERLIBS2=
fi
if test "$BCSGPUSOLVER" = BCSGPU
then
BCSSOLVERLIBS="${MARC_LIB}/bcsgpulib.a "
MARCCUDALIBS1="-L${MARC_LIB}/cuda_dummy -lmarccuda "
MARCCUDALIBS2="-L${MARC_LIB}/cuda -lmarccuda "
MARCCUDALIBS=$MARCCUDALIBS1
else
BCSSOLVERLIBS="${MARC_LIB}/bcslib.a "
fi
SOLVER2LIBS=
if test "$SOLVER2GPU" = GPU
then
SOLVER2LIBS="-L$MARC_LIB/cuda_dummy -lsolver2gpu"
fi
if test "$AEM_DLL" -eq 1
then
BCSSOLVERLIBS=
fi
if test "$MARC_INTEGER_SIZE" = "i4" ; then
MKLLIB="$MARC_MKL/libmkl_scalapack_lp64.a -Wl,--start-group $MARC_MKL/libmkl_intel_lp64.a $MARC_MKL/libmkl_intel_thread.a $MARC_MKL/libmkl_core.a $MARC_MKL/libmkl_blacs_intelmpi_lp64.a $MUMPSSOLVERLIBS2 -Wl,--end-group"
else
MKLLIB="$MARC_MKL/libmkl_scalapack_ilp64.a -Wl,--start-group $MARC_MKL/libmkl_intel_ilp64.a $MARC_MKL/libmkl_intel_thread.a $MARC_MKL/libmkl_core.a $MARC_MKL/libmkl_blacs_intelmpi_ilp64.a $MUMPSSOLVERLIBS2 -Wl,--end-group"
fi
SECLIBS="-L$MARC_LIB -llapi"
SOLVERLIBS="${BCSSOLVERLIBS} ${VKISOLVERLIBS} ${CASISOLVERLIBS} ${MF2SOLVERLIBS} \
$MKLLIB -L$MARC_MKL -liomp5 \
$MARC_LIB/blas_src.a ${ACSI_LIB}/ACSI_MarcLib.a $KDTREE2_LIB/libkdtree2.a $HDF5_LIB $SOLVER2LIBS"
SOLVERLIBS_DLL=${SOLVERLIBS}
if test "$AEM_DLL" -eq 1
then
SOLVERLIBS_DLL="$MKLLIB -L$MARC_MKL -liomp5 $MARC_LIB/blas_src.a"
fi
MRCLIBS="$MARC_LIB/clib.a ${CASISOLVERLIBS}"
MRCLIBSPAR="$MARC_LIB/clib.a"
STUBS="$MARC_LIB/stubs.a "
MNFLIBS="$MARC_LIB/libmnf.a"
MDUSER="$MARC_LIB/md_user.a"
if test "X$MARC_SIMUFACT" != "X"
then
SFLIB="-L$SFMATDIR -lMBA_Grain"
else
SFLIB=" "
fi
OPENMP="-qopenmp"
if test "$AEM_DLL" -eq 1
then
LOAD_DLL=$LOAD
OPENMP=
LIBMNF=
OPENSSL=NONE
fi
SYSLIBS=" $OPENMP -lpthread -cxxlib"
# Uncomment the following lines to turn on the trace and comment out the next 4 lines
# if test $MPITYPE = intelmpi
# then
# SYSLIBS="-L${VT_ROOT}/lib -lVT -ldwarf -lelf -lm -lpthread \
# -L${MPI_ROOT}/lib64 -lmpi -lmpiif -lmpigi -lrt"
# fi
if test $MPITYPE = intelmpi
then
SYSLIBS="-L${MPI_ROOT}/lib/release -lmpi -L${MPI_ROOT}/lib -lmpifort -lrt $OPENMP -threads -lpthread -cxxlib"
fi
SYSLIBSPAR=" "
MARC_DLL_CODES="runmarc.f"
BLAS_SRC="dzero.f icopy.f izero.f"
if test "$_OEM_NASTRAN" -ne 0
then
if test "$MARC_INTEGER_SIZE" = "i4" ; then
BLAS_SRC="$BLAS_SRC dsctr.f zsctr.f dzasum.f daxpyi.f zaxpyi.f dgthr.f zgthr.f"
else
BLAS_SRC="ALL"
fi
fi
LOW_OPT_CODES="are163.f contro.f ndext.f omarc.f omarca.f omarcb.f omarcc.f \
omars.f fixbc.f triang.f bet049.f norst3.f eldata.f \
elec*.f elct*.f fmeig.f oada00.f ogeig.f updtrbe2.f cycrota.f \
cordef.f ogpk.f ogtan.f eldam.f formrbe3.f \
inertie.f em_sso072.f cn_fol3d_qpatch6.f cosim_begin.f"
if test "$MARC_INTEGER_SIZE" = "i8" ; then
LOW_OPT_CODES="$LOW_OPT_CODES bbcseg.f"
fi
HIGH_OPT_CODES="dpsmsa1.f dpsmsa2.f dpsmsa3.f dpsmsa4.f dpsmsa5.f dpsmsa6.f \
dpsmsa7.f dpsmsa8.f dpsmsa9.f dpsmsa10.f dpsmsa11.f dpsmsa12.f \
dpsmsa13.f dpsmsa14.f dpsmsa15.f dpsmsa16.f dpsmsah.f tpsmsah.f cn_qsort4_11.f \
prei11.f prei12.f prei31.f prei32.f prei41.f prei42.f prei61.f prei62.f \
prei1n.f prei2n.f cgfullnts1.f cgfullnts2.f cg1n.f cg2n.f cg3n.f \
cg4n.f cg5n.f cg6n.f cgnn.f sortab.f sortab1.f triann1.f preinv_nts.f \
cn_sur_patchl.f cn_quad_3e.f"
MAXNUM=1000000

View File

@ -0,0 +1,785 @@
#
# General definitions for the Marc 2020 version
#
# EM64T
#
# Linux RedHat 7.3 / SuSE 12 SP1
#
# 64 bit MPI version
#
# Intel(R) Fortran Intel(R) 64 Compiler XE for applications
# running on Intel(R) 64, Version 19.0.4.243 Build 20190416
#
# Intel(R) C Intel(R) 64 Compiler XE for applications
# running on Intel(R) 64, Version 19.0.4.243 Build 20190416
#
# To check the O/S level, type:
# uname -a
#
# Distributed parallel MPI libraries:
# Intel MPI 2019 Update 4
# To check the mpi version, type:
# mpiexec.hydra -version
#
# To check the Compiler level, type using the compiler
# installation path:
# ifort -V
# icc -V
#
# REMARKS : This file contains the definitions of variables used during
# compilation loading and use of the MARC programmes . The
# current machine type is identified by means of the variable
# MACHINE , defined below.
#
#
# MPI_ROOT: root directory in which mpi shared libraries, etc. are located
# DIRJOB : directory in which spawned jobs should look for Marc input
# MPI_ARCH: system architecture
# MPI_EPATH: path where executable resides
#
REVISION="VERSION, BUILD"
HOSTNAME=`hostname`
# find available memory in Mbyte on the machine
# can be set explicitly
MEMLIMIT=`free -m | awk '/Mem:/ {print $2}'`
# set _OEM_NASTRAN to 1 for MD Nastran build
# override _OEM_NASTRAN setting with MARC_MD_NASTRAN environment variable
_OEM_NASTRAN="${MARC_MD_NASTRAN:-0}"
# uncomment the following line for an autoforge build
#AUTOFORGE=1
AUTOFORGE=0
export AUTOFORGE
# integer size
if test "$MARC_INTEGER_SIZE" = "" ; then
INTEGER_PATH=
else
INTEGER_PATH=/$MARC_INTEGER_SIZE
fi
FCOMP=ifort
INTELPATH="/opt/intel/compilers_and_libraries_2019/linux"
# find the root directory of the compiler installation:
# - if ifort is found in $PATH, then the root directory is derived
# from the path to ifort
# - if ifort is not found in $PATH, the root directory is assumed
# to be $INTELPATH and the directory in which ifort is found is
# added to $PATH
FCOMPPATH=`which "$FCOMP" 2>/dev/null`
if test -n "$FCOMPPATH"; then
# derive the root directory from $FCOMPPATH
FCOMPROOT="${FCOMPPATH%/bin/intel64/$FCOMP}"
if test "$FCOMPROOT" = "$FCOMPPATH"; then
FCOMPROOT="${FCOMPPATH%/bin/$FCOMP}"
fi
if test "$FCOMPROOT" = "$FCOMPPATH"; then
FCOMPROOT=
fi
elif test -d "$INTELPATH"; then
# check for compiler in $INTELPATH
if test -d "$INTELPATH/bin/intel64" -a \
-x "$INTELPATH/bin/intel64/$FCOMP" ; then
FCOMPROOT="$INTELPATH"
PATH="$INTELPATH/bin/intel64:$PATH"
elif test -d "$INTELPATH/bin" -a \
-x "$INTELPATH/bin/$FCOMP"; then
FCOMPROOT="$INTELPATH"
PATH="$INTELPATH/bin:$PATH"
else
FCOMPROOT=
fi
else
FCOMPROOT=
fi
# AEM
if test "$MARCDLLOUTDIR" = ""; then
DLLOUTDIR="$MARC_LIB"
else
DLLOUTDIR="$MARCDLLOUTDIR"
fi
# settings for MKL
if test "$IMKLDIR" = ""; then
MARC_MKL="$FCOMPROOT/mkl/lib/intel64"
else
MARC_MKL=$IMKLDIR/lib/intel64
fi
#
# settings for Metis
#
METIS="-I$METIS_SOURCE/include"
METISLIBS="$METISLIB_DIR/libmarcddm.a $METISLIB_DIR/libmarcmetis.a "
#
# settings for MPI
#
# RCP and RSH are used for parallel network runs
# replace with similar commands like rsh if needed
RCP=/usr/bin/scp
RSH=/usr/bin/ssh
#
MPI_DEFAULT=intelmpi
MPI_OTHER=
MPITYPE=$MPI_DEFAULT
if test $AUTOFORGE
then
if test $AUTOFORGE = 1
then
MPITYPE=none
fi
fi
# overrule MPITYPE setting with environmental variable MARC_MPITYPE
if test $MARC_MPITYPE
then
MPITYPE=$MARC_MPITYPE
fi
# always set MPITYPE to none for MD Nastran
if test "$_OEM_NASTRAN" -ne 0
then
MPITYPE=none
fi
# Edit following lines to build with GPGPU version of BCS Solver for
# NVIDIA platforms
#BCSGPUSOLVER=NONE
BCSGPUSOLVER=BCSGPU
# Edit following lines to build solver 2 with GPU support
SOLVER2GPU=GPU
#SOLVER2GPU=NONE
# Edit following lines to set the openssl library
if test "$OPENSSL" != "NONE"
then
OPENSSL_LIB="$MARC_LIB/libcrypto.a"
fi
OPENSSL_INCLUDE=-I"$MARC_OPENSSL/include/"
MARCHDF_HDF=HDF
#MARCHDF_HDF=NONE
# activate contact component build if flagged
AEM_DLL=0
if test "$AEM_BUILD" = "ON" ; then
AEM_DLL=1
LINK_MARC_DLL="-shared -fPIC"
EXT_DLL="so"
MPITYPE=none
MPI_OTHER=
MARCHDF_HDF=NONE
BCSGPUSOLVER=NONE
SOLVER2GPU=NONE
MUMPSSOLVER=NONE
CASISOLVER=NONE
fi
if test "$MARCHDF_HDF" = "HDF"; then
HDF_INCLUDE="-I$MARC_HDF/include"
HDF_LIBS="$MARC_LIB/libhdf5_fortran.so.100 $MARC_LIB/libhdf5.so.103"
fi
SOLVERFLAGS=
if test "$BCSGPUSOLVER" = BCSGPU
then
SOLVERFLAGS="$SOLVERFLAGS -DBCSGPU -DCUDA"
BCS_DIR=bcsgpusolver
else
BCS_DIR=bcssolver
fi
#
# settings for MPI
#
DDM=
if test $MPITYPE != none
then
if test $MPITYPE = intelmpi
then
INTELMPI_VERSION=HYDRA
FCOMPMPI=mpiifort
if test -n "$MARC_INTELMPI" ; then
MPI_ROOT=$MARC_INTELMPI
else
MPI_ROOT=/opt/intel/compilers_and_libraries_2019/linux/mpi/intel64
MARC_INTELMPI=$MPI_ROOT
fi
DDM="-I${MPI_ROOT}/include -DDDM"
PATH=$MPI_ROOT/bin:$PATH
export PATH
LD_LIBRARY_PATH=$MPI_ROOT/lib:$MPI_ROOT/lib/release:$MPI_ROOT/libfabric/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH
if test $INTELMPI_VERSION = HYDRA
then
RUN_JOB1="${MPI_ROOT}/bin/mpiexec.hydra -genvall -n "
RUN_JOB2="${MPI_ROOT}/bin/mpiexec.hydra -genvall"
else
RUN_JOB1="${MPI_ROOT}/bin/mpiexec -n "
RUN_JOB2="${MPI_ROOT}/bin/mpiexec -configfile "
fi
RUN_JOB0=
MPI_CLEAN=
MPI_EPATH=$MARC_BIN
MPIR_HOME=$MPI_ROOT
MPICH_F77=$FCOMP
MPICH_F77LINKER=$FCOMP
export MPI_ROOT MPI_EPATH MPIR_HOME MPICH_F77 MPICH_F77LINKER
FI_PROVIDER_PATH=$MPI_ROOT/libfabric/lib/prov
export FI_PROVIDER_PATH
fi
else
MPI_ROOT=$MARC_DUMMYMPI
export MPI_ROOT=$MARC_DUMMYMPI
DDM="-I$MPI_ROOT/include"
fi
#
# variables for the "maintain" script
#
MACHINENAME=LINUX
MACHINE64BIT=yes
MACHINE=Linux_EM64T
DEV=/dev/tape
GETLOG="whoami"
CLEAR="clear"
MY_UNAME=`uname -a`
# Edit following 2 lines to build with VKI Solver
#VKISOLVER=VKI
VKISOLVER=NONE
# Edit following 2 lines to build with CASI Solver
CASISOLVER=CASI
if test "$MARC_CASISOLVER" = "NONE" ; then
CASISOLVER=NONE
fi
#CASISOLVER=NONE
# Edit following 2 lines to build with MF2 Solver
MF2SOLVER=NONE
#MF2SOLVER=SERIAL
#MF2SOLVER=MF2PARALLEL
# Edit following lines to build with Intel(c) Multithreaded solver (PARDISO)
#INTELSOLVER=NONE
INTELSOLVER=PARDISO
# Edit following lines to build with MUMPS
if test "$MARC_INTEGER_SIZE" = "i4" ; then
#MUMPSSOLVER=NONE
MUMPSSOLVER=MUMPS
else
#MUMPSSOLVER=NONE
MUMPSSOLVER=MUMPS
fi
# Edit following 2 lines to build MARC dynamic shared library
MARC_DLL=MARC_DLL
MARC_DLL=NONE
# always set VKISOLVER, CASISOLVER, BCSGPUSOLVER, and MARC_DLL to NONE for MD Nastran
if test "$_OEM_NASTRAN" -ne 0
then
VKISOLVER=NONE
CASISOLVER=NONE
MF2SOLVER=NONE
INTELSOLVER=NONE
MUMPSSOLVER=NONE
BCSGPUSOLVER=NONE
SOLVER2GPU=NONE
MARC_DLL=NONE
fi
if test "$AEM_DLL" -eq 1
then
VKISOLVER=NONE
CASISOLVER=NONE
MF2SOLVER=NONE
INTELSOLVER=NONE
MUMPSSOLVER=NONE
BCSGPUSOLVER=NONE
SOLVER2GPU=NONE
fi
#
# define Fortran and C compile syntax
#
if test "$VKISOLVER" = VKI
then
SOLVERFLAGS="$SOLVERFLAGS -DVKI"
fi
if test "$CASISOLVER" = CASI
then
SOLVERFLAGS="$SOLVERFLAGS -DCASI"
fi
if test "$MF2SOLVER" = MF2PARALLEL
then
SOLVERFLAGS="$SOLVERFLAGS -DMF2PARALLEL"
fi
if test "$MF2SOLVER" = MF2SERIAL
then
SOLVERFLAGS="$SOLVERFLAGS -DMF2SERIAL"
fi
if test "$INTELSOLVER" = PARDISO
then
SOLVERFLAGS="$SOLVERFLAGS -DPARDISO"
fi
if test "$MUMPSSOLVER" = MUMPS
then
SOLVERFLAGS="$SOLVERFLAGS -DMUMPS"
fi
if test "$MARC_DLL" = MARC_DLL
then
SOLVERFLAGS="$SOLVERFLAGS -DMARC_DLL"
fi
if test "$SOLVER2GPU" = GPU
then
SOLVERFLAGS="$SOLVERFLAGS -DSOLVER2GPU"
fi
LINK_OPT=
DEBUG_OPT=
C_DEBUG_OPT=
#Uncomment following line to build Marc in debuggable mode
MARCDEBUG=
#MARCDEBUG="ON"
if test "$MARCDEBUG" = "ON"
then
LINK_OPT="-debug -traceback"
DEBUG_OPT="-debug -traceback"
C_DEBUG_OPT="-debug -traceback"
fi
MARCCHECK=
#MARCCHECK="ON"
if test "$MARCCHECK" = "ON"
then
DEBUG_OPT="$DEBUG_OPT -fpe0 -fp-stack-check -check all -ftrapuv "
C_DEBUG_OPT="$C_DEBUG_OPT -fp-stack-check -check-uninit -Wformat -ftrapuv "
fi
MARCCODECOV=
#MARCCODECOV="ON"
MARCCODEPROF=
#MARCCODEPROF="ON"
if test "$MARC_INTEGER_SIZE" = "i4" ; then
I8FFLAGS=
I8DEFINES=
I8CDEFINES=
else
I8FFLAGS="-i8"
I8DEFINES="-DI64"
I8CDEFINES="-U_DOUBLE -D_SINGLE"
fi
MTHREAD=OPENMP
if test "$MARC_OPENMP" = "NONE" ; then
MTHREAD=NONE
fi
#MTHREAD=NONE
if test "$_OEM_NASTRAN" -ne 0
then
MTHREAD=NONE
fi
if test "$AEM_DLL" -eq 1
then
MTHREAD=NONE
CASISOLVER=NONE
VKISOLVER=NONE
MF2SOLVER=NONE
INTELSOLVER=NONE
BCSGPUSOLVER=NONE
SOLVER2GPU=NONE
OPENSSL_LIB=
MARC_DLL=NONE
METISLIBS=
fi
OMP_COMPAT=NO
OMP_COMPAT=YES
if test "$MTHREAD" = "NONE"
then
OMP_COMPAT=NO
fi
CDEFINES=
FDEFINES=
if test "$_OEM_NASTRAN" -ne 0
then
CDEFINES="$CDEFINES -D_OEM_NASTRAN"
FDEFINES="$FDEFINES -D_OEM_NASTRAN"
fi
FDEFINES="$FDEFINES -D_IMPLICITNONE"
if test "$_OEM_NASTRAN" -eq 0
then
FDEFINES="$FDEFINES -DMKL -DOPENMP"
fi
if test "$OMP_COMPAT" = "YES"
then
FDEFINES="$FDEFINES -DOMP_COMPAT"
fi
# -D_MSCMARC
FDEFINES="$FDEFINES -D_MSCMARC $DEBUG_OPT $MARC_SIMUFACT"
CDEFINES="$CDEFINES -D_MSCMARC $C_DEBUG_OPT $I8CDEFINES"
if test "$AEM_DLL" -eq 1
then
FDEFINES="$FDEFINES -D_AEMNL -DAAA"
CDEFINES="$CDEFINES -D_AEMNL -DAAA"
fi
CINCL="-I$MARC_SOURCE/mdsrc -I$MARC_SOURCE/csource $METIS -I$LAPI_IMPORTS/common/include"
if test "$_OEM_NASTRAN" -ne 0
then
CINCL="$CINCL -I../../include"
fi
CC_OPT=
if test "$MTHREAD" = "OPENMP"
then
CC_OPT=" $CC_OPT -qopenmp"
fi
CC="icc -c $CC_OPT -O1 $I8DEFINES -DLinux -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
CCLOW="icc -c $CC_OPT -O0 $I8DEFINES -DLinux -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
CCHIGH="icc -c $CC_OPT -O3 $I8DEFINES -DLinux -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
if test "$MARCDEBUG" = "ON"
then
CC="icc -c $CC_OPT -DLinux $I8DEFINES -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
CCLOW="icc $CC_OPT -c -DLinux $I8DEFINES -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
CCHIGH="icc $CC_OPT -c -DLinux $I8DEFINES -DLINUX -DLinux_intel $CDEFINES $CINCL $SOLVERFLAGS $OPENSSL_INCLUDE "
fi
LOAD_CC="icc $CC_OPT -O1 -DLinux -DLINUX -DLinux_intel"
CCT="$CC"
CCTLOW="$CCLOW"
CCTHIGH="$CCHIGH"
#PROFILE="-Mprof=func"
#PROFILE="-Mprof=lines"
#PROFILE="-Mprof=func,mpi"
PROFILE=
#PROFILE="-init=snan,arrays -CB -traceback -fpe0 -fp-stack-check -check all -check uninit -ftrapuv"
if test "$MARCCODECOV" = "ON"
then
PROFILE="-prof-gen=srcpos"
fi
if test "$MARCCODEPROF" = "ON"
then
PROFILE=" $PROFILE -pg"
fi
FORT_OPT="-c -assume byterecl -safe_cray_ptr -mp1 -WB -fp-model source"
if test "$MTHREAD" = "OPENMP"
then
FORT_OPT=" $FORT_OPT -qopenmp"
if test "$OMP_COMPAT" = "YES"
then
FORT_OPT=" $FORT_OPT -qopenmp-threadprivate=compat"
fi
else
# FORT_OPT=" $FORT_OPT -auto "
FORT_OPT=" $FORT_OPT -save -zero"
fi
if test "$MARCHDF_HDF" = "HDF"; then
FORT_OPT="$FORT_OPT -DMARCHDF_HDF=$MARCHDF_HDF $HDF_INCLUDE"
fi
FORTLOW="$FCOMP $FORT_OPT $PROFILE -O0 $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTRAN="$FCOMP $FORT_OPT $PROFILE -O1 $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTHIGH="$FCOMP $FORT_OPT $PROFILE -fno-alias -O3 $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTNA="$FCOMP $FORT_OPT -fno-alias -O3 $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM"
# for compiling free form f90 files. high opt, integer(4)
FORTF90="$FCOMP -c -O3"
if test "$MARCDEBUG" = "ON"
then
FORTLOW="$FCOMP $FORT_OPT $PROFILE $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTRAN="$FCOMP $FORT_OPT $PROFILE $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTHIGH="$FCOMP $FORT_OPT $PROFILE -fno-alias $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM $SOLVERFLAGS -I$KDTREE2_MOD"
FORTNA="$FCOMP $FORT_OPT -fno-alias $I8FFLAGS -I$MARC_SOURCE/common \
$MUMPS_INCLUDE $I8DEFINES -DLinux -DLINUX -DLinux_intel $FDEFINES $DDM"
fi
FORTLOWT="$FORTLOW"
FORTRANT="$FORTRAN"
FORTHIGHT="$FORTHIGH"
FORTRANMNF="$FCOMP -c $FDEFINES "
CCMNF="icc -c -O1 -DLinux -DLINUX -DLinux_intel -Dport2egcs -I$MARC_SOURCE/marctoadams/mnf/include -D_LARGEFILE64_SOURCE"
if test $MPITYPE != none
then
if test $MPITYPE = hpmpi
then
LOAD="$MPI_ROOT/bin/$FCOMPMPI ${LOADOPTIONS} -L$MPI_ROOT/lib/$ARCHITECTURE $PROFILE $LINK_OPT -o "
LOADT="$MPI_ROOT/bin/$FCOMPMPI ${LOADOPTIONS} -L$MPI_ROOT/lib/$ARCHITECTURE $PROFILE $LINK_OPT -o "
fi
# Uncomment the following lines to turn on the tracer and commnet out the next 5 lines
# if test $MPITYPE = intelmpi
# then
# INCLUDEMPI="-I$MPI_ROOT/include -I$VT_ROOT/include"
# LOAD="$MPI_ROOT/bin/$FCOMPMPI $PROFILE $INCLUDEMPI -g -t=log $LINK_OPT -o "
# LOADT="$MPI_ROOT/bin/$FCOMPMPI $PROFILE $INCLUDEMPI -g -t=log $LINK_OPT -o "
# fi
if test $MPITYPE = intelmpi
then
LOAD="ifort $PROFILE $LINK_OPT -o "
LOADT="ifort $PROFILE $LINK_OPT -o "
fi
else
LOAD="$FCOMP $LINK_OPT -o "
LOADT="$FCOMP $LINK_OPT -o "
fi
if test "$MARC_DLL" = MARC_DLL
then
FORTLOW="$FORTLOW -fpp -fPIC"
FORTRAN="$FORTRAN -fpp -fPIC"
FORTHIGH="$FORTHIGH -fpp -fPIC"
FORTRANMNF="$FORTRANMNF -fpp -fPIC"
CC="$CC -fPIC"
CCMNF="$CCMNF -fPIC"
LINK_EXE_MARC="-L$MARC_LIB -lmarc -L$MARC_LIB_SHARED -lguide -lpthread"
LINK_MARC_DLL="-shared -fPIC"
LOAD_DLL=$LOAD
LOADT_DLL=$LOADT
EXT_DLL="so"
fi
if test "$AEM_DLL" -eq 1
then
FORTLOW="$FORTLOW -fpp -fPIC"
FORTRAN="$FORTRAN -fpp -fPIC"
FORTHIGH="$FORTHIGH -fpp -fPIC"
FORTRANMNF="$FORTRANMNF -fpp -fPIC"
CC="$CC -fPIC"
CCMNF="$CCMNF -fPIC"
LINK_EXE_MARC="-L$MARC_LIB -lmarc -L$MARC_LIB_SHARED -lguide"
LINK_MARC_DLL="-shared -fPIC"
LOAD_DLL=$LOAD
LOADT_DLL=$LOADT
EXT_DLL="so"
fi
XLIBS="-L/usr/X11/lib -lX11 "
#
# define archive and ranlib syntax
#
ARC="ar rvl"
ARD="ar dvl"
ARX="ar xl"
RAN=""
#
# choose which libraries you want to use ( e.g. blas )
#
if test "$VKISOLVER" = VKI
then
VKISOLVERLIBS="$MARC_LIB/vkisolver.a"
else
VKISOLVERLIBS=
fi
if test "$CASISOLVER" = CASI
then
CASISOLVERLIBS="$CASILIB_DIR/libmarccasi.a $CASILIB_DIR/libcasi.a"
else
CASISOLVERLIBS=
fi
MF2SOLVERLIBS=
if test "$MF2SOLVER" = MF2PARALLEL
then
MF2SOLVERLIBS="$MARC_LIB/mf2parallel/libseq.a \
$MARC_LIB/mf2parallel/libsym.a \
$MARC_LIB/mf2parallel/libmet.a \
$MARC_LIB/mf2parallel/libmf2.a \
$MARC_LIB/mf2parallel/libgauss.a \
$MARC_LIB/mf2parallel/libmf2.a \
$MARC_LIB/mf2parallel/libgauss.a \
$MARC_LIB/mf2parallel/libnum.a \
$MARC_LIB/mf2parallel/libutl.a \
$MARC_LIB/mf2parallel/libr8.a \
$MARC_LIB/mf2parallel/libz.a "
fi
if test "$MUMPSSOLVER" = MUMPS
then
MUMPSSOLVERLIBS="$MUMPSLIB_DIR/libmumps.a"
if test $MPITYPE = none
then
MUMPSSOLVERLIBS2=
echo hello > /dev/null
fi
if test $MPITYPE = intelmpi
then
MUMPSSOLVERLIBS="$MUMPSLIB_DIR/libmumps_intelmpi.a"
if test "$MARC_INTEGER_SIZE" = "i4" ; then
MUMPSSOLVERLIBS2=" $MARC_MKL/libmkl_blacs_intelmpi_lp64.a "
else
MUMPSSOLVERLIBS2=" $MARC_MKL/libmkl_blacs_intelmpi_ilp64.a "
fi
fi
if test $MPITYPE = hpmpi
then
MUMPSSOLVERLIBS="$MUMPSLIB_DIR/libmumps_hpmpi.a"
if test "$MARC_INTEGER_SIZE" = "i4" ; then
MUMPSSOLVERLIBS2=" $MARC_MKL/libmkl_blacs_intelmpi_lp64.a"
else
MUMPSSOLVERLIBS2=" $MARC_MKL/libmkl_blacs_intelmpi_ilp64.a"
fi
fi
else
MUMPSSOLVERLIBS=
MUMPSSOLVERLIBS2=
fi
if test "$BCSGPUSOLVER" = BCSGPU
then
BCSSOLVERLIBS="${MARC_LIB}/bcsgpulib.a "
MARCCUDALIBS1="-L${MARC_LIB}/cuda_dummy -lmarccuda "
MARCCUDALIBS2="-L${MARC_LIB}/cuda -lmarccuda "
MARCCUDALIBS=$MARCCUDALIBS1
else
BCSSOLVERLIBS="${MARC_LIB}/bcslib.a "
fi
SOLVER2LIBS=
if test "$SOLVER2GPU" = GPU
then
SOLVER2LIBS="-L$MARC_LIB/cuda_dummy -lsolver2gpu"
fi
if test "$AEM_DLL" -eq 1
then
BCSSOLVERLIBS=
fi
if test "$MARC_INTEGER_SIZE" = "i4" ; then
MKLLIB="$MARC_MKL/libmkl_scalapack_lp64.a -Wl,--start-group $MARC_MKL/libmkl_intel_lp64.a $MARC_MKL/libmkl_intel_thread.a $MARC_MKL/libmkl_core.a $MUMPSSOLVERLIBS2 -Wl,--end-group"
else
MKLLIB="$MARC_MKL/libmkl_scalapack_ilp64.a -Wl,--start-group $MARC_MKL/libmkl_intel_ilp64.a $MARC_MKL/libmkl_intel_thread.a $MARC_MKL/libmkl_core.a $MUMPSSOLVERLIBS2 -Wl,--end-group"
fi
SECLIBS="-L$MARC_LIB -llapi"
SOLVERLIBS="${BCSSOLVERLIBS} ${VKISOLVERLIBS} ${CASISOLVERLIBS} ${MF2SOLVERLIBS} \
$MKLLIB -L$MARC_MKL -liomp5 \
$MARC_LIB/blas_src.a ${ACSI_LIB}/ACSI_MarcLib.a $KDTREE2_LIB/libkdtree2.a $HDF_LIBS $SOLVER2LIBS"
SOLVERLIBS_DLL=${SOLVERLIBS}
if test "$AEM_DLL" -eq 1
then
SOLVERLIBS_DLL="$MKLLIB -L$MARC_MKL -liomp5 $MARC_LIB/blas_src.a"
fi
MRCLIBS="$MARC_LIB/clib.a ${CASISOLVERLIBS}"
MRCLIBSPAR="$MARC_LIB/clib.a"
STUBS="$MARC_LIB/stubs.a "
MNFLIBS="$MARC_LIB/libmnf.a"
MDUSER="$MARC_LIB/md_user.a"
if test "X$MARC_SIMUFACT" != "X"
then
SFLIB="-L$SFMATDIR -lMBA_Grain"
else
SFLIB=" "
fi
OPENMP="-qopenmp"
if test "$AEM_DLL" -eq 1
then
LOAD_DLL=$LOAD
OPENMP=
LIBMNF=
OPENSSL=NONE
fi
SYSLIBS=" $OPENMP -lpthread -shared-intel -cxxlib"
# Uncomment the following lines to turn on the trace and comment out the next 4 lines
# if test $MPITYPE = intelmpi
# then
# SYSLIBS="-L${VT_ROOT}/lib -lVT -ldwarf -lelf -lm -lpthread \
# -L${MPI_ROOT}/lib64 -lmpi -lmpiif -lmpigi -lrt"
# fi
if test $MPITYPE = intelmpi
then
SYSLIBS="-L${MPI_ROOT}/lib/release -lmpi -L${MPI_ROOT}/lib -lmpifort -lrt $OPENMP -threads -lpthread -shared-intel -cxxlib"
fi
SYSLIBSPAR=" "
MARC_DLL_CODES="runmarc.f"
BLAS_SRC="dzero.f icopy.f izero.f"
if test "$_OEM_NASTRAN" -ne 0
then
if test "$MARC_INTEGER_SIZE" = "i4" ; then
BLAS_SRC="$BLAS_SRC dsctr.f zsctr.f dzasum.f daxpyi.f zaxpyi.f dgthr.f zgthr.f"
else
BLAS_SRC="ALL"
fi
fi
LOW_OPT_CODES="are163.f contro.f ndext.f omarc.f omarca.f omarcb.f omarcc.f \
omars.f fixbc.f triang.f bet049.f norst3.f eldata.f \
elec*.f elct*.f fmeig.f oada00.f ogeig.f updtrbe2.f cycrota.f \
cordef.f ogpk.f ogtan.f eldam.f formrbe3.f \
inertie.f em_sso072.f cn_fol3d_qpatch6.f cosim_begin.f"
if test "$MARC_INTEGER_SIZE" = "i8" ; then
LOW_OPT_CODES="$LOW_OPT_CODES bbcseg.f"
fi
HIGH_OPT_CODES="dpsmsa1.f dpsmsa2.f dpsmsa3.f dpsmsa4.f dpsmsa5.f dpsmsa6.f \
dpsmsa7.f dpsmsa8.f dpsmsa9.f dpsmsa10.f dpsmsa11.f dpsmsa12.f \
dpsmsa13.f dpsmsa14.f dpsmsa15.f dpsmsa16.f dpsmsah.f tpsmsah.f cn_qsort4_11.f \
prei11.f prei12.f prei31.f prei32.f prei41.f prei42.f prei61.f prei62.f \
prei1n.f prei2n.f cgfullnts1.f cgfullnts2.f cg1n.f cg2n.f cg3n.f \
cg4n.f cg5n.f cg6n.f cgnn.f sortab.f sortab1.f triann1.f preinv_nts.f \
cn_sur_patchl.f cn_quad_3e.f"
MAXNUM=1000000

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@ -0,0 +1,5 @@
#!/bin/sh
# This script opens a window running an editor.
# The command to invoke the editor is specified during DAMASK installation
%EDITOR% $*

View File

@ -0,0 +1,18 @@
#!/bin/sh
# This script opens a window running an editor. The default window is an
# xterm, and the default editor is vi. These may be customized.
dir=
for d in /usr/bin /usr/bin/X11; do
if test -x "$d/xterm"; then
dir="$d"
break
fi
done
if test -z "$dir"; then
echo "$0: Could not find xterm"
exit 1
fi
"$dir/xterm" -T "vi $*" -n "vi $*" -e vi $*

View File

@ -0,0 +1,8 @@
#!/bin/sh
if [ "$1" = "" ]; then
echo "usage: $0 job_name"
exit 1
fi
echo STOP > $1.cnt

View File

@ -0,0 +1,8 @@
#!/bin/sh
if [ "$1" = "" ]; then
echo "usage: $0 job_name"
exit 1
fi
echo STOP > $1.cnt

View File

@ -0,0 +1,8 @@
#!/bin/sh
if [ "$1" = "" ]; then
echo "usage: $0 job_name"
exit 1
fi
echo STOP > $1.cnt

View File

@ -0,0 +1,8 @@
#!/bin/sh
if [ "$1" = "" ]; then
echo "usage: $0 job_name"
exit 1
fi
echo STOP > $1.cnt

View File

@ -0,0 +1,189 @@
#!/bin/sh
#
# The exit status of this script is read by Mentat.
# Normal exit status is 0.
#
DIR=/tmp/msc/marc2020
if test $MARCDIR1
then
DIR=$MARCDIR1
fi
if test -z "$DIR"; then
REALCOM="`ls -l $0 |awk '{ print $NF; }'`"
DIRSCRIPT=`dirname $REALCOM`
case $DIRSCRIPT in
\/*)
;;
*)
DIRSCRIPT=`pwd`/$DIRSCRIPT
;;
esac
. $DIRSCRIPT/getarch
DIR="$MENTAT_MARCDIR"
fi
SRCEXT=.f
SRCEXTC=.F
RSTEXT=.t08
PSTEXT=.t19
PSTEXTB=.t16
VWFCEXT=.vfs
slv=$1
version=$2
ndom_fea_solver=$3
ndom_preprocessor=$4
hostfile=$5
compat=$6
job=$7
srcfile=$8
srcmeth=$9
shift 9 # cannot use $10, $11, ...
restart=$1
postfile=$2
viewfactorsfile=$3
copy_datfile="-ci $4"
copy_postfile="-cr $5"
scr_dir=$6
dcoup=$7
assem_recov_nthread=$8
nthread=$9
shift 9 # cannot use $10, $11, ...
nsolver=$1
mode=$2
gpu=$3
if [ "$slv" != "" -a "$slv" != "marc" -a "$slv" != "datfit" ]; then
slv="-iam sfm"
fi
if [ "$slv" == "marc" ]; then
slv=""
fi
if [ "$slv" == "datfit" ]; then
slv="-iam datfit"
fi
if [ "$ndom_fea_solver" != "" -a "$ndom_fea_solver" != "1" ]; then
nprocds="-nprocds $ndom_fea_solver"
else
nprocd=""
if [ "$ndom_preprocessor" != "" -a "$ndom_preprocessor" != "1" ]; then
nprocd="-nprocd $ndom_preprocessor"
else
nprocd=""
fi
fi
if [ "$srcfile" != "" -a "$srcfile" != "-" ]; then
srcfile=`echo $srcfile | sed "s/$SRCEXT$//" | sed "s/$SRCEXTC$//"`
case "$srcmeth" in
-)
srcfile="-u $srcfile"
;;
compsave)
srcfile="-u $srcfile -save y"
;;
runsaved)
srcfile="-prog $srcfile"
;;
esac
else
srcfile=""
fi
if [ "$restart" != "" -a "$restart" != "-" ]; then
restart=`echo $restart | sed "s/$RSTEXT$//"`
restart="-r $restart"
else
restart=""
fi
if [ "$postfile" != "" -a "$postfile" != "-" ]; then
postfile=`echo $postfile | sed "s/$PSTEXT$//"`
postfile=`echo $postfile | sed "s/$PSTEXTB$//"`
postfile="-pid $postfile"
else
postfile=""
fi
if [ "$viewfactorsfile" != "" -a "$viewfactorsfile" != "-" ]; then
viewfactorsfile=`echo $viewfactorsfile | sed "s/$VWFCEXT$//"`
viewfactorsfile="-vf $viewfactorsfile"
else
viewfactorsfile=""
fi
if [ "$hostfile" != "" -a "$hostfile" != "-" ]; then
hostfile="-ho $hostfile"
else
hostfile=""
fi
if [ "$compat" != "" -a "$compat" != "-" ]; then
compat="-co $compat"
else
compat=""
fi
if [ "$scr_dir" != "" -a "$scr_dir" != "-" ]; then
scr_dir="-sd $scr_dir"
else
scr_dir=""
fi
if [ "$dcoup" != "" -a "$dcoup" != "0" ]; then
dcoup="-dcoup $dcoup"
else
dcoup=""
fi
if [ "$assem_recov_nthread" != "" -a "$assem_recov_nthread" != "1" ]; then
assem_recov_nthread="-nthread_elem $assem_recov_nthread"
else
assem_recov_nthread=""
fi
if [ "$nthread" != "" -a "$nthread" != "0" -a "$nthread" != "1" ]; then
nthread="-nthread $nthread"
else
nthread=""
fi
if [ "$nsolver" != "" -a "$nsolver" != "0" ]; then
nsolver="-nsolver $nsolver"
else
nsolver=""
fi
case "$mode" in
4) mode="-mo i4" ;;
8) mode="-mo i8" ;;
*) mode= ;;
esac
if [ "$gpu" != "" -a "$gpu" != "-" ]; then
gpu="-gpu $gpu"
else
gpu=""
fi
rm -f $job.cnt
rm -f $job.sts
rm -f $job.out
rm -f $job.log
# To prevent a mismatch with the python version used by the solver
# do *not* prepend $MENTAT_INSTALL_DIR/python/bin to environment variable PATH
# unset environment variables PYTHONHOME and PYTHONPATH
unset PYTHONHOME
unset PYTHONPATH
"${DIR}/tools/run_marc" $slv -j $job -v n -b y $nprocds $nprocd \
$srcfile $restart $postfile $viewfactorsfile $hostfile \
$compat $copy_datfile $copy_postfile $scr_dir $dcoup \
$assem_recov_nthread $nthread $nsolver $mode $gpu > /dev/null 2>&1
sleep 1
exit 0

View File

@ -0,0 +1,191 @@
#!/bin/sh
#
# The exit status of this script is read by Mentat.
# Normal exit status is 0.
#
DIR=%INSTALLDIR%/marc%VERSION%
if test $MARCDIR1
then
DIR=$MARCDIR1
fi
if test -z "$DIR"; then
REALCOM="`ls -l $0 |awk '{ print $NF; }'`"
DIRSCRIPT=`dirname $REALCOM`
case $DIRSCRIPT in
\/*)
;;
*)
DIRSCRIPT=`pwd`/$DIRSCRIPT
;;
esac
. $DIRSCRIPT/getarch
DIR="$MENTAT_MARCDIR"
fi
SRCEXT=.f
SRCEXTC=.F
RSTEXT=.t08
PSTEXT=.t19
PSTEXTB=.t16
VWFCEXT=.vfs
slv=$1
version=$2
ndom_fea_solver=$3
ndom_preprocessor=$4
hostfile=$5
compat=$6
job=$7
srcfile=$8
srcmeth=$9
shift 9 # cannot use $10, $11, ...
restart=$1
postfile=$2
viewfactorsfile=$3
copy_datfile="-ci $4"
copy_postfile="-cr $5"
scr_dir=$6
dcoup=$7
assem_recov_nthread=$8
nthread=$9
shift 9 # cannot use $10, $11, ...
nsolver=$1
mode=$2
gpu=$3
if [ "$slv" != "" -a "$slv" != "marc" -a "$slv" != "datfit" ]; then
slv="-iam sfm"
fi
if [ "$slv" = "marc" ]; then
slv=""
fi
if [ "$slv" = "datfit" ]; then
slv="-iam datfit"
fi
if [ "$ndom_fea_solver" != "" -a "$ndom_fea_solver" != "1" ]; then
nprocds="-nprocds $ndom_fea_solver"
else
nprocd=""
if [ "$ndom_preprocessor" != "" -a "$ndom_preprocessor" != "1" ]; then
nprocd="-nprocd $ndom_preprocessor"
else
nprocd=""
fi
fi
if [ "$srcfile" != "" -a "$srcfile" != "-" ]; then
srcfile=`echo $srcfile | sed "s/$SRCEXT$//" | sed "s/$SRCEXTC$//"`
case "$srcmeth" in
-)
srcfile="-u $srcfile"
;;
compsave)
srcfile="-u $srcfile -save y"
;;
runsaved)
srcfile=${srcfile%.*}".marc"
srcfile="-prog $srcfile"
;;
esac
else
srcfile=""
fi
if [ "$restart" != "" -a "$restart" != "-" ]; then
restart=`echo $restart | sed "s/$RSTEXT$//"`
restart="-r $restart"
else
restart=""
fi
if [ "$postfile" != "" -a "$postfile" != "-" ]; then
postfile=`echo $postfile | sed "s/$PSTEXT$//"`
postfile=`echo $postfile | sed "s/$PSTEXTB$//"`
postfile="-pid $postfile"
else
postfile=""
fi
if [ "$viewfactorsfile" != "" -a "$viewfactorsfile" != "-" ]; then
viewfactorsfile=`echo $viewfactorsfile | sed "s/$VWFCEXT$//"`
viewfactorsfile="-vf $viewfactorsfile"
else
viewfactorsfile=""
fi
if [ "$hostfile" != "" -a "$hostfile" != "-" ]; then
hostfile="-ho $hostfile"
else
hostfile=""
fi
if [ "$compat" != "" -a "$compat" != "-" ]; then
compat="-co $compat"
else
compat=""
fi
if [ "$scr_dir" != "" -a "$scr_dir" != "-" ]; then
scr_dir="-sd $scr_dir"
else
scr_dir=""
fi
if [ "$dcoup" != "" -a "$dcoup" != "0" ]; then
dcoup="-dcoup $dcoup"
else
dcoup=""
fi
if [ "$assem_recov_nthread" != "" -a "$assem_recov_nthread" != "1" ]; then
assem_recov_nthread="-nthread_elem $assem_recov_nthread"
else
assem_recov_nthread=""
fi
if [ "$nthread" != "" -a "$nthread" != "0" -a "$nthread" != "1" ]; then
nthread="-nthread $nthread"
else
nthread=""
fi
if [ "$nsolver" != "" -a "$nsolver" != "0" ]; then
nsolver="-nsolver $nsolver"
else
nsolver=""
fi
case "$mode" in
4) mode="-mo i4" ;;
8) mode="-mo i8" ;;
*) mode= ;;
esac
if [ "$gpu" != "" -a "$gpu" != "-" ]; then
gpu="-gpu $gpu"
else
gpu=""
fi
rm -f $job.cnt
rm -f $job.sts
rm -f $job.out
rm -f $job.log
# To prevent a mismatch with the python version used by the solver
# do *not* prepend $MENTAT_INSTALL_DIR/python/bin to environment variable PATH
# unset environment variables PYTHONHOME and PYTHONPATH
unset PYTHONHOME
unset PYTHONPATH
"${DIR}/tools/run_damask_hmp" $slv -j $job -v n -b y $nprocds $nprocd \
$srcfile $restart $postfile $viewfactorsfile $hostfile \
$compat $copy_datfile $copy_postfile $scr_dir $dcoup \
$assem_recov_nthread $nthread $nsolver $mode $gpu > /dev/null 2>&1
sleep 1
exit 0

View File

@ -0,0 +1,191 @@
#!/bin/sh
#
# The exit status of this script is read by Mentat.
# Normal exit status is 0.
#
DIR=%INSTALLDIR%/marc%VERSION%
if test $MARCDIR1
then
DIR=$MARCDIR1
fi
if test -z "$DIR"; then
REALCOM="`ls -l $0 |awk '{ print $NF; }'`"
DIRSCRIPT=`dirname $REALCOM`
case $DIRSCRIPT in
\/*)
;;
*)
DIRSCRIPT=`pwd`/$DIRSCRIPT
;;
esac
. $DIRSCRIPT/getarch
DIR="$MENTAT_MARCDIR"
fi
SRCEXT=.f
SRCEXTC=.F
RSTEXT=.t08
PSTEXT=.t19
PSTEXTB=.t16
VWFCEXT=.vfs
slv=$1
version=$2
ndom_fea_solver=$3
ndom_preprocessor=$4
hostfile=$5
compat=$6
job=$7
srcfile=$8
srcmeth=$9
shift 9 # cannot use $10, $11, ...
restart=$1
postfile=$2
viewfactorsfile=$3
copy_datfile="-ci $4"
copy_postfile="-cr $5"
scr_dir=$6
dcoup=$7
assem_recov_nthread=$8
nthread=$9
shift 9 # cannot use $10, $11, ...
nsolver=$1
mode=$2
gpu=$3
if [ "$slv" != "" -a "$slv" != "marc" -a "$slv" != "datfit" ]; then
slv="-iam sfm"
fi
if [ "$slv" = "marc" ]; then
slv=""
fi
if [ "$slv" = "datfit" ]; then
slv="-iam datfit"
fi
if [ "$ndom_fea_solver" != "" -a "$ndom_fea_solver" != "1" ]; then
nprocds="-nprocds $ndom_fea_solver"
else
nprocd=""
if [ "$ndom_preprocessor" != "" -a "$ndom_preprocessor" != "1" ]; then
nprocd="-nprocd $ndom_preprocessor"
else
nprocd=""
fi
fi
if [ "$srcfile" != "" -a "$srcfile" != "-" ]; then
srcfile=`echo $srcfile | sed "s/$SRCEXT$//" | sed "s/$SRCEXTC$//"`
case "$srcmeth" in
-)
srcfile="-u $srcfile"
;;
compsave)
srcfile="-u $srcfile -save y"
;;
runsaved)
srcfile=${srcfile%.*}".marc"
srcfile="-prog $srcfile"
;;
esac
else
srcfile=""
fi
if [ "$restart" != "" -a "$restart" != "-" ]; then
restart=`echo $restart | sed "s/$RSTEXT$//"`
restart="-r $restart"
else
restart=""
fi
if [ "$postfile" != "" -a "$postfile" != "-" ]; then
postfile=`echo $postfile | sed "s/$PSTEXT$//"`
postfile=`echo $postfile | sed "s/$PSTEXTB$//"`
postfile="-pid $postfile"
else
postfile=""
fi
if [ "$viewfactorsfile" != "" -a "$viewfactorsfile" != "-" ]; then
viewfactorsfile=`echo $viewfactorsfile | sed "s/$VWFCEXT$//"`
viewfactorsfile="-vf $viewfactorsfile"
else
viewfactorsfile=""
fi
if [ "$hostfile" != "" -a "$hostfile" != "-" ]; then
hostfile="-ho $hostfile"
else
hostfile=""
fi
if [ "$compat" != "" -a "$compat" != "-" ]; then
compat="-co $compat"
else
compat=""
fi
if [ "$scr_dir" != "" -a "$scr_dir" != "-" ]; then
scr_dir="-sd $scr_dir"
else
scr_dir=""
fi
if [ "$dcoup" != "" -a "$dcoup" != "0" ]; then
dcoup="-dcoup $dcoup"
else
dcoup=""
fi
if [ "$assem_recov_nthread" != "" -a "$assem_recov_nthread" != "1" ]; then
assem_recov_nthread="-nthread_elem $assem_recov_nthread"
else
assem_recov_nthread=""
fi
if [ "$nthread" != "" -a "$nthread" != "0" -a "$nthread" != "1" ]; then
nthread="-nthread $nthread"
else
nthread=""
fi
if [ "$nsolver" != "" -a "$nsolver" != "0" ]; then
nsolver="-nsolver $nsolver"
else
nsolver=""
fi
case "$mode" in
4) mode="-mo i4" ;;
8) mode="-mo i8" ;;
*) mode= ;;
esac
if [ "$gpu" != "" -a "$gpu" != "-" ]; then
gpu="-gpu $gpu"
else
gpu=""
fi
rm -f $job.cnt
rm -f $job.sts
rm -f $job.out
rm -f $job.log
# To prevent a mismatch with the python version used by the solver
# do *not* prepend $MENTAT_INSTALL_DIR/python/bin to environment variable PATH
# unset environment variables PYTHONHOME and PYTHONPATH
unset PYTHONHOME
unset PYTHONPATH
"${DIR}/tools/run_damask_mp" $slv -j $job -v n -b y $nprocds $nprocd \
$srcfile $restart $postfile $viewfactorsfile $hostfile \
$compat $copy_datfile $copy_postfile $scr_dir $dcoup \
$assem_recov_nthread $nthread $nsolver $mode $gpu > /dev/null 2>&1
sleep 1
exit 0

View File

@ -0,0 +1,191 @@
#!/bin/sh
#
# The exit status of this script is read by Mentat.
# Normal exit status is 0.
#
DIR=%INSTALLDIR%/marc%VERSION%
if test $MARCDIR1
then
DIR=$MARCDIR1
fi
if test -z "$DIR"; then
REALCOM="`ls -l $0 |awk '{ print $NF; }'`"
DIRSCRIPT=`dirname $REALCOM`
case $DIRSCRIPT in
\/*)
;;
*)
DIRSCRIPT=`pwd`/$DIRSCRIPT
;;
esac
. $DIRSCRIPT/getarch
DIR="$MENTAT_MARCDIR"
fi
SRCEXT=.f
SRCEXTC=.F
RSTEXT=.t08
PSTEXT=.t19
PSTEXTB=.t16
VWFCEXT=.vfs
slv=$1
version=$2
ndom_fea_solver=$3
ndom_preprocessor=$4
hostfile=$5
compat=$6
job=$7
srcfile=$8
srcmeth=$9
shift 9 # cannot use $10, $11, ...
restart=$1
postfile=$2
viewfactorsfile=$3
copy_datfile="-ci $4"
copy_postfile="-cr $5"
scr_dir=$6
dcoup=$7
assem_recov_nthread=$8
nthread=$9
shift 9 # cannot use $10, $11, ...
nsolver=$1
mode=$2
gpu=$3
if [ "$slv" != "" -a "$slv" != "marc" -a "$slv" != "datfit" ]; then
slv="-iam sfm"
fi
if [ "$slv" = "marc" ]; then
slv=""
fi
if [ "$slv" = "datfit" ]; then
slv="-iam datfit"
fi
if [ "$ndom_fea_solver" != "" -a "$ndom_fea_solver" != "1" ]; then
nprocds="-nprocds $ndom_fea_solver"
else
nprocd=""
if [ "$ndom_preprocessor" != "" -a "$ndom_preprocessor" != "1" ]; then
nprocd="-nprocd $ndom_preprocessor"
else
nprocd=""
fi
fi
if [ "$srcfile" != "" -a "$srcfile" != "-" ]; then
srcfile=`echo $srcfile | sed "s/$SRCEXT$//" | sed "s/$SRCEXTC$//"`
case "$srcmeth" in
-)
srcfile="-u $srcfile"
;;
compsave)
srcfile="-u $srcfile -save y"
;;
runsaved)
srcfile=${srcfile%.*}".marc"
srcfile="-prog $srcfile"
;;
esac
else
srcfile=""
fi
if [ "$restart" != "" -a "$restart" != "-" ]; then
restart=`echo $restart | sed "s/$RSTEXT$//"`
restart="-r $restart"
else
restart=""
fi
if [ "$postfile" != "" -a "$postfile" != "-" ]; then
postfile=`echo $postfile | sed "s/$PSTEXT$//"`
postfile=`echo $postfile | sed "s/$PSTEXTB$//"`
postfile="-pid $postfile"
else
postfile=""
fi
if [ "$viewfactorsfile" != "" -a "$viewfactorsfile" != "-" ]; then
viewfactorsfile=`echo $viewfactorsfile | sed "s/$VWFCEXT$//"`
viewfactorsfile="-vf $viewfactorsfile"
else
viewfactorsfile=""
fi
if [ "$hostfile" != "" -a "$hostfile" != "-" ]; then
hostfile="-ho $hostfile"
else
hostfile=""
fi
if [ "$compat" != "" -a "$compat" != "-" ]; then
compat="-co $compat"
else
compat=""
fi
if [ "$scr_dir" != "" -a "$scr_dir" != "-" ]; then
scr_dir="-sd $scr_dir"
else
scr_dir=""
fi
if [ "$dcoup" != "" -a "$dcoup" != "0" ]; then
dcoup="-dcoup $dcoup"
else
dcoup=""
fi
if [ "$assem_recov_nthread" != "" -a "$assem_recov_nthread" != "1" ]; then
assem_recov_nthread="-nthread_elem $assem_recov_nthread"
else
assem_recov_nthread=""
fi
if [ "$nthread" != "" -a "$nthread" != "0" -a "$nthread" != "1" ]; then
nthread="-nthread $nthread"
else
nthread=""
fi
if [ "$nsolver" != "" -a "$nsolver" != "0" ]; then
nsolver="-nsolver $nsolver"
else
nsolver=""
fi
case "$mode" in
4) mode="-mo i4" ;;
8) mode="-mo i8" ;;
*) mode= ;;
esac
if [ "$gpu" != "" -a "$gpu" != "-" ]; then
gpu="-gpu $gpu"
else
gpu=""
fi
rm -f $job.cnt
rm -f $job.sts
rm -f $job.out
rm -f $job.log
# To prevent a mismatch with the python version used by the solver
# do *not* prepend $MENTAT_INSTALL_DIR/python/bin to environment variable PATH
# unset environment variables PYTHONHOME and PYTHONPATH
unset PYTHONHOME
unset PYTHONPATH
"${DIR}/tools/run_damask_lmp" $slv -j $job -v n -b y $nprocds $nprocd \
$srcfile $restart $postfile $viewfactorsfile $hostfile \
$compat $copy_datfile $copy_postfile $scr_dir $dcoup \
$assem_recov_nthread $nthread $nsolver $mode $gpu > /dev/null 2>&1
sleep 1
exit 0

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@ -8,6 +8,8 @@ from pathlib import Path
import damask
marc_version = float(damask.environment.options['MARC_VERSION'])
if int(marc_version) == marc_version:
marc_version = int(marc_version)
msc_root = Path(damask.environment.options['MSC_ROOT'])
damask_root = damask.environment.root_dir
@ -21,8 +23,8 @@ parser.add_argument('--editor', dest='editor', metavar='string', default='vi',
def copy_and_replace(in_file,dst):
with open(in_file) as f:
content = f.read()
content = content.replace('%INSTALLDIR%',str(damask_root/'bin'))
content = content.replace('%VERSION%',str(marc_version).replace('.0',''))
content = content.replace('%INSTALLDIR%',msc_root)
content = content.replace('%VERSION%',str(marc_version))
content = content.replace('%EDITOR%', parser.parse_args().editor)
with open(dst/Path(in_file).name,'w') as f:
f.write(content)

View File

@ -39,21 +39,21 @@ for filename in options.filenames:
N_digits = 5 # hack to keep test intact
for inc in damask.util.show_progress(results.iterate('increments'),len(results.increments)):
table = damask.Table(np.ones(np.product(results.grid),dtype=int)*int(inc[3:]),{'inc':(1,)})
table.add('pos',coords.reshape(-1,3))
table = table.add('pos',coords.reshape(-1,3))
results.pick('materialpoints',False)
results.pick('constituents', True)
for label in options.con:
x = results.get_dataset_location(label)
if len(x) != 0:
table.add(label,results.read_dataset(x,0,plain=True).reshape(results.grid.prod(),-1))
table = table.add(label,results.read_dataset(x,0,plain=True).reshape(results.grid.prod(),-1))
results.pick('constituents', False)
results.pick('materialpoints',True)
for label in options.mat:
x = results.get_dataset_location(label)
if len(x) != 0:
table.add(label,results.read_dataset(x,0,plain=True).reshape(results.grid.prod(),-1))
table = table.add(label,results.read_dataset(x,0,plain=True).reshape(results.grid.prod(),-1))
dirname = os.path.abspath(os.path.join(os.path.dirname(filename),options.dir))
if not os.path.isdir(dirname):

View File

@ -181,14 +181,14 @@ for name in filenames:
if options.shape:
centers = damask.grid_filters.cell_coord(size,F)
shapeMismatch = shapeMismatch(size,F,nodes,centers)
table.add('shapeMismatch(({}))'.format(options.defgrad),
shapeMismatch.reshape(-1,1,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('shapeMismatch(({}))'.format(options.defgrad),
shapeMismatch.reshape(-1,1,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
if options.volume:
volumeMismatch = volumeMismatch(size,F,nodes)
table.add('volMismatch(({}))'.format(options.defgrad),
volumeMismatch.reshape(-1,1,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('volMismatch(({}))'.format(options.defgrad),
volumeMismatch.reshape(-1,1,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -51,8 +51,8 @@ for name in filenames:
shape = (3,) if np.prod(field.shape)//np.prod(grid) == 3 else (3,3) # vector or tensor
field = field.reshape(tuple(grid)+(-1,),order='F').reshape(tuple(grid)+shape)
curl = damask.grid_filters.curl(size,field)
table.add('curlFFT({})'.format(label),
curl.reshape(tuple(grid)+(-1,)).reshape(-1,np.prod(shape),order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('curlFFT({})'.format(label),
curl.reshape(tuple(grid)+(-1,)).reshape(-1,np.prod(shape),order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -67,8 +67,8 @@ for name in filenames:
table = damask.Table.from_ASCII(StringIO(''.join(sys.stdin.read())) if name is None else name)
for label in options.labels:
table.add('d({})/d({})'.format(label,options.coordinates),
derivative(table.get(options.coordinates),table.get(label)),
scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('d({})/d({})'.format(label,options.coordinates),
derivative(table.get(options.coordinates),table.get(label)),
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -53,19 +53,19 @@ for name in filenames:
F = table.get(options.f).reshape(tuple(grid)+(-1,),order='F').reshape(tuple(grid)+(3,3))
if options.nodal:
table = damask.Table(damask.grid_filters.node_coord0(grid,size).reshape(-1,3,order='F'),
{'pos':(3,)})
table.add('avg({}).{}'.format(options.f,options.pos),
damask.grid_filters.node_displacement_avg(size,F).reshape(-1,3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table.add('fluct({}).{}'.format(options.f,options.pos),
damask.grid_filters.node_displacement_fluct(size,F).reshape(-1,3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
{'pos':(3,)})\
.add('avg({}).{}'.format(options.f,options.pos),
damask.grid_filters.node_displacement_avg(size,F).reshape(-1,3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))\
.add('fluct({}).{}'.format(options.f,options.pos),
damask.grid_filters.node_displacement_fluct(size,F).reshape(-1,3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else os.path.splitext(name)[0]+'_nodal.txt')
else:
table.add('avg({}).{}'.format(options.f,options.pos),
table = table.add('avg({}).{}'.format(options.f,options.pos),
damask.grid_filters.cell_displacement_avg(size,F).reshape(-1,3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table.add('fluct({}).{}'.format(options.f,options.pos),
scriptID+' '+' '.join(sys.argv[1:]))\
.add('fluct({}).{}'.format(options.f,options.pos),
damask.grid_filters.cell_displacement_fluct(size,F).reshape(-1,3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -51,8 +51,8 @@ for name in filenames:
shape = (3,) if np.prod(field.shape)//np.prod(grid) == 3 else (3,3) # vector or tensor
field = field.reshape(tuple(grid)+(-1,),order='F').reshape(tuple(grid)+shape)
div = damask.grid_filters.divergence(size,field)
table.add('divFFT({})'.format(label),
div.reshape(tuple(grid)+(-1,)).reshape(-1,np.prod(shape)//3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('divFFT({})'.format(label),
div.reshape(tuple(grid)+(-1,)).reshape(-1,np.prod(shape)//3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -180,8 +180,8 @@ for name in filenames:
for i,feature in enumerate(feature_list):
table.add('ED_{}({})'.format(features[feature]['names'][0],options.id),
distance[i,:],
scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('ED_{}({})'.format(features[feature]['names'][0],options.id),
distance[i,:],
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -67,10 +67,10 @@ for name in filenames:
damask.grid_filters.coord0_check(table.get(options.pos))
for label in options.labels:
table.add('Gauss{}({})'.format(options.sigma,label),
ndimage.filters.gaussian_filter(table.get(label).reshape(-1),
options.sigma,options.order,
mode = 'wrap' if options.periodic else 'nearest'),
scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('Gauss{}({})'.format(options.sigma,label),
ndimage.filters.gaussian_filter(table.get(label).reshape(-1),
options.sigma,options.order,
mode = 'wrap' if options.periodic else 'nearest'),
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -51,8 +51,8 @@ for name in filenames:
shape = (1,) if np.prod(field.shape)//np.prod(grid) == 1 else (3,) # scalar or vector
field = field.reshape(tuple(grid)+(-1,),order='F')
grad = damask.grid_filters.gradient(size,field)
table.add('gradFFT({})'.format(label),
grad.reshape(tuple(grid)+(-1,)).reshape(-1,np.prod(shape)*3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('gradFFT({})'.format(label),
grad.reshape(tuple(grid)+(-1,)).reshape(-1,np.prod(shape)*3,order='F'),
scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -1,64 +0,0 @@
#!/usr/bin/env python3
import os
import sys
from io import StringIO
from optparse import OptionParser
import numpy as np
import damask
scriptName = os.path.splitext(os.path.basename(__file__))[0]
scriptID = ' '.join([scriptName,damask.version])
# --------------------------------------------------------------------
# MAIN
# --------------------------------------------------------------------
parser = OptionParser(option_class=damask.extendableOption, usage='%prog options [ASCIItable(s)]', description = """
Add data in column(s) of mapped ASCIItable selected from the row indexed by the value in a mapping column.
Row numbers start at 1.
""", version = scriptID)
parser.add_option('--index',
dest = 'index',
type = 'string', metavar = 'string',
help = 'column label containing row index')
parser.add_option('-o','--offset',
dest = 'offset',
type = 'int', metavar = 'int',
help = 'constant offset for index column value [%default]')
parser.add_option('-l','--label',
dest = 'label',
action = 'extend', metavar = '<string LIST>',
help = 'column label(s) to be appended')
parser.add_option('-a','--asciitable',
dest = 'asciitable',
type = 'string', metavar = 'string',
help = 'indexed ASCIItable')
parser.set_defaults(offset = 0,
)
(options,filenames) = parser.parse_args()
if filenames == []: filenames = [None]
if options.label is None:
parser.error('no data columns specified.')
if options.index is None:
parser.error('no index column given.')
for name in filenames:
damask.util.report(scriptName,name)
table = damask.Table.from_ASCII(StringIO(''.join(sys.stdin.read())) if name is None else name)
indexedTable = damask.Table.from_ASCII(options.asciitable)
idx = np.reshape(table.get(options.index).astype(int) + options.offset,(-1))-1
for data in options.label:
table.add(data+'_addIndexed',indexedTable.get(data)[idx],scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -1,118 +0,0 @@
#!/usr/bin/env python3
import os
import sys
from optparse import OptionParser
import numpy as np
import damask
scriptName = os.path.splitext(os.path.basename(__file__))[0]
scriptID = ' '.join([scriptName,damask.version])
# --------------------------------------------------------------------
# MAIN
# --------------------------------------------------------------------
parser = OptionParser(option_class=damask.extendableOption, usage='%prog options [ASCIItable(s)]', description = """
Add data of selected column(s) from (first) row of linked ASCIItable that shares the linking column value.
""", version = scriptID)
parser.add_option('--link',
dest = 'link', nargs = 2,
type = 'string', metavar = 'string string',
help = 'column labels of table and linked table containing linking values')
parser.add_option('-l','--label',
dest = 'label',
action = 'extend', metavar = '<string LIST>',
help = 'column label(s) to add from linked ASCIItable')
parser.add_option('-a','--asciitable',
dest = 'asciitable',
type = 'string', metavar = 'string',
help = 'linked ASCIItable')
parser.set_defaults()
(options,filenames) = parser.parse_args()
if options.label is None:
parser.error('no data columns specified.')
if options.link is None:
parser.error('no linking columns given.')
# -------------------------------------- process linked ASCIItable --------------------------------
if options.asciitable is not None and os.path.isfile(options.asciitable):
linkedTable = damask.ASCIItable(name = options.asciitable, readonly = True)
linkedTable.head_read() # read ASCII header info of linked table
linkDim = linkedTable.label_dimension(options.link[1]) # dimension of linking column
missing_labels = linkedTable.data_readArray([options.link[1]]+options.label) # try reading linked ASCII table
linkedTable.close() # close linked ASCII table
if len(missing_labels) > 0:
damask.util.croak('column{} {} not found...'.format('s' if len(missing_labels) > 1 else '',', '.join(missing_labels)))
if len(missing_labels) >= len(options.label):
damask.util.croak('aborting...')
sys.exit()
index = linkedTable.data[:,:linkDim]
data = linkedTable.data[:,linkDim:]
else:
parser.error('no linked ASCIItable given.')
# --- loop over input files -----------------------------------------------------------------------
if filenames == []: filenames = [None]
for name in filenames:
try:
table = damask.ASCIItable(name = name)
except IOError:
continue
damask.util.report(scriptName,"{} {} <== {} {}".format(name,damask.util.deemph('@ '+options.link[0]),
options.asciitable,damask.util.deemph('@ '+options.link[1])))
# ------------------------------------------ read header ------------------------------------------
table.head_read()
# ------------------------------------------ sanity checks ----------------------------------------
errors = []
myLink = table.label_index (options.link[0])
myLinkDim = table.label_dimension(options.link[0])
if myLink < 0: errors.append('linking column {} not found.'.format(options.link[0]))
if myLinkDim != linkDim: errors.append('dimension mismatch for column {}.'.format(options.link[0]))
if errors != []:
damask.util.croak(errors)
table.close(dismiss = True)
continue
# ------------------------------------------ assemble header --------------------------------------
table.info_append(scriptID + '\t' + ' '.join(sys.argv[1:]))
table.labels_append(linkedTable.labels(raw = True)[linkDim:]) # extend with new labels (except for linked column)
table.head_write()
# ------------------------------------------ process data ------------------------------------------
outputAlive = True
while outputAlive and table.data_read(): # read next data line of ASCII table
try:
table.data_append(data[np.argwhere(np.all((list(map(float,table.data[myLink:myLink+myLinkDim])) - index)==0,
axis=1))[0]]) # add data of first matching line
except IndexError:
table.data_append(np.nan*np.ones_like(data[0])) # or add NaNs
outputAlive = table.data_write() # output processed line
# ------------------------------------------ output finalization -----------------------------------
table.close() # close ASCII tables

View File

@ -137,14 +137,14 @@ for name in filenames:
if 'rodrigues' in options.output:
table.add('ro({})'.format(label),o.as_Rodrigues(), scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('ro({})'.format(label),o.as_Rodrigues(), scriptID+' '+' '.join(sys.argv[1:]))
if 'eulers' in options.output:
table.add('eu({})'.format(label),o.as_Eulers(options.degrees), scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('eu({})'.format(label),o.as_Eulers(options.degrees), scriptID+' '+' '.join(sys.argv[1:]))
if 'quaternion' in options.output:
table.add('qu({})'.format(label),o.as_quaternion(), scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('qu({})'.format(label),o.as_quaternion(), scriptID+' '+' '.join(sys.argv[1:]))
if 'matrix' in options.output:
table.add('om({})'.format(label),o.as_matrix(), scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('om({})'.format(label),o.as_matrix(), scriptID+' '+' '.join(sys.argv[1:]))
if 'axisangle' in options.output:
table.add('om({})'.format(label),o.as_axisangle(options.degrees), scriptID+' '+' '.join(sys.argv[1:]))
table = table.add('om({})'.format(label),o.as_axisangle(options.degrees), scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -187,6 +187,6 @@ for name in filenames:
np.einsum('ijk,ik->ij',slip_normal, (o@normal)))
for i,label in enumerate(labels):
table.add(label,S[:,i],scriptID+' '+' '.join(sys.argv[1:]))
table = table.add(label,S[:,i],scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -1,142 +0,0 @@
#!/usr/bin/env python3
import os
import sys
from optparse import OptionParser, OptionGroup
import math # noqa
import numpy as np
import damask
def periodicAverage(coords, limits):
"""Centroid in periodic domain, see https://en.wikipedia.org/wiki/Center_of_mass#Systems_with_periodic_boundary_conditions."""
theta = 2.0*np.pi * (coords - limits[0])/(limits[1] - limits[0])
theta_avg = np.pi + np.arctan2(-np.sin(theta).mean(axis=0), -np.cos(theta).mean(axis=0))
return limits[0] + theta_avg * (limits[1] - limits[0])/2.0/np.pi
scriptName = os.path.splitext(os.path.basename(__file__))[0]
scriptID = ' '.join([scriptName,damask.version])
# --------------------------------------------------------------------
# MAIN
# --------------------------------------------------------------------
parser = OptionParser(option_class=damask.extendableOption, usage='%prog options [ASCIItable(s)]', description = """
Apply a user-specified function to condense into a single row all those rows for which columns 'label' have identical values.
Output table will contain as many rows as there are different (unique) values in the grouping column(s).
Periodic domain averaging of coordinate values is supported.
Examples:
For grain averaged values, replace all rows of particular 'texture' with a single row containing their average.
{name} --label texture --function np.average data.txt
""".format(name = scriptName), version = scriptID)
parser.add_option('-l','--label',
dest = 'label',
action = 'extend', metavar = '<string LIST>',
help = 'column label(s) for grouping rows')
parser.add_option('-f','--function',
dest = 'function',
type = 'string', metavar = 'string',
help = 'mapping function [%default]')
parser.add_option('-a','--all',
dest = 'all',
action = 'store_true',
help = 'apply mapping function also to grouping column(s)')
group = OptionGroup(parser, "periodic averaging", "")
group.add_option ('-p','--periodic',
dest = 'periodic',
action = 'extend', metavar = '<string LIST>',
help = 'coordinate label(s) to average across periodic domain')
group.add_option ('--limits',
dest = 'boundary',
type = 'float', metavar = 'float float', nargs = 2,
help = 'min and max of periodic domain %default')
parser.add_option_group(group)
parser.set_defaults(function = 'np.average',
all = False,
label = [],
boundary = [0.0, 1.0])
(options,filenames) = parser.parse_args()
funcModule,funcName = options.function.split('.')
try:
mapFunction = getattr(locals().get(funcModule) or
globals().get(funcModule) or
__import__(funcModule),
funcName)
except Exception:
mapFunction = None
if options.label is []:
parser.error('no grouping column specified.')
if not hasattr(mapFunction,'__call__'):
parser.error('function "{}" is not callable.'.format(options.function))
# --- loop over input files -------------------------------------------------------------------------
if filenames == []: filenames = [None]
for name in filenames:
try:
table = damask.ASCIItable(name = name)
except IOError:
continue
damask.util.report(scriptName,name)
# ------------------------------------------ sanity checks ---------------------------------------
remarks = []
errors = []
table.head_read()
grpColumns = table.label_index(options.label)[::-1]
grpColumns = grpColumns[np.where(grpColumns>=0)]
if len(grpColumns) == 0: errors.append('no valid grouping column present.')
if remarks != []: damask.util.croak(remarks)
if errors != []:
damask.util.croak(errors)
table.close(dismiss=True)
continue
# ------------------------------------------ assemble info ---------------------------------------
table.info_append(scriptID + '\t' + ' '.join(sys.argv[1:]))
table.head_write()
# ------------------------------------------ process data --------------------------------
table.data_readArray()
indexrange = table.label_indexrange(options.periodic) if options.periodic is not None else []
rows,cols = table.data.shape
table.data = table.data[np.lexsort(table.data[:,grpColumns].T)] # sort data by grpColumn(s)
values,index = np.unique(table.data[:,grpColumns], axis=0, return_index=True) # unique grpColumn values and their positions
index = sorted(np.append(index,rows)) # add termination position
grpTable = np.empty((len(values), cols)) # initialize output
for i in range(len(values)): # iterate over groups (unique values in grpColumn)
grpTable[i] = np.apply_along_axis(mapFunction,0,table.data[index[i]:index[i+1]]) # apply (general) mapping function
grpTable[i,indexrange] = \
periodicAverage(table.data[index[i]:index[i+1],indexrange],options.boundary) # apply periodicAverage mapping function
if not options.all: grpTable[i,grpColumns] = table.data[index[i],grpColumns] # restore grouping column value
table.data = grpTable
# ------------------------------------------ output result -------------------------------
table.data_writeArray()
table.close() # close ASCII table

View File

@ -56,6 +56,6 @@ for name in filenames:
data = table.get(label)
uniques,inverse = np.unique(data,return_inverse=True,axis=0) if options.unique else (data,np.arange(len(data)))
rng.shuffle(uniques)
table.set(label,uniques[inverse], scriptID+' '+' '.join(sys.argv[1:]))
table = table.set(label,uniques[inverse], scriptID+' '+' '.join(sys.argv[1:]))
table.to_file(sys.stdout if name is None else name)

View File

@ -64,6 +64,5 @@ for name in filenames:
'homogenization\t{}'.format(geom.homogenization)]
table = damask.Table(seeds[mask],{'pos':(3,)},comments)
table.add('microstructure',microstructure[mask])
table.to_file(sys.stdout if name is None else \
os.path.splitext(name)[0]+'.seeds')
table = table.add('microstructure',microstructure[mask])
table.to_file(sys.stdout if name is None else os.path.splitext(name)[0]+'.seeds')

View File

@ -155,11 +155,11 @@ for name in filenames:
]
table = damask.Table(np.hstack((seeds,eulers)),{'pos':(3,),'euler':(3,)},comments)
table.add('microstructure',np.arange(options.microstructure,options.microstructure + options.N,dtype=int))
table = table.add('microstructure',np.arange(options.microstructure,options.microstructure + options.N,dtype=int))
if options.weights:
weights = np.random.uniform(low = 0, high = options.max, size = options.N) if options.max > 0.0 \
else np.random.normal(loc = options.mean, scale = options.sigma, size = options.N)
table.add('weight',weights)
table = table.add('weight',weights)
table.to_file(sys.stdout if name is None else name)

View File

@ -434,10 +434,11 @@ class Rotation:
if P == 1: ax[...,0:3] *= -1
if degrees: ax[..., 3] = np.radians(ax[...,3])
if normalize: ax[...,0:3] /= np.linalg.norm(ax[...,0:3],axis=-1)
if normalize: ax[...,0:3] /= np.linalg.norm(ax[...,0:3],axis=-1,keepdims=True)
if np.any(ax[...,3] < 0.0) or np.any(ax[...,3] > np.pi):
raise ValueError('Axis angle rotation angle outside of [0..π].')
if not np.all(np.isclose(np.linalg.norm(ax[...,0:3],axis=-1), 1.0)):
print(np.linalg.norm(ax[...,0:3],axis=-1))
raise ValueError('Axis angle rotation axis is not of unit length.')
return Rotation(Rotation._ax2qu(ax))
@ -516,7 +517,7 @@ class Rotation:
raise ValueError('P ∉ {-1,1}')
if P == 1: ro[...,0:3] *= -1
if normalize: ro[...,0:3] /= np.linalg.norm(ro[...,0:3],axis=-1)
if normalize: ro[...,0:3] /= np.linalg.norm(ro[...,0:3],axis=-1,keepdims=True)
if np.any(ro[...,3] < 0.0):
raise ValueError('Rodrigues vector rotation angle not positive.')
if not np.all(np.isclose(np.linalg.norm(ro[...,0:3],axis=-1), 1.0)):
@ -613,13 +614,29 @@ class Rotation:
return Rotation.from_quaternion(np.real(vec.T[eig.argmax()]),accept_homomorph = True)
@staticmethod
def from_random(shape=None):
def from_random(shape=None,seed=None):
"""
Draw random rotation.
Rotations are uniformly distributed.
Parameters
----------
shape : tuple of ints, optional
Shape of the sample. Defaults to None which gives a
single rotation
seed : {None, int, array_like[ints], SeedSequence, BitGenerator, Generator}, optional
A seed to initialize the BitGenerator. Defaults to None.
If None, then fresh, unpredictable entropy will be pulled from the OS.
"""
rng = np.random.default_rng(seed)
if shape is None:
r = np.random.random(3)
r = rng.random(3)
elif hasattr(shape, '__iter__'):
r = np.random.random(tuple(shape)+(3,))
r = rng.random(tuple(shape)+(3,))
else:
r = np.random.rand(shape,3)
r = rng.random((shape,3))
A = np.sqrt(r[...,2])
B = np.sqrt(1.0-r[...,2])
@ -636,6 +653,84 @@ class Rotation:
asAxisAngle = as_axis_angle
__mul__ = __matmul__
@staticmethod
def from_spherical_component(center,sigma,N=500,degrees=True,seed=None):
"""
Calculate set of rotations with Gaussian distribution around center.
Parameters
----------
center : Rotation
Central Rotation.
sigma : float
Standard deviation of (Gaussian) misorientation distribution.
N : int, optional
Number of samples, defaults to 500.
degrees : boolean, optional
sigma is given in degrees.
seed : {None, int, array_like[ints], SeedSequence, BitGenerator, Generator}, optional
A seed to initialize the BitGenerator. Defaults to None, i.e. unpredictable entropy
will be pulled from the OS.
"""
rng = np.random.default_rng(seed)
sigma = np.radians(sigma) if degrees else sigma
u,Theta = (rng.random((N,2)) * 2.0 * np.array([1,np.pi]) - np.array([1.0, 0])).T
omega = abs(rng.normal(scale=sigma,size=N))
p = np.column_stack([np.sqrt(1-u**2)*np.cos(Theta),
np.sqrt(1-u**2)*np.sin(Theta),
u, omega])
return Rotation.from_axis_angle(p) @ center
@staticmethod
def from_fiber_component(alpha,beta,sigma=0.0,N=500,degrees=True,seed=None):
"""
Calculate set of rotations with Gaussian distribution around direction.
Parameters
----------
alpha : numpy.ndarray of size 2
Polar coordinates (phi from x,theta from z) of fiber direction in crystal frame.
beta : numpy.ndarray of size 2
Polar coordinates (phi from x,theta from z) of fiber direction in sample frame.
sigma : float, optional
Standard deviation of (Gaussian) misorientation distribution.
Defaults to 0.
N : int, optional
Number of samples, defaults to 500.
degrees : boolean, optional
sigma, alpha, and beta are given in degrees.
seed : {None, int, array_like[ints], SeedSequence, BitGenerator, Generator}, optional
A seed to initialize the BitGenerator. Defaults to None, i.e. unpredictable entropy
will be pulled from the OS.
"""
rng = np.random.default_rng(seed)
sigma_,alpha_,beta_ = map(np.radians,(sigma,alpha,beta)) if degrees else (sigma,alpha,beta)
d_cr = np.array([np.sin(alpha_[0])*np.cos(alpha_[1]), np.sin(alpha_[0])*np.sin(alpha_[1]), np.cos(alpha_[0])])
d_lab = np.array([np.sin( beta_[0])*np.cos( beta_[1]), np.sin( beta_[0])*np.sin( beta_[1]), np.cos( beta_[0])])
ax_align = np.append(np.cross(d_lab,d_cr), np.arccos(np.dot(d_lab,d_cr)))
if np.isclose(ax_align[3],0.0): ax_align[:3] = np.array([1,0,0])
R_align = Rotation.from_axis_angle(ax_align if ax_align[3] > 0.0 else -ax_align,normalize=True) # rotate fiber axis from sample to crystal frame
u,Theta = (rng.random((N,2)) * 2.0 * np.array([1,np.pi]) - np.array([1.0, 0])).T
omega = abs(rng.normal(scale=sigma_,size=N))
p = np.column_stack([np.sqrt(1-u**2)*np.cos(Theta),
np.sqrt(1-u**2)*np.sin(Theta),
u, omega])
p[:,:3] = np.einsum('ij,...j',np.eye(3)-np.outer(d_lab,d_lab),p[:,:3]) # remove component along fiber axis
f = np.column_stack((np.broadcast_to(d_lab,(N,3)),rng.random(N)*np.pi))
f[::2,:3] *= -1 # flip half the rotation axes to negative sense
return R_align.broadcast_to(N) \
@ Rotation.from_axis_angle(p,normalize=True) \
@ Rotation.from_axis_angle(f)
####################################################################################################
# Code below available according to the following conditions on https://github.com/MarDiehl/3Drotations
####################################################################################################

View File

@ -1,4 +1,5 @@
import re
import copy
import pandas as pd
import numpy as np
@ -29,6 +30,15 @@ class Table:
self._label_condensed()
def __copy__(self):
"""Copy Table."""
return copy.deepcopy(self)
def copy(self):
"""Copy Table."""
return self.__copy__()
def _label_flat(self):
"""Label data individually, e.g. v v v ==> 1_v 2_v 3_v."""
labels = []
@ -191,15 +201,16 @@ class Table:
Human-readable information about the new data.
"""
self._add_comment(label,data.shape[1:],info)
dup = self.copy()
dup._add_comment(label,data.shape[1:],info)
if re.match(r'[0-9]*?_',label):
idx,key = label.split('_',1)
iloc = self.data.columns.get_loc(key).tolist().index(True) + int(idx) -1
self.data.iloc[:,iloc] = data
iloc = dup.data.columns.get_loc(key).tolist().index(True) + int(idx) -1
dup.data.iloc[:,iloc] = data
else:
self.data[label] = data.reshape(self.data[label].shape)
dup.data[label] = data.reshape(dup.data[label].shape)
return dup
def add(self,label,data,info=None):
"""
@ -215,15 +226,17 @@ class Table:
Human-readable information about the modified data.
"""
self._add_comment(label,data.shape[1:],info)
dup = self.copy()
dup._add_comment(label,data.shape[1:],info)
self.shapes[label] = data.shape[1:] if len(data.shape) > 1 else (1,)
dup.shapes[label] = data.shape[1:] if len(data.shape) > 1 else (1,)
size = np.prod(data.shape[1:],dtype=int)
new = pd.DataFrame(data=data.reshape(-1,size),
columns=[label]*size,
)
new.index = self.data.index
self.data = pd.concat([self.data,new],axis=1)
new.index = dup.data.index
dup.data = pd.concat([dup.data,new],axis=1)
return dup
def delete(self,label):
@ -236,25 +249,31 @@ class Table:
Column label.
"""
self.data.drop(columns=label,inplace=True)
del self.shapes[label]
dup = self.copy()
dup.data.drop(columns=label,inplace=True)
del dup.shapes[label]
return dup
def rename(self,label_old,label_new,info=None):
def rename(self,old,new,info=None):
"""
Rename column data.
Parameters
----------
label_old : str
Old column label.
label_new : str
New column label.
label_old : str or iterable of str
Old column label(s).
label_new : str or iterable of str
New column label(s).
"""
self.data.rename(columns={label_old:label_new},inplace=True)
self.comments.append(f'{label_old} => {label_new}'+('' if info is None else f': {info}'))
self.shapes = {(label if label != label_old else label_new):self.shapes[label] for label in self.shapes}
dup = self.copy()
columns = dict(zip([old] if isinstance(old,str) else old,
[new] if isinstance(new,str) else new))
dup.data.rename(columns=columns,inplace=True)
dup.comments.append(f'{old} => {new}'+('' if info is None else f': {info}'))
dup.shapes = {(label if label not in columns else columns[label]):dup.shapes[label] for label in dup.shapes}
return dup
def sort_by(self,labels,ascending=True):
@ -269,10 +288,12 @@ class Table:
Set sort order.
"""
self._label_flat()
self.data.sort_values(labels,axis=0,inplace=True,ascending=ascending)
self._label_condensed()
self.comments.append(f'sorted by [{", ".join(labels)}]')
dup = self.copy()
dup._label_flat()
dup.data.sort_values(labels,axis=0,inplace=True,ascending=ascending)
dup._label_condensed()
dup.comments.append(f'sorted {"ascending" if ascending else "descending"} by {labels}')
return dup
def append(self,other):
@ -290,7 +311,9 @@ class Table:
if self.shapes != other.shapes or not self.data.columns.equals(other.data.columns):
raise KeyError('Labels or shapes or order do not match')
else:
self.data = self.data.append(other.data,ignore_index=True)
dup = self.copy()
dup.data = dup.data.append(other.data,ignore_index=True)
return dup
def join(self,other):
@ -308,9 +331,11 @@ class Table:
if set(self.shapes) & set(other.shapes) or self.data.shape[0] != other.data.shape[0]:
raise KeyError('Dublicated keys or row count mismatch')
else:
self.data = self.data.join(other.data)
dup = self.copy()
dup.data = dup.data.join(other.data)
for key in other.shapes:
self.shapes[key] = other.shapes[key]
dup.shapes[key] = other.shapes[key]
return dup
def to_file(self,fname,format='ASCII',new_style=False):

View File

@ -105,7 +105,7 @@ class TestOrientation:
if update:
coords = np.array([(1,i+1) for i,x in enumerate(eu)])
table = Table(eu,{'Eulers':(3,)})
table.add('pos',coords)
table = table.add('pos',coords)
table.to_ASCII(reference)
assert np.allclose(eu,Table.from_ASCII(reference).get('Eulers'))

View File

@ -2,6 +2,7 @@ import os
import pytest
import numpy as np
from scipy import stats
from damask import Rotation
from damask import _rotation
@ -686,6 +687,15 @@ class TestRotation:
print(u,c)
assert np.allclose(single(u),c) and np.allclose(single(u),vectorized(u))
@pytest.mark.parametrize('func',[Rotation.from_axis_angle])
def test_normalization_vectorization(self,func):
"""Check vectorized implementation normalization."""
vec = np.random.rand(5,4)
ori = func(vec,normalize=True)
for v,o in zip(vec,ori):
assert np.allclose(func(v,normalize=True).as_quaternion(),o.as_quaternion())
@pytest.mark.parametrize('degrees',[True,False])
def test_Eulers(self,set_of_rotations,degrees):
for rot in set_of_rotations:
@ -911,3 +921,39 @@ class TestRotation:
R_2 = Rotation.from_axis_angle([0,0,1,angle],degrees=True)
avg_angle = R_1.average(R_2).as_axis_angle(degrees=True,pair=True)[1]
assert np.isclose(avg_angle,10+(angle-10)/2.)
@pytest.mark.parametrize('sigma',[5,10,15,20])
@pytest.mark.parametrize('N',[1000,10000,100000])
def test_spherical_component(self,N,sigma):
c = Rotation.from_random()
o = Rotation.from_spherical_component(c,sigma,N)
_, angles = c.misorientation(o).as_axis_angle(pair=True,degrees=True)
angles[::2] *= -1 # flip angle for every second to symmetrize distribution
p = stats.normaltest(angles)[1]
sigma_out = np.std(angles)
print(f'\np: {p}, sigma ratio {sigma/sigma_out}')
assert (.9 < sigma/sigma_out < 1.1) and p > 0.001
@pytest.mark.parametrize('sigma',[5,10,15,20])
@pytest.mark.parametrize('N',[1000,10000,100000])
def test_from_fiber_component(self,N,sigma):
"""https://en.wikipedia.org/wiki/Full_width_at_half_maximum."""
alpha = np.random.random(2)*np.pi
beta = np.random.random(2)*np.pi
f_in_C = np.array([np.sin(alpha[0])*np.cos(alpha[1]), np.sin(alpha[0])*np.sin(alpha[1]), np.cos(alpha[0])])
f_in_S = np.array([np.sin(beta[0] )*np.cos(beta[1] ), np.sin(beta[0] )*np.sin(beta[1] ), np.cos(beta[0] )])
ax = np.append(np.cross(f_in_C,f_in_S), - np.arccos(np.dot(f_in_C,f_in_S)))
n = Rotation.from_axis_angle(ax if ax[3] > 0.0 else ax*-1.0 ,normalize=True) # rotation to align fiber axis in crystal and sample system
o = Rotation.from_fiber_component(alpha,beta,np.radians(sigma),N,False)
angles = np.arccos(np.clip(np.dot(o@np.broadcast_to(f_in_S,(N,3)),n@f_in_S),-1,1))
dist = np.array(angles) * (np.random.randint(0,2,N)*2-1)
p = stats.normaltest(dist)[1]
sigma_out = np.degrees(np.std(dist))
print(f'\np: {p}, sigma ratio {sigma/sigma_out}')
assert (.9 < sigma/sigma_out < 1.1) and p > 0.001

View File

@ -81,13 +81,11 @@ class TestTable:
Table.from_ASCII(f)
def test_set(self,default):
default.set('F',np.zeros((5,3,3)),'set to zero')
d=default.get('F')
d = default.set('F',np.zeros((5,3,3)),'set to zero').get('F')
assert np.allclose(d,0.0) and d.shape[1:] == (3,3)
def test_set_component(self,default):
default.set('1_F',np.zeros((5)),'set to zero')
d=default.get('F')
d = default.set('1_F',np.zeros((5)),'set to zero').get('F')
assert np.allclose(d[...,0,0],0.0) and d.shape[1:] == (3,3)
def test_labels(self,default):
@ -95,36 +93,34 @@ class TestTable:
def test_add(self,default):
d = np.random.random((5,9))
default.add('nine',d,'random data')
assert np.allclose(d,default.get('nine'))
assert np.allclose(d,default.add('nine',d,'random data').get('nine'))
def test_rename_equivalent(self):
x = np.random.random((5,13))
t = Table(x,{'F':(3,3),'v':(3,),'s':(1,)},['random test data'])
s = t.get('s')
t.rename('s','u')
u = t.get('u')
u = t.rename('s','u').get('u')
assert np.all(s == u)
def test_rename_gone(self,default):
default.rename('v','V')
assert 'v' not in default.shapes and 'v' not in default.data.columns
gone = default.rename('v','V')
assert 'v' not in gone.shapes and 'v' not in gone.data.columns
with pytest.raises(KeyError):
default.get('v')
gone.get('v')
def test_delete(self,default):
default.delete('v')
assert 'v' not in default.shapes and 'v' not in default.data.columns
delete = default.delete('v')
assert 'v' not in delete.shapes and 'v' not in delete.data.columns
with pytest.raises(KeyError):
default.get('v')
delete.get('v')
def test_join(self):
x = np.random.random((5,13))
a = Table(x,{'F':(3,3),'v':(3,),'s':(1,)},['random test data'])
y = np.random.random((5,3))
b = Table(y,{'u':(3,)},['random test data'])
a.join(b)
assert np.array_equal(a.get('u'), b.get('u'))
c = a.join(b)
assert np.array_equal(c.get('u'), b.get('u'))
def test_join_invalid(self):
x = np.random.random((5,13))
@ -135,8 +131,8 @@ class TestTable:
def test_append(self):
x = np.random.random((5,13))
a = Table(x,{'F':(3,3),'v':(3,),'s':(1,)},['random test data'])
a.append(a)
assert np.array_equal(a.data[:5].to_numpy(),a.data[5:].to_numpy())
b = a.append(a)
assert np.array_equal(b.data[:5].to_numpy(),b.data[5:].to_numpy())
def test_append_invalid(self):
x = np.random.random((5,13))
@ -163,29 +159,26 @@ class TestTable:
x = np.random.random((5,13))
t = Table(x,{'F':(3,3),'v':(3,),'s':(1,)},['random test data'])
unsort = t.get('s')
t.sort_by('s')
sort = t.get('s')
sort = t.sort_by('s').get('s')
assert np.all(np.sort(unsort,0)==sort)
def test_sort_component(self):
x = np.random.random((5,12))
t = Table(x,{'F':(3,3),'v':(3,)},['random test data'])
unsort = t.get('4_F')
t.sort_by('4_F')
sort = t.get('4_F')
sort = t.sort_by('4_F').get('4_F')
assert np.all(np.sort(unsort,0)==sort)
def test_sort_revert(self):
x = np.random.random((5,12))
t = Table(x,{'F':(3,3),'v':(3,)},['random test data'])
t.sort_by('4_F',ascending=False)
sort = t.get('4_F')
sort = t.sort_by('4_F',ascending=False).get('4_F')
assert np.all(np.sort(sort,0)==sort[::-1,:])
def test_sort(self):
t = Table(np.array([[0,1,],[2,1,]]),
{'v':(2,)},
['test data'])
t.add('s',np.array(['b','a']))
t.sort_by('s')
['test data'])\
.add('s',np.array(['b','a']))\
.sort_by('s')
assert np.all(t.get('1_v') == np.array([2,0]).reshape(2,1))

View File

@ -48,7 +48,7 @@ module CPFEM
end type tNumerics
type(tNumerics), private :: num
type, private :: tDebugOptions
logical :: &
basic, &
@ -58,9 +58,9 @@ module CPFEM
element, &
ip
end type tDebugOptions
type(tDebugOptions), private :: debugCPFEM
public :: &
CPFEM_general, &
CPFEM_initAll, &
@ -74,6 +74,7 @@ contains
!--------------------------------------------------------------------------------------------------
subroutine CPFEM_initAll
call parallelization_init
call DAMASK_interface_init
call prec_init
call IO_init
@ -81,7 +82,7 @@ subroutine CPFEM_initAll
call math_init
call rotations_init
call YAML_types_init
call YAML_init
call YAML_parse_init
call HDF5_utilities_init
call results_init(.false.)
call discretization_marc_init
@ -105,33 +106,32 @@ subroutine CPFEM_init
num_commercialFEM, &
debug_CPFEM
write(6,'(/,a)') ' <<<+- CPFEM init -+>>>'
flush(6)
print'(/,a)', ' <<<+- CPFEM init -+>>>'; flush(6)
allocate(CPFEM_cs( 6,discretization_nIP,discretization_nElem), source= 0.0_pReal)
allocate(CPFEM_dcsdE( 6,6,discretization_nIP,discretization_nElem), source= 0.0_pReal)
allocate(CPFEM_dcsdE_knownGood(6,6,discretization_nIP,discretization_nElem), source= 0.0_pReal)
!------------------------------------------------------------------------------
! read numerical parameters and do sanity check
num_commercialFEM => numerics_root%get('commercialFEM',defaultVal=emptyDict)
! read numerical parameters and do sanity check
num_commercialFEM => config_numerics%get('commercialFEM',defaultVal=emptyDict)
num%iJacoStiffness = num_commercialFEM%get_asInt('ijacostiffness',defaultVal=1)
if (num%iJacoStiffness < 1) call IO_error(301,ext_msg='iJacoStiffness')
!------------------------------------------------------------------------------
! read debug options
! read debug options
debug_CPFEM => debug_root%get('cpfem',defaultVal=emptyList)
debug_CPFEM => config_debug%get('cpfem',defaultVal=emptyList)
debugCPFEM%basic = debug_CPFEM%contains('basic')
debugCPFEM%extensive = debug_CPFEM%contains('extensive')
debugCPFEM%selective = debug_CPFEM%contains('selective')
debugCPFEM%element = debug_root%get_asInt('element',defaultVal = 1)
debugCPFEM%ip = debug_root%get_asInt('integrationpoint',defaultVal = 1)
debugCPFEM%element = config_debug%get_asInt('element',defaultVal = 1)
debugCPFEM%ip = config_debug%get_asInt('integrationpoint',defaultVal = 1)
if(debugCPFEM%basic) then
write(6,'(a32,1x,6(i8,1x))') 'CPFEM_cs: ', shape(CPFEM_cs)
write(6,'(a32,1x,6(i8,1x))') 'CPFEM_dcsdE: ', shape(CPFEM_dcsdE)
write(6,'(a32,1x,6(i8,1x),/)') 'CPFEM_dcsdE_knownGood: ', shape(CPFEM_dcsdE_knownGood)
print'(a32,1x,6(i8,1x))', 'CPFEM_cs: ', shape(CPFEM_cs)
print'(a32,1x,6(i8,1x))', 'CPFEM_dcsdE: ', shape(CPFEM_dcsdE)
print'(a32,1x,6(i8,1x),/)', 'CPFEM_dcsdE_knownGood: ', shape(CPFEM_dcsdE_knownGood)
flush(6)
endif
@ -170,14 +170,14 @@ subroutine CPFEM_general(mode, ffn, ffn1, temperature_inp, dt, elFE, ip, cauchyS
elCP = mesh_FEM2DAMASK_elem(elFE)
if (debugCPFEM%basic .and. elCP == debugCPFEM%element .and. ip == debugCPFEM%ip) then
write(6,'(/,a)') '#############################################'
write(6,'(a1,a22,1x,i8,a13)') '#','element', elCP, '#'
write(6,'(a1,a22,1x,i8,a13)') '#','ip', ip, '#'
write(6,'(a1,a22,1x,i8,a13)') '#','cycleCounter', cycleCounter, '#'
write(6,'(a1,a22,1x,i8,a13)') '#','computationMode',mode, '#'
print'(/,a)', '#############################################'
print'(a1,a22,1x,i8,a13)', '#','element', elCP, '#'
print'(a1,a22,1x,i8,a13)', '#','ip', ip, '#'
print'(a1,a22,1x,i8,a13)', '#','cycleCounter', cycleCounter, '#'
print'(a1,a22,1x,i8,a13)', '#','computationMode',mode, '#'
if (terminallyIll) &
write(6,'(a,/)') '# --- terminallyIll --- #'
write(6,'(a,/)') '#############################################'; flush (6)
print'(a,/)', '# --- terminallyIll --- #'
print'(a,/)', '#############################################'; flush (6)
endif
if (iand(mode, CPFEM_BACKUPJACOBIAN) /= 0_pInt) &
@ -201,14 +201,14 @@ subroutine CPFEM_general(mode, ffn, ffn1, temperature_inp, dt, elFE, ip, cauchyS
call random_number(rnd)
if (rnd < 0.5_pReal) rnd = rnd - 1.0_pReal
CPFEM_cs(1:6,ip,elCP) = ODD_STRESS * rnd
CPFEM_dcsde(1:6,1:6,ip,elCP) = ODD_JACOBIAN * math_identity2nd(6)
CPFEM_dcsde(1:6,1:6,ip,elCP) = ODD_JACOBIAN * math_eye(6)
else validCalculation
updateJaco = mod(cycleCounter,num%iJacoStiffness) == 0
FEsolving_execElem = elCP
FEsolving_execIP = ip
if (debugCPFEM%extensive) &
write(6,'(a,i8,1x,i2)') '<< CPFEM >> calculation for elFE ip ',elFE,ip
print'(a,i8,1x,i2)', '<< CPFEM >> calculation for elFE ip ',elFE,ip
call materialpoint_stressAndItsTangent(updateJaco, dt)
terminalIllness: if (terminallyIll) then
@ -216,7 +216,7 @@ subroutine CPFEM_general(mode, ffn, ffn1, temperature_inp, dt, elFE, ip, cauchyS
call random_number(rnd)
if (rnd < 0.5_pReal) rnd = rnd - 1.0_pReal
CPFEM_cs(1:6,ip,elCP) = ODD_STRESS * rnd
CPFEM_dcsde(1:6,1:6,ip,elCP) = ODD_JACOBIAN * math_identity2nd(6)
CPFEM_dcsde(1:6,1:6,ip,elCP) = ODD_JACOBIAN * math_eye(6)
else terminalIllness
@ -246,9 +246,9 @@ subroutine CPFEM_general(mode, ffn, ffn1, temperature_inp, dt, elFE, ip, cauchyS
if (debugCPFEM%extensive &
.and. ((debugCPFEM%element == elCP .and. debugCPFEM%ip == ip) .or. .not. debugCPFEM%selective)) then
write(6,'(a,i8,1x,i2,/,12x,6(f10.3,1x)/)') &
print'(a,i8,1x,i2,/,12x,6(f10.3,1x)/)', &
'<< CPFEM >> stress/MPa at elFE ip ', elFE, ip, CPFEM_cs(1:6,ip,elCP)*1.0e-6_pReal
write(6,'(a,i8,1x,i2,/,6(12x,6(f10.3,1x)/))') &
print'(a,i8,1x,i2,/,6(12x,6(f10.3,1x)/))', &
'<< CPFEM >> Jacobian/GPa at elFE ip ', elFE, ip, transpose(CPFEM_dcsdE(1:6,1:6,ip,elCP))*1.0e-9_pReal
flush(6)
endif

View File

@ -40,6 +40,7 @@ contains
!--------------------------------------------------------------------------------------------------
subroutine CPFEM_initAll
call parallelization_init
call DAMASK_interface_init ! Spectral and FEM interface to commandline
call prec_init
call IO_init
@ -51,7 +52,7 @@ subroutine CPFEM_initAll
call math_init
call rotations_init
call YAML_types_init
call YAML_init
call YAML_parse_init
call lattice_init
call HDF5_utilities_init
call results_init(restart=interface_restartInc>0)
@ -75,7 +76,7 @@ end subroutine CPFEM_initAll
!--------------------------------------------------------------------------------------------------
subroutine CPFEM_init
write(6,'(/,a)') ' <<<+- CPFEM init -+>>>'; flush(6)
print'(/,a)', ' <<<+- CPFEM init -+>>>'; flush(6)
if (interface_restartInc > 0) call crystallite_restartRead

View File

@ -19,26 +19,27 @@ module DAMASK_interface
use PETScSys
use prec
use parallelization
use system_routines
implicit none
private
logical, volatile, public, protected :: &
SIGTERM, & !< termination signal
SIGUSR1, & !< 1. user-defined signal
SIGUSR2 !< 2. user-defined signal
interface_SIGTERM, & !< termination signal
interface_SIGUSR1, & !< 1. user-defined signal
interface_SIGUSR2 !< 2. user-defined signal
integer, public, protected :: &
interface_restartInc = 0 !< Increment at which calculation starts
character(len=:), allocatable, public, protected :: &
geometryFile, & !< parameter given for geometry file
loadCaseFile !< parameter given for load case file
interface_geomFile, & !< parameter given for geometry file
interface_loadFile !< parameter given for load case file
public :: &
getSolverJobName, &
DAMASK_interface_init, &
setSIGTERM, &
setSIGUSR1, &
setSIGUSR2
interface_setSIGTERM, &
interface_setSIGUSR1, &
interface_setSIGUSR2
contains
@ -72,144 +73,94 @@ subroutine DAMASK_interface_init
userName !< name of user calling the executable
integer :: &
stat, &
i, &
#ifdef _OPENMP
threadLevel, &
#endif
worldrank = 0, &
worldsize = 0, &
typeSize
i
integer, dimension(8) :: &
dateAndTime
integer :: err
PetscErrorCode :: petsc_err
external :: &
quit
print'(/,a)', ' <<<+- DAMASK_interface init -+>>>'
open(6, encoding='UTF-8') ! for special characters in output
!--------------------------------------------------------------------------------------------------
! PETSc Init
#ifdef _OPENMP
! If openMP is enabled, check if the MPI libary supports it and initialize accordingly.
! Otherwise, the first call to PETSc will do the initialization.
call MPI_Init_Thread(MPI_THREAD_FUNNELED,threadLevel,err)
if (err /= 0) call quit(1)
if (threadLevel<MPI_THREAD_FUNNELED) then
write(6,'(/,a)') ' ERROR: MPI library does not support OpenMP'
call quit(1)
endif
! http://patorjk.com/software/taag/#p=display&f=Lean&t=DAMASK%203
#ifdef DEBUG
print*, achar(27)//'[31m'
print'(a,/)', ' debug version - debug version - debug version - debug version - debug version'
#else
print*, achar(27)//'[94m'
#endif
call PETScInitializeNoArguments(petsc_err) ! according to PETSc manual, that should be the first line in the code
CHKERRQ(petsc_err) ! this is a macro definition, it is case sensitive
print*, ' _/_/_/ _/_/ _/ _/ _/_/ _/_/_/ _/ _/ _/_/_/'
print*, ' _/ _/ _/ _/ _/_/ _/_/ _/ _/ _/ _/ _/ _/'
print*, ' _/ _/ _/_/_/_/ _/ _/ _/ _/_/_/_/ _/_/ _/_/ _/_/'
print*, ' _/ _/ _/ _/ _/ _/ _/ _/ _/ _/ _/ _/'
print*, ' _/_/_/ _/ _/ _/ _/ _/ _/ _/_/_/ _/ _/ _/_/_/'
#ifdef DEBUG
print'(/,a)', ' debug version - debug version - debug version - debug version - debug version'
#endif
print*, achar(27)//'[0m'
call MPI_Comm_rank(PETSC_COMM_WORLD,worldrank,err)
if (err /= 0) call quit(1)
call MPI_Comm_size(PETSC_COMM_WORLD,worldsize,err)
if (err /= 0) call quit(1)
print'(a)', ' Roters et al., Computational Materials Science 158:420478, 2019'
print'(a)', ' https://doi.org/10.1016/j.commatsci.2018.04.030'
mainProcess: if (worldrank == 0) then
if (output_unit /= 6) then
write(output_unit,'(/,a)') ' ERROR: STDOUT != 6'
call quit(1)
endif
if (error_unit /= 0) then
write(output_unit,'(/,a)') ' ERROR: STDERR != 0'
call quit(1)
endif
else mainProcess
close(6) ! disable output for non-master processes (open 6 to rank specific file for debug)
open(6,file='/dev/null',status='replace') ! close(6) alone will leave some temp files in cwd
endif mainProcess
write(6,'(/,a)') ' <<<+- DAMASK_interface init -+>>>'
! http://patorjk.com/software/taag/#p=display&f=Lean&t=DAMASK
write(6,*) achar(27)//'[94m'
write(6,*) ' _/_/_/ _/_/ _/ _/ _/_/ _/_/_/ _/ _/'
write(6,*) ' _/ _/ _/ _/ _/_/ _/_/ _/ _/ _/ _/ _/'
write(6,*) ' _/ _/ _/_/_/_/ _/ _/ _/ _/_/_/_/ _/_/ _/_/'
write(6,*) ' _/ _/ _/ _/ _/ _/ _/ _/ _/ _/ _/'
write(6,*) ' _/_/_/ _/ _/ _/ _/ _/ _/ _/_/_/ _/ _/'
write(6,*) achar(27)//'[0m'
write(6,'(/,a)') ' Roters et al., Computational Materials Science 158:420478, 2019'
write(6,'(a)') ' https://doi.org/10.1016/j.commatsci.2018.04.030'
write(6,'(/,a)') ' Version: '//DAMASKVERSION
print'(/,a)', ' Version: '//DAMASKVERSION
! https://github.com/jeffhammond/HPCInfo/blob/master/docs/Preprocessor-Macros.md
#if defined(__GFORTRAN__) || __INTEL_COMPILER >= 1800
write(6,'(/,a)') ' Compiled with: '//compiler_version()
write(6,'(a)') ' Compiler options: '//compiler_options()
#elif defined(__INTEL_COMPILER)
write(6,'(/,a,i4.4,a,i8.8)') ' Compiled with Intel fortran version :', __INTEL_COMPILER,&
', build date :', __INTEL_COMPILER_BUILD_DATE
#elif defined(__PGI)
write(6,'(a,i4.4,a,i8.8)') ' Compiled with PGI fortran version :', __PGIC__,&
'.', __PGIC_MINOR__
#if defined(__PGI)
print'(/,a,i4.4,a,i8.8)', ' Compiled with PGI fortran version :', __PGIC__,&
'.', __PGIC_MINOR__
#else
print'(/,a)', ' Compiled with: '//compiler_version()
print'(a)', ' Compiler options: '//compiler_options()
#endif
write(6,'(/,a)') ' Compiled on: '//__DATE__//' at '//__TIME__
print'(/,a)', ' Compiled on: '//__DATE__//' at '//__TIME__
call date_and_time(values = dateAndTime)
write(6,'(/,a,2(i2.2,a),i4.4)') ' Date: ',dateAndTime(3),'/',dateAndTime(2),'/', dateAndTime(1)
write(6,'(a,2(i2.2,a),i2.2)') ' Time: ',dateAndTime(5),':', dateAndTime(6),':', dateAndTime(7)
call MPI_Type_size(MPI_INTEGER,typeSize,err)
if (err /= 0) call quit(1)
if (typeSize*8 /= bit_size(0)) then
write(6,'(a)') ' Mismatch between MPI and DAMASK integer'
call quit(1)
endif
call MPI_Type_size(MPI_DOUBLE,typeSize,err)
if (err /= 0) call quit(1)
if (typeSize*8 /= storage_size(0.0_pReal)) then
write(6,'(a)') ' Mismatch between MPI and DAMASK real'
call quit(1)
endif
print'(/,a,2(i2.2,a),i4.4)', ' Date: ',dateAndTime(3),'/',dateAndTime(2),'/', dateAndTime(1)
print'(a,2(i2.2,a),i2.2)', ' Time: ',dateAndTime(5),':', dateAndTime(6),':', dateAndTime(7)
do i = 1, command_argument_count()
call get_command_argument(i,arg,status=err)
if (err /= 0) call quit(1)
select case(trim(arg)) ! extract key
case ('-h','--help')
write(6,'(a)') ' #######################################################################'
write(6,'(a)') ' DAMASK Command Line Interface:'
write(6,'(a)') ' For PETSc-based solvers for the Düsseldorf Advanced Material Simulation Kit'
write(6,'(a,/)')' #######################################################################'
write(6,'(a,/)')' Valid command line switches:'
write(6,'(a)') ' --geom (-g, --geometry)'
write(6,'(a)') ' --load (-l, --loadcase)'
write(6,'(a)') ' --workingdir (-w, --wd, --workingdirectory)'
write(6,'(a)') ' --restart (-r, --rs)'
write(6,'(a)') ' --help (-h)'
write(6,'(/,a)')' -----------------------------------------------------------------------'
write(6,'(a)') ' Mandatory arguments:'
write(6,'(/,a)')' --geom PathToGeomFile/NameOfGeom'
write(6,'(a)') ' Specifies the location of the geometry definition file.'
write(6,'(/,a)')' --load PathToLoadFile/NameOfLoadFile'
write(6,'(a)') ' Specifies the location of the load case definition file.'
write(6,'(/,a)')' -----------------------------------------------------------------------'
write(6,'(a)') ' Optional arguments:'
write(6,'(/,a)')' --workingdirectory PathToWorkingDirectory'
write(6,'(a)') ' Specifies the working directory and overwrites the default ./'
write(6,'(a)') ' Make sure the file "material.config" exists in the working'
write(6,'(a)') ' directory.'
write(6,'(a)') ' For further configuration place "numerics.config"'
write(6,'(a)')' and "debug.config" in that directory.'
write(6,'(/,a)')' --restart N'
write(6,'(a)') ' Reads in increment N and continues with calculating'
write(6,'(a)') ' increment N+1 based on this.'
write(6,'(a)') ' Appends to existing results file'
write(6,'(a)') ' "NameOfGeom_NameOfLoadFile.hdf5".'
write(6,'(a)') ' Works only if the restart information for increment N'
write(6,'(a)') ' is available in the working directory.'
write(6,'(/,a)')' -----------------------------------------------------------------------'
write(6,'(a)') ' Help:'
write(6,'(/,a)')' --help'
write(6,'(a,/)')' Prints this message and exits'
print'(a)', ' #######################################################################'
print'(a)', ' DAMASK Command Line Interface:'
print'(a)', ' For PETSc-based solvers for the Düsseldorf Advanced Material Simulation Kit'
print'(a,/)',' #######################################################################'
print'(a,/)',' Valid command line switches:'
print'(a)', ' --geom (-g, --geometry)'
print'(a)', ' --load (-l, --loadcase)'
print'(a)', ' --workingdir (-w, --wd, --workingdirectory)'
print'(a)', ' --restart (-r, --rs)'
print'(a)', ' --help (-h)'
print'(/,a)',' -----------------------------------------------------------------------'
print'(a)', ' Mandatory arguments:'
print'(/,a)',' --geom PathToGeomFile/NameOfGeom'
print'(a)', ' Specifies the location of the geometry definition file.'
print'(/,a)',' --load PathToLoadFile/NameOfLoadFile'
print'(a)', ' Specifies the location of the load case definition file.'
print'(/,a)',' -----------------------------------------------------------------------'
print'(a)', ' Optional arguments:'
print'(/,a)',' --workingdirectory PathToWorkingDirectory'
print'(a)', ' Specifies the working directory and overwrites the default ./'
print'(a)', ' Make sure the file "material.config" exists in the working'
print'(a)', ' directory.'
print'(a)', ' For further configuration place "numerics.config"'
print'(a)',' and "debug.config" in that directory.'
print'(/,a)',' --restart N'
print'(a)', ' Reads in increment N and continues with calculating'
print'(a)', ' increment N+1 based on this.'
print'(a)', ' Appends to existing results file'
print'(a)', ' "NameOfGeom_NameOfLoadFile.hdf5".'
print'(a)', ' Works only if the restart information for increment N'
print'(a)', ' is available in the working directory.'
print'(/,a)',' -----------------------------------------------------------------------'
print'(a)', ' Help:'
print'(/,a)',' --help'
print'(a,/)',' Prints this message and exits'
call quit(0) ! normal Termination
case ('-l', '--load', '--loadcase')
call get_command_argument(i+1,loadCaseArg,status=err)
@ -221,7 +172,7 @@ subroutine DAMASK_interface_init
call get_command_argument(i+1,arg,status=err)
read(arg,*,iostat=stat) interface_restartInc
if (interface_restartInc < 0 .or. stat /=0) then
write(6,'(/,a)') ' ERROR: Could not parse restart increment: '//trim(arg)
print'(/,a)', ' ERROR: Could not parse restart increment: '//trim(arg)
call quit(1)
endif
end select
@ -229,40 +180,38 @@ subroutine DAMASK_interface_init
enddo
if (len_trim(loadcaseArg) == 0 .or. len_trim(geometryArg) == 0) then
write(6,'(/,a)') ' ERROR: Please specify geometry AND load case (-h for help)'
print'(/,a)', ' ERROR: Please specify geometry AND load case (-h for help)'
call quit(1)
endif
if (len_trim(workingDirArg) > 0) call setWorkingDirectory(trim(workingDirArg))
geometryFile = getGeometryFile(geometryArg)
loadCaseFile = getLoadCaseFile(loadCaseArg)
interface_geomFile = getGeometryFile(geometryArg)
interface_loadFile = getLoadCaseFile(loadCaseArg)
call get_command(commandLine)
call get_environment_variable('USER',userName)
! ToDo: https://stackoverflow.com/questions/8953424/how-to-get-the-username-in-c-c-in-linux
write(6,'(/,a,i4.1)') ' MPI processes: ',worldsize
write(6,'(a,a)') ' Host name: ', trim(getHostName())
write(6,'(a,a)') ' User name: ', trim(userName)
print'(a)', ' Host name: '//trim(getHostName())
print'(a)', ' User name: '//trim(userName)
write(6,'(/a,a)') ' Command line call: ', trim(commandLine)
print'(/a)', ' Command line call: '//trim(commandLine)
if (len_trim(workingDirArg) > 0) &
write(6,'(a,a)') ' Working dir argument: ', trim(workingDirArg)
write(6,'(a,a)') ' Geometry argument: ', trim(geometryArg)
write(6,'(a,a)') ' Load case argument: ', trim(loadcaseArg)
write(6,'(a,a)') ' Working directory: ', getCWD()
write(6,'(a,a)') ' Geometry file: ', geometryFile
write(6,'(a,a)') ' Loadcase file: ', loadCaseFile
write(6,'(a,a)') ' Solver job name: ', getSolverJobName()
print'(a)', ' Working dir argument: '//trim(workingDirArg)
print'(a)', ' Geometry argument: '//trim(geometryArg)
print'(a)', ' Load case argument: '//trim(loadcaseArg)
print'(a)', ' Working directory: '//getCWD()
print'(a)', ' Geometry file: '//interface_geomFile
print'(a)', ' Loadcase file: '//interface_loadFile
print'(a)', ' Solver job name: '//getSolverJobName()
if (interface_restartInc > 0) &
write(6,'(a,i6.6)') ' Restart from increment: ', interface_restartInc
print'(a,i6.6)', ' Restart from increment: ', interface_restartInc
!call signalterm_c(c_funloc(catchSIGTERM))
call signalusr1_c(c_funloc(catchSIGUSR1))
call signalusr2_c(c_funloc(catchSIGUSR2))
call setSIGTERM(.false.)
call setSIGUSR1(.false.)
call setSIGUSR2(.false.)
call interface_setSIGTERM(.false.)
call interface_setSIGUSR1(.false.)
call interface_setSIGUSR2(.false.)
end subroutine DAMASK_interface_init
@ -288,7 +237,7 @@ subroutine setWorkingDirectory(workingDirectoryArg)
workingDirectory = trim(rectifyPath(workingDirectory))
error = setCWD(trim(workingDirectory))
if(error) then
write(6,'(/,a)') ' ERROR: Invalid Working directory: '//trim(workingDirectory)
print*, 'ERROR: Invalid Working directory: '//trim(workingDirectory)
call quit(1)
endif
@ -303,15 +252,15 @@ function getSolverJobName()
character(len=:), allocatable :: getSolverJobName
integer :: posExt,posSep
posExt = scan(geometryFile,'.',back=.true.)
posSep = scan(geometryFile,'/',back=.true.)
posExt = scan(interface_geomFile,'.',back=.true.)
posSep = scan(interface_geomFile,'/',back=.true.)
getSolverJobName = geometryFile(posSep+1:posExt-1)
getSolverJobName = interface_geomFile(posSep+1:posExt-1)
posExt = scan(loadCaseFile,'.',back=.true.)
posSep = scan(loadCaseFile,'/',back=.true.)
posExt = scan(interface_loadFile,'.',back=.true.)
posSep = scan(interface_loadFile,'/',back=.true.)
getSolverJobName = getSolverJobName//'_'//loadCaseFile(posSep+1:posExt-1)
getSolverJobName = getSolverJobName//'_'//interface_loadFile(posSep+1:posExt-1)
end function getSolverJobName
@ -332,7 +281,7 @@ function getGeometryFile(geometryParameter)
inquire(file=getGeometryFile, exist=file_exists)
if (.not. file_exists) then
write(6,'(/,a)') ' ERROR: Geometry file does not exists ('//trim(getGeometryFile)//')'
print*, 'ERROR: Geometry file does not exists: '//trim(getGeometryFile)
call quit(1)
endif
@ -355,7 +304,7 @@ function getLoadCaseFile(loadCaseParameter)
inquire(file=getLoadCaseFile, exist=file_exists)
if (.not. file_exists) then
write(6,'(/,a)') ' ERROR: Load case file does not exists ('//trim(getLoadCaseFile)//')'
print*, 'ERROR: Load case file does not exists: '//trim(getLoadCaseFile)
call quit(1)
endif
@ -438,75 +387,78 @@ end function makeRelativePath
!--------------------------------------------------------------------------------------------------
!> @brief sets global variable SIGTERM to .true.
!> @brief Set global variable interface_SIGTERM to .true.
!> @details This function can be registered to catch signals send to the executable.
!--------------------------------------------------------------------------------------------------
subroutine catchSIGTERM(signal) bind(C)
integer(C_INT), value :: signal
SIGTERM = .true.
interface_SIGTERM = .true.
write(6,'(a,i2.2,a)') ' received signal ',signal, ', set SIGTERM'
print'(a,i2.2,a)', ' received signal ',signal, ', set SIGTERM=TRUE'
end subroutine catchSIGTERM
!--------------------------------------------------------------------------------------------------
!> @brief sets global variable SIGTERM
!> @brief Set global variable interface_SIGTERM.
!--------------------------------------------------------------------------------------------------
subroutine setSIGTERM(state)
subroutine interface_setSIGTERM(state)
logical, intent(in) :: state
SIGTERM = state
interface_SIGTERM = state
end subroutine setSIGTERM
end subroutine interface_setSIGTERM
!--------------------------------------------------------------------------------------------------
!> @brief sets global variable SIGUSR1 to .true.
!> @brief Set global variable interface_SIGUSR1 to .true.
!> @details This function can be registered to catch signals send to the executable.
!--------------------------------------------------------------------------------------------------
subroutine catchSIGUSR1(signal) bind(C)
integer(C_INT), value :: signal
SIGUSR1 = .true.
interface_SIGUSR1 = .true.
write(6,'(a,i2.2,a)') ' received signal ',signal, ', set SIGUSR1'
print'(a,i2.2,a)', ' received signal ',signal, ', set SIGUSR1=TRUE'
end subroutine catchSIGUSR1
!--------------------------------------------------------------------------------------------------
!> @brief sets global variable SIGUSR1
!> @brief Set global variable interface_SIGUSR.
!--------------------------------------------------------------------------------------------------
subroutine setSIGUSR1(state)
subroutine interface_setSIGUSR1(state)
logical, intent(in) :: state
SIGUSR1 = state
interface_SIGUSR1 = state
end subroutine setSIGUSR1
end subroutine interface_setSIGUSR1
!--------------------------------------------------------------------------------------------------
!> @brief sets global variable SIGUSR2 to .true. if program receives SIGUSR2
!> @brief Set global variable interface_SIGUSR2 to .true.
!> @details This function can be registered to catch signals send to the executable.
!--------------------------------------------------------------------------------------------------
subroutine catchSIGUSR2(signal) bind(C)
integer(C_INT), value :: signal
SIGUSR2 = .true.
interface_SIGUSR2 = .true.
write(6,'(a,i2.2,a)') ' received signal ',signal, ', set SIGUSR2'
print'(a,i2.2,a)', ' received signal ',signal, ', set SIGUSR2=TRUE'
end subroutine catchSIGUSR2
!--------------------------------------------------------------------------------------------------
!> @brief sets global variable SIGUSR2
!> @brief Set global variable interface_SIGUSR2.
!--------------------------------------------------------------------------------------------------
subroutine setSIGUSR2(state)
subroutine interface_setSIGUSR2(state)
logical, intent(in) :: state
SIGUSR2 = state
interface_SIGUSR2 = state
end subroutine setSIGUSR2
end subroutine interface_setSIGUSR2
end module

View File

@ -42,7 +42,6 @@ module DAMASK_interface
logical, protected, public :: symmetricSolver
character(len=*), parameter, public :: INPUTFILEEXTENSION = '.dat'
public :: &
DAMASK_interface_init, &
@ -59,33 +58,33 @@ subroutine DAMASK_interface_init
integer :: ierr
character(len=pPathLen) :: wd
write(6,'(/,a)') ' <<<+- DAMASK_marc init -+>>>'
print'(/,a)', ' <<<+- DAMASK_marc init -+>>>'
write(6,'(/,a)') ' Roters et al., Computational Materials Science 158:420478, 2019'
write(6,'(a)') ' https://doi.org/10.1016/j.commatsci.2018.04.030'
print*, 'Roters et al., Computational Materials Science 158:420478, 2019'
print*, 'https://doi.org/10.1016/j.commatsci.2018.04.030'
write(6,'(/,a)') ' Version: '//DAMASKVERSION
print'(/,a)', ' Version: '//DAMASKVERSION
! https://github.com/jeffhammond/HPCInfo/blob/master/docs/Preprocessor-Macros.md
#if __INTEL_COMPILER >= 1800
write(6,'(/,a)') ' Compiled with: '//compiler_version()
write(6,'(a)') ' Compiler options: '//compiler_options()
print'(/,a)', ' Compiled with: '//compiler_version()
print'(a)', ' Compiler options: '//compiler_options()
#else
write(6,'(/,a,i4.4,a,i8.8)') ' Compiled with Intel fortran version :', __INTEL_COMPILER,&
print'(/,a,i4.4,a,i8.8)', ' Compiled with Intel fortran version :', __INTEL_COMPILER,&
', build date :', __INTEL_COMPILER_BUILD_DATE
#endif
write(6,'(/,a)') ' Compiled on: '//__DATE__//' at '//__TIME__
print'(/,a)', ' Compiled on: '//__DATE__//' at '//__TIME__
call date_and_time(values = dateAndTime)
write(6,'(/,a,2(i2.2,a),i4.4)') ' Date: ',dateAndTime(3),'/',dateAndTime(2),'/', dateAndTime(1)
write(6,'(a,2(i2.2,a),i2.2)') ' Time: ',dateAndTime(5),':', dateAndTime(6),':', dateAndTime(7)
print'(/,a,2(i2.2,a),i4.4)', ' Date: ',dateAndTime(3),'/',dateAndTime(2),'/', dateAndTime(1)
print'(a,2(i2.2,a),i2.2)', ' Time: ',dateAndTime(5),':', dateAndTime(6),':', dateAndTime(7)
inquire(5, name=wd)
wd = wd(1:scan(wd,'/',back=.true.))
ierr = CHDIR(wd)
if (ierr /= 0) then
write(6,'(a20,a,a16)') ' working directory "',trim(wd),'" does not exist'
print*, 'working directory "'//trim(wd)//'" does not exist'
call quit(1)
endif
symmetricSolver = solverIsSymmetric()
@ -258,14 +257,14 @@ subroutine hypela2(d,g,e,de,s,t,dt,ngens,m,nn,kcus,matus,ndi,nshear,disp, &
debug_Marc ! pointer to Marc debug options
if(debug_basic) then
write(6,'(a,/,i8,i8,i2)') ' MSC.MARC information on shape of element(2), IP:', m, nn
write(6,'(a,2(i1))') ' Jacobian: ', ngens,ngens
write(6,'(a,i1)') ' Direct stress: ', ndi
write(6,'(a,i1)') ' Shear stress: ', nshear
write(6,'(a,i2)') ' DoF: ', ndeg
write(6,'(a,i2)') ' Coordinates: ', ncrd
write(6,'(a,i12)') ' Nodes: ', nnode
write(6,'(a,i1)') ' Deformation gradient: ', itel
print'(a,/,i8,i8,i2)', ' MSC.MARC information on shape of element(2), IP:', m, nn
print'(a,2(i1))', ' Jacobian: ', ngens,ngens
print'(a,i1)', ' Direct stress: ', ndi
print'(a,i1)', ' Shear stress: ', nshear
print'(a,i2)', ' DoF: ', ndeg
print'(a,i2)', ' Coordinates: ', ncrd
print'(a,i12)', ' Nodes: ', nnode
print'(a,i1)', ' Deformation gradient: ', itel
write(6,'(/,a,/,3(3(f12.7,1x)/))',advance='no') ' Deformation gradient at t=n:', &
transpose(ffn)
write(6,'(/,a,/,3(3(f12.7,1x)/))',advance='no') ' Deformation gradient at t=n+1:', &
@ -278,7 +277,7 @@ subroutine hypela2(d,g,e,de,s,t,dt,ngens,m,nn,kcus,matus,ndi,nshear,disp, &
if (.not. CPFEM_init_done) then
CPFEM_init_done = .true.
call CPFEM_initAll
debug_Marc => debug_root%get('marc',defaultVal=emptyList)
debug_Marc => config_debug%get('marc',defaultVal=emptyList)
debug_basic = debug_Marc%contains('basic')
endif
@ -296,22 +295,22 @@ subroutine hypela2(d,g,e,de,s,t,dt,ngens,m,nn,kcus,matus,ndi,nshear,disp, &
lastIncConverged = .false.
outdatedByNewInc = .false.
lastLovl = lovl ! pretend that this is NOT the first after a lovl change
write(6,'(a,i6,1x,i2)') '<< HYPELA2 >> start of analysis..! ',m(1),nn
print'(a,i6,1x,i2)', '<< HYPELA2 >> start of analysis..! ',m(1),nn
else if (inc - theInc > 1) then ! >> restart of broken analysis <<
lastIncConverged = .false.
outdatedByNewInc = .false.
write(6,'(a,i6,1x,i2)') '<< HYPELA2 >> restart of analysis..! ',m(1),nn
print'(a,i6,1x,i2)', '<< HYPELA2 >> restart of analysis..! ',m(1),nn
else ! >> just the next inc <<
lastIncConverged = .true.
outdatedByNewInc = .true.
write(6,'(a,i6,1x,i2)') '<< HYPELA2 >> new increment..! ',m(1),nn
print'(a,i6,1x,i2)', '<< HYPELA2 >> new increment..! ',m(1),nn
endif
else if ( timinc < theDelta ) then ! >> cutBack <<
lastIncConverged = .false.
outdatedByNewInc = .false.
terminallyIll = .false.
cycleCounter = -1 ! first calc step increments this to cycle = 0
write(6,'(a,i6,1x,i2)') '<< HYPELA2 >> cutback detected..! ',m(1),nn
print'(a,i6,1x,i2)', '<< HYPELA2 >> cutback detected..! ',m(1),nn
endif ! convergence treatment end
flush(6)

View File

@ -11,9 +11,9 @@ module HDF5_utilities
#endif
use prec
use parallelization
use IO
use rotations
use config
implicit none
public
@ -88,7 +88,7 @@ subroutine HDF5_utilities_init
integer :: hdferr
integer(SIZE_T) :: typeSize
write(6,'(/,a)') ' <<<+- HDF5_Utilities init -+>>>'
print'(/,a)', ' <<<+- HDF5_Utilities init -+>>>'
!--------------------------------------------------------------------------------------------------
!initialize HDF5 library and check if integer and float type size match
@ -98,12 +98,12 @@ subroutine HDF5_utilities_init
call h5tget_size_f(H5T_NATIVE_INTEGER,typeSize, hdferr)
if (hdferr < 0) call IO_error(1,ext_msg='HDF5_Utilities_init: h5tget_size_f (int)')
if (int(bit_size(0),SIZE_T)/=typeSize*8) &
call IO_error(0,ext_msg='Default integer size does not match H5T_NATIVE_INTEGER')
error stop 'Default integer size does not match H5T_NATIVE_INTEGER'
call h5tget_size_f(H5T_NATIVE_DOUBLE,typeSize, hdferr)
if (hdferr < 0) call IO_error(1,ext_msg='HDF5_Utilities_init: h5tget_size_f (double)')
if (int(storage_size(0.0_pReal),SIZE_T)/=typeSize*8) &
call IO_error(0,ext_msg='pReal does not match H5T_NATIVE_DOUBLE')
error stop 'pReal does not match H5T_NATIVE_DOUBLE'
end subroutine HDF5_utilities_init

View File

@ -22,19 +22,11 @@ module IO
'───────────────────'//&
'────────────'
! Obsolete alias
interface IO_read_ASCII
module procedure IO_readlines
end interface IO_read_ASCII
public :: &
IO_init, &
IO_read, &
IO_readlines, &
IO_read_ASCII, &
IO_open_binary, &
IO_isBlank, &
IO_getTag, &
IO_stringPos, &
IO_stringValue, &
IO_intValue, &
@ -51,11 +43,11 @@ contains
!--------------------------------------------------------------------------------------------------
!> @brief do self test
!> @brief Do self test.
!--------------------------------------------------------------------------------------------------
subroutine IO_init
write(6,'(/,a)') ' <<<+- IO init -+>>>'; flush(6)
print'(/,a)', ' <<<+- IO init -+>>>'; flush(6)
call selfTest
@ -63,7 +55,7 @@ end subroutine IO_init
!--------------------------------------------------------------------------------------------------
!> @brief read ASCII file and split at EOL
!> @brief Read ASCII file and split at EOL.
!--------------------------------------------------------------------------------------------------
function IO_readlines(fileName) result(fileContent)
@ -114,7 +106,7 @@ end function IO_readlines
!--------------------------------------------------------------------------------------------------
!> @brief read ASCII file into a string
!> @brief Read whole file.
!> @details ensures that the string ends with a new line (expected UNIX behavior)
!--------------------------------------------------------------------------------------------------
function IO_read(fileName) result(fileContent)
@ -146,40 +138,7 @@ end function IO_read
!--------------------------------------------------------------------------------------------------
!> @brief opens an existing file for reading or a new file for writing.
!> @details replaces an existing file when writing
!--------------------------------------------------------------------------------------------------
integer function IO_open_binary(fileName,mode)
character(len=*), intent(in) :: fileName
character, intent(in), optional :: mode
character :: m
integer :: ierr
if (present(mode)) then
m = mode
else
m = 'r'
endif
if (m == 'w') then
open(newunit=IO_open_binary, file=trim(fileName),&
status='replace',access='stream',action='write',iostat=ierr)
if (ierr /= 0) call IO_error(100,ext_msg='could not open file (w): '//trim(fileName))
elseif(m == 'r') then
open(newunit=IO_open_binary, file=trim(fileName),&
status='old', access='stream',action='read', iostat=ierr)
if (ierr /= 0) call IO_error(100,ext_msg='could not open file (r): '//trim(fileName))
else
call IO_error(100,ext_msg='unknown access mode: '//m)
endif
end function IO_open_binary
!--------------------------------------------------------------------------------------------------
!> @brief identifies strings without content
!> @brief Identifiy strings without content.
!--------------------------------------------------------------------------------------------------
logical pure function IO_isBlank(string)
@ -194,34 +153,8 @@ end function IO_isBlank
!--------------------------------------------------------------------------------------------------
!> @brief get tagged content of string
!--------------------------------------------------------------------------------------------------
pure function IO_getTag(string,openChar,closeChar)
character(len=*), intent(in) :: string !< string to check for tag
character, intent(in) :: openChar, & !< indicates beginning of tag
closeChar !< indicates end of tag
character(len=:), allocatable :: IO_getTag
integer :: left,right
left = scan(string,openChar)
right = merge(scan(string,closeChar), &
left + merge(scan(string(left+1:),openChar),0,len(string) > left), &
openChar /= closeChar)
foundTag: if (left == verify(string,IO_WHITESPACE) .and. right > left) then
IO_getTag = string(left+1:right-1)
else foundTag
IO_getTag = ''
endif foundTag
end function IO_getTag
!--------------------------------------------------------------------------------------------------
!> @brief locates all whitespace-separated chunks in given string and returns array containing
!! number them and the left/right position to be used by IO_xxxVal
!> @brief Locate all whitespace-separated chunks in given string and returns array containing
!! number them and the left/right position to be used by IO_xxxVal.
!! Array size is dynamically adjusted to number of chunks found in string
!! IMPORTANT: first element contains number of chunks!
!--------------------------------------------------------------------------------------------------
@ -251,7 +184,7 @@ end function IO_stringPos
!--------------------------------------------------------------------------------------------------
!> @brief reads string value at myChunk from string
!> @brief Read string value at myChunk from string.
!--------------------------------------------------------------------------------------------------
function IO_stringValue(string,chunkPos,myChunk)
@ -271,7 +204,7 @@ end function IO_stringValue
!--------------------------------------------------------------------------------------------------
!> @brief reads integer value at myChunk from string
!> @brief Read integer value at myChunk from string.
!--------------------------------------------------------------------------------------------------
integer function IO_intValue(string,chunkPos,myChunk)
@ -285,7 +218,7 @@ end function IO_intValue
!--------------------------------------------------------------------------------------------------
!> @brief reads float value at myChunk from string
!> @brief Read float value at myChunk from string.
!--------------------------------------------------------------------------------------------------
real(pReal) function IO_floatValue(string,chunkPos,myChunk)
@ -299,7 +232,7 @@ end function IO_floatValue
!--------------------------------------------------------------------------------------------------
!> @brief changes characters in string to lower case
!> @brief Convert characters in string to lower case.
!--------------------------------------------------------------------------------------------------
pure function IO_lc(string)
@ -324,7 +257,7 @@ end function IO_lc
!--------------------------------------------------------------------------------------------------
! @brief Remove comments (characters beyond '#') and trailing space
! @brief Remove comments (characters beyond '#') and trailing space.
! ToDo: Discuss name (the trim aspect is not clear)
!--------------------------------------------------------------------------------------------------
function IO_rmComment(line)
@ -345,7 +278,7 @@ end function IO_rmComment
!--------------------------------------------------------------------------------------------------
!> @brief return verified integer value in given string
!> @brief Return integer value from given string.
!--------------------------------------------------------------------------------------------------
integer function IO_stringAsInt(string)
@ -366,7 +299,7 @@ end function IO_stringAsInt
!--------------------------------------------------------------------------------------------------
!> @brief return verified float value in given string
!> @brief Return float value from given string.
!--------------------------------------------------------------------------------------------------
real(pReal) function IO_stringAsFloat(string)
@ -387,7 +320,7 @@ end function IO_stringAsFloat
!--------------------------------------------------------------------------------------------------
!> @brief return verified logical value in given string
!> @brief Return logical value from given string.
!--------------------------------------------------------------------------------------------------
logical function IO_stringAsBool(string)
@ -406,7 +339,7 @@ end function IO_stringAsBool
!--------------------------------------------------------------------------------------------------
!> @brief write error statements to standard out and terminate the Marc/spectral run with exit #9xxx
!> @brief Write error statements to standard out and terminate the run with exit #9xxx
!--------------------------------------------------------------------------------------------------
subroutine IO_error(error_ID,el,ip,g,instance,ext_msg)
@ -635,7 +568,7 @@ end subroutine IO_error
!--------------------------------------------------------------------------------------------------
!> @brief writes warning statement to standard out
!> @brief Write warning statement to standard out.
!--------------------------------------------------------------------------------------------------
subroutine IO_warning(warning_ID,el,ip,g,ext_msg)
@ -717,56 +650,56 @@ end subroutine IO_warning
!--------------------------------------------------------------------------------------------------
!> @brief check correctness of some IO functions
!> @brief Check correctness of some IO functions.
!--------------------------------------------------------------------------------------------------
subroutine selfTest
integer, dimension(:), allocatable :: chunkPos
character(len=:), allocatable :: str
if(dNeq(1.0_pReal, IO_stringAsFloat('1.0'))) call IO_error(0,ext_msg='IO_stringAsFloat')
if(dNeq(1.0_pReal, IO_stringAsFloat('1e0'))) call IO_error(0,ext_msg='IO_stringAsFloat')
if(dNeq(0.1_pReal, IO_stringAsFloat('1e-1'))) call IO_error(0,ext_msg='IO_stringAsFloat')
if(dNeq(1.0_pReal, IO_stringAsFloat('1.0'))) error stop 'IO_stringAsFloat'
if(dNeq(1.0_pReal, IO_stringAsFloat('1e0'))) error stop 'IO_stringAsFloat'
if(dNeq(0.1_pReal, IO_stringAsFloat('1e-1'))) error stop 'IO_stringAsFloat'
if(3112019 /= IO_stringAsInt( '3112019')) call IO_error(0,ext_msg='IO_stringAsInt')
if(3112019 /= IO_stringAsInt(' 3112019')) call IO_error(0,ext_msg='IO_stringAsInt')
if(-3112019 /= IO_stringAsInt('-3112019')) call IO_error(0,ext_msg='IO_stringAsInt')
if(3112019 /= IO_stringAsInt('+3112019 ')) call IO_error(0,ext_msg='IO_stringAsInt')
if(3112019 /= IO_stringAsInt( '3112019')) error stop 'IO_stringAsInt'
if(3112019 /= IO_stringAsInt(' 3112019')) error stop 'IO_stringAsInt'
if(-3112019 /= IO_stringAsInt('-3112019')) error stop 'IO_stringAsInt'
if(3112019 /= IO_stringAsInt('+3112019 ')) error stop 'IO_stringAsInt'
if(.not. IO_stringAsBool(' true')) call IO_error(0,ext_msg='IO_stringAsBool')
if(.not. IO_stringAsBool(' True ')) call IO_error(0,ext_msg='IO_stringAsBool')
if( IO_stringAsBool(' false')) call IO_error(0,ext_msg='IO_stringAsBool')
if( IO_stringAsBool('False')) call IO_error(0,ext_msg='IO_stringAsBool')
if(.not. IO_stringAsBool(' true')) error stop 'IO_stringAsBool'
if(.not. IO_stringAsBool(' True ')) error stop 'IO_stringAsBool'
if( IO_stringAsBool(' false')) error stop 'IO_stringAsBool'
if( IO_stringAsBool('False')) error stop 'IO_stringAsBool'
if(any([1,1,1] /= IO_stringPos('a'))) call IO_error(0,ext_msg='IO_stringPos')
if(any([2,2,3,5,5] /= IO_stringPos(' aa b'))) call IO_error(0,ext_msg='IO_stringPos')
if(any([1,1,1] /= IO_stringPos('a'))) error stop 'IO_stringPos'
if(any([2,2,3,5,5] /= IO_stringPos(' aa b'))) error stop 'IO_stringPos'
str=' 1.0 xxx'
chunkPos = IO_stringPos(str)
if(dNeq(1.0_pReal,IO_floatValue(str,chunkPos,1))) call IO_error(0,ext_msg='IO_floatValue')
if(dNeq(1.0_pReal,IO_floatValue(str,chunkPos,1))) error stop 'IO_floatValue'
str='M 3112019 F'
chunkPos = IO_stringPos(str)
if(3112019 /= IO_intValue(str,chunkPos,2)) call IO_error(0,ext_msg='IO_intValue')
if(3112019 /= IO_intValue(str,chunkPos,2)) error stop 'IO_intValue'
if(.not. IO_isBlank(' ')) call IO_error(0,ext_msg='IO_isBlank/1')
if(.not. IO_isBlank(' #isBlank')) call IO_error(0,ext_msg='IO_isBlank/2')
if( IO_isBlank(' i#s')) call IO_error(0,ext_msg='IO_isBlank/3')
if(.not. IO_isBlank(' ')) error stop 'IO_isBlank/1'
if(.not. IO_isBlank(' #isBlank')) error stop 'IO_isBlank/2'
if( IO_isBlank(' i#s')) error stop 'IO_isBlank/3'
str = IO_rmComment('#')
if (str /= '' .or. len(str) /= 0) call IO_error(0,ext_msg='IO_rmComment/1')
if (str /= '' .or. len(str) /= 0) error stop 'IO_rmComment/1'
str = IO_rmComment(' #')
if (str /= '' .or. len(str) /= 0) call IO_error(0,ext_msg='IO_rmComment/2')
if (str /= '' .or. len(str) /= 0) error stop 'IO_rmComment/2'
str = IO_rmComment(' # ')
if (str /= '' .or. len(str) /= 0) call IO_error(0,ext_msg='IO_rmComment/3')
if (str /= '' .or. len(str) /= 0) error stop 'IO_rmComment/3'
str = IO_rmComment(' # a')
if (str /= '' .or. len(str) /= 0) call IO_error(0,ext_msg='IO_rmComment/4')
if (str /= '' .or. len(str) /= 0) error stop 'IO_rmComment/4'
str = IO_rmComment(' # a')
if (str /= '' .or. len(str) /= 0) call IO_error(0,ext_msg='IO_rmComment/5')
if (str /= '' .or. len(str) /= 0) error stop 'IO_rmComment/5'
str = IO_rmComment(' a#')
if (str /= ' a' .or. len(str) /= 2) call IO_error(0,ext_msg='IO_rmComment/6')
if (str /= ' a' .or. len(str) /= 2) error stop 'IO_rmComment/6'
str = IO_rmComment(' ab #')
if (str /= ' ab'.or. len(str) /= 3) call IO_error(0,ext_msg='IO_rmComment/7')
if (str /= ' ab'.or. len(str) /= 3) error stop 'IO_rmComment/7'
end subroutine selfTest

View File

@ -11,27 +11,39 @@ module YAML_parse
implicit none
private
public :: &
YAML_init, &
parse_flow, &
to_flow
YAML_parse_init, &
YAML_parse_file
contains
!--------------------------------------------------------------------------------------------------
!> @brief do sanity checks
!> @brief Do sanity checks.
!--------------------------------------------------------------------------------------------------
subroutine YAML_init
subroutine YAML_parse_init
call selfTest
end subroutine YAML_init
end subroutine YAML_parse_init
!--------------------------------------------------------------------------------------------------
!> @brief Parse a YAML file into a a structure of nodes.
!--------------------------------------------------------------------------------------------------
function YAML_parse_file(fname) result(node)
character(len=*), intent(in) :: fname
class (tNode), pointer :: node
node => parse_flow(to_flow(IO_read(fname)))
end function YAML_parse_file
!--------------------------------------------------------------------------------------------------
!> @brief reads the flow style string and stores it in the form of dictionaries, lists and scalars.
!> @details A node type pointer can either point to a dictionary, list or scalar type entities.
!> @details A node type pointer can either point to a dictionary, list or scalar type entities.
!--------------------------------------------------------------------------------------------------
recursive function parse_flow(YAML_flow) result(node)
@ -47,7 +59,7 @@ recursive function parse_flow(YAML_flow) result(node)
e, & ! end position of dictionary or list
s, & ! start position of dictionary or list
d ! position of key: value separator (':')
flow_string = trim(adjustl(YAML_flow(:)))
if (len_trim(flow_string) == 0) then
node => emptyDict
@ -57,12 +69,12 @@ recursive function parse_flow(YAML_flow) result(node)
allocate(tDict::node)
do while (e < len_trim(flow_string))
s = e
d = s + scan(flow_string(s+1:),':')
e = d + find_end(flow_string(d+1:),'}')
d = s + scan(flow_string(s+1:),':')
e = d + find_end(flow_string(d+1:),'}')
key = trim(adjustl(flow_string(s+1:d-1)))
myVal => parse_flow(flow_string(d+1:e-1)) ! parse items (recursively)
select type (node)
class is (tDict)
call node%set(key,myVal)
@ -208,7 +220,7 @@ logical function isKey(line)
if(len(IO_rmComment(line)) == 0) then
isKey = .false.
else
isKey = IO_rmComment(line(len(IO_rmComment(line)):len(IO_rmComment(line)))) == ':' &
isKey = IO_rmComment(line(len(IO_rmComment(line)):len(IO_rmComment(line)))) == ':' &
.and. .not. isFlow(line)
endif
@ -224,19 +236,19 @@ recursive subroutine line_isFlow(flow,s_flow,line)
character(len=*), intent(inout) :: flow !< YAML in flow style only
integer, intent(inout) :: s_flow !< start position in flow
character(len=*), intent(in) :: line
integer :: &
s, &
list_chunk, &
dict_chunk
if(index(adjustl(line),'[') == 1) then
s = index(line,'[')
flow(s_flow:s_flow) = '['
s_flow = s_flow +1
do while(s < len_trim(line))
list_chunk = s + find_end(line(s+1:),']')
if(iskeyValue(line(s+1:list_chunk-1))) then
if(iskeyValue(line(s+1:list_chunk-1))) then
flow(s_flow:s_flow) = '{'
s_flow = s_flow +1
call keyValue_toFlow(flow,s_flow,line(s+1:list_chunk-1))
@ -252,10 +264,10 @@ recursive subroutine line_isFlow(flow,s_flow,line)
s = s + find_end(line(s+1:),']')
enddo
s_flow = s_flow - 1
if (flow(s_flow-1:s_flow-1) == ',') s_flow = s_flow - 1
if (flow(s_flow-1:s_flow-1) == ',') s_flow = s_flow - 1
flow(s_flow:s_flow) = ']'
s_flow = s_flow+1
elseif(index(adjustl(line),'{') == 1) then
s = index(line,'{')
flow(s_flow:s_flow) = '{'
@ -275,21 +287,21 @@ recursive subroutine line_isFlow(flow,s_flow,line)
else
call line_toFlow(flow,s_flow,line)
endif
end subroutine line_isFlow
!-------------------------------------------------------------------------------------------------
! @brief reads a line of YAML block of type <key>: <value> and places it in the YAML flow style structure
! @details Makes sure that the <value> is consistent with the input required in DAMASK YAML parser
! @details Makes sure that the <value> is consistent with the input required in DAMASK YAML parser
!-------------------------------------------------------------------------------------------------
recursive subroutine keyValue_toFlow(flow,s_flow,line)
character(len=*), intent(inout) :: flow !< YAML in flow style only
integer, intent(inout) :: s_flow !< start position in flow
character(len=*), intent(in) :: line
character(len=:), allocatable :: line_asStandard ! standard form of <key>: <value>
character(len=:), allocatable :: line_asStandard ! standard form of <key>: <value>
integer :: &
d_flow, &
col_pos, &
@ -318,7 +330,7 @@ subroutine line_toFlow(flow,s_flow,line)
character(len=*), intent(inout) :: flow !< YAML in flow style only
integer, intent(inout) :: s_flow !< start position in flow
character(len=*), intent(in) :: line
integer :: &
d_flow
@ -398,7 +410,7 @@ recursive subroutine lst(blck,flow,s_blck,s_flow,offset)
if(isScalar(line) .or. isFlow(line)) then
flow(s_flow:s_flow+1) = ', '
s_flow = s_flow +2
s_flow = s_flow + 2
endif
end do
@ -441,7 +453,7 @@ recursive subroutine dct(blck,flow,s_blck,s_flow,offset)
elseif(indentDepth(line,offset) < indent) then
if(isScalar(line) .or. isFlow(line) .and. previous_isKey) &
call IO_error(701,ext_msg=line)
offset = 0
offset = 0
exit ! job done (lower level)
elseif(indentDepth(line,offset) > indent .or. isListItem(line)) then
offset = 0
@ -455,20 +467,20 @@ recursive subroutine dct(blck,flow,s_blck,s_flow,offset)
flow(s_flow-1:s_flow) = ', '
s_flow = s_flow + 1
endif
if(isKeyValue(line)) then
call keyValue_toFlow(flow,s_flow,line)
else
call line_toFlow(flow,s_flow,line)
endif
s_blck = e_blck +2
end if
if(isScalar(line) .or. isKeyValue(line)) then
flow(s_flow:s_flow) = ','
s_flow = s_flow + 1
previous_isKey = .false.
previous_isKey = .false.
else
previous_isKey = .true.
endif
@ -540,7 +552,7 @@ function to_flow(blck)
s_flow, & !< start position in flow
offset, & !< counts leading '- ' in nested lists
end_line
if(isFlow(blck)) then
if(isFlow(blck)) then
to_flow = trim(adjustl(blck))
else
allocate(character(len=len(blck)*2)::to_flow)
@ -552,43 +564,45 @@ function to_flow(blck)
to_flow = trim(to_flow(:s_flow-1))
endif
end_line = index(to_flow,IO_EOL)
if(end_line > 0) to_flow = to_flow(:end_line-1)
if(end_line > 0) to_flow = to_flow(:end_line-1)
end function to_flow
!--------------------------------------------------------------------------------------------------
subroutine selfTest()
!> @brief Check correctness of some YAML functions.
!--------------------------------------------------------------------------------------------------
subroutine selfTest
if (indentDepth(' a') /= 1) call IO_error(0,ext_msg='indentDepth')
if (indentDepth('a') /= 0) call IO_error(0,ext_msg='indentDepth')
if (indentDepth('x ') /= 0) call IO_error(0,ext_msg='indentDepth')
if (indentDepth(' a') /= 1) error stop 'indentDepth'
if (indentDepth('a') /= 0) error stop 'indentDepth'
if (indentDepth('x ') /= 0) error stop 'indentDepth'
if ( isFlow(' a')) call IO_error(0,ext_msg='isFLow')
if (.not. isFlow('{')) call IO_error(0,ext_msg='isFlow')
if (.not. isFlow(' [')) call IO_error(0,ext_msg='isFlow')
if ( isFlow(' a')) error stop 'isFLow'
if (.not. isFlow('{')) error stop 'isFlow'
if (.not. isFlow(' [')) error stop 'isFlow'
if ( isListItem(' a')) call IO_error(0,ext_msg='isListItem')
if ( isListItem(' -b')) call IO_error(0,ext_msg='isListItem')
if (.not. isListItem('- a ')) call IO_error(0,ext_msg='isListItem')
if (.not. isListItem('- -a ')) call IO_error(0,ext_msg='isListItem')
if ( isListItem(' a')) error stop 'isListItem'
if ( isListItem(' -b')) error stop 'isListItem'
if (.not. isListItem('- a ')) error stop 'isListItem'
if (.not. isListItem('- -a ')) error stop 'isListItem'
if ( isKeyValue(' a')) call IO_error(0,ext_msg='isKeyValue')
if ( isKeyValue(' a: ')) call IO_error(0,ext_msg='isKeyValue')
if (.not. isKeyValue(' a: b')) call IO_error(0,ext_msg='isKeyValue')
if ( isKeyValue(' a')) error stop 'isKeyValue'
if ( isKeyValue(' a: ')) error stop 'isKeyValue'
if (.not. isKeyValue(' a: b')) error stop 'isKeyValue'
if ( isKey(' a')) call IO_error(0,ext_msg='isKey')
if ( isKey('{a:b}')) call IO_error(0,ext_msg='isKey')
if ( isKey(' a:b')) call IO_error(0,ext_msg='isKey')
if (.not. isKey(' a: ')) call IO_error(0,ext_msg='isKey')
if (.not. isKey(' a:')) call IO_error(0,ext_msg='isKey')
if (.not. isKey(' a: #')) call IO_error(0,ext_msg='isKey')
if ( isKey(' a')) error stop 'isKey'
if ( isKey('{a:b}')) error stop 'isKey'
if ( isKey(' a:b')) error stop 'isKey'
if (.not. isKey(' a: ')) error stop 'isKey'
if (.not. isKey(' a:')) error stop 'isKey'
if (.not. isKey(' a: #')) error stop 'isKey'
if( isScalar('a: ')) call IO_error(0,ext_msg='isScalar')
if( isScalar('a: b')) call IO_error(0,ext_msg='isScalar')
if( isScalar('{a:b}')) call IO_error(0,ext_msg='isScalar')
if( isScalar('- a:')) call IO_error(0,ext_msg='isScalar')
if(.not. isScalar(' a')) call IO_error(0,ext_msg='isScalar')
if( isScalar('a: ')) error stop 'isScalar'
if( isScalar('a: b')) error stop 'isScalar'
if( isScalar('{a:b}')) error stop 'isScalar'
if( isScalar('- a:')) error stop 'isScalar'
if(.not. isScalar(' a')) error stop 'isScalar'
basic_list: block
character(len=*), parameter :: block_list = &
@ -602,8 +616,8 @@ subroutine selfTest()
character(len=*), parameter :: flow_list = &
"[Casablanca, North by Northwest]"
if (.not. to_flow(block_list) == flow_list) call IO_error(0,ext_msg='to_flow')
if (.not. to_flow(block_list_newline) == flow_list) call IO_error(0,ext_msg='to_flow')
if (.not. to_flow(block_list) == flow_list) error stop 'to_flow'
if (.not. to_flow(block_list_newline) == flow_list) error stop 'to_flow'
end block basic_list
basic_dict: block
@ -618,10 +632,10 @@ subroutine selfTest()
character(len=*), parameter :: flow_dict = &
"{aa: Casablanca, bb: North by Northwest}"
if (.not. to_flow(block_dict) == flow_dict) call IO_error(0,ext_msg='to_flow')
if (.not. to_flow(block_dict_newline) == flow_dict) call IO_error(0,ext_msg='to_flow')
if (.not. to_flow(block_dict) == flow_dict) error stop 'to_flow'
if (.not. to_flow(block_dict_newline) == flow_dict) error stop 'to_flow'
end block basic_dict
basic_flow: block
character(len=*), parameter :: flow_braces = &
" source: [{param: 1}, {param: 2}, {param: 3}, {param: 4}]"//IO_EOL
@ -629,9 +643,9 @@ subroutine selfTest()
" source: [param: 1, {param: 2}, param: 3, {param: 4}]"//IO_EOL
character(len=*), parameter :: flow = &
"{source: [{param: 1}, {param: 2}, {param: 3}, {param: 4}]}"
if (.not. to_flow(flow_braces) == flow) call IO_error(0,ext_msg='to_flow')
if (.not. to_flow(flow_mixed_braces) == flow) call IO_error(0,ext_msg='to_flow')
if (.not. to_flow(flow_braces) == flow) error stop 'to_flow'
if (.not. to_flow(flow_mixed_braces) == flow) error stop 'to_flow'
end block basic_flow
basic_mixed: block
@ -644,8 +658,8 @@ subroutine selfTest()
" - {param_1: [{a: b}, c, {d: {e: [{f: g}, h]}}]}"//IO_EOL
character(len=*), parameter :: mixed_flow = &
"{aa: [{param_1: [{a: b}, c, {d: {e: [{f: g}, h]}}]}, {c: d}], bb: [{param_1: [{a: b}, c, {d: {e: [{f: g}, h]}}]}]}"
if(.not. to_flow(block_flow) == mixed_flow) call IO_error(0,ext_msg='to_flow')
if(.not. to_flow(block_flow) == mixed_flow) error stop 'to_flow'
end block basic_mixed
end subroutine selfTest

View File

@ -185,11 +185,11 @@ module YAML_types
contains
!--------------------------------------------------------------------------------------------------
!> @brief do sanity checks
!> @brief Do sanity checks.
!--------------------------------------------------------------------------------------------------
subroutine YAML_types_init
write(6,'(/,a)') ' <<<+- YAML_types init -+>>>'
print'(/,a)', ' <<<+- YAML_types init -+>>>'
call selfTest
@ -197,7 +197,7 @@ end subroutine YAML_types_init
!--------------------------------------------------------------------------------------------------
!> @brief check correctness of some type bound procedures
!> @brief Check correctness of some type bound procedures.
!--------------------------------------------------------------------------------------------------
subroutine selfTest
@ -207,11 +207,11 @@ subroutine selfTest
select type(s1)
class is(tScalar)
s1 = '1'
if(s1%asInt() /= 1) call IO_error(0,ext_msg='tScalar_asInt')
if(dNeq(s1%asFloat(),1.0_pReal)) call IO_error(0,ext_msg='tScalar_asFloat')
if(s1%asInt() /= 1) error stop 'tScalar_asInt'
if(dNeq(s1%asFloat(),1.0_pReal)) error stop 'tScalar_asFloat'
s1 = 'true'
if(.not. s1%asBool()) call IO_error(0,ext_msg='tScalar_asBool')
if(s1%asString() /= 'true') call IO_error(0,ext_msg='tScalar_asString')
if(.not. s1%asBool()) error stop 'tScalar_asBool'
if(s1%asString() /= 'true') error stop 'tScalar_asString'
end select
block
@ -232,18 +232,18 @@ subroutine selfTest
call l1%append(s1)
call l1%append(s2)
n => l1
if(any(l1%asInts() /= [2,3])) call IO_error(0,ext_msg='tList_asInts')
if(any(dNeq(l1%asFloats(),[2.0_pReal,3.0_pReal]))) call IO_error(0,ext_msg='tList_asFloats')
if(n%get_asInt(1) /= 2) call IO_error(0,ext_msg='byIndex_asInt')
if(dNeq(n%get_asFloat(2),3.0_pReal)) call IO_error(0,ext_msg='byIndex_asFloat')
if(any(l1%asInts() /= [2,3])) error stop 'tList_asInts'
if(any(dNeq(l1%asFloats(),[2.0_pReal,3.0_pReal]))) error stop 'tList_asFloats'
if(n%get_asInt(1) /= 2) error stop 'byIndex_asInt'
if(dNeq(n%get_asFloat(2),3.0_pReal)) error stop 'byIndex_asFloat'
endselect
allocate(tList::l2)
select type(l2)
class is(tList)
call l2%append(l1)
if(any(l2%get_asInts(1) /= [2,3])) call IO_error(0,ext_msg='byIndex_asInts')
if(any(dNeq(l2%get_asFloats(1),[2.0_pReal,3.0_pReal]))) call IO_error(0,ext_msg='byIndex_asFloats')
if(any(l2%get_asInts(1) /= [2,3])) error stop 'byIndex_asInts'
if(any(dNeq(l2%get_asFloats(1),[2.0_pReal,3.0_pReal]))) error stop 'byIndex_asFloats'
n => l2
end select
deallocate(n)
@ -265,10 +265,10 @@ subroutine selfTest
call l1%append(s2)
n => l1
if(any(l1%asBools() .neqv. [.true., .false.])) call IO_error(0,ext_msg='tList_asBools')
if(any(l1%asStrings() /= ['true ','False'])) call IO_error(0,ext_msg='tList_asStrings')
if(n%get_asBool(2)) call IO_error(0,ext_msg='byIndex_asBool')
if(n%get_asString(1) /= 'true') call IO_error(0,ext_msg='byIndex_asString')
if(any(l1%asBools() .neqv. [.true., .false.])) error stop 'tList_asBools'
if(any(l1%asStrings() /= ['true ','False'])) error stop 'tList_asStrings'
if(n%get_asBool(2)) error stop 'byIndex_asBool'
if(n%get_asString(1) /= 'true') error stop 'byIndex_asString'
end block
end subroutine selfTest

View File

@ -27,7 +27,7 @@ contains
!--------------------------------------------------------------------------------------------------
subroutine base64_init
write(6,'(/,a)') ' <<<+- base64 init -+>>>'; flush(6)
print'(/,a)', ' <<<+- base64 init -+>>>'; flush(6)
call selfTest
@ -167,59 +167,59 @@ subroutine selfTest
character(len=*), parameter :: zero_to_three = 'AAECAw=='
! https://en.wikipedia.org/wiki/Base64#Output_padding
if(base64_nChar(20_pI64) /= 28_pI64) call IO_error(0,ext_msg='base64_nChar/20/28')
if(base64_nChar(19_pI64) /= 28_pI64) call IO_error(0,ext_msg='base64_nChar/19/28')
if(base64_nChar(18_pI64) /= 24_pI64) call IO_error(0,ext_msg='base64_nChar/18/24')
if(base64_nChar(17_pI64) /= 24_pI64) call IO_error(0,ext_msg='base64_nChar/17/24')
if(base64_nChar(16_pI64) /= 24_pI64) call IO_error(0,ext_msg='base64_nChar/16/24')
if(base64_nChar(20_pI64) /= 28_pI64) error stop 'base64_nChar/20/28'
if(base64_nChar(19_pI64) /= 28_pI64) error stop 'base64_nChar/19/28'
if(base64_nChar(18_pI64) /= 24_pI64) error stop 'base64_nChar/18/24'
if(base64_nChar(17_pI64) /= 24_pI64) error stop 'base64_nChar/17/24'
if(base64_nChar(16_pI64) /= 24_pI64) error stop 'base64_nChar/16/24'
if(base64_nByte(4_pI64) /= 3_pI64) call IO_error(0,ext_msg='base64_nByte/4/3')
if(base64_nByte(8_pI64) /= 6_pI64) call IO_error(0,ext_msg='base64_nByte/8/6')
if(base64_nByte(4_pI64) /= 3_pI64) error stop 'base64_nByte/4/3'
if(base64_nByte(8_pI64) /= 6_pI64) error stop 'base64_nByte/8/6'
bytes = base64_to_bytes(zero_to_three)
if(any(bytes /= int([0,1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 4) call IO_error(0,ext_msg='base64_to_bytes//')
if(any(bytes /= int([0,1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 4) error stop 'base64_to_bytes//'
bytes = base64_to_bytes(zero_to_three,e=1_pI64)
if(any(bytes /= int([0],C_SIGNED_CHAR)) .or. size(bytes) /= 1) call IO_error(0,ext_msg='base64_to_bytes//1')
if(any(bytes /= int([0],C_SIGNED_CHAR)) .or. size(bytes) /= 1) error stop 'base64_to_bytes//1'
bytes = base64_to_bytes(zero_to_three,e=2_pI64)
if(any(bytes /= int([0,1],C_SIGNED_CHAR)) .or. size(bytes) /= 2) call IO_error(0,ext_msg='base64_to_bytes//2')
if(any(bytes /= int([0,1],C_SIGNED_CHAR)) .or. size(bytes) /= 2) error stop 'base64_to_bytes//2'
bytes = base64_to_bytes(zero_to_three,e=3_pI64)
if(any(bytes /= int([0,1,2],C_SIGNED_CHAR)) .or. size(bytes) /= 3) call IO_error(0,ext_msg='base64_to_bytes//3')
if(any(bytes /= int([0,1,2],C_SIGNED_CHAR)) .or. size(bytes) /= 3) error stop 'base64_to_bytes//3'
bytes = base64_to_bytes(zero_to_three,e=4_pI64)
if(any(bytes /= int([0,1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 4) call IO_error(0,ext_msg='base64_to_bytes//4')
if(any(bytes /= int([0,1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 4) error stop 'base64_to_bytes//4'
bytes = base64_to_bytes(zero_to_three,s=1_pI64)
if(any(bytes /= int([0,1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 4) call IO_error(0,ext_msg='base64_to_bytes/1/')
if(any(bytes /= int([0,1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 4) error stop 'base64_to_bytes/1/'
bytes = base64_to_bytes(zero_to_three,s=2_pI64)
if(any(bytes /= int([1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 3) call IO_error(0,ext_msg='base64_to_bytes/2/')
if(any(bytes /= int([1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 3) error stop 'base64_to_bytes/2/'
bytes = base64_to_bytes(zero_to_three,s=3_pI64)
if(any(bytes /= int([2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 2) call IO_error(0,ext_msg='base64_to_bytes/3/')
if(any(bytes /= int([2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 2) error stop 'base64_to_bytes/3/'
bytes = base64_to_bytes(zero_to_three,s=4_pI64)
if(any(bytes /= int([3],C_SIGNED_CHAR)) .or. size(bytes) /= 1) call IO_error(0,ext_msg='base64_to_bytes/4/')
if(any(bytes /= int([3],C_SIGNED_CHAR)) .or. size(bytes) /= 1) error stop 'base64_to_bytes/4/'
bytes = base64_to_bytes(zero_to_three,s=1_pI64,e=1_pI64)
if(any(bytes /= int([0],C_SIGNED_CHAR)) .or. size(bytes) /= 1) call IO_error(0,ext_msg='base64_to_bytes/1/1')
if(any(bytes /= int([0],C_SIGNED_CHAR)) .or. size(bytes) /= 1) error stop 'base64_to_bytes/1/1'
bytes = base64_to_bytes(zero_to_three,s=2_pI64,e=2_pI64)
if(any(bytes /= int([1],C_SIGNED_CHAR)) .or. size(bytes) /= 1) call IO_error(0,ext_msg='base64_to_bytes/2/2')
if(any(bytes /= int([1],C_SIGNED_CHAR)) .or. size(bytes) /= 1) error stop 'base64_to_bytes/2/2'
bytes = base64_to_bytes(zero_to_three,s=3_pI64,e=3_pI64)
if(any(bytes /= int([2],C_SIGNED_CHAR)) .or. size(bytes) /= 1) call IO_error(0,ext_msg='base64_to_bytes/3/3')
if(any(bytes /= int([2],C_SIGNED_CHAR)) .or. size(bytes) /= 1) error stop 'base64_to_bytes/3/3'
bytes = base64_to_bytes(zero_to_three,s=4_pI64,e=4_pI64)
if(any(bytes /= int([3],C_SIGNED_CHAR)) .or. size(bytes) /= 1) call IO_error(0,ext_msg='base64_to_bytes/4/4')
if(any(bytes /= int([3],C_SIGNED_CHAR)) .or. size(bytes) /= 1) error stop 'base64_to_bytes/4/4'
bytes = base64_to_bytes(zero_to_three,s=1_pI64,e=2_pI64)
if(any(bytes /= int([0,1],C_SIGNED_CHAR)) .or. size(bytes) /= 2) call IO_error(0,ext_msg='base64_to_bytes/1/2')
if(any(bytes /= int([0,1],C_SIGNED_CHAR)) .or. size(bytes) /= 2) error stop 'base64_to_bytes/1/2'
bytes = base64_to_bytes(zero_to_three,s=2_pI64,e=3_pI64)
if(any(bytes /= int([1,2],C_SIGNED_CHAR)) .or. size(bytes) /= 2) call IO_error(0,ext_msg='base64_to_bytes/2/3')
if(any(bytes /= int([1,2],C_SIGNED_CHAR)) .or. size(bytes) /= 2) error stop 'base64_to_bytes/2/3'
bytes = base64_to_bytes(zero_to_three,s=3_pI64,e=4_pI64)
if(any(bytes /= int([2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 2) call IO_error(0,ext_msg='base64_to_bytes/3/4')
if(any(bytes /= int([2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 2) error stop 'base64_to_bytes/3/4'
bytes = base64_to_bytes(zero_to_three,s=1_pI64,e=3_pI64)
if(any(bytes /= int([0,1,2],C_SIGNED_CHAR)) .or. size(bytes) /= 3) call IO_error(0,ext_msg='base64_to_bytes/1/3')
if(any(bytes /= int([0,1,2],C_SIGNED_CHAR)) .or. size(bytes) /= 3) error stop 'base64_to_bytes/1/3'
bytes = base64_to_bytes(zero_to_three,s=2_pI64,e=4_pI64)
if(any(bytes /= int([1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 3) call IO_error(0,ext_msg='base64_to_bytes/2/4')
if(any(bytes /= int([1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 3) error stop 'base64_to_bytes/2/4'
bytes = base64_to_bytes(zero_to_three,s=1_pI64,e=4_pI64)
if(any(bytes /= int([0,1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 4) call IO_error(0,ext_msg='base64_to_bytes/1/4')
if(any(bytes /= int([0,1,2,3],C_SIGNED_CHAR)) .or. size(bytes) /= 4) error stop 'base64_to_bytes/1/4'
end subroutine selfTest

View File

@ -3,6 +3,7 @@
!> @brief all DAMASK files without solver
!> @details List of files needed by MSC.Marc
!--------------------------------------------------------------------------------------------------
#include "parallelization.f90"
#include "IO.f90"
#include "YAML_types.f90"
#include "YAML_parse.f90"

View File

@ -2,7 +2,7 @@
!> @author Martin Diehl, Max-Planck-Institut für Eisenforschung GmbH
!> @brief Reads in the material, numerics & debug configuration from their respective file
!> @details Reads the material configuration file, where solverJobName.yaml takes
!! precedence over material.yaml.
!! precedence over material.yaml.
!--------------------------------------------------------------------------------------------------
module config
use prec
@ -15,22 +15,14 @@ module config
#include <petsc/finclude/petscsys.h>
use petscsys
#endif
!$ use OMP_LIB
implicit none
private
class(tNode), pointer, public :: &
material_root, &
numerics_root, &
debug_root
integer, protected, public :: &
worldrank = 0, & !< MPI worldrank (/=0 for MPI simulations only)
worldsize = 1 !< MPI worldsize (/=1 for MPI simulations only)
integer(4), protected, public :: &
DAMASK_NumThreadsInt = 0 !< value stored in environment variable DAMASK_NUM_THREADS, set to zero if no OpenMP directive
config_material, &
config_numerics, &
config_debug
public :: &
config_init, &
@ -39,27 +31,26 @@ module config
contains
!--------------------------------------------------------------------------------------------------
!> @brief calls subroutines that reads material, numerics and debug configuration files
!> @brief Real *.yaml configuration files.
!--------------------------------------------------------------------------------------------------
subroutine config_init
write(6,'(/,a)') ' <<<+- config init -+>>>'; flush(6)
print'(/,a)', ' <<<+- config init -+>>>'; flush(6)
call parse_material
call parse_numerics
call parse_debug
end subroutine config_init
!--------------------------------------------------------------------------------------------------
!> @brief reads material.yaml
!> @brief Read material.yaml or <jobname>.yaml.
!--------------------------------------------------------------------------------------------------
subroutine parse_material
logical :: fileExists
character(len=:), allocatable :: fname,flow
character(len=:), allocatable :: fname
fname = getSolverJobName()//'.yaml'
inquire(file=fname,exist=fileExists)
@ -68,88 +59,54 @@ subroutine parse_material
inquire(file=fname,exist=fileExists)
if(.not. fileExists) call IO_error(100,ext_msg=fname)
endif
write(6,'(/,a)') ' reading '//fname; flush(6)
flow = to_flow(IO_read(fname))
material_root => parse_flow(flow)
print*, 'reading '//fname; flush(6)
config_material => YAML_parse_file(fname)
end subroutine parse_material
!--------------------------------------------------------------------------------------------------
!> @brief reads in parameters from numerics.yaml and sets openMP related parameters. Also does
! a sanity check
!> @brief Read numerics.yaml.
!--------------------------------------------------------------------------------------------------
subroutine parse_numerics
!$ integer :: gotDAMASK_NUM_THREADS = 1
integer :: ierr
character(len=:), allocatable :: &
numerics_inFlow
logical :: fexist
!$ character(len=6) DAMASK_NumThreadsString ! environment variable DAMASK_NUM_THREADS
#ifdef PETSc
call MPI_Comm_rank(PETSC_COMM_WORLD,worldrank,ierr);CHKERRQ(ierr)
call MPI_Comm_size(PETSC_COMM_WORLD,worldsize,ierr);CHKERRQ(ierr)
#endif
!$ call GET_ENVIRONMENT_VARIABLE(NAME='DAMASK_NUM_THREADS',VALUE=DAMASK_NumThreadsString,STATUS=gotDAMASK_NUM_THREADS) ! get environment variable DAMASK_NUM_THREADS...
!$ if(gotDAMASK_NUM_THREADS /= 0) then ! could not get number of threads, set it to 1
!$ call IO_warning(35,ext_msg='BEGIN:'//DAMASK_NumThreadsString//':END')
!$ DAMASK_NumThreadsInt = 1_4
!$ else
!$ read(DAMASK_NumThreadsString,'(i6)') DAMASK_NumThreadsInt ! read as integer
!$ if (DAMASK_NumThreadsInt < 1_4) DAMASK_NumThreadsInt = 1_4 ! in case of string conversion fails, set it to one
!$ endif
!$ call omp_set_num_threads(DAMASK_NumThreadsInt) ! set number of threads for parallel execution
numerics_root => emptyDict
config_numerics => emptyDict
inquire(file='numerics.yaml', exist=fexist)
if (fexist) then
write(6,'(a,/)') ' using values from config.yaml file'
flush(6)
numerics_inFlow = to_flow(IO_read('numerics.yaml'))
numerics_root => parse_flow(numerics_inFlow)
print*, 'reading numerics.yaml'; flush(6)
config_numerics => YAML_parse_file('numerics.yaml')
endif
!--------------------------------------------------------------------------------------------------
! openMP parameter
!$ write(6,'(a24,1x,i8,/)') ' number of threads: ',DAMASK_NumThreadsInt
end subroutine parse_numerics
!--------------------------------------------------------------------------------------------------
!> @brief reads in parameters from debug.yaml
!> @brief Read debug.yaml.
!--------------------------------------------------------------------------------------------------
subroutine parse_debug
character(len=:), allocatable :: debug_inFlow
logical :: fexist
logical :: fexist
#ifdef DEBUG
write(6,'(a)') achar(27)//'[31m <<<+- DEBUG version -+>>>'//achar(27)//'[0m'
#endif
debug_root => emptyDict
config_debug => emptyDict
inquire(file='debug.yaml', exist=fexist)
fileExists: if (fexist) then
debug_inFlow = to_flow(IO_read('debug.yaml'))
debug_root => parse_flow(debug_inFlow)
print*, 'reading debug.yaml'; flush(6)
config_debug => YAML_parse_file('debug.yaml')
endif fileExists
end subroutine parse_debug
!--------------------------------------------------------------------------------------------------
!> @brief deallocates material.yaml structure
!> @brief Deallocate config_material.
!ToDo: deallocation of numerics debug (optional)
!--------------------------------------------------------------------------------------------------
subroutine config_deallocate
deallocate(material_root) !ToDo: deallocation of numerics and debug (slightly different for optional files)
deallocate(config_material)
end subroutine config_deallocate
end module config

View File

@ -140,7 +140,7 @@ module constitutive
el !< current element number
end subroutine plastic_nonlocal_dotState
module subroutine source_damage_anisoBrittle_dotState(S, ipc, ip, el)
integer, intent(in) :: &
ipc, & !< component-ID of integration point
@ -212,7 +212,7 @@ module constitutive
real(pReal), dimension(3,3) :: &
initialStrain
end function kinematics_thermal_expansion_initialStrain
module subroutine plastic_nonlocal_updateCompatibility(orientation,instance,i,e)
integer, intent(in) :: &
instance, &
@ -269,7 +269,7 @@ module constitutive
Li !< thermal velocity gradient
real(pReal), intent(out), dimension(3,3,3,3) :: &
dLi_dTstar !< derivative of Li with respect to Tstar (4th-order tensor defined to be zero)
end subroutine kinematics_thermal_expansion_LiAndItsTangent
end subroutine kinematics_thermal_expansion_LiAndItsTangent
module subroutine plastic_kinehardening_deltaState(Mp,instance,of)
@ -303,7 +303,7 @@ module constitutive
module subroutine plastic_results
end subroutine plastic_results
module subroutine damage_results
end subroutine damage_results
@ -339,7 +339,7 @@ module constitutive
real(pReal), intent(in), dimension(3,3) :: &
F, & !< elastic deformation gradient
Fp !< plastic deformation gradient
end subroutine constitutive_plastic_dependentState
end subroutine constitutive_plastic_dependentState
end interface constitutive_dependentState
@ -356,7 +356,7 @@ module constitutive
end type tDebugOptions
type(tDebugOptions) :: debugConstitutive
public :: &
constitutive_init, &
constitutive_homogenizedC, &
@ -379,7 +379,7 @@ contains
!--------------------------------------------------------------------------------------------------
!> @brief Initialze constitutive models for individual physics
!> @brief Initialze constitutive models for individual physics
!--------------------------------------------------------------------------------------------------
subroutine constitutive_init
@ -394,17 +394,17 @@ subroutine constitutive_init
elastic, &
stiffDegradation
debug_constitutive => debug_root%get('constitutive', defaultVal=emptyList)
debugConstitutive%basic = debug_constitutive%contains('basic')
debugConstitutive%extensive = debug_constitutive%contains('extensive')
debug_constitutive => config_debug%get('constitutive', defaultVal=emptyList)
debugConstitutive%basic = debug_constitutive%contains('basic')
debugConstitutive%extensive = debug_constitutive%contains('extensive')
debugConstitutive%selective = debug_constitutive%contains('selective')
debugConstitutive%element = debug_root%get_asInt('element',defaultVal = 1)
debugConstitutive%ip = debug_root%get_asInt('integrationpoint',defaultVal = 1)
debugConstitutive%grain = debug_root%get_asInt('grain',defaultVal = 1)
debugConstitutive%element = config_debug%get_asInt('element',defaultVal = 1)
debugConstitutive%ip = config_debug%get_asInt('integrationpoint',defaultVal = 1)
debugConstitutive%grain = config_debug%get_asInt('grain',defaultVal = 1)
!-------------------------------------------------------------------------------------------------
! initialize elasticity (hooke) !ToDO: Maybe move to elastic submodule along with function homogenizedC?
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(phase_elasticity(phases%length), source = ELASTICITY_undefined_ID)
allocate(phase_elasticityInstance(phases%length), source = 0)
allocate(phase_NstiffnessDegradations(phases%length),source=0)
@ -446,7 +446,7 @@ subroutine constitutive_init
call damage_init
call thermal_init
write(6,'(/,a)') ' <<<+- constitutive init -+>>>'; flush(6)
print'(/,a)', ' <<<+- constitutive init -+>>>'; flush(6)
constitutive_source_maxSizeDotState = 0
PhaseLoop2:do p = 1,phases%length
@ -472,7 +472,7 @@ end subroutine constitutive_init
!--------------------------------------------------------------------------------------------------
module function source_active(source_label,src_length) result(active_source)
character(len=*), intent(in) :: source_label !< name of source mechanism
character(len=*), intent(in) :: source_label !< name of source mechanism
integer, intent(in) :: src_length !< max. number of sources in system
logical, dimension(:,:), allocatable :: active_source
@ -480,10 +480,10 @@ module function source_active(source_label,src_length) result(active_source)
phases, &
phase, &
sources, &
src
src
integer :: p,s
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(active_source(src_length,phases%length), source = .false. )
do p = 1, phases%length
phase => phases%get(p)
@ -512,10 +512,10 @@ module function kinematics_active(kinematics_label,kinematics_length) result(ac
phases, &
phase, &
kinematics, &
kinematics_type
kinematics_type
integer :: p,k
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(active_kinematics(kinematics_length,phases%length), source = .false. )
do p = 1, phases%length
phase => phases%get(p)
@ -528,7 +528,7 @@ module function kinematics_active(kinematics_label,kinematics_length) result(ac
end function kinematics_active
!--------------------------------------------------------------------------------------------------
!> @brief returns the homogenize elasticity matrix

View File

@ -117,7 +117,7 @@ module subroutine damage_init
sources, &
kinematics
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(sourceState (phases%length))
allocate(phase_Nsources(phases%length),source = 0) ! same for kinematics

View File

@ -1,5 +1,5 @@
!----------------------------------------------------------------------------------------------------
!> @brief internal microstructure state for all plasticity constitutive models
!> @brief internal microstructure state for all plasticity constitutive models
!----------------------------------------------------------------------------------------------------
submodule(constitutive) constitutive_plastic
@ -198,7 +198,9 @@ module subroutine plastic_init
integer :: p
class(tNode), pointer :: phases
phases => material_root%get('phase')
print'(/,a)', ' <<<+- constitutive_plastic init -+>>>'
phases => config_material%get('phase')
allocate(plasticState(phases%length))
allocate(phase_plasticity(phases%length),source = PLASTICITY_undefined_ID)
@ -215,7 +217,7 @@ module subroutine plastic_init
do p = 1, phases%length
phase_plasticityInstance(p) = count(phase_plasticity(1:p) == phase_plasticity(p))
enddo
enddo
end subroutine plastic_init
@ -235,7 +237,7 @@ module function plastic_active(plastic_label) result(active_plastic)
pl
integer :: p
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(active_plastic(phases%length), source = .false. )
do p = 1, phases%length
phase => phases%get(p)
@ -355,7 +357,7 @@ end subroutine constitutive_plastic_LpAndItsTangents
!--------------------------------------------------------------------------------------------
!> @brief writes plasticity constitutive results to HDF5 output file
!--------------------------------------------------------------------------------------------
!--------------------------------------------------------------------------------------------
module subroutine plastic_results
integer :: p

View File

@ -96,23 +96,22 @@ module function plastic_disloTungsten_init() result(myPlasticity)
phase, &
pl
write(6,'(/,a)') ' <<<+- plastic_disloTungsten init -+>>>'
write(6,'(/,a)') ' Cereceda et al., International Journal of Plasticity 78:242256, 2016'
write(6,'(a)') ' https://dx.doi.org/10.1016/j.ijplas.2015.09.002'
print'(/,a)', ' <<<+- plastic_dislotungsten init -+>>>'
myPlasticity = plastic_active('disloTungsten')
Ninstance = count(myPlasticity)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
print*, 'Cereceda et al., International Journal of Plasticity 78:242256, 2016'
print*, 'https://dx.doi.org/10.1016/j.ijplas.2015.09.002'
allocate(param(Ninstance))
allocate(state(Ninstance))
allocate(dotState(Ninstance))
allocate(dependentState(Ninstance))
phases => material_root%get('phase')
phases => config_material%get('phase')
i = 0
do p = 1, phases%length
phase => phases%get(p)
@ -179,7 +178,7 @@ module function plastic_disloTungsten_init() result(myPlasticity)
prm%Q_cl = pl%get_asFloat('Q_cl')
prm%atomicVolume = pl%get_asFloat('f_at') * prm%b_sl**3.0_pReal
prm%D_a = pl%get_asFloat('D_a') * prm%b_sl
prm%dipoleformation = pl%get_asBool('dipole_formation_factor', defaultVal = .true.)
! expand: family => system
@ -410,19 +409,19 @@ module subroutine plastic_disloTungsten_results(instance,group)
associate(prm => param(instance), stt => state(instance), dst => dependentState(instance))
outputsLoop: do o = 1,size(prm%output)
select case(trim(prm%output(o)))
case('rho_mob')
case('rho_mob')
if(prm%sum_N_sl>0) call results_writeDataset(group,stt%rho_mob,trim(prm%output(o)), &
'mobile dislocation density','1/m²')
case('rho_dip')
case('rho_dip')
if(prm%sum_N_sl>0) call results_writeDataset(group,stt%rho_dip,trim(prm%output(o)), &
'dislocation dipole density''1/m²')
case('gamma_sl')
case('gamma_sl')
if(prm%sum_N_sl>0) call results_writeDataset(group,stt%gamma_sl,trim(prm%output(o)), &
'plastic shear','1')
case('Lambda_sl')
case('Lambda_sl')
if(prm%sum_N_sl>0) call results_writeDataset(group,dst%Lambda_sl,trim(prm%output(o)), &
'mean free path for slip','m')
case('tau_pass')
case('tau_pass')
if(prm%sum_N_sl>0) call results_writeDataset(group,dst%threshold_stress,trim(prm%output(o)), &
'threshold stress for slip','Pa')
end select

View File

@ -143,29 +143,28 @@ module function plastic_dislotwin_init() result(myPlasticity)
phase, &
pl
write(6,'(/,a)') ' <<<+- constitutive_dislotwin init -+>>>'
write(6,'(/,a)') ' Ma and Roters, Acta Materialia 52(12):36033612, 2004'
write(6,'(a)') ' https://doi.org/10.1016/j.actamat.2004.04.012'
write(6,'(/,a)') ' Roters et al., Computational Materials Science 39:9195, 2007'
write(6,'(a)') ' https://doi.org/10.1016/j.commatsci.2006.04.014'
write(6,'(/,a)') ' Wong et al., Acta Materialia 118:140151, 2016'
write(6,'(a,/)') ' https://doi.org/10.1016/j.actamat.2016.07.032'
print'(/,a)', ' <<<+- plastic_dislotwin init -+>>>'
myPlasticity = plastic_active('dislotwin')
Ninstance = count(myPlasticity)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
print*, 'Ma and Roters, Acta Materialia 52(12):36033612, 2004'
print*, 'https://doi.org/10.1016/j.actamat.2004.04.012'//IO_EOL
print*, 'Roters et al., Computational Materials Science 39:9195, 2007'
print*, 'https://doi.org/10.1016/j.commatsci.2006.04.014'//IO_EOL
print*, 'Wong et al., Acta Materialia 118:140151, 2016'
print*, 'https://doi.org/10.1016/j.actamat.2016.07.032'
allocate(param(Ninstance))
allocate(state(Ninstance))
allocate(dotState(Ninstance))
allocate(dependentState(Ninstance))
phases => material_root%get('phase')
phases => config_material%get('phase')
i = 0
do p = 1, phases%length
phase => phases%get(p)
@ -414,7 +413,7 @@ module function plastic_dislotwin_init() result(myPlasticity)
+ size(['f_tr']) * prm%sum_N_tr
sizeState = sizeDotState
call constitutive_allocateState(plasticState(p),NipcMyPhase,sizeState,sizeDotState,0)
!--------------------------------------------------------------------------------------------------

View File

@ -49,7 +49,7 @@ contains
!> @brief Perform module initialization.
!> @details reads in material parameters, allocates arrays, and does sanity checks
!--------------------------------------------------------------------------------------------------
module function plastic_isotropic_init() result(myPlasticity)
module function plastic_isotropic_init() result(myPlasticity)
logical, dimension(:), allocatable :: myPlasticity
integer :: &
@ -67,23 +67,21 @@ module function plastic_isotropic_init() result(myPlasticity)
phase, &
pl
write(6,'(/,a)') ' <<<+- plastic_isotropic init -+>>>'
write(6,'(/,a)') ' Maiti and Eisenlohr, Scripta Materialia 145:3740, 2018'
write(6,'(a)') ' https://doi.org/10.1016/j.scriptamat.2017.09.047'
print'(/,a)', ' <<<+- plastic_isotropic init -+>>>'
myPlasticity = plastic_active('isotropic')
Ninstance = count(myPlasticity)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
print*, 'Maiti and Eisenlohr, Scripta Materialia 145:3740, 2018'
print*, 'https://doi.org/10.1016/j.scriptamat.2017.09.047'
allocate(param(Ninstance))
allocate(state(Ninstance))
allocate(dotState(Ninstance))
phases => material_root%get('phase')
phases => config_material%get('phase')
i = 0
do p = 1, phases%length
phase => phases%get(p)
@ -96,7 +94,7 @@ module function plastic_isotropic_init() result(myPlasticity)
pl => phase%get('plasticity')
#if defined (__GFORTRAN__)
#if defined (__GFORTRAN__)
prm%output = output_asStrings(pl)
#else
prm%output = pl%get_asStrings('output',defaultVal=emptyStringArray)
@ -203,10 +201,10 @@ module subroutine plastic_isotropic_LpAndItsTangent(Lp,dLp_dMp,Mp,instance,of)
Lp = dot_gamma/prm%M * Mp_dev/norm_Mp_dev
#ifdef DEBUG
if (debugConstitutive%extensive .and. (of == prm%of_debug .or. .not. debugConstitutive%selective)) then
write(6,'(/,a,/,3(12x,3(f12.4,1x)/))') '<< CONST isotropic >> Tstar (dev) / MPa', &
print'(/,a,/,3(12x,3(f12.4,1x)/))', '<< CONST isotropic >> Tstar (dev) / MPa', &
transpose(Mp_dev)*1.0e-6_pReal
write(6,'(/,a,/,f12.5)') '<< CONST isotropic >> norm Tstar / MPa', norm_Mp_dev*1.0e-6_pReal
write(6,'(/,a,/,f12.5)') '<< CONST isotropic >> gdot', dot_gamma
print'(/,a,/,f12.5)', '<< CONST isotropic >> norm Tstar / MPa', norm_Mp_dev*1.0e-6_pReal
print'(/,a,/,f12.5)', '<< CONST isotropic >> gdot', dot_gamma
end if
#endif
forall (k=1:3,l=1:3,m=1:3,n=1:3) &
@ -258,8 +256,8 @@ module subroutine plastic_isotropic_LiAndItsTangent(Li,dLi_dMi,Mi,instance,of)
#ifdef DEBUG
if (debugConstitutive%extensive .and. (of == prm%of_debug .or. .not. debugConstitutive%selective)) then
write(6,'(/,a,/,f12.5)') '<< CONST isotropic >> pressure / MPa', tr/3.0_pReal*1.0e-6_pReal
write(6,'(/,a,/,f12.5)') '<< CONST isotropic >> gdot', prm%dot_gamma_0 * (3.0_pReal*prm%M*stt%xi(of))**(-prm%n) &
print'(/,a,/,f12.5)', '<< CONST isotropic >> pressure / MPa', tr/3.0_pReal*1.0e-6_pReal
print'(/,a,/,f12.5)', '<< CONST isotropic >> gdot', prm%dot_gamma_0 * (3.0_pReal*prm%M*stt%xi(of))**(-prm%n) &
* tr * abs(tr)**(prm%n-1.0_pReal)
end if
#endif
@ -339,7 +337,7 @@ module subroutine plastic_isotropic_results(instance,group)
associate(prm => param(instance), stt => state(instance))
outputsLoop: do o = 1,size(prm%output)
select case(trim(prm%output(o)))
case ('xi')
case ('xi')
call results_writeDataset(group,stt%xi,trim(prm%output(o)), &
'resistance against plastic flow','Pa')
end select

View File

@ -58,7 +58,7 @@ contains
!> @brief Perform module initialization.
!> @details reads in material parameters, allocates arrays, and does sanity checks
!--------------------------------------------------------------------------------------------------
module function plastic_kinehardening_init() result(myPlasticity)
module function plastic_kinehardening_init() result(myPlasticity)
logical, dimension(:), allocatable :: myPlasticity
integer :: &
@ -79,12 +79,11 @@ module function plastic_kinehardening_init() result(myPlasticity)
phase, &
pl
write(6,'(/,a)') ' <<<+- plastic_kinehardening init -+>>>'
print'(/,a)', ' <<<+- plastic_kinehardening init -+>>>'
myPlasticity = plastic_active('kinehardening')
Ninstance = count(myPlasticity)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
allocate(param(Ninstance))
@ -92,7 +91,7 @@ module function plastic_kinehardening_init() result(myPlasticity)
allocate(dotState(Ninstance))
allocate(deltaState(Ninstance))
phases => material_root%get('phase')
phases => config_material%get('phase')
i = 0
do p = 1, phases%length
phase => phases%get(p)
@ -346,8 +345,8 @@ module subroutine plastic_kinehardening_deltaState(Mp,instance,of)
#ifdef DEBUG
if (debugConstitutive%extensive &
.and. (of == prm%of_debug .or. .not. debugConstitutive%selective)) then
write(6,'(a)') '======= kinehardening delta state ======='
write(6,*) sense,state(instance)%sense(:,of)
print*, '======= kinehardening delta state ======='
print*, sense,state(instance)%sense(:,of)
endif
#endif
@ -384,7 +383,7 @@ module subroutine plastic_kinehardening_results(instance,group)
case('xi')
if(prm%sum_N_sl>0) call results_writeDataset(group,stt%crss,trim(prm%output(o)), &
'resistance against plastic slip','Pa')
case('tau_b')
case('tau_b')
if(prm%sum_N_sl>0) call results_writeDataset(group,stt%crss_back,trim(prm%output(o)), &
'back stress against plastic slip','Pa')
case ('sgn(gamma)')

View File

@ -9,10 +9,10 @@ submodule(constitutive:constitutive_plastic) plastic_none
contains
!--------------------------------------------------------------------------------------------------
!> @brief module initialization
!> @brief Perform module initialization.
!> @details reads in material parameters, allocates arrays, and does sanity checks
!--------------------------------------------------------------------------------------------------
module function plastic_none_init() result(myPlasticity)
module function plastic_none_init() result(myPlasticity)
logical, dimension(:), allocatable :: myPlasticity
integer :: &
@ -24,20 +24,20 @@ module function plastic_none_init() result(myPlasticity)
phase, &
pl
write(6,'(/,a)') ' <<<+- plastic_none init -+>>>'
print'(/,a)', ' <<<+- plastic_none init -+>>>'
phases => material_root%get('phase')
allocate(myPlasticity(phases%length), source = .false. )
phases => config_material%get('phase')
allocate(myPlasticity(phases%length), source = .false.)
do p = 1, phases%length
phase => phases%get(p)
pl => phase%get('plasticity')
if(pl%get_asString('type') == 'none') myPlasticity(p) = .true.
enddo
Ninstance = count(myPlasticity)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
Ninstance = count(myPlasticity)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
do p = 1, phases%length
phase => phases%get(p)
if(.not. myPlasticity(p)) cycle

View File

@ -159,8 +159,9 @@ submodule(constitutive:constitutive_plastic) plastic_nonlocal
contains
!--------------------------------------------------------------------------------------------------
!> @brief module initialization
!> @brief Perform module initialization.
!> @details reads in material parameters, allocates arrays, and does sanity checks
!--------------------------------------------------------------------------------------------------
module function plastic_nonlocal_init() result(myPlasticity)
@ -183,24 +184,22 @@ module function plastic_nonlocal_init() result(myPlasticity)
phases, &
phase, &
pl
write(6,'(/,a)') ' <<<+- plastic_nonlocal init -+>>>'
write(6,'(/,a)') ' Reuber et al., Acta Materialia 71:333348, 2014'
write(6,'(a)') ' https://doi.org/10.1016/j.actamat.2014.03.012'
write(6,'(/,a)') ' Kords, Dissertation RWTH Aachen, 2014'
write(6,'(a)') ' http://publications.rwth-aachen.de/record/229993'
print'(/,a)', ' <<<+- plastic_nonlocal init -+>>>'
myPlasticity = plastic_active('nonlocal')
Ninstance = count(myPlasticity)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) then
call geometry_plastic_nonlocal_disable
return
endif
print*, 'Reuber et al., Acta Materialia 71:333348, 2014'
print*, 'https://doi.org/10.1016/j.actamat.2014.03.012'//IO_EOL
print*, 'Kords, Dissertation RWTH Aachen, 2014'
print*, 'http://publications.rwth-aachen.de/record/229993'//IO_EOL
allocate(param(Ninstance))
allocate(state(Ninstance))
@ -209,7 +208,7 @@ module function plastic_nonlocal_init() result(myPlasticity)
allocate(deltaState(Ninstance))
allocate(microstructure(Ninstance))
phases => material_root%get('phase')
phases => config_material%get('phase')
i = 0
do p = 1, phases%length
phase => phases%get(p)
@ -224,14 +223,14 @@ module function plastic_nonlocal_init() result(myPlasticity)
dst => microstructure(i))
pl => phase%get('plasticity')
phase_localPlasticity(p) = .not. pl%contains('nonlocal')
phase_localPlasticity(p) = .not. pl%contains('nonlocal')
#if defined (__GFORTRAN__)
prm%output = output_asStrings(pl)
#else
prm%output = pl%get_asStrings('output',defaultVal=emptyStringArray)
#endif
prm%atol_rho = pl%get_asFloat('atol_rho',defaultVal=1.0e4_pReal)
! This data is read in already in lattice
@ -519,7 +518,7 @@ module function plastic_nonlocal_init() result(myPlasticity)
if(.not. myPlasticity(p)) cycle
i = i + 1
NipcMyPhase = count(material_phaseAt==p) * discretization_nIP
l = 0
do t = 1,4
@ -543,7 +542,7 @@ module function plastic_nonlocal_init() result(myPlasticity)
enddo
if (iD(param(i)%sum_N_sl,2,i) /= plasticState(p)%sizeState) &
call IO_error(0, ext_msg = 'state indices not properly set (nonlocal)')
enddo
enddo
end function plastic_nonlocal_init
@ -1275,7 +1274,7 @@ function rhoDotFlux(F,Fp,timestep, instance,of,ip,el)
.and. prm%CFLfactor * abs(v0) * timestep &
> IPvolume(ip,el) / maxval(IParea(:,ip,el))), &
' at a timestep of ',timestep
write(6,'(a)') '<< CONST >> enforcing cutback !!!'
print*, '<< CONST >> enforcing cutback !!!'
endif
#endif
rhoDotFlux = IEEE_value(1.0_pReal,IEEE_quiet_NaN) ! enforce cutback

View File

@ -66,7 +66,7 @@ contains
!> @brief Perform module initialization.
!> @details reads in material parameters, allocates arrays, and does sanity checks
!--------------------------------------------------------------------------------------------------
module function plastic_phenopowerlaw_init() result(myPlasticity)
module function plastic_phenopowerlaw_init() result(myPlasticity)
logical, dimension(:), allocatable :: myPlasticity
integer :: &
@ -88,20 +88,18 @@ module function plastic_phenopowerlaw_init() result(myPlasticity)
phase, &
pl
write(6,'(/,a)') ' <<<+- plastic_phenopowerlaw init -+>>>'
print'(/,a)', ' <<<+- plastic_phenopowerlaw init -+>>>'
myPlasticity = plastic_active('phenopowerlaw')
Ninstance = count(myPlasticity)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
allocate(param(Ninstance))
allocate(state(Ninstance))
allocate(dotState(Ninstance))
phases => material_root%get('phase')
phases => config_material%get('phase')
i = 0
do p = 1, phases%length
phase => phases%get(p)
@ -231,7 +229,7 @@ module function plastic_phenopowerlaw_init() result(myPlasticity)
+ size(['xi_tw ','gamma_tw']) * prm%sum_N_tw
sizeState = sizeDotState
call constitutive_allocateState(plasticState(p),NipcMyPhase,sizeState,sizeDotState,0)
!--------------------------------------------------------------------------------------------------

View File

@ -10,6 +10,7 @@
module crystallite
use prec
use parallelization
use IO
use HDF5_utilities
use DAMASK_interface
@ -81,8 +82,6 @@ module crystallite
iJacoLpresiduum, & !< frequency of Jacobian update of residuum in Lp
nState, & !< state loop limit
nStress !< stress loop limit
character(len=:), allocatable :: &
integrator !< integration scheme
real(pReal) :: &
subStepMinCryst, & !< minimum (relative) size of sub-step allowed during cutback
subStepSizeCryst, & !< size of first substep when cutback
@ -147,15 +146,15 @@ subroutine crystallite_init
phase, &
generic_param
write(6,'(/,a)') ' <<<+- crystallite init -+>>>'
print'(/,a)', ' <<<+- crystallite init -+>>>'
debug_crystallite => debug_root%get('crystallite', defaultVal=emptyList)
debug_crystallite => config_debug%get('crystallite', defaultVal=emptyList)
debugCrystallite%basic = debug_crystallite%contains('basic')
debugCrystallite%extensive = debug_crystallite%contains('extensive')
debugCrystallite%selective = debug_crystallite%contains('selective')
debugCrystallite%element = debug_root%get_asInt('element', defaultVal=1)
debugCrystallite%ip = debug_root%get_asInt('integrationpoint', defaultVal=1)
debugCrystallite%grain = debug_root%get_asInt('grain', defaultVal=1)
debugCrystallite%element = config_debug%get_asInt('element', defaultVal=1)
debugCrystallite%ip = config_debug%get_asInt('integrationpoint', defaultVal=1)
debugCrystallite%grain = config_debug%get_asInt('grain', defaultVal=1)
cMax = homogenization_maxNgrains
iMax = discretization_nIP
@ -188,7 +187,7 @@ subroutine crystallite_init
allocate(crystallite_requested(cMax,iMax,eMax), source=.false.)
allocate(crystallite_converged(cMax,iMax,eMax), source=.true.)
num_crystallite => numerics_root%get('crystallite',defaultVal=emptyDict)
num_crystallite => config_numerics%get('crystallite',defaultVal=emptyDict)
num%subStepMinCryst = num_crystallite%get_asFloat ('subStepMin', defaultVal=1.0e-3_pReal)
num%subStepSizeCryst = num_crystallite%get_asFloat ('subStepSize', defaultVal=0.25_pReal)
@ -199,7 +198,6 @@ subroutine crystallite_init
num%rtol_crystalliteStress = num_crystallite%get_asFloat ('rtol_Stress', defaultVal=1.0e-6_pReal)
num%atol_crystalliteStress = num_crystallite%get_asFloat ('atol_Stress', defaultVal=1.0e-8_pReal)
num%iJacoLpresiduum = num_crystallite%get_asInt ('iJacoLpresiduum', defaultVal=1)
num%integrator = num_crystallite%get_asString('integrator', defaultVal='FPI')
num%nState = num_crystallite%get_asInt ('nState', defaultVal=20)
num%nStress = num_crystallite%get_asInt ('nStress', defaultVal=40)
@ -219,8 +217,7 @@ subroutine crystallite_init
if(num%nState < 1) call IO_error(301,ext_msg='nState')
if(num%nStress< 1) call IO_error(301,ext_msg='nStress')
select case(num%integrator)
select case(num_crystallite%get_asString('integrator',defaultVal='FPI'))
case('FPI')
integrateState => integrateStateFPI
case('Euler')
@ -235,7 +232,7 @@ subroutine crystallite_init
call IO_error(301,ext_msg='integrator')
end select
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(output_constituent(phases%length))
do c = 1, phases%length
@ -255,7 +252,7 @@ subroutine crystallite_init
do e = FEsolving_execElem(1),FEsolving_execElem(2)
myNcomponents = homogenization_Ngrains(material_homogenizationAt(e))
do i = FEsolving_execIP(1), FEsolving_execIP(2); do c = 1, myNcomponents
crystallite_Fp0(1:3,1:3,c,i,e) = material_orientation0(c,i,e)%asMatrix() ! plastic def gradient reflects init orientation
crystallite_Fp0(1:3,1:3,c,i,e) = material_orientation0(c,i,e)%asMatrix() ! Fp reflects initial orientation (see 10.1016/j.actamat.2006.01.005)
crystallite_Fp0(1:3,1:3,c,i,e) = crystallite_Fp0(1:3,1:3,c,i,e) &
/ math_det33(crystallite_Fp0(1:3,1:3,c,i,e))**(1.0_pReal/3.0_pReal)
crystallite_Fi0(1:3,1:3,c,i,e) = constitutive_initialFi(c,i,e)
@ -294,9 +291,9 @@ subroutine crystallite_init
#ifdef DEBUG
if (debugCrystallite%basic) then
write(6,'(a42,1x,i10)') ' # of elements: ', eMax
write(6,'(a42,1x,i10)') ' # of integration points/element: ', iMax
write(6,'(a42,1x,i10)') 'max # of constituents/integration point: ', cMax
print'(a42,1x,i10)', ' # of elements: ', eMax
print'(a42,1x,i10)', ' # of integration points/element: ', iMax
print'(a42,1x,i10)', 'max # of constituents/integration point: ', cMax
flush(6)
endif
@ -327,24 +324,24 @@ function crystallite_stress()
if (debugCrystallite%selective &
.and. FEsolving_execElem(1) <= debugCrystallite%element &
.and. debugCrystallite%element <= FEsolving_execElem(2)) then
write(6,'(/,a,i8,1x,i2,1x,i3)') '<< CRYST stress >> boundary and initial values at el ip ipc ', &
print'(/,a,i8,1x,i2,1x,i3)', '<< CRYST stress >> boundary and initial values at el ip ipc ', &
debugCrystallite%element,debugCrystallite%ip, debugCrystallite%grain
write(6,'(a,/,3(12x,3(f14.9,1x)/))') '<< CRYST stress >> F ', &
print'(a,/,3(12x,3(f14.9,1x)/))', '<< CRYST stress >> F ', &
transpose(crystallite_partionedF(1:3,1:3,debugCrystallite%grain, &
debugCrystallite%ip,debugCrystallite%element))
write(6,'(a,/,3(12x,3(f14.9,1x)/))') '<< CRYST stress >> F0 ', &
print'(a,/,3(12x,3(f14.9,1x)/))', '<< CRYST stress >> F0 ', &
transpose(crystallite_partionedF0(1:3,1:3,debugCrystallite%grain, &
debugCrystallite%ip,debugCrystallite%element))
write(6,'(a,/,3(12x,3(f14.9,1x)/))') '<< CRYST stress >> Fp0', &
print'(a,/,3(12x,3(f14.9,1x)/))', '<< CRYST stress >> Fp0', &
transpose(crystallite_partionedFp0(1:3,1:3,debugCrystallite%grain, &
debugCrystallite%ip,debugCrystallite%element))
write(6,'(a,/,3(12x,3(f14.9,1x)/))') '<< CRYST stress >> Fi0', &
print'(a,/,3(12x,3(f14.9,1x)/))', '<< CRYST stress >> Fi0', &
transpose(crystallite_partionedFi0(1:3,1:3,debugCrystallite%grain, &
debugCrystallite%ip,debugCrystallite%element))
write(6,'(a,/,3(12x,3(f14.9,1x)/))') '<< CRYST stress >> Lp0', &
print'(a,/,3(12x,3(f14.9,1x)/))', '<< CRYST stress >> Lp0', &
transpose(crystallite_partionedLp0(1:3,1:3,debugCrystallite%grain, &
debugCrystallite%ip,debugCrystallite%element))
write(6,'(a,/,3(12x,3(f14.9,1x)/))') '<< CRYST stress >> Li0', &
print'(a,/,3(12x,3(f14.9,1x)/))', '<< CRYST stress >> Li0', &
transpose(crystallite_partionedLi0(1:3,1:3,debugCrystallite%grain, &
debugCrystallite%ip,debugCrystallite%element))
endif
@ -393,7 +390,7 @@ function crystallite_stress()
#ifdef DEBUG
if (debugCrystallite%extensive) &
write(6,'(a,i6)') '<< CRYST stress >> crystallite iteration ',NiterationCrystallite
print'(a,i6)', '<< CRYST stress >> crystallite iteration ',NiterationCrystallite
#endif
!$OMP PARALLEL DO PRIVATE(formerSubStep)
elementLooping3: do e = FEsolving_execElem(1),FEsolving_execElem(2)
@ -575,7 +572,7 @@ subroutine crystallite_stressTangent
lhs_3333 = crystallite_subdt(c,i,e)*math_mul3333xx3333(dSdFe,temp_3333) &
+ math_mul3333xx3333(dSdFi,dFidS)
call math_invert(temp_99,error,math_identity2nd(9)+math_3333to99(lhs_3333))
call math_invert(temp_99,error,math_eye(9)+math_3333to99(lhs_3333))
if (error) then
call IO_warning(warning_ID=600,el=e,ip=i,g=c, &
ext_msg='inversion error in analytic tangent calculation')
@ -946,7 +943,7 @@ function integrateStress(ipc,ip,el,timeFraction) result(broken)
do o=1,3; do p=1,3
dFe_dLp(o,1:3,p,1:3) = - dt * A(o,p)*transpose(invFi_new) ! dFe_dLp(i,j,k,l) = -dt * A(i,k) invFi(l,j)
enddo; enddo
dRLp_dLp = math_identity2nd(9) &
dRLp_dLp = math_eye(9) &
- math_3333to99(math_mul3333xx3333(math_mul3333xx3333(dLp_dS,dS_dFe),dFe_dLp))
temp_9 = math_33to9(residuumLp)
call dgesv(9,1,dRLp_dLp,9,devNull_9,temp_9,9,ierr) ! solve dRLp/dLp * delta Lp = -res for delta Lp
@ -991,7 +988,7 @@ function integrateStress(ipc,ip,el,timeFraction) result(broken)
do o=1,3; do p=1,3
dFi_dLi(1:3,1:3,o,p) = matmul(matmul(Fi_new,dFi_dLi(1:3,1:3,o,p)),Fi_new)
enddo; enddo
dRLi_dLi = math_identity2nd(9) &
dRLi_dLi = math_eye(9) &
- math_3333to99(math_mul3333xx3333(dLi_dS, math_mul3333xx3333(dS_dFe, dFe_dLi) &
+ math_mul3333xx3333(dS_dFi, dFi_dLi))) &
- math_3333to99(math_mul3333xx3333(dLi_dFi, dFi_dLi))
@ -1564,7 +1561,7 @@ subroutine crystallite_restartWrite
integer(HID_T) :: fileHandle, groupHandle
character(len=pStringLen) :: fileName, datasetName
write(6,'(a)') ' writing field and constitutive data required for restart to file';flush(6)
print*, ' writing field and constitutive data required for restart to file';flush(6)
write(fileName,'(a,i0,a)') trim(getSolverJobName())//'_',worldrank,'.hdf5'
fileHandle = HDF5_openFile(fileName,'a')
@ -1605,7 +1602,7 @@ subroutine crystallite_restartRead
integer(HID_T) :: fileHandle, groupHandle
character(len=pStringLen) :: fileName, datasetName
write(6,'(/,a,i0,a)') ' reading restart information of increment from file'
print'(/,a,i0,a)', ' reading restart information of increment from file'
write(fileName,'(a,i0,a)') trim(getSolverJobName())//'_',worldrank,'.hdf5'
fileHandle = HDF5_openFile(fileName)

View File

@ -49,18 +49,18 @@ subroutine damage_local_init
homog, &
homogDamage
write(6,'(/,a)') ' <<<+- damage_local init -+>>>'; flush(6)
print'(/,a)', ' <<<+- damage_local init -+>>>'; flush(6)
!----------------------------------------------------------------------------------------------
! read numerics parameter and do sanity check
num_generic => numerics_root%get('generic',defaultVal=emptyDict)
num_generic => config_numerics%get('generic',defaultVal=emptyDict)
num%residualStiffness = num_generic%get_asFloat('residualStiffness', defaultVal=1.0e-6_pReal)
if (num%residualStiffness < 0.0_pReal) call IO_error(301,ext_msg='residualStiffness')
Ninstance = count(damage_type == DAMAGE_local_ID)
allocate(param(Ninstance))
material_homogenization => material_root%get('homogenization')
material_homogenization => config_material%get('homogenization')
do h = 1, material_homogenization%length
if (damage_type(h) /= DAMAGE_LOCAL_ID) cycle
homog => material_homogenization%get(h)

View File

@ -18,7 +18,7 @@ subroutine damage_none_init
integer :: h,NofMyHomog
write(6,'(/,a)') ' <<<+- damage_none init -+>>>'; flush(6)
print'(/,a)', ' <<<+- damage_none init -+>>>'; flush(6)
do h = 1, material_Nhomogenization
if (damage_type(h) /= DAMAGE_NONE_ID) cycle

View File

@ -53,17 +53,17 @@ subroutine damage_nonlocal_init
homog, &
homogDamage
write(6,'(/,a)') ' <<<+- damage_nonlocal init -+>>>'; flush(6)
print'(/,a)', ' <<<+- damage_nonlocal init -+>>>'; flush(6)
!------------------------------------------------------------------------------------
! read numerics parameter
num_generic => numerics_root%get('generic',defaultVal= emptyDict)
num_generic => config_numerics%get('generic',defaultVal= emptyDict)
num%charLength = num_generic%get_asFloat('charLength',defaultVal=1.0_pReal)
Ninstance = count(damage_type == DAMAGE_nonlocal_ID)
allocate(param(Ninstance))
material_homogenization => material_root%get('homogenization')
material_homogenization => config_material%get('homogenization')
do h = 1, material_homogenization%length
if (damage_type(h) /= DAMAGE_NONLOCAL_ID) cycle
homog => material_homogenization%get(h)

View File

@ -49,7 +49,7 @@ subroutine discretization_init(microstructureAt,&
integer, optional, intent(in) :: &
sharedNodesBegin !< index of first node shared among different processes (MPI)
write(6,'(/,a)') ' <<<+- discretization init -+>>>'; flush(6)
print'(/,a)', ' <<<+- discretization init -+>>>'; flush(6)
discretization_nElem = size(microstructureAt,1)
discretization_nIP = size(IPcoords0,2)/discretization_nElem

View File

@ -4,9 +4,9 @@
!--------------------------------------------------------------------------------------------------
module element
use IO
implicit none
private
private
!---------------------------------------------------------------------------------------------------
!> Properties of a single element
@ -39,7 +39,7 @@ module element
integer, parameter :: &
NELEMTYPE = 13
integer, dimension(NELEMTYPE), parameter :: NNODE = &
[ &
3, & ! 2D, 1 IP
@ -57,7 +57,7 @@ module element
20, & ! 3D, 8 IP
20 & ! 3D, 27 IP
] !< number of nodes that constitute a specific type of element
integer, dimension(NELEMTYPE), parameter :: GEOMTYPE = &
[ &
1, & ! 1 triangle
@ -75,7 +75,7 @@ module element
9, & ! 8 hexahedrons
10 & ! 27 hexahedrons
] !< geometry type (same number of cell nodes and IPs)
integer, dimension(maxval(GEOMTYPE)), parameter :: NCELLNODE = &
[ &
3, &
@ -89,21 +89,21 @@ module element
27, &
64 &
] !< number of cell nodes
integer, dimension(maxval(GEOMTYPE)), parameter :: NIP = &
[ &
1, &
3, &
4, &
9, &
1, &
4, &
6, &
1, &
8, &
27 &
1, &
3, &
4, &
9, &
1, &
4, &
6, &
1, &
8, &
27 &
] !< number of IPs
integer, dimension(maxval(GEOMTYPE)), parameter :: CELLTYPE = &
[ &
1, & ! 2D, 3 node (Triangle)
@ -147,7 +147,7 @@ module element
! It is sorted in (local) +x,-x, +y,-y, +z,-z direction.
! Positive integers denote an intra-element IP identifier.
! Negative integers denote the interface behind which the neighboring (extra-element) IP will be located.
integer, dimension(NIPNEIGHBOR(CELLTYPE(1)),NIP(1)), parameter :: IPNEIGHBOR1 = &
reshape([&
-2,-3,-1 &
@ -156,7 +156,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(1)),NIP(1)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(2)),NIP(2)), parameter :: IPNEIGHBOR2 = &
reshape([&
2,-3, 3,-1, &
@ -167,7 +167,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(2)),NIP(2)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(3)),NIP(3)), parameter :: IPNEIGHBOR3 = &
reshape([&
2,-4, 3,-1, &
@ -179,7 +179,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(3)),NIP(3)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(4)),NIP(4)), parameter :: IPNEIGHBOR4 = &
reshape([&
2,-4, 4,-1, &
@ -196,7 +196,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(4)),NIP(4)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(5)),NIP(5)), parameter :: IPNEIGHBOR5 = &
reshape([&
-1,-2,-3,-4 &
@ -205,7 +205,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(5)),NIP(5)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(6)),NIP(6)), parameter :: IPNEIGHBOR6 = &
reshape([&
2,-4, 3,-2, 4,-1, &
@ -217,7 +217,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(6)),NIP(6)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(7)),NIP(7)), parameter :: IPNEIGHBOR7 = &
reshape([&
2,-4, 3,-2, 4,-1, &
@ -231,7 +231,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(7)),NIP(7)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(8)),NIP(8)), parameter :: IPNEIGHBOR8 = &
reshape([&
-3,-5,-4,-2,-6,-1 &
@ -240,7 +240,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(8)),NIP(8)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(9)),NIP(9)), parameter :: IPNEIGHBOR9 = &
reshape([&
2,-5, 3,-2, 5,-1, &
@ -256,7 +256,7 @@ module element
#else
],[NIPNEIGHBOR(CELLTYPE(9)),NIP(9)])
#endif
integer, dimension(NIPNEIGHBOR(CELLTYPE(10)),NIP(10)), parameter :: IPNEIGHBOR10 = &
reshape([&
2,-5, 4,-2,10,-1, &
@ -292,7 +292,7 @@ module element
],[NIPNEIGHBOR(CELLTYPE(10)),NIP(10)])
#endif
integer, dimension(NNODE(1),NCELLNODE(GEOMTYPE(1))), parameter :: CELLNODEPARENTNODEWEIGHTS1 = &
reshape([&
1, 0, 0, &
@ -303,7 +303,7 @@ module element
#else
],[NNODE(1),NCELLNODE(GEOMTYPE(1))])
#endif
integer, dimension(NNODE(2),NCELLNODE(GEOMTYPE(2))), parameter :: CELLNODEPARENTNODEWEIGHTS2 = &
reshape([&
1, 0, 0, 0, 0, 0, &
@ -318,7 +318,7 @@ module element
#else
],[NNODE(2),NCELLNODE(GEOMTYPE(2))])
#endif
integer, dimension(NNODE(3),NCELLNODE(GEOMTYPE(3))), parameter :: CELLNODEPARENTNODEWEIGHTS3 = &
reshape([&
1, 0, 0, 0, &
@ -335,7 +335,7 @@ module element
#else
],[NNODE(3),NCELLNODE(GEOMTYPE(3))])
#endif
integer, dimension(NNODE(4),NCELLNODE(GEOMTYPE(4))), parameter :: CELLNODEPARENTNODEWEIGHTS4 = &
reshape([&
1, 0, 0, 0, 0, 0, 0, 0, &
@ -359,7 +359,7 @@ module element
#else
],[NNODE(4),NCELLNODE(GEOMTYPE(4))])
#endif
integer, dimension(NNODE(5),NCELLNODE(GEOMTYPE(5))), parameter :: CELLNODEPARENTNODEWEIGHTS5 = &
reshape([&
1, 0, 0, 0, 0, 0, 0, 0, &
@ -376,7 +376,7 @@ module element
#else
],[NNODE(5),NCELLNODE(GEOMTYPE(5))])
#endif
integer, dimension(NNODE(6),NcellNode(GEOMTYPE(6))), parameter :: CELLNODEPARENTNODEWEIGHTS6 = &
reshape([&
1, 0, 0, 0, &
@ -388,7 +388,7 @@ module element
#else
],[NNODE(6),NcellNode(GEOMTYPE(6))])
#endif
integer, dimension(NNODE(7),NCELLNODE(GEOMTYPE(7))), parameter :: CELLNODEPARENTNODEWEIGHTS7 = &
reshape([&
1, 0, 0, 0, 0, &
@ -411,7 +411,7 @@ module element
#else
],[NNODE(7),NCELLNODE(GEOMTYPE(7))])
#endif
integer, dimension(NNODE(8),NCELLNODE(GEOMTYPE(8))), parameter :: CELLNODEPARENTNODEWEIGHTS8 = &
reshape([&
1, 0, 0, 0, 0, 0, 0, 0, 0, 0, &
@ -434,7 +434,7 @@ module element
#else
],[NNODE(8),NCELLNODE(GEOMTYPE(8))])
#endif
integer, dimension(NNODE(9),NCELLNODE(GEOMTYPE(9))), parameter :: CELLNODEPARENTNODEWEIGHTS9 = &
reshape([&
1, 0, 0, 0, 0, 0, &
@ -463,7 +463,7 @@ module element
#else
],[NNODE(9),NCELLNODE(GEOMTYPE(9))])
#endif
integer, dimension(NNODE(10),NCELLNODE(GEOMTYPE(10))), parameter :: CELLNODEPARENTNODEWEIGHTS10 = &
reshape([&
1, 0, 0, 0, 0, 0, 0, 0, &
@ -479,7 +479,7 @@ module element
#else
],[NNODE(10),NCELLNODE(GEOMTYPE(10))])
#endif
integer, dimension(NNODE(11),NCELLNODE(GEOMTYPE(11))), parameter :: CELLNODEPARENTNODEWEIGHTS11 = &
reshape([&
1, 0, 0, 0, 0, 0, 0, 0, & !
@ -514,7 +514,7 @@ module element
#else
],[NNODE(11),NCELLNODE(GEOMTYPE(11))])
#endif
integer, dimension(NNODE(12),NCELLNODE(GEOMTYPE(12))), parameter :: CELLNODEPARENTNODEWEIGHTS12 = &
reshape([&
1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, & !
@ -549,7 +549,7 @@ module element
#else
],[NNODE(12),NCELLNODE(GEOMTYPE(12))])
#endif
integer, dimension(NNODE(13),NCELLNODE(GEOMTYPE(13))), parameter :: CELLNODEPARENTNODEWEIGHTS13 = &
reshape([&
1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, & !
@ -621,8 +621,8 @@ module element
#else
],[NNODE(13),NCELLNODE(GEOMTYPE(13))])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(1)),NIP(1)), parameter :: CELL1 = &
reshape([&
1,2,3 &
@ -631,7 +631,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(1)),NIP(1)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(2)),NIP(2)), parameter :: CELL2 = &
reshape([&
1, 4, 7, 6, &
@ -642,7 +642,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(2)),NIP(2)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(3)),NIP(3)), parameter :: CELL3 = &
reshape([&
1, 5, 9, 8, &
@ -654,7 +654,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(3)),NIP(3)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(4)),NIP(4)), parameter :: CELL4 = &
reshape([&
1, 5,13,12, &
@ -671,7 +671,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(4)),NIP(4)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(5)),NIP(5)), parameter :: CELL5 = &
reshape([&
1, 2, 3, 4 &
@ -680,7 +680,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(5)),NIP(5)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(6)),NIP(6)), parameter :: CELL6 = &
reshape([&
1, 5,11, 7, 8,12,15,14, &
@ -692,7 +692,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(6)),NIP(6)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(7)),NIP(7)), parameter :: CELL7 = &
reshape([&
1, 7,16, 9,10,17,21,19, &
@ -706,7 +706,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(7)),NIP(7)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(8)),NIP(8)), parameter :: CELL8 = &
reshape([&
1, 2, 3, 4, 5, 6, 7, 8 &
@ -715,7 +715,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(8)),NIP(8)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(9)),NIP(9)), parameter :: CELL9 = &
reshape([&
1, 9,21,12,17,22,27,25, &
@ -731,7 +731,7 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(9)),NIP(9)])
#endif
integer, dimension(NCELLNODEPERCELL(CELLTYPE(10)),NIP(10)), parameter :: CELL10 = &
reshape([&
1, 9,33,16,17,37,57,44, &
@ -766,8 +766,8 @@ module element
#else
],[NCELLNODEPERCELL(CELLTYPE(10)),NIP(10)])
#endif
integer, dimension(NCELLNODEPERCELLFACE(1),NIPNEIGHBOR(1)), parameter :: CELLFACE1 = &
reshape([&
2,3, &
@ -778,7 +778,7 @@ module element
#else
],[NCELLNODEPERCELLFACE(1),NIPNEIGHBOR(1)])
#endif
integer, dimension(NCELLNODEPERCELLFACE(2),NIPNEIGHBOR(2)), parameter :: CELLFACE2 = &
reshape([&
2,3, &
@ -790,7 +790,7 @@ module element
#else
],[NCELLNODEPERCELLFACE(2),NIPNEIGHBOR(2)])
#endif
integer, dimension(NCELLNODEPERCELLFACE(3),NIPNEIGHBOR(3)), parameter :: CELLFACE3 = &
reshape([&
1,3,2, &
@ -802,7 +802,7 @@ module element
#else
],[NCELLNODEPERCELLFACE(3),NIPNEIGHBOR(3)])
#endif
integer, dimension(NCELLNODEPERCELLFACE(4),NIPNEIGHBOR(4)), parameter :: CELLFACE4 = &
reshape([&
2,3,7,6, &
@ -816,10 +816,10 @@ module element
#else
],[NCELLNODEPERCELLFACE(4),NIPNEIGHBOR(4)])
#endif
contains
!---------------------------------------------------------------------------------------------------
!> define properties of an element
@ -828,9 +828,9 @@ subroutine tElement_init(self,elemType)
class(tElement) :: self
integer, intent(in) :: elemType
self%elemType = elemType
self%Nnodes = NNODE (self%elemType)
self%geomType = GEOMTYPE(self%elemType)
@ -864,12 +864,12 @@ subroutine tElement_init(self,elemType)
case default
call IO_error(0,ext_msg='invalid element type')
end select
self%NcellNodes = NCELLNODE(self%geomType)
self%nIPs = NIP (self%geomType)
self%cellType = CELLTYPE (self%geomType)
select case (self%geomType)
case(1)
self%IPneighbor = IPNEIGHBOR1
@ -904,7 +904,7 @@ subroutine tElement_init(self,elemType)
end select
self%NcellnodesPerCell = NCELLNODEPERCELL(self%cellType)
select case(self%cellType)
case(1)
self%cellFace = CELLFACE1
@ -919,20 +919,20 @@ subroutine tElement_init(self,elemType)
self%cellFace = CELLFACE4
self%vtkType = 'HEXAHEDRON'
end select
self%nIPneighbors = size(self%IPneighbor,1)
write(6,'(/,a)') ' <<<+- element_init -+>>>'; flush(6)
write(6,*) ' element type: ',self%elemType
write(6,*) ' geom type: ',self%geomType
write(6,*) ' cell type: ',self%cellType
write(6,*) ' # node: ',self%Nnodes
write(6,*) ' # IP: ',self%nIPs
write(6,*) ' # cellnode: ',self%Ncellnodes
write(6,*) ' # cellnode/cell: ',self%NcellnodesPerCell
write(6,*) ' # IP neighbor: ',self%nIPneighbors
print'(/,a)', ' <<<+- element_init -+>>>'; flush(6)
print*, 'element type: ',self%elemType
print*, ' geom type: ',self%geomType
print*, ' cell type: ',self%cellType
print*, ' # node: ',self%Nnodes
print*, ' # IP: ',self%nIPs
print*, ' # cellnode: ',self%Ncellnodes
print*, ' # cellnode/cell: ',self%NcellnodesPerCell
print*, ' # IP neighbor: ',self%nIPneighbors
end subroutine tElement_init
end module element

View File

@ -10,6 +10,7 @@ program DAMASK_grid
#include <petsc/finclude/petscsys.h>
use PETScsys
use prec
use parallelization
use DAMASK_interface
use IO
use config
@ -23,7 +24,7 @@ program DAMASK_grid
use grid_damage_spectral
use grid_thermal_spectral
use results
implicit none
!--------------------------------------------------------------------------------------------------
@ -87,7 +88,7 @@ program DAMASK_grid
mech_updateCoords
procedure(grid_mech_spectral_basic_restartWrite), pointer :: &
mech_restartWrite
external :: &
quit
class (tNode), pointer :: &
@ -96,10 +97,10 @@ program DAMASK_grid
!--------------------------------------------------------------------------------------------------
! init DAMASK (all modules)
call CPFEM_initAll
call CPFEM_initAll
write(6,'(/,a)') ' <<<+- DAMASK_spectral init -+>>>'; flush(6)
write(6,'(/,a)') ' Shanthraj et al., Handbook of Mechanics of Materials, 2019'
write(6,'(a)') ' https://doi.org/10.1007/978-981-10-6855-3_80'
@ -113,7 +114,7 @@ program DAMASK_grid
!-------------------------------------------------------------------------------------------------
! reading field paramters from numerics file and do sanity checks
num_grid => numerics_root%get('grid', defaultVal=emptyDict)
num_grid => config_numerics%get('grid', defaultVal=emptyDict)
stagItMax = num_grid%get_asInt('maxStaggeredIter',defaultVal=10)
maxCutBack = num_grid%get_asInt('maxCutBack',defaultVal=3)
@ -122,16 +123,16 @@ program DAMASK_grid
!--------------------------------------------------------------------------------------------------
! assign mechanics solver depending on selected type
debug_grid => debug_root%get('grid',defaultVal=emptyList)
select case (trim(num_grid%get_asString('solver', defaultVal = 'Basic')))
debug_grid => config_debug%get('grid',defaultVal=emptyList)
select case (trim(num_grid%get_asString('solver', defaultVal = 'Basic')))
case ('Basic')
mech_init => grid_mech_spectral_basic_init
mech_forward => grid_mech_spectral_basic_forward
mech_solution => grid_mech_spectral_basic_solution
mech_updateCoords => grid_mech_spectral_basic_updateCoords
mech_restartWrite => grid_mech_spectral_basic_restartWrite
case ('Polarisation')
if(debug_grid%contains('basic')) &
call IO_warning(42, ext_msg='debug Divergence')
@ -140,7 +141,7 @@ program DAMASK_grid
mech_solution => grid_mech_spectral_polarisation_solution
mech_updateCoords => grid_mech_spectral_polarisation_updateCoords
mech_restartWrite => grid_mech_spectral_polarisation_restartWrite
case ('FEM')
if(debug_grid%contains('basic')) &
call IO_warning(42, ext_msg='debug Divergence')
@ -149,24 +150,24 @@ program DAMASK_grid
mech_solution => grid_mech_FEM_solution
mech_updateCoords => grid_mech_FEM_updateCoords
mech_restartWrite => grid_mech_FEM_restartWrite
case default
call IO_error(error_ID = 891, ext_msg = trim(num_grid%get_asString('solver')))
end select
!--------------------------------------------------------------------------------------------------
! reading information from load case file and to sanity checks
fileContent = IO_readlines(trim(loadCaseFile))
! reading information from load case file and to sanity checks
fileContent = IO_readlines(trim(interface_loadFile))
if(size(fileContent) == 0) call IO_error(307,ext_msg='No load case specified')
allocate (loadCases(0)) ! array of load cases
do currentLoadCase = 1, size(fileContent)
line = fileContent(currentLoadCase)
if (IO_isBlank(line)) cycle
chunkPos = IO_stringPos(line)
do i = 1, chunkPos(1) ! reading compulsory parameters for loadcase
select case (IO_lc(IO_stringValue(line,chunkPos,i)))
case('l','fdot','dotf','f')
@ -178,8 +179,8 @@ program DAMASK_grid
end select
enddo
if ((N_def /= N_n) .or. (N_n /= N_t) .or. N_n < 1) & ! sanity check
call IO_error(error_ID=837,el=currentLoadCase,ext_msg = trim(loadCaseFile)) ! error message for incomplete loadcase
call IO_error(error_ID=837,el=currentLoadCase,ext_msg = trim(interface_loadFile)) ! error message for incomplete loadcase
newLoadCase%stress%myType='stress'
field = 1
newLoadCase%ID(field) = FIELD_MECH_ID ! mechanical active by default
@ -191,7 +192,7 @@ program DAMASK_grid
field = field + 1
newLoadCase%ID(field) = FIELD_DAMAGE_ID
endif damageActive
call newLoadCase%rot%fromEulers(real([0.0,0.0,0.0],pReal))
readIn: do i = 1, chunkPos(1)
select case (IO_lc(IO_stringValue(line,chunkPos,i)))
@ -257,9 +258,9 @@ program DAMASK_grid
call newLoadCase%rot%fromMatrix(math_9to33(temp_valueVector))
end select
enddo readIn
newLoadCase%followFormerTrajectory = merge(.true.,.false.,currentLoadCase > 1) ! by default, guess from previous load case
reportAndCheck: if (worldrank == 0) then
write (loadcase_string, '(i0)' ) currentLoadCase
write(6,'(/,1x,a,i0)') 'load case: ', currentLoadCase
@ -324,13 +325,13 @@ program DAMASK_grid
select case (loadCases(1)%ID(field))
case(FIELD_MECH_ID)
call mech_init
case(FIELD_THERMAL_ID)
call grid_thermal_spectral_init
case(FIELD_DAMAGE_ID)
call grid_damage_spectral_init
end select
enddo
@ -348,16 +349,16 @@ program DAMASK_grid
'.sta',form='FORMATTED', position='APPEND', status='OLD')
endif writeHeader
endif
writeUndeformed: if (interface_restartInc < 1) then
write(6,'(1/,a)') ' ... writing initial configuration to file ........................'
call CPFEM_results(0,0.0_pReal)
endif writeUndeformed
loadCaseLooping: do currentLoadCase = 1, size(loadCases)
time0 = time ! load case start time
guess = loadCases(currentLoadCase)%followFormerTrajectory ! change of load case? homogeneous guess for the first inc
incLooping: do inc = 1, loadCases(currentLoadCase)%incs
totalIncsCounter = totalIncsCounter + 1
@ -382,13 +383,13 @@ program DAMASK_grid
endif
endif
timeinc = timeinc * real(subStepFactor,pReal)**real(-cutBackLevel,pReal) ! depending on cut back level, decrease time step
skipping: if (totalIncsCounter <= interface_restartInc) then ! not yet at restart inc?
time = time + timeinc ! just advance time, skip already performed calculation
guess = .true. ! QUESTION:why forced guessing instead of inheriting loadcase preference
else skipping
stepFraction = 0 ! fraction scaled by stepFactor**cutLevel
subStepLooping: do while (stepFraction < subStepFactor**cutBackLevel)
remainingLoadCaseTime = loadCases(currentLoadCase)%time+time0 - time
time = time + timeinc ! forward target time
@ -417,7 +418,7 @@ program DAMASK_grid
deformation_BC = loadCases(currentLoadCase)%deformation, &
stress_BC = loadCases(currentLoadCase)%stress, &
rotation_BC = loadCases(currentLoadCase)%rot)
case(FIELD_THERMAL_ID); call grid_thermal_spectral_forward(cutBack)
case(FIELD_DAMAGE_ID); call grid_damage_spectral_forward(cutBack)
end select
@ -432,21 +433,21 @@ program DAMASK_grid
do field = 1, nActiveFields
select case(loadCases(currentLoadCase)%ID(field))
case(FIELD_MECH_ID)
solres(field) = mech_solution(&
solres(field) = mech_solution (&
incInfo, &
stress_BC = loadCases(currentLoadCase)%stress, &
rotation_BC = loadCases(currentLoadCase)%rot)
case(FIELD_THERMAL_ID)
solres(field) = grid_thermal_spectral_solution(timeinc,timeIncOld)
case(FIELD_DAMAGE_ID)
solres(field) = grid_damage_spectral_solution(timeinc,timeIncOld)
end select
if (.not. solres(field)%converged) exit ! no solution found
enddo
stagIter = stagIter + 1
stagIterate = stagIter < stagItMax &
@ -480,17 +481,17 @@ program DAMASK_grid
if (worldrank == 0) close(statUnit)
call quit(0) ! quit
endif
enddo subStepLooping
cutBackLevel = max(0, cutBackLevel - 1) ! try half number of subincs next inc
if (all(solres(:)%converged)) then
write(6,'(/,a,i0,a)') ' increment ', totalIncsCounter, ' converged'
else
write(6,'(/,a,i0,a)') ' increment ', totalIncsCounter, ' NOT converged'
endif; flush(6)
if (mod(inc,loadCases(currentLoadCase)%outputFrequency) == 0) then ! at output frequency
write(6,'(1/,a)') ' ... writing results to file ......................................'
flush(6)
@ -501,17 +502,17 @@ program DAMASK_grid
call CPFEM_restartWrite
endif
endif skipping
enddo incLooping
enddo loadCaseLooping
!--------------------------------------------------------------------------------------------------
! report summary of whole calculation
write(6,'(/,a)') ' ###########################################################################'
if (worldrank == 0) close(statUnit)
call quit(0) ! no complains ;)
end program DAMASK_grid

View File

@ -9,6 +9,7 @@ module discretization_grid
use PETScsys
use prec
use parallelization
use system_routines
use base64
use zlib
@ -66,12 +67,16 @@ subroutine discretization_grid_init(restart)
write(6,'(/,a)') ' <<<+- discretization_grid init -+>>>'; flush(6)
if(index(geometryFile,'.vtr') /= 0) then
if(index(interface_geomFile,'.vtr') /= 0) then
call readVTR(grid,geomSize,origin,microstructureAt)
else
call readGeom(grid,geomSize,origin,microstructureAt)
endif
print'(/,a,3(i12 ))', ' grid a b c: ', grid
print'(a,3(es12.5))', ' size x y z: ', geomSize
print'(a,3(es12.5))', ' origin x y z: ', origin
!--------------------------------------------------------------------------------------------------
! grid solver specific quantities
if(worldsize>grid(3)) call IO_error(894, ext_msg='number of processes exceeds grid(3)')
@ -92,8 +97,8 @@ subroutine discretization_grid_init(restart)
!-------------------------------------------------------------------------------------------------
! debug parameters
debug_element = debug_root%get_asInt('element',defaultVal=1)
debug_ip = debug_root%get_asInt('integrationpoint',defaultVal=1)
debug_element = config_debug%get_asInt('element',defaultVal=1)
debug_ip = config_debug%get_asInt('integrationpoint',defaultVal=1)
!--------------------------------------------------------------------------------------------------
! general discretization
@ -172,10 +177,10 @@ subroutine readGeom(grid,geomSize,origin,microstructure)
!--------------------------------------------------------------------------------------------------
! read raw data as stream
inquire(file = trim(geometryFile), size=fileLength)
open(newunit=fileUnit, file=trim(geometryFile), access='stream',&
inquire(file = trim(interface_geomFile), size=fileLength)
open(newunit=fileUnit, file=trim(interface_geomFile), access='stream',&
status='old', position='rewind', action='read',iostat=myStat)
if(myStat /= 0) call IO_error(100,ext_msg=trim(geometryFile))
if(myStat /= 0) call IO_error(100,ext_msg=trim(interface_geomFile))
allocate(character(len=fileLength)::rawData)
read(fileUnit) rawData
close(fileUnit)
@ -326,10 +331,10 @@ subroutine readVTR(grid,geomSize,origin,microstructure)
!--------------------------------------------------------------------------------------------------
! read raw data as stream
inquire(file = trim(geometryFile), size=fileLength)
open(newunit=fileUnit, file=trim(geometryFile), access='stream',&
inquire(file = trim(interface_geomFile), size=fileLength)
open(newunit=fileUnit, file=trim(interface_geomFile), access='stream',&
status='old', position='rewind', action='read',iostat=myStat)
if(myStat /= 0) call IO_error(100,ext_msg=trim(geometryFile))
if(myStat /= 0) call IO_error(100,ext_msg=trim(interface_geomFile))
allocate(character(len=fileLength)::fileContent)
read(fileUnit) fileContent
close(fileUnit)
@ -433,7 +438,7 @@ subroutine readVTR(grid,geomSize,origin,microstructure)
coords = as_pReal(base64_str,headerType,compressed,dataType)
delta = coords(2:) - coords(:size(coords)-1)
if(any(delta<0.0_pReal) .or. dNeq(maxval(delta),minval(delta))) &
if(any(delta<0.0_pReal) .or. dNeq(maxval(delta),minval(delta),1.0e-8_pReal*maxval(abs(coords)))) &
call IO_error(error_ID = 844, ext_msg = 'grid spacing')
grid(direction) = size(coords)-1
@ -457,13 +462,13 @@ subroutine readVTR(grid,geomSize,origin,microstructure)
select case(dataType)
case('Int32')
as_Int = int(bytes_to_C_INT32_T(asBytes(base64_str,headerType,compressed)))
as_Int = int(prec_bytesToC_INT32_T(asBytes(base64_str,headerType,compressed)))
case('Int64')
as_Int = int(bytes_to_C_INT64_T(asBytes(base64_str,headerType,compressed)))
as_Int = int(prec_bytesToC_INT64_T(asBytes(base64_str,headerType,compressed)))
case('Float32')
as_Int = int(bytes_to_C_FLOAT (asBytes(base64_str,headerType,compressed)))
as_Int = int(prec_bytesToC_FLOAT (asBytes(base64_str,headerType,compressed)))
case('Float64')
as_Int = int(bytes_to_C_DOUBLE (asBytes(base64_str,headerType,compressed)))
as_Int = int(prec_bytesToC_DOUBLE (asBytes(base64_str,headerType,compressed)))
case default
call IO_error(844_pInt,ext_msg='unknown data type: '//trim(dataType))
end select
@ -485,13 +490,13 @@ subroutine readVTR(grid,geomSize,origin,microstructure)
select case(dataType)
case('Int32')
as_pReal = real(bytes_to_C_INT32_T(asBytes(base64_str,headerType,compressed)),pReal)
as_pReal = real(prec_bytesToC_INT32_T(asBytes(base64_str,headerType,compressed)),pReal)
case('Int64')
as_pReal = real(bytes_to_C_INT64_T(asBytes(base64_str,headerType,compressed)),pReal)
as_pReal = real(prec_bytesToC_INT64_T(asBytes(base64_str,headerType,compressed)),pReal)
case('Float32')
as_pReal = real(bytes_to_C_FLOAT (asBytes(base64_str,headerType,compressed)),pReal)
as_pReal = real(prec_bytesToC_FLOAT (asBytes(base64_str,headerType,compressed)),pReal)
case('Float64')
as_pReal = real(bytes_to_C_DOUBLE (asBytes(base64_str,headerType,compressed)),pReal)
as_pReal = real(prec_bytesToC_DOUBLE (asBytes(base64_str,headerType,compressed)),pReal)
case default
call IO_error(844_pInt,ext_msg='unknown data type: '//trim(dataType))
end select
@ -538,15 +543,15 @@ subroutine readVTR(grid,geomSize,origin,microstructure)
integer(pI64) :: headerLen, nBlock, b,s,e
if (headerType == 'UInt32') then
temp = int(bytes_to_C_INT32_T(base64_to_bytes(base64_str(:base64_nChar(4_pI64)))),pI64)
temp = int(prec_bytesToC_INT32_T(base64_to_bytes(base64_str(:base64_nChar(4_pI64)))),pI64)
nBlock = int(temp(1),pI64)
headerLen = 4_pI64 * (3_pI64 + nBlock)
temp = int(bytes_to_C_INT32_T(base64_to_bytes(base64_str(:base64_nChar(headerLen)))),pI64)
temp = int(prec_bytesToC_INT32_T(base64_to_bytes(base64_str(:base64_nChar(headerLen)))),pI64)
elseif(headerType == 'UInt64') then
temp = int(bytes_to_C_INT64_T(base64_to_bytes(base64_str(:base64_nChar(8_pI64)))),pI64)
temp = int(prec_bytesToC_INT64_T(base64_to_bytes(base64_str(:base64_nChar(8_pI64)))),pI64)
nBlock = int(temp(1),pI64)
headerLen = 8_pI64 * (3_pI64 + nBlock)
temp = int(bytes_to_C_INT64_T(base64_to_bytes(base64_str(:base64_nChar(headerLen)))),pI64)
temp = int(prec_bytesToC_INT64_T(base64_to_bytes(base64_str(:base64_nChar(headerLen)))),pI64)
endif
allocate(size_inflated(nBlock),source=temp(2))
@ -584,13 +589,13 @@ subroutine readVTR(grid,geomSize,origin,microstructure)
s=0_pI64
if (headerType == 'UInt32') then
do while(s+base64_nChar(4_pI64)<(len(base64_str,pI64)))
nByte = int(bytes_to_C_INT32_T(base64_to_bytes(base64_str(s+1_pI64:s+base64_nChar(4_pI64)))),pI64)
nByte = int(prec_bytesToC_INT32_T(base64_to_bytes(base64_str(s+1_pI64:s+base64_nChar(4_pI64)))),pI64)
bytes = [bytes,base64_to_bytes(base64_str(s+1_pI64:s+base64_nChar(4_pI64+nByte(1))),5_pI64)]
s = s + base64_nChar(4_pI64+nByte(1))
enddo
elseif(headerType == 'UInt64') then
do while(s+base64_nChar(8_pI64)<(len(base64_str,pI64)))
nByte = int(bytes_to_C_INT64_T(base64_to_bytes(base64_str(s+1_pI64:s+base64_nChar(8_pI64)))),pI64)
nByte = int(prec_bytesToC_INT64_T(base64_to_bytes(base64_str(s+1_pI64:s+base64_nChar(8_pI64)))),pI64)
bytes = [bytes,base64_to_bytes(base64_str(s+1_pI64:s+base64_nChar(8_pI64+nByte(1))),9_pI64)]
s = s + base64_nChar(8_pI64+nByte(1))
enddo

View File

@ -11,12 +11,13 @@ module grid_damage_spectral
use PETScsnes
use prec
use parallelization
use IO
use spectral_utilities
use discretization_grid
use damage_nonlocal
use config
use YAML_types
use config
implicit none
private
@ -83,12 +84,12 @@ subroutine grid_damage_spectral_init
!-------------------------------------------------------------------------------------------------
! read numerical parameters and do sanity checks
num_grid => numerics_root%get('grid',defaultVal=emptyDict)
num_grid => config_numerics%get('grid',defaultVal=emptyDict)
num%itmax = num_grid%get_asInt ('itmax',defaultVal=250)
num%eps_damage_atol = num_grid%get_asFloat ('eps_damage_atol',defaultVal=1.0e-2_pReal)
num%eps_damage_rtol = num_grid%get_asFloat ('eps_damage_rtol',defaultVal=1.0e-6_pReal)
num_generic => numerics_root%get('generic',defaultVal=emptyDict)
num_generic => config_numerics%get('generic',defaultVal=emptyDict)
num%residualStiffness = num_generic%get_asFloat('residualStiffness', defaultVal=1.0e-6_pReal)
if (num%residualStiffness < 0.0_pReal) call IO_error(301,ext_msg='residualStiffness')

View File

@ -11,6 +11,7 @@ module grid_mech_FEM
use PETScsnes
use prec
use parallelization
use DAMASK_interface
use HDF5_utilities
use math
@ -125,12 +126,12 @@ subroutine grid_mech_FEM_init
!-----------------------------------------------------------------------------------------------
! debugging options
debug_grid => debug_root%get('grid', defaultVal=emptyList)
debug_grid => config_debug%get('grid', defaultVal=emptyList)
debugRotation = debug_grid%contains('rotation')
!-------------------------------------------------------------------------------------------------
! read numerical parameter and do sanity checks
num_grid => numerics_root%get('grid',defaultVal=emptyDict)
num_grid => config_numerics%get('grid',defaultVal=emptyDict)
num%eps_div_atol = num_grid%get_asFloat ('eps_div_atol', defaultVal=1.0e-4_pReal)
num%eps_div_rtol = num_grid%get_asFloat ('eps_div_rtol', defaultVal=5.0e-4_pReal)
num%eps_stress_atol = num_grid%get_asFloat ('eps_stress_atol', defaultVal=1.0e3_pReal)

View File

@ -11,6 +11,7 @@ module grid_mech_spectral_basic
use PETScsnes
use prec
use parallelization
use DAMASK_interface
use HDF5_utilities
use math
@ -119,12 +120,12 @@ subroutine grid_mech_spectral_basic_init
!-------------------------------------------------------------------------------------------------
! debugging options
debug_grid => debug_root%get('grid', defaultVal=emptyList)
debug_grid => config_debug%get('grid', defaultVal=emptyList)
debugRotation = debug_grid%contains('rotation')
!-------------------------------------------------------------------------------------------------
! read numerical parameters and do sanity checks
num_grid => numerics_root%get('grid',defaultVal=emptyDict)
num_grid => config_numerics%get('grid',defaultVal=emptyDict)
num%update_gamma = num_grid%get_asBool ('update_gamma', defaultVal=.false.)
num%eps_div_atol = num_grid%get_asFloat ('eps_div_atol', defaultVal=1.0e-4_pReal)

View File

@ -11,6 +11,7 @@ module grid_mech_spectral_polarisation
use PETScsnes
use prec
use parallelization
use DAMASK_interface
use HDF5_utilities
use math
@ -129,12 +130,12 @@ subroutine grid_mech_spectral_polarisation_init
!------------------------------------------------------------------------------------------------
! debugging options
debug_grid => debug_root%get('grid',defaultVal=emptyList)
debug_grid => config_debug%get('grid',defaultVal=emptyList)
debugRotation = debug_grid%contains('rotation')
!-------------------------------------------------------------------------------------------------
! read numerical parameters
num_grid => numerics_root%get('grid',defaultVal=emptyDict)
num_grid => config_numerics%get('grid',defaultVal=emptyDict)
num%update_gamma = num_grid%get_asBool ('update_gamma', defaultVal=.false.)
num%eps_div_atol = num_grid%get_asFloat ('eps_div_atol', defaultVal=1.0e-4_pReal)

View File

@ -11,6 +11,7 @@ module grid_thermal_spectral
use PETScsnes
use prec
use parallelization
use IO
use spectral_utilities
use discretization_grid
@ -80,7 +81,7 @@ subroutine grid_thermal_spectral_init
!-------------------------------------------------------------------------------------------------
! read numerical parameter and do sanity checks
num_grid => numerics_root%get('grid',defaultVal=emptyDict)
num_grid => config_numerics%get('grid',defaultVal=emptyDict)
num%itmax = num_grid%get_asInt ('itmax', defaultVal=250)
num%eps_thermal_atol = num_grid%get_asFloat ('eps_thermal_atol',defaultVal=1.0e-2_pReal)
num%eps_thermal_rtol = num_grid%get_asFloat ('eps_thermal_rtol',defaultVal=1.0e-6_pReal)

View File

@ -10,11 +10,11 @@ module spectral_utilities
use prec
use DAMASK_interface
use parallelization
use math
use rotations
use IO
use discretization_grid
use config
use discretization
use homogenization
@ -109,15 +109,10 @@ module spectral_utilities
end type tSolutionParams
type, private :: tNumerics
real(pReal) :: &
FFTW_timelimit !< timelimit for FFTW plan creation, see www.fftw.org
integer :: &
divergence_correction !< scale divergence/curl calculation: [0: no correction, 1: size scaled to 1, 2: size scaled to Npoints]
logical :: &
memory_efficient !< calculate gamma operator on the fly
character(len=:), allocatable :: &
spectral_derivative, & !< approximation used for derivatives in Fourier space
FFTW_plan_mode !< FFTW plan mode, see www.fftw.org
end type tNumerics
type(tNumerics), private :: num ! numerics parameters. Better name?
@ -191,23 +186,23 @@ subroutine spectral_utilities_init
num_grid, &
debug_grid ! pointer to grid debug options
write(6,'(/,a)') ' <<<+- spectral_utilities init -+>>>'
print'(/,a)', ' <<<+- spectral_utilities init -+>>>'
write(6,'(/,a)') ' Diehl, Diploma Thesis TU München, 2010'
write(6,'(a)') ' https://doi.org/10.13140/2.1.3234.3840'
print*, 'Diehl, Diploma Thesis TU München, 2010'
print*, 'https://doi.org/10.13140/2.1.3234.3840'//IO_EOL
write(6,'(/,a)') ' Eisenlohr et al., International Journal of Plasticity 46:3753, 2013'
write(6,'(a)') ' https://doi.org/10.1016/j.ijplas.2012.09.012'
print*, 'Eisenlohr et al., International Journal of Plasticity 46:3753, 2013'
print*, 'https://doi.org/10.1016/j.ijplas.2012.09.012'//IO_EOL
write(6,'(/,a)') ' Shanthraj et al., International Journal of Plasticity 66:3145, 2015'
write(6,'(a)') ' https://doi.org/10.1016/j.ijplas.2014.02.006'
print*, 'Shanthraj et al., International Journal of Plasticity 66:3145, 2015'
print*, 'https://doi.org/10.1016/j.ijplas.2014.02.006'//IO_EOL
write(6,'(/,a)') ' Shanthraj et al., Handbook of Mechanics of Materials, 2019'
write(6,'(a)') ' https://doi.org/10.1007/978-981-10-6855-3_80'
print*, 'Shanthraj et al., Handbook of Mechanics of Materials, 2019'
print*, 'https://doi.org/10.1007/978-981-10-6855-3_80'
!--------------------------------------------------------------------------------------------------
! set debugging parameters
debug_grid => debug_root%get('grid',defaultVal=emptyList)
debug_grid => config_debug%get('grid',defaultVal=emptyList)
debugGeneral = debug_grid%contains('basic')
debugRotation = debug_grid%contains('rotation')
debugPETSc = debug_grid%contains('petsc')
@ -218,7 +213,7 @@ subroutine spectral_utilities_init
trim(PETScDebug), &
' add more using the PETSc_Options keyword in numerics.yaml '; flush(6)
num_grid => numerics_root%get('grid',defaultVal=emptyDict)
num_grid => config_numerics%get('grid',defaultVal=emptyDict)
call PETScOptionsClear(PETSC_NULL_OPTIONS,ierr)
CHKERRQ(ierr)
@ -231,19 +226,13 @@ subroutine spectral_utilities_init
grid1Red = grid(1)/2 + 1
wgt = 1.0/real(product(grid),pReal)
write(6,'(/,a,3(i12 ))') ' grid a b c: ', grid
write(6,'(a,3(es12.5))') ' size x y z: ', geomSize
num%memory_efficient = num_grid%get_asInt ('memory_efficient', defaultVal=1) > 0
num%FFTW_timelimit = num_grid%get_asFloat ('fftw_timelimit', defaultVal=-1.0_pReal)
num%divergence_correction = num_grid%get_asInt ('divergence_correction', defaultVal=2)
num%spectral_derivative = num_grid%get_asString('derivative', defaultVal='continuous')
num%FFTW_plan_mode = num_grid%get_asString('fftw_plan_mode', defaultVal='FFTW_MEASURE')
num%memory_efficient = num_grid%get_asInt('memory_efficient', defaultVal=1) > 0 ! ToDo: should be logical in YAML file
num%divergence_correction = num_grid%get_asInt('divergence_correction', defaultVal=2)
if (num%divergence_correction < 0 .or. num%divergence_correction > 2) &
call IO_error(301,ext_msg='divergence_correction')
select case (num%spectral_derivative)
select case (num_grid%get_asString('derivative',defaultVal='continuous'))
case ('continuous')
spectral_derivative_ID = DERIVATIVE_CONTINUOUS_ID
case ('central_difference')
@ -251,7 +240,7 @@ subroutine spectral_utilities_init
case ('FWBW_difference')
spectral_derivative_ID = DERIVATIVE_FWBW_DIFF_ID
case default
call IO_error(892,ext_msg=trim(num%spectral_derivative))
call IO_error(892,ext_msg=trim(num_grid%get_asString('derivative')))
end select
!--------------------------------------------------------------------------------------------------
@ -272,8 +261,7 @@ subroutine spectral_utilities_init
scaledGeomSize = geomSize
endif
select case(IO_lc(num%FFTW_plan_mode)) ! setting parameters for the plan creation of FFTW. Basically a translation from fftw3.f
select case(IO_lc(num_grid%get_asString('fftw_plan_mode',defaultVal='FFTW_MEASURE')))
case('fftw_estimate') ! ordered from slow execution (but fast plan creation) to fast execution
FFTW_planner_flag = FFTW_ESTIMATE
case('fftw_measure')
@ -283,14 +271,14 @@ subroutine spectral_utilities_init
case('fftw_exhaustive')
FFTW_planner_flag = FFTW_EXHAUSTIVE
case default
call IO_warning(warning_ID=47,ext_msg=trim(IO_lc(num%FFTW_plan_mode)))
call IO_warning(warning_ID=47,ext_msg=trim(IO_lc(num_grid%get_asString('fftw_plan_mode'))))
FFTW_planner_flag = FFTW_MEASURE
end select
!--------------------------------------------------------------------------------------------------
! general initialization of FFTW (see manual on fftw.org for more details)
if (pReal /= C_DOUBLE .or. kind(1) /= C_INT) call IO_error(0,ext_msg='Fortran to C') ! check for correct precision in C
call fftw_set_timelimit(num%FFTW_timelimit) ! set timelimit for plan creation
if (pReal /= C_DOUBLE .or. kind(1) /= C_INT) error stop 'C and Fortran datatypes do not match'
call fftw_set_timelimit(num_grid%get_asFloat('fftw_timelimit',defaultVal=-1.0_pReal))
if (debugGeneral) write(6,'(/,a)') ' FFTW initialized'; flush(6)
@ -731,7 +719,7 @@ function utilities_maskedCompliance(rot_BC,mask_stress,C)
!--------------------------------------------------------------------------------------------------
! check if inversion was successful
sTimesC = matmul(c_reduced,s_reduced)
errmatinv = errmatinv .or. any(dNeq(sTimesC,math_identity2nd(size_reduced),1.0e-12_pReal))
errmatinv = errmatinv .or. any(dNeq(sTimesC,math_eye(size_reduced),1.0e-12_pReal))
if (debugGeneral .or. errmatinv) then
write(formatString, '(i2)') size_reduced
formatString = '(/,a,/,'//trim(formatString)//'('//trim(formatString)//'(2x,es9.2,1x)/))'
@ -1104,11 +1092,13 @@ end subroutine utilities_updateCoords
subroutine utilities_saveReferenceStiffness
integer :: &
fileUnit
fileUnit,ierr
if (worldrank == 0) then
write(6,'(a)') ' writing reference stiffness data required for restart to file'; flush(6)
fileUnit = IO_open_binary(trim(getSolverJobName())//'.C_ref','w')
open(newunit=fileUnit, file=getSolverJobName()//'.C_ref',&
status='replace',access='stream',action='write',iostat=ierr)
if(ierr /=0) call IO_error(100,ext_msg='could not open file '//getSolverJobName()//'.C_ref')
write(fileUnit) C_ref
close(fileUnit)
endif

View File

@ -149,20 +149,20 @@ subroutine homogenization_init
num_homogGeneric, &
debug_homogenization
debug_homogenization => debug_root%get('homogenization', defaultVal=emptyList)
debug_homogenization => config_debug%get('homogenization', defaultVal=emptyList)
debugHomog%basic = debug_homogenization%contains('basic')
debugHomog%extensive = debug_homogenization%contains('extensive')
debugHomog%selective = debug_homogenization%contains('selective')
debugHomog%element = debug_root%get_asInt('element',defaultVal = 1)
debugHomog%ip = debug_root%get_asInt('integrationpoint',defaultVal = 1)
debugHomog%grain = debug_root%get_asInt('grain',defaultVal = 1)
debugHomog%element = config_debug%get_asInt('element',defaultVal = 1)
debugHomog%ip = config_debug%get_asInt('integrationpoint',defaultVal = 1)
debugHomog%grain = config_debug%get_asInt('grain',defaultVal = 1)
if (debugHomog%grain < 1 &
.or. debugHomog%grain > homogenization_Ngrains(material_homogenizationAt(debugHomog%element))) &
call IO_error(602,ext_msg='constituent', el=debugHomog%element, g=debugHomog%grain)
num_homog => numerics_root%get('homogenization',defaultVal=emptyDict)
num_homog => config_numerics%get('homogenization',defaultVal=emptyDict)
num_homogMech => num_homog%get('mech',defaultVal=emptyDict)
num_homogGeneric => num_homog%get('generic',defaultVal=emptyDict)
@ -186,7 +186,7 @@ subroutine homogenization_init
materialpoint_F = materialpoint_F0 ! initialize to identity
allocate(materialpoint_P(3,3,discretization_nIP,discretization_nElem), source=0.0_pReal)
write(6,'(/,a)') ' <<<+- homogenization init -+>>>'; flush(6)
print'(/,a)', ' <<<+- homogenization init -+>>>'; flush(6)
num%nMPstate = num_homogGeneric%get_asInt ('nMPstate', defaultVal=10)
num%subStepMinHomog = num_homogGeneric%get_asFloat('subStepMin', defaultVal=1.0e-3_pReal)
@ -228,11 +228,11 @@ subroutine materialpoint_stressAndItsTangent(updateJaco,dt)
#ifdef DEBUG
if (debugHomog%basic) then
write(6,'(/a,i5,1x,i2)') '<< HOMOG >> Material Point start at el ip ', debugHomog%element, debugHomog%ip
print'(/a,i5,1x,i2)', ' << HOMOG >> Material Point start at el ip ', debugHomog%element, debugHomog%ip
write(6,'(a,/,3(12x,3(f14.9,1x)/))') '<< HOMOG >> F0', &
print'(a,/,3(12x,3(f14.9,1x)/))', ' << HOMOG >> F0', &
transpose(materialpoint_F0(1:3,1:3,debugHomog%ip,debugHomog%element))
write(6,'(a,/,3(12x,3(f14.9,1x)/))') '<< HOMOG >> F', &
print'(a,/,3(12x,3(f14.9,1x)/))', ' << HOMOG >> F', &
transpose(materialpoint_F(1:3,1:3,debugHomog%ip,debugHomog%element))
endif
#endif
@ -292,12 +292,11 @@ subroutine materialpoint_stressAndItsTangent(updateJaco,dt)
if (converged(i,e)) then
#ifdef DEBUG
if (debugHomog%extensive &
.and. ((e == debugHomog%element .and. i == debugHomog%ip) &
.or. .not. debugHomog%selective)) then
write(6,'(a,1x,f12.8,1x,a,1x,f12.8,1x,a,i8,1x,i2/)') '<< HOMOG >> winding forward from', &
subFrac(i,e), 'to current subFrac', &
subFrac(i,e)+subStep(i,e),'in materialpoint_stressAndItsTangent at el ip',e,i
if (debugHomog%extensive .and. ((e == debugHomog%element .and. i == debugHomog%ip) &
.or. .not. debugHomog%selective)) then
print'(a,f12.8,a,f12.8,a,i8,1x,i2/)', ' << HOMOG >> winding forward from ', &
subFrac(i,e), ' to current subFrac ', &
subFrac(i,e)+subStep(i,e),' in materialpoint_stressAndItsTangent at el ip ',e,i
endif
#endif
@ -342,20 +341,17 @@ subroutine materialpoint_stressAndItsTangent(updateJaco,dt)
num%subStepSizeHomog * subStep(i,e) <= num%subStepMinHomog ) then ! would require too small subStep
! cutback makes no sense
if (.not. terminallyIll) then ! so first signals terminally ill...
!$OMP CRITICAL (write2out)
write(6,*) 'Integration point ', i,' at element ', e, ' terminally ill'
!$OMP END CRITICAL (write2out)
print*, ' Integration point ', i,' at element ', e, ' terminally ill'
endif
terminallyIll = .true. ! ...and kills all others
else ! cutback makes sense
subStep(i,e) = num%subStepSizeHomog * subStep(i,e) ! crystallite had severe trouble, so do a significant cutback
#ifdef DEBUG
if (debugHomog%extensive &
.and. ((e == debugHomog%element .and. i == debugHomog%ip) &
.or. .not. debugHomog%selective)) then
write(6,'(a,1x,f12.8,a,i8,1x,i2/)') &
'<< HOMOG >> cutback step in materialpoint_stressAndItsTangent with new subStep:',&
if (debugHomog%extensive .and. ((e == debugHomog%element .and. i == debugHomog%ip) &
.or. .not. debugHomog%selective)) then
print'(a,f12.8,a,i8,1x,i2/)', &
'<< HOMOG >> cutback step in materialpoint_stressAndItsTangent with new subStep: ',&
subStep(i,e),' at el ip',e,i
endif
#endif
@ -469,7 +465,7 @@ subroutine materialpoint_stressAndItsTangent(updateJaco,dt)
enddo elementLooping4
!$OMP END PARALLEL DO
else
write(6,'(/,a,/)') '<< HOMOG >> Material Point terminally ill'
print'(/,a,/)', ' << HOMOG >> Material Point terminally ill'
endif
end subroutine materialpoint_stressAndItsTangent

View File

@ -92,16 +92,18 @@ module subroutine mech_RGC_init(num_homogMech)
homog, &
homogMech
write(6,'(/,a)') ' <<<+- homogenization_mech_rgc init -+>>>'
write(6,'(/,a)') ' Tjahjanto et al., International Journal of Material Forming 2(1):939942, 2009'
write(6,'(a)') ' https://doi.org/10.1007/s12289-009-0619-1'
write(6,'(/,a)') ' Tjahjanto et al., Modelling and Simulation in Materials Science and Engineering 18:015006, 2010'
write(6,'(a)') ' https://doi.org/10.1088/0965-0393/18/1/015006'
print'(/,a)', ' <<<+- homogenization_mech_rgc init -+>>>'
Ninstance = count(homogenization_type == HOMOGENIZATION_RGC_ID)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
print*, 'Tjahjanto et al., International Journal of Material Forming 2(1):939942, 2009'
print*, 'https://doi.org/10.1007/s12289-009-0619-1'//IO_EOL
print*, 'Tjahjanto et al., Modelling and Simulation in Materials Science and Engineering 18:015006, 2010'
print*, 'https://doi.org/10.1088/0965-0393/18/1/015006'//IO_EOL
allocate(param(Ninstance))
allocate(state(Ninstance))
@ -139,7 +141,7 @@ module subroutine mech_RGC_init(num_homogMech)
if (num%volDiscrPow <= 0.0_pReal) call IO_error(301,ext_msg='volDiscrPw_RGC')
material_homogenization => material_root%get('homogenization')
material_homogenization => config_material%get('homogenization')
do h = 1, size(homogenization_type)
if (homogenization_type(h) /= HOMOGENIZATION_RGC_ID) cycle
homog => material_homogenization%get(h)
@ -240,11 +242,11 @@ module subroutine mech_RGC_partitionDeformation(F,avgF,instance,of)
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a32,1x,i3)')'Deformation gradient of grain: ',iGrain
print'(a,i3)',' Deformation gradient of grain: ',iGrain
do i = 1,3
write(6,'(1x,3(e15.8,1x))')(F(i,j,iGrain), j = 1,3)
print'(1x,3(e15.8,1x))',(F(i,j,iGrain), j = 1,3)
enddo
write(6,*)' '
print*,' '
flush(6)
endif
#endif
@ -303,11 +305,11 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a30)')'Obtained state: '
print*, 'Obtained state: '
do i = 1,size(stt%relaxationVector(:,of))
write(6,'(1x,2(e15.8,1x))') stt%relaxationVector(i,of)
print'(1x,2(e15.8,1x))', stt%relaxationVector(i,of)
enddo
write(6,*)' '
print*,' '
endif
#endif
@ -319,22 +321,6 @@ module procedure mech_RGC_updateState
! calculating volume discrepancy and stress penalty related to overall volume discrepancy
call volumePenalty(D,dst%volumeDiscrepancy(of),avgF,F,nGrain,instance,of)
#ifdef DEBUG
if (debugHomog%extensive) then
do iGrain = 1,nGrain
write(6,'(1x,a30,1x,i3,1x,a4,3(1x,e15.8))')'Mismatch magnitude of grain(',iGrain,') :',&
NN(1,iGrain),NN(2,iGrain),NN(3,iGrain)
write(6,'(/,1x,a30,1x,i3)')'Stress and penalties of grain: ',iGrain
do i = 1,3
write(6,'(1x,3(e15.8,1x),1x,3(e15.8,1x),1x,3(e15.8,1x))')(P(i,j,iGrain), j = 1,3), &
(R(i,j,iGrain), j = 1,3), &
(D(i,j,iGrain), j = 1,3)
enddo
write(6,*)' '
enddo
endif
#endif
!------------------------------------------------------------------------------------------------
! computing the residual stress from the balance of traction at all (interior) interfaces
do iNum = 1,nIntFaceTot
@ -369,9 +355,9 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a30,1x,i3)')'Traction at interface: ',iNum
write(6,'(1x,3(e15.8,1x))')(tract(iNum,j), j = 1,3)
write(6,*)' '
print'(a,i3)',' Traction at interface: ',iNum
print'(1x,3(e15.8,1x))',(tract(iNum,j), j = 1,3)
print*,' '
endif
#endif
enddo
@ -385,12 +371,11 @@ module procedure mech_RGC_updateState
if (debugHomog%extensive .and. prm%of_debug == of) then
stresLoc = maxloc(abs(P))
residLoc = maxloc(abs(tract))
write(6,'(1x,a)')' '
write(6,'(1x,a,1x,i2,1x,i4)')'RGC residual check ...',ip,el
write(6,'(1x,a15,1x,e15.8,1x,a7,i3,1x,a12,i2,i2)')'Max stress: ',stresMax, &
'@ grain',stresLoc(3),'in component',stresLoc(1),stresLoc(2)
write(6,'(1x,a15,1x,e15.8,1x,a7,i3,1x,a12,i2)')'Max residual: ',residMax, &
'@ iface',residLoc(1),'in direction',residLoc(2)
print'(a,i2,1x,i4)',' RGC residual check ... ',ip,el
print'(a,e15.8,a,i3,a,i2,i2)', ' Max stress: ',stresMax, &
'@ grain ',stresLoc(3),' in component ',stresLoc(1),stresLoc(2)
print'(a,e15.8,a,i3,a,i2)',' Max residual: ',residMax, &
' @ iface ',residLoc(1),' in direction ',residLoc(2)
flush(6)
endif
#endif
@ -403,7 +388,7 @@ module procedure mech_RGC_updateState
mech_RGC_updateState = .true.
#ifdef DEBUG
if (debugHomog%extensive .and. prm%of_debug == of) &
write(6,'(1x,a55,/)')'... done and happy'; flush(6)
print*, '... done and happy'; flush(6)
#endif
!--------------------------------------------------------------------------------------------------
@ -423,14 +408,14 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive .and. prm%of_debug == of) then
write(6,'(1x,a30,1x,e15.8)') 'Constitutive work: ',stt%work(of)
write(6,'(1x,a30,3(1x,e15.8))')'Magnitude mismatch: ',dst%mismatch(1,of), &
dst%mismatch(2,of), &
dst%mismatch(3,of)
write(6,'(1x,a30,1x,e15.8)') 'Penalty energy: ', stt%penaltyEnergy(of)
write(6,'(1x,a30,1x,e15.8,/)') 'Volume discrepancy: ', dst%volumeDiscrepancy(of)
write(6,'(1x,a30,1x,e15.8)') 'Maximum relaxation rate: ', dst%relaxationRate_max(of)
write(6,'(1x,a30,1x,e15.8,/)') 'Average relaxation rate: ', dst%relaxationRate_avg(of)
print'(a,e15.8)', ' Constitutive work: ',stt%work(of)
print'(a,3(1x,e15.8))', ' Magnitude mismatch: ',dst%mismatch(1,of), &
dst%mismatch(2,of), &
dst%mismatch(3,of)
print'(a,e15.8)', ' Penalty energy: ', stt%penaltyEnergy(of)
print'(a,e15.8,/)', ' Volume discrepancy: ', dst%volumeDiscrepancy(of)
print'(a,e15.8)', ' Maximum relaxation rate: ', dst%relaxationRate_max(of)
print'(a,e15.8,/)', ' Average relaxation rate: ', dst%relaxationRate_avg(of)
flush(6)
endif
#endif
@ -444,7 +429,7 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive .and. prm%of_debug == of) &
write(6,'(1x,a,/)') '... broken'; flush(6)
print'(a,/)', ' ... broken'; flush(6)
#endif
return
@ -452,7 +437,7 @@ module procedure mech_RGC_updateState
else ! proceed with computing the Jacobian and state update
#ifdef DEBUG
if (debugHomog%extensive .and. prm%of_debug == of) &
write(6,'(1x,a,/)') '... not yet done'; flush(6)
print'(a,/)', ' ... not yet done'; flush(6)
#endif
endif
@ -509,11 +494,11 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a30)')'Jacobian matrix of stress'
print*, 'Jacobian matrix of stress'
do i = 1,3*nIntFaceTot
write(6,'(1x,100(e11.4,1x))')(smatrix(i,j), j = 1,3*nIntFaceTot)
print'(1x,100(e11.4,1x))',(smatrix(i,j), j = 1,3*nIntFaceTot)
enddo
write(6,*)' '
print*,' '
flush(6)
endif
#endif
@ -569,11 +554,11 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a30)')'Jacobian matrix of penalty'
print*, 'Jacobian matrix of penalty'
do i = 1,3*nIntFaceTot
write(6,'(1x,100(e11.4,1x))')(pmatrix(i,j), j = 1,3*nIntFaceTot)
print'(1x,100(e11.4,1x))',(pmatrix(i,j), j = 1,3*nIntFaceTot)
enddo
write(6,*)' '
print*,' '
flush(6)
endif
#endif
@ -588,11 +573,11 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a30)')'Jacobian matrix of penalty'
print*, 'Jacobian matrix of penalty'
do i = 1,3*nIntFaceTot
write(6,'(1x,100(e11.4,1x))')(rmatrix(i,j), j = 1,3*nIntFaceTot)
print'(1x,100(e11.4,1x))',(rmatrix(i,j), j = 1,3*nIntFaceTot)
enddo
write(6,*)' '
print*,' '
flush(6)
endif
#endif
@ -603,11 +588,11 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a30)')'Jacobian matrix (total)'
print*, 'Jacobian matrix (total)'
do i = 1,3*nIntFaceTot
write(6,'(1x,100(e11.4,1x))')(jmatrix(i,j), j = 1,3*nIntFaceTot)
print'(1x,100(e11.4,1x))',(jmatrix(i,j), j = 1,3*nIntFaceTot)
enddo
write(6,*)' '
print*,' '
flush(6)
endif
#endif
@ -619,11 +604,11 @@ module procedure mech_RGC_updateState
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a30)')'Jacobian inverse'
print*, 'Jacobian inverse'
do i = 1,3*nIntFaceTot
write(6,'(1x,100(e11.4,1x))')(jnverse(i,j), j = 1,3*nIntFaceTot)
print'(1x,100(e11.4,1x))',(jnverse(i,j), j = 1,3*nIntFaceTot)
enddo
write(6,*)' '
print*,' '
flush(6)
endif
#endif
@ -638,19 +623,19 @@ module procedure mech_RGC_updateState
if (any(abs(drelax) > num%maxdRelax)) then ! Forcing cutback when the incremental change of relaxation vector becomes too large
mech_RGC_updateState = [.true.,.false.]
!$OMP CRITICAL (write2out)
write(6,'(1x,a,1x,i3,1x,a,1x,i3,1x,a)')'RGC_updateState: ip',ip,'| el',el,'enforces cutback'
write(6,'(1x,a,1x,e15.8)')'due to large relaxation change =',maxval(abs(drelax))
print'(a,i3,a,i3,a)',' RGC_updateState: ip ',ip,' | el ',el,' enforces cutback'
print'(a,e15.8)',' due to large relaxation change = ',maxval(abs(drelax))
flush(6)
!$OMP END CRITICAL (write2out)
endif
#ifdef DEBUG
if (debugHomog%extensive) then
write(6,'(1x,a30)')'Returned state: '
print*, 'Returned state: '
do i = 1,size(stt%relaxationVector(:,of))
write(6,'(1x,2(e15.8,1x))') stt%relaxationVector(i,of)
print'(1x,2(e15.8,1x))', stt%relaxationVector(i,of)
enddo
write(6,*)' '
print*,' '
flush(6)
endif
#endif
@ -678,9 +663,6 @@ module procedure mech_RGC_updateState
integer :: iGrain,iGNghb,iFace,i,j,k,l
real(pReal) :: muGrain,muGNghb,nDefNorm,bgGrain,bgGNghb
real(pReal), parameter :: nDefToler = 1.0e-10_pReal
#ifdef DEBUG
logical :: debugActive
#endif
nGDim = param(instance)%Nconstituents
rPen = 0.0_pReal
@ -695,11 +677,9 @@ module procedure mech_RGC_updateState
associate(prm => param(instance))
#ifdef DEBUG
debugActive = debugHomog%extensive .and. prm%of_debug == of
if (debugActive) then
write(6,'(1x,a20,2(1x,i3))')'Correction factor: ',ip,el
write(6,*) surfCorr
if (debugHomog%extensive .and. prm%of_debug == of) then
print'(a,2(1x,i3))', ' Correction factor: ',ip,el
print*, surfCorr
endif
#endif
@ -738,10 +718,10 @@ module procedure mech_RGC_updateState
nDefNorm = max(nDefToler,sqrt(nDefNorm)) ! approximation to zero mismatch if mismatch is zero (singularity)
nMis(abs(intFace(1)),iGrain) = nMis(abs(intFace(1)),iGrain) + nDefNorm ! total amount of mismatch experienced by the grain (at all six interfaces)
#ifdef DEBUG
if (debugActive) then
write(6,'(1x,a20,i2,1x,a20,1x,i3)')'Mismatch to face: ',intFace(1),'neighbor grain: ',iGNghb
write(6,*) transpose(nDef)
write(6,'(1x,a20,e11.4)')'with magnitude: ',nDefNorm
if (debugHomog%extensive .and. prm%of_debug == of) then
print'(a,i2,a,i3)',' Mismatch to face: ',intFace(1),' neighbor grain: ',iGNghb
print*, transpose(nDef)
print'(a,e11.4)', ' with magnitude: ',nDefNorm
endif
#endif
@ -756,9 +736,9 @@ module procedure mech_RGC_updateState
enddo; enddo;enddo; enddo
enddo interfaceLoop
#ifdef DEBUG
if (debugActive) then
write(6,'(1x,a20,i2)')'Penalty of grain: ',iGrain
write(6,*) transpose(rPen(1:3,1:3,iGrain))
if (debugHomog%extensive .and. prm%of_debug == of) then
print'(a,i2)', ' Penalty of grain: ',iGrain
print*, transpose(rPen(1:3,1:3,iGrain))
endif
#endif
@ -805,10 +785,9 @@ module procedure mech_RGC_updateState
gVol(i)*transpose(math_inv33(fDef(:,:,i)))
#ifdef DEBUG
if (debugHomog%extensive &
.and. param(instance)%of_debug == of) then
write(6,'(1x,a30,i2)')'Volume penalty of grain: ',i
write(6,*) transpose(vPen(:,:,i))
if (debugHomog%extensive .and. param(instance)%of_debug == of) then
print'(a,i2)',' Volume penalty of grain: ',i
print*, transpose(vPen(:,:,i))
endif
#endif
enddo

View File

@ -10,16 +10,16 @@ submodule(homogenization) homogenization_mech_isostrain
parallel_ID, &
average_ID
end enum
type :: tParameters !< container type for internal constitutive parameters
integer :: &
Nconstituents
integer(kind(average_ID)) :: &
mapping
end type
type(tParameters), dimension(:), allocatable :: param !< containers of constitutive parameters (len Ninstance)
contains
@ -36,21 +36,21 @@ module subroutine mech_isostrain_init
material_homogenization, &
homog, &
homogMech
write(6,'(/,a)') ' <<<+- homogenization_mech_isostrain init -+>>>'
print'(/,a)', ' <<<+- homogenization_mech_isostrain init -+>>>'
Ninstance = count(homogenization_type == HOMOGENIZATION_ISOSTRAIN_ID)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
allocate(param(Ninstance)) ! one container of parameters per instance
material_homogenization => material_root%get('homogenization')
material_homogenization => config_material%get('homogenization')
do h = 1, size(homogenization_type)
if (homogenization_type(h) /= HOMOGENIZATION_ISOSTRAIN_ID) cycle
homog => material_homogenization%get(h)
homogMech => homog%get('mech')
associate(prm => param(homogenization_typeInstance(h)))
prm%Nconstituents = homogMech%get_asInt('N_constituents')
select case(homogMech%get_asString('mapping',defaultVal = 'sum'))
case ('sum')
@ -60,15 +60,15 @@ module subroutine mech_isostrain_init
case default
call IO_error(211,ext_msg='sum'//' (mech_isostrain)')
end select
NofMyHomog = count(material_homogenizationAt == h)
homogState(h)%sizeState = 0
allocate(homogState(h)%state0 (0,NofMyHomog))
allocate(homogState(h)%subState0(0,NofMyHomog))
allocate(homogState(h)%state (0,NofMyHomog))
end associate
enddo
end subroutine mech_isostrain_init
@ -78,30 +78,30 @@ end subroutine mech_isostrain_init
!> @brief partitions the deformation gradient onto the constituents
!--------------------------------------------------------------------------------------------------
module subroutine mech_isostrain_partitionDeformation(F,avgF)
real(pReal), dimension (:,:,:), intent(out) :: F !< partitioned deformation gradient
real(pReal), dimension (3,3), intent(in) :: avgF !< average deformation gradient at material point
F = spread(avgF,3,size(F,3))
end subroutine mech_isostrain_partitionDeformation
!--------------------------------------------------------------------------------------------------
!> @brief derive average stress and stiffness from constituent quantities
!> @brief derive average stress and stiffness from constituent quantities
!--------------------------------------------------------------------------------------------------
module subroutine mech_isostrain_averageStressAndItsTangent(avgP,dAvgPdAvgF,P,dPdF,instance)
real(pReal), dimension (3,3), intent(out) :: avgP !< average stress at material point
real(pReal), dimension (3,3,3,3), intent(out) :: dAvgPdAvgF !< average stiffness at material point
real(pReal), dimension (:,:,:), intent(in) :: P !< partitioned stresses
real(pReal), dimension (:,:,:,:,:), intent(in) :: dPdF !< partitioned stiffnesses
integer, intent(in) :: instance
integer, intent(in) :: instance
associate(prm => param(instance))
select case (prm%mapping)
case (parallel_ID)
avgP = sum(P,3)
@ -110,7 +110,7 @@ module subroutine mech_isostrain_averageStressAndItsTangent(avgP,dAvgPdAvgF,P,dP
avgP = sum(P,3) /real(prm%Nconstituents,pReal)
dAvgPdAvgF = sum(dPdF,5)/real(prm%Nconstituents,pReal)
end select
end associate
end subroutine mech_isostrain_averageStressAndItsTangent

View File

@ -18,10 +18,10 @@ module subroutine mech_none_init
h, &
NofMyHomog
write(6,'(/,a)') ' <<<+- homogenization_mech_none init -+>>>'
print'(/,a)', ' <<<+- homogenization_mech_none init -+>>>'
Ninstance = count(homogenization_type == HOMOGENIZATION_NONE_ID)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
do h = 1, size(homogenization_type)
if (homogenization_type(h) /= HOMOGENIZATION_NONE_ID) cycle

View File

@ -45,15 +45,14 @@ module function kinematics_cleavage_opening_init(kinematics_length) result(myKin
kinematics, &
kinematic_type
write(6,'(/,a)') ' <<<+- kinematics_cleavage_opening init -+>>>'
print'(/,a)', ' <<<+- kinematics_cleavage_opening init -+>>>'
myKinematics = kinematics_active('cleavage_opening',kinematics_length)
Ninstance = count(myKinematics)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(param(Ninstance))
allocate(kinematics_cleavage_opening_instance(phases%length), source=0)

View File

@ -48,15 +48,14 @@ module function kinematics_slipplane_opening_init(kinematics_length) result(myKi
kinematics, &
kinematic_type
write(6,'(/,a)') ' <<<+- kinematics_slipplane init -+>>>'
print'(/,a)', ' <<<+- kinematics_slipplane init -+>>>'
myKinematics = kinematics_active('slipplane_opening',kinematics_length)
Ninstance = count(myKinematics)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(kinematics_slipplane_opening_instance(phases%length), source=0)
allocate(param(Ninstance))

View File

@ -38,15 +38,14 @@ module function kinematics_thermal_expansion_init(kinematics_length) result(myKi
kinematics, &
kinematic_type
write(6,'(/,a)') ' <<<+- kinematics_thermal_expansion init -+>>>'
print'(/,a)', ' <<<+- kinematics_thermal_expansion init -+>>>'
myKinematics = kinematics_active('thermal_expansion',kinematics_length)
Ninstance = count(myKinematics)
write(6,'(a16,1x,i5,/)') '# instances:',Ninstance; flush(6)
print'(a,i2)', ' # instances: ',Ninstance; flush(6)
if(Ninstance == 0) return
phases => material_root%get('phase')
phases => config_material%get('phase')
allocate(param(Ninstance))
allocate(kinematics_thermal_expansion_instance(phases%length), source=0)

View File

@ -457,9 +457,9 @@ subroutine lattice_init
phase, &
elasticity
write(6,'(/,a)') ' <<<+- lattice init -+>>>'; flush(6)
print'(/,a)', ' <<<+- lattice init -+>>>'; flush(6)
phases => material_root%get('phase')
phases => config_material%get('phase')
Nphases = phases%length
allocate(lattice_structure(Nphases),source = lattice_UNDEFINED_ID)
@ -738,7 +738,7 @@ function lattice_nonSchmidMatrix(Nslip,nonSchmidCoefficients,sense) result(nonSc
type(rotation) :: R
integer :: i
if (abs(sense) /= 1) call IO_error(0,ext_msg='lattice_nonSchmidMatrix')
if (abs(sense) /= 1) error stop 'Sense in lattice_nonSchmidMatrix'
coordinateSystem = buildCoordinateSystem(Nslip,BCC_NSLIPSYSTEM,BCC_SYSTEMSLIP,&
'bcc',0.0_pReal)
@ -2299,7 +2299,7 @@ end function equivalent_mu
!--------------------------------------------------------------------------------------------------
!> @brief check correctness of some lattice functions
!> @brief Check correctness of some lattice functions.
!--------------------------------------------------------------------------------------------------
subroutine selfTest
@ -2314,21 +2314,19 @@ subroutine selfTest
system = reshape([1.0_pReal+r(1),0.0_pReal,0.0_pReal, 0.0_pReal,1.0_pReal+r(2),0.0_pReal],[6,1])
CoSy = buildCoordinateSystem([1],[1],system,'fcc',0.0_pReal)
if(any(dNeq(CoSy(1:3,1:3,1),math_I3))) &
call IO_error(0)
if(any(dNeq(CoSy(1:3,1:3,1),math_I3))) error stop 'buildCoordinateSystem'
call random_number(C)
C(1,1) = C(1,1) + 1.0_pReal
C = applyLatticeSymmetryC66(C,'iso')
if(dNeq(C(6,6),equivalent_mu(C,'voigt'),1.0e-12_pReal)) &
call IO_error(0,ext_msg='equivalent_mu/voigt')
if(dNeq(C(6,6),equivalent_mu(C,'voigt'),1.0e-12_pReal)) &
call IO_error(0,ext_msg='equivalent_mu/reuss')
if(dNeq(C(6,6),equivalent_mu(C,'voigt'),1.0e-12_pReal)) error stop 'equivalent_mu/voigt'
if(dNeq(C(6,6),equivalent_mu(C,'voigt'),1.0e-12_pReal)) error stop 'equivalent_mu/reuss'
lambda = C(1,2)
if(dNeq(lambda*0.5_pReal/(lambda+equivalent_mu(C,'voigt')),equivalent_nu(C,'voigt'),1.0e-12_pReal)) &
call IO_error(0,ext_msg='equivalent_nu/voigt')
error stop 'equivalent_nu/voigt'
if(dNeq(lambda*0.5_pReal/(lambda+equivalent_mu(C,'reuss')),equivalent_nu(C,'reuss'),1.0e-12_pReal)) &
call IO_error(0,ext_msg='equivalent_nu/reuss')
error stop 'equivalent_nu/reuss'
end subroutine selfTest

View File

@ -1,453 +0,0 @@
!-------------------------------------------------------------------------------------------------
!> @author Martin Diehl, Max-Planck-Institut für Eisenforschung GmbH
!> @brief Linked list
!--------------------------------------------------------------------------------------------------
module list
use prec
use IO
implicit none
private
type, private :: tPartitionedString
character(len=:), allocatable :: val
integer, dimension(:), allocatable :: pos
end type tPartitionedString
type, public :: tPartitionedStringList
type(tPartitionedString) :: string
type(tPartitionedStringList), pointer :: next => null()
contains
procedure :: add => add
procedure :: show => show
procedure :: free => free
! currently, a finalize is needed for all shapes of tPartitionedStringList.
! with Fortran 2015, we can define one recursive elemental function
! https://software.intel.com/en-us/forums/intel-visual-fortran-compiler-for-windows/topic/543326
final :: finalize, &
finalizeArray
procedure :: keyExists => keyExists
procedure :: countKeys => countKeys
procedure :: getFloat => getFloat
procedure :: getInt => getInt
procedure :: getString => getString
procedure :: getFloats => getFloats
procedure :: getInts => getInts
procedure :: getStrings => getStrings
end type tPartitionedStringList
contains
!--------------------------------------------------------------------------------------------------
!> @brief add element
!> @details Adds a string together with the start/end position of chunks in this string. The new
!! element is added at the end of the list. Empty strings are not added. All strings are converted
!! to lower case. The data is not stored in the new element but in the current.
!--------------------------------------------------------------------------------------------------
subroutine add(this,string)
class(tPartitionedStringList), target, intent(in) :: this
character(len=*), intent(in) :: string
type(tPartitionedStringList), pointer :: new, temp
if (IO_isBlank(string)) return
allocate(new)
temp => this
do while (associated(temp%next))
temp => temp%next
enddo
temp%string%val = IO_lc (trim(string))
temp%string%pos = IO_stringPos(trim(string))
temp%next => new
end subroutine add
!--------------------------------------------------------------------------------------------------
!> @brief prints all elements
!> @details Strings are printed in order of insertion (FIFO)
!--------------------------------------------------------------------------------------------------
subroutine show(this)
class(tPartitionedStringList), target, intent(in) :: this
type(tPartitionedStringList), pointer :: item
item => this
do while (associated(item%next))
write(6,'(a)') ' '//trim(item%string%val)
item => item%next
enddo
end subroutine show
!--------------------------------------------------------------------------------------------------
!> @brief empties list and frees associated memory
!> @details explicit interface to reset list. Triggers final statement (and following chain reaction)
!--------------------------------------------------------------------------------------------------
subroutine free(this)
class(tPartitionedStringList), intent(inout) :: this
if(associated(this%next)) deallocate(this%next)
end subroutine free
!--------------------------------------------------------------------------------------------------
!> @brief empties list and frees associated memory
!> @details called when variable goes out of scope. Triggers chain reaction for list
!--------------------------------------------------------------------------------------------------
recursive subroutine finalize(this)
type(tPartitionedStringList), intent(inout) :: this
if(associated(this%next)) deallocate(this%next)
end subroutine finalize
!--------------------------------------------------------------------------------------------------
!> @brief cleans entire array of linke lists
!> @details called when variable goes out of scope and deallocates the list at each array entry
!--------------------------------------------------------------------------------------------------
subroutine finalizeArray(this)
integer :: i
type(tPartitionedStringList), intent(inout), dimension(:) :: this
type(tPartitionedStringList), pointer :: temp ! bug in Gfortran?
do i=1, size(this)
if (associated(this(i)%next)) then
temp => this(i)%next
!deallocate(this(i)) !internal compiler error: in gfc_build_final_call, at fortran/trans.c:975
deallocate(temp)
endif
enddo
end subroutine finalizeArray
!--------------------------------------------------------------------------------------------------
!> @brief reports wether a given key (string value at first position) exists in the list
!--------------------------------------------------------------------------------------------------
logical function keyExists(this,key)
class(tPartitionedStringList), target, intent(in) :: this
character(len=*), intent(in) :: key
type(tPartitionedStringList), pointer :: item
keyExists = .false.
item => this
do while (associated(item%next) .and. .not. keyExists)
keyExists = trim(IO_stringValue(item%string%val,item%string%pos,1)) == trim(key)
item => item%next
enddo
end function keyExists
!--------------------------------------------------------------------------------------------------
!> @brief count number of key appearances
!> @details traverses list and counts each occurrence of specified key
!--------------------------------------------------------------------------------------------------
integer function countKeys(this,key)
class(tPartitionedStringList), target, intent(in) :: this
character(len=*), intent(in) :: key
type(tPartitionedStringList), pointer :: item
countKeys = 0
item => this
do while (associated(item%next))
if (trim(IO_stringValue(item%string%val,item%string%pos,1)) == trim(key)) &
countKeys = countKeys + 1
item => item%next
enddo
end function countKeys
!--------------------------------------------------------------------------------------------------
!> @brief gets float value of for a given key from a linked list
!> @details gets the last value if the key occurs more than once. If key is not found exits with
!! error unless default is given
!--------------------------------------------------------------------------------------------------
real(pReal) function getFloat(this,key,defaultVal)
class(tPartitionedStringList), target, intent(in) :: this
character(len=*), intent(in) :: key
real(pReal), intent(in), optional :: defaultVal
type(tPartitionedStringList), pointer :: item
logical :: found
getFloat = huge(1.0) ! suppress warning about unitialized value
found = present(defaultVal)
if (found) getFloat = defaultVal
item => this
do while (associated(item%next))
if (trim(IO_stringValue(item%string%val,item%string%pos,1)) == trim(key)) then
found = .true.
if (item%string%pos(1) < 2) call IO_error(143,ext_msg=key)
getFloat = IO_FloatValue(item%string%val,item%string%pos,2)
endif
item => item%next
enddo
if (.not. found) call IO_error(140,ext_msg=key)
end function getFloat
!--------------------------------------------------------------------------------------------------
!> @brief gets integer value of for a given key from a linked list
!> @details gets the last value if the key occurs more than once. If key is not found exits with
!! error unless default is given
!--------------------------------------------------------------------------------------------------
integer function getInt(this,key,defaultVal)
class(tPartitionedStringList), target, intent(in) :: this
character(len=*), intent(in) :: key
integer, intent(in), optional :: defaultVal
type(tPartitionedStringList), pointer :: item
logical :: found
getInt = huge(1) ! suppress warning about unitialized value
found = present(defaultVal)
if (found) getInt = defaultVal
item => this
do while (associated(item%next))
if (trim(IO_stringValue(item%string%val,item%string%pos,1)) == trim(key)) then
found = .true.
if (item%string%pos(1) < 2) call IO_error(143,ext_msg=key)
getInt = IO_IntValue(item%string%val,item%string%pos,2)
endif
item => item%next
enddo
if (.not. found) call IO_error(140,ext_msg=key)
end function getInt
!--------------------------------------------------------------------------------------------------
!> @brief gets string value of for a given key from a linked list
!> @details gets the last value if the key occurs more than once. If key is not found exits with
!! error unless default is given. If raw is true, the the complete string is returned, otherwise
!! the individual chunks are returned
!--------------------------------------------------------------------------------------------------
character(len=pStringLen) function getString(this,key,defaultVal,raw)
class(tPartitionedStringList), target, intent(in) :: this
character(len=*), intent(in) :: key
character(len=*), intent(in), optional :: defaultVal
logical, intent(in), optional :: raw
type(tPartitionedStringList), pointer :: item
logical :: found, &
whole
if (present(raw)) then
whole = raw
else
whole = .false.
endif
found = present(defaultVal)
if (found) then
if (len_trim(defaultVal) > len(getString)) call IO_error(0,ext_msg='getString')
getString = trim(defaultVal)
endif
item => this
do while (associated(item%next))
if (trim(IO_stringValue(item%string%val,item%string%pos,1)) == trim(key)) then
found = .true.
if (item%string%pos(1) < 2) call IO_error(143,ext_msg=key)
if (whole) then
getString = trim(item%string%val(item%string%pos(4):)) ! raw string starting a second chunk
else
getString = IO_StringValue(item%string%val,item%string%pos,2)
endif
endif
item => item%next
enddo
if (.not. found) call IO_error(140,ext_msg=key)
end function getString
!--------------------------------------------------------------------------------------------------
!> @brief gets array of float values of for a given key from a linked list
!> @details for cumulative keys, "()", values from all occurrences are return. Otherwise only all
!! values from the last occurrence. If key is not found exits with error unless default is given.
!--------------------------------------------------------------------------------------------------
function getFloats(this,key,defaultVal,requiredSize)
real(pReal), dimension(:), allocatable :: getFloats
class(tPartitionedStringList), target, intent(in) :: this
character(len=*), intent(in) :: key
real(pReal), dimension(:), intent(in), optional :: defaultVal
integer, intent(in), optional :: requiredSize
type(tPartitionedStringList), pointer :: item
integer :: i
logical :: found, &
cumulative
cumulative = (key(1:1) == '(' .and. key(len_trim(key):len_trim(key)) == ')')
found = .false.
allocate(getFloats(0))
item => this
do while (associated(item%next))
if (trim(IO_stringValue(item%string%val,item%string%pos,1)) == trim(key)) then
found = .true.
if (.not. cumulative) getFloats = [real(pReal)::]
if (item%string%pos(1) < 2) call IO_error(143,ext_msg=key)
do i = 2, item%string%pos(1)
getFloats = [getFloats,IO_FloatValue(item%string%val,item%string%pos,i)]
enddo
endif
item => item%next
enddo
if (.not. found) then
if (present(defaultVal)) then; getFloats = defaultVal; else; call IO_error(140,ext_msg=key); endif
endif
if (present(requiredSize)) then
if(requiredSize /= size(getFloats)) call IO_error(146,ext_msg=key)
endif
end function getFloats
!--------------------------------------------------------------------------------------------------
!> @brief gets array of integer values of for a given key from a linked list
!> @details for cumulative keys, "()", values from all occurrences are return. Otherwise only all
!! values from the last occurrence. If key is not found exits with error unless default is given.
!--------------------------------------------------------------------------------------------------
function getInts(this,key,defaultVal,requiredSize)
integer, dimension(:), allocatable :: getInts
class(tPartitionedStringList), target, intent(in) :: this
character(len=*), intent(in) :: key
integer, dimension(:), intent(in), optional :: defaultVal
integer, intent(in), optional :: requiredSize
type(tPartitionedStringList), pointer :: item
integer :: i
logical :: found, &
cumulative
cumulative = (key(1:1) == '(' .and. key(len_trim(key):len_trim(key)) == ')')
found = .false.
allocate(getInts(0))
item => this
do while (associated(item%next))
if (trim(IO_stringValue(item%string%val,item%string%pos,1)) == trim(key)) then
found = .true.
if (.not. cumulative) getInts = [integer::]
if (item%string%pos(1) < 2) call IO_error(143,ext_msg=key)
do i = 2, item%string%pos(1)
getInts = [getInts,IO_IntValue(item%string%val,item%string%pos,i)]
enddo
endif
item => item%next
enddo
if (.not. found) then
if (present(defaultVal)) then; getInts = defaultVal; else; call IO_error(140,ext_msg=key); endif
endif
if (present(requiredSize)) then
if(requiredSize /= size(getInts)) call IO_error(146,ext_msg=key)
endif
end function getInts
!--------------------------------------------------------------------------------------------------
!> @brief gets array of string values of for a given key from a linked list
!> @details for cumulative keys, "()", values from all occurrences are return. Otherwise only all
!! values from the last occurrence. If key is not found exits with error unless default is given.
!! If raw is true, the the complete string is returned, otherwise the individual chunks are returned
!--------------------------------------------------------------------------------------------------
function getStrings(this,key,defaultVal,raw)
character(len=pStringLen),dimension(:), allocatable :: getStrings
class(tPartitionedStringList),target, intent(in) :: this
character(len=*), intent(in) :: key
character(len=*), dimension(:), intent(in), optional :: defaultVal
logical, intent(in), optional :: raw
type(tPartitionedStringList), pointer :: item
character(len=pStringLen) :: str
integer :: i
logical :: found, &
whole, &
cumulative
cumulative = (key(1:1) == '(' .and. key(len_trim(key):len_trim(key)) == ')')
if (present(raw)) then
whole = raw
else
whole = .false.
endif
found = .false.
item => this
do while (associated(item%next))
if (trim(IO_stringValue(item%string%val,item%string%pos,1)) == trim(key)) then
found = .true.
if (allocated(getStrings) .and. .not. cumulative) deallocate(getStrings)
if (item%string%pos(1) < 2) call IO_error(143,ext_msg=key)
notAllocated: if (.not. allocated(getStrings)) then
if (whole) then
str = item%string%val(item%string%pos(4):)
getStrings = [str]
else
str = IO_StringValue(item%string%val,item%string%pos,2)
allocate(getStrings(1),source=str)
do i=3,item%string%pos(1)
str = IO_StringValue(item%string%val,item%string%pos,i)
getStrings = [getStrings,str]
enddo
endif
else notAllocated
if (whole) then
str = item%string%val(item%string%pos(4):)
getStrings = [getStrings,str]
else
do i=2,item%string%pos(1)
str = IO_StringValue(item%string%val,item%string%pos,i)
getStrings = [getStrings,str]
enddo
endif
endif notAllocated
endif
item => item%next
enddo
if (.not. found) then
if (present(defaultVal)) then
if (len(defaultVal) > len(getStrings)) call IO_error(0,ext_msg='getStrings')
getStrings = defaultVal
else
call IO_error(140,ext_msg=key)
endif
endif
end function getStrings
end module list

View File

@ -74,12 +74,12 @@ subroutine discretization_marc_init
!---------------------------------------------------------------------------------
! read debug parameters
debug_e = debug_root%get_asInt('element',defaultVal=1)
debug_i = debug_root%get_asInt('integrationpoint',defaultVal=1)
debug_e = config_debug%get_asInt('element',defaultVal=1)
debug_i = config_debug%get_asInt('integrationpoint',defaultVal=1)
!--------------------------------------------------------------------------------
! read numerics parameter and do sanity check
num_commercialFEM => numerics_root%get('commercialFEM',defaultVal = emptyDict)
num_commercialFEM => config_numerics%get('commercialFEM',defaultVal = emptyDict)
mesh_unitlength = num_commercialFEM%get_asFloat('unitlength',defaultVal=1.0_pReal) ! set physical extent of a length unit in mesh
if (mesh_unitlength <= 0.0_pReal) call IO_error(301,ext_msg='unitlength')
@ -197,7 +197,7 @@ subroutine inputRead(elem,node0_elem,connectivity_elem,microstructureAt)
integer, dimension(:,:), allocatable :: &
mapElemSet !< list of elements in elementSet
inputFile = IO_read_ASCII(trim(getSolverJobName())//trim(InputFileExtension))
inputFile = IO_readlines(trim(getSolverJobName())//trim(InputFileExtension))
call inputRead_fileFormat(fileFormatVersion, &
inputFile)
call inputRead_tableStyles(initialcondTableStyle,hypoelasticTableStyle, &
@ -723,7 +723,7 @@ subroutine inputRead_microstructure(microstructureAt,&
endif
enddo
if(any(microstructureAt < 1)) call IO_error(190)
if(any(microstructureAt < 1)) call IO_error(180)
end subroutine inputRead_microstructure

443
src/marc/include/concom2020 vendored Normal file
View File

@ -0,0 +1,443 @@
! common block definition file taken from respective MSC.Marc release and reformated to free format
!***********************************************************************
!
! File: concom.cmn
!
! MSC.Marc include file
!
integer &
iacous, iasmbl, iautth, ibear, icompl, iconj, icreep, ideva, idyn, idynt,&
ielas, ielcma, ielect, iform, ifour, iharm, ihcps, iheat, iheatt, ihresp,&
ijoule, ilem, ilnmom, iloren, inc, incext, incsub, ipass, iplres, ipois,&
ipoist, irpflo, ismall, ismalt, isoil, ispect, ispnow, istore, iswep, ithcrp,&
itherm, iupblg, iupdat, jacflg, jel, jparks, largst, lfond, loadup, loaduq,&
lodcor, lovl, lsub, magnet, ncycle, newtnt, newton, noshr, linear, ivscpl,&
icrpim, iradrt, ipshft, itshr, iangin, iupmdr, iconjf, jincfl, jpermg, jhour,&
isolvr, jritz, jtable, jshell, jdoubl, jform, jcentr, imini, kautth, iautof,&
ibukty, iassum, icnstd, icnstt, kmakmas, imethvp, iradrte, iradrtp, iupdate, iupdatp,&
ncycnt, marmen , idynme, ihavca, ispf, kmini, imixex, largtt, kdoela, iautofg,&
ipshftp, idntrc, ipore, jtablm, jtablc, isnecma, itrnspo, imsdif, jtrnspo, mcnear,&
imech, imecht, ielcmat, ielectt, magnett, imsdift, noplas, jtabls, jactch, jtablth,&
kgmsto , jpzo, ifricsh, iremkin, iremfor, ishearp, jspf, machining, jlshell, icompsol,&
iupblgfo, jcondir, nstcrp, nactive, ipassref, nstspnt, ibeart, icheckmpc, noline, icuring,&
ishrink, ioffsflg, isetoff, ioffsetm,iharmt, inc_incdat, iautspc, ibrake, icbush, istream_input,&
iprsinp, ivlsinp, ifirst_time,ipin_m, jgnstr_glb, imarc_return,iqvcinp, nqvceid, istpnx, imicro1,&
iaxisymm, jbreakglue,iglstif, jfastasm,iwear, iwearcf, imixmeth, ielcmadyn, idinout, igena_meth,&
magf_meth, non_assumed, iredoboudry, ioffsz0,icomplt, mesh_dual, iactrp, mgnewton, iusedens,igsigd0,&
iaem, icosim, inodels, nlharm, iampini, iphasetr, inonlcl, inonlct, iforminp,ispecerror,&
icsprg, imol, imolt, idatafit,iharmpar, inclcase, imultifreq,init_elas, ifatig, iftgmat
dimension :: ideva(60)
integer num_concom
parameter(num_concom=260)
common/marc_concom/&
iacous, iasmbl, iautth, ibear, icompl, iconj, icreep, ideva, idyn, idynt,&
ielas, ielcma, ielect, iform, ifour, iharm, ihcps, iheat, iheatt, ihresp,&
ijoule, ilem, ilnmom, iloren, inc, incext, incsub, ipass, iplres, ipois,&
ipoist, irpflo, ismall, ismalt, isoil, ispect, ispnow, istore, iswep, ithcrp,&
itherm, iupblg, iupdat, jacflg, jel, jparks, largst, lfond, loadup, loaduq,&
lodcor, lovl, lsub, magnet, ncycle, newtnt, newton, noshr, linear, ivscpl,&
icrpim, iradrt, ipshft, itshr, iangin, iupmdr, iconjf, jincfl, jpermg, jhour,&
isolvr, jritz, jtable, jshell, jdoubl, jform, jcentr, imini, kautth, iautof,&
ibukty, iassum, icnstd, icnstt, kmakmas, imethvp, iradrte, iradrtp, iupdate, iupdatp,&
ncycnt, marmen, idynme, ihavca, ispf, kmini, imixex, largtt, kdoela, iautofg,&
ipshftp, idntrc, ipore, jtablm, jtablc, isnecma, itrnspo, imsdif, jtrnspo, mcnear,&
imech, imecht, ielcmat, ielectt, magnett, imsdift, noplas, jtabls, jactch, jtablth,&
kgmsto , jpzo, ifricsh, iremkin, iremfor, ishearp, jspf, machining, jlshell, icompsol,&
iupblgfo, jcondir, nstcrp, nactive, ipassref, nstspnt, ibeart, icheckmpc, noline, icuring,&
ishrink, ioffsflg, isetoff, ioffsetm,iharmt, inc_incdat, iautspc, ibrake, icbush, istream_input,&
iprsinp, ivlsinp, ifirst_time,ipin_m, jgnstr_glb, imarc_return,iqvcinp, nqvceid, istpnx, imicro1,&
iaxisymm, jbreakglue,iglstif, jfastasm,iwear, iwearcf, imixmeth, ielcmadyn, idinout, igena_meth,&
magf_meth, non_assumed, iredoboudry, ioffsz0,icomplt, mesh_dual, iactrp, mgnewton, iusedens,igsigd0,&
iaem, icosim, inodels, nlharm, iampini, iphasetr, inonlcl, inonlct, iforminp,ispecerror,&
icsprg, imol, imolt, idatafit,iharmpar, inclcase, imultifreq,init_elas, ifatig, iftgmat
!
! comments of variables:
!
! iacous Control flag for acoustic analysis. Input data.
! iacous=1 modal acoustic analysis.
! iacous=2 harmonic acoustic-structural analysis.
! iasmbl Control flag to indicate that operator matrix should be
! recalculated.
! iautth Control flag for AUTO THERM option.
! ibear Control flag for bearing analysis. Input data.
! icompl Control variable to indicate that a complex analysis is
! being performed. Either a Harmonic analysis with damping,
! or a harmonic electro-magnetic analysis. Input data.
! iconj Flag for EBE conjugate gradient solver (=solver 1, retired)
! Also used for VKI iterative solver.
! icreep Control flag for creep analysis. Input data.
! ideva(60) - debug print out flag
! 1 print element stiffness matrices, mass matrix
! 2 output matrices used in tying
! 3 force the solution of a nonpositive definite matrix
! 4 print info of connections to each node
! 5 info of gap convergence, internal heat generated, contact
! touching and separation
! 6 nodal value array during rezoning
! 7 tying info in CONRAD GAP option, fluid element numbers in
! CHANNEL option
! 8 output incremental displacements in local coord. system
! 9 latent heat output
! 10 stress-strain in local coord. system
! 11 additional info on interlaminar stress
! 12 output right hand side and solution vector
! 13 info of CPU resources used and memory available on NT
! 14 info of mesh adaption process, 2D outline information
! info of penetration checking for remeshing
! save .fem files after afmesh3d meshing
! 15 surface energy balance flag
! 16 print info regarding pyrolysis
! 17 print info of "streamline topology"
! 18 print mesh data changes after remeshing
! 19 print material flow stress data read in from *.mat file
! if unit flag is on, print out flow stress after conversion
! 20 print information on table input
! 21 print out information regarding kinematic boundary conditions
! 22 print out information regarding dist loads, point loads, film
! and foundations
! 23 print out information about automatic domain decomposition
! 24 print out iteration information in SuperForm status report file
! 25 print out information for ablation
! 26 print out information for films - Table input
! 27 print out the tying forces
! 28 print out for CASI solver, convection,
! 29 DDM single file debug printout
! 30 print out cavity debug info
! 31 print out welding related info
! 32 prints categorized DDM memory usage
! 33 print out the cutting info regarding machining feature
! 34 print out the list of quantities which can be defined via a table
! and for each quantity the supported independent variables
! 35 print out detailed coupling region info
! 36 print out solver debug info level 1 (Least Detailed)
! 37 print out solver debug info level 1 (Medium Detailed)
! 38 print out solver debug info level 1 (Very Detailed)
! 39 print detailed memory allocation info
! 40 print out marc-adams debug info
! 41 output rezone mapping post file for debugging
! 42 output post file after calling oprofos() for debugging
! 43 debug printout for vcct
! 44 debug printout for progressive failure
! 45 print out automatically generated midside node coordinates (arecrd)
! 46 print out message about routine and location, where the ibort is raised (ibort_inc)
! 47 print out summary message of element variables on a
! group-basis after all the automatic changes have been
! made (em_ellibp)
! 48 Automatically generate check results based on max and min vals.
! These vals are stored in the checkr file, which is inserted
! into the *dat file by the generate_check_results script from /marc/tools
! 49 Automatically generate check results based on the real calculated values
! at the sppecified check result locations.
! These vals are stored in the checkr file, which is inserted
! into the *dat file by the update_check_results script from /marc/tools
! 50 generate a file containing the resistance or capacity matrix;
! this file can be used to compare results with a reference file
! 51 print out detailed information for segment-to-segment contact
! 52 print out detailed relative displacement information
! for uniaxial sliding contact
! 53 print out detailed sliding direction information for
! uniaxial sliding contact
! 54 print out detailed information for edges attached to a curve
! 55 print information related to viscoelasticity calculations
! 56 print out detailed information for element coloring for multithreading
! 57 print out extra overheads due to multi-threading.
! These overhead includes (i) time and (ii) memory.
! The memory report will be summed over all the children.
!
!
! 58 debug output for ELSTO usage
!
! idyn Control flag for dynamics. Input data.
! 1 = eigenvalue extraction and / or modal superposition
! 2 = Newmark Beta and Single Step Houbolt (ssh with idynme=1)
! 3 = Houbolt
! 4 = Central difference
! 5 = Newer central difference
! idynt Copy of idyn at begining of increment
! ielas Control flag for ELASTIC analysis. Input data.
! Set by user or automatically turned on by Fourier option.
! Implies that each load case is treated separately.
! In Adaptive meshing analysis , forces re-analysis until
! convergence obtained.
! Also seriously misused to indicate no convergence.
! = 1 elastic option with fourier analysis
! = 2 elastic option without fourier analysis
! =-1 no convergence in recycles or max # increments reached
! Set to 1 if ELASTIC or SUBSTRUC parameter cards are used,
! or if fourier option is used.
! Then set to 2 if not fourier analysis.
! ielcma Control flag for electromagnetic analysis. Input data.
! ielcma = 1 Harmonic formulation
! ielcma = 2 Transient formulation
! ielect Control flag for electrostatic option. Input data.
! iform Control flag indicating that contact will be performed.
! ifour Control flag for Fourier analysis.
! 0 = Odd and even terms.
! 1 = symmetric (cosine) terms
! 2 = antisymmetric (sine) terms.
! iharm Control flag to indicate that a harmonic analysis will
! be performed. May change between passes.
! ihcps Control flag for coupled thermal - stress analysis.
! iheat Control flag for heat transfer analysis. Input data.
! iheatt Permanent control flag for heat transfer analysis.
! Note in coupled analysis iheatt will remain as one,
! but iheat will be zero in stress pass.
! ihresp Control flag to indicate to perform a harmonic subincrement.
! ijoule Control flag for Joule heating.
! ilem Control flag to determin which vector is to be transformed.
! Control flag to see where one is:
! ilem = 1 - elem.f
! ilem = 2 - initst.f
! ilem = 3 - pressr.f
! ilem = 3 - fstif.f
! ilem = 4 - jflux.f
! ilem = 4 - strass.f
! ilem = 5 - mass.f
! ilem = 5 - osolty.f
! ilnmom Control flag for soil - pore pressure calculation. Input data.
! ilnmom = 0 - perform only pore pressure calculation.
! = 1 - couples pore pressure - displacement analysis
! iloren Control flag for DeLorenzi J-Integral evaluation. Input data.
! inc Increment number.
! incext Control flag indicating that currently working on a
! subincrement.
! Could be due to harmonics , damping component (bearing),
! stiffness component (bearing), auto therm creep or
! old viscoplaticity
! incsub Sub-increment number.
! ipass Control flag for which part of coupled analysis.
! ipass = -1 - reset to base values
! ipass = 0 - do nothing
! ipass = 1 - stress part
! ipass = 2 - heat transfer part
! iplres Flag indicating that either second matrix is stored.
! dynamic analysis - mass matrix
! heat transfer - specific heat matrix
! buckle - initial stress stiffness
! ipois Control flag indicating Poisson type analysis
! ipois = 1 for heat transfer
! = 1 for heat transfer part of coupled
! = 1 for bearing
! = 1 for electrostatic
! = 1 for magnetostatic
! ipoist Permanent copy of ipois. In coupled analysis , ipois = 0
! in stress portion, yet ipoist will still =1.
! irpflo global flag for rigid plastic flow analysis
! = 1 eularian formulation
! = 2 regular formulation; rigid material present in the analysis
! ismall control flag to indicate small displacement analysis. input data.
! ismall = 0 - large disp included.
! ismall = 1 - small displacement.
! the flag is changing between passes.
! ismalt permanent copy of ismall . in heat transfer portion of
! coupled analysis ismall =0 , but ismalt remains the same.
! isoil control flag indicating that soil / pore pressure
! calculation . input data.
! ispect control flag for response spectrum calculation. input data.
! ispnow control flag to indicate to perform a spectrum response
! calculation now.
! istore store stresses flag.
! istore = 0 in elem.f and if first pass of creep
! convergence checking in ogetst.f
! or harmonic analysis or thruc.f if not
! converged.
! iswep control flag for eigenvalue analysis.
! iswep=1 - go do extraction process
! ithcrp control flag for auto therm creep option. input data.
! itherm control flag for either temperature dependent material
! properties and/or thermal loads.
! iupblg control flag for follower force option. input data.
! iupdat control flag for update lagrange option for current element.
! jacflg control flag for lanczos iteration method. input data.
! jel control flag indicating that total load applied in
! increment, ignore previous solution.
! jel = 1 in increment 0
! = 1 if elastic or fourier
! = 1 in subincrements with elastic and adaptive
! jparks control flag for j integral by parks method. input data.
! largst control flag for finite strain plasticity. input data.
! lfond control variable that indicates if doing elastic
! foundation or film calculation. influences whether
! this is volumetric or surface integration.
! loadup control flag that indicates that nonlinearity occurred
! during previous increment.
! loaduq control flag that indicates that nonlinearity occurred.
! lodcor control flag for switching on the residual load correction.
! notice in input stage lodcor=0 means no loadcor,
! after omarc lodcor=1 means no loadcor
! lovl control flag for determining which "overlay" is to
! be called from ellib.
! lovl = 1 omarc
! = 2 oaread
! = 3 opress
! = 4 oasemb
! = 5 osolty
! = 6 ogetst
! = 7 oscinc
! = 8 odynam
! = 9 opmesh
! = 10 omesh2
! = 11 osetz
! = 12 oass
! = 13 oincdt
! = 14 oasmas
! = 15 ofluas
! = 16 ofluso
! = 17 oshtra
! = 18 ocass
! = 19 osoltc
! = 20 orezon
! = 21 otest
! = 22 oeigen
! lsub control variable to determine which part of element
! assembly function is being done.
! lsub = 1 - no longer used
! = 2 - beta*
! = 3 - cons*
! = 4 - ldef*
! = 5 - posw*
! = 6 - theta*
! = 7 - tmarx*
! = 8 - geom*
! magnet control flag for magnetostatic analysis. input data.
! ncycle cycle number. accumulated in osolty.f
! note first time through oasemb.f , ncycle = 0.
! newtnt control flag for permanent copy of newton.
! newton iteration type. input data.
! newton : = 1 full newton raphson
! 2 modified newton raphson
! 3 newton raphson with strain correct.
! 4 direct substitution
! 5 direct substitution followed by n.r.
! 6 direct substitution with line search
! 7 full newton raphson with secant initial stress
! 8 secant method
! 9 full newton raphson with line search
! noshr control flag for calculation interlaminar shears for
! elements 22,45, and 75. input data.
!ees
!
! jactch = 1 or 2 if elements are activated or deactivated
! = 3 if elements are adaptively remeshed or rezoned
! = 0 normally / reset to 0 when assembly is done
! ifricsh = 0 call to fricsh in otest not needed
! = 1 call to fricsh (nodal friction) in otest needed
! iremkin = 0 remove deactivated kinematic boundary conditions
! immediately - only in new input format (this is default)
! = 1 remove deactivated kinematic boundary conditions
! gradually - only in new input format
! iremfor = 0 remove force boundary conditions immediately -
! only in new input format (this is default)
! = 1 remove force boundary conditions gradually -
! only in new input format (this is default)
! ishearp set to 1 if shear panel elements are present in the model
!
! jspf = 0 not in spf loadcase
! > 0 in spf loadcase (jspf=1 during first increment)
! machining = 1 if the metal cutting feature is used, for memory allocation purpose
! = 0 (default) if no metal cutting feature required
!
! jlshell = 1 if there is a shell element in the mesh
! icompsol = 1 if there is a composite solid element in the mesh
! iupblgfo = 1 if follower force for point loads
! jcondir = 1 if contact priority option is used
! nstcrp = 0 (default) steady state creep flag (undocumented feature.
! if not 0, turns off special ncycle = 0 code in radial.f)
! nactive = number of active passes, if =1 then it's not a coupled analysis
! ipassref = reference ipass, if not in a multiphysics pass ipass=ipassref
! icheckmpc = value of mpc-check parameter option
! noline = set to 1 in osolty if no line seacrh should be done in ogetst
! icuring = set to 1 if the curing is included for the heat transfer analysis.
! ishrink = set to 1 if shrinkage strain is included for mechancial analysis.
! ioffsflg = 1 for small displacement beam/shell offsets
! = 2 for large displacement beam/shell offsets
! isetoff = 0 - do not apply beam/shell offsets
! = 1 - apply beam/shell offsets
! ioffsetm = min. value of offset flag
! iharmt = 1 global flag if a coupled analysis contains an harmonic pass
! inc_incdat = flag to record increment number of a new loadcase in incdat.f
! iautspc = flag for AutoSPC option
! ibrake = brake squeal in this increment
! icbush = set to 1 if cbush elements present in model
! istream_input = set to 1 for streaming input calling Marc as library
! iprsinp = set to 1 if pressure input, introduced so other variables
! such as h could be a function of pressure
! ivlsinp = set to 1 if velocity input, introduced so other variables
! such as h could be a function of velocity
! ipin_m = # of beam element with PIN flag
! jgnstr_glb = global control over pre or fast integrated composite shells
! imarc_return = Marc return flag for streaming input control
! iqvcimp = if non-zero, then the number of QVECT boundary conditions
! nqvceid = number of QVECT boundary conditions, where emisivity/absorbtion id entered
! istpnx = 1 if to stop at end of increment
! imicro1 = 1 if micro1 interface is used
! iaxisymm = set to 1 if axisymmetric analysis
! jbreakglue = set to 1 if breaking glued option is used
! iglstif = 1 if ddm and global stiffness matrix formed (sgi solver 6 or solver9)
! jfastasm = 1 do fast assembly using SuperForm code
! iwear = set to 1 if wear model, set to 2 if wear model and coordinates updated
! iwearcf = set to 1 to store nodal coefficient of friction for wear calculation
! imixmeth = set=1 then use nonlinear mixture material - allocate memory
! ielcmadyn = flag for magnetodynamics
! 0 - electromagnetics using newmark beta
! 1 - transient magnetics using backward euler
! idinout = flag to control if inside out elements should be deactivated
! igena_meth = 0 - generalized alpha parameters depend on whether or not contact
! is flagged (dynamic,7)
! 10 - generalized alpha parameters are optimized for a contact
! analysis (dynamic,8)
! 11 - generalized alpha parameters are optimized for an analysis
! without contact (dynamic,8)
! magf_meth = - Method to compute force in magnetostatic - structural
! = 1 - Virtual work method based on finite difference for the force computation
! = 2 - Maxwell stress tensor
! = 3 - Virtual work method based on local derivative for the force computation
! non_assumed = 1 no assumed strain formulation (forced)
! iredoboudry set to 1 if contact boundary needs to be recalculated
! ioffsz0 = 1 if composite are used with reference position.ne.0
! icomplt = 1 global flag if a coupled analysis contains an complex pass
! mesh_dual = 1 two independent meshes are used in magnetodynamic/thermal/structural
! one for magnetodynamic and the other for the remaining passes
! iactrp = 1 in an analysis with global remeshing, include inactive
! rigid bodies on post file
! mgnewton = 1 Use full Newton Raphson iteration for magnetostatic pass
!
! iusedens > 0 if mass density is used in the analysis (dynamics, mass dependent loading)
! igsigd0 = 1 set varselem(igsigd) to zero in next oasemb
! iaem = 1 if marc is called from aem (0 - off - default)
! icosim = 1 if marc is used in co-simulation software (ADAMS-MARC)
! inodels = 1 nodal integration elements 239/240/241 present
! nlharm = 0 harmonic subincrements are linear
! = 1 harmonic subincrements are nonlinear
! iampini = 0 amplitude of previous harmonic subinc is initial estimate (default)
! = 1 zero amplitude is initial estimate
! iphasetr = 1 phase transformation material model is used
! iforminp flag indicating that contact is switched on via the CONTACT
! option in the input file (as opposed to the case that contact
! is switched on internally due to cyclic symmetry or model
! section creation)
! ispecerror = a+10*b (only for spectrum response analysis with missing mass option)
! a=0 or a=1 (modal shape with non-zero shift)
! b=0 or b=1 (recover with new assembly of stiffness matrix)
! icsprg = set to 1 if spring elements present in model
! imol Control flag for molecualr diffusion pass
! imolt Permanent control flag for molecualr diffusion pass
! Note in coupled analysis imolt will remain as one,
! but imol will be zero in stress pass or thermal pass.
! idatafit = run Marc to fit parameters
! iharmpar = 1 if harmonic parameter option is used
! inclcase load case increment use for cyclic plasticity data fitting
! imultifreq flag to indicate how many harmonic magnetodynamic passes are computed in coupled
! magnetodynamic/thermal(/structural) analyses.
! 0 or 1 one pass 2 two passes 3 or more is not supported
! init_elas use elastic stress-strain law as the material tangent for
! the first cycle of an increment
! ifatig = 1 stress-life fatigue
! = 2 strain-life fatigue
! iftgmat = 0 no fatigue material properties in the dat file
! = 1 fatigue material properties in the dat file
!
!***********************************************************************
!$omp threadprivate(/marc_concom/)
!!

66
src/marc/include/creeps2020 vendored Normal file
View File

@ -0,0 +1,66 @@
! common block definition file taken from respective MSC.Marc release and reformated to free format
!***********************************************************************
!
! File: creeps.cmn
!
! MSC.Marc include file
!
real(pReal) cptim,timinc,timinc_p,timinc_s,timincm,timinc_a,timinc_b
integer icfte,icfst,icfeq,icftm,icetem,mcreep,jcreep,icpa,icftmp,icfstr,&
icfqcp,icfcpm,icrppr,icrcha,icpb,iicpmt,iicpa
real(pReal) time_beg_lcase,time_beg_inc,fractol,time_beg_pst
real(pReal) fraction_donn,timinc_ol2
!
integer num_creepsr,num_creepsi,num_creeps2r
parameter(num_creepsr=7)
parameter(num_creepsi=17)
parameter(num_creeps2r=6)
common/marc_creeps/cptim,timinc,timinc_p,timinc_s,timincm,timinc_a,timinc_b,icfte,icfst,&
icfeq,icftm,icetem,mcreep,jcreep,icpa,icftmp,icfstr,icfqcp,icfcpm,icrppr,icrcha,icpb,iicpmt,iicpa
common/marc_creeps2/time_beg_lcase,time_beg_inc,fractol,time_beg_pst,fraction_donn,timinc_ol2
!
! cptim Total time at begining of increment.
! timinc Incremental time for this step.
! icfte Local copy number of slopes of creep strain rate function
! versus temperature. Is -1 if exponent law used.
! icfst Local copy number of slopes of creep strain rate function
! versus equivalent stress. Is -1 if exponent law used.
! icfeq Local copy number of slopes of creep strain rate function
! versus equivalent strain. Is -1 if exponent law used.
! icftm Local copy number of slopes of creep strain rate function
! versus time. Is -1 if exponent law used.
! icetem Element number that needs to be checked for creep convergence
! or, if negative, the number of elements that need to
! be checked. In the latter case the elements to check
! are stored in ielcp.
! mcreep Maximum nuber of iterations for explicit creep.
! jcreep Counter of number of iterations for explicit creep
! procedure. jcreep must be .le. mcreep
! icpa Pointer to constant in creep strain rate expression.
! icftmp Pointer to temperature dependent creep strain rate data.
! icfstr Pointer to equivalent stress dependent creep strain rate data.
! icfqcp Pointer to equivalent creep strain dependent creep strain
! rate data.
! icfcpm Pointer to equivalent creep strain rate dependent
! creep strain rate data.
! icrppr Permanent copy of icreep
! icrcha Control flag for creep convergence checking , if set to
! 1 then testing on absolute change in stress and creep
! strain, not relative testing. Input data.
! icpb Pointer to storage of material id cross reference numbers.
! iicpmt
! iicpa Pointer to constant in creep strain rate expression
!
! time_beg_lcase time at the beginning of the current load case
! time_beg_inc time at the beginning of the current increment
! fractol fraction of loadcase or increment time when we
! consider it to be finished
! time_beg_pst time corresponding to first increment to be
! read in from thermal post file for auto step
!
! timinc_old Time step of the previous increment
!
!***********************************************************************
!!$omp threadprivate(/marc_creeps/)
!!$omp threadprivate(/marc_creeps2/)
!!

View File

@ -55,7 +55,7 @@ module material
character(len=pStringLen), public, protected, allocatable, dimension(:) :: &
material_name_phase, & !< name of each phase
material_name_homogenization !< name of each homogenization
integer(kind(THERMAL_isothermal_ID)), dimension(:), allocatable, public, protected :: &
thermal_type !< thermal transport model
integer(kind(DAMAGE_none_ID)), dimension(:), allocatable, public, protected :: &
@ -160,34 +160,31 @@ subroutine material_init(restart)
integer :: ph, myHomog
class(tNode), pointer :: &
debug_material, & ! pointer to material debug options
phases, &
material_homogenization
character(len=pStringLen) :: sectionName
write(6,'(/,a)') ' <<<+- material init -+>>>'; flush(6)
phases => material_root%get('phase')
print'(/,a)', ' <<<+- material init -+>>>'; flush(6)
phases => config_material%get('phase')
allocate(material_name_phase(phases%length))
do ph = 1, phases%length
write(sectionName,'(i0,a)') ph,'_'
material_name_phase(ph) = trim(adjustl(sectionName))//phases%getKey(ph) !ToDO: No reason to do. Update damage tests
material_name_phase(ph) = trim(adjustl(sectionName))//phases%getKey(ph) !ToDO: No reason to do. Update damage tests
enddo
material_homogenization => material_root%get('homogenization')
material_homogenization => config_material%get('homogenization')
allocate(material_name_homogenization(material_homogenization%length))
do myHomog = 1, material_homogenization%length
write(sectionName,'(i0,a)') myHomog,'_'
write(sectionName,'(i0,a)') myHomog,'_'
material_name_homogenization(myHomog) = trim(adjustl(sectionName))//material_homogenization%getKey(myHomog)
enddo
debug_material => debug_root%get('material',defaultVal=emptyList)
call material_parseMicrostructure
print*, ' Microstructure parsed'
call material_parseMicrostructure()
if (debug_material%contains('basic')) write(6,'(a)') ' Microstructure parsed'; flush(6)
call material_parseHomogenization()
if (debug_material%contains('basic')) write(6,'(a)') ' Homogenization parsed'; flush(6)
call material_parseHomogenization
print*, ' Homogenization parsed'
if(homogenization_maxNgrains > size(material_phaseAt,1)) call IO_error(148)
@ -242,7 +239,7 @@ subroutine material_parseHomogenization
integer :: h
material_homogenization => material_root%get('homogenization')
material_homogenization => config_material%get('homogenization')
material_Nhomogenization = material_homogenization%length
allocate(homogenization_type(material_Nhomogenization), source=HOMOGENIZATION_undefined_ID)
@ -325,18 +322,18 @@ subroutine material_parseMicrostructure
class(tNode), pointer :: microstructure, & !> pointer to microstructure list
constituentsInMicrostructure, & !> pointer to a microstructure list item
constituents, & !> pointer to constituents list
constituent, & !> pointer to each constituent
constituents, & !> pointer to constituents list
constituent, & !> pointer to each constituent
phases, &
homogenization
integer, dimension(:), allocatable :: &
CounterPhase, &
CounterHomogenization
real(pReal), dimension(:,:), allocatable :: &
microstructure_fraction !< vol fraction of each constituent in microstrcuture
microstructure_fraction !< vol fraction of each constituent in microstrcuture
integer :: &
e, &
@ -347,11 +344,11 @@ subroutine material_parseMicrostructure
real(pReal), dimension(4) :: phase_orientation
homogenization => material_root%get('homogenization')
phases => material_root%get('phase')
microstructure => material_root%get('microstructure')
homogenization => config_material%get('homogenization')
phases => config_material%get('phase')
microstructure => config_material%get('microstructure')
allocate(microstructure_Nconstituents(microstructure%length), source = 0)
if(any(discretization_microstructureAt > microstructure%length)) &
call IO_error(155,ext_msg='More microstructures in geometry than sections in material.yaml')
@ -360,7 +357,7 @@ subroutine material_parseMicrostructure
constituents => constituentsInMicrostructure%get('constituents')
microstructure_Nconstituents(m) = constituents%length
enddo
microstructure_maxNconstituents = maxval(microstructure_Nconstituents)
allocate(microstructure_fraction(microstructure_maxNconstituents,microstructure%length), source =0.0_pReal)
allocate(material_phaseAt(microstructure_maxNconstituents,discretization_nElem), source =0)
@ -379,7 +376,7 @@ subroutine material_parseMicrostructure
constituent => constituents%get(c)
microstructure_fraction(c,m) = constituent%get_asFloat('fraction')
enddo
if (dNeq(sum(microstructure_fraction(:,m)),1.0_pReal)) call IO_error(153,ext_msg='constituent')
if (dNeq(sum(microstructure_fraction(:,m)),1.0_pReal)) call IO_error(153,ext_msg='constituent')
enddo
do e = 1, discretization_nElem
@ -394,11 +391,11 @@ subroutine material_parseMicrostructure
enddo
enddo
enddo
do e = 1, discretization_nElem
do i = 1, discretization_nIP
constituentsInMicrostructure => microstructure%get(discretization_microstructureAt(e))
material_homogenizationAt(e) = homogenization%getIndex(constituentsInMicrostructure%get_asString('homogenization'))
constituentsInMicrostructure => microstructure%get(discretization_microstructureAt(e))
material_homogenizationAt(e) = homogenization%getIndex(constituentsInMicrostructure%get_asString('homogenization'))
CounterHomogenization(material_homogenizationAt(e)) = CounterHomogenization(material_homogenizationAt(e)) + 1
material_homogenizationMemberAt(i,e) = CounterHomogenization(material_homogenizationAt(e))
enddo
@ -419,5 +416,5 @@ subroutine material_parseMicrostructure
end subroutine material_parseMicrostructure
end module material

Some files were not shown because too many files have changed in this diff Show More