-
Notifications
You must be signed in to change notification settings - Fork 72
Replies: 2 comments · 5 replies
-
Dear Francesco,
or whatever the path to your gnu compilers is. And then run the cmake command without the The fact that you cannot pull the dependencies automatically is certainly annoying here, but it should work. After I see the cmake output we can have a look. Also have a look here: https://github.com/flatironinstitute/ccq-software-build/blob/main/triqs/3_stable_nix2.1_llvm/install.sh to see how we build triqs with a bash script on our cluster if that helps in any way. |
Beta Was this translation helpful? Give feedback.
All reactions
-
Dear, Thank you very much for your fast reply. So I am doing what you suggested replacing mpicc with g++. export CC=/ccc/products2/gcc-11.1.0/Rhel_8__x86_64/system/default/bin/gcc cmake .. The output of cmake is: -- Installation directory will be /ccc/cont003/dsku/blanchet/home/user/impmc/cassolfr/Code/Build/TRIQS Then make: Scanning dependencies of target h5_c but in the end make test output: 92% tests passed, 21 tests failed out of 251 Total Test time (real) = 168.15 sec The following tests FAILED: Could it be related to my python3 virtual environment? In any case, thanks a lot, the situation seems to be already improved. Concerning the mpi version, I am nearly sure that the mpi version has been compiled with gnu/11.1.0. The location of mpicc is: Best Regards, Francesco |
Beta Was this translation helpful? Give feedback.
All reactions
-
okay that looks not too bad now. Good progress. Could you go into the build folder with your modules loaded and run:
and show me the output? We need to get an idea what fails here. It seems related to python / your virtual env, but I am not sure yet how. |
Beta Was this translation helpful? Give feedback.
All reactions
-
Dear Alex, Thank you for your precious help. The compilation process went fine. The issue with the test was a problem with the fftw3 version, that I solved changing the module. To recap I finally installed TRIQS by using the following script: ################################################################################ module purge #export LD_PRELOAD=/ccc/products2/gcc-11.1.0/Rhel_8__x86_64/system/default/lib64/libstdc++.so.6.0.29 source /ccc/work/cont003/gen12043/cassolfr/triqs-env/bin/activate cmake .. make -j 8 &> Out.Make make test &> Out.test make install Everything worked fine. python3 Bethe_Vanilla.py the code works perfectly. Unfortunately to submit job in the cluster I have to do it properly by the script: ############################################################################################ !/bin/bash module purge export LD_PRELOAD=/ccc/products2/gcc-11.1.0/Rhel_8__x86_64/system/default/lib64/libstdc++.so.6.0.29 source /ccc/work/cont003/gen12043/cassolfr/triqs-env/bin/activate export LD_LIBRARY_PATH=/ccc/products2/gcc-11.1.0/Rhel_8__x86_64/system/default/lib64:$LD_LIBRARY_PATH #export OMP_NUM_THREADS=1 ccc_mprun -n 2 python3 Bethe_Vanilla.py 1>test.out 2>test.err ################################################################################# ################################################################################### Starting serial run at: 2023-06-15 14:42:48.496317 .. Error .. in calling C++ overload of constructor : That shows that the mpi environment is not recognized by the code. All the Do you have any idea about why the ccc_mprun flag could not work? I reply a bit late because I wanted to figure it out myself but without success. Best Regards, Francesco |
Beta Was this translation helpful? Give feedback.
All reactions
-
Hi Francesco, I see you are running the code in the scratch directory. Can you try to run it in the work directory instead? Just change #MSUB -m scratch into #MSUB -m work And run the code somewhere in your $CCCWORKDIR directory. Best, Michel |
Beta Was this translation helpful? Give feedback.
All reactions
-
Dear Michel, Thank you for your suggestion! Best, Francesco |
Beta Was this translation helpful? Give feedback.
All reactions
-
Hi, I had this error also multiple times. Apart from Michel already suggested, sometimes I had the feeling that the cluster file systems work slow on these files and files are not given free in time. Can you try to rename in the script the name of the h5 archive and run again? And also make sure that no other process like ipython, jupyter is accessing it? Just to check, you are using the same script, i.e. using the If this does not help, we have to see how your h5 library is compiled. Maybe something is wrong with it... |
Beta Was this translation helpful? Give feedback.
-
Dear all,
My name is Francesco Cassol, I am a PhD at Sorbonne University. I am a TRIQS user since a year and an half to perform Dynamical Mean Field Theory calculations. Recently I have to move to a new cluster, TGCC Irene, in the Rome partition. It is being a couple of week that I am trying to install it there, without success. So, my question is: Is there anyone in the community that was able to perform the installation? Could someone provide me the right combination of modules and the cmake flags to compile it?
I also put here information on my first attempts.
First, since the normal installation of TRIQS would require a internet connection to download all the dependencies from GitHub but TGCC does not allow this,I had to download all the sources by hand and copy them in the path:
$TRIQS_SRC/deps/$name
where $name is the dependency. In total we have that $name= Cpp2py, GTest, mpi , h5, itertools, nda.
Then I have to upload all the modules. I selected the following:
module load gnu/11.1.0
module load mpi/openmpi/4.1.4
module load flavor/python3/cuda
module load python3/3.10.6
module load boost/1.74
module load hdf5
module load fftw3/mkl
module load eigen
module load blas/mkl
Since the python3 version on the cluster does not contains the package mako, I installed a virtual python environment called 'TRIQS_ENV' that has it installed and I activate it by means of:
source $PATH_TO_TRIOS_ENV/bin/activate
then, doing cmake:
cmake ..
-DCMAKE_INSTALL_PREFIX=$WHERE_TO_INSTALL
-DCMAKE_C_COMPILER=$WHICH_GCC
-DCMAKE_CXX_COMPILER=$WHICH_MPICC
-DCMAKE_CXX_FLAGS='-L/$PATH_TO_LIBINTL.SO -lintl' \ # I had to include it to manually specify the position of libinlt.so, not found by cmake
-DCMAKE_VERBOSE_MAKEFILE:BOOL=ON \
It construct the building files without any issue.ss
But then, calling make I get the following error:
../../../../lib/libgtest.so.1.13.0: error: undefined reference to 'std::runtime_error::~runtime_error()'
../../../../lib/libgtest.so.1.13.0: error: undefined reference to 'std::basic_ostream<char, std::char_traits >& std::operator<< <std::char_traits >(std::basic_ostream<char, std::char_traits >&, char const*)'
../../../../lib/libgtest.so.1.13.0: error: undefined reference to 'std::__cxx11::basic_string<char, std::char_traits, std::allocator >::_M_create(unsigned long&, unsigned long)'
../../../../lib/libgtest.so.1.13.0: error: undefined reference to 'std::__cxx11::basic_string<char, std::char_traits, std::allocator >::find(char, unsigned long) const'
And after a lot of errors on GTest.
I also tried to rewrite the -DCMAKE_CXX_FLAGS adding at the end 'std=c++11' to manually specify the standart but nothing actually changed.
What could this error be related?
However, as the error is related to GTest, I also tried the following:
cmake ..
-DCMAKE_INSTALL_PREFIX=$WHERE_TO_INSTALL
-DCMAKE_C_COMPILER=$WHICH_GCC
-DCMAKE_CXX_COMPILER=$WHICH_MPICC
-DCMAKE_CXX_FLAGS='-L/$PATH_TO_LIBINTL.SO -lintl' \ # I had to include it to manually specify the position of libinlt.so, not found by cmake
-DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
-DBuild_Tests=OFF
where the -DBuild_Tests=OFF has been added. In that case the installation is complete but I cannot run any test to check whether everything is correct.
Unfortunately updating the envorinment variables by:
source $PATH_TO_TRIQS/share/triqsvars.sh
and launching the interpreter I fail loading triqs module as in the following example, where I try to load triqs/gf:
python 3.10.6 (main, Feb 3 2023, 12:15:43) [GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
None of the two strategies then seemed to work. Are the two error related? Could be something related to the wrong modules I am uploading? Is there any of you that is currently using TRIQS on that machine that could provide me the module list and the cmake command to install it?
I thank you for the time to read,
Best Regards,
Francesco Cassol
Beta Was this translation helpful? Give feedback.
All reactions