Skip to content

Latest commit

 

History

History
190 lines (148 loc) · 5.42 KB

INSTALL.md

File metadata and controls

190 lines (148 loc) · 5.42 KB
Raw

Installation

These instructions show how to install SOOPERCOOL and setting up a working environment to run the pipeline and loading all its dependencies on NERSC.

Option 1 - Load master_env

Load the master_env module

The module can be loaded with

module use --append /pscratch/sd/s/susannaz/conda_envs/master_env/modulefiles
module load master_env/0.0.3

This will automatically load an existing conda environment and all the modules needed to run SOOPERCOOL and its dependencies, including toast and NaMaster.

Make your own master_env conda environment

If you wanted to create your own conda environment based on master_env and activate it,

conda create --clone /pscratch/sd/s/susannaz/conda_envs/master_env -p <path-to-your-env>
conda activate <path-to-your-env>

This will allow you to access all the pip-installed packages that are in the existing conda environment, but will not load any additional modules.

Option 2 - Create a custom environment using micromamba

Micromamba is a programming environment similar to conda, but based on C and therefore much quicker at resolving and installing packages.

"${SHELL}" <(curl -L micro.mamba.pm/install.sh)

You will see some installation messages showing up and are asked to confirm the location of your micromamba environment files. Then apply the following settings:

source ~/.bashrc
micromamba config append channels conda-forge
micromamba config set channel_priority strict

SOOPERCOOL environment

  • Set up environment (python=3.6 proved to work with NaMaster)

    micromamba create -n soopercool python=3.6 numpy scipy ipython matplotlib healpy astropy -y
micromamba activate soopercool

  • Install GSL from source (otherwise NaMaster won't recognize it)

    wget https://mirror.kumi.systems/gnu/gsl/gsl-2.7.tar.gz
tar -zxvf gsl-2.7.tar.gz
cd gsl-2.7
./configure --prefix=/global/u2/k/kwolz/software/gsl
make && make -j4 install
echo "export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/global/u2/k/kwolz/software/lib" >> ~/.bashrc
echo "export CFLAGS='-I/global/u2/k/kwolz/software/include'" >> ~/.bashrc
echo "export LDFLAGS='-I/global/u2/k/kwolz/software/lib'" >> ~/.bashrc
source ~/.bashrc

  • Install NaMaster and its dependencies

    micromamba install -c conda-forge cfitsio fftw
micromamba install -c conda-forge namaster

  • Install soopercool pipeline and its dependencies

    git clone git@github.com:simonsobs/so_noise_models.git
cd so_noise_models; pip install -e .

TOAST environment

  • Set up environment

    micromamba create -n toast python=3.9 numpy scipy ipython matplotlib pyfftw=0.12 \
make cmake c-compiler cxx-compiler fortran-compiler -y
micromamba activate toast

  • Install dependencies

    module load cpu PrgEnv-gnu cray-mpich craype-accel-nvidia80
micromamba install -c conda-forge cudatoolkit
micromamba install -c conda-forge cuda-nvcc
env MPICC=/opt/cray/pe/mpich/8.1.25/ofi/gnu/9.1 pip install --force --no-cache-dir \
--no-binary=mpi4py mpi4py

    (This will install mpi4py at ${HOME}/micromamba/envs/toast/lib/python3.9/site-packages/mpi4py.)

    pip install pshmem posix_ipc astropy==5.2 healpy pytest

    (Don't use conda for the preceding packages; this leads to conflicts!)

  • Install TOAST

    pip install --pre toast

    (--pre causes the latest toast3 version to get installed)

  • Run toast test (passes in ~15 min):

    python -c "import toast.tests; toast.tests.run()"

    (SOLVED: issue regarding posix_ipc (undefined symbol: shm_unlink): Install pshmem and posix_ipc with pip, not micromamba.

  • Install spt3g and so3g

    • cd ${HOME}/spt3g_software
mkdir -p build
cd build
cmake \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_C_COMPILER="gcc" \
  -DCMAKE_CXX_COMPILER="g++" \
  -DCMAKE_C_FLAGS="-O3 -g -fPIC" \
  -DCMAKE_CXX_FLAGS="-O3 -g -fPIC -std=c++11" \
  -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON \
  -DPython_EXECUTABLE:FILEPATH=$(which python3) \
  -DPYTHON_MODULE_DIR="${HOME}/toast/lib/python3.9/site-packages" \
  -DCMAKE_INSTALL_PREFIX="${HOME}/toast" \
  ..
make -j 2 install

    • micromamba install boost libflac
cd ~/git/OpenBLAS
make
make install PREFIX=""${HOME}/micromamba/envs/toast"
cd ${HOME}/so3g
mkdir -p build
cd build
cmake \
  -DCMAKE_PREFIX_PATH=""${HOME}/spt3g_software/build" \
  -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON \
  -DPYTHON_INSTALL_DEST="${HOME}/toast" \
  -DCMAKE_INSTALL_PREFIX="${HOME}/toast" \
  ..
make -j 2 install

    • pytest ~/so3g/test passes (38 passed, 3 warnings).

  • Install sotodlib following Bai-Chang's instructions.

    cd ~
git clone https://github.com/simonsobs/sotodlib.git
cd sotodlib
pip install -e .

    python setup.py test passes in ~3 min (78 tests).

    ERROR (solved):

    MPICH ERROR [Rank 0] [job id ] [Thu Sep 21 06:09:17 2023] [login17] - Abort(-1) (rank 0 in comm 0): MPIDI_CRAY_init: GPU_SUPPORT_ENABLED is requested, but GTL library is not linked
 (Other MPI error)

aborting job:
MPIDI_CRAY_init: GPU_SUPPORT_ENABLED is requested, but GTL library is not linked

    Solution: module unload gpu; module load cpu cudatoolkit

  • Test script using ${HOME}/sotodlib/workflows/toast_so_sim.py