Library aimed to enable experimentation with different floating point formats, compression, and introducing randomness/stochasticity in codes based on C++ and Fortran.
The idea is to provide a lightweight wrapper for different libraries and functionalities that modify IEEE floating point values, be it to a lower floating point precision, compression, or adding noise or other random disturbances.
The library revolves around the object pcs_struct which is initialized with the parameters one wants to modify and provides an in-place function pcs which carries out the transformation of the input vector.
git clone --recurse-submodules git@github.com:MartinKarp/pcs.git
Set correct compilers in makefile and for cpfloat in external/cpfloat.
make cpfloat
export INSTALL_LOCATION=WHEREVER
make PREFIX=${INSTALL_LOCATION}
default prefix is pcs/install.
After this using pcs should be a matter of adding
-L${INSTALL_LOCATION}/lib -I${INSTALL_LOCATION}/include
linking with
-lpcs_f -lpcs -lcpfloat -lpcg_random
and adding
${INSTALL_LOCATION}/lib
to the LD_LIBRARY_PATH.