Some patches for the CUDA version of GPAW to make it run with newer versions of the CUDA software stack.
Source code is available in GPAW's repository as a separate branch called 'cuda'. To obtain the code, use e.g. the following commands:
git clone https://gitlab.com/gpaw/gpaw.git
cd gpaw
git checkout cuda
Script apply.sh can be used to automatically apply all the patches to the
source code, e.g.
bash apply.sh /path/to/gpaw-cuda-source-code
or alternatively you can copy the patches to the source code directory and apply them manually with
# patch CUDA version limit
patch gpaw/gpuarray.py patch-gpuarray.diff
# patch XC headers
patch c/xc/xc_gpaw.h patch-xc_gpaw.diff
patch c/xc/xc_mgga.h patch-xc_mgga.diff
# patch eigensolver/mode check
patch gpaw/paw.py patch-paw.diff
# patch GPGPU pinning on exclusive nodes
patch gpaw/cuda.py patch-cuda.diff
Example GPAW setup script (customize-cuda.py) and Make include file for the
CUDA part (make.inc) are also available. Modify them to match your system
and then copy them to the source code directory, e.g. with
tgt=/path/to/gpaw-cuda-source-code
cp customize-cuda.py $tgt/
cp make.inc $tgt/c/cuda/