You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Trying to compile the main branch using the conda compiler that comes with the Rubin Science Pipelines on MacOS 12.6.3. The compiler is identified with Yay! This version of clang supports GPU! (which seems wrong as macs don't really support nvptx nowadays, but maybe it's capable of cross-compiling?). When checking for cpp11 a bit later in setup.py though, it errors out with:
Trying compile command:
x86_64-apple-darwin13.4.0-clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -fwrapv -O2 -Wall -fPIC -O2 -isystem /Users/josh/src/lsst/conda/miniconda3-py38_4.9.2/envs/lsst-scipipe-5.1.0/include -fPIC -O2 -isystem /Users/josh/src/lsst/conda/miniconda3-py38_4.9.2/envs/lsst-scipipe-5.1.0/include -march=core2 -mtune=haswell -mssse3 -ftree-vectorize -fPIC -fPIE -fstack-protector-strong -O2 -pipe -isystem /Users/josh/src/lsst/conda/miniconda3-py38_4.9.2/envs/lsst-scipipe-5.1.0/include -D_FORTIFY_SOURCE=2 -isystem /Users/josh/src/lsst/conda/miniconda3-py38_4.9.2/envs/lsst-scipipe-5.1.0/include -O2 -msse2 -std=c++11 -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda -Wno-openmp-mapping -Wno-unknown-cuda-version -Wno-shorten-64-to-32 -fvisibility=hidden -DGALSIM_USE_GPU -c /Users/josh/src/GalSim/tmp/tmpmofvgpp3.cpp -o /Users/josh/src/GalSim/tmp/tmpy5z_ilkd.o
Output was:
clang-14: error: cannot find libdevice for sm_35; provide path to different CUDA installation via '--cuda-path', or pass '-nocudalib' to build without linking with libdevice
The compiler x86_64-apple-darwin13.4.0-clang with flags -O2 -msse2 -std=c++11 -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda -Wno-openmp-mapping -Wno-unknown-cuda-version -Wno-shorten-64-to-32 -fvisibility=hidden -DGALSIM_USE_GPU did not successfully compile C++11 code
error: Compiler is not C++-11 compatible
So looks like a GPU problem. Is there a way to force no-gpu with an env variable or something?
The text was updated successfully, but these errors were encountered:
OK, it looks as if it's not safe to assume that a compiler that reports nvptx support can actually perform a successful GPU build. Maybe Apple supplies their core compiler with nvptx enabled but it requires other software to function properly. I'll replace this with a check that actually builds a small GPU program and ensures there are no errors.
Trying to compile the main branch using the conda compiler that comes with the Rubin Science Pipelines on MacOS 12.6.3. The compiler is identified with
Yay! This version of clang supports GPU!
(which seems wrong as macs don't really support nvptx nowadays, but maybe it's capable of cross-compiling?). When checking for cpp11 a bit later in setup.py though, it errors out with:So looks like a GPU problem. Is there a way to force no-gpu with an env variable or something?
The text was updated successfully, but these errors were encountered: