Closed iesahin closed 3 years ago
By the way, when I install via pip3
CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda CUDA_ARCH=52 pip3 install --user 'git+https://github.com/src-d/kmcuda.git#subdirectory=src'
Collecting git+https://github.com/src-d/kmcuda.git#subdirectory=src
Cloning https://github.com/src-d/kmcuda.git to /tmp/pip-6t9eg3vb-build
Collecting numpy (from libKMCUDA==6.2.2)
Using cached https://files.pythonhosted.org/packages/35/d5/4f8410ac303e690144f0a0603c4b8fd3b986feb2749c435f7cdbb288f17e/numpy-1.16.2-cp36-cp36m-manylinux1_x86_64.whl
Installing collected packages: numpy, libKMCUDA
Running setup.py install for libKMCUDA ... done
Successfully installed libKMCUDA-6.2.2 numpy-1.16.2
it gives import error for numpy.core.multiarray
python3
Python 3.6.7 (default, Oct 22 2018, 11:32:17)
[GCC 8.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import libKMCUDA
ImportError: numpy.core.multiarray failed to import
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ImportError: numpy.core.multiarray failed to import
>>>
Sorry, the last one with pip
is from botched numpy versions or something like that. I installed via pip
as root
and it looks OK.
Is the problem from installng as user/virtual env?
Most likely yes, it requires native compilation and detects numpy, which may leak through the virtual environment.
First
cmake
Then,
make
seems OK.
Now, Python
This is in Ubuntu 18.04 with CUDA 10. Device is GeForce 980 Ti, hence capability = 5.2.
It's in a virtual environment, but I tried without it too and no difference. Probably a link error but where may I look at?