-
Notifications
You must be signed in to change notification settings - Fork 375
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Installation issues #88
Comments
It seems that Also, the error says that |
I've found that i can get a working install if I run this docker script without the pip3 install MinkowskiEngine, but instead allow it to finish, run the dockerfile, git clone the repository, and edit the Makefile as the following diff --git a/Makefile b/Makefile index 86d8267..f9cdac8 100644 --- a/Makefile +++ b/Makefile @@ -8,8 +8,7 @@ Q ?= @ # CPU_ONLY := 1 CXX ?= g++ -PYTHON ?= python - +PYTHON ?= python3.7 EXTENSION_NAME := minkowski # BLAS choice: @@ -38,7 +37,7 @@ INCLUDE_DIRS += $(PYTORCH_INCLUDES) LIBRARY_DIRS := $(PYTORCH_LIBRARIES) # Determine ABI support -WITH_ABI := $(shell python -c 'import torch; print(int(torch._C._GLIBCXX_USE_CXX11_ABI))') +WITH_ABI := $(shell $(PYTHON) -c 'import torch; print(int(torch._C._GLIBCXX_USE_CXX11_ABI))') # Determine platform UNAME := $(shell uname -s) @@ -58,8 +57,8 @@ endif ifneq ($(CPU_ONLY), 1) # CUDA ROOT DIR that contains bin/ lib64/ and include/ # CUDA_DIR := /usr/local/cuda - CUDA_DIR := $(shell python -c 'from torch.utils.cpp_extension import _find_cuda_home; print(_find_cuda_home())') - + CUDA_DIR := $(shell $(PYTHON) -c 'from torch.utils.cpp_extension import _find_cuda_home; print(_find_cuda_home())') + INCLUDE_DIRS += ./ $(CUDA_DIR)/include LIBRARY_DIRS += $(CUDA_DIR)/lib64 endif @@ -105,7 +104,7 @@ else ifeq ($(BLAS), blas) else # ATLAS LIBRARIES += atlas - ATLAS_PATH := $(shell python -c "import numpy.distutils.system_info as si; ai = si.atlas_info(); [print(p) for p in ai.get_lib_dirs()]") + ATLAS_PATH := $(shell $(PYTHON) -c "import numpy.distutils.system_info as si; ai = si.atlas_info(); [print(p) for p in ai.get_lib_dirs()]") LIBRARY_DIRS += $(ATLAS_PATH) endif then I can run python3.7 setup.py install and it all works fine. When I googled symlinking python3.7 to python I saw stack exchanges of warnings not to since it could mess up some low level ubuntu stuff?I should mention I am only using python 3.7 here over 3.6 as I want this install to be able to run FCGF which states that it needs 3.7. I wondered what you thought about making these changes to the Makefile on pip as this seems to just grab the correct version of python? |
Thanks for the updated Makefile. I'll update the pip package in a few days. |
strange, when I execute the line: python -c 'from torch.utils.cpp_extension import _find_cuda_home; print(_find_cuda_home())' it output: /usr my After uncommented the Using mkl
CXX src/region.cpp
...
NVCC src/pooling_avg.cu
nvcc fatal : Path to libdevice library not specified
Makefile:165: recipe for target 'objs/cuda/union.o' failed I solved the issue by export export PATH="/usr/local/cuda-10.1/bin:$PATH" The reason it can't be found is that I installed the cuda libary with next time just install by wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-ubuntu1604.pin
sudo mv cuda-ubuntu1604.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget http://developer.download.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda-repo-ubuntu1604-10-1-local-10.1.243-418.87.00_1.0-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1604-10-1-local-10.1.243-418.87.00_1.0-1_amd64.deb
sudo apt-key add /var/cuda-repo-10-1-local-10.1.243-418.87.00/7fa2af80.pub
sudo apt-get updatesudo apt-get -y install cuda |
* modelnet40 * WIP * training working * Working * Save checkpoint * InfSampler * No removed print * cache / iter based training * test logit move * reduce size * num worker update * remove reset_seed
Hi!
I've tried to install MinkowskiEngine via pip and have had different issues on every machine I've tried. To try isolate this install from any previous install/uninstalls, I'm trying to make a docker container that runs Minkowski Engine. Below is my current Dockerfile.
Building this without RUN python3.7 -m pip install -U MinkowskiEngine works fine and posts appropriate nvidia-smi and nvcc -V results (Driver Version: 418.87.00 CUDA Version: 10.1) but running the ME pip install gives errors (middle part removed as it's just loads of similar 'In file included from...' errors):
Any ideas? I'd love the help!
The text was updated successfully, but these errors were encountered: