-
Notifications
You must be signed in to change notification settings - Fork 372
New issue
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # 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: