-
Notifications
You must be signed in to change notification settings - Fork 1.9k
error: mismatched argument pack lengths while expanding ‘std::is_constructible #1898
Comments
Hi @elcou sorry for the delay. Is there a particular reason why you need this flag -DCMAKE_CXX_COMPILER=g++-6 ? Can you try building without it? |
No problem, and thanks for the reply. |
Hi @elcou @pjh5. I too am experiencing this same build error. System information Operating system: Ubuntu 18.04 Has any progress been made on this issue? Or any recommendations? |
I've never seen an error coming from a place like this before '/usr/include/c++/6/tuple' Do you also know if your CUDA setup is working? Do the included demos / samples work okay? |
I also have a similar problem to the one described here. I can use/compile cuda software. For instance I have built and run the Tensorflow package with GPU support. System InformationOperating system: Fedora 27 Error/usr/lib64/gcc/x86_64-redhat-linux/6.4.0/include/c++/tuple: In instantiation of ‘static constexpr bool std::_TC<, _Elements>::_NonNestedTuple() [with _SrcTuple = const std::tuple<long unsigned int, long unsigned int, long unsigned int>&; bool = true; _Elements = {long unsigned int&, long unsigned int&, long unsigned int&}]’: make[2]: *** [caffe2/CMakeFiles/caffe2_gpu.dir/build.make:1351: caffe2/CMakeFiles/caffe2_gpu.dir/operators/caffe2_gpu_generated_abs_op.cu.o] Error 1 |
@pjh5 Yes my cuda setup is working fine. |
Actually I see that this was addressed in #1636 which in turn links to https://devtalk.nvidia.com/default/topic/1028112/cuda-setup-and-installation/nvcc-bug-related-to-gcc-6-lt-tuple-gt-header-/ |
Due to the bug described by @pjh5 on the last comment I have tried to use a older version of the GCC (5.5). Using this version, I was able to compile the package with gpu support (Cuda 9.1/Cudnn 7.0.5). However it was also necessary to compile the intel TBB library to avoid linking errors. Since the system version used on Fedora 27 was compiled with a newer GCC version. Is possible to compile the cuda code of caffe2 using the clang compiler ? |
@labor00 we do test Caffe2 with clang on Ubuntu 16.04. The code used to switch gcc to clang is here https://github.com/caffe2/caffe2/blob/master/docker/jenkins/common/install_clang.sh |
hey @elcou , did you manage to solve this? |
@elcou any update? |
I'm running into a very similar issue. I'm using an Intel 8700k processor, which pretty much forced me onto a more up to date linux kernel. I've got a 1080ti which as far as I can tell NVIDIA isn't yet supporting with newer compilers / kernels. I had a stint into Manjaro Linux, then shifted to Ubuntu 18 beta. After reading up on these issues, I got gcc 6.3 going through a ppa (ppa:jonathonf/gcc-6.3)... Then installed the NVIDIA stuff...
|
@iGoog could you try gcc 5 based on #1636 which in turn links to https://devtalk.nvidia.com/default/topic/1028112/cuda-setup-and-installation/nvcc-bug-related-to-gcc-6-lt-tuple-gt-header-/ ? |
cmake -DUSE_CUDA=ON -DBLAS=MKL -DCUDA_HOST_COMPILER=/usr/bin/gcc-5 .. The install part failed... Arguably an even further different issue, possibly my pretty fresh Ubuntu 18 setup (first tried conda, but that didn't work so I uninstalled that and went to pip... Installing libboost-all-dev seemed to help). I can't seem to do the "from caffe2.python import core" outside of the build directory. Having "PYTHONPATH=/usr/local:/usr/bin/python/../.. " doesn't seem to help. It might be my pip itself is messed up, as it pretty much demands I use sudo for anything to install.
|
It's best to run Caffe2 from anywhere except for the build folder, since the local directories can confuse Python import statements. @iGoog if your pip needs sudo then I suggest using a python which does not have this requirement. Can you read https://caffe2.ai/docs/faq.html#why-do-i-get-import-errors-in-python-when-i-try-to-use-caffe2 and https://caffe2.ai/docs/faq.html#how-can-i-find-a-file-library-or-package-on-my-computer and tell me where your libcaffe2.so is and what it's linked against? |
Hi, I'm trying to build caffe2 with GPU support.
cmake configuration runs fine but then when building I get the output as below. Can someone help me with that please ?
Thanks a lot !
System information
CMake summary output
The text was updated successfully, but these errors were encountered: