Search Criteria
Package Details: caffe-cuda-git 1.0.r136.g9b8915401-3
Package Actions
Git Clone URL: | https://aur.archlinux.org/caffe-cuda-git.git (read-only, click to copy) |
---|---|
Package Base: | caffe-cuda-git |
Description: | A deep learning framework made with expression, speed, and modularity in mind (with cuda support, git version) |
Upstream URL: | https://caffe.berkeleyvision.org/ |
Licenses: | BSD |
Conflicts: | caffe |
Provides: | caffe, caffe-cuda, caffe-git |
Submitter: | dbermond |
Maintainer: | dbermond |
Last Packager: | dbermond |
Votes: | 0 |
Popularity: | 0.000000 |
First Submitted: | 2018-08-22 18:02 (UTC) |
Last Updated: | 2022-03-01 14:46 (UTC) |
Dependencies (52)
- boost-libs (boost-libs-gitAUR)
- cuda (cuda11.1AUR)
- gflags (gflags-gitAUR)
- google-glog (glog-gitAUR)
- hdf5 (hdf5-gitAUR, hdf5-openmpi)
- lapack (aocl-libflame-aoccAUR, aocl-libflameAUR, blas-mklAUR, lapack-gitAUR, atlas-lapackAUR, blas-aocl-gccAUR, blas-aocl-aoccAUR, openblas-lapackAUR, blas-openblas)
- leveldb (leveldb-gitAUR)
- lmdb (lumosqlAUR, lmdb-gitAUR)
- nccl (nccl-gitAUR)
- openblas (openblas-lapackAUR)
- opencv (opencv-cuda)
- protobuf (protobuf-gitAUR)
- python (python32AUR, python37AUR, python310AUR)
- python-numpy (python-numpy-flameAUR, python-numpy-mkl-binAUR, python-numpy-gitAUR, python-numpy-mklAUR)
- python-pandas
- boost (boost-gitAUR) (make)
- boost-libs (boost-libs-gitAUR) (make)
- cuda (cuda11.1AUR) (make)
- doxygen (doxygen-gitAUR) (make)
- gflags (gflags-gitAUR) (make)
- ghostscript (make)
- git (git-gitAUR) (make)
- google-glog (glog-gitAUR) (make)
- hdf5 (hdf5-gitAUR, hdf5-openmpi) (make)
- lapack (aocl-libflame-aoccAUR, aocl-libflameAUR, blas-mklAUR, lapack-gitAUR, atlas-lapackAUR, blas-aocl-gccAUR, blas-aocl-aoccAUR, openblas-lapackAUR, blas-openblas) (make)
- leveldb (leveldb-gitAUR) (make)
- lmdb (lumosqlAUR, lmdb-gitAUR) (make)
- nccl (nccl-gitAUR) (make)
- openblas (openblas-lapackAUR) (make)
- opencv (opencv-cuda) (make)
- protobuf (protobuf-gitAUR) (make)
- python (python32AUR, python37AUR, python310AUR) (make)
- python-numpy (python-numpy-flameAUR, python-numpy-mkl-binAUR, python-numpy-gitAUR, python-numpy-mklAUR) (make)
- python-pandas (make)
- texlive-core (texlive-installerAUR, texlive-fullAUR, texlive-basic) (make)
- texlive-latexextra (texlive-installerAUR, texlive-fullAUR) (make)
- cython (cython-gitAUR, cython0) (optional)
- ipython (ipython-gitAUR) (optional)
- python-dateutil (optional)
- python-gflags (optional)
- python-h5py (python-h5py-gitAUR, python-h5py-openmpi) (optional)
- python-leveldbAUR (optional)
- python-matplotlib (python-matplotlib-gitAUR) (optional)
- python-networkx (python-networkx-gitAUR) (optional)
- python-nose (optional)
- python-pillow (python-pillow-gitAUR) (optional)
- python-protobuf (python-protobuf-gitAUR) (optional)
- python-pydotplusAUR (optional)
- python-scikit-imageAUR (optional)
- python-scipy (python-scipy-mkl-binAUR, python-scipy-gitAUR, python-scipy-mklAUR) (optional)
- python-six (optional)
- python-yaml (python-yaml-gitAUR) (optional)
Required by (2)
- multitask-network-cascades-dr-git (requires caffe)
- openpose (requires caffe)
Latest Comments
dbermond commented on 2024-03-06 01:10 (UTC)
@gardotd426 This package itself is not abandoned. What seems to be abandoned is the upstream caffe project, which was superseded by caffe2, which in turn was incorporated by pytorch. The latest caffe upstream release was in 2017, and the latest caffe upstream commit was in 2020. That's why you do not see any updates here the packaging side. In such scenario, package updates are only needed here if something breaks and/or needs adjustment. It looks like that the building is broken again, as it fails to compile against the latest glog. I will look at the source and try to add a fix, but I cannot guarantee anything. If you have a fix for this problem, please post a link to a patch. If we cannot fix it, the best thing to do is to delete the caffe packages from the AUR, since it's pretty abandoned upstream and mostly unmaintainable. As upstream caffe does not get any official attention and/or fixes anymore, it becomes more and more difficult to maintain this package, since it tends to always break when newer versions of dependencies are released, without any official upstream maintenance over that.
gardotd426 commented on 2024-03-05 09:34 (UTC)
This is all sorts of broken right now. Won't compile whatsoever.
I ran an update which updated protobuf and gflags, and at the end of the update where it says "checking for packages that need to be rebuilt" or whatever, it listed caffe-cuda-git (as well as a few others). The others all built fine (I don't believe they use the same dependencies, it was a large update), but caffe-cuda-git does NOT build. There are a MILLION
error: #error <glog/libraryname.h> was not included correctly. See the documentation for how to consume the library
, with<glog/libraryname.h>
being basically every glog library.Obviously the comments here are old as dirt, so they're irrelevant, but I did try uninstalling caffe-cuda-git and any caffe packages and reinstalling from scratch, but it changed absolutely nothing.
caffe-cuda
non-git fails for entirely different reasons.Is this package abandoned? It's not been updated in over 2 years at this point, and there hasn't been a comment on this page for almost 4 years.
Here's a sample of some of the error messages, but it's only like a third or half of them, but they're all the same thing mostly:
dbermond commented on 2018-10-03 17:47 (UTC)
@muffehazard Uninstall the any caffe package before building. It should work fine.
muffehazard commented on 2018-10-03 05:44 (UTC)
@dbermond I don't think libprotobuf was causing this, because the linker failed again when I tried to upgrade today. This time it said it can't find libcudart.so.9.2, libcublas.so.9.2 and libcurand.so.9.2, but again when I renamed libcaffe.so to libcaffe.so.old in /usr/lib the build completed without errors. Checking ldd /usr/lib/libcaffe.so after the build, it's now using 10.0 -versions of the aforementioned libraries. I still think that the build is trying to use the old .so from the system, which causes the build to fail if any of the packages providing any of the linked libraries have been upgraded. I don't know why but this seems to be the case and my workaround also seems to indicate this.
dbermond commented on 2018-09-29 01:08 (UTC)
Guys, this highly looks like an issue with your system. You should not have libprotobuf.so.16 installed because it's not on the official repositories. I wonder why this file is on your system. I do not have this file on my system.
Use
pacman -Qo /usr/lib/libprotobuf.so.16
to find what package owns it and remove this given package. If no package owns this file, you can delete libprotobuf.so.16. After this, reinstall protobuf and rebuild caffe-cuda-git.Package is building fine for me. I've have even updated the package to include nccl since it is now open source again and has landed on the official repositories.
muffehazard commented on 2018-09-26 05:31 (UTC)
Deleting or renaming /usr/lib/libcaffe.so before building resolved this for me, maybe there is an issue where the linker tries to use the old libcaffe.so from /usr/lib instead of the newly built libcaffe.so during build?
Ape commented on 2018-09-24 13:44 (UTC)
I also have this issue with libprotobuf 16 vs 17. I cannot find anything on my system that would provide the older version, but caffe still seems to want it.
dbermond commented on 2018-09-21 23:15 (UTC)
@elmuz I cannot reproduce your issue. Package is building fine for me. My libcaffe.so links against libprotobuf.so.17, which is the current protofub library on the official repositories.
It seems that your libcaffe.so is wrongly linking against libprotobuf.so.16. Try to find what whatever is providing it and remove it.
elmuz commented on 2018-09-18 12:58 (UTC) (edited on 2018-09-18 13:09 (UTC) by elmuz)
Hi I cannot build against libprotobuf. In fact I get warning: libprotobuf.so.16, needed by /usr/lib/libcaffe.so, not found (try using -rpath or -rpath-link). Problem is that I have version 17 instead of 16. Is there any workaround for that?