LinuxQuestions.org - Installing and packaging Caffe on Slackware

- Slackware (https://www.linuxquestions.org/questions/slackware-14/)

- - Installing and packaging Caffe on Slackware (https://www.linuxquestions.org/questions/slackware-14/installing-and-packaging-caffe-on-slackware-4175632155/)

Installing and packaging Caffe on Slackware

Hi,

I use Slackware64-current.
I have a working nvidia/cudatoolkit/cudnn setup.

I have this in my .bashrc:

Code:

CUDA_HOME=/usr/share/cuda

I installed those deps from sbopkg (may have a few others previously installed not listed here):
openblas
protobuf
hdf5
gflags
glog
lmdb
leveldb
snappy

I did:

Code:



git clone https://github.com/BVLC/caffe.git



sed -e "s/# USE_CUDNN := 1/USE_CUDNN := 1/" caffe/Makefile.config.example > caffe/Makefile.config

sed -i -e "s/CUDA_DIR := \/usr\/local\/cuda/CUDA_DIR := \/usr\/share\/cuda/" caffe/Makefile.config

sed -i -e "s/BLAS := atlas/BLAS := open/" caffe/Makefile.config

sed -i -e "s/\/usr\/lib/\/usr\/lib64/" caffe/Makefile.config

sed -i -e "s/\/usr\/local\/include/\/usr\/include/" caffe/Makefile.config



sed -i -e "s/# OPENCV_VERSION := 3/OPENCV_VERSION := 3/" caffe/Makefile.config

# to get rid of => nvcc fatal  : Unsupported gpu architecture 'compute_20'

sed -i -e "s/-gencode arch=compute_20,code=sm_20//" caffe/Makefile.config

sed -i -e "s/-gencode arch=compute_20,code=sm_21//" caffe/Makefile.config





make all -j8

make test

make runtest

Everything is ok, but now I have no idea how to make a package from this.

Any help is appreciated.

Code:

makepkg

? No?

Neither do I, frankly, because the information is a bit garbled and the make procedure a bit odd. Some of those appear to have non standard make systems.

The usual thing I do is grep the Makefile for DESTDIR

Code:

grep DESTDIR Makefile*

If the DESTDIR is there, simply use an install directory. I use /tmp/pack. Run

Code:

make DESTDIR=/tmp/pack install

at the install stage

If there's a non standard make system, you may have to copy files into /tmp/pack. /tmp/pack is the / for your package, so you would use /tmp/pack/usr/lib64 or whatever.

Finally, chdir to /tmp/pack and run

Code:

makepkg ../packagename.txz

to make up the package.

It's not that simple as there's no rule for "install".

It's simple. Create the package directories manually, copy the built binaries and other stuff you need into those directories, add a slack-desc file and some documentation, then call makepkg.
If you think that that's not simple then simply search slackbuilds.org for a SlackBuild script that does roughly the same and use that for your own package.

Unfortunately, the folder seems to contain only includes, build is empty.
$ ls
CMakeLists.txt CONTRIBUTORS.md LICENSE Makefile.config README.md caffe.cloc data docker examples matlab python src
CONTRIBUTING.md INSTALL.md Makefile Makefile.config.example build cmake distribute docs include models scripts tools

I would expect replacements for libGL.so and other libraries, and a few executable binaries. Nvidia is a reasonably orderly in their packages. Is there a Makefile? It will have an install section. Otherwise, it's in the bash script.Includes are usually given, not made. Check subdirectories.

I have this in the Makefile:
LIB_BUILD_DIR := $(BUILD_DIR)/lib

I searched the folders, there's only empty bin and lib folders in ./distribute
Something also annoying is that the CMake build doesn't seem to use the Makefile.config

I made a little progress:
Because of the error with make distribute:
/usr/bin/ld: cannot find -lboost_python

I did:

ln -s /usr/lib64/libboost_python27.so /usr/lib64/libboost_python.so

and

sed -i -e "s/dist-packages/site-packages/g" caffe/Makefile.config

permitted to use the

make distribute

command.
The ./distribute folder is now populated.

Have you libGL{version}.so, and the other libGL* libs - about 3 of them?

Nope, there's nothing GL-related in the whole source/build folder.

I was able to use caffe with linking here, not packaging and installing, I guess that's ok for me now.