Hello, everyone.

I have written a few SlackBuilds, which I needed for myself, and they sort of work on my machine, but I'd like some second opinion before I can relax about their quality.

Could someone test if they work on your machines?

Googletest and benchmark should need glog and gflags from SBo,
suitesparse should need CUDA and gcc-5, and ceres should need all of the above.

Ceres installs tests and demos in /usr/lib64/ceres-solver/, and data into /usr/share/ceres-$VERSION, maybe not the best place for those, as they both can't fine each other, but in the default build they are also runnable only if given special care (that is, data files given explicitly).

So any feedback would be welcome. Thanks in advance.

https://gitlab.com/Lockywolf/lwfslackbuilds/

I'm not sure if you already spotted it, but googletest is already on SBo

https://slackbuilds.org/repository/1...lopment/gtest/

if you need version 1.8.1 or some particular build options maybe you can ping its maintainer...

also, I don't know if you want to push this stuff later on SBo, but it seems you are using different variables compared to SBo templates: as they are standards in all of SBo's scripts you might have to alter your SlackBuilds prior to submit them there...

I did use SBo's original variables at first, but found them so confusing that it proved to be quicker to write an lwf2sbo conversion sed script rather than try to remember that $TMP is not /tmp

I will proofread the script's output before submitting. Moreover, the submissions are not yet open for 15.0

The environment variables are all the same though (TAG and such), so the scripts should be runnable from sbopkg without modification.

Since you had (in another thread) the problem with gcc version for cudatoolkit on -current such that it needed gcc-5, have you considered the possibility that a more recent version of cuda might solve the problem of needing gcc-5? The SBo cudatoolkit is very old - maybe there are reasons for not updating it or maybe no one has tried. There have been 5 cuda releases since the version at SBo (the latest Feb 2019). I can't believe NVidia would release newer versions that don't compile with the latest or recent gcc.

Also, since cudatoolkit is already such a big install, I've been hesitant to make it a mandatory requirement for suitesparse on 14.2, where suitesparse detects cuda if it is installed and compiles without problems. Then (at least on -current) it seems another gcc is needed too with existing cudatoolkit.

chris

Well, you have a point.

The problem in my particular case is that I use Nvidia GF119, which is old as hell and only supports sm_21 and compute_20, and SBo's CUDA is the last one supporting this architecture. Although even it throws warnings about this architecture being deprecated "soon".

The point about mandatory CUDA is also totally valid, and I will make it optional, thanks for the suggestion.