Data

[gsvgit]

Generators for graphs. Huge graphs are compressed. I'm not sure should we provide GTgraph as sources or as precompiled binaries.

[SokolovYaroslav]

We should not store .o files in git.
I did not understand where Data.tar.xz is. I see only Matrices.tar.xz (by the way, in several places).
Why do we need all these .c and .h files? Does this program generate test files or something like that?

[gsvgit]

@SokolovYaroslav 1) I think, that we can compile it from sources. Yes, .o files in git is a bad idea.
2) It is my mistake. Now I compress only big matrices for some cases. Readme should be updated. 3) It is a special graph generator. We use it for random sparse graphs generation.

[SokolovYaroslav]

So, waiting for the changes that you described.

[SokolovYaroslav]

@gsvgit
3) Are these random graphs automatically generated as part of benchmarking, or user should generate them manually?

[gsvgit]

@SokolovYaroslav

1. Docker image is changed. GTgraph is added.
2. Readme is updated (in PR)
3. Random graphs (and other graphs) are automatically generated by init script which should be run manually (https://github.com/SokolovYaroslav/CFPQ-on-GPGPU/pull/34/files#diff-200fddbfb083dd7a5f02df17f800800a). This script unpacks graphs for RDFs and static code analysis (it is too hard to generate these files) and generates other files (Random, Full, Worst).

[SokolovYaroslav]

0. I haven't computer with NVidia. so I can't test anything.
1. Such python call CMD ["python", "testing_system.py", "tests.csv", "init.py"] actually works? And what its behavior? It first runs testing_system.py with argument tests.csv and then runs init.py? Seems very strange.
2. In readme: 3. Run 'init.py' script to initialize environment.. You want to user runs init.py from Docker? I think, we should describe this as standalone step from the same docker image, but with different script.

[gsvgit]

@SokolovYaroslav Should we move existing implementations compilation to Docker image initialization step? For example, I can include the existing Python code into init.py. In my opinion, the testing system should not include a compilation step.

[gsvgit]

@SokolovYaroslav

1. Fixed. Now python interpreter is specified as the entry point. So, the user can execute arbitrary python script inside docker.
2. init.py should be executed by the user inside docker. Instructions are added into Readme.

[SokolovYaroslav]

Now it seems conveniently. If it works (I can't test), here is my approve.

[SokolovYaroslav]

We have two Matrices.tar.xz. Is it okay? I can't verify even that they are not same, because of the git lfs quota (at the moment no one can download the data of this repo from lfs).

[gsvgit]

@SokolovYaroslav Yes, we should have 2 different Matrices.tar.xz. They should be placed in different folders.