Spliced Experiment Runs

Hoping we can do a subset of runs on here using the Docker experiment container! See the splice-experiment repository for details, and the .github/analysis.yaml for how the automation works, and here for the actions.

Manual Symbols

If you want to do a run with manual libraries (e.g., from spack) it is a simple case of building a library with spack, and then comparing across versions. Spliced can do this internally with a spack splice, but instead we really want to do the same procedure as with two different libs and install the libs separately and provide the paths. Here is how to do that. Clone spack:

$ git clone git@github.com:spack/spack

On the host ensure debug flags are on and install across versions of a spack library.

$ export SPACK_DEBUG_FLAGS=true
$ spack compiler find

This develop branch doesn't have an option to list without the extra, so we do:

for package in binutils; do
    for version in $(spack versions --safe ${package}); do
       printf "Installing ${package}@${version}\n"
       spack install ${package}@${version}
    done
done

Now we can shell into the container (with the test_libs_manual.py in your PWD and test for the pattern. Since I wound up running this locally I used docker, but you could pull a Singularity container instead.

$ mkdir -p results/manual/binutils
$ docker run -it -v $PWD:/runs -v /home/vanessa/Desktop/Code/spack-vsoch:/spack ghcr.io/buildsi/splice-experiment:ubuntu-20.04 

Then run the script. I targeted the exact spack compiler directory I wanted to ensure staying in the same compiler version/space.

$ python /runs/test_libs_manual.py /spack/opt/spack/linux-ubuntu20.04-skylake/gcc-9.4.0 binutils --outdir /runs/results/manual/binutils

That will save results (json files) to the directory you created.