Some useful files and scripts for the fast parallel tree decomposition projects.
I generated several variations of the graphs for different circuits of interest as it may be interesting to compare these in a couple of weeks. Initially, however we can just stick to graphs in circuit_graphs/qflex_line_graph_files_decomposed_true_hyper_true.
Each graph comes from a different circuit. The graphs named bristlecone and sycamore come from circuits ran on google's bristlecone and sycamore quantum processors for example. These can get quite big, so maybe the smaller versions are best to start testing with: test.gr, rectangular_2x2_1-16-1_0.gr, rectangular_4x4_1-16-1_0.gr, sycamore_53_4_0.gr, bristlecone_48_1-16-1_0.gr.
The first line in the .gr files contain the number of vertices and edges in the graph. Each subsequent line has two integers (source and destination vertices) describing an edge of the graph
You can use the code from the script plot_graph_from_gr_file.jl to read the .gr files and create a graph.
From your home directory, download julia 1.6.2 and extract from the tar file
wget https://julialang-s3.julialang.org/bin/linux/x64/1.6/julia-1.6.2-linux-x86_64.tar.gz
tar -xf julia-1.6.2-linux-x86_64.tar.gz
Then create a symbolic link to the extracted binary
ln -s ./julia-1.6.2/bin/julia .local/bin/julia
Run interactive job on kay
srun -p DevQ -N 1 -A iccom013c -t 0:20:00 --pty bash
First load a module with mpi.
module load openmpi/intel/3.1.2
Then start julia and add the following environment variable. This will tell MPI.jl to use the system mpi and not install its own.
ENV["JULIA_MPI_BINARY"]="system"
Then in the julia package manager add MPI.jl
]add MPI
You can build the package with
]build MPI