This repository contains the artifacts necessarty to reproduce results for the paper:

FaSTCC: Fast Sparse Tensor Contractions on CPUs

Get the code

Clone this repository recursively so that all the dependencies are cloned correctly at the right location. git clone --recursive [email protected]:HPCRL/sparse_benchmark.git

Get the data

Download the following tensors from FROSTT: http://frostt.io/tensors/

• NIPS
• Chicago-crime-comm
• vast-2015-mc1-5d.tns.gz
• Uber

The chemistry tensors will be packaged with the repository.

Build the source

mkdir build
cd build
cmake ..
make -j16

Run the experiments

Inside the build folder after running make:

./sc_ae_speedups <<path to frostt tensors>> <<path to caffeine tensors>> <<path to guanine tensors>>

This creates two CSV files in the build folder chemistry_times.csv frostt_times.csv with the wall-time for FaSTCC kernel running with the data given.