SSpMM

Env

g++-7.5.0
cuda-11.8
sputnik

Install Sputnik

git clone https://github.com/google-research/sputnik.git
cd sputnik
cd third_party
git clone https://github.com/abseil/abseil-cpp.git
git clone https://github.com/google/benchmark.git
git clone https://github.com/google/googletest.git
cd ..
mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release -DBUILD_TEST=ON -DBUILD_BENCHMARK=ON -DCUDA_ARCHS="70;75;80;89"
make -j12
cd ../..

Get Dataset

wget https://storage.googleapis.com/sgk-sc2020/dlmc.tar.gz
tar -xvf dlmc.tar.gz
rm dlmc.tar.gz

Run Sparse Vector Compression (SVC) and Produce New Dataset

cd RowMerge
bash mkdir.sh
python row_merge.py
python row_merge_reorder.py
cp ./csv_data/* ../data/
cd ..

wait about 6hours IF you don't want to wait, you can download the data(dlmc-v8, dlmc-v16, dataset-v8, dataset-v16) from https://github.com/xuezy-mmi/dlmc.git(TBD)

SSpMM

cd SSpMM

Build and Run

You need to change the NVCC direction (line 1) in "Makefile_xxx". (Four Makfile-files: Makefile_volta, Makefile_turing, Makefile_ampere, Makefile_ada) We have 4 shell script files, aimed to run on different GPU architectures. xxx depends on your GPU architexture. xxx includes volta, turing, ampere, ada.

bash ./run_xxx.sh

After programming executing, some csv files will be generated in ./data/. And the shell scropt has cpoied them to ../plt/ for plot figures.

Baseline

cd vectorsparse

You need to edit "MakeFile_xxx", change the NVCC direction (line 1) and your Sputnik direction (line 2).

Build and Run

bash ./run_baseline_xxx.sh

Plot the Figures of Experimental Results

cd plt
bash ./plt_all.sh

After programming executing, all pdf files of figures will be generated in ./fig/.