Another Linnear Algebra Benchmark is a benchmark closely inspired by the HPCCG benchmark. This project is on early development and there will be a roadmap bellow for future work.
To install this program you need to make sure your compiler supports OpenMP. If it does support it proceed with the following commands:
git clone git@github.com:TolisSth/ALAB.git
cd ALAB
make clean
make -j
In general the --help argument explains everything but because this is a README.md file let's plagiarize.
Usage: alab [options]
Options:
--size, -s <size> Set the size (default: 1000)
--threads, -t <num> Set the number of threads (default: 2)
--seed, -sd <seed> Set the random seed (default: 42)
--help, -h Display this help message
--serial Run the serial version of the benchmark (default)
--parallel Run the parallel version of the benchmark (please specify number of thread or else it will run in two threads)
This is the --help message and as it is evident you don't need to do much to run the program because it will handle the configuration by itself running the equal: ./alab --size 1000 --threads 2 --serial --seed 42, which is a bit misleading since --threads are only used when the kernel is running at parallel but oh well, more thing to add to the roadmap.
A good starting configuration to play with the parallel kernel is:
./alab --size 100000 --threads 8 --parallel.
Right now we are measuring MFLOPS GFLOPS and time in seconds.
The aforementioned roadmap is finally here! The whole roadmap is splitted in two categories, the infrastructure part and the kernels part namely:
- C++ Multitreading through OpenMP.
- MPI or Chapel support.
- A way for the user to extract the data in a data analysis ready format.
- DDOT
- Gauss Elimination
- Sparse Matrix-Vector Multiplication
Hey! I don't bite and I am looking for people to work with me on this project so feel free to contact me or just fork the project and start writing your first pull request.