Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/sparselinearalgebra/graphblas-sharp

GPGPU-based GraphBLAS-like API implementation in F# (using Brahma.FSharp and OpenCL)
https://github.com/sparselinearalgebra/graphblas-sharp

dotnet fsharp gpu graph-algorithms graphblas linear-algebra matrix sparse-linear-algebra sparse-matrix

Last synced: 2 months ago
JSON representation

GPGPU-based GraphBLAS-like API implementation in F# (using Brahma.FSharp and OpenCL)

Host: GitHub
URL: https://github.com/sparselinearalgebra/graphblas-sharp
Owner: SparseLinearAlgebra
License: mit
Created: 2020-09-20T13:18:07.000Z (over 4 years ago)
Default Branch: master
Last Pushed: 2024-06-02T21:27:06.000Z (8 months ago)
Last Synced: 2024-11-21T02:08:36.674Z (2 months ago)
Topics: dotnet, fsharp, gpu, graph-algorithms, graphblas, linear-algebra, matrix, sparse-linear-algebra, sparse-matrix
Language: F#
Homepage: https://yaccconstructor.github.io/GraphBLAS-sharp/
Size: 16.6 MB
Stars: 4
Watchers: 14
Forks: 7
Open Issues: 1
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- License: LICENSE.md

Awesome Lists containing this project

README

        # GraphBLAS-sharp

[![FAKE Build](https://github.com/YaccConstructor/GraphBLAS-sharp/actions/workflows/build-on-push.yml/badge.svg)](https://github.com/YaccConstructor/GraphBLAS-sharp/actions/workflows/build-on-push.yml) 

[![NuGet Badge](https://buildstats.info/nuget/GraphBLAS-sharp)](https://www.nuget.org/packages/GraphBLAS-sharp/)

[![License](https://img.shields.io/badge/License-MIT-red.svg)](https://opensource.org/licenses/MIT)

GraphBLAS# is a GPGPU-based [GraphBLAS](https://graphblas.org/)-like API implementation in F#. To utilize GPGPUs we use [Brahma.FSharp](https://github.com/YaccConstructor/Brahma.FSharp). So, GraphBLAS# can utilize any OpenCL-compatible device.

### Features

- ```Option<'t>``` to solve [explicit/implicit zeroes problem](https://github.com/GraphBLAS/LAGraph/issues/28#issuecomment-542952115). If graph has labels of type ```'t``` then adjacency matrix is ```Matrix>```. Sparse storage contains only values for ```Some<'t>``` cells. 

- Elementwise operations have type ```AtLeastOne<'t1,'t2> -> Option<'t3>``` to be shure that ```None op None``` is ```None```. Also developer explicitly controls what should be ```None```. ```AtLeastOne``` defined as fallows:

  ```fsharp

  type AtLeastOne<'t1, 't2> =

     | Both of 't1*'t2

     | Left of 't1

     | Right of 't2

  ```

  So, type of matrix-matrix elementwise operation is ```Matrix> -> Matrix> -> (AtLeastOne<'t1,'t2> -> Option<'t3>) -> Matrix>```. 

- No semirings. Just functions. Of course one can implement semirings on the top of provided API.

- Minimal core: high-order functions allows us to minimaze core by functions unification. For example, such functions as matrix-matrix addition, matrix-matrix element-wise multiplication, masking all are partial case of `map2` function.

### Operations

- **Matrix-Matrix**

  - [x] CSR-CSR `map2`

  - [x] CSR-CSR `map2AtLeastOne`

  - [x] COO-COO `map2`

  - [x] COO-COO `map2AtLeastOne`

  - [x] CSR-CSR multiplication

  - [x] CSR-CSR Kronecker product

- **Vector-Matrix**

  - [x] Dense-CSR multiplication

  - [ ] Sparse-CSR multiplication

- **Vector-Vector**

  - [x] Dense-Dense `map2`

  - [x] Dense-Dense `map2AtLeastOne`

  - [x] Sparse-Sparse `map2`

  - [x] Sparse-Sparse `map2AtLeastOne`

  - [ ] ...

- **Matrix**

  - [x] `copy` 

  - [x] `map`

  - [x] COO transpose

  - [x] CSR transpose

  - [x] CSC transpose

  - [ ] ...

- **Vector**

  - [x] `zeroCreate` 

  - [x] `ofList`

  - [x] `copy`

  - [x] `reduce`

  - [ ] ...  

### Graph Analysis Algorithms

- [ ] BFS

- [ ] Parent BFS

- [ ] Single Source Shortest Path

- [ ] Triangles Counting

- [ ] Local Clustering Coefficient

- [ ] Community Detection using Label Propagation

- [ ] Weakly Connected Components

- [ ] PageRank

### Evaluation

Matrices from [SuiteSparse matrix collection](https://sparse.tamu.edu/) which we choose for evaluation.

  

    Matrix

    Size

    NNZ

    Squared matrix NNZ

  

  

    wing

    62 032

    243 088

    714 200

  

  

    luxembourg osm

    114 599

    119 666

    393 261

  

  

    amazon0312

    400 727

    3 200 440

    14 390 544

  

  

    amazon-2008

    735 323

    5 158 388

    25 366 745

  

  

    web-Google

    916 428

    5 105 039

    29 710 164

  

  

    webbase-1M

    1 000 005

    3 105 536

    51 111 996

  

  

    cit-Patents

    3 774 768

    16 518 948

    469

  

Element-wise matrix-matrix evaluation results presented below. Time is measured in milliseconds. We perform our experiments on the PC with Ubuntu 20.04 installed and with the following hardware configuration: Intel Core i7–4790 CPU, 3.60GHz, 32GB DDR4 RAM with GeForce GTX 2070, 8GB GDDR6, 1.41GHz. 

  

    Matrix

    Elemint-wise addition

    Elemint-wise multiplication

  

  

    GraphBLAS-sharp

    SuiteSparse

    CUSP

    GraphBLAS-sharp

    SuiteSparse

  

  

    No AtLeastOne

    AtLeastOne

  

  

    wing

    4,3 ± 0,8

    4,3 ± 0,6

    2,7 ± 0,9

    1,5 ± 0,0

    3,7 ± 0,5

    3,5 ± 0,4

  

  

    luxembourg osm

    4,9 ± 0,7

    4,1 ± 0,5

    3,0 ± 1,1

    1,2 ± 0,1

    3,8 ± 0,6

    3,0 ± 0,6

  

  

    amazon0312

    22,3 ± 1,3

    22,1 ± 1,3

    33,4 ± 0,8

    11,0 ± 1,4

    18,7 ± 0,9

    35,7 ± 1,4

  

  

    amazon-2008

    38,7 ± 0,8

    39,0 ± 1,0

    55,9 ± 1,0

    19,1 ± 1,4

    34,5 ± 1,0

    58,9 ± 1,9

  

  

    web-Google

    43,4 ± 0,8

    43,4 ± 1,1

    67,2 ± 7,5

    21,3 ± 1,3

    39,0 ± 1,2

    66,2 ± 0,4

  

  

    webbase-1M

    63,6 ± 1,1

    63,7 ± 1,3

    86,5 ± 2,0

    24,3 ± 1,3

    54,5 ± 0,7

    37,6 ± 5,6

  

  

    cit-Patents

    26,9 ± 0,7

    26,0 ± 0,7

    183,4 ± 5,4

    10,8 ± 0,6

    24,3 ± 0,7

    162,2 ± 1,7

  

## Contributing

Contributions, issues and feature requests are welcome.

Feel free to check [issues](https://github.com/YaccConstructor/GraphBLAS-sharp/issues) page if you want to contribute.

## Build

Make sure the following **requirements** are installed on your system:

- [dotnet SDK](https://dotnet.microsoft.com/en-us/download/dotnet/5.0) 5.0 or higher

- OpenCL-compatible device and respective OpenCL driver

To build and run all tests:

- on Windows

```cmd

build.cmd 

```

- on Linux/macOS

```shell

./build.sh 

```

To find more options look at [MiniScaffold](https://github.com/TheAngryByrd/MiniScaffold). We use it in our project.

### Release

The release process is automated: NuGet packages publishing process is triggered by tag pushing to any branch.

To release new version one should

1. [Add release notes to CHANGELOG](https://github.com/TheAngryByrd/MiniScaffold/blob/master/Content/Library/README.md#releasing)

2. Run ```./build.sh Release [version]``` (on local machine)

## License

This project licensed under MIT License. License text can be found in the [license file](https://github.com/YaccConstructor/GraphBLAS-sharp/blob/master/LICENSE.md).