https://github.com/chrisdalvit/gpu-matrix-transpose

Implementation and benchmarking of different matrix transpose with CUDA
https://github.com/chrisdalvit/gpu-matrix-transpose

c cpp cuda cuda-kernels cuda-programming gpu-acceleration gpu-computing gpu-programming matrix-transpose nvidia-gpu

Last synced: about 2 months ago
JSON representation

Implementation and benchmarking of different matrix transpose with CUDA

• Host: GitHub
• URL: https://github.com/chrisdalvit/gpu-matrix-transpose
• Owner: chrisdalvit
• Created: 2024-12-19T17:53:06.000Z (about 2 months ago)
• Default Branch: main
• Last Pushed: 2024-12-19T19:22:03.000Z (about 2 months ago)
• Last Synced: 2024-12-19T20:24:28.035Z (about 2 months ago)
• Topics: c, cpp, cuda, cuda-kernels, cuda-programming, gpu-acceleration, gpu-computing, gpu-programming, matrix-transpose, nvidia-gpu
• Language: C++
• Homepage: https://chrisdalvit.github.io/gpu-matrix-transpose
• Size: 193 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# GPU Matrix Transpose

This repository implements and benchmarks different matrix transpose algorithms for GPU's using CUDA. Definitely check out the [corresponding blog post](https://chrisdalvit.github.io/gpu-matrix-transpose)

## Repository structure

The ```data``` folder contains the original benchmark data from the tested architectures that was used in the experimental analyses. 

The ```lib``` folder contains C-functions used by all tested algorithms with the corresponding header file.

The ```src``` folder contains C files with the different algorithms for matrix transposition.

## Setup project

After cloning the repository you can run 

```

make

```

and the files in ```src``` are compiled, and the benchmark test is started and stored in the ```stats``` folder (that is created by the Makefile). __Waring: It can take a lot of time for the benchmarks to finish!__

If you only want to compile the files in ```src``` create a folder ```bin``` and run 

```

make compile_c

```

This should compile all C files in ```src``` and store them into the ```bin``` folder without starting the benchmarks. Compiled binaries follow the naming convention of ```-```

In order to run the experiments on the Marzola cluster run 

```

make marzola

```

The script compiles the source code and launches SLURM jobs on the cluster. Results are stored in the ```stats/``` folder.

## Validate implementations

The correctness of the provided implementations can be verified by running the compiled binaries in 'debug mode'. After compilation you can run 

```

./bin/  --debug

```

For example 

```

./bin/naive-0 2 --debug

```

Should output a randomly initialized matrix with dimension 2^2 and the corresponding transposed matrix. Additionaly the execution time and the effective bandwidth are displayed.