https://github.com/guoci/sycl-examples
Simple SYCL examples
https://github.com/guoci/sycl-examples
cuda-by-example gpu gpu-programming julia-set maxwell-boltzmann-distribution sycl
Last synced: about 2 months ago
JSON representation
Simple SYCL examples
- Host: GitHub
- URL: https://github.com/guoci/sycl-examples
- Owner: guoci
- License: mit
- Created: 2023-09-21T18:21:22.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-10-03T21:50:28.000Z (7 months ago)
- Last Synced: 2025-01-23T05:34:29.361Z (3 months ago)
- Topics: cuda-by-example, gpu, gpu-programming, julia-set, maxwell-boltzmann-distribution, sycl
- Language: C
- Homepage:
- Size: 22.2 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# SYCL examples
Some simple SYCL code examples.
## N-body Maxwell-Boltzmann distribution
Uses N-body simulation to
approximate [Maxwell-Boltzmann distribution](https://en.wikipedia.org/wiki/Maxwell%E2%80%93Boltzmann_distribution).
https://github.com/guoci/sycl-examples/assets/1260178/161a5c62-2759-4e99-92a2-502f1f6e88da
Notes
Another implementation written in PyTorch can be
found [here](https://github.com/lukepolson/youtube_channel/blob/main/Python%20GPU/multibody_boltzmann.ipynb).
I found an [issue](https://github.com/lukepolson/youtube_channel/issues/12) with that code. The total energy do not
remain constant over time due to multiple collisions per particle at an iteration.
For my SYCL implementation, I resolved that with allowing only one collision per particle at an iteration.
## Ripple
SYCL rewrite
of [CUDA code](https://github.com/CodedK/CUDA-by-Example-source-code-for-the-book-s-examples-/blob/master/chapter05/ripple.cu)
in the "CUDA By Example" book.
https://github.com/guoci/sycl-examples/assets/1260178/e31a7f02-8a2e-40f5-a549-9171d0b4b5c9
Notes
The update_host member function can be very useful when we have a loop where a buffer is used repeatedly and we need the host data to update at each iteration.
Since we cannot destroy a buffer in the loop, the only way of forcing the update is to use update_host.
## Julia set
SYCL rewrite
of [CUDA code](https://github.com/CodedK/CUDA-by-Example-source-code-for-the-book-s-examples-/blob/master/chapter04/julia_gpu.cu)
in the "CUDA By Example" book.
## Ray tracing
SYCL rewrite
of [CUDA code](https://github.com/CodedK/CUDA-by-Example-source-code-for-the-book-s-examples-/blob/master/chapter06/ray.cu)
in the "CUDA By Example" book.
## build and run
A SYCL compiler is required to compiler SYCL code. A normal C++ compiler would not work. See https://sycl.tech/ for more information.
This is built with the SYCL [compiler](https://software.intel.com/oneapi) from Intel.
`cd` into the directory of the example, then run the following in shell
```shell
cmake -DCMAKE_BUILD_TYPE=Release -S . -B ./cmake-build-release
cmake --build ./cmake-build-release --target -j
cmake-build-release/
```