Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/elftausend/custos
A minimal OpenCL, CUDA, Vulkan and host CPU array manipulation engine / framework.
https://github.com/elftausend/custos
array-manipulations autograd automatic-differentiation cpu cuda cuda-support custos framework gpu lazy-evaluation no-std opencl rust vulkan wgsl
Last synced: 4 days ago
JSON representation
A minimal OpenCL, CUDA, Vulkan and host CPU array manipulation engine / framework.
- Host: GitHub
- URL: https://github.com/elftausend/custos
- Owner: elftausend
- License: mit
- Created: 2022-03-08T08:32:39.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2024-04-17T05:32:37.000Z (9 months ago)
- Last Synced: 2024-04-17T17:10:10.101Z (9 months ago)
- Topics: array-manipulations, autograd, automatic-differentiation, cpu, cuda, cuda-support, custos, framework, gpu, lazy-evaluation, no-std, opencl, rust, vulkan, wgsl
- Language: Rust
- Homepage:
- Size: 2.58 MB
- Stars: 60
- Watchers: 2
- Forks: 7
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome-cuda-triton-hpc - custos
- awesome-cuda-triton-hpc - custos
- awesome-rust-list - custos
- awesome-rust-list - custos
README
[](https://crates.io/crates/custos)
[](https://docs.rs/custos/0.7.0/custos/)
[](https://github.com/elftausend/custos/actions/workflows/rust.yml)
[](https://github.com/elftausend/custos/actions/workflows/gpu.yml)
[](https://github.com/elftausend/custos/actions/workflows/rust-clippy.yml)
[](https://github.com/elftausend/custos/actions/workflows/android.yml)
A minimal, extensible OpenCL, Vulkan (with WGSL), CUDA, NNAPI (Android) and host CPU array manipulation engine / framework written in Rust.
This crate provides tools for executing custom array and automatic differentiation operations.
## Installation
The latest published version is of `0.7.x` (April 14th, 2023). A lot has changed since then. `0.7.x` can be found in the `custos-0.7` branch.
Add "custos" as a dependency:
```toml
[dependencies]
custos = "0.7.0"
# to disable the default features (cpu, cuda, opencl, static-api, blas, macro) and use an own set of features:
#custos = {version = "0.7.0", default-features=false, features=["opencl", "blas"]}
```
### Available features:
To make specific devices useable, activate the corresponding features:
Feature | Device | Notes
--- | --- | ---
cpu | `CPU` | Uses heap allocations.
stack | `Stack` | Useable in `no-std` environments as it uses stack allocated `Buffer`s without requiring `alloc` or `std`. Practically only supports the `Base` module.
opencl | `OpenCL` | Automatically maps unified memory.
cuda | `CUDA` |
vulkan | `Vulkan` | Shaders are written in WGSL. + unified memory
nnapi | `NnapiDevice` | `Lazy` module is mandatory.
untyped | `Untyped` | Removes the need of `Buffer`'s generic parameters. (CPU and CUDA only for now)
custos ships combineable modules. Different selected modules result in different behaviour when executing operations.
New modules can be added in user code.
```rust
use custos::prelude::*;
// Autograd, Base = Modules
let device = CPU::>::new();
```
To make specific modules useable for building a device, activate the corresponding features:
Feature | Module | Description
--- | --- | ---
*on by default* | `Base` | Default behaviour.
autograd | `Autograd` | Enables running automatic differentiation.
cached | `Cached` | Reuses allocations on demand.
fork | `Fork` | Decides whether the CPU or GPU is faster for an operation. It then uses the faster device for following computations. (unified memory devices)
lazy | `Lazy` | Lazy execution of operations and lazy intermediate allocations. Enables support for CUDA graphs.
graph | `Graph` | Adds a memory usage optimizeable graph and fusing of unary operations in combination with `Lazy`.
Usage of these modules when writing custom operations: [`modules.md`](modules.md) and [`modules_usage.rs`](examples/modules_usage.rs).
If an operations wants to be affected by a module, specific custos code must be called in that operation.
Remaining features:
Feature | Description
--- | ---
static-api | Enables the creation of `Buffer`s without providing a device.
std | Adds standard library support.
no-std | For no std environments, activates `stack` feature.
macro | Reexport of [custos-macro]
blas | Adds gemm functions of the system's (selected) BLAS library.
half | Adds support for half precision floats.
serde | Adds serialization and deserialization support.
json | Adds convenience functions for serialization and deserialization to and from json.
[custos-macro]: https://github.com/elftausend/custos-macro
## [Examples]
[examples]: https://github.com/elftausend/custos/tree/main/examples
[unary]: https://github.com/elftausend/custos/blob/main/src/unary.rs
Implement an operation for `CPU`:
- If you want to implement your own operations for all compute devices, consider looking here: [implement_operations.rs](examples/implement_operations.rs) or ["modules_usage.rs"](examples/modules_usage.rs)
or to see it at a larger scale, look here [`custos-math`](https://github.com/elftausend/custos-math) (outdated, requires custos 0.7) or here [`sliced`](https://github.com/elftausend/sliced) (for automatic diff examples).
This operation is only affected by the `Cached` module (and partially `Autograd`).
```rust
use custos::prelude::*;
use std::ops::{Deref, Mul};
pub trait MulBuf: Sized + Device {
fn mul(&self, lhs: &Buffer, rhs: &Buffer) -> Buffer;
}
impl MulBuf for CPU
where
Mods: Retrieve,
T: Unit + Mul + Copy + 'static,
S: Shape,
D: Device,
D::Base: Deref,
{
fn mul(&self, lhs: &Buffer, rhs: &Buffer) -> Buffer {
let mut out = self.retrieve(lhs.len(), (lhs, rhs)).unwrap(); // unwrap or return error (update trait)
for ((lhs, rhs), out) in lhs.iter().zip(rhs.iter()).zip(&mut out) {
*out = *lhs * *rhs;
}
out
}
}
```
A lot more usage examples can be found in the [tests] and [examples] folders.
(Or in the [unary] operation file, [custos-math](https://github.com/elftausend/custos-math) and [`sliced`](https://github.com/elftausend/sliced))
[tests]: https://github.com/elftausend/custos/tree/main/tests