10 releases (4 breaking)
0.6.3 | Feb 11, 2023 |
---|---|
0.6.2 |
|
0.5.0 | Sep 10, 2022 |
0.4.6 | Aug 16, 2022 |
0.1.4 | Jul 26, 2022 |
#630 in Machine learning
Used in gradients
145KB
3.5K
SLoC
custos-math
This crate provides CUDA, OpenCL, CPU (and Stack) based matrix operations using custos.
Installation
Add "custos-math" as a dependency: You will also need custos, if you want to run an example.
[dependencies]
custos-math = "0.6.3"
# to disable the default features (cuda, opencl) and use an own set of features:
#custos-math = { version="0.6.3", default-features = false, features=["opencl"]}
custos-math
supports no-std via the no-std
feature. This activates the "stack" feature, providing a Stack
device.
custos is accessible via custos_math::custos::{..}
Example
use custos_math::{Matrix, custos::CPU};
fn main() {
let device = CPU::new();
let a = Matrix::from((&device, (2, 3), [1., 2., 3., 4., 5., 6.,]));
let b = Matrix::from((&device, (3, 2), [6., 5., 4., 3., 2., 1.,]));
let c = a.gemm(&b);
assert_eq!(c.read(), vec![20., 14., 56., 41.,]);
}
Many more examples can be found in the tests and examples folder.
Dependencies
~0.6–1.1MB
~26K SLoC