Struct GpuArrayOps 

pub struct GpuArrayOps;

GPU array operations.

add_arrays/multiply_arrays/matrix_multiply dispatch to a real oxicuda-backed GPU (via sklears_core::gpu) when built with the gpu feature and device_id names an actual present device; otherwise (no gpu feature, or no such device on this host) they transparently fall back to the CPU implementation below – same pattern as sklears-svm’s DeviceNotAvailable skip. apply_activation/ reduce_sum/reduce_max remain CPU-only: there is no oxicuda-primitives elementwise/reduction kernel wired for them yet (deferred 2026-07-06, see TODO.md).

Implementations

impl GpuArrayOps

pub fn add_arrays( a: &[f32], b: &[f32], device_id: u32, ) -> Result<Vec<f32>, GpuError>

Add two arrays, on GPU if device_id names a present device and this crate was built with the gpu feature, on CPU otherwise.

pub fn multiply_arrays( a: &[f32], b: &[f32], device_id: u32, ) -> Result<Vec<f32>, GpuError>

Multiply two arrays elementwise, on GPU if device_id names a present device and this crate was built with the gpu feature, on CPU otherwise.

pub fn matrix_multiply( a: &[f32], b: &[f32], m: usize, n: usize, k: usize, device_id: u32, ) -> Result<Vec<f32>, GpuError>

Matrix multiplication (a is m x k, b is k x n, row-major), on GPU if device_id names a present device and this crate was built with the gpu feature, on CPU otherwise.

pub fn apply_activation( input: &[f32], activation: ActivationFunction, _device_id: u32, ) -> Result<Vec<f32>, GpuError>

Apply activation function.

CPU-only: no oxicuda elementwise kernel is wired for activation functions yet (deferred 2026-07-06, see TODO.md). device_id is accepted for API symmetry with the other ops but unused.

pub fn reduce_sum(input: &[f32], _device_id: u32) -> Result<f32, GpuError>

Compute the sum reduction.

CPU-only: no oxicuda reduction kernel is wired for this op yet (deferred 2026-07-06, see TODO.md). device_id is accepted for API symmetry with the other ops but unused.

pub fn reduce_max(input: &[f32], _device_id: u32) -> Result<f32, GpuError>

Compute the max reduction.

CPU-only: no oxicuda reduction kernel is wired for this op yet (deferred 2026-07-06, see TODO.md). device_id is accepted for API symmetry with the other ops but unused.