Struct CpuBackend 

pub struct CpuBackend;

CPU backend using strided-kernel and GEMM.

Dispatched automatically when tensors reside on LogicalMemorySpace::MainMemory. Implements TensorPrims<Standard> for standard arithmetic.

Examples

use tenferro_prims::{CpuBackend, TensorPrims, PrimDescriptor};
use strided_view::StridedArray;

// Transpose a matrix
let desc = PrimDescriptor::Permute {
    modes_a: vec![0, 1],
    modes_b: vec![1, 0],
};
let plan = CpuBackend::plan::<f64>(&desc, &[&[3, 4], &[4, 3]]).unwrap();
let a = StridedArray::<f64>::col_major(&[3, 4]);
let mut b = StridedArray::<f64>::col_major(&[4, 3]);
CpuBackend::execute(&plan, 1.0, &[&a.view()], 0.0, &mut b.view_mut()).unwrap();

Trait Implementations

impl TensorPrims<Standard> for CpuBackend

type Plan<T: ScalarBase> = CpuPlan<T>

Backend-specific plan type (no type erasure).

fn plan<T: ScalarBase>( _desc: &PrimDescriptor, _shapes: &[&[usize]], ) -> Result<CpuPlan<T>>

Create an execution plan from an operation descriptor.

fn execute<T: ScalarBase>( _plan: &CpuPlan<T>, _alpha: T, _inputs: &[&StridedView<'_, T>], _beta: T, _output: &mut StridedViewMut<'_, T>, ) -> Result<()>

Execute a plan with the given scaling factors and tensor views.

fn has_extension_for<T: ScalarBase>(_ext: Extension) -> bool

Query whether an extended operation is available for scalar type T.