Trait TensorSortPrims

Source

pub trait TensorSortPrims<Alg: Algebra>: Sizedwhere
    Alg::Scalar: PartialOrd,{
    type Plan;
    type Context;

    // Required methods
    fn plan(
        ctx: &mut Self::Context,
        desc: &SortPrimsDescriptor,
        shapes: &[&[usize]],
    ) -> Result<Self::Plan>;
    fn execute(
        ctx: &mut Self::Context,
        plan: &Self::Plan,
        input: &Tensor<Alg::Scalar>,
        values_out: &mut Tensor<Alg::Scalar>,
        indices_out: &mut Tensor<i64>,
    ) -> Result<()>;
    fn has_sort_support(desc: &SortPrimsDescriptor) -> bool;
}

Expand description

Sort execution protocol family.

Provides sorting, argsort, and top-k operations on tensors. The indices output tensor uses i64 elements representing positions along the target axis (same convention as crate::TensorIndexingPrims).

Unlike semiring and scalar families, sort does not use alpha/beta scaling – operations are pure data rearrangement.

§Examples

use tenferro_algebra::Standard;
use tenferro_prims::{CpuBackend, CpuContext, SortPrimsDescriptor, TensorSortPrims};

let mut ctx = CpuContext::new(1);
let desc = SortPrimsDescriptor::Sort {
    axis: 0,
    descending: false,
    stable: true,
};
let _plan = <CpuBackend as TensorSortPrims<Standard<f64>>>::plan(
    &mut ctx,
    &desc,
    &[&[5]],
)
.unwrap();

Required Associated Types§

Source

type Plan

Backend-specific execution plan.

Source

type Context

Backend execution context.

Required Methods§

Source

fn plan( ctx: &mut Self::Context, desc: &SortPrimsDescriptor, shapes: &[&[usize]], ) -> Result<Self::Plan>

Plan a sort operation for the given input shape.

shapes contains: [input_shape].

Source

fn execute( ctx: &mut Self::Context, plan: &Self::Plan, input: &Tensor<Alg::Scalar>, values_out: &mut Tensor<Alg::Scalar>, indices_out: &mut Tensor<i64>, ) -> Result<()>

Execute a previously planned sort operation.

For Sort: writes sorted values to values_out and permutation indices to indices_out.
For Argsort: writes permutation indices to indices_out; values_out is left unmodified.
For Topk: writes top-k values to values_out and their original indices to indices_out.