strided_map_simd

fn strided_map_simd(self, f: F, vec_f: F2) -> ParStridedMap

Applies two functions to process data: one for scalar operations and one for vectorized (SIMD) operations, enabling faster processing through hardware acceleration.

Notes

• Automatically falls back to scalar function when SIMD is not possible

Parameters

• self: The parallel SIMD iterator
  • Type: ParStridedSimd<T>
• f: The scalar mapping function
  • Type: FnMut((&mut T, T))
  • Used for non-vectorized elements
• vec_f: The SIMD mapping function
  • Type: FnMut((&mut Simd<T>, Simd<T>))
  • Used for vectorized operations

Returns

A ParStridedMapSimd iterator

Examples:

use hpt::Tensor;
use hpt::iter::TensorIterator;
use hpt::types::math::FloatOutUnary;

fn main() -> anyhow::Result<()> {
    let x = Tensor::<f64>::new(&[1f64, 2., 3.]);

    let res = x.par_iter_simd().strided_map_simd(
        |(res, x)| {
            *res = x.sin();
        },
        |(res, x)| {
            res.write_unaligned(x._sin());
        },
    ).collect::<Tensor<f64>>();

    println!("{}", res);
    Ok(())
}

Backend Support