par_iter_simd

fn par_iter_simd(x: &Tensor<T>) -> ParStridedSimd<T>

Convert Tensor to ParStridedSimd iterator which will use simd register to do computation when it is possible, ParStridedSimd will split the tasks and execute the method the user provides

Parameters:

x: Tensor to iterate

Returns:

ParStridedSimd

Note

Simd closure will only be execute when the result simd type number of lane is equal to input's.
In Simd closure, result can only call write_unaligned method to write data.

Examples:

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

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

Backend	Supported
CPU	✅
Cuda	❌