tract-data 0.23.0-dev.4

Tiny, no-nonsense, self contained, TensorFlow and ONNX inference
Documentation
use crate::prelude::*;
use ndarray::Dimension;

pub(crate) unsafe fn scatter_contig_data<T: Datum>(
    mut src: *const T,
    dst: *mut T,
    dst_len_and_strides: &[(usize, usize)],
) {
    unsafe {
        match *dst_len_and_strides {
            [(len_a, stride_a)] => {
                for a in 0..len_a {
                    *dst.add(a * stride_a) = (*src).clone();
                    src = src.offset(1);
                }
            }
            [(len_a, stride_a), (len_b, stride_b)] => {
                for a in 0..len_a {
                    for b in 0..len_b {
                        *dst.add(a * stride_a + b * stride_b) = (*src).clone();
                        src = src.offset(1);
                    }
                }
            }
            [(len_a, stride_a), (len_b, stride_b), (len_c, stride_c)] => {
                for a in 0..len_a {
                    for b in 0..len_b {
                        for c in 0..len_c {
                            *dst.add(a * stride_a + b * stride_b + c * stride_c) = (*src).clone();
                            src = src.offset(1);
                        }
                    }
                }
            }
            [(len_a, stride_a), (len_b, stride_b), (len_c, stride_c), (len_d, stride_d)] => {
                for a in 0..len_a {
                    for b in 0..len_b {
                        for c in 0..len_c {
                            for d in 0..len_d {
                                *dst.add(
                                    a * stride_a + b * stride_b + c * stride_c + d * stride_d,
                                ) = (*src).clone();
                                src = src.offset(1);
                            }
                        }
                    }
                }
            }
            [
                (len_a, stride_a),
                (len_b, stride_b),
                (len_c, stride_c),
                (len_d, stride_d),
                (len_e, stride_e),
            ] => {
                for a in 0..len_a {
                    for b in 0..len_b {
                        for c in 0..len_c {
                            for d in 0..len_d {
                                for e in 0..len_e {
                                    *dst.add(
                                        a * stride_a
                                            + b * stride_b
                                            + c * stride_c
                                            + d * stride_d
                                            + e * stride_e,
                                    ) = (*src).clone();
                                    src = src.offset(1);
                                }
                            }
                        }
                    }
                }
            }
            _ => {
                let shape: TVec<usize> = dst_len_and_strides.iter().map(|pair| pair.0).collect();
                for coords in ndarray::indices(&*shape) {
                    let offset = coords
                        .slice()
                        .iter()
                        .zip(dst_len_and_strides.iter())
                        .map(|(x, (_len, stride))| x * stride)
                        .sum::<usize>();
                    *dst.add(offset) = (*src).clone();
                    src = src.offset(1);
                }
            }
        }
    }
}