redstone-ml 0.0.0

High-performance Machine Learning, Auto-Differentiation and Tensor Algebra crate for Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65

use crate::dtype::RawDataType;
use crate::iterator::collapse_contiguous::collapse_to_uniform_stride;
use crate::iterator::flat_index_generator::FlatIndexGenerator;
use crate::{Constructors, NdArray, StridedMemory};
use std::ptr::copy_nonoverlapping;

impl<'a, T: RawDataType> NdArray<'a, T> {
    #[allow(clippy::should_implement_trait)]
    pub fn clone<'r>(&self) -> NdArray<'r, T> {
        unsafe { NdArray::from_contiguous_owned_buffer(self.shape.clone(), self.clone_data()) }
    }

    pub(super) fn clone_data(&self) -> Vec<T> {
        if self.is_contiguous() {
            return unsafe { self.clone_data_contiguous() };
        }
        unsafe { self.clone_data_non_contiguous() }
    }

    /// Safety: expects ndarray buffer is contiguously stored
    unsafe fn clone_data_contiguous(&self) -> Vec<T> {
        let mut data = Vec::with_capacity(self.len);

        copy_nonoverlapping(self.ptr(), data.as_mut_ptr(), self.len);
        data.set_len(self.len);
        data
    }

    /// Safety: expects ndarray buffer is not contiguously stored
    unsafe fn clone_data_non_contiguous(&self) -> Vec<T> {
        let size = self.size();
        let mut data = Vec::with_capacity(size);

        let (mut shape, mut stride) = collapse_to_uniform_stride(&self.shape, &self.stride);

        // safe to unwrap because if stride has no elements, this would be a scalar ndarray
        // however, scalar arrays are contiguously stored so this method wouldn't be called
        let &mut mut contiguous_stride = stride.last_mut().unwrap();

        // if elements along the last axis are located contiguously,
        // we can collapse the last dimension and copy contiguous_stride elements at once
        if contiguous_stride == 1 {
            contiguous_stride = shape.pop().unwrap();
            stride.pop();
        }

        // if elements along the last axis aren't located contiguously,
        // they must correspond to an NdArray view with a step-size along the last axis of > 1
        // this is equivalent to 1 contiguous element along the last axis
        else {
            contiguous_stride = 1;
        }

        let src = self.ptr();
        let mut dst = data.as_mut_ptr();

        for i in FlatIndexGenerator::from(&shape, &stride) {
            copy_nonoverlapping(src.add(i), dst, contiguous_stride);
            dst = dst.add(contiguous_stride);
        }

        data.set_len(size);
        data
    }
}