leto 0.41.0

N-dimensional strided arrays, zero-copy views, and storage backends
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
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174

//! Validated convolution parameter vocabulary.

use super::error::{LetoError, Result};

/// Spatial regular-convolution parameters.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub struct ConvolutionParameters<const D: usize> {
    stride: [usize; D],
    padding: [usize; D],
    dilation: [usize; D],
}

impl<const D: usize> ConvolutionParameters<D> {
    /// Construct validated spatial convolution parameters.
    ///
    /// # Errors
    ///
    /// Returns [`LetoError::InvalidInput`] when a stride or dilation is zero.
    pub fn new(stride: [usize; D], padding: [usize; D], dilation: [usize; D]) -> Result<Self> {
        if stride.contains(&0) {
            return Err(LetoError::InvalidInput(
                "convolution stride must be nonzero".to_string(),
            ));
        }
        if dilation.contains(&0) {
            return Err(LetoError::InvalidInput(
                "convolution dilation must be nonzero".to_string(),
            ));
        }
        Ok(Self {
            stride,
            padding,
            dilation,
        })
    }

    /// Return the per-axis traversal stride.
    #[must_use]
    pub const fn stride(&self) -> &[usize; D] {
        &self.stride
    }

    /// Return the symmetric per-axis padding.
    #[must_use]
    pub const fn padding(&self) -> &[usize; D] {
        &self.padding
    }

    /// Return the per-axis kernel dilation.
    #[must_use]
    pub const fn dilation(&self) -> &[usize; D] {
        &self.dilation
    }
}

/// Spatial transposed-convolution parameters.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub struct TransposedConvolutionParameters<const D: usize> {
    stride: [usize; D],
    padding: [usize; D],
    output_padding: [usize; D],
    dilation: [usize; D],
}

impl<const D: usize> TransposedConvolutionParameters<D> {
    /// Construct validated spatial transposed-convolution parameters.
    ///
    /// `output_padding` changes only the derived output shape. It does not add
    /// values or padding during the scatter operation.
    ///
    /// # Errors
    ///
    /// Returns [`LetoError::InvalidInput`] when a stride or dilation is zero.
    pub fn new(
        stride: [usize; D],
        padding: [usize; D],
        output_padding: [usize; D],
        dilation: [usize; D],
    ) -> Result<Self> {
        if stride.contains(&0) {
            return Err(LetoError::InvalidInput(
                "transposed convolution stride must be nonzero".to_string(),
            ));
        }
        if dilation.contains(&0) {
            return Err(LetoError::InvalidInput(
                "transposed convolution dilation must be nonzero".to_string(),
            ));
        }
        Ok(Self {
            stride,
            padding,
            output_padding,
            dilation,
        })
    }

    /// Return the per-axis scatter stride.
    #[must_use]
    pub const fn stride(&self) -> &[usize; D] {
        &self.stride
    }

    /// Return the symmetric per-axis padding.
    #[must_use]
    pub const fn padding(&self) -> &[usize; D] {
        &self.padding
    }

    /// Return the per-axis shape-only output padding.
    #[must_use]
    pub const fn output_padding(&self) -> &[usize; D] {
        &self.output_padding
    }

    /// Return the per-axis kernel dilation.
    #[must_use]
    pub const fn dilation(&self) -> &[usize; D] {
        &self.dilation
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn rejects_zero_regular_stride_and_dilation() {
        assert_eq!(
            ConvolutionParameters::new([0], [0], [1]),
            Err(LetoError::InvalidInput(
                "convolution stride must be nonzero".to_string()
            ))
        );
        assert_eq!(
            ConvolutionParameters::new([1], [0], [0]),
            Err(LetoError::InvalidInput(
                "convolution dilation must be nonzero".to_string()
            ))
        );
    }

    #[test]
    fn rejects_zero_transposed_stride_and_dilation() {
        assert_eq!(
            TransposedConvolutionParameters::new([0], [0], [0], [1]),
            Err(LetoError::InvalidInput(
                "transposed convolution stride must be nonzero".to_string()
            ))
        );
        assert_eq!(
            TransposedConvolutionParameters::new([1], [0], [0], [0]),
            Err(LetoError::InvalidInput(
                "transposed convolution dilation must be nonzero".to_string()
            ))
        );
    }

    #[test]
    fn preserves_validated_parameter_values() {
        let regular = ConvolutionParameters::new([2, 3], [4, 5], [6, 7])
            .expect("valid regular convolution parameters");
        assert_eq!(regular.stride(), &[2, 3]);
        assert_eq!(regular.padding(), &[4, 5]);
        assert_eq!(regular.dilation(), &[6, 7]);

        let transposed = TransposedConvolutionParameters::new([2, 3], [4, 5], [1, 2], [6, 7])
            .expect("valid transposed convolution parameters");
        assert_eq!(transposed.stride(), &[2, 3]);
        assert_eq!(transposed.padding(), &[4, 5]);
        assert_eq!(transposed.output_padding(), &[1, 2]);
        assert_eq!(transposed.dilation(), &[6, 7]);
    }
}