executorch 0.9.0

Rust bindings for ExecuTorch - On-device AI across mobile, embedded and edge for PyTorch
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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265

use std::ops::{Index, IndexMut};

use super::{SizesType, StridesType};

pub(crate) struct TensorAccessorInner<'a, T, const N: usize> {
    data: *const T,
    sizes: [SizesType; N],
    strides: [StridesType; N],
    phantom: std::marker::PhantomData<&'a ()>,
}

impl<'a, T, const N: usize> TensorAccessorInner<'a, T, N> {
    pub(crate) unsafe fn new(
        data: *const T,
        sizes: &'a [SizesType],
        strides: &'a [StridesType],
    ) -> Self {
        Self {
            data,
            sizes: sizes.try_into().unwrap(),
            strides: strides.try_into().unwrap(),
            phantom: std::marker::PhantomData,
        }
    }

    fn offset_of(&self, index: [usize; N]) -> Option<isize> {
        // Safety: we check the index validity before calling the unsafe function
        self.is_valid_index(index)
            .then(|| unsafe { self.offset_of_unchecked_impl(index) })
    }

    /// # Safety
    ///
    /// The caller must ensure that the index is within bounds.
    unsafe fn offset_of_unchecked(&self, index: [usize; N]) -> isize {
        debug_assert!(self.is_valid_index(index));
        // Safety: enforced by the caller
        unsafe { self.offset_of_unchecked_impl(index) }
    }

    /// # Safety
    ///
    /// The caller must ensure that the index is within bounds.
    unsafe fn offset_of_unchecked_impl(&self, index: [usize; N]) -> isize {
        let mut offset = 0isize;
        for (&idx, stride) in index.iter().zip(self.strides) {
            offset += idx as isize * stride as isize;
        }
        offset
    }

    fn is_valid_index(&self, index: [usize; N]) -> bool {
        index
            .iter()
            .zip(self.sizes)
            .all(|(&idx, size)| idx < size as usize)
    }
}

/// A fast accessor for a tensor.
///
/// The accessor is a utility struct, templated over the type of the tensor elements and the number
/// of dimensions, which make it very efficient to access tensor elements by index.
/// A regular Tensor stores its number of dimensions dynamically, and its typed can be stored
/// dynamically or known at compile time
/// (see [Tensor][`crate::tensor::Tensor`] and [TensorAny][`crate::tensor::TensorAny`]).
/// If you know the rank (number of dimensions) and the type of the tensor elements at compile time,
/// you can use this accessor to access the tensor elements efficiently.
pub struct TensorAccessor<'a, T, const N: usize>(pub(crate) TensorAccessorInner<'a, T, N>);
impl<'a, T, const N: usize> TensorAccessor<'a, T, N> {
    /// Get a reference to the tensor element at the given index.
    ///
    /// Returns the element at the given index, or `None` if the index is out of bounds.
    pub fn get(&self, index: [usize; N]) -> Option<&T> {
        let offset = self.0.offset_of(index)?;
        Some(unsafe { &*self.0.data.offset(offset) })
    }

    /// Get a reference to the tensor element at the given index, without bounds checking.
    ///
    /// # Safety
    ///
    /// The caller must ensure that the index is within bounds.
    pub unsafe fn get_unchecked(&self, index: [usize; N]) -> &T {
        let offset = unsafe { self.0.offset_of_unchecked(index) };
        unsafe { &*self.0.data.offset(offset) }
    }
}
impl<'a, T> Index<usize> for TensorAccessor<'a, T, 1> {
    type Output = T;
    #[inline]
    #[track_caller]
    fn index(&self, index: usize) -> &Self::Output {
        self.get([index]).unwrap()
    }
}
impl<'a, T, const N: usize> Index<[usize; N]> for TensorAccessor<'a, T, N> {
    type Output = T;
    #[inline]
    #[track_caller]
    fn index(&self, index: [usize; N]) -> &Self::Output {
        self.get(index).unwrap()
    }
}

/// A mutable accessor for a tensor.
///
/// The accessor is a utility struct, templated over the type of the tensor elements and the number
/// of dimensions, which make it very efficient to access tensor elements by index.
/// This is similar to [TensorAccessor], but allows for mutable access to the tensor elements.
/// See the immutable accessor for more details.
pub struct TensorAccessorMut<'a, T, const N: usize>(pub(crate) TensorAccessorInner<'a, T, N>);
impl<'a, T, const N: usize> TensorAccessorMut<'a, T, N> {
    /// Get a reference to the tensor element at the given index.
    ///
    /// Returns the element at the given index, or `None` if the index is out of bounds.
    pub fn get(&self, index: [usize; N]) -> Option<&T> {
        let offset = self.0.offset_of(index)?;
        Some(unsafe { &*self.0.data.offset(offset) })
    }

    /// Get a reference to the tensor element at the given index, without bounds checking.
    ///
    /// # Safety
    ///
    /// The caller must ensure that the index is within bounds.
    pub unsafe fn get_unchecked(&self, index: [usize; N]) -> &T {
        let offset = unsafe { self.0.offset_of_unchecked(index) };
        unsafe { &*self.0.data.offset(offset) }
    }

    /// Get a mutable reference to the tensor element at the given index.
    ///
    /// Returns the element at the given index, or `None` if the index is out of bounds.
    pub fn get_mut(&mut self, index: [usize; N]) -> Option<&mut T> {
        let offset = self.0.offset_of(index)?;
        Some(unsafe { &mut *self.0.data.cast_mut().offset(offset) })
    }

    /// Get a mutable reference to the tensor element at the given index, without bounds checking.
    ///
    /// # Safety
    ///
    /// The caller must ensure that the index is within bounds.
    pub unsafe fn get_mut_unchecked(&mut self, index: [usize; N]) -> &mut T {
        let offset = unsafe { self.0.offset_of_unchecked(index) };
        unsafe { &mut *self.0.data.cast_mut().offset(offset) }
    }
}
impl<'a, T> Index<usize> for TensorAccessorMut<'a, T, 1> {
    type Output = T;
    #[inline]
    #[track_caller]
    fn index(&self, index: usize) -> &Self::Output {
        self.get([index]).unwrap()
    }
}
impl<'a, T, const N: usize> Index<[usize; N]> for TensorAccessorMut<'a, T, N> {
    type Output = T;
    #[inline]
    #[track_caller]
    fn index(&self, index: [usize; N]) -> &Self::Output {
        self.get(index).unwrap()
    }
}
impl<'a, T> IndexMut<usize> for TensorAccessorMut<'a, T, 1> {
    #[inline]
    #[track_caller]
    fn index_mut(&mut self, index: usize) -> &mut Self::Output {
        self.get_mut([index]).unwrap()
    }
}
impl<'a, T, const N: usize> IndexMut<[usize; N]> for TensorAccessorMut<'a, T, N> {
    #[inline]
    #[track_caller]
    fn index_mut(&mut self, index: [usize; N]) -> &mut Self::Output {
        self.get_mut(index).unwrap()
    }
}

#[cfg(test)]
mod tests {
    #[cfg(feature = "alloc")]
    use crate::tensor::{RawTensor, RawTensorImpl};

    #[cfg(feature = "alloc")]
    #[test]
    fn get() {
        let sizes = [2, 3];
        let data: [i32; 6] = [1, 2, 3, 4, 5, 6];
        let dim_order = [0, 1];
        let strides = [3, 1];
        let tensor_impl = unsafe {
            RawTensorImpl::from_ptr(&sizes, data.as_ptr().cast_mut(), &dim_order, &strides).unwrap()
        };
        let tensor = unsafe { RawTensor::new(&tensor_impl) };

        assert!(tensor.accessor::<i8, 2>().is_none());
        assert!(tensor.accessor::<i16, 2>().is_none());
        assert!(tensor.accessor::<i64, 2>().is_none());
        assert!(tensor.accessor::<f32, 2>().is_none());
        assert!(tensor.accessor::<f64, 2>().is_none());
        assert!(tensor.accessor::<i32, 1>().is_none());
        assert!(tensor.accessor::<i32, 3>().is_none());
        assert!(tensor.accessor::<i32, 4>().is_none());
        let accessor = tensor.accessor::<i32, 2>().unwrap();
        assert_eq!(accessor.get([0, 0]), Some(&1));
        assert_eq!(accessor.get([0, 1]), Some(&2));
        assert_eq!(accessor.get([0, 2]), Some(&3));
        assert_eq!(accessor.get([0, 3]), None);
        assert_eq!(accessor.get([1, 0]), Some(&4));
        assert_eq!(accessor.get([1, 1]), Some(&5));
        assert_eq!(accessor.get([1, 2]), Some(&6));
        assert_eq!(accessor.get([2, 0]), None);
        assert_eq!(accessor[[0, 0]], 1);
        assert_eq!(accessor[[0, 1]], 2);
        assert_eq!(accessor[[0, 2]], 3);
        assert_eq!(accessor[[1, 0]], 4);
        assert_eq!(accessor[[1, 1]], 5);
        assert_eq!(accessor[[1, 2]], 6);
    }

    #[cfg(feature = "alloc")]
    #[test]
    fn get_mut() {
        let sizes = [2, 3];
        let mut data: [i32; 6] = [1, 2, 3, 4, 5, 6];
        let dim_order = [0, 1];
        let strides = [3, 1];
        let mut tensor_impl = unsafe {
            RawTensorImpl::from_ptr(&sizes, data.as_mut_ptr(), &dim_order, &strides).unwrap()
        };
        #[allow(clippy::unnecessary_mut_passed)]
        let mut tensor = unsafe { RawTensor::new(&mut tensor_impl) };

        unsafe {
            assert!(tensor.accessor_mut::<i8, 2>().is_none());
            assert!(tensor.accessor_mut::<i16, 2>().is_none());
            assert!(tensor.accessor_mut::<i64, 2>().is_none());
            assert!(tensor.accessor_mut::<f32, 2>().is_none());
            assert!(tensor.accessor_mut::<f64, 2>().is_none());
            assert!(tensor.accessor_mut::<i32, 1>().is_none());
            assert!(tensor.accessor_mut::<i32, 3>().is_none());
            assert!(tensor.accessor_mut::<i32, 4>().is_none());
            let mut accessor = tensor.accessor_mut::<i32, 2>().unwrap();
            assert_eq!(accessor.get([0, 0]), Some(&1));
            assert_eq!(accessor.get([0, 1]), Some(&2));
            assert_eq!(accessor.get([0, 2]), Some(&3));
            assert_eq!(accessor.get([0, 3]), None);
            assert_eq!(accessor.get([1, 0]), Some(&4));
            assert_eq!(accessor.get([1, 1]), Some(&5));
            assert_eq!(accessor.get([1, 2]), Some(&6));
            assert_eq!(accessor.get([2, 0]), None);

            *accessor.get_mut([0, 0]).unwrap() = 10;
            accessor[[1, 1]] = 25;
            assert_eq!(accessor[[0, 0]], 10);
            assert_eq!(accessor[[0, 1]], 2);
            assert_eq!(accessor[[0, 2]], 3);
            assert_eq!(accessor[[1, 0]], 4);
            assert_eq!(accessor[[1, 1]], 25);
            assert_eq!(accessor[[1, 2]], 6);
        }
    }
}