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
|
#![allow(dead_code)]
pub type Shape<const D: usize> = [usize; D];
pub type Index<const D: usize> = Shape<D>;
pub struct Tensor<T, const D: usize> {
data: Vec<T>,
shape: Shape<D>,
strides: Shape<D>
}
pub type Tensor1<T> = Tensor<T, 1>;
pub type Tensor2<T> = Tensor<T, 2>;
pub type Tensor3<T> = Tensor<T, 3>;
pub type Tensor4<T> = Tensor<T, 4>;
pub type Index1 = Index<1>;
pub type Index2 = Index<2>;
pub type Index3 = Index<3>;
pub type Index4 = Index<4>;
impl<T: Copy, const D: usize> Tensor<T, D> {
pub fn new(shape: Shape<D>, x: T) -> Self {
let dim = D;
if dim == 0 { panic!("Empty shape not allowed."); }
let mut len = shape[dim-1];
let mut strides: Shape<D> = [0; D];
strides[dim-1] = 1;
for d in (1..dim).rev() { // d=dim-1, …, 1.
strides[d-1] = shape[d]*strides[d];
len *= shape[d-1];
}
if len == 0 { panic!("Empty dimensions not allowed."); }
Self {
data: vec![x; len],
shape: shape,
strides: strides,
}
}
pub fn new_from(shape: Shape<D>, x: Vec<T>) -> Self {
let dim = D;
if dim == 0 { panic!("Empty shape not allowed."); }
let mut len = shape[dim-1];
let mut strides: Shape<D> = [0; D];
strides[dim-1] = 1;
for d in (1..dim).rev() { // d=dim-1, …, 1.
strides[d-1] = shape[d]*strides[d];
len *= shape[d-1];
}
if len == 0 { panic!("Empty dimensions not allowed."); }
if len != x.len() { panic!("Vector of length {} cannot fill tensor with {} entries.", x.len(), len); }
Self {
data: x,
shape: shape,
strides: strides,
}
}
#[inline(always)]
fn flatten_idx(self: & Self, idx: & Index<D>) -> usize {
// NOTE: This is a very hot code path. Should benchmark versus explicit loop.
idx.iter().zip(self.strides.iter()).fold(0, |sum, (i, s)| sum + i*s)
}
fn bound_check_panic(self: & Self, idx: & Index<D>) -> () {
for d in 0..self.dim() {
let i = *(unsafe { idx.get_unchecked(d) });
if i >= self.shape[d] {
panic!("{}-dimensional tensor index is out of bounds in dimension {} ({} >= {}).", self.dim(), d, i, self.shape[d])
}
}
}
pub fn in_bounds(self: & Self, idx: & Index<D>) -> bool {
for d in 0..self.dim() {
let i = *(unsafe { idx.get_unchecked(d) });
if i >= self.shape[d] {
return false;
}
}
true
}
pub fn dim(self: & Self) -> usize { D }
pub fn shape(self: & Self) -> & Shape<D> { &self.shape }
pub fn el(self: & Self, idx: & Index<D>) -> Option<& T> {
if self.in_bounds(idx) { Some(unsafe { self.el_unchecked(idx) }) }
else { None }
}
pub unsafe fn el_unchecked(self: & Self, idx: & Index<D>) -> & T { self.data.get_unchecked(self.flatten_idx(idx)) }
pub unsafe fn el_unchecked_mut(self: &mut Self, idx: & Index<D>) -> &mut T {
let flat_idx = self.flatten_idx(idx);
self.data.get_unchecked_mut(flat_idx)
}
pub fn flat_len(self: & Self) -> usize { self.data.len() }
pub fn size(self: & Self) -> usize { self.flat_len()*std::mem::size_of::<T>() }
pub fn fill_with(self: &mut Self, x: & [T]) -> () {
// Already panics on size mismatch.
self.data.copy_from_slice(x)
}
pub fn fill(self: &mut Self, x: T) -> () {
self.data.fill(x);
}
pub fn data(self: & Self) -> & [T] { &self.data }
}
impl<T: Copy + std::fmt::Display> Tensor<T, 2> {
pub fn dirty_print(self: & Self) {
for i in 0..self.shape[0] {
for j in 0..self.shape[1] {
print!("{} ", self[[i, j]]);
}
println!("");
}
}
}
impl<T: Copy, const D: usize> std::ops::Index<Index<D>> for Tensor<T, D> {
type Output = T;
fn index(self: & Self, idx: Index<D>) -> & Self::Output {
self.bound_check_panic(&idx);
unsafe { self.el_unchecked(&idx) }
}
}
impl<T: Copy, const D: usize> std::ops::Index<& Index<D>> for Tensor<T, D> {
type Output = T;
fn index(self: & Self, idx: & Index<D>) -> & Self::Output {
self.bound_check_panic(idx);
unsafe { self.el_unchecked(idx) }
}
}
impl<T: Copy, const D: usize> std::ops::IndexMut<Index<D>> for Tensor<T, D> {
fn index_mut(self: &mut Self, idx: Index<D>) -> &mut Self::Output {
self.bound_check_panic(&idx);
unsafe { self.el_unchecked_mut(&idx) }
}
}
impl<T: Copy, const D: usize> std::ops::IndexMut<& Index<D>> for Tensor<T, D> {
fn index_mut(self: &mut Self, idx: & Index<D>) -> &mut Self::Output {
self.bound_check_panic(idx);
unsafe { self.el_unchecked_mut(idx) }
}
}
impl<T: Copy, const D: usize> IntoIterator for Tensor<T, D> {
type Item = T;
type IntoIter = std::vec::IntoIter<Self::Item>;
fn into_iter(self) -> Self::IntoIter { self.data.into_iter() }
}
// FIXME: Should have a proper IntoIter implementing IntoIter for &'a Tensor<T, D>.
// Note: Tensor is also sliceable (due to the Index implementations)
|
