burn_tch/ops/
bool_tensor.rs

1use super::TchOps;
2use crate::{element::TchElement, LibTorch, LibTorchDevice, QuantElement, TchShape, TchTensor};
3use burn_tensor::{backend::Backend, ops::BoolTensorOps, Shape, TensorData, TensorMetadata};
4use std::ops::Range;
5
6impl<E: TchElement, Q: QuantElement> BoolTensorOps<Self> for LibTorch<E, Q> {
7    fn bool_from_data(data: TensorData, device: &LibTorchDevice) -> TchTensor {
8        TchTensor::from_data::<bool>(data, (*device).into())
9    }
10
11    fn bool_repeat_dim(tensor: TchTensor, dim: usize, times: usize) -> TchTensor {
12        TchOps::repeat_dim(tensor, dim, times)
13    }
14
15    async fn bool_into_data(tensor: TchTensor) -> TensorData {
16        let shape = tensor.shape();
17        let tensor = Self::bool_reshape(tensor.clone(), Shape::new([shape.num_elements()]));
18        let values: Result<Vec<bool>, tch::TchError> = tensor.tensor.shallow_clone().try_into();
19        TensorData::new(values.unwrap(), shape)
20    }
21
22    fn bool_to_device(tensor: TchTensor, device: &LibTorchDevice) -> TchTensor {
23        TchOps::to_device(tensor, device)
24    }
25
26    fn bool_reshape(tensor: TchTensor, shape: Shape) -> TchTensor {
27        TchOps::reshape(tensor, shape)
28    }
29
30    fn bool_device(tensor: &TchTensor) -> LibTorchDevice {
31        tensor.tensor.device().into()
32    }
33
34    fn bool_empty(shape: Shape, device: &<LibTorch<E> as Backend>::Device) -> TchTensor {
35        let tensor = tch::Tensor::empty(
36            TchShape::from(shape).dims,
37            (tch::Kind::Bool, (*device).into()),
38        );
39
40        TchTensor::new(tensor)
41    }
42
43    fn bool_slice(tensor: TchTensor, ranges: &[Range<usize>]) -> TchTensor {
44        TchOps::slice(tensor, ranges)
45    }
46
47    fn bool_slice_assign(
48        tensor: TchTensor,
49        ranges: &[Range<usize>],
50        value: TchTensor,
51    ) -> TchTensor {
52        TchOps::slice_assign(tensor, ranges, value)
53    }
54
55    fn bool_cat(tensors: Vec<TchTensor>, dim: usize) -> TchTensor {
56        TchOps::cat(tensors, dim)
57    }
58
59    fn bool_equal(lhs: TchTensor, rhs: TchTensor) -> TchTensor {
60        TchOps::equal(lhs, rhs)
61    }
62
63    fn bool_not(tensor: TchTensor) -> TchTensor {
64        tensor.unary_ops(
65            |mut tensor| tensor.eq_(0).to_kind(tch::Kind::Bool),
66            |tensor| tensor.eq(0),
67        )
68    }
69
70    fn bool_into_int(tensor: TchTensor) -> TchTensor {
71        let tensor = tensor.tensor.to_kind(tch::Kind::Int64);
72        TchTensor::new(tensor)
73    }
74
75    fn bool_into_float(tensor: TchTensor) -> TchTensor {
76        let tensor = tensor.tensor.to_kind(E::KIND);
77        TchTensor::new(tensor)
78    }
79
80    fn bool_swap_dims(tensor: TchTensor, dim1: usize, dim2: usize) -> TchTensor {
81        TchOps::swap_dims(tensor, dim1, dim2)
82    }
83
84    fn bool_narrow(tensor: TchTensor, dim: usize, start: usize, length: usize) -> TchTensor {
85        TchOps::narrow(tensor, dim, start, length)
86    }
87
88    fn bool_chunk(tensor: TchTensor, chunks: usize, dim: usize) -> Vec<TchTensor> {
89        TchOps::chunk(tensor, chunks, dim)
90    }
91
92    fn bool_split(tensor: TchTensor, split_size: usize, dim: usize) -> Vec<TchTensor> {
93        TchOps::split(tensor, split_size, dim)
94    }
95
96    fn bool_split_with_sizes(
97        tensor: TchTensor,
98        split_sizes: Vec<usize>,
99        dim: usize,
100    ) -> Vec<TchTensor> {
101        TchOps::split_with_sizes(tensor, split_sizes, dim)
102    }
103
104    fn bool_permute(tensor: TchTensor, axes: &[usize]) -> TchTensor {
105        TchOps::permute(tensor, axes)
106    }
107
108    fn bool_flip(tensor: TchTensor, axes: &[usize]) -> TchTensor {
109        TchOps::flip(tensor, axes)
110    }
111
112    async fn bool_argwhere(tensor: TchTensor) -> TchTensor {
113        TchTensor::new(tensor.tensor.argwhere())
114    }
115
116    async fn bool_nonzero(tensor: TchTensor) -> Vec<TchTensor> {
117        tensor
118            .tensor
119            .nonzero_numpy()
120            .into_iter()
121            // As opposed to tch, the resulting vector should be empty for zero tensors
122            .filter_map(|t| if t.numel() > 0 { Some(t) } else { None })
123            .map(TchTensor::new)
124            .collect()
125    }
126
127    fn bool_expand(tensor: TchTensor, shape: Shape) -> TchTensor {
128        TchOps::expand(tensor, shape)
129    }
130}
burn_tch/ops/bool_tensor.rs

burn_tch/ops/
bool_tensor.rs
