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

use ndarray::*;

use crate::internal::*;

#[derive(Debug, Clone, new)]
pub struct ConstantOfShape {
    value: Arc<Tensor>,
}

impl ConstantOfShape {
    pub fn make<T>(&self, shape: &Arc<Tensor>) -> TractResult<Arc<Tensor>>
    where
        T: Datum + Copy,
    {
        let shape: TVec<usize> =
            shape.cast_to::<i64>()?.as_slice::<i64>()?.iter().map(|&x| x as usize).collect();
        Ok(Array::<T, _>::from_elem(&*shape, *self.value.to_scalar()?).into_arc_tensor())
    }
}

impl Op for ConstantOfShape {
    fn name(&self) -> Cow<str> {
        "ConstantOfShape".into()
    }
}

impl StatelessOp for ConstantOfShape {
    /// Evaluates the operation given the input tensors.
    fn eval(&self, inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        Ok(tvec!(dispatch_numbers!(Self::make(self.value.datum_type())(self, &inputs[0]))?))
    }
}

impl InferenceRulesOp for ConstantOfShape {
    fn rules<'r, 'p: 'r, 's: 'r>(
        &'s self,
        s: &mut Solver<'r>,
        inputs: &'p [TensorProxy],
        outputs: &'p [TensorProxy],
    ) -> InferenceResult {
        check_input_arity(&inputs, 1)?;
        check_output_arity(&outputs, 1)?;
        s.equals(&outputs[0].datum_type, self.value.datum_type())?;
        s.equals(&inputs[0].rank, 1)?;
        s.equals(&inputs[0].shape[0], outputs[0].rank.bex().to_dim())?;
        // FIXME: inputs[0] is int64 and to_dim only works on i32.
        //        s.given(&outputs[0].rank, move |s, rank| {
        //            for idx in 0..(rank as usize) {
        //                s.equals(inputs[0].value[idx].bex().to_dim(), &outputs[0].shape[idx])?;
        //            }
        //            Ok(())
        //        })?;
        Ok(())
    }
}