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

use crate::internal::*;
use tract_ndarray::*;
use tract_num_traits::{AsPrimitive, One, Zero};

#[derive(Debug, Clone, new, Default, Educe)]
#[educe(Hash)]
pub struct ConstantLike {
    #[educe(Hash(method = "hash_f32"))]
    value: f32,
}

impl_dyn_hash!(ConstantLike);

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

    op_hir!();
    op_as_typed_op!();
}

impl EvalOp for ConstantLike {
    fn is_stateless(&self) -> bool {
        true
    }

    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let input = args_1!(inputs);
        Ok(tvec!(tensor0(self.value).broadcast_scalar_to_shape(input.shape())?.into_arc_tensor()))
    }
}

impl InferenceRulesOp for ConstantLike {
    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(&inputs[0].datum_type, &outputs[0].datum_type)?;
        s.equals(&inputs[0].rank, &outputs[0].rank)?;
        s.equals(&inputs[0].shape, &outputs[0].shape)?;
        s.given_2(&inputs[0].shape, &inputs[0].datum_type, move |s, shape, dt| {
            if shape.iter().all(|d| d.to_usize().is_ok()) {
                let shape: Vec<usize> = shape.iter().map(|d| d.to_usize().unwrap()).collect();
                let value = tensor0(self.value)
                    .cast_to_dt(dt)?
                    .broadcast_scalar_to_shape(&*shape)?
                    .into_arc_tensor();
                s.equals(&outputs[0].value, value)?;
            }
            Ok(())
        })
    }

    as_op!();
    to_typed!();
}

impl TypedOp for ConstantLike {
    as_op!();

    fn output_facts(&self, inputs: &[&TypedFact]) -> TractResult<TVec<TypedFact>> {
        Ok(tvec!(inputs[0].clone()))
    }
}

#[derive(Debug, Clone, new, Default, Hash)]
pub struct EyeLike {
    dt: Option<DatumType>,
    k: isize,
}

impl_dyn_hash!(EyeLike);

impl EyeLike {
    pub fn make<T>(&self, (r, c): (usize, usize)) -> TractResult<Arc<Tensor>>
    where
        T: Copy + Datum + One + Zero,
        f32: AsPrimitive<T>,
    {
        let mut array = Array2::<T>::zeros((r, c));
        for y in 0..r {
            let x = y as isize + self.k;
            if x >= 0 && x < c as isize {
                array[(y, x as usize)] = T::one()
            }
        }
        Ok(array.into_dyn().into_arc_tensor())
    }
}

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

    op_hir!();
    op_as_typed_op!();
}

impl EvalOp for EyeLike {
    fn is_stateless(&self) -> bool {
        true
    }

    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let input = args_1!(inputs);
        let dt = self.dt.unwrap_or(input.datum_type());
        Ok(tvec!(dispatch_numbers!(Self::make(dt)(self, (input.shape()[0], input.shape()[1])))?))
    }
}

impl InferenceRulesOp for EyeLike {
    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)?;
        if let Some(dt) = self.dt {
            s.equals(&outputs[0].datum_type, dt)?;
        } else {
            s.equals(&inputs[0].datum_type, &outputs[0].datum_type)?;
        }
        s.equals(&inputs[0].rank, 2)?;
        s.equals(&inputs[0].shape, &outputs[0].shape)?;
        s.given(&inputs[0].shape, move |s, shape| {
            if let (Ok(r), Ok(c)) = (shape[0].to_usize(), shape[1].to_usize()) {
                let shape = (r, c);
                if let Some(dt) = self.dt {
                    let value = dispatch_numbers!(Self::make(dt)(self, shape))?;
                    s.equals(&outputs[0].value, value)?;
                } else {
                    s.given(&inputs[0].datum_type, move |s, dt| {
                        let value = dispatch_numbers!(Self::make(dt)(self, shape))?;
                        s.equals(&outputs[0].value, value)
                    })?;
                }
            }
            Ok(())
        })
    }

    as_op!();
    to_typed!();
}

impl TypedOp for EyeLike {
    as_op!();

    fn output_facts(&self, inputs: &[&TypedFact]) -> TractResult<TVec<TypedFact>> {
        Ok(tvec!(TypedFact::dt_shape(
            self.dt.unwrap_or(inputs[0].datum_type),
            inputs[0].shape.iter()
        )))
    }
}