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

use crate::internal::*;

#[derive(Debug, Clone, new, Default)]
pub struct MultiBroadcastTo;

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

    not_a_typed_op!();
}

impl StatelessOp for MultiBroadcastTo {
    /// Evaluates the operation given the input tensors.
    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let (input, dims) = args_2!(inputs);
        let dims: Vec<usize> = dims.to_array_view::<i64>()?.iter().map(|i| *i as usize).collect();
        let dims = crate::broadcast::multi_broadcast(&[&*dims, &*input.shape()])
            .ok_or("incompatible shapes")?;
        dispatch_datum!(self::eval_t(input.datum_type())(&*input, &*dims))
    }
}

impl InferenceRulesOp for MultiBroadcastTo {
    fn rules<'r, 'p: 'r, 's: 'r>(
        &'s self,
        s: &mut Solver<'r>,
        inputs: &'p [TensorProxy],
        outputs: &'p [TensorProxy],
    ) -> InferenceResult {
        check_input_arity(&inputs, 2)?;
        check_output_arity(&outputs, 1)?;
        s.equals(&outputs[0].datum_type, &inputs[0].datum_type)?;
        s.equals(&inputs[1].rank, 1)?;
        s.given(&inputs[0].shape, move |s, shape| {
            s.given(&inputs[1].value, move |s, dims| {
                let dims = dims.cast_to::<TDim>()?;
                let dims =
                    crate::broadcast::multi_broadcast(&[&*dims.as_slice::<TDim>()?, &*shape])
                        .ok_or("incompatible shapes")
                        .unwrap();
                s.equals(&outputs[0].shape, ShapeFact::from(dims))
            })
        })
    }

    fn to_typed(
        &self,
        source: &InferenceModel,
        node: &InferenceNode,
        target: &mut TypedModel,
        mapping: &HashMap<OutletId, OutletId>,
    ) -> TractResult<TVec<OutletId>> {
        if let Some(ref shape) = source.outlet_fact(node.inputs[1])?.value.concretize() {
            let shape = shape.cast_to::<TDim>()?;
            let op = TypedMultiBroadcastTo::new(shape.as_slice::<TDim>()?.into());
            return target.wire_node(&*node.name, op, [mapping[&node.inputs[0]]].as_ref());
        }
        bail!("shape input is variable")
    }

    inference_op_as_op!();
}

#[derive(Debug, Clone, new, Default)]
pub struct TypedMultiBroadcastTo {
    shape: TVec<TDim>,
}

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

    canonic!();
    op_as_typed_op!();
}

impl StatelessOp for TypedMultiBroadcastTo {
    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let input = args_1!(inputs);
        let dims: Vec<usize> =
            self.shape.iter().map(|d| Ok(d.to_integer()? as usize)).collect::<TractResult<_>>()?;
        dispatch_datum!(self::eval_t(input.datum_type())(&*input, &*dims))
    }
}

impl TypedOp for TypedMultiBroadcastTo {
    fn output_facts(&self, inputs: &[&TypedTensorInfo]) -> TractResult<TVec<TypedTensorInfo>> {
        Ok(tvec!(TypedTensorInfo::dt_shape(inputs[0].datum_type, &*self.shape)?))
    }

    typed_op_as_op!();
}

fn eval_t<T: Datum>(input: &Tensor, shape: &[usize]) -> TractResult<TVec<Arc<Tensor>>> {
    let input = input.to_array_view::<T>()?;
    let output = input.broadcast(&*shape).ok_or("incompatible shapes")?;
    Ok(tvec![output.to_owned().into_arc_tensor()])
}