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
176
177
178
179
180
181
182

use crate::internal::*;
use ndarray::prelude::*;

#[derive(Debug, Clone, new, Default, Hash)]
pub struct GlobalAvgPool {
    //    data_is_nhwc: bool, // default is nchw (onnx)
}

impl GlobalAvgPool {
    fn eval_t<D: Datum + ::num_traits::Float + ::num_traits::FromPrimitive>(
        &self,
        input: Arc<Tensor>,
    ) -> TractResult<TVec<Arc<Tensor>>> {
        let array = input.to_array_view::<D>()?;
        let n = array.shape()[0];
        let c = array.shape()[1];
        let mut final_shape = array.shape().to_vec();
        for dim in final_shape[2..].iter_mut() {
            *dim = 1;
        }
        let divisor_int = array.len() / (n * c);
        let divisor = D::from(divisor_int).unwrap().recip();
        let result: Tensor = array
            .into_shape(((n * c), divisor_int))?
            .sum_axis(Axis(1))
            .map(|x| *x * divisor)
            .into_shape(final_shape)?
            .into();
        Ok(tvec!(result.into()))
    }
}

tract_linalg::impl_dyn_hash!(GlobalAvgPool);

impl Op for GlobalAvgPool {
    fn name(&self) -> Cow<str> {
        "GlobalAvgPool".into()
    }
    fn validation(&self) -> Validation {
        Validation::Rounding
    }
    op_core_mir!();
    op_as_typed_op!();
    not_a_pulsed_op!();
}

impl StatelessOp for GlobalAvgPool {
    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let input = args_1!(inputs);
        dispatch_floatlike!(Self::eval_t(input.datum_type())(self, input))
    }
}

impl TypedOp for GlobalAvgPool {
    as_op!();

    fn output_facts(&self, inputs: &[&TypedFact]) -> TractResult<TVec<TypedFact>> {
        output_facts(inputs)
    }
}

#[derive(Debug, Clone, new, Default, Hash)]
pub struct GlobalLpPool {
    p: usize, //    data_is_nhwc: bool, // default is nchw (onnx)
}

tract_linalg::impl_dyn_hash!(GlobalLpPool);

impl GlobalLpPool {
    fn eval_t<D: Datum + ::num_traits::Float>(
        &self,
        input: Arc<Tensor>,
    ) -> TractResult<TVec<Arc<Tensor>>> {
        let array = input.to_array_view::<D>()?;
        let n = array.shape()[0];
        let c = array.shape()[1];
        let mut final_shape = array.shape().to_vec();
        for dim in final_shape[2..].iter_mut() {
            *dim = 1;
        }
        let divisor = array.len() / (n * c);
        let input = array.into_shape(((n * c), divisor))?;
        let divisor = D::from(divisor).unwrap().recip();
        let result = if self.p == 1 {
            input.fold_axis(Axis(1), D::zero(), |&a, &b| a + b.abs()).map(|a| *a * divisor)
        } else if self.p == 2 {
            input.fold_axis(Axis(1), D::zero(), |&a, &b| a + b * b).map(|a| a.sqrt() * divisor)
        } else {
            input
                .fold_axis(Axis(1), D::zero(), |&a, &b| a + b.abs().powi(self.p as i32))
                .map(|a| a.powf(D::from(self.p).unwrap().recip()) * divisor)
        };
        Ok(tvec!(result.into_shape(final_shape)?.into_arc_tensor()))
    }
}

impl Op for GlobalLpPool {
    fn name(&self) -> Cow<str> {
        "GlobalLpPool".into()
    }
    fn validation(&self) -> Validation {
        Validation::Rounding
    }
    op_core_mir!();
    op_as_typed_op!();
    not_a_pulsed_op!();
}

impl StatelessOp for GlobalLpPool {
    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let input = args_1!(inputs);
        dispatch_floatlike!(Self::eval_t(input.datum_type())(self, input))
    }
}
impl TypedOp for GlobalLpPool {
    as_op!();

    fn output_facts(&self, inputs: &[&TypedFact]) -> TractResult<TVec<TypedFact>> {
        output_facts(inputs)
    }
}

#[derive(Debug, Clone, new, Default, Hash)]
pub struct GlobalMaxPool {
    //    data_is_nhwc: bool, // default is nchw (onnx)
}

tract_linalg::impl_dyn_hash!(GlobalMaxPool);

impl GlobalMaxPool {
    fn eval_t<D: Datum + ::num_traits::Float>(
        &self,
        input: Arc<Tensor>,
    ) -> TractResult<TVec<Arc<Tensor>>> {
        let array = input.to_array_view::<D>()?;
        let n = array.shape()[0];
        let c = array.shape()[1];
        let mut final_shape = array.shape().to_vec();
        for dim in final_shape[2..].iter_mut() {
            *dim = 1;
        }
        let divisor = array.len() / (n * c);
        let result: Tensor = array
            .into_shape(((n * c), divisor))?
            .fold_axis(Axis(1), D::min_value(), |a, b| a.max(*b))
            .into_shape(final_shape)?
            .into();
        Ok(tvec!(result.into()))
    }
}

impl Op for GlobalMaxPool {
    fn name(&self) -> Cow<str> {
        "GlobalMaxPool".into()
    }
    op_core_mir!();
    op_as_typed_op!();
    not_a_pulsed_op!();
}

impl StatelessOp for GlobalMaxPool {
    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let input = args_1!(inputs);
        dispatch_floatlike!(Self::eval_t(input.datum_type())(self, input))
    }
}

impl TypedOp for GlobalMaxPool {
    as_op!();

    fn output_facts(&self, inputs: &[&TypedFact]) -> TractResult<TVec<TypedFact>> {
        output_facts(inputs)
    }
}

fn output_facts(inputs: &[&TypedFact]) -> TractResult<TVec<TypedFact>> {
    let mut output = inputs[0].clone();
    for i in 2..output.shape.rank() {
        output.shape.set_dim(i, TDim::from(1))?
    }
    Ok(tvec!(output))
}