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

use tract_hir::internal::*;
use tract_ndarray::prelude::*;

use tract_hir::ops::cnn::{ConvUnary, PoolSpec};

#[derive(Debug, Copy, Clone, Hash)]
pub enum PaddingStrat {
    FlexFixed(usize),
    FixedFlex(usize),
    FixedFixed(usize, usize),
}

#[derive(Debug, Clone, new, Hash)]
pub struct SpaceToBatchUnary {
    pub datum_type: DatumType,
    pub space_shape: TVec<TDim>,
    pub batch_shape: TVec<TDim>,
    pub block_shape: Array1<i32>,
    pub pad: TVec<PaddingStrat>,
}

impl_dyn_hash!(SpaceToBatchUnary);

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

    op_tf!();
    op_as_typed_op!();
}

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

    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let input = args_1!(inputs);
        let mut paddings = unsafe { Array2::uninitialized((self.block_shape.len(), 2)) };
        for (ax, &strat) in self.pad.iter().enumerate() {
            let spread = (self.batch_shape[2 + ax].clone() * self.block_shape[ax]
                - &self.space_shape[2 + ax])
                .to_usize()?;
            let (bef, aft) = match strat {
                PaddingStrat::FlexFixed(f) => (spread - f, f),
                PaddingStrat::FixedFlex(f) => (f, spread - f),
                PaddingStrat::FixedFixed(a, b) => (a, b),
            };
            paddings[(ax, 0)] = bef as i32;
            paddings[(ax, 1)] = aft as i32;
        }
        let r = dispatch_numbers!(super::space_to_batch(input.datum_type())(
            input,
            &self.block_shape.view(),
            &paddings.view()
        ))?;
        Ok(tvec!(r))
    }
}

impl TypedOp for SpaceToBatchUnary {
    fn output_facts(&self, inputs: &[&TypedFact]) -> TractResult<TVec<TypedFact>> {
        Ok(tvec!(TypedFact::dt_shape(inputs[0].datum_type, &self.batch_shape)))
    }

    fn declutter(
        &self,
        model: &TypedModel,
        node: &TypedNode,
    ) -> TractResult<Option<TypedModelPatch>> {
        if let Some(conv_node) = model.single_succ(node.id)? {
            if let Some(b2s_node) = model.single_succ(conv_node.id)? {
                if let (Some(conv_op), Some(_)) =
                    (conv_node.op_as::<ConvUnary>(), b2s_node.op_as::<BatchToSpaceUnary>())
                {
                    let op = ConvUnary {
                        pool_spec: PoolSpec {
                            data_format: conv_op.pool_spec.data_format,
                            padding: conv_op.pool_spec.padding.clone(), // FIXME
                            strides: conv_op.pool_spec.strides.clone(),
                            kernel_shape: conv_op.pool_spec.kernel_shape.clone(),
                            output_channel_override: conv_op.pool_spec.output_channel_override,
                            dilations: Some(self.block_shape.iter().map(|&i| i as usize).collect()),
                        },
                        kernel_fmt: conv_op.kernel_fmt,
                        kernel: conv_op.kernel.clone(),
                        group: conv_op.group,
                        bias: None,
                        q_params: None,
                    };
                    let mut patch = TypedModelPatch::default();
                    let tap = patch.tap_model(&model, node.inputs[0])?;
                    let out = patch.model.wire_node(&*conv_node.name, op, &[tap])?[0];
                    patch.shunt_outside(model, OutletId::new(b2s_node.id, 0), out)?;
                    return Ok(Some(patch));
                }
            }
        }
        Ok(None)
    }

    as_op!();
}

#[derive(Debug, Clone, new, Hash)]
pub struct BatchToSpaceUnary {
    datum_type: DatumType,
    batch_shape: TVec<TDim>,
    space_shape: TVec<TDim>,
    block_shape: Array1<i32>,
    pad: Vec<PaddingStrat>,
}

impl_dyn_hash!(BatchToSpaceUnary);

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

    op_tf!();
    op_as_typed_op!();
}

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

    fn eval(&self, mut inputs: TVec<Arc<Tensor>>) -> TractResult<TVec<Arc<Tensor>>> {
        let input = args_1!(inputs);
        let mut paddings = unsafe { Array2::uninitialized((self.block_shape.len(), 2)) };
        for (ax, &strat) in self.pad.iter().enumerate() {
            let spread = (self.batch_shape[2 + ax].clone() * self.block_shape[ax]
                - &self.space_shape[2 + ax])
                .to_usize()?;
            let (bef, aft) = match strat {
                PaddingStrat::FlexFixed(f) => (spread - f, f),
                PaddingStrat::FixedFlex(f) => (f, spread - f),
                PaddingStrat::FixedFixed(a, b) => (a, b),
            };
            paddings[(ax, 0)] = bef as i32;
            paddings[(ax, 1)] = aft as i32;
        }
        let r = dispatch_numbers!(super::batch_to_space(input.datum_type())(
            input,
            &self.block_shape.view(),
            &paddings.view()
        ))?;
        Ok(tvec!(r))
    }
}

impl TypedOp for BatchToSpaceUnary {
    as_op!();

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