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

use crate::axes::Axis;
use crate::internal::*;
use ndarray::*;

use tract_linalg::frame::Packer;

#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct MatMatMulPack {
    pub(crate) packer: Packer,
    pub(crate) k_axis: usize,
    pub(crate) mn_axis: usize,
    pub(crate) output_shape_fact: ShapeFact,
}

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

    fn same_as(&self, other: &dyn Op) -> bool {
        other.downcast_ref::<Self>().map(|other| other == self).unwrap_or(false)
    }

    op_as_typed_op!();
}

impl EvalOp for MatMatMulPack {
    fn is_stateless(&self) -> bool {
        self.output_shape_fact.is_concrete()
    }

    fn eval(&self, inputs: TVec<TValue>) -> TractResult<TVec<TValue>> {
        let output_shape = self.output_shape_fact.as_concrete().unwrap();
        self.do_eval(&inputs[0], output_shape)
    }

    fn state(
        &self,
        _session: &mut SessionState,
        _node_id: usize,
    ) -> TractResult<Option<Box<dyn OpState>>> {
        Ok(Some(Box::new(self.clone())))
    }
}

impl OpState for MatMatMulPack {
    fn eval(
        &mut self,
        session: &mut SessionState,
        _op: &dyn Op,
        inputs: TVec<TValue>,
    ) -> TractResult<TVec<TValue>> {
        let output_shape = self.output_shape_fact.eval_to_usize(&session.resolved_symbols)?;
        self.do_eval(&inputs[0], &output_shape)
    }
}

trivial_op_state_freeeze!(MatMatMulPack);

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

    fn axes_mapping(
        &self,
        inputs: &[&TypedFact],
        outputs: &[&TypedFact],
    ) -> TractResult<AxesMapping> {
        let mut axes: Vec<Axis> = (0..inputs[0].rank())
            .filter(|&ix| ix != self.k_axis && ix != self.mn_axis)
            .enumerate()
            .zip('a'..)
            .map(|((o, i), repr)| Axis::new(repr, 1, 1).input(0, i).output(0, o))
            .collect();
        axes.push(Axis::new('K', 1, 1).input(0, self.k_axis));
        axes.push(Axis::new('M', 1, 1).input(0, self.mn_axis));
        axes.push(Axis::new('P', 1, 1).output(0, outputs[0].rank() - 1));
        AxesMapping::new(1, 1, axes)
    }

    fn concretize_dims(
        &self,
        _source: &TypedModel,
        node: &TypedNode,
        target: &mut TypedModel,
        mapping: &HashMap<OutletId, OutletId>,
        values: &SymbolValues,
    ) -> TractResult<TVec<OutletId>> {
        let output_shape_fact = self.output_shape_fact.eval(values)?.into_owned();
        let inputs: TVec<OutletId> = node.inputs.iter().map(|o| mapping[o]).collect();
        target.wire_node(&node.name, MatMatMulPack { output_shape_fact, ..self.clone() }, &inputs)
    }

    as_op!();
}

impl MatMatMulPack {
    fn do_eval(&self, input: &Tensor, output_shape: &[usize]) -> TractResult<TVec<TValue>> {
        let dt = input.datum_type();
        unsafe {
            let mut packed =
                Tensor::uninitialized_aligned_dt(dt, output_shape, self.packer.alignment())
                    .unwrap();
            let mut bc_shape: TVec<usize> = input.shape().into();
            bc_shape[self.k_axis] = 1;
            bc_shape[self.mn_axis] = 1;
            for coord in indices(&*bc_shape) {
                let offset = coord
                    .as_array_view()
                    .iter()
                    .zip(input.strides())
                    .map(|(x, s)| *x as isize * s)
                    .sum::<isize>()
                    * input.datum_type().size_of() as isize;
                let mut prefix: TVec<usize> = coord.slice().into();
                prefix.remove(self.k_axis.max(self.mn_axis));
                prefix.remove(self.k_axis.min(self.mn_axis));
                self.packer.pack(
                    &mut packed.view_at_prefix_mut(&prefix)?,
                    TensorView::from_bytes(input, offset, input.shape(), input.strides()),
                    self.k_axis,
                    self.mn_axis,
                )
            }
            Ok(tvec!(packed.into_tvalue()))
        }
    }

    pub fn output_shape<D: DimLike>(
        input: &[D],
        packer: &Packer,
        mn_axis: usize,
        k_axis: usize,
    ) -> ShapeFact {
        let mut packed_shape: TVec<D> = input.into();
        packed_shape.remove(mn_axis.max(k_axis));
        packed_shape.remove(mn_axis.min(k_axis));
        packed_shape.push(packer.len(input[k_axis].clone(), input[mn_axis].clone()));
        packed_shape.into()
    }
}