tract_core/ops/matmul/
pack.rs

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
use crate::axes::Axis;
use crate::internal::*;
use ndarray::*;
use tract_linalg::frame::PackedFormat;

use super::ModePicker;

#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct OptMatMulPack {
    pub(crate) packers: Vec<PackedFormat>,
    pub(crate) mode_picker: ModePicker,
    pub(crate) k_axis: usize,
    pub(crate) mn_axis: usize,
}

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

    fn info(&self) -> TractResult<Vec<String>> {
        Ok(vec![format!("{:?}. k axis: {}, mn axis: {}", self.packers, self.k_axis, self.mn_axis)])
    }

    op_as_typed_op!();
    impl_op_same_as!();
}

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

    fn eval_with_session(
        &self,
        session: &SessionState,
        mut inputs: TVec<TValue>,
    ) -> TractResult<TVec<TValue>> {
        self.do_eval(session, inputs.remove(0))
    }
}

impl TypedOp for OptMatMulPack {
    fn output_facts(&self, inputs: &[&TypedFact]) -> TractResult<TVec<TypedFact>> {
        let k = inputs[0].shape[self.k_axis].clone();
        let mn = inputs[0].shape[self.mn_axis].clone();
        let opaque_fact = PackedOpaqueFact { k, mn, packers: self.packers.clone() };
        Ok(tvec!(Opaque::datum_type()
            .fact(self.output_shape(&inputs[0].shape))
            .with_opaque_fact(opaque_fact)))
    }

    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()));
        AxesMapping::new(1, 1, axes)
    }

    as_op!();
}

impl OptMatMulPack {
    fn do_eval(&self, _session: &SessionState, input: TValue) -> TractResult<TVec<TValue>> {
        unsafe {
            let mode = self.mode_picker.pick(input.shape()[self.mn_axis])?;
            let packer = &self.packers[mode];
            let output_shape: TVec<usize> = self.output_shape(input.shape());
            let stores = if output_shape.iter().all(|d| *d == 1) {
                tensor0::<Opaque>(
                    packer.pack_tensor_view(&input.view(), self.k_axis, self.mn_axis)?.into(),
                )
                .into_shape(&output_shape)?
            } else {
                let mut stores = Tensor::uninitialized_dt(Opaque::datum_type(), &output_shape)?;
                let mut stores_view = stores.to_array_view_mut::<Opaque>()?;
                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 pack_coords: TVec<usize> = coord.slice().into();
                    pack_coords.remove(self.k_axis.max(self.mn_axis));
                    pack_coords.remove(self.k_axis.min(self.mn_axis));
                    stores_view[&*pack_coords] = packer
                        .pack_tensor_view(
                            &TensorView::from_bytes(&input, offset, input.shape(), input.strides()),
                            self.k_axis,
                            self.mn_axis,
                        )?
                        .into();
                }
                stores
            };
            Ok(tvec!(stores.into_tvalue()))
        }
    }

    pub fn output_shape<D: DimLike>(&self, input: &[D]) -> TVec<D> {
        let mut packed_shape: TVec<D> = input.into();
        packed_shape.remove(self.mn_axis.max(self.k_axis));
        packed_shape.remove(self.mn_axis.min(self.k_axis));
        packed_shape
    }
}

#[derive(Hash, Clone, Debug, PartialEq, Eq)]
pub struct PackedOpaqueFact {
    pub k: TDim,
    pub mn: TDim,
    pub packers: Vec<PackedFormat>,
}

impl OpaqueFact for PackedOpaqueFact {
    fn mem_size(&self) -> TDim {
        self.k.clone() * &self.mn * self.packers[0].dt.size_of()
    }
}