candela-tensor 0.2.0

A lazy, graph-based tensor engine in Rust
Documentation 
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

extern crate cblas;
extern crate intel_mkl_src;
extern crate intel_mkl_sys;

mod f32;
mod f64;
mod kernels;
mod mkl_extension;

use crate::Layout;
use crate::tensor::backend::{Backend, ComputeFor, Dtype};
use crate::tensor::ops::def_op::OpKind;
use crate::tensor::storage::TensorData;

#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Hash)]
pub struct CpuMkl;

impl Backend for CpuMkl {
    const SUPPORTS_2D_TRANSPOSED_MATMUL: bool = true;
    const SUPPORTS_NON_CONTIGUOUS_MATMUL: bool = false;

    fn compute<T>(
        op: &OpKind<T>,
        output_buffer: Vec<T>,
        output_layout: &Layout,
        inputs: &[TensorData<T>],
    ) -> TensorData<T>
    where
        T: Dtype + ComputeFor<CpuMkl>,
    {
        T::compute(op, output_buffer, output_layout, inputs)
    }

    fn compute_inplace<T>(
        op: &OpKind<T>,
        output_layout: &Layout,
        inputs: Vec<TensorData<T>>,
        output_idx: usize,
    ) -> TensorData<T>
    where
        T: Dtype + ComputeFor<Self>,
    {
        T::compute_inplace(op, output_layout, inputs, output_idx)
    }
}

impl ComputeFor<CpuMkl> for f64 {
    fn compute(
        op: &OpKind<f64>,
        output_buffer: Vec<f64>,
        output_layout: &Layout,
        inputs: &[TensorData<f64>],
    ) -> TensorData<f64> {
        f64::compute_op(op, output_buffer, output_layout, inputs)
    }

    fn compute_inplace(
        op: &OpKind<Self>,
        output_layout: &Layout,
        inputs: Vec<TensorData<Self>>,
        output_idx: usize,
    ) -> TensorData<Self> {
        f64::compute_op_inplace(op, output_layout, inputs, output_idx)
    }
}

impl ComputeFor<CpuMkl> for f32 {
    fn compute(
        op: &OpKind<f32>,
        output_buffer: Vec<f32>,
        output_layout: &Layout,
        inputs: &[TensorData<f32>],
    ) -> TensorData<f32> {
        f32::compute_op(op, output_buffer, output_layout, inputs)
    }

    fn compute_inplace(
        op: &OpKind<Self>,
        output_layout: &Layout,
        inputs: Vec<TensorData<Self>>,
        output_idx: usize,
    ) -> TensorData<Self> {
        f32::compute_op_inplace(op, output_layout, inputs, output_idx)
    }
}