numr 0.5.1

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
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
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

//! CPU implementation of einsum operations.

use crate::error::Result;
use crate::ops::EinsumOps;
use crate::ops::impl_generic::einsum::einsum_impl;
use crate::runtime::cpu::{CpuClient, CpuRuntime};
use crate::tensor::Tensor;

impl EinsumOps<CpuRuntime> for CpuClient {
    fn einsum(&self, notation: &str, inputs: &[&Tensor<CpuRuntime>]) -> Result<Tensor<CpuRuntime>> {
        einsum_impl(self, notation, inputs)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::ops::{BinaryOps, MatmulOps};
    use crate::runtime::Runtime;
    use crate::runtime::cpu::CpuDevice;

    fn create_client() -> (CpuClient, CpuDevice) {
        let device = CpuDevice::new();
        let client = CpuRuntime::default_client(&device);
        (client, device)
    }

    #[test]
    fn test_einsum_matmul() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0, 4.0], &[2, 2], &device);
        let b = Tensor::<CpuRuntime>::from_slice(&[5.0f32, 6.0, 7.0, 8.0], &[2, 2], &device);

        let result = client.einsum("ij,jk->ik", &[&a, &b]).unwrap();
        let expected = client.matmul(&a, &b).unwrap();

        let r: Vec<f32> = result.to_vec();
        let e: Vec<f32> = expected.to_vec();
        assert_eq!(r, e);
    }

    #[test]
    fn test_einsum_transpose() {
        let (client, device) = create_client();
        let a =
            Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0, 4.0, 5.0, 6.0], &[2, 3], &device);

        let result = client.einsum("ij->ji", &[&a]).unwrap();
        assert_eq!(result.shape(), &[3, 2]);
        let data: Vec<f32> = result.to_vec();
        assert_eq!(data, vec![1.0, 4.0, 2.0, 5.0, 3.0, 6.0]);
    }

    #[test]
    fn test_einsum_trace() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0, 4.0], &[2, 2], &device);

        let result = client.einsum("ii->", &[&a]).unwrap();
        let data: Vec<f32> = result.to_vec();
        assert_eq!(data, vec![5.0]); // 1 + 4
    }

    #[test]
    fn test_einsum_diagonal() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0, 4.0], &[2, 2], &device);

        let result = client.einsum("ii->i", &[&a]).unwrap();
        let data: Vec<f32> = result.to_vec();
        assert_eq!(data, vec![1.0, 4.0]);
    }

    #[test]
    fn test_einsum_sum_all() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0, 4.0], &[2, 2], &device);

        let result = client.einsum("ij->", &[&a]).unwrap();
        let data: Vec<f32> = result.to_vec();
        assert_eq!(data, vec![10.0]);
    }

    #[test]
    fn test_einsum_sum_rows() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0, 4.0], &[2, 2], &device);

        let result = client.einsum("ij->i", &[&a]).unwrap();
        let data: Vec<f32> = result.to_vec();
        assert_eq!(data, vec![3.0, 7.0]);
    }

    #[test]
    fn test_einsum_outer_product() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0], &[3], &device);
        let b = Tensor::<CpuRuntime>::from_slice(&[4.0f32, 5.0], &[2], &device);

        let result = client.einsum("i,j->ij", &[&a, &b]).unwrap();
        assert_eq!(result.shape(), &[3, 2]);
        let data: Vec<f32> = result.to_vec();
        assert_eq!(data, vec![4.0, 5.0, 8.0, 10.0, 12.0, 15.0]);
    }

    #[test]
    fn test_einsum_hadamard() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0, 4.0], &[2, 2], &device);
        let b = Tensor::<CpuRuntime>::from_slice(&[5.0f32, 6.0, 7.0, 8.0], &[2, 2], &device);

        let result = client.einsum("ij,ij->ij", &[&a, &b]).unwrap();
        let expected = client.mul(&a, &b).unwrap();

        let r: Vec<f32> = result.to_vec();
        let e: Vec<f32> = expected.to_vec();
        assert_eq!(r, e);
    }

    #[test]
    fn test_einsum_dot_product() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0], &[3], &device);
        let b = Tensor::<CpuRuntime>::from_slice(&[4.0f32, 5.0, 6.0], &[3], &device);

        let result = client.einsum("i,i->", &[&a, &b]).unwrap();
        let data: Vec<f32> = result.to_vec();
        assert_eq!(data, vec![32.0]); // 1*4 + 2*5 + 3*6
    }

    #[test]
    fn test_einsum_implicit_notation() {
        let (client, device) = create_client();
        let a = Tensor::<CpuRuntime>::from_slice(&[1.0f32, 2.0, 3.0, 4.0], &[2, 2], &device);
        let b = Tensor::<CpuRuntime>::from_slice(&[5.0f32, 6.0, 7.0, 8.0], &[2, 2], &device);

        // Implicit "ij,jk" = "ij,jk->ik"
        let result = client.einsum("ij,jk", &[&a, &b]).unwrap();
        let expected = client.matmul(&a, &b).unwrap();

        let r: Vec<f32> = result.to_vec();
        let e: Vec<f32> = expected.to_vec();
        assert_eq!(r, e);
    }
}