numr 0.5.2

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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167

//! LogicalOps implementation for WebGPU runtime.
//!
//! WebGPU does not support U8, so logical operations use U32 tensors
//! where 0 = false and non-zero = true.

use crate::dtype::DType;
use crate::error::{Error, Result};
use crate::ops::LogicalOps;
use crate::runtime::wgpu::ops::native::logical::{
    native_logical_and, native_logical_not, native_logical_or, native_logical_xor,
};
use crate::runtime::wgpu::{WgpuClient, WgpuRuntime};
use crate::tensor::Tensor;

impl LogicalOps<WgpuRuntime> for WgpuClient {
    fn logical_and(
        &self,
        a: &Tensor<WgpuRuntime>,
        b: &Tensor<WgpuRuntime>,
    ) -> Result<Tensor<WgpuRuntime>> {
        // WebGPU uses U32 for boolean tensors (no U8 support)
        validate_logical_inputs(a, b)?;
        native_logical_and(self, a, b)
    }

    fn logical_or(
        &self,
        a: &Tensor<WgpuRuntime>,
        b: &Tensor<WgpuRuntime>,
    ) -> Result<Tensor<WgpuRuntime>> {
        validate_logical_inputs(a, b)?;
        native_logical_or(self, a, b)
    }

    fn logical_xor(
        &self,
        a: &Tensor<WgpuRuntime>,
        b: &Tensor<WgpuRuntime>,
    ) -> Result<Tensor<WgpuRuntime>> {
        validate_logical_inputs(a, b)?;
        native_logical_xor(self, a, b)
    }

    fn logical_not(&self, a: &Tensor<WgpuRuntime>) -> Result<Tensor<WgpuRuntime>> {
        // WebGPU uses U32 for boolean tensors (no U8 support)
        if a.dtype() != DType::U32 {
            return Err(Error::DTypeMismatch {
                lhs: DType::U32,
                rhs: a.dtype(),
            });
        }
        native_logical_not(self, a)
    }
}

/// Validate inputs for binary logical operations.
/// WebGPU requires U32 dtype (no U8 support).
fn validate_logical_inputs(a: &Tensor<WgpuRuntime>, b: &Tensor<WgpuRuntime>) -> Result<()> {
    // WebGPU uses U32 for boolean tensors
    if a.dtype() != DType::U32 {
        return Err(Error::DTypeMismatch {
            lhs: DType::U32,
            rhs: a.dtype(),
        });
    }
    if b.dtype() != DType::U32 {
        return Err(Error::DTypeMismatch {
            lhs: DType::U32,
            rhs: b.dtype(),
        });
    }

    // Shapes must match
    if a.shape() != b.shape() {
        return Err(Error::ShapeMismatch {
            expected: a.shape().to_vec(),
            got: b.shape().to_vec(),
        });
    }

    Ok(())
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::runtime::wgpu::device::WgpuDevice;
    use crate::runtime::wgpu::runtime::WgpuRuntime as WgpuRt;
    use crate::runtime::{Runtime, RuntimeClient};

    fn create_client() -> WgpuClient {
        let device = WgpuDevice::new(0);
        WgpuRt::default_client(&device)
    }

    #[test]
    fn test_logical_and() {
        let client = create_client();

        let a = Tensor::<WgpuRuntime>::from_slice(&[1u32, 0, 1, 0], &[4], client.device());
        let b = Tensor::<WgpuRuntime>::from_slice(&[1u32, 1, 0, 0], &[4], client.device());

        let result = client.logical_and(&a, &b).unwrap();
        let data: Vec<u32> = result.to_vec();

        assert_eq!(data, vec![1, 0, 0, 0]);
    }

    #[test]
    fn test_logical_or() {
        let client = create_client();

        let a = Tensor::<WgpuRuntime>::from_slice(&[1u32, 0, 1, 0], &[4], client.device());
        let b = Tensor::<WgpuRuntime>::from_slice(&[1u32, 1, 0, 0], &[4], client.device());

        let result = client.logical_or(&a, &b).unwrap();
        let data: Vec<u32> = result.to_vec();

        assert_eq!(data, vec![1, 1, 1, 0]);
    }

    #[test]
    fn test_logical_xor() {
        let client = create_client();

        let a = Tensor::<WgpuRuntime>::from_slice(&[1u32, 0, 1, 0], &[4], client.device());
        let b = Tensor::<WgpuRuntime>::from_slice(&[1u32, 1, 0, 0], &[4], client.device());

        let result = client.logical_xor(&a, &b).unwrap();
        let data: Vec<u32> = result.to_vec();

        assert_eq!(data, vec![0, 1, 1, 0]);
    }

    #[test]
    fn test_logical_not() {
        let client = create_client();

        let a = Tensor::<WgpuRuntime>::from_slice(&[1u32, 0, 5, 0], &[4], client.device());

        let result = client.logical_not(&a).unwrap();
        let data: Vec<u32> = result.to_vec();

        assert_eq!(data, vec![0, 1, 0, 1]);
    }

    #[test]
    fn test_logical_rejects_non_u32() {
        let client = create_client();

        let a = Tensor::<WgpuRuntime>::from_slice(&[1.0f32, 0.0], &[2], client.device());
        let b = Tensor::<WgpuRuntime>::from_slice(&[1.0f32, 1.0], &[2], client.device());

        assert!(client.logical_and(&a, &b).is_err());
        assert!(client.logical_not(&a).is_err());
    }

    #[test]
    fn test_logical_shape_mismatch() {
        let client = create_client();

        let a = Tensor::<WgpuRuntime>::from_slice(&[1u32, 0, 1], &[3], client.device());
        let b = Tensor::<WgpuRuntime>::from_slice(&[1u32, 1], &[2], client.device());

        assert!(client.logical_and(&a, &b).is_err());
    }
}