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

//! Native GPU implementation for logical operations on WebGPU.
//!
//! WebGPU uses U32 for boolean tensors (0 = false, non-zero = true).

use super::helpers::*;
use crate::dtype::DType;
use crate::error::Result;
use crate::runtime::ensure_contiguous;
use crate::runtime::wgpu::shaders::logical::{
    launch_logical_and, launch_logical_not, launch_logical_or, launch_logical_xor,
};
use crate::runtime::wgpu::{WgpuClient, WgpuRuntime};
use crate::tensor::Tensor;

/// Parameters for binary logical operations.
#[repr(C)]
#[derive(Clone, Copy, bytemuck::Pod, bytemuck::Zeroable)]
pub(crate) struct LogicalBinaryParams {
    pub(crate) numel: u32,
}

/// Parameters for unary logical operations.
#[repr(C)]
#[derive(Clone, Copy, bytemuck::Pod, bytemuck::Zeroable)]
pub(crate) struct LogicalUnaryParams {
    pub(crate) numel: u32,
}

/// Native logical AND operation.
#[allow(dead_code)]
pub(crate) fn native_logical_and(
    client: &WgpuClient,
    a: &Tensor<WgpuRuntime>,
    b: &Tensor<WgpuRuntime>,
) -> Result<Tensor<WgpuRuntime>> {
    let a_contig = ensure_contiguous(a);
    let b_contig = ensure_contiguous(b);
    let numel = a.numel();

    let out = alloc_output(client, a.shape(), DType::U32);

    let a_buf = get_tensor_buffer(&a_contig)?;
    let b_buf = get_tensor_buffer(&b_contig)?;
    let out_buf = get_tensor_buffer(&out)?;

    let params = LogicalBinaryParams {
        numel: numel as u32,
    };
    let params_buf = create_params_buffer(client, &params);

    launch_logical_and(
        client.pipeline_cache(),
        client.wgpu_queue(),
        &a_buf,
        &b_buf,
        &out_buf,
        &params_buf,
        numel,
    )?;

    Ok(out)
}

/// Native logical OR operation.
#[allow(dead_code)]
pub(crate) fn native_logical_or(
    client: &WgpuClient,
    a: &Tensor<WgpuRuntime>,
    b: &Tensor<WgpuRuntime>,
) -> Result<Tensor<WgpuRuntime>> {
    let a_contig = ensure_contiguous(a);
    let b_contig = ensure_contiguous(b);
    let numel = a.numel();

    let out = alloc_output(client, a.shape(), DType::U32);

    let a_buf = get_tensor_buffer(&a_contig)?;
    let b_buf = get_tensor_buffer(&b_contig)?;
    let out_buf = get_tensor_buffer(&out)?;

    let params = LogicalBinaryParams {
        numel: numel as u32,
    };
    let params_buf = create_params_buffer(client, &params);

    launch_logical_or(
        client.pipeline_cache(),
        client.wgpu_queue(),
        &a_buf,
        &b_buf,
        &out_buf,
        &params_buf,
        numel,
    )?;

    Ok(out)
}

/// Native logical XOR operation.
#[allow(dead_code)]
pub(crate) fn native_logical_xor(
    client: &WgpuClient,
    a: &Tensor<WgpuRuntime>,
    b: &Tensor<WgpuRuntime>,
) -> Result<Tensor<WgpuRuntime>> {
    let a_contig = ensure_contiguous(a);
    let b_contig = ensure_contiguous(b);
    let numel = a.numel();

    let out = alloc_output(client, a.shape(), DType::U32);

    let a_buf = get_tensor_buffer(&a_contig)?;
    let b_buf = get_tensor_buffer(&b_contig)?;
    let out_buf = get_tensor_buffer(&out)?;

    let params = LogicalBinaryParams {
        numel: numel as u32,
    };
    let params_buf = create_params_buffer(client, &params);

    launch_logical_xor(
        client.pipeline_cache(),
        client.wgpu_queue(),
        &a_buf,
        &b_buf,
        &out_buf,
        &params_buf,
        numel,
    )?;

    Ok(out)
}

/// Native logical NOT operation.
#[allow(dead_code)]
pub(crate) fn native_logical_not(
    client: &WgpuClient,
    a: &Tensor<WgpuRuntime>,
) -> Result<Tensor<WgpuRuntime>> {
    let a_contig = ensure_contiguous(a);
    let numel = a.numel();

    let out = alloc_output(client, a.shape(), DType::U32);

    let a_buf = get_tensor_buffer(&a_contig)?;
    let out_buf = get_tensor_buffer(&out)?;

    let params = LogicalUnaryParams {
        numel: numel as u32,
    };
    let params_buf = create_params_buffer(client, &params);

    launch_logical_not(
        client.pipeline_cache(),
        client.wgpu_queue(),
        &a_buf,
        &out_buf,
        &params_buf,
        numel,
    )?;

    Ok(out)
}