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

//! Reduction operations for WebGPU runtime

use crate::error::Result;
use crate::ops::{AccumulationPrecision, ReduceOps};
use crate::runtime::wgpu::WgpuClient;
use crate::runtime::wgpu::WgpuRuntime;
use crate::runtime::wgpu::ops::native::native_reduce_op;
use crate::tensor::Tensor;

impl ReduceOps<WgpuRuntime> for WgpuClient {
    fn sum(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
    ) -> Result<Tensor<WgpuRuntime>> {
        native_reduce_op(self, "sum", a, dims, keepdim)
    }

    fn sum_with_precision(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
        _precision: AccumulationPrecision,
    ) -> Result<Tensor<WgpuRuntime>> {
        // WebGPU doesn't support accumulation precision control
        // Just delegate to standard sum
        native_reduce_op(self, "sum", a, dims, keepdim)
    }

    fn mean(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
    ) -> Result<Tensor<WgpuRuntime>> {
        native_reduce_op(self, "mean", a, dims, keepdim)
    }

    fn max(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
    ) -> Result<Tensor<WgpuRuntime>> {
        native_reduce_op(self, "max", a, dims, keepdim)
    }

    fn max_with_precision(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
        _precision: AccumulationPrecision,
    ) -> Result<Tensor<WgpuRuntime>> {
        // WebGPU doesn't support accumulation precision control
        // Just delegate to standard max
        native_reduce_op(self, "max", a, dims, keepdim)
    }

    fn min(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
    ) -> Result<Tensor<WgpuRuntime>> {
        native_reduce_op(self, "min", a, dims, keepdim)
    }

    fn min_with_precision(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
        _precision: AccumulationPrecision,
    ) -> Result<Tensor<WgpuRuntime>> {
        // WebGPU doesn't support accumulation precision control
        // Just delegate to standard min
        native_reduce_op(self, "min", a, dims, keepdim)
    }

    fn prod(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
    ) -> Result<Tensor<WgpuRuntime>> {
        native_reduce_op(self, "prod", a, dims, keepdim)
    }

    fn prod_with_precision(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
        _precision: AccumulationPrecision,
    ) -> Result<Tensor<WgpuRuntime>> {
        // WebGPU doesn't support accumulation precision control
        // Just delegate to standard prod
        native_reduce_op(self, "prod", a, dims, keepdim)
    }

    fn any(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
    ) -> Result<Tensor<WgpuRuntime>> {
        native_reduce_op(self, "any", a, dims, keepdim)
    }

    fn all(
        &self,
        a: &Tensor<WgpuRuntime>,
        dims: &[usize],
        keepdim: bool,
    ) -> Result<Tensor<WgpuRuntime>> {
        native_reduce_op(self, "all", a, dims, keepdim)
    }
}