hpt 0.1.3

High Performance Tensor (HPT) - A fast, efficient, and user-friendly tensor computation library for 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
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

use crate::backends::cuda::{
    cuda_utils::get_module_name_1, utils::unary::unary::uary_fn_with_out_simd,
};
use crate::{backend::Cuda, tensor_base::_Tensor};
use cudarc::driver::DeviceRepr;
use hpt_allocator::traits::{Allocator, AllocatorOutputRetrive};
use hpt_common::error::base::TensorError;
use hpt_traits::ops::creation::TensorCreator;
use hpt_traits::ops::windows::WindowOps;
use hpt_traits::tensor::CommonBounds;
use hpt_types::cuda_types::scalar::Scalar;
use hpt_types::dtype::CudaType;
use hpt_types::{
    dtype::FloatConst,
    into_scalar::Cast,
    type_promote::{FloatOutBinary, FloatOutUnary},
};

impl<T, const DEVICE_ID: usize, Al> WindowOps for _Tensor<T, Cuda, DEVICE_ID, Al>
where
    f64: Cast<T>,
    T: CommonBounds
        + FloatOutBinary<Output = T>
        + FloatOutUnary<Output = T>
        + FloatConst
        + DeviceRepr
        + CudaType,
    usize: Cast<T>,
    i64: Cast<T>,
    Al: Allocator,
    Al::Output: AllocatorOutputRetrive,
{
    type Output = _Tensor<T, Cuda, DEVICE_ID, Al>;
    type Meta = T;
    #[track_caller]
    fn hamming_window(
        window_length: i64,
        periodic: bool,
    ) -> std::result::Result<Self::Output, TensorError> {
        __hamming_window::<T, DEVICE_ID, Al>(window_length, (0.54).cast(), (0.46).cast(), periodic)
    }

    #[track_caller]
    fn hann_window(
        window_length: i64,
        periodic: bool,
    ) -> std::result::Result<Self::Output, TensorError> {
        __hamming_window::<T, DEVICE_ID, Al>(window_length, (0.5).cast(), (0.5).cast(), periodic)
    }

    #[track_caller]
    fn blackman_window(
        window_length: i64,
        periodic: bool,
    ) -> std::result::Result<Self::Output, TensorError> {
        let length_usize = if periodic {
            window_length
        } else {
            window_length - 1
        };
        let length: T = length_usize.cast();
        let ret = _Tensor::<T, Cuda, DEVICE_ID, Al>::empty(&[length_usize])?;
        uary_fn_with_out_simd(
            &ret,
            &get_module_name_1("blackman_window", &ret),
            |out, idx| {
                let res = match T::STR {
                    "f32" => {
                        format!(
                            "
                            float n = (float){idx};
                            float w1 = 2.0f * M_PI * n / {length}f;
                            float w2 = 2.0f * w1;  // 4π * n / (N-1)
                            {out} = 0.42f - 0.5f * cosf(w1) + 0.08f * cosf(w2);"
                        )
                    }
                    "f64" => {
                        format!(
                            "
                            double n = (double){idx};
                            double w1 = 2.0 * M_PI * n / {length};
                            double w2 = 2.0 * w1;
                            {out} = 0.42 - 0.5 * cos(w1) + 0.08 * cos(w2);"
                        )
                    }
                    "f16" => {
                        format!(
                            "
                            float n = (float){idx};
                            float w1 = 2.0f * M_PI * n / {length}f;
                            float w2 = 2.0f * w1;
                            {out} = __float2half(0.42f - 0.5f * cosf(w1) + 0.08f * cosf(w2));"
                        )
                    }
                    _ => unreachable!(),
                };
                Scalar::<T>::new(res)
            },
            None::<Self::Output>,
        )
    }
}

#[track_caller]
fn __hamming_window<T, const DEVICE_ID: usize, Al>(
    window_length: i64,
    alpha: T,
    beta: T,
    periodic: bool,
) -> std::result::Result<_Tensor<T, Cuda, DEVICE_ID, Al>, TensorError>
where
    T: CommonBounds + DeviceRepr + CudaType,
    usize: Cast<T>,
    Al: Allocator,
    Al::Output: AllocatorOutputRetrive,
{
    let length_usize = (if periodic {
        window_length
    } else {
        window_length - 1
    }) as usize;
    let length: T = length_usize.cast();
    let ret = _Tensor::<T, Cuda, DEVICE_ID, Al>::empty(&[length_usize as i64])?;
    uary_fn_with_out_simd(
        &ret,
        &get_module_name_1("hamming_window", &ret),
        |out, idx| {
            let res = match T::STR {
                "f32" => {
                    format!(
                        "
                    float n = (float){idx};
                    {out} = {alpha} - {beta} * cosf(2.0f * M_PI * n / {length});"
                    )
                }
                "f64" => {
                    format!(
                        "
                    double n = (double){idx};
                    {out} = {alpha} - {beta} * cos(2.0 * M_PI * n / {length});"
                    )
                }
                "f16" => {
                    format!(
                        "
                    float n = (float){idx};
                    {out} = __float2half({alpha}f - {beta}f * cosf(2.0f * M_PI * n / {length}));"
                    )
                }
                _ => unreachable!(),
            };
            Scalar::<T>::new(res)
        },
        None::<_Tensor<T, Cuda, DEVICE_ID, Al>>,
    )
}