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
157
158

#![allow(unused)]

use cudarc::{
    cudnn::{ConvForward, Cudnn, CudnnDataType},
    driver::{CudaSlice, DeviceRepr},
};
use hpt_common::{error::shape::ShapeError, utils::conv_algos::ConvAlgo};
use hpt_traits::ops::{conv::CudaConvBatchNorm, creation::TensorCreator};
use hpt_traits::{
    ops::conv::{ConvBatchNorm, CudaConv},
    tensor::{CommonBounds, TensorInfo},
};
use hpt_types::{
    cuda_types::scalar::Scalar,
    dtype::CudaType,
    into_scalar::Cast,
    type_promote::{FloatOutBinary, FloatOutUnary, NormalOut},
};

use crate::{backends::common::conv::cal_conv2d_output_shape, tensor_base::_Tensor, Tensor, CUDNN};
use hpt_allocator::{
    traits::{Allocator, AllocatorOutputRetrive},
    Cuda,
};

impl<T, const DEVICE: usize, Al> CudaConv<T> for Tensor<T, Cuda, DEVICE, Al>
where
    T: CommonBounds + DeviceRepr + CudaType + CudnnDataType,
    bool: Cast<T>,
    Al: Allocator + Send + Sync,
    Al::Output: AllocatorOutputRetrive,
    Scalar<T>: NormalOut<Scalar<T>, Output = Scalar<T>>,
{
    type Output = Tensor<T, Cuda, DEVICE, Al>;

    fn conv2d(
        &self,
        kernels: &Self::Output,
        bias: Option<&Self::Output>,
        steps: [i64; 2],
        padding: [i64; 2],
        dilation: [i64; 2],
        algo: Option<ConvAlgo>,
    ) -> Result<Self::Output, hpt_common::error::base::TensorError> {
        Ok(self
            .inner
            .conv2d(
                kernels.inner.as_ref(),
                bias.map(|b| b.inner.as_ref()),
                steps,
                padding,
                dilation,
                algo,
            )?
            .into())
    }

    fn conv2d_group(
        &self,
        kernels: &Self::Output,
        bias: Option<&Self::Output>,
        steps: [i64; 2],
        padding: [i64; 2],
        dilation: [i64; 2],
        groups: i64,
        algo: Option<ConvAlgo>,
    ) -> Result<Self::Output, hpt_common::error::base::TensorError> {
        Ok(self
            .inner
            .conv2d_group(
                kernels.inner.as_ref(),
                bias.map(|b| b.inner.as_ref()),
                steps,
                padding,
                dilation,
                groups,
                algo,
            )?
            .into())
    }

    fn dwconv2d(
        &self,
        kernels: &Self::Output,
        bias: Option<&Self::Output>,
        steps: [i64; 2],
        padding: [i64; 2],
        dilation: [i64; 2],
        algo: Option<ConvAlgo>,
    ) -> Result<Self::Output, hpt_common::error::base::TensorError> {
        Ok(self
            .inner
            .dwconv2d(
                kernels.inner.as_ref(),
                bias.map(|b| b.inner.as_ref()),
                steps,
                padding,
                dilation,
                algo,
            )?
            .into())
    }

    fn conv2d_transpose(
        &self,
        kernels: &Self::Output,
        steps: [i64; 2],
        padding: [(i64, i64); 2],
        output_padding: [i64; 2],
        dilation: [i64; 2],
    ) -> Result<Self::Output, hpt_common::error::base::TensorError> {
        unimplemented!()
    }
}

impl<T, const DEVICE: usize, A> CudaConvBatchNorm<T> for Tensor<T, Cuda, DEVICE, A>
where
    T: CommonBounds + DeviceRepr + CudaType + CudnnDataType,
    T::Vec: FloatOutBinary<Output = T::Vec> + FloatOutUnary<Output = T::Vec>,
    T: FloatOutBinary<Output = T> + FloatOutUnary<Output = T>,
    Scalar<T>: NormalOut<Scalar<T>, Output = Scalar<T>>,
    bool: Cast<T>,
    A: Allocator + Send + Sync,
    A::Output: AllocatorOutputRetrive,
{
    type Output = Tensor<T, Cuda, DEVICE, A>;
    fn batchnorm_conv2d(
        &self,
        kernels: &Self::Output,
        mean: &Self::Output,
        var: &Self::Output,
        gamma: &Self::Output,
        beta: &Self::Output,
        bias: Option<&Self::Output>,
        eps: T,
        steps: [i64; 2],
        padding: [i64; 2],
        dilation: [i64; 2],
        algo: Option<ConvAlgo>,
    ) -> Result<Self::Output, hpt_common::error::base::TensorError> {
        Ok(self
            .inner
            .batchnorm_conv2d(
                kernels.inner.as_ref(),
                mean.inner.as_ref(),
                var.inner.as_ref(),
                gamma.inner.as_ref(),
                beta.inner.as_ref(),
                bias.map(|b| b.inner.as_ref()),
                eps,
                steps,
                padding,
                dilation,
                algo,
            )?
            .into())
    }
}