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

use crate::Matrix;
use custos::{get_device, number::Number, CDatatype, Cache, CPU};

#[cfg(feature = "opencl")]
use super::{cl_to_cpu_s, cl_to_cpu_scalar};
#[cfg(feature = "opencl")]
use custos::CLDevice;

#[cfg(feature = "cuda")]
use crate::{cu_to_cpu_s, cu_to_cpu_scalar};
#[cfg(feature = "cuda")]
use custos::CudaDevice;

impl<'a, T: CDatatype> Matrix<'a, T> {
    pub fn sum(&self) -> T {
        get_device!(self.device(), SumOps<T>).sum(self)
    }

    pub fn mean(&self) -> T {
        get_device!(self.device(), SumOps<T>).mean(self)
    }

    pub fn sum_rows(&self) -> Matrix<'a, T> {
        get_device!(self.device(), SumOps<T>).sum_rows(self)
    }

    pub fn sum_cols(&self) -> Matrix<'a, T> {
        get_device!(self.device(), SumOps<T>).sum_cols(self)
    }
}

pub trait SumOps<T> {
    fn sum(&self, x: &Matrix<T>) -> T;
    fn mean(&self, x: &Matrix<T>) -> T;
    fn sum_rows(&self, x: &Matrix<T>) -> Matrix<T>;
    fn sum_cols(&self, x: &Matrix<T>) -> Matrix<T>;
}

impl<T: Number> SumOps<T> for CPU {
    fn sum(&self, x: &Matrix<T>) -> T {
        x.iter().copied().sum()
        /*let mut sum = T::default();
        for value in x.as_slice() {
            sum += *value;
        }
        sum*/
    }

    fn mean(&self, x: &Matrix<T>) -> T {
        let sum = self.sum(x);
        sum / T::from_usize(x.size())
    }

    fn sum_rows(&self, x: &Matrix<T>) -> Matrix<T> {
        let mut out = Cache::get(self, x.cols(), x.node.idx);

        let data = x.as_slice();
        let sum_slice = out.as_mut_slice();

        for value in sum_slice.iter_mut() {
            *value = T::default();
        }

        for idx in 0..x.rows() {
            let index = idx * x.cols();
            let row = &data[index..index + x.cols()];

            for (i, value) in row.iter().enumerate() {
                sum_slice[i] += *value;
            }
        }
        (out, 1, x.cols()).into()
    }

    fn sum_cols(&self, x: &Matrix<T>) -> Matrix<T> {
        let mut out = Cache::get(self, x.rows(), x.node.idx);

        let data = x.as_slice();
        let sum_slice = out.as_mut_slice();

        for (idx, col_vec_value) in sum_slice.iter_mut().enumerate().take(x.rows()) {
            let index = idx * x.cols();
            let row = &data[index..index + x.cols()];
            let mut sum = T::default();

            for data in row {
                sum += *data;
            }
            *col_vec_value = sum;
        }
        (out, x.rows(), 1).into()
    }
}

#[cfg(feature = "opencl")]
impl<T: CDatatype> SumOps<T> for CLDevice {
    fn sum(&self, x: &Matrix<T>) -> T {
        cl_to_cpu_scalar(self, x, |device, x| device.sum(x))
    }

    fn mean(&self, x: &Matrix<T>) -> T {
        cl_to_cpu_scalar(self, x, |device, x| device.mean(x))
    }

    fn sum_rows<'a>(&'a self, x: &Matrix<T>) -> Matrix<'a, T> {
        cl_to_cpu_s(self, x, |device, x| device.sum_rows(x))
    }

    fn sum_cols(&self, x: &Matrix<T>) -> Matrix<T> {
        cl_to_cpu_s(self, x, |device, x| device.sum_cols(x))
    }
}

#[cfg(feature = "cuda")]
impl<T: CDatatype> SumOps<T> for CudaDevice {
    fn sum(&self, x: &Matrix<T>) -> T {
        cu_to_cpu_scalar(self, x, |device, x| device.sum(&x))
    }

    fn mean(&self, x: &Matrix<T>) -> T {
        cu_to_cpu_scalar(self, x, |device, x| device.mean(&x))
    }

    fn sum_rows(&self, x: &Matrix<T>) -> Matrix<T> {
        cu_to_cpu_s(self, x, |device, x| device.sum_rows(&x))
    }

    fn sum_cols(&self, x: &Matrix<T>) -> Matrix<T> {
        cu_to_cpu_s(self, x, |device, x| device.sum_cols(&x))
    }
}