blasoxide 0.3.2

BLAS implementation in 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
159
160
161

pub unsafe fn sgemm_16x4_packed(
    k: usize,
    a: *const f32,
    b: *const f32,
    beta: f32,
    c: *mut f32,
    ldc: usize,
) {
    #[cfg(target_arch = "x86_64")]
    {
        if is_x86_feature_detected!("fma") {
            crate::fma::sgemm_16x4_packed(k, a, b, beta, c, ldc);
            return;
        }
    }

    let cptr0 = c;
    let cptr1 = c.add(ldc);
    let cptr2 = c.add(ldc * 2);
    let cptr3 = c.add(ldc * 3);

    for i in 0..16 {
        let mut creg0 = *cptr0.add(i) * beta;
        let mut creg1 = *cptr1.add(i) * beta;
        let mut creg2 = *cptr2.add(i) * beta;
        let mut creg3 = *cptr3.add(i) * beta;

        let mut bptr = b;

        for p in 0..k {
            let areg = *a.add(i + p * 16);

            let breg0 = *bptr;
            let breg1 = *bptr.add(1);
            let breg2 = *bptr.add(2);
            let breg3 = *bptr.add(3);

            creg0 += breg0 * areg;
            creg1 += breg1 * areg;
            creg2 += breg2 * areg;
            creg3 += breg3 * areg;

            bptr = bptr.add(4);
        }

        *cptr0.add(i) = creg0;
        *cptr1.add(i) = creg1;
        *cptr2.add(i) = creg2;
        *cptr3.add(i) = creg3;
    }
}

pub unsafe fn s_pack_a(
    k: usize,
    alpha: f32,
    mut a: *const f32,
    lda: usize,
    mut packed_a: *mut f32,
) {
    #[cfg(target_arch = "x86_64")]
    {
        if is_x86_feature_detected!("fma") {
            crate::fma::s_pack_a(k, alpha, a, lda, packed_a);
            return;
        }
    }

    for _ in 0..k {
        let mut aptr = a;

        for _ in 0..16 {
            *packed_a = alpha * *aptr;

            aptr = aptr.add(1);
            packed_a = packed_a.add(1);
        }

        a = a.add(lda);
    }
}

pub unsafe fn dgemm_8x4_packed(
    k: usize,
    a: *const f64,
    b: *const f64,
    beta: f64,
    c: *mut f64,
    ldc: usize,
) {
    #[cfg(target_arch = "x86_64")]
    {
        if is_x86_feature_detected!("fma") {
            crate::fma::dgemm_8x4_packed(k, a, b, beta, c, ldc);
            return;
        }
    }

    let cptr0 = c;
    let cptr1 = c.add(ldc);
    let cptr2 = c.add(ldc * 2);
    let cptr3 = c.add(ldc * 3);

    for i in 0..8 {
        let mut creg0 = *cptr0.add(i) * beta;
        let mut creg1 = *cptr1.add(i) * beta;
        let mut creg2 = *cptr2.add(i) * beta;
        let mut creg3 = *cptr3.add(i) * beta;

        let mut bptr = b;

        for p in 0..k {
            let areg = *a.add(i + p * 8);

            let breg0 = *bptr;
            let breg1 = *bptr.add(1);
            let breg2 = *bptr.add(2);
            let breg3 = *bptr.add(3);

            creg0 += breg0 * areg;
            creg1 += breg1 * areg;
            creg2 += breg2 * areg;
            creg3 += breg3 * areg;

            bptr = bptr.add(4);
        }

        *cptr0.add(i) = creg0;
        *cptr1.add(i) = creg1;
        *cptr2.add(i) = creg2;
        *cptr3.add(i) = creg3;
    }
}

pub unsafe fn d_pack_a(
    k: usize,
    alpha: f64,
    mut a: *const f64,
    lda: usize,
    mut packed_a: *mut f64,
) {
    #[cfg(target_arch = "x86_64")]
    {
        if is_x86_feature_detected!("fma") {
            crate::fma::d_pack_a(k, alpha, a, lda, packed_a);
            return;
        }
    }

    for _ in 0..k {
        let mut aptr = a;

        for _ in 0..8 {
            *packed_a = alpha * *aptr;

            aptr = aptr.add(1);
            packed_a = packed_a.add(1);
        }

        a = a.add(lda);
    }
}