x86intrin 0.4.5

x86intrin implementation
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
162
163
164
165
166
167
168
169
170
171
172
173
174

#![allow(improper_ctypes)]  // TODO(mayah): Remove this flag

use super::*;
use super::{simd_shuffle2, simd_shuffle4};

extern "platform-intrinsic" {
    fn x86_mm_addsub_pd(x: m128d, y: m128d) -> m128d;
    fn x86_mm_addsub_ps(x: m128, y: m128) -> m128;
    fn x86_mm_hadd_pd(x: m128d, y: m128d) -> m128d;
    fn x86_mm_hadd_ps(x: m128, y: m128) -> m128;
    fn x86_mm_hsub_pd(x: m128d, y: m128d) -> m128d;
    fn x86_mm_hsub_ps(x: m128, y: m128) -> m128;
}

extern {
    #[link_name = "llvm.x86.sse3.ldu.dq"]
    fn sse3_ldu_dq(a: *mut u8) -> i8x16;
}

// addsubpd
// __m128d _mm_addsub_pd (__m128d a, __m128d b)
#[inline]
pub fn mm_addsub_pd(a: m128d, b: m128d) -> m128d {
    unsafe { x86_mm_addsub_pd(a, b) }
}

// addsubps
// __m128 _mm_addsub_ps (__m128 a, __m128 b)
#[inline]
pub fn mm_addsub_ps(a: m128, b: m128) -> m128 {
    unsafe { x86_mm_addsub_ps(a, b) }
}

// haddpd
// __m128d _mm_hadd_pd (__m128d a, __m128d b)
#[inline]
pub fn mm_hadd_pd(a: m128d, b: m128d) -> m128d {
    unsafe { x86_mm_hadd_pd(a, b) }
}

// haddps
// __m128 _mm_hadd_ps (__m128 a, __m128 b)
#[inline]
pub fn mm_hadd_ps(a: m128, b: m128) -> m128 {
    unsafe { x86_mm_hadd_ps(a, b) }
}

// hsubpd
// __m128d _mm_hsub_pd (__m128d a, __m128d b)
#[inline]
pub fn mm_hsub_pd(a: m128d, b: m128d) -> m128d {
    unsafe { x86_mm_hsub_pd(a, b) }
}

// hsubps
// __m128 _mm_hsub_ps (__m128 a, __m128 b)
#[inline]
pub fn mm_hsub_ps(a: m128, b: m128) -> m128 {
    unsafe { x86_mm_hsub_ps(a, b) }
}

// lddqu
// __m128i _mm_lddqu_si128 (__m128i const* mem_addr)
#[inline]
pub unsafe fn mm_lddqu_si128(mem_addr: *const m128i) -> m128i {
    sse3_ldu_dq(mem_addr as *mut u8).as_m128i()
}

// movddup
// __m128d _mm_loaddup_pd (double const* mem_addr)
#[inline]
pub unsafe fn mm_loaddup_pd(mem_addr: *const f64) ->  m128d {
    mm_load1_pd(mem_addr)
}

// movddup
// __m128d _mm_movedup_pd (__m128d a)
#[inline]
pub fn mm_movedup_pd(a: m128d) -> m128d {
    let a64 = a.as_f64x2();
    unsafe { simd_shuffle2(a64, a64, [0, 0]) }
}

// movshdup
// __m128 _mm_movehdup_ps (__m128 a)
#[inline]
pub fn mm_movehdup_ps(a: m128) -> m128 {
    let a32 = a.as_f32x4();
    unsafe { simd_shuffle4(a32, a32, [1, 1, 3, 3]) }
}

// movsldup
// __m128 _mm_moveldup_ps (__m128 a)
#[inline]
pub fn mm_moveldup_ps(a: m128) -> m128 {
    let a32 = a.as_f32x4();
    unsafe { simd_shuffle4(a32, a32, [0, 0, 2, 2]) }
}

// #define _MM_DENORMALS_ZERO_ON   (0x0040)
pub const MM_DENORMALS_ZERO_ON: u32 = 0x0040;
// #define _MM_DENORMALS_ZERO_OFF  (0x0000)
pub const MM_DENORMALS_ZERO_OFF: u32 = 0x0000;
// #define _MM_DENORMALS_ZERO_MASK (0x0040)
pub const MM_DENORMALS_ZERO_MASK: u32 = 0x0040;

// #define _MM_GET_DENORMALS_ZERO_MODE() (_mm_getcsr() & _MM_DENORMALS_ZERO_MASK)
#[inline]
pub fn mm_get_denormals_zero_mode() -> u32 {
    mm_getcsr() & MM_DENORMALS_ZERO_MASK
}

// #define _MM_SET_DENORMALS_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_DENORMALS_ZERO_MASK) | (x)))
#[inline]
pub fn mm_set_denormals_zero_mode(x: u32) {
    mm_setcsr((mm_getcsr() & !MM_DENORMALS_ZERO_MASK) | x)
}

#[cfg(test)]
mod test {
    use super::super::*;

    #[test]
    fn test_arith_ps() {
        let x = mm_setr_ps(1.0, 2.0, 3.0, 4.0);
        let y = mm_setr_ps(2.0, 1.0, 2.0, 4.0);

        assert_eq!(mm_addsub_ps(x, y).as_f32x4().as_array(), [-1.0, 3.0, 1.0, 8.0]);
        assert_eq!(mm_hadd_ps(x, y).as_f32x4().as_array(), [3.0, 7.0, 3.0, 6.0]);
        assert_eq!(mm_hsub_ps(x, y).as_f32x4().as_array(), [-1.0, -1.0, 1.0, -2.0]);
    }

    #[test]
    fn test_arith_pd() {
        let x = mm_setr_pd(1.0, 2.0);
        let y = mm_setr_pd(3.0, 5.0);

        assert_eq!(mm_addsub_pd(x, y).as_f64x2().as_array(), [-2.0, 7.0]);
        assert_eq!(mm_hadd_pd(x, y).as_f64x2().as_array(), [3.0, 8.0]);
        assert_eq!(mm_hsub_pd(x, y).as_f64x2().as_array(), [-1.0, -2.0]);
    }

    #[test]
    fn test_move_pd() {
        let x = mm_setr_pd(1.0, 2.0);
        assert_eq!(mm_movedup_pd(x).as_f64x2().as_array(), [1.0, 1.0]);
    }

    #[test]
    fn test_move_ps() {
        let x = mm_setr_ps(1.0, 2.0, 3.0, 4.0);
        assert_eq!(mm_movehdup_ps(x).as_f32x4().as_array(), [2.0, 2.0, 4.0, 4.0]);
        assert_eq!(mm_moveldup_ps(x).as_f32x4().as_array(), [1.0, 1.0, 3.0, 3.0]);
    }

    #[test]
    fn test_lddqu() {
        let x = mm_setr_epi32(1, 2, 3, 4);
        let p = &x as *const m128i;

        let r = unsafe { mm_lddqu_si128(p) };
        assert_eq!(r.as_i32x4().as_array(), [1, 2, 3, 4]);
    }

    #[test]
    fn test_denormal() {
        let initial = mm_get_denormals_zero_mode();
        mm_set_denormals_zero_mode(MM_DENORMALS_ZERO_ON);
        assert_eq!(mm_get_denormals_zero_mode(), MM_DENORMALS_ZERO_ON);
        mm_set_denormals_zero_mode(MM_DENORMALS_ZERO_OFF);
        assert_eq!(mm_get_denormals_zero_mode(), MM_DENORMALS_ZERO_OFF);
        mm_set_denormals_zero_mode(initial);
    }
}