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

//! `i686`'s Streaming SIMD Extensions 4a (`SSE4a`)

use crate::{
    core_arch::{simd::*, x86::*},
    mem::transmute,
};

#[cfg(test)]
use stdarch_test::assert_instr;

#[allow(improper_ctypes)]
extern "C" {
    #[link_name = "llvm.x86.sse4a.extrq"]
    fn extrq(x: i64x2, y: i8x16) -> i64x2;
    #[link_name = "llvm.x86.sse4a.insertq"]
    fn insertq(x: i64x2, y: i64x2) -> i64x2;
    #[link_name = "llvm.x86.sse4a.movnt.sd"]
    fn movntsd(x: *mut f64, y: __m128d);
    #[link_name = "llvm.x86.sse4a.movnt.ss"]
    fn movntss(x: *mut f32, y: __m128);
}

// FIXME(blocked on #248): _mm_extracti_si64(x, len, idx) // EXTRQ
// FIXME(blocked on #248): _mm_inserti_si64(x, y, len, idx) // INSERTQ

/// Extracts the bit range specified by `y` from the lower 64 bits of `x`.
///
/// The `[13:8]` bits of `y` specify the index of the bit-range to extract. The
/// `[5:0]` bits of `y` specify the length of the bit-range to extract. All
/// other bits are ignored.
///
/// If the length is zero, it is interpreted as `64`. If the length and index
/// are zero, the lower 64 bits of `x` are extracted.
///
/// If `length == 0 && index > 0` or `length + index > 64` the result is
/// undefined.
#[inline]
#[target_feature(enable = "sse4a")]
#[cfg_attr(test, assert_instr(extrq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_extract_si64(x: __m128i, y: __m128i) -> __m128i {
    transmute(extrq(x.as_i64x2(), y.as_i8x16()))
}

/// Inserts the `[length:0]` bits of `y` into `x` at `index`.
///
/// The bits of `y`:
///
/// - `[69:64]` specify the `length`,
/// - `[77:72]` specify the index.
///
/// If the `length` is zero it is interpreted as `64`. If `index + length > 64`
/// or `index > 0 && length == 0` the result is undefined.
#[inline]
#[target_feature(enable = "sse4a")]
#[cfg_attr(test, assert_instr(insertq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_insert_si64(x: __m128i, y: __m128i) -> __m128i {
    transmute(insertq(x.as_i64x2(), y.as_i64x2()))
}

/// Non-temporal store of `a.0` into `p`.
///
/// Writes 64-bit data to a memory location without polluting the caches.
#[inline]
#[target_feature(enable = "sse4a")]
#[cfg_attr(test, assert_instr(movntsd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_stream_sd(p: *mut f64, a: __m128d) {
    movntsd(p, a);
}

/// Non-temporal store of `a.0` into `p`.
///
/// Writes 32-bit data to a memory location without polluting the caches.
#[inline]
#[target_feature(enable = "sse4a")]
#[cfg_attr(test, assert_instr(movntss))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_stream_ss(p: *mut f32, a: __m128) {
    movntss(p, a);
}

#[cfg(test)]
mod tests {
    use crate::core_arch::x86::*;
    use stdarch_test::simd_test;

    #[simd_test(enable = "sse4a")]
    unsafe fn test_mm_extract_si64() {
        let b = 0b0110_0000_0000_i64;
        //        ^^^^ bit range extracted
        let x = _mm_setr_epi64x(b, 0);
        let v = 0b001000___00___000100_i64;
        //        ^idx: 2^3 = 8 ^length = 2^2 = 4
        let y = _mm_setr_epi64x(v, 0);
        let e = _mm_setr_epi64x(0b0110_i64, 0);
        let r = _mm_extract_si64(x, y);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4a")]
    unsafe fn test_mm_insert_si64() {
        let i = 0b0110_i64;
        //        ^^^^ bit range inserted
        let z = 0b1010_1010_1010i64;
        //        ^^^^ bit range replaced
        let e = 0b0110_1010_1010i64;
        //        ^^^^ replaced 1010 with 0110
        let x = _mm_setr_epi64x(z, 0);
        let expected = _mm_setr_epi64x(e, 0);
        let v = 0b001000___00___000100_i64;
        //        ^idx: 2^3 = 8 ^length = 2^2 = 4
        let y = _mm_setr_epi64x(i, v);
        let r = _mm_insert_si64(x, y);
        assert_eq_m128i(r, expected);
    }

    #[repr(align(16))]
    struct MemoryF64 {
        data: [f64; 2],
    }

    #[simd_test(enable = "sse4a")]
    unsafe fn test_mm_stream_sd() {
        let mut mem = MemoryF64 {
            data: [1.0_f64, 2.0],
        };
        {
            let vals = &mut mem.data;
            let d = vals.as_mut_ptr();

            let x = _mm_setr_pd(3.0, 4.0);

            _mm_stream_sd(d, x);
        }
        assert_eq!(mem.data[0], 3.0);
        assert_eq!(mem.data[1], 2.0);
    }

    #[repr(align(16))]
    struct MemoryF32 {
        data: [f32; 4],
    }

    #[simd_test(enable = "sse4a")]
    unsafe fn test_mm_stream_ss() {
        let mut mem = MemoryF32 {
            data: [1.0_f32, 2.0, 3.0, 4.0],
        };
        {
            let vals = &mut mem.data;
            let d = vals.as_mut_ptr();

            let x = _mm_setr_ps(5.0, 6.0, 7.0, 8.0);

            _mm_stream_ss(d, x);
        }
        assert_eq!(mem.data[0], 5.0);
        assert_eq!(mem.data[1], 2.0);
        assert_eq!(mem.data[2], 3.0);
        assert_eq!(mem.data[3], 4.0);
    }
}