rustlane 0.1.0

ISPC-style SPMD programming model for Rust: natural control flow lowered to masked SIMD by proc macros over nightly std::simd
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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228

//! Varying RNG — a bit-exact port of ISPC's stdlib pseudo-random generator
//! (the L'Ecuyer LFSR113 / combined-Tausworthe generator that the `ao`
//! benchmark's sample distribution depends on).
//!
//! Ported verbatim from the ISPC standard library:
//!   - `random` / `frandom` / `seed_rng`: `stdlib/stdlib.ispc`, lines
//!     7005-7057 (the `varying RNGState *uniform` overloads).
//!   - `struct RNGState { unsigned int z1, z2, z3, z4; }`:
//!     `stdlib/include/core.isph`, lines 395-396.
//! Source repo: github.com/ispc/ispc, commit
//! `e99a37840cd7d83c84e56e97a03eab6049b59fe7` (main HEAD 2026-07-23).
//! License of the ported material: BSD-3-Clause, Copyright (c) Intel
//! Corporation (ISPC stdlib). See THIRD-PARTY.md.
//!
//! Every lane carries an independent `(z1,z2,z3,z4)` state; the arithmetic is
//! `unsigned int` (u32) throughout, matching ISPC's wrapping shifts and
//! logical right-shifts exactly, so identical seeds produce identical streams
//! bit-for-bit against ISPC (and against the `N = 1` scalar instantiation).

use crate::varying::Varying;
use core::simd::num::SimdFloat;
use core::simd::{LaneCount, Simd, SupportedLaneCount};

/// Per-lane RNG state — four `u32` Tausworthe components, one set per lane.
/// Mirrors ISPC's `struct RNGState { unsigned int z1, z2, z3, z4; }`.
#[derive(Copy, Clone, Debug, PartialEq)]
pub struct RNGState<const N: usize>
where
    LaneCount<N>: SupportedLaneCount,
{
    pub z1: Varying<u32, N>,
    pub z2: Varying<u32, N>,
    pub z3: Varying<u32, N>,
    pub z4: Varying<u32, N>,
}

impl<const N: usize> RNGState<N>
where
    LaneCount<N>: SupportedLaneCount,
{
    /// Construct and seed in one step (`seed_rng` on a fresh state).
    #[inline(always)]
    pub fn new(seeds: Varying<u32, N>) -> Self {
        let mut s = RNGState {
            z1: Varying::splat(0),
            z2: Varying::splat(0),
            z3: Varying::splat(0),
            z4: Varying::splat(0),
        };
        s.seed_rng(seeds);
        s
    }

    /// Port of ISPC `seed_rng(varying RNGState *uniform, unsigned int seed)`.
    #[inline(always)]
    pub fn seed_rng(&mut self, seed: Varying<u32, N>) {
        self.z1 = seed;
        self.z2 = seed ^ 0xbeeff00d_u32;
        self.z3 = ((seed & 0x0000ffff_u32) << 16u32) | (seed >> 16u32);
        self.z4 = ((seed & 0x000000ff_u32) << 24u32)
            | ((seed & 0x0000ff00_u32) << 8u32)
            | ((seed & 0x00ff0000_u32) >> 8u32)
            | ((seed & 0xff000000_u32) >> 24u32);
    }

    /// Port of ISPC `random(varying RNGState *uniform)`: advances the state
    /// and returns the combined 32-bit output per lane.
    #[inline(always)]
    pub fn random_u32(&mut self) -> Varying<u32, N> {
        let mut z1 = self.z1;
        let mut z2 = self.z2;
        let mut z3 = self.z3;
        let mut z4 = self.z4;

        let mut b = ((z1 << 6u32) ^ z1) >> 13u32;
        z1 = ((z1 & 0xfffffffe_u32) << 18u32) ^ b;
        b = ((z2 << 2u32) ^ z2) >> 27u32;
        z2 = ((z2 & 0xfffffff8_u32) << 2u32) ^ b;
        b = ((z3 << 13u32) ^ z3) >> 21u32;
        z3 = ((z3 & 0xfffffff0_u32) << 7u32) ^ b;
        b = ((z4 << 3u32) ^ z4) >> 12u32;
        z4 = ((z4 & 0xffffff80_u32) << 13u32) ^ b;

        self.z1 = z1;
        self.z2 = z2;
        self.z3 = z3;
        self.z4 = z4;
        z1 ^ z2 ^ z3 ^ z4
    }

    /// Port of ISPC `frandom(varying RNGState *uniform)`: a `f32` in `[0, 1)`.
    /// Takes the low 23 bits of `random_u32`, injects them into the mantissa
    /// of `1.0`, and subtracts 1.0 (identical bit layout to ISPC's
    /// `floatbits(0x3F800000 | irand) - 1.0f`).
    #[inline(always)]
    pub fn frandom(&mut self) -> Varying<f32, N> {
        let irand = self.random_u32() & 0x007fffff_u32;
        let bits = irand | 0x3f800000_u32;
        let f = <Simd<f32, N> as SimdFloat>::from_bits(bits.0);
        Varying(f) - 1.0f32
    }
}

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

    const VECTORS: &[(u32, [u32; 4], [u32; 3])] = &[
        (0x1, [1, 3203395596, 65536, 16777216], [0xfb3f5128, 0xbcff80e6, 0xbbb4200a]),
        (0x2a, [42, 3203395623, 2752512, 704643072], [0xee106289, 0x9aff5462, 0xee091f2b]),
        (
            0x12345678,
            [305419896, 2900076149, 1450709556, 2018915346],
            [0x1008d415, 0x58c11c4d, 0x8ebb09ea],
        ),
        (0x7, [7, 3203395594, 458752, 117440512], [0xf8243728, 0x50fcb1be, 0x7904ef6a]),
        (0x0, [0, 3203395597, 0, 0], [0xfbbfc028, 0xeeff00a2, 0xbbfc028a]),
    ];

    struct ScalarRng {
        z: [u32; 4],
    }
    impl ScalarRng {
        fn seed(seed: u32) -> Self {
            let z1 = seed;
            let z2 = seed ^ 0xbeeff00d;
            let z3 = ((seed & 0xffff) << 16) | (seed >> 16);
            let z4 = ((seed & 0xff) << 24)
                | ((seed & 0xff00) << 8)
                | ((seed & 0xff0000) >> 8)
                | ((seed & 0xff000000) >> 24);
            ScalarRng { z: [z1, z2, z3, z4] }
        }
        fn next(&mut self) -> u32 {
            let [mut z1, mut z2, mut z3, mut z4] = self.z;
            let mut b = ((z1 << 6) ^ z1) >> 13;
            z1 = ((z1 & 0xfffffffe) << 18) ^ b;
            b = ((z2 << 2) ^ z2) >> 27;
            z2 = ((z2 & 0xfffffff8) << 2) ^ b;
            b = ((z3 << 13) ^ z3) >> 21;
            z3 = ((z3 & 0xfffffff0) << 7) ^ b;
            b = ((z4 << 3) ^ z4) >> 12;
            z4 = ((z4 & 0xffffff80) << 13) ^ b;
            self.z = [z1, z2, z3, z4];
            z1 ^ z2 ^ z3 ^ z4
        }
    }

    fn frand_ref(u: u32) -> f32 {
        f32::from_bits(0x3f80_0000 | (u & 0x007f_ffff)) - 1.0
    }

    #[test]
    fn seed_matches_ispc_state() {
        for &(seed, state, _) in VECTORS {
            let s = RNGState::<8>::new(Varying::splat(seed));
            assert_eq!(s.z1.to_array(), [state[0]; 8], "z1 seed={seed:#x}");
            assert_eq!(s.z2.to_array(), [state[1]; 8], "z2 seed={seed:#x}");
            assert_eq!(s.z3.to_array(), [state[2]; 8], "z3 seed={seed:#x}");
            assert_eq!(s.z4.to_array(), [state[3]; 8], "z4 seed={seed:#x}");
        }
    }

    #[test]
    fn random_u32_bit_parity_n8() {
        for &(seed, _, outs) in VECTORS {
            let mut s = RNGState::<8>::new(Varying::splat(seed));
            for &expect in &outs {
                let got = s.random_u32();
                assert_eq!(got.to_array(), [expect; 8], "seed={seed:#x} expect={expect:#010x}");
            }
        }
    }

    #[test]
    fn random_u32_bit_parity_n1() {
        for &(seed, _, outs) in VECTORS {
            let mut s = RNGState::<1>::new(Varying::splat(seed));
            for &expect in &outs {
                assert_eq!(s.random_u32().to_array(), [expect], "seed={seed:#x}");
            }
        }
    }

    #[test]
    fn n1_agrees_with_n8_and_scalar() {
        for &(seed, _, _) in VECTORS {
            let mut s1 = RNGState::<1>::new(Varying::splat(seed));
            let mut s8 = RNGState::<8>::new(Varying::splat(seed));
            let mut sref = ScalarRng::seed(seed);
            for _ in 0..16 {
                let r1 = s1.random_u32().to_array()[0];
                let r8 = s8.random_u32().to_array();
                let rref = sref.next();
                assert_eq!(r1, rref, "N=1 vs scalar seed={seed:#x}");
                assert_eq!(r8, [rref; 8], "N=8 vs scalar seed={seed:#x}");
            }
        }
    }

    #[test]
    fn frandom_matches_reference_and_is_unit_interval() {
        for &(seed, _, _) in VECTORS {
            let mut s = RNGState::<8>::new(Varying::splat(seed));
            let mut sref = ScalarRng::seed(seed);
            for _ in 0..16 {
                let f = s.frandom().to_array();
                let expect = frand_ref(sref.next());
                for &lane in &f {
                    assert_eq!(lane, expect, "frandom seed={seed:#x}");
                    assert!((0.0..1.0).contains(&lane), "frandom out of [0,1): {lane}");
                }
            }
        }
    }

    #[test]
    fn varying_seeds_are_independent_per_lane() {
        let seeds = [1u32, 42, 0x12345678, 7, 0, 999, 0xdead_beef, 0x8000_0001];
        let mut s = RNGState::<8>::new(Varying::from_array(seeds));
        let mut refs: [ScalarRng; 8] = core::array::from_fn(|i| ScalarRng::seed(seeds[i]));
        for _ in 0..16 {
            let got = s.random_u32().to_array();
            let want: [u32; 8] = core::array::from_fn(|i| refs[i].next());
            assert_eq!(got, want);
        }
    }
}