fgf 1.1.0

Faster Galois Fields: SIMD and portable finite-field arithmetic (binary towers and prime fields) and vector kernels
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
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
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247

//! Mersenne31 (`u32` lanes) SSE4.2 kernels.
//!
//! Four lanes per vector, same Mersenne fold as the AVX2 path. Serves the
//! `V2` backend (x86-64-v2 guarantees SSE4.1 `pminud` and SSE4.2
//! `pcmpgtq`): each entry is a safe capability-token function taking an
//! `X64V2Token` and validating its own geometry.

use super::{M31_P, check_elem_multiple, check_equal, m31_mulmod128};
use crate::field::mersenne31::{self, Mersenne31};
use crate::kernel::prime;

#[cfg(target_arch = "x86")]
use core::arch::x86::*;
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64::*;

/// Fold four raw `u32` lanes to `0..=p` (`2^31 ≡ 1`, no conditional
/// subtract), tolerating `= p` outputs like the AVX2 twin.
#[archmage::rite(v2)]
fn m31_fold128(x: __m128i, p: __m128i) -> __m128i {
    let lo = _mm_and_si128(x, p);
    let hi = _mm_srli_epi32(x, 31);
    _mm_add_epi32(lo, hi)
}

/// Canonicalize lanes in `0..=2p`: two unconditional subtract-and-min steps.
#[archmage::rite(v2)]
fn m31_min_chain128(x: __m128i, p: __m128i) -> __m128i {
    let u = _mm_min_epu32(x, _mm_sub_epi32(x, p));
    _mm_min_epu32(u, _mm_sub_epi32(u, p))
}

/// `dst += src (mod p)`, Mersenne31, SSE4.2.
///
/// # Panics
/// Panics if the slices differ in length or hold a partial lane.
#[allow(clippy::used_underscore_binding)]
#[archmage::arcane(import_intrinsics)]
pub fn add_assign_m31_sse42(_token: archmage::X64V2Token, dst: &mut [u8], src: &[u8]) {
    check_equal("m31::add_assign_sse42", "dst", dst.len(), "src", src.len());
    check_elem_multiple("m31::add_assign_sse42", dst.len(), 4);
    let p = _mm_set1_epi32(M31_P);
    let (dst_tiles, dst_rest) = dst.as_chunks_mut::<64>();
    let (src_tiles, src_rest) = src.as_chunks::<64>();
    for (dst_tile, src_tile) in dst_tiles.iter_mut().zip(src_tiles) {
        let (dst_lanes, _) = dst_tile.as_chunks_mut::<16>();
        let (src_lanes, _) = src_tile.as_chunks::<16>();
        for (dst_lane, src_lane) in dst_lanes.iter_mut().zip(src_lanes) {
            let d = m31_fold128(_mm_loadu_si128(&*dst_lane), p);
            let s = m31_fold128(_mm_loadu_si128(src_lane), p);
            let sum = _mm_add_epi32(d, s); // <= 2p
            _mm_storeu_si128(dst_lane, m31_min_chain128(sum, p));
        }
    }
    let (dst_lanes, dst_tail) = dst_rest.as_chunks_mut::<16>();
    let (src_lanes, src_tail) = src_rest.as_chunks::<16>();
    for (dst_lane, src_lane) in dst_lanes.iter_mut().zip(src_lanes) {
        let d = m31_fold128(_mm_loadu_si128(&*dst_lane), p);
        let s = m31_fold128(_mm_loadu_si128(src_lane), p);
        let sum = _mm_add_epi32(d, s);
        _mm_storeu_si128(dst_lane, m31_min_chain128(sum, p));
    }
    prime::add_assign::<Mersenne31>(dst_tail, src_tail);
}

/// `dst -= src (mod p)`, Mersenne31, SSE4.2.
///
/// # Panics
/// Panics if the slices differ in length or hold a partial lane.
#[allow(clippy::used_underscore_binding)]
#[archmage::arcane(import_intrinsics)]
pub fn sub_assign_m31_sse42(_token: archmage::X64V2Token, dst: &mut [u8], src: &[u8]) {
    check_equal("m31::sub_assign_sse42", "dst", dst.len(), "src", src.len());
    check_elem_multiple("m31::sub_assign_sse42", dst.len(), 4);
    let p = _mm_set1_epi32(M31_P);
    let (dst_tiles, dst_rest) = dst.as_chunks_mut::<64>();
    let (src_tiles, src_rest) = src.as_chunks::<64>();
    for (dst_tile, src_tile) in dst_tiles.iter_mut().zip(src_tiles) {
        let (dst_lanes, _) = dst_tile.as_chunks_mut::<16>();
        let (src_lanes, _) = src_tile.as_chunks::<16>();
        for (dst_lane, src_lane) in dst_lanes.iter_mut().zip(src_lanes) {
            let d = m31_fold128(_mm_loadu_si128(&*dst_lane), p);
            let s = m31_fold128(_mm_loadu_si128(src_lane), p);
            let diff = _mm_sub_epi32(d, s); // wraps iff negative
            let fixed = _mm_min_epu32(diff, _mm_add_epi32(diff, p));
            _mm_storeu_si128(dst_lane, m31_min_chain128(fixed, p));
        }
    }
    let (dst_lanes, dst_tail) = dst_rest.as_chunks_mut::<16>();
    let (src_lanes, src_tail) = src_rest.as_chunks::<16>();
    for (dst_lane, src_lane) in dst_lanes.iter_mut().zip(src_lanes) {
        let d = m31_fold128(_mm_loadu_si128(&*dst_lane), p);
        let s = m31_fold128(_mm_loadu_si128(src_lane), p);
        let diff = _mm_sub_epi32(d, s);
        let fixed = _mm_min_epu32(diff, _mm_add_epi32(diff, p));
        _mm_storeu_si128(dst_lane, m31_min_chain128(fixed, p));
    }
    prime::sub_assign::<Mersenne31>(dst_tail, src_tail);
}

/// `dst += coeff * src (mod p)`, Mersenne31, SSE4.2. `coeff` must be canonical.
///
/// # Panics
/// Panics if the slices differ in length or hold a partial lane.
#[allow(clippy::cast_possible_wrap, clippy::used_underscore_binding)]
#[archmage::arcane(import_intrinsics)]
pub fn mul_add_m31_sse42(_token: archmage::X64V2Token, dst: &mut [u8], coeff: u32, src: &[u8]) {
    check_equal("m31::mul_add_sse42", "dst", dst.len(), "src", src.len());
    check_elem_multiple("m31::mul_add_sse42", dst.len(), 4);
    let p = _mm_set1_epi32(M31_P);
    // The coefficient is loop-invariant and contract-canonical: fold it once.
    let cred = {
        let cvec = _mm_set1_epi32(coeff as i32);
        let f = m31_fold128(cvec, p);
        let u = _mm_min_epu32(f, _mm_sub_epi32(f, p));
        _mm_min_epu32(u, _mm_sub_epi32(u, p))
    };
    let (dst_tiles, dst_rest) = dst.as_chunks_mut::<64>();
    let (src_tiles, src_rest) = src.as_chunks::<64>();
    for (dst_tile, src_tile) in dst_tiles.iter_mut().zip(src_tiles) {
        let (dst_lanes, _) = dst_tile.as_chunks_mut::<16>();
        let (src_lanes, _) = src_tile.as_chunks::<16>();
        for (dst_lane, src_lane) in dst_lanes.iter_mut().zip(src_lanes) {
            let prod = m31_mulmod128(cred, m31_fold128(_mm_loadu_si128(src_lane), p), p);
            let d = m31_fold128(_mm_loadu_si128(&*dst_lane), p);
            let sum = _mm_add_epi32(d, prod); // <= 2p - 1
            let u = _mm_min_epu32(sum, _mm_sub_epi32(sum, p));
            _mm_storeu_si128(dst_lane, u);
        }
    }
    let (dst_lanes, dst_tail) = dst_rest.as_chunks_mut::<16>();
    let (src_lanes, src_tail) = src_rest.as_chunks::<16>();
    for (dst_lane, src_lane) in dst_lanes.iter_mut().zip(src_lanes) {
        let prod = m31_mulmod128(cred, m31_fold128(_mm_loadu_si128(src_lane), p), p);
        let d = m31_fold128(_mm_loadu_si128(&*dst_lane), p);
        let sum = _mm_add_epi32(d, prod);
        let u = _mm_min_epu32(sum, _mm_sub_epi32(sum, p));
        _mm_storeu_si128(dst_lane, u);
    }
    prime::mul_add::<Mersenne31>(dst_tail, mersenne31::Elem(coeff), src_tail);
}

/// `dst *= coeff (mod p)`, Mersenne31, SSE4.2. `coeff` must be canonical.
///
/// # Panics
/// Panics on a partial trailing lane.
#[allow(clippy::cast_possible_wrap, clippy::used_underscore_binding)]
#[archmage::arcane(import_intrinsics)]
pub fn mul_assign_m31_sse42(_token: archmage::X64V2Token, dst: &mut [u8], coeff: u32) {
    check_elem_multiple("m31::mul_assign_sse42", dst.len(), 4);
    let p = _mm_set1_epi32(M31_P);
    // The coefficient is loop-invariant and contract-canonical: fold it once.
    let cred = {
        let f = m31_fold128(_mm_set1_epi32(coeff as i32), p);
        let u = _mm_min_epu32(f, _mm_sub_epi32(f, p));
        _mm_min_epu32(u, _mm_sub_epi32(u, p))
    };
    let (dst_tiles, dst_rest) = dst.as_chunks_mut::<64>();
    for dst_tile in dst_tiles {
        let (dst_lanes, _) = dst_tile.as_chunks_mut::<16>();
        for dst_lane in dst_lanes {
            let d = m31_fold128(_mm_loadu_si128(&*dst_lane), p);
            _mm_storeu_si128(dst_lane, m31_mulmod128(cred, d, p));
        }
    }
    let (dst_lanes, dst_tail) = dst_rest.as_chunks_mut::<16>();
    for dst_lane in dst_lanes {
        let d = m31_fold128(_mm_loadu_si128(&*dst_lane), p);
        _mm_storeu_si128(dst_lane, m31_mulmod128(cred, d, p));
    }
    prime::mul_assign::<Mersenne31>(dst_tail, mersenne31::Elem(coeff));
}

/// `dst = coeff * src (mod p)`, Mersenne31, SSE4.2. `coeff` must be canonical.
///
/// # Panics
/// Panics if the slices differ in length or hold a partial lane.
#[allow(clippy::cast_possible_wrap, clippy::used_underscore_binding)]
#[archmage::arcane(import_intrinsics)]
pub fn mul_into_m31_sse42(_token: archmage::X64V2Token, dst: &mut [u8], coeff: u32, src: &[u8]) {
    check_equal("m31::mul_into_sse42", "dst", dst.len(), "src", src.len());
    check_elem_multiple("m31::mul_into_sse42", dst.len(), 4);
    let p = _mm_set1_epi32(M31_P);
    // The coefficient is loop-invariant and contract-canonical: fold it once.
    let cred = {
        let f = m31_fold128(_mm_set1_epi32(coeff as i32), p);
        let u = _mm_min_epu32(f, _mm_sub_epi32(f, p));
        _mm_min_epu32(u, _mm_sub_epi32(u, p))
    };
    let (dst_tiles, dst_rest) = dst.as_chunks_mut::<64>();
    let (src_tiles, src_rest) = src.as_chunks::<64>();
    for (dst_tile, src_tile) in dst_tiles.iter_mut().zip(src_tiles) {
        let (dst_lanes, _) = dst_tile.as_chunks_mut::<16>();
        let (src_lanes, _) = src_tile.as_chunks::<16>();
        for (dst_lane, src_lane) in dst_lanes.iter_mut().zip(src_lanes) {
            let s = m31_fold128(_mm_loadu_si128(src_lane), p);
            _mm_storeu_si128(dst_lane, m31_mulmod128(cred, s, p));
        }
    }
    let (dst_lanes, dst_tail) = dst_rest.as_chunks_mut::<16>();
    let (src_lanes, src_tail) = src_rest.as_chunks::<16>();
    for (dst_lane, src_lane) in dst_lanes.iter_mut().zip(src_lanes) {
        let s = m31_fold128(_mm_loadu_si128(src_lane), p);
        _mm_storeu_si128(dst_lane, m31_mulmod128(cred, s, p));
    }
    // Scalar tail.
    for (d, s) in dst_tail.chunks_exact_mut(4).zip(src_tail.chunks_exact(4)) {
        let v =
            mersenne31::Elem(coeff) * mersenne31::Elem(u32::from_le_bytes(s.try_into().unwrap()));
        d.copy_from_slice(&v.to_raw().to_le_bytes());
    }
}

/// `dst[i] = a[i] * b[i] (mod p)`, Mersenne31, SSE4.2.
///
/// # Panics
/// Panics unless all three buffers match in length (whole lanes).
#[allow(clippy::used_underscore_binding)]
#[archmage::arcane(import_intrinsics)]
pub fn mul_elementwise_m31_sse42(_token: archmage::X64V2Token, dst: &mut [u8], a: &[u8], b: &[u8]) {
    check_equal("m31::mul_elementwise_sse42", "dst", dst.len(), "a", a.len());
    check_equal("m31::mul_elementwise_sse42", "dst", dst.len(), "b", b.len());
    check_elem_multiple("m31::mul_elementwise_sse42", dst.len(), 4);
    let p = _mm_set1_epi32(M31_P);
    let (dst_tiles, dst_rest) = dst.as_chunks_mut::<64>();
    let (a_tiles, a_rest) = a.as_chunks::<64>();
    let (b_tiles, b_rest) = b.as_chunks::<64>();
    for (dst_tile, (a_tile, b_tile)) in dst_tiles.iter_mut().zip(a_tiles.iter().zip(b_tiles)) {
        let (dst_lanes, _) = dst_tile.as_chunks_mut::<16>();
        let (a_lanes, _) = a_tile.as_chunks::<16>();
        let (b_lanes, _) = b_tile.as_chunks::<16>();
        for ((dst_lane, a_lane), b_lane) in dst_lanes.iter_mut().zip(a_lanes).zip(b_lanes) {
            let va = m31_fold128(_mm_loadu_si128(a_lane), p);
            let vb = m31_fold128(_mm_loadu_si128(b_lane), p);
            _mm_storeu_si128(dst_lane, m31_mulmod128(va, vb, p));
        }
    }
    let (dst_lanes, dst_tail) = dst_rest.as_chunks_mut::<16>();
    let (a_lanes, a_tail) = a_rest.as_chunks::<16>();
    let (b_lanes, b_tail) = b_rest.as_chunks::<16>();
    for ((dst_lane, a_lane), b_lane) in dst_lanes.iter_mut().zip(a_lanes).zip(b_lanes) {
        let va = m31_fold128(_mm_loadu_si128(a_lane), p);
        let vb = m31_fold128(_mm_loadu_si128(b_lane), p);
        _mm_storeu_si128(dst_lane, m31_mulmod128(va, vb, p));
    }
    prime::mul_elementwise::<Mersenne31>(dst_tail, a_tail, b_tail);
}