fgf 3.0.2

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

//! x86 tier dispatch for the GF(2) bit-packed interop seam.
//!
//! The GitHub-hosted coverage runner resolves `V3` without GFNI or AVX-512,
//! so the `V3GfniCrypto` and `V4x` arms below cannot execute there — the
//! same situation as the GFNI dispatch arms of `kernel/gf8.rs` and
//! `kernel/tower.rs`, and this dispatch file is excluded from coverage the
//! same way (`COV_IGNORE` in `crate.just`).
//!
//! Tier floors are measured, not reasoned: a row shorter than the tier's
//! vector unit pays setup without vector work and loses to the portable
//! body; from one full unit up every tier wins. Each arm therefore requires
//! one full vector unit of row coverage: 16 bytes at `V2`, 32 at `V3` and
//! `V3GfniCrypto`, 64 at `V4x`. The floor benchmark record is the bit interop
//! tier-floors panel in `benchmarks/v3/gf1.md`.

use super::super::{X86Proof, x86, x86_proof};

/// Dispatch [`super::expand`] to the selected tier. Returns `true` when an
/// arm handled the row; `false` leaves the portable body to run.
pub(crate) fn expand(dst: &mut [u8], one_w: u64, src: &[u8], bits: usize, bytes: usize) -> bool {
    match x86_proof() {
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 64 =>
        {
            x86::gf1::expand_avx512(token, dst, one_w, src, bits, bytes);
            true
        }
        // A short V4x row falls through to the narrower vector tiers
        // instead of the portable body: 16-63 byte rows still cover a
        // full SSE4.2 or AVX2 unit.
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::expand_avx2(token.v3(), dst, one_w, src, bits, bytes);
            true
        }
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 16 =>
        {
            x86::gf1::expand_sse42(token.v3().v2(), dst, one_w, src, bits, bytes);
            true
        }
        X86Proof::V3GfniCrypto(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::expand_avx2(token.v3(), dst, one_w, src, bits, bytes);
            true
        }
        X86Proof::V3(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::expand_avx2(token, dst, one_w, src, bits, bytes);
            true
        }
        X86Proof::V2(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 16 =>
        {
            x86::gf1::expand_sse42(token, dst, one_w, src, bits, bytes);
            true
        }
        _ => false,
    }
}

/// Dispatch [`super::compress`] to the selected tier. Returns `true` when an
/// arm handled the row; `false` leaves the portable body to run.
pub(crate) fn compress(
    dst: &mut [u8],
    src: &[u8],
    bits: usize,
    bytes: usize,
    test: &super::LaneTest,
) -> bool {
    match x86_proof() {
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 64 =>
        {
            x86::gf1::compress_avx512(token, dst, src, bits, bytes, test);
            true
        }
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::compress_avx2(token.v3(), dst, src, bits, bytes, test);
            true
        }
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 16 =>
        {
            x86::gf1::compress_sse42(token.v3().v2(), dst, src, bits, bytes, test);
            true
        }
        X86Proof::V3GfniCrypto(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::compress_avx2(token.v3(), dst, src, bits, bytes, test);
            true
        }
        X86Proof::V3(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::compress_avx2(token, dst, src, bits, bytes, test);
            true
        }
        X86Proof::V2(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 16 =>
        {
            x86::gf1::compress_sse42(token, dst, src, bits, bytes, test);
            true
        }
        _ => false,
    }
}

/// Dispatch [`super::restrict`] to the selected tier. Returns `true` when an
/// arm handled the row; `false` leaves the portable body to run.
#[allow(clippy::too_many_arguments)]
pub(crate) fn restrict(
    dst: &mut [u8],
    valid: &mut [u8],
    one_w: u64,
    src: &[u8],
    bits: usize,
    bytes: usize,
) -> bool {
    match x86_proof() {
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 64 =>
        {
            x86::gf1::restrict_avx512(token, dst, valid, one_w, src, bits, bytes);
            true
        }
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::restrict_avx2(token.v3(), dst, valid, one_w, src, bits, bytes);
            true
        }
        #[cfg(feature = "simd512")]
        X86Proof::V4x(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 16 =>
        {
            x86::gf1::restrict_sse42(token.v3().v2(), dst, valid, one_w, src, bits, bytes);
            true
        }
        X86Proof::V3GfniCrypto(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::restrict_avx2(token.v3(), dst, valid, one_w, src, bits, bytes);
            true
        }
        X86Proof::V3(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 32 =>
        {
            x86::gf1::restrict_avx2(token, dst, valid, one_w, src, bits, bytes);
            true
        }
        X86Proof::V2(token)
            if matches!(bytes, 1 | 2 | 4 | 8) && bits.saturating_mul(bytes) >= 16 =>
        {
            x86::gf1::restrict_sse42(token, dst, valid, one_w, src, bits, bytes);
            true
        }
        _ => false,
    }
}