opthash 0.7.0

Rust implementations of Elastic Hashing and Funnel Hashing
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
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290

#[cfg(target_arch = "aarch64")]
use core::arch::aarch64::{
    uint8x16_t, vandq_u8, vceqq_u8, vdupq_n_u8, vget_lane_u64, vld1q_u8, vreinterpret_u64_u8,
    vreinterpretq_u16_u8, vshrn_n_u16,
};
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64::{_MM_HINT_T0, _mm_prefetch};
#[cfg(target_arch = "x86_64")]
use std::arch::x86_64::{
    __m128i, __m256i, _mm_and_si128, _mm_cmpeq_epi8, _mm_cmpgt_epi8, _mm_loadu_si128,
    _mm_movemask_epi8, _mm_set1_epi8, _mm_setzero_si128, _mm256_cmpeq_epi8, _mm256_loadu_si256,
    _mm256_movemask_epi8, _mm256_set1_epi8,
};

use super::bitmask::BitMask;
use super::config::GROUP_SIZE;
#[allow(unused_imports)]
use super::control::{CTRL_EMPTY, CTRL_TOMBSTONE, FINGERPRINT_MASK};

/// 1-bit-per-byte u32 mask over a 16-byte chunk
#[inline]
pub(super) fn match_fingerprint_group_u32(ptr: *const u8, target: u8) -> u32 {
    #[cfg(target_arch = "x86_64")]
    #[allow(clippy::cast_ptr_alignment)]
    unsafe {
        let data = _mm_loadu_si128(ptr.cast::<__m128i>());
        #[allow(clippy::cast_possible_wrap)]
        let cmp = _mm_cmpeq_epi8(data, _mm_set1_epi8(target as i8));
        #[allow(clippy::cast_sign_loss)]
        {
            (_mm_movemask_epi8(cmp) as u32) & 0xFFFF
        }
    }
    #[cfg(not(target_arch = "x86_64"))]
    {
        let mut m = 0u32;
        for i in 0..GROUP_SIZE {
            if unsafe { *ptr.add(i) } == target {
                m |= 1 << i;
            }
        }
        m
    }
}

// ---------------------------------------------------------------------------
// SIMD mask functions
// ---------------------------------------------------------------------------

/// # Safety
///
/// `ptr` must be valid to read `GROUP_SIZE` bytes.
#[inline]
#[must_use]
pub(crate) unsafe fn eq_mask_16(ptr: *const u8, target: u8) -> BitMask {
    #[cfg(target_arch = "aarch64")]
    unsafe {
        eq_mask_16_neon(ptr, target)
    }
    #[cfg(target_arch = "x86_64")]
    unsafe {
        eq_mask_16_sse2(ptr, target)
    }
    #[cfg(not(any(target_arch = "x86_64", target_arch = "aarch64")))]
    {
        let mut m: u16 = 0;
        for i in 0..GROUP_SIZE {
            if unsafe { *ptr.add(i) } == target {
                m |= 1u16 << i;
            }
        }
        BitMask(m)
    }
}

/// # Safety
///
/// `ptr` must be valid to read `GROUP_SIZE` bytes.
#[inline]
#[must_use]
pub(crate) unsafe fn free_mask_16(ptr: *const u8) -> BitMask {
    #[cfg(target_arch = "aarch64")]
    unsafe {
        free_mask_16_neon(ptr)
    }
    #[cfg(target_arch = "x86_64")]
    unsafe {
        free_mask_16_sse2(ptr)
    }
    #[cfg(not(any(target_arch = "x86_64", target_arch = "aarch64")))]
    {
        let mut m: u16 = 0;
        for i in 0..GROUP_SIZE {
            let b = unsafe { *ptr.add(i) };
            if b == CTRL_EMPTY || b == CTRL_TOMBSTONE {
                m |= 1u16 << i;
            }
        }
        BitMask(m)
    }
}

/// Bitmask of occupied slots (low-7-bit fingerprint set, high bit clear).
/// Padding and tombstones are excluded.
///
/// # Safety
///
/// `ptr` must be valid to read `GROUP_SIZE` bytes.
#[inline]
#[must_use]
pub(crate) unsafe fn occupied_mask_16(ptr: *const u8) -> BitMask {
    #[cfg(target_arch = "aarch64")]
    unsafe {
        occupied_mask_16_neon(ptr)
    }
    #[cfg(target_arch = "x86_64")]
    unsafe {
        occupied_mask_16_sse2(ptr)
    }
    #[cfg(not(any(target_arch = "x86_64", target_arch = "aarch64")))]
    {
        let mut m: u16 = 0;
        for i in 0..GROUP_SIZE {
            let b = unsafe { *ptr.add(i) };
            if (b & FINGERPRINT_MASK) != 0 {
                m |= 1u16 << i;
            }
        }
        BitMask(m)
    }
}

/// # Safety
///
/// `ptr` must be valid to read 32 bytes.
#[inline]
#[must_use]
pub(crate) unsafe fn eq_mask_32(ptr: *const u8, target: u8) -> u32 {
    #[cfg(target_arch = "x86_64")]
    {
        if std::is_x86_feature_detected!("avx2") {
            unsafe { return eq_mask_32_avx2(ptr, target) };
        }
    }

    let lo = match_fingerprint_group_u32(ptr, target);
    let hi = match_fingerprint_group_u32(unsafe { ptr.add(GROUP_SIZE) }, target);
    lo | (hi << GROUP_SIZE)
}

// ---------------------------------------------------------------------------
// Prefetch
// ---------------------------------------------------------------------------

/// # Safety
///
/// `ptr` must be a valid, aligned pointer to readable memory (or null, in which
/// case the prefetch is silently ignored by the hardware).
#[inline]
pub(crate) unsafe fn prefetch_read(ptr: *const u8) {
    #[cfg(target_arch = "aarch64")]
    unsafe {
        core::arch::asm!("prfm pldl1keep, [{}]", in(reg) ptr, options(nostack, preserves_flags));
    }
    #[cfg(target_arch = "x86_64")]
    unsafe {
        _mm_prefetch(ptr.cast::<i8>(), _MM_HINT_T0);
    }
}

// ---------------------------------------------------------------------------
// Platform-specific SIMD implementations
// ---------------------------------------------------------------------------

#[cfg(target_arch = "aarch64")]
#[inline]
unsafe fn nibble_mask_from_cmp(cmp: uint8x16_t) -> BitMask {
    // vshrn narrows 16×u8 → 8 bytes: each source byte of 0xFF becomes a nibble
    // of 0xF in the output, 0x00 becomes 0x0. Result u64 has 4 bits per slot.
    unsafe {
        let narrowed = vshrn_n_u16(vreinterpretq_u16_u8(cmp), 4);
        BitMask(vget_lane_u64(vreinterpret_u64_u8(narrowed), 0))
    }
}

#[cfg(target_arch = "aarch64")]
#[inline]
unsafe fn eq_mask_16_neon(ptr: *const u8, target: u8) -> BitMask {
    unsafe {
        let bytes = vld1q_u8(ptr);
        let cmp = vceqq_u8(bytes, vdupq_n_u8(target));
        nibble_mask_from_cmp(cmp)
    }
}

#[cfg(target_arch = "aarch64")]
#[inline]
unsafe fn free_mask_16_neon(ptr: *const u8) -> BitMask {
    unsafe {
        let bytes = vld1q_u8(ptr);
        let masked = vandq_u8(bytes, vdupq_n_u8(FINGERPRINT_MASK));
        let free_cmp = vceqq_u8(masked, vdupq_n_u8(0));
        nibble_mask_from_cmp(free_cmp)
    }
}

#[cfg(target_arch = "aarch64")]
#[inline]
unsafe fn occupied_mask_16_neon(ptr: *const u8) -> BitMask {
    // vtstq_u8(a, b) gives 0xFF where (a AND b) != 0 — with b = 0x7F that's
    // exactly the occupied set.
    unsafe {
        let bytes = vld1q_u8(ptr);
        let occ_cmp = core::arch::aarch64::vtstq_u8(bytes, vdupq_n_u8(FINGERPRINT_MASK));
        nibble_mask_from_cmp(occ_cmp)
    }
}

#[allow(
    clippy::cast_possible_wrap,
    clippy::cast_sign_loss,
    clippy::cast_ptr_alignment
)]
#[cfg(target_arch = "x86_64")]
#[inline]
unsafe fn eq_mask_16_sse2(ptr: *const u8, target: u8) -> BitMask {
    unsafe {
        let data = _mm_loadu_si128(ptr.cast::<__m128i>());
        let target_vec = _mm_set1_epi8(target as i8);
        let cmp = _mm_cmpeq_epi8(data, target_vec);
        #[allow(clippy::cast_possible_truncation)]
        {
            BitMask(_mm_movemask_epi8(cmp) as u16)
        }
    }
}

#[allow(
    clippy::cast_possible_wrap,
    clippy::cast_sign_loss,
    clippy::cast_ptr_alignment
)]
#[cfg(target_arch = "x86_64")]
#[inline]
unsafe fn free_mask_16_sse2(ptr: *const u8) -> BitMask {
    unsafe {
        let data = _mm_loadu_si128(ptr.cast::<__m128i>());
        let masked = _mm_and_si128(data, _mm_set1_epi8(FINGERPRINT_MASK as i8));
        let free = _mm_cmpeq_epi8(masked, _mm_setzero_si128());
        #[allow(clippy::cast_possible_truncation)]
        {
            BitMask(_mm_movemask_epi8(free) as u16)
        }
    }
}

#[allow(
    clippy::cast_possible_wrap,
    clippy::cast_sign_loss,
    clippy::cast_ptr_alignment
)]
#[cfg(target_arch = "x86_64")]
#[inline]
unsafe fn occupied_mask_16_sse2(ptr: *const u8) -> BitMask {
    // cmpgt against zero on the low-7-bit mask yields 0xFF for occupied lanes.
    unsafe {
        let data = _mm_loadu_si128(ptr.cast::<__m128i>());
        let masked = _mm_and_si128(data, _mm_set1_epi8(FINGERPRINT_MASK as i8));
        let occ = _mm_cmpgt_epi8(masked, _mm_setzero_si128());
        #[allow(clippy::cast_possible_truncation)]
        {
            BitMask(_mm_movemask_epi8(occ) as u16)
        }
    }
}

#[allow(clippy::cast_possible_wrap, clippy::cast_ptr_alignment)]
#[cfg(target_arch = "x86_64")]
#[inline]
unsafe fn eq_mask_32_avx2(ptr: *const u8, target: u8) -> u32 {
    unsafe {
        let data = _mm256_loadu_si256(ptr.cast::<__m256i>());
        let target_vec = _mm256_set1_epi8(target as i8);
        let cmp = _mm256_cmpeq_epi8(data, target_vec);
        #[allow(clippy::cast_sign_loss)]
        {
            _mm256_movemask_epi8(cmp) as u32
        }
    }
}