opthash 0.4.1

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
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
use std::borrow::Borrow;
use std::hash::{BuildHasher, Hash};

use crate::common::DefaultHashBuilder;
use crate::common::simd::ProbeOps;

use crate::common::{
    config::{DEFAULT_RESERVE_FRACTION, INITIAL_CAPACITY},
    control::{CTRL_EMPTY, CTRL_TOMBSTONE, ControlByte, ControlOps},
    layout::{Entry, GROUP_SIZE, RawTable},
    math::{
        advance_wrapping_index, ceil_three_quarters, fastmod_magic, fastmod_u32,
        floor_half_reserve_slots, level_salt, max_insertions, sanitize_reserve_fraction,
        usize_to_f64,
    },
};

const DEFAULT_PROBE_SCALE: f64 = 16.0;

/// Construction-time tuning for `ElasticHashMap`.
#[derive(Debug, Clone, Copy)]
pub struct ElasticOptions {
    /// Target initial capacity. The map sizes its level partition so
    /// `capacity * (1 - reserve_fraction)` inserts fit before the next resize.
    capacity: usize,
    /// Fraction of slots kept free as headroom. Lower means higher load
    /// factor but more probing on collisions.
    reserve_fraction: f64,
    /// Multiplier on per-level probe budgets. Higher means more thorough
    /// probing within a level before falling through to the next.
    probe_scale: f64,
}

impl Default for ElasticOptions {
    fn default() -> Self {
        Self {
            capacity: 0,
            reserve_fraction: DEFAULT_RESERVE_FRACTION,
            probe_scale: DEFAULT_PROBE_SCALE,
        }
    }
}

impl ElasticOptions {
    #[must_use]
    pub fn with_capacity(capacity: usize) -> Self {
        Self {
            capacity,
            ..Self::default()
        }
    }

    #[must_use]
    pub fn capacity(mut self, capacity: usize) -> Self {
        self.capacity = capacity;
        self
    }

    #[must_use]
    pub fn reserve_fraction(mut self, reserve_fraction: f64) -> Self {
        self.reserve_fraction = reserve_fraction;
        self
    }

    #[must_use]
    pub fn probe_scale(mut self, probe_scale: f64) -> Self {
        self.probe_scale = probe_scale;
        self
    }
}

/// One level in elastic hashing's geometric partition.
///
/// Each level is an independent open-addressed table sized to roughly half
/// the capacity of the previous level. Inserts cascade from level 0 outward
/// per the active batch plan; lookups probe every populated level.
struct Level<K, V> {
    /// Structure of Arrays control bytes + entries.
    table: RawTable<Entry<K, V>>,
    /// Live entry count.
    len: usize,
    /// Deleted-slot count.
    tombstones: usize,
    /// Cached `floor(reserve * cap / 2)` for the
    /// `current_free_slots > threshold` branch in slot selection.
    half_reserve_slot_threshold: usize,
    /// Per-(free-slot count) probe budget for limited probing.
    /// Indexed by `free_slots()`.
    limited_probe_budgets: Box<[usize]>,
    /// Precomputed double-hashing step set.
    group_steps: Box<[usize]>,
    /// Per-level salt mixed into the key hash so each level probes a
    /// different sequence.
    salt: u64,
    /// Fastmod magic for `group_count`.
    group_count_magic: u64,
    /// Fastmod magic for `group_steps.len()`.
    step_count_magic: u64,
}

impl<K, V> Level<K, V> {
    fn with_capacity(
        capacity: usize,
        reserve_fraction: f64,
        probe_scale: f64,
        level_idx: usize,
    ) -> Self {
        let table = RawTable::new(capacity);
        let group_count = table.group_count();
        let group_steps = ProbeOps::build_group_steps(group_count);
        let limited_probe_budgets =
            build_probe_budgets(capacity, group_count, reserve_fraction, probe_scale);
        let group_count_magic = if group_count > 1 {
            fastmod_magic(group_count)
        } else {
            0
        };
        let step_count_magic = if group_steps.len() > 1 {
            fastmod_magic(group_steps.len())
        } else {
            0
        };

        Self {
            table,
            len: 0,
            tombstones: 0,
            half_reserve_slot_threshold: floor_half_reserve_slots(reserve_fraction, capacity),
            limited_probe_budgets,
            group_steps,
            salt: level_salt(level_idx),
            group_count_magic,
            step_count_magic,
        }
    }

    #[inline]
    fn capacity(&self) -> usize {
        self.table.capacity()
    }

    /// Slots minus live entries. Includes tombstone slots since they're
    /// reusable on insert (control byte FREE-or-TOMBSTONE).
    #[inline]
    fn free_slots(&self) -> usize {
        self.capacity().saturating_sub(self.len)
    }

    /// Probe-group budget at the current fill level for limited (early-stop)
    /// probing. Tighter as the level fills.
    #[inline]
    fn limited_group_budget(&self) -> usize {
        self.limited_probe_budgets[self.free_slots()]
    }

    /// Triggers a no-grow rehash on remove when tombstones outnumber half
    /// the slots. Keeps probe sequences from degrading after delete-heavy
    /// workloads.
    #[inline]
    fn needs_cleanup(&self) -> bool {
        self.tombstones > self.capacity() / 2
    }
}

impl<K, V> Drop for Level<K, V> {
    fn drop(&mut self) {
        for idx in 0..self.table.capacity() {
            if self.table.control_at(idx).is_occupied() {
                unsafe { self.table.drop_in_place(idx) };
            }
        }
    }
}

/// Open-addressed hash map using elastic hashing.
///
/// Splits capacity across geometrically shrinking `levels` and routes inserts
/// through a `batch_plan`: early batches concentrate on level 0; later
/// batches push toward deeper levels. Lookups probe every level whose
/// `len > 0`. Unlike standard open addressing, expected probe count stays
/// low even at high load.
pub struct ElasticHashMap<K, V> {
    /// Geometrically shrinking partition of capacity.
    levels: Vec<Level<K, V>>,
    /// Total live entries.
    len: usize,
    /// Total slot count across all levels.
    capacity: usize,
    /// Insert count that triggers `resize(2x)`.
    max_insertions: usize,
    /// Slot reserve fraction per level. See `ElasticOptions`.
    reserve_fraction: f64,
    /// Probe-budget multiplier. See `ElasticOptions`.
    probe_scale: f64,
    /// Per-batch insert quota; drives `current_batch_index` advancement.
    batch_plan: Vec<usize>,
    /// Index into `batch_plan`. Selects which level pair new keys target.
    current_batch_index: usize,
    /// Remaining inserts in the current batch before advancing.
    batch_remaining: usize,
    /// Highest level index ever written; bounds the lookup probe loop.
    max_populated_level: usize,
    /// Hash builder. Cloned across resizes to preserve probe sequences.
    hash_builder: DefaultHashBuilder,
}

impl<K: std::fmt::Debug, V: std::fmt::Debug> std::fmt::Debug for ElasticHashMap<K, V> {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("ElasticHashMap")
            .field("len", &self.len)
            .field("capacity", &self.capacity)
            .field("max_populated_level", &self.max_populated_level)
            .finish_non_exhaustive()
    }
}

impl<K, V> Default for ElasticHashMap<K, V>
where
    K: Eq + Hash,
{
    fn default() -> Self {
        Self::new()
    }
}

impl<K, V> ElasticHashMap<K, V>
where
    K: Eq + Hash,
{
    #[must_use]
    pub fn new() -> Self {
        Self::with_options(ElasticOptions::default())
    }

    #[must_use]
    pub fn with_capacity(capacity: usize) -> Self {
        Self::with_options(ElasticOptions::with_capacity(capacity))
    }

    #[must_use]
    pub fn with_hasher(hash_builder: DefaultHashBuilder) -> Self {
        Self::with_options_and_hasher(ElasticOptions::default(), hash_builder)
    }

    #[must_use]
    pub fn with_capacity_and_hasher(capacity: usize, hash_builder: DefaultHashBuilder) -> Self {
        Self::with_options_and_hasher(ElasticOptions::with_capacity(capacity), hash_builder)
    }

    #[must_use]
    pub fn with_options(options: ElasticOptions) -> Self {
        Self::with_options_and_hasher(options, DefaultHashBuilder::default())
    }

    /// Full constructor. `resize` also calls this with the existing
    /// `hash_builder` so all keys keep the same hash sequence across grows.
    #[must_use]
    pub fn with_options_and_hasher(
        options: ElasticOptions,
        hash_builder: DefaultHashBuilder,
    ) -> Self {
        let reserve_fraction = sanitize_reserve_fraction(options.reserve_fraction);
        let probe_scale = sanitize_probe_scale(options.probe_scale);
        let capacity = options.capacity;
        let max_insertions = max_insertions(capacity, reserve_fraction);

        let level_capacities = partition_levels(capacity);
        let levels = level_capacities
            .iter()
            .enumerate()
            .map(|(level_idx, &cap)| {
                Level::with_capacity(cap, reserve_fraction, probe_scale, level_idx)
            })
            .collect::<Vec<_>>();

        let batch_plan = build_batch_plan(&level_capacities, reserve_fraction, max_insertions);
        let batch_remaining = batch_plan.first().copied().unwrap_or(0);

        Self {
            levels,
            len: 0,
            capacity,
            max_insertions,
            reserve_fraction,
            probe_scale,
            batch_plan,
            current_batch_index: 0,
            batch_remaining,
            max_populated_level: 0,
            hash_builder,
        }
    }

    #[must_use]
    pub fn len(&self) -> usize {
        self.len
    }

    #[must_use]
    pub fn is_empty(&self) -> bool {
        self.len == 0
    }

    #[must_use]
    pub fn capacity(&self) -> usize {
        self.capacity
    }

    /// Grow capacity so at least `additional` more inserts fit without
    /// triggering an internal resize. No-op if already large enough.
    pub fn reserve(&mut self, additional: usize) {
        let needed = self.len.saturating_add(additional);
        if needed <= self.max_insertions {
            return;
        }
        let mut new_capacity = self.capacity.max(INITIAL_CAPACITY);
        while max_insertions(new_capacity, self.reserve_fraction) < needed {
            new_capacity = new_capacity.saturating_mul(2);
        }
        self.resize(new_capacity);
    }

    /// # Panics
    ///
    /// Panics if a resize succeeds but no free slot can be found for the new key.
    pub fn insert(&mut self, key: K, value: V) -> Option<V> {
        let key_hash = self.hash_key(&key);
        let key_fingerprint = ControlOps::control_fingerprint(key_hash);

        if let Some((level_idx, slot_idx)) =
            self.find_slot_indices_with_hash(&key, key_hash, key_fingerprint)
        {
            let entry = unsafe { self.levels[level_idx].table.get_mut(slot_idx) };
            let old = std::mem::replace(&mut entry.value, value);
            return Some(old);
        }

        if self.len >= self.max_insertions {
            let new_capacity = if self.capacity == 0 {
                INITIAL_CAPACITY
            } else {
                self.capacity.saturating_mul(2)
            };
            self.resize(new_capacity);
        }

        self.advance_batch_window();
        let (level_idx, slot_idx) = self
            .choose_slot_for_new_key(key_hash)
            .expect("no free slot found after resize");

        let level = &mut self.levels[level_idx];
        let prev_ctrl = level.table.control_at(slot_idx);
        level
            .table
            .write_with_control(slot_idx, Entry { key, value }, key_fingerprint);
        level.len += 1;
        if prev_ctrl == CTRL_TOMBSTONE {
            level.tombstones -= 1;
        }
        if level_idx > self.max_populated_level {
            self.max_populated_level = level_idx;
        }
        self.len += 1;
        if self.batch_remaining > 0 {
            self.batch_remaining -= 1;
        }
        None
    }

    pub fn get<Q>(&self, key: &Q) -> Option<&V>
    where
        K: Borrow<Q>,
        Q: Hash + Eq + ?Sized,
    {
        let key_hash = self.hash_key(key);
        let key_fingerprint = ControlOps::control_fingerprint(key_hash);
        let (level_idx, slot_idx) =
            self.find_slot_indices_with_hash(key, key_hash, key_fingerprint)?;
        Some(unsafe { &self.levels[level_idx].table.get_ref(slot_idx).value })
    }

    pub fn get_mut<Q>(&mut self, key: &Q) -> Option<&mut V>
    where
        K: Borrow<Q>,
        Q: Hash + Eq + ?Sized,
    {
        let key_hash = self.hash_key(key);
        let key_fingerprint = ControlOps::control_fingerprint(key_hash);
        let (level_idx, slot_idx) =
            self.find_slot_indices_with_hash(key, key_hash, key_fingerprint)?;
        Some(unsafe { &mut self.levels[level_idx].table.get_mut(slot_idx).value })
    }

    pub fn contains_key<Q>(&self, key: &Q) -> bool
    where
        K: Borrow<Q>,
        Q: Hash + Eq + ?Sized,
    {
        let key_hash = self.hash_key(key);
        let key_fingerprint = ControlOps::control_fingerprint(key_hash);
        self.find_slot_indices_with_hash(key, key_hash, key_fingerprint)
            .is_some()
    }

    pub fn remove<Q>(&mut self, key: &Q) -> Option<V>
    where
        K: Borrow<Q>,
        Q: Hash + Eq + ?Sized,
    {
        let key_hash = self.hash_key(key);
        let key_fingerprint = ControlOps::control_fingerprint(key_hash);
        let (level_idx, slot_idx) =
            self.find_slot_indices_with_hash(key, key_hash, key_fingerprint)?;

        let removed_entry = {
            let level = &mut self.levels[level_idx];
            let removed = unsafe { level.table.take(slot_idx) };
            level.table.mark_tombstone(slot_idx);
            level.len -= 1;
            level.tombstones += 1;
            removed
        };

        self.len -= 1;
        let needs_resize = self.levels[level_idx].needs_cleanup();
        self.shrink_max_populated_level();
        if needs_resize {
            self.resize(self.capacity);
        }
        Some(removed_entry.value)
    }

    pub fn clear(&mut self) {
        for level in &mut self.levels {
            for idx in 0..level.table.capacity() {
                if level.table.control_at(idx).is_occupied() {
                    unsafe { level.table.drop_in_place(idx) };
                }
            }
            level.table.clear_all_controls();
            level.len = 0;
            level.tombstones = 0;
        }
        self.len = 0;
        self.current_batch_index = 0;
        self.batch_remaining = self.batch_plan.first().copied().unwrap_or(0);
        self.max_populated_level = 0;
    }

    #[must_use]
    pub fn iter(&self) -> ElasticIter<'_, K, V> {
        ElasticIter {
            levels: &self.levels,
            level_idx: 0,
            slot_idx: 0,
        }
    }
}

/// Borrowing iterator over occupied entries. Walks levels in order, scanning
/// each level's slot array linearly. Skips FREE and TOMBSTONE control bytes.
pub struct ElasticIter<'a, K, V> {
    levels: &'a [Level<K, V>],
    level_idx: usize,
    slot_idx: usize,
}

impl<'a, K, V> Iterator for ElasticIter<'a, K, V> {
    type Item = (&'a K, &'a V);

    fn next(&mut self) -> Option<Self::Item> {
        while self.level_idx < self.levels.len() {
            let level = &self.levels[self.level_idx];
            while self.slot_idx < level.table.capacity() {
                let idx = self.slot_idx;
                self.slot_idx += 1;
                if level.table.control_at(idx).is_occupied() {
                    let entry = unsafe { level.table.get_ref(idx) };
                    return Some((&entry.key, &entry.value));
                }
            }
            self.level_idx += 1;
            self.slot_idx = 0;
        }
        None
    }
}

impl<'a, K, V> IntoIterator for &'a ElasticHashMap<K, V>
where
    K: Eq + Hash,
{
    type Item = (&'a K, &'a V);
    type IntoIter = ElasticIter<'a, K, V>;

    fn into_iter(self) -> Self::IntoIter {
        self.iter()
    }
}

impl<K, V> ElasticHashMap<K, V>
where
    K: Eq + Hash,
{
    /// Drain all live entries into a temp Vec, build a fresh map at
    /// `new_capacity`, reinsert. Passing the current capacity performs a
    /// no-grow rehash that flushes accumulated tombstones.
    fn resize(&mut self, new_capacity: usize) {
        let mut entries = Vec::with_capacity(self.len);

        for level in &mut self.levels {
            for idx in 0..level.table.capacity() {
                if level.table.control_at(idx).is_occupied() {
                    let entry = unsafe { level.table.take(idx) };
                    entries.push((entry.key, entry.value));
                }
            }
            level.table.clear_all_controls();
            level.len = 0;
            level.tombstones = 0;
        }

        self.len = 0;
        self.max_populated_level = 0;

        let hash_builder = std::mem::take(&mut self.hash_builder);
        let mut new_map = Self::with_options_and_hasher(
            ElasticOptions {
                capacity: new_capacity,
                reserve_fraction: self.reserve_fraction,
                probe_scale: self.probe_scale,
            },
            hash_builder,
        );

        for (key, value) in entries {
            new_map.insert(key, value);
        }

        *self = new_map;
    }

    #[inline]
    fn hash_key<Q>(&self, key: &Q) -> u64
    where
        Q: Hash + ?Sized,
    {
        self.hash_builder.hash_one(key)
    }

    /// Advance the batch state machine past any zero-quota batches so the
    /// next insert routes to the correct level pair.
    #[inline]
    fn advance_batch_window(&mut self) {
        while self.batch_remaining == 0 && self.current_batch_index + 1 < self.batch_plan.len() {
            self.current_batch_index += 1;
            self.batch_remaining = self.batch_plan[self.current_batch_index];
        }
    }

    /// Pick the (level, slot) pair to write a new key into. Tries the
    /// batch-targeted level pair first (`choose_slot_targeted`); falls back
    /// to a full sweep across all levels when the targeted slot is full
    /// (e.g. tombstones in earlier levels are the only reusable slots).
    fn choose_slot_for_new_key(&mut self, key_hash: u64) -> Option<(usize, usize)> {
        if self.levels.is_empty() {
            return None;
        }

        if let Some(pair) = self.choose_slot_targeted(key_hash) {
            return Some(pair);
        }

        for li in 0..self.levels.len() {
            if let Some(slot_idx) = self.first_free_uniform(key_hash, li) {
                return Some((li, slot_idx));
            }
        }
        None
    }

    /// Batch-driven slot selection. Reads `current_batch_index` to pick the
    /// level pair `(li, li+1)`, then steers between them based on
    /// `current_free_slots > half_reserve_threshold` and `next_free_slots`
    /// thresholds. Per the elastic-hashing schedule, this is what keeps
    /// expected probe count low at high load.
    fn choose_slot_targeted(&self, key_hash: u64) -> Option<(usize, usize)> {
        if self.current_batch_index == 0 {
            return self
                .first_free_uniform(key_hash, 0)
                .map(|slot_idx| (0, slot_idx));
        }

        let level_idx = self.current_batch_index.saturating_sub(1);
        if level_idx + 1 >= self.levels.len() {
            let last = self.levels.len() - 1;
            return self
                .first_free_uniform(key_hash, last)
                .map(|slot_idx| (last, slot_idx));
        }

        let current_level = &self.levels[level_idx];
        let next_level = &self.levels[level_idx + 1];
        let current_free_slots = current_level.free_slots();
        let next_free_slots = next_level.free_slots();

        if current_free_slots > current_level.half_reserve_slot_threshold
            && next_free_slots.saturating_mul(4) > next_level.capacity()
        {
            let limited_budget = current_level.limited_group_budget();
            if let Some(slot_idx) = self.first_free_limited(key_hash, level_idx, limited_budget) {
                return Some((level_idx, slot_idx));
            }
            if let Some(slot_idx) = self.first_free_uniform(key_hash, level_idx + 1) {
                return Some((level_idx + 1, slot_idx));
            }
            return self
                .first_free_uniform(key_hash, level_idx)
                .map(|slot_idx| (level_idx, slot_idx));
        }

        if current_free_slots <= current_level.half_reserve_slot_threshold {
            if let Some(slot_idx) = self.first_free_uniform(key_hash, level_idx + 1) {
                return Some((level_idx + 1, slot_idx));
            }
            return self
                .first_free_uniform(key_hash, level_idx)
                .map(|slot_idx| (level_idx, slot_idx));
        }

        if let Some(slot_idx) = self.first_free_uniform(key_hash, level_idx) {
            return Some((level_idx, slot_idx));
        }
        self.first_free_uniform(key_hash, level_idx + 1)
            .map(|slot_idx| (level_idx + 1, slot_idx))
    }

    /// Locate `key` across all populated levels. Returns `(level, slot)` on
    /// hit. Bounded by `max_populated_level + 1` so empty trailing levels
    /// don't get probed.
    fn find_slot_indices_with_hash<Q>(
        &self,
        key: &Q,
        key_hash: u64,
        key_fingerprint: u8,
    ) -> Option<(usize, usize)>
    where
        K: Borrow<Q>,
        Q: Eq + ?Sized,
    {
        let search_limit = (self.max_populated_level + 1).min(self.levels.len());
        for (level_idx, level) in self.levels[..search_limit].iter().enumerate() {
            if let Some(slot_idx) =
                Self::find_in_level_by_probe(key_hash, key_fingerprint, key, level)
            {
                return Some((level_idx, slot_idx));
            }
        }
        None
    }

    /// Probe one level for `key`. Walks groups via the level's double-hashing
    /// step, SIMD-matches the fingerprint byte, then key-compares only the
    /// matched slots. Stops on FREE byte (group has space) when no
    /// tombstones exist.
    #[inline]
    fn find_in_level_by_probe<Q>(
        key_hash: u64,
        key_fingerprint: u8,
        key: &Q,
        level: &Level<K, V>,
    ) -> Option<usize>
    where
        K: Borrow<Q>,
        Q: Eq + ?Sized,
    {
        if level.capacity() == 0 || level.len == 0 {
            return None;
        }

        let (group_start, group_step) = Self::group_probe_params(level, key_hash);
        let group_count = level.table.group_count();
        let mut group_idx = group_start;

        let capacity = level.capacity();
        for _ in 0..group_count {
            let match_mask = level.table.group_match_mask(group_idx, key_fingerprint);
            for relative_idx in match_mask {
                let slot_idx = group_idx * GROUP_SIZE + relative_idx;
                let entry = unsafe { level.table.get_ref(slot_idx) };
                if entry.key.borrow() == key {
                    return Some(slot_idx);
                }
            }

            let empty_mask = level.table.group_match_mask(group_idx, CTRL_EMPTY);
            if let Some(off) = empty_mask.lowest()
                && group_idx * GROUP_SIZE + off < capacity
            {
                return None;
            }

            group_idx = advance_wrapping_index(group_idx, group_step, group_count);
        }

        None
    }

    /// Probe-bounded variant of `first_free_uniform`: scans at most
    /// `max_groups` groups before giving up. Used by the elastic schedule
    /// when `current_level` still has reserve headroom.
    fn first_free_limited(
        &self,
        key_hash: u64,
        level_idx: usize,
        max_groups: usize,
    ) -> Option<usize> {
        let level = &self.levels[level_idx];
        if level.capacity() == 0 || level.len >= level.capacity() {
            return None;
        }

        let (group_start, group_step) = Self::group_probe_params(level, key_hash);
        let group_count = level.table.group_count();
        let mut group_idx = group_start;
        let max_groups = max_groups.min(group_count.max(1));

        for _ in 0..max_groups {
            if let Some(slot_idx) = level.table.first_free_in_group(group_idx) {
                return Some(slot_idx);
            }
            group_idx = advance_wrapping_index(group_idx, group_step, group_count);
        }

        None
    }

    /// Linear scan over all groups in `level_idx` for the first FREE-or-
    /// TOMBSTONE slot, following the level's double-hashing step. Returns
    /// `None` only if the level is completely full of OCCUPIED bytes.
    fn first_free_uniform(&self, key_hash: u64, level_idx: usize) -> Option<usize> {
        let level = &self.levels[level_idx];
        if level.capacity() == 0 || level.len >= level.capacity() {
            return None;
        }

        let (group_start, group_step) = Self::group_probe_params(level, key_hash);
        let group_count = level.table.group_count();
        let mut group_idx = group_start;

        for _ in 0..group_count {
            if let Some(slot_idx) = level.table.first_free_in_group(group_idx) {
                return Some(slot_idx);
            }
            group_idx = advance_wrapping_index(group_idx, group_step, group_count);
        }

        None
    }

    #[inline]
    /// Compute `(group_start, group_step)` for double-hashing within
    /// `level`. Mixes `key_hash` with the level's salt and rotates for the
    /// step index so each level walks the group ring differently.
    fn group_probe_params(level: &Level<K, V>, key_hash: u64) -> (usize, usize) {
        let group_count = level.table.group_count();
        if group_count <= 1 {
            return (0, 1);
        }

        let mixed = key_hash ^ level.salt;
        let group_start = fastmod_u32(mixed, level.group_count_magic, group_count);
        let step = if level.group_steps.len() > 1 {
            let step_idx = fastmod_u32(
                mixed.rotate_left(29),
                level.step_count_magic,
                level.group_steps.len(),
            );
            level.group_steps[step_idx]
        } else {
            level.group_steps[0]
        };
        (group_start, step)
    }

    /// After a remove, walk down `max_populated_level` past any now-empty
    /// trailing levels so subsequent lookups don't probe them.
    fn shrink_max_populated_level(&mut self) {
        while self.max_populated_level > 0
            && self.levels[self.max_populated_level].len == 0
            && self.levels[self.max_populated_level].tombstones == 0
        {
            self.max_populated_level -= 1;
        }
        if self.levels.is_empty() || (self.levels[0].len == 0 && self.levels[0].tombstones == 0) {
            self.max_populated_level = 0;
        }
    }
}

fn sanitize_probe_scale(probe_scale: f64) -> f64 {
    if probe_scale.is_finite() && probe_scale > 0.0 {
        probe_scale
    } else {
        DEFAULT_PROBE_SCALE
    }
}

#[allow(
    clippy::cast_precision_loss,
    clippy::cast_possible_truncation,
    clippy::cast_sign_loss
)]
fn build_probe_budgets(
    capacity: usize,
    group_count: usize,
    reserve_fraction: f64,
    probe_scale: f64,
) -> Box<[usize]> {
    let mut budgets = vec![1usize; capacity.saturating_add(1)];
    if capacity == 0 {
        return budgets.into_boxed_slice();
    }

    let max_budget = group_count.max(1);
    let cap_f = usize_to_f64(capacity);
    let log_cap = (1.0 / reserve_fraction).log2();

    // Budget(fs) is a non-increasing staircase function of free_slots.
    // Instead of computing log2/ceil per slot, find the threshold free_slots
    // where each budget level transitions, then fill segments.
    //
    // Budget >= b when: fs < capacity / 2^sqrt((b-1)*GROUP_SIZE / probe_scale)
    let mut thresholds: Vec<(usize, usize)> = Vec::new();
    for b in 2..=max_budget {
        let ratio = ((b - 1) * GROUP_SIZE) as f64 / probe_scale;
        if ratio >= log_cap {
            break;
        }
        let exact = cap_f / f64::exp2(ratio.sqrt());
        let threshold = (exact.ceil() as usize).saturating_sub(1).min(capacity);
        if threshold == 0 {
            break;
        }
        thresholds.push((b, threshold));
    }

    // Fill from highest budget inward (thresholds decrease with increasing b).
    let mut prev_end = 0;
    for &(b, threshold) in thresholds.iter().rev() {
        if threshold > prev_end {
            budgets[(prev_end + 1)..=threshold].fill(b);
            prev_end = threshold;
        }
    }

    budgets.into_boxed_slice()
}

/// Split `total_capacity` into geometrically halving level sizes.
/// First level is `ceil(total / 2)`; each subsequent level halves until the
/// remaining budget is exhausted. Returns `[]` for capacity 0.
fn partition_levels(total_capacity: usize) -> Vec<usize> {
    if total_capacity == 0 {
        return Vec::new();
    }

    let mut sizes = Vec::new();
    let mut remaining = total_capacity;
    let mut next_size = total_capacity.div_ceil(2);

    while remaining > 0 {
        let size = next_size.min(remaining).max(1);
        sizes.push(size);
        remaining -= size;
        if remaining == 0 {
            break;
        }
        next_size = (size / 2).max(1);
    }

    sizes
}

/// Build the per-batch insertion quota that drives `current_batch_index`.
/// Batch 0 fills level 0 to ~3/4 occupancy. Each subsequent batch tops up
/// the previous level toward its reserve threshold while priming the next
/// level. Total quota equals `max_insertions`.
fn build_batch_plan(
    level_capacities: &[usize],
    reserve_fraction: f64,
    max_insertions: usize,
) -> Vec<usize> {
    if level_capacities.is_empty() || max_insertions == 0 {
        return Vec::new();
    }

    let mut plan = Vec::with_capacity(level_capacities.len() + 1);
    plan.push(ceil_three_quarters(level_capacities[0]));

    for level_index in 1..level_capacities.len() {
        let current_level_capacity = level_capacities[level_index - 1];
        let next_level_capacity = level_capacities[level_index];

        let target_current_level_occupancy = current_level_capacity.saturating_sub(
            floor_half_reserve_slots(reserve_fraction, current_level_capacity),
        );
        let initial_current_level_occupancy = ceil_three_quarters(current_level_capacity);
        let initial_next_level_occupancy = ceil_three_quarters(next_level_capacity);

        let batch_size = target_current_level_occupancy
            .saturating_sub(initial_current_level_occupancy)
            .saturating_add(initial_next_level_occupancy);
        plan.push(batch_size);
    }

    let mut inserted = 0;
    for size in &mut plan {
        if inserted >= max_insertions {
            *size = 0;
            continue;
        }
        let room = max_insertions - inserted;
        if *size > room {
            *size = room;
        }
        inserted += *size;
    }

    if inserted < max_insertions {
        plan.push(max_insertions - inserted);
    }

    plan
}

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

    #[test]
    fn level_partition_keeps_capacity_and_halving_shape() {
        let sizes = partition_levels(127);
        assert_eq!(sizes.iter().sum::<usize>(), 127);
        assert!(!sizes.is_empty());

        for window in sizes.windows(2) {
            let current_level_size = usize_to_f64(window[0]);
            let next_level_size = usize_to_f64(window[1]);
            assert!(next_level_size >= (current_level_size / 2.0 - 1.0));
            assert!(next_level_size <= (current_level_size / 2.0 + 1.0));
        }
    }

    #[test]
    fn insert_get_and_update_work() {
        let mut map = ElasticHashMap::with_capacity(64);

        for key in 0..20 {
            assert_eq!(map.insert(key, key * 10), None);
        }
        for key in 0..20 {
            assert_eq!(map.get(&key), Some(&(key * 10)));
        }

        let replaced = map.insert(7, 777).expect("update should succeed");
        assert_eq!(replaced, 70);
        assert_eq!(map.get(&7), Some(&777));
    }

    #[test]
    fn get_mut_and_contains_key_work() {
        let mut map = ElasticHashMap::new();
        assert_eq!(map.insert("alpha", 1), None);
        assert!(map.contains_key("alpha"));

        if let Some(v) = map.get_mut("alpha") {
            *v = 2;
        }
        assert_eq!(map.get("alpha"), Some(&2));
    }

    #[test]
    fn remove_supports_borrowed_key_and_updates_len() {
        let mut map: ElasticHashMap<String, i32> = ElasticHashMap::new();
        assert_eq!(map.insert("alpha".to_string(), 1), None);
        assert_eq!(map.insert("beta".to_string(), 2), None);

        assert_eq!(map.remove("alpha"), Some(1));
        assert_eq!(map.remove("alpha"), None);
        assert_eq!(map.len(), 1);
        assert_eq!(map.get("beta"), Some(&2));
    }

    #[test]
    fn clear_removes_all_entries_and_resets_map() {
        let mut map = ElasticHashMap::with_capacity(64);
        for key in 0..10 {
            assert_eq!(map.insert(key, key * 10), None);
        }

        map.clear();
        assert!(map.is_empty());
        for key in 0..10 {
            assert_eq!(map.get(&key), None);
        }

        assert_eq!(map.insert(99, 990), None);
        assert_eq!(map.get(&99), Some(&990));
    }

    #[test]
    fn new_starts_with_zero_capacity() {
        let map: ElasticHashMap<i32, i32> = ElasticHashMap::new();
        assert_eq!(map.capacity(), 0);
        assert_eq!(map.len(), 0);
    }

    #[test]
    fn insert_resizes_when_threshold_is_reached() {
        let capacity = 40;
        let mut map = ElasticHashMap::with_capacity(capacity);
        let max_insertions = max_insertions(capacity, DEFAULT_RESERVE_FRACTION);

        for key in 0..max_insertions + 10 {
            assert_eq!(map.insert(key, key), None);
        }

        for key in 0..max_insertions + 10 {
            assert_eq!(map.get(&key), Some(&key));
        }

        assert!(map.capacity() > capacity);
    }

    #[test]
    fn insert_resizes_from_zero_capacity() {
        let mut map: ElasticHashMap<i32, i32> = ElasticHashMap::new();
        map.insert(1, 10);
        assert_eq!(map.get(&1), Some(&10));
        assert!(map.capacity() > 0);
    }

    #[test]
    fn options_constructor_preserves_capacity() {
        let map: ElasticHashMap<i32, i32> = ElasticHashMap::with_options(ElasticOptions {
            capacity: 96,
            reserve_fraction: DEFAULT_RESERVE_FRACTION,
            probe_scale: 8.0,
        });
        assert_eq!(map.capacity(), 96);
    }

    #[test]
    fn delete_heavy_preserves_correctness() {
        let n = 10_000;
        let cutoff = (n * 4) / 5;
        for trial in 0..50 {
            let mut map = ElasticHashMap::new();
            for i in 0..n {
                map.insert(i, i * 10);
            }
            // Delete the first 80% of keys.
            for i in 0..cutoff {
                assert_eq!(
                    map.remove(&i),
                    Some(i * 10),
                    "trial {trial}: missing key {i} during delete"
                );
            }
            // Lookup remaining keys (post-tombstone state).
            for i in cutoff..n {
                assert_eq!(
                    map.get(&i),
                    Some(&(i * 10)),
                    "trial {trial}: key {i} missing after deletes"
                );
            }
            assert_eq!(map.len(), (n - cutoff) as usize);
            // Re-insert into tombstone-heavy map.
            for i in n..(n + n / 5) {
                assert_eq!(map.insert(i, i), None);
            }
            for i in n..(n + n / 5) {
                assert_eq!(
                    map.get(&i),
                    Some(&i),
                    "trial {trial}: key {i} missing after re-insert"
                );
            }
        }
    }

    #[test]
    fn large_map_correctness() {
        let n = 10_000;
        let mut map = ElasticHashMap::with_capacity(n * 2);
        for i in 0..n {
            assert_eq!(map.insert(i, i), None);
        }
        for i in 0..n {
            assert_eq!(map.get(&i), Some(&i), "key {i} missing");
        }
        assert_eq!(map.len(), n);
    }

    #[test]
    fn partial_group_capacity_works() {
        // Capacity 18 creates a partial last group (2 valid slots out of 16).
        let mut map = ElasticHashMap::with_capacity(18);
        for i in 0..15 {
            assert_eq!(map.insert(i, i), None);
        }
        for i in 0..15 {
            assert_eq!(map.get(&i), Some(&i));
        }
    }

    #[test]
    fn iter_yields_every_inserted_pair_once() {
        let mut map: ElasticHashMap<i32, i32> = ElasticHashMap::with_capacity(64);
        for i in 0..50 {
            map.insert(i, i * 10);
        }
        let mut collected: Vec<(i32, i32)> = map.iter().map(|(&k, &v)| (k, v)).collect();
        collected.sort();
        let expected: Vec<(i32, i32)> = (0..50).map(|i| (i, i * 10)).collect();
        assert_eq!(collected, expected);
    }

    #[test]
    fn iter_skips_tombstones_after_remove() {
        let mut map: ElasticHashMap<i32, i32> = ElasticHashMap::with_capacity(32);
        for i in 0..20 {
            map.insert(i, i);
        }
        for i in (0..20).step_by(2) {
            map.remove(&i);
        }
        let keys: Vec<i32> = map.iter().map(|(&k, _)| k).collect();
        assert_eq!(keys.len(), 10);
        let mut sorted = keys;
        sorted.sort();
        assert_eq!(sorted, (1..20).step_by(2).collect::<Vec<_>>());
    }

    #[test]
    fn iter_empty_map_is_empty() {
        let map: ElasticHashMap<i32, i32> = ElasticHashMap::new();
        assert_eq!(map.iter().count(), 0);
    }
}