ihash 0.1.2

World's fastest high-quality non-cryptographic hash, deterministic cross-platform XOF, and seekable stream cipher
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
use core::array::from_fn;

use crate::{
  backend::{
    aes128::Aes128,
    group::{GroupV, LaneGroup, MidGroup},
    mode::PostXorAes,
    tree::finalize_merge4_lk,
  },
  const_fn::{CTR_STEP_LI, M_CTR, lane_ptr},
  state::BLOCK_LEN,
};

/// Multi-block SIMD AES primitives shared by 256-bit and 512-bit wide vector backends.
pub(crate) trait WideAes: Copy {
  type L: PostXorAes;
  type W: Copy;
  const BLOCK_PER_VEC: usize;

  fn lane(self) -> Self::L;
  unsafe fn zero_w(self) -> Self::W;
  unsafe fn ld_w(self, ptr: *const u8) -> Self::W;
  unsafe fn st_w(self, ptr: *mut u8, v: Self::W);
  unsafe fn zext128(self, v: <Self::L as Aes128>::V) -> Self::W;
  unsafe fn broadcast(self, v: <Self::L as Aes128>::V) -> Self::W;
  unsafe fn xor_w(self, a: Self::W, b: Self::W) -> Self::W;
  unsafe fn add_u64_w(self, a: Self::W, b: Self::W) -> Self::W;
  unsafe fn aesenc_w(self, a: Self::W, rk: Self::W) -> Self::W;
  unsafe fn aesenclast_w(self, a: Self::W, rk: Self::W) -> Self::W;
}

/// 256-bit (2x128-bit) SIMD AES extension providing 128-bit pair split primitive.
pub(crate) trait Lane256x4: WideAes {
  unsafe fn split_pair(self, v: Self::W) -> (<Self::L as Aes128>::V, <Self::L as Aes128>::V);
}

#[cfg(not(all(
  target_arch = "x86_64",
  target_feature = "avx512f",
  target_feature = "vaes",
  target_feature = "avx2",
  target_feature = "aes",
  target_feature = "sse2"
)))]
mod w256 {
  use super::*;

  /// Adapter mapping a [`Lane256x4`] backend onto [`MidGroup`] and [`LaneGroup`].
  #[derive(Clone, Copy)]
  #[repr(transparent)]
  pub(crate) struct Wide256<B>(pub(in super::super) B);

  #[inline(always)]
  unsafe fn wide_absorb_swap32<B: Lane256x4, const N: usize>(
    b: B,
    acc_li: [B::W; N],
    ptr: *const u8,
  ) -> [B::W; N] {
    // SAFETY: caller ensures target features and readable bytes across N vectors at ptr.
    unsafe { from_fn(|i| b.aesenc_w(acc_li[i], b.ld_w(ptr.add((i ^ 1) * 32)))) }
  }

  impl<B: Lane256x4> MidGroup for Wide256<B> {
    type G4 = [B::W; 2];
    type G8 = [B::W; 4];
    type L = B::L;

    #[inline(always)]
    fn lane(self) -> Self::L {
      self.0.lane()
    }

    #[inline(always)]
    unsafe fn init4<const SEED0: bool>(self, sv: GroupV<Self>) -> Self::G4 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_init::<B, 2, SEED0>(self.0, sv, 0) }
    }

    #[inline(always)]
    unsafe fn absorb4_init(self, key_li: Self::G4, ptr: *const u8) -> Self::G4 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_absorb_init(self.0, key_li, ptr) }
    }

    #[inline(always)]
    unsafe fn absorb4_linear(self, acc: Self::G4, ptr: *const u8) -> Self::G4 {
      let b = self.0;
      // SAFETY: caller ensures target features and >= 64 readable bytes at ptr.
      unsafe { from_fn(|i| b.aesenc_w(acc[i], b.ld_w(ptr.add(i * 32)))) }
    }

    #[inline(always)]
    unsafe fn absorb4_swap32(self, acc: Self::G4, ptr: *const u8) -> Self::G4 {
      // SAFETY: caller ensures target features and >= 64 readable bytes at ptr.
      unsafe { wide_absorb_swap32(self.0, acc, ptr) }
    }

    #[inline(always)]
    unsafe fn merge4<const HI: bool>(
      self,
      acc: Self::G4,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      let b = self.0;
      // SAFETY: caller ensures target features and merge4 inputs.
      unsafe {
        let l01_lk = b.aesenc_w(acc[0], b.zext128(lk));
        let l23 = b.aesenc_w(acc[1], b.zero_w());
        wide_finalize_merge4_256_lk::<B, HI>(b, l01_lk, l23, lk)
      }
    }

    #[inline(always)]
    unsafe fn init8<const SEED0: bool>(self, sv: GroupV<Self>, lane_base: usize) -> Self::G8 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_init::<B, 4, SEED0>(self.0, sv, lane_base) }
    }

    #[inline(always)]
    unsafe fn absorb8_init(self, key_li: Self::G8, ptr: *const u8) -> Self::G8 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_absorb_init(self.0, key_li, ptr) }
    }

    #[inline(always)]
    unsafe fn absorb8_pair64(self, acc: Self::G8, p_lo: *const u8, p_hi: *const u8) -> Self::G8 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_absorb8_pair64(self.0, acc, p_lo, p_hi) }
    }

    #[inline(always)]
    unsafe fn absorb8_swap32(self, acc: Self::G8, last128: *const u8) -> Self::G8 {
      // SAFETY: caller ensures target features and 128 readable bytes at last128.
      unsafe { wide_absorb_swap32(self.0, acc, last128) }
    }

    #[inline(always)]
    unsafe fn merge8<const HI: bool>(
      self,
      acc: Self::G8,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      let b = self.0;
      // SAFETY: caller ensures target features and merge8 inputs.
      unsafe {
        let [p01_lk, p23] = wide_fold_pair::<B, 2, 0>(b, [acc[0], acc[1]], [acc[2], acc[3]], lk);
        wide_finalize_merge4_256_lk::<B, HI>(b, p01_lk, p23, lk)
      }
    }

    #[inline(always)]
    unsafe fn merge16<const HI: bool>(
      self,
      lo: Self::G8,
      hi: Self::G8,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      let b = self.0;
      // SAFETY: caller ensures target features and merge16 inputs.
      unsafe {
        let m_li = wide_fold_pair::<B, 4, 2>(b, lo, hi, lk);
        let p01_lk = b.aesenc_w(m_li[0], m_li[2]);
        let p23 = b.aesenc_w(m_li[1], m_li[3]);
        wide_finalize_merge4_256_lk::<B, HI>(b, p01_lk, p23, lk)
      }
    }
  }

  impl<B: Lane256x4> LaneGroup for Wide256<B> {
    #[inline(always)]
    unsafe fn compress_65_128<const HI: bool, const SEED0: bool>(
      self,
      ptr: *const u8,
      len: usize,
      sv: GroupV<Self>,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      let b = self.0;
      // SAFETY: caller ensures target features and 65 <= len <= 128.
      unsafe {
        let key_li = self.init4::<SEED0>(sv);
        let h0 = b.ld_w(ptr);
        let h1 = b.ld_w(ptr.add(32));
        let tail = ptr.add(len - 64);
        let t0 = b.ld_w(tail);
        let t1 = b.ld_w(tail.add(32));
        let acc_li = [
          b.aesenc_w(b.xor_w(key_li[0], h0), b.add_u64_w(t1, h1)),
          b.aesenc_w(b.xor_w(key_li[1], h1), b.add_u64_w(t0, h0)),
        ];
        self.merge4::<HI>(acc_li, lk)
      }
    }

    #[inline(always)]
    unsafe fn compress_33_64<const HI: bool, const SEED0: bool>(
      self,
      ptr: *const u8,
      len: usize,
      sv: GroupV<Self>,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      // SAFETY: caller ensures target features and 33 <= len <= 64.
      unsafe { wide256_compress_33_64::<B, HI, SEED0>(self.0, ptr, len, sv, lk) }
    }

    #[inline(always)]
    unsafe fn absorb16_swapped(
      self,
      lo: Self::G8,
      hi: Self::G8,
      ptr: *const u8,
    ) -> (Self::G8, Self::G8) {
      let b = self.0;
      // SAFETY: caller ensures target features and 256 readable bytes at ptr.
      unsafe {
        // Load sequentially from ptr + 0 to ptr + 256 before assigning back into lo/hi lanes.
        let hi2 = b.aesenc_w(hi[2], b.ld_w(ptr));
        let hi3 = b.aesenc_w(hi[3], b.ld_w(ptr.add(32)));
        let lo0 = b.aesenc_w(lo[0], b.ld_w(ptr.add(64)));
        let lo1 = b.aesenc_w(lo[1], b.ld_w(ptr.add(96)));
        let lo2 = b.aesenc_w(lo[2], b.ld_w(ptr.add(128)));
        let lo3 = b.aesenc_w(lo[3], b.ld_w(ptr.add(160)));
        let hi0 = b.aesenc_w(hi[0], b.ld_w(ptr.add(192)));
        let hi1 = b.aesenc_w(hi[1], b.ld_w(ptr.add(224)));
        ([lo0, lo1, lo2, lo3], [hi0, hi1, hi2, hi3])
      }
    }
  }
}

#[cfg(not(all(
  target_arch = "x86_64",
  target_feature = "avx512f",
  target_feature = "vaes",
  target_feature = "avx2",
  target_feature = "aes",
  target_feature = "sse2"
)))]
pub(crate) use w256::Wide256;

#[inline(always)]
unsafe fn wide_init<W: WideAes, const N: usize, const SEED0: bool>(
  ctx: W,
  sv: <W::L as Aes128>::V,
  lane_base: usize,
) -> [W::W; N] {
  let step = W::BLOCK_PER_VEC * BLOCK_LEN;
  // SAFETY: caller guarantees target features and lane_base in {0, 8}.
  unsafe {
    let lp = lane_ptr::<SEED0>().add(lane_base * BLOCK_LEN);
    if SEED0 {
      from_fn(|i| ctx.ld_w(lp.add(i * step)))
    } else {
      let sv_w = ctx.broadcast(sv);
      from_fn(|i| ctx.xor_w(sv_w, ctx.ld_w(lp.add(i * step))))
    }
  }
}

#[inline(always)]
unsafe fn wide_absorb_init<W: WideAes, const N: usize>(
  ctx: W,
  key_li: [W::W; N],
  ptr: *const u8,
) -> [W::W; N] {
  let step = W::BLOCK_PER_VEC * BLOCK_LEN;
  // SAFETY: caller guarantees target features and readable bytes across N vectors at ptr.
  from_fn(|i| unsafe { ctx.xor_w(key_li[i], ctx.ld_w(ptr.add(i * step))) })
}

#[inline(always)]
unsafe fn wide_absorb8_pair64<W: WideAes, const N: usize>(
  ctx: W,
  acc_li: [W::W; N],
  p_lo: *const u8,
  p_hi: *const u8,
) -> [W::W; N] {
  let step = W::BLOCK_PER_VEC * BLOCK_LEN;
  let half = N / 2;
  // SAFETY: caller guarantees target features and 64 readable bytes at p_lo and p_hi.
  from_fn(|i| unsafe {
    let p = if i < half {
      p_lo.add(i * step)
    } else {
      p_hi.add((i - half) * step)
    };
    ctx.aesenc_w(acc_li[i], ctx.ld_w(p))
  })
}

#[inline(always)]
unsafe fn wide_fold_pair<W: WideAes, const N: usize, const LK_IDX: usize>(
  w: W,
  a_li: [W::W; N],
  b_li: [W::W; N],
  lk: <W::L as Aes128>::V,
) -> [W::W; N] {
  // SAFETY: caller guarantees target features and valid inputs.
  unsafe {
    let z = w.zero_w();
    let lk_w = w.zext128(lk);
    from_fn(|i| {
      let b_rk = if i == LK_IDX { lk_w } else { z };
      w.aesenc_w(w.aesenc_w(a_li[i], z), w.aesenc_w(b_li[i], b_rk))
    })
  }
}

#[inline(always)]
unsafe fn wide_finalize_merge4_256_lk<H: Lane256x4, const HI: bool>(
  h: H,
  l01_lk: H::W,
  l23: H::W,
  lk: <H::L as Aes128>::V,
) -> (<H::L as Aes128>::V, <H::L as Aes128>::V) {
  let l = h.lane();
  // SAFETY: caller guarantees target features and valid inputs.
  unsafe {
    let (l0_lk, l1) = h.split_pair(l01_lk);
    let (l2, l3) = h.split_pair(l23);
    finalize_merge4_lk::<H::L, HI>(l, [l0_lk, l1, l2, l3], lk)
  }
}

#[inline(always)]
unsafe fn wide256_compress_33_64<H: Lane256x4, const HI: bool, const SEED0: bool>(
  h: H,
  ptr: *const u8,
  len: usize,
  sv: <H::L as Aes128>::V,
  lk: <H::L as Aes128>::V,
) -> (<H::L as Aes128>::V, <H::L as Aes128>::V) {
  // SAFETY: caller guarantees target features and 33 <= len <= 64.
  unsafe {
    let [k01, k23] = wide_init::<H, 2, SEED0>(h, sv, 0);
    let tail = ptr.add(len - 32);
    let z = h.zero_w();
    let l01_lk = h.aesenc_w(h.xor_w(h.ld_w(ptr), k01), h.zext128(lk));
    let l23 = h.aesenc_w(h.xor_w(h.ld_w(tail), k23), z);
    wide_finalize_merge4_256_lk::<H, HI>(h, l01_lk, l23, lk)
  }
}

#[cfg(all(
  target_arch = "x86_64",
  any(
    feature = "std",
    all(
      target_feature = "avx512f",
      target_feature = "vaes",
      target_feature = "avx2",
      target_feature = "aes",
      target_feature = "sse2"
    )
  )
))]
mod w512 {
  use super::*;

  /// 512-bit (4x128-bit) SIMD AES extension delegating finalization to `Half`.
  pub(in super::super) trait Lane512x2: WideAes {
    type Half: Lane256x4<L = Self::L>;

    fn half(self) -> Self::Half;
    unsafe fn split_half(
      self,
      v: Self::W,
    ) -> (<Self::Half as WideAes>::W, <Self::Half as WideAes>::W);
    unsafe fn swap32_in_64(self, v: Self::W) -> Self::W;
  }

  /// Adapter mapping a [`Lane512x2`] backend onto [`MidGroup`] and [`LaneGroup`].
  #[derive(Clone, Copy)]
  #[repr(transparent)]
  pub(in super::super) struct Wide512<B>(pub(in super::super) B);

  #[inline(always)]
  unsafe fn wide_finalize_merge4_512_lk<B: Lane512x2, const HI: bool>(
    b: B,
    v_lk: B::W,
    lk: <B::L as Aes128>::V,
  ) -> (<B::L as Aes128>::V, <B::L as Aes128>::V) {
    // SAFETY: caller guarantees target features and valid inputs.
    unsafe {
      let (l01_lk, l23) = b.split_half(v_lk);
      wide_finalize_merge4_256_lk::<B::Half, HI>(b.half(), l01_lk, l23, lk)
    }
  }

  impl<B: Lane512x2> MidGroup for Wide512<B> {
    type G4 = B::W;
    type G8 = [B::W; 2];
    type L = B::L;

    #[inline(always)]
    fn lane(self) -> Self::L {
      self.0.lane()
    }

    #[inline(always)]
    unsafe fn init4<const SEED0: bool>(self, sv: GroupV<Self>) -> Self::G4 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_init::<B, 1, SEED0>(self.0, sv, 0)[0] }
    }

    #[inline(always)]
    unsafe fn absorb4_init(self, key_li: Self::G4, ptr: *const u8) -> Self::G4 {
      let b = self.0;
      // SAFETY: caller ensures target features and >= 64 readable bytes at ptr.
      unsafe { b.xor_w(key_li, b.ld_w(ptr)) }
    }

    #[inline(always)]
    unsafe fn absorb4_linear(self, acc: Self::G4, ptr: *const u8) -> Self::G4 {
      let b = self.0;
      // SAFETY: caller ensures target features and >= 64 readable bytes at ptr.
      unsafe { b.aesenc_w(acc, b.ld_w(ptr)) }
    }

    #[inline(always)]
    unsafe fn absorb4_swap32(self, acc: Self::G4, ptr: *const u8) -> Self::G4 {
      let b = self.0;
      // SAFETY: caller ensures target features and >= 64 readable bytes at ptr.
      unsafe { b.aesenc_w(acc, b.swap32_in_64(b.ld_w(ptr))) }
    }

    #[inline(always)]
    unsafe fn merge4<const HI: bool>(
      self,
      acc: Self::G4,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      let b = self.0;
      // SAFETY: caller ensures target features and merge4 inputs.
      unsafe { wide_finalize_merge4_512_lk::<B, HI>(b, b.aesenc_w(acc, b.zext128(lk)), lk) }
    }

    #[inline(always)]
    unsafe fn init8<const SEED0: bool>(self, sv: GroupV<Self>, lane_base: usize) -> Self::G8 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_init::<B, 2, SEED0>(self.0, sv, lane_base) }
    }

    #[inline(always)]
    unsafe fn absorb8_init(self, key_li: Self::G8, ptr: *const u8) -> Self::G8 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_absorb_init(self.0, key_li, ptr) }
    }

    #[inline(always)]
    unsafe fn absorb8_pair64(self, acc: Self::G8, p_lo: *const u8, p_hi: *const u8) -> Self::G8 {
      // SAFETY: forwarded from caller contract.
      unsafe { wide_absorb8_pair64(self.0, acc, p_lo, p_hi) }
    }

    #[inline(always)]
    unsafe fn absorb8_swap32(self, acc: Self::G8, last128: *const u8) -> Self::G8 {
      let b = self.0;
      // SAFETY: caller ensures target features and 128 readable bytes at last128.
      unsafe { from_fn(|i| b.aesenc_w(acc[i], b.swap32_in_64(b.ld_w(last128.add(i * 64))))) }
    }

    #[inline(always)]
    unsafe fn merge8<const HI: bool>(
      self,
      acc: Self::G8,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      let b = self.0;
      // SAFETY: caller ensures target features and merge8 inputs.
      unsafe {
        let [v_lk] = wide_fold_pair::<B, 1, 0>(b, [acc[0]], [acc[1]], lk);
        wide_finalize_merge4_512_lk::<B, HI>(b, v_lk, lk)
      }
    }

    #[inline(always)]
    unsafe fn merge16<const HI: bool>(
      self,
      lo: Self::G8,
      hi: Self::G8,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      let b = self.0;
      // SAFETY: caller ensures target features and merge16 inputs.
      unsafe {
        let [m0, m1_lk] = wide_fold_pair::<B, 2, 1>(b, lo, hi, lk);
        wide_finalize_merge4_512_lk::<B, HI>(b, b.aesenc_w(m0, m1_lk), lk)
      }
    }
  }

  impl<B: Lane512x2> LaneGroup for Wide512<B> {
    #[inline(always)]
    unsafe fn compress_65_128<const HI: bool, const SEED0: bool>(
      self,
      ptr: *const u8,
      len: usize,
      sv: GroupV<Self>,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      let b = self.0;
      // SAFETY: caller ensures target features and 65 <= len <= 128.
      unsafe {
        let key = self.init4::<SEED0>(sv);
        let h = b.ld_w(ptr);
        let t = b.ld_w(ptr.add(len - 64));
        let acc = b.aesenc_w(b.xor_w(key, h), b.swap32_in_64(b.add_u64_w(t, h)));
        self.merge4::<HI>(acc, lk)
      }
    }

    #[inline(always)]
    unsafe fn compress_33_64<const HI: bool, const SEED0: bool>(
      self,
      ptr: *const u8,
      len: usize,
      sv: GroupV<Self>,
      lk: GroupV<Self>,
    ) -> (GroupV<Self>, GroupV<Self>) {
      // SAFETY: caller ensures target features and 33 <= len <= 64.
      unsafe { wide256_compress_33_64::<B::Half, HI, SEED0>(self.0.half(), ptr, len, sv, lk) }
    }

    #[inline(always)]
    unsafe fn absorb16_swapped(
      self,
      lo: Self::G8,
      hi: Self::G8,
      ptr: *const u8,
    ) -> (Self::G8, Self::G8) {
      let b = self.0;
      // SAFETY: caller ensures target features and 256 readable bytes at ptr.
      unsafe {
        // Load sequentially from ptr + 0 to ptr + 256 before assigning back into lo/hi lanes.
        let hi1 = b.aesenc_w(hi[1], b.ld_w(ptr));
        let lo0 = b.aesenc_w(lo[0], b.ld_w(ptr.add(64)));
        let lo1 = b.aesenc_w(lo[1], b.ld_w(ptr.add(128)));
        let hi0 = b.aesenc_w(hi[0], b.ld_w(ptr.add(192)));
        ([lo0, lo1], [hi0, hi1])
      }
    }
  }
}

#[cfg(all(
  target_arch = "x86_64",
  any(
    feature = "std",
    all(
      target_feature = "avx512f",
      target_feature = "vaes",
      target_feature = "avx2",
      target_feature = "aes",
      target_feature = "sse2"
    )
  )
))]
pub(super) use w512::{Lane512x2, Wide512};

/// Processes chunks of `4 * BLOCK_PER_VEC` 16-byte CTR blocks across wide SIMD vectors.
#[inline(always)]
pub(super) unsafe fn ctr_4_wide<W: WideAes, const XOR: bool>(
  wide: W,
  rk_li: [<W::L as Aes128>::V; 4],
  blk: &mut u64,
  ptr: &mut *mut u8,
  rem: &mut usize,
) {
  let vec_byte = W::BLOCK_PER_VEC * BLOCK_LEN;
  let chunk_byte = 4 * vec_byte;
  if *rem < chunk_byte {
    return;
  }
  let lane = wide.lane();
  // SAFETY: caller guarantees target features and *rem >= chunk_byte readable/writable bytes.
  unsafe {
    let rk0_w = wide.broadcast(rk_li[0]);
    let rk1_w = wide.broadcast(rk_li[1]);
    let rk2_w = wide.broadcast(rk_li[2]);
    let rk3_w = wide.broadcast(rk_li[3]);
    let step_lane = wide.broadcast(lane.ld(CTR_STEP_LI[W::BLOCK_PER_VEC].as_ptr()));
    let step_2lane = wide.broadcast(lane.ld(CTR_STEP_LI[2 * W::BLOCK_PER_VEC].as_ptr()));
    let step_loop = wide.broadcast(lane.ld(CTR_STEP_LI[4 * W::BLOCK_PER_VEC].as_ptr()));

    let c0 = wide.add_u64_w(
      wide.broadcast(lane.pair_v(*blk, blk.wrapping_mul(M_CTR))),
      wide.ld_w(CTR_STEP_LI.as_ptr().cast()),
    );
    let c1 = wide.add_u64_w(c0, step_lane);
    let mut c_li = [
      c0,
      c1,
      wide.add_u64_w(c0, step_2lane),
      wide.add_u64_w(c1, step_2lane),
    ];

    let blk_step = (4 * W::BLOCK_PER_VEC) as u64;
    while *rem >= chunk_byte {
      let mut k_li: [W::W; 4] = from_fn(|i| wide.xor_w(c_li[i], rk0_w));
      for rk in [rk1_w, rk2_w, rk3_w] {
        for k in &mut k_li {
          *k = wide.aesenc_w(*k, rk);
        }
      }
      for (i, k) in k_li.into_iter().enumerate() {
        let ks = wide.aesenclast_w(k, rk0_w);
        let p = ptr.add(i * vec_byte);
        if XOR {
          wide.st_w(p, wide.xor_w(wide.ld_w(p), ks));
        } else {
          wide.st_w(p, ks);
        }
      }
      for c in &mut c_li {
        *c = wide.add_u64_w(*c, step_loop);
      }
      *ptr = ptr.add(chunk_byte);
      *rem -= chunk_byte;
      *blk = blk.wrapping_add(blk_step);
    }
  }
}