summaryrefslogtreecommitdiffstats
path: root/src/vppinfra/vector_avx512.h
blob: eda65caed35b656219d20588ba987ac7d145f4e4 (plain)
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
/*
 * Copyright (c) 2015 Cisco and/or its affiliates.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at:
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#ifndef included_vector_avx512_h
#define included_vector_avx512_h

#include <vppinfra/clib.h>
#include <x86intrin.h>

/* *INDENT-OFF* */
#define foreach_avx512_vec512i \
  _(i,8,64,epi8) _(i,16,32,epi16) _(i,32,16,epi32)  _(i,64,8,epi64)
#define foreach_avx512_vec512u \
  _(u,8,64,epi8) _(u,16,32,epi16) _(u,32,16,epi32)  _(u,64,8,epi64)
#define foreach_avx512_vec512f \
  _(f,32,8,ps) _(f,64,4,pd)

/* splat, load_unaligned, store_unaligned, is_all_zero, is_equal,
   is_all_equal, is_zero_mask */
#define _(t, s, c, i)                                                         \
  static_always_inline t##s##x##c t##s##x##c##_splat (t##s x)                 \
  {                                                                           \
    return (t##s##x##c) _mm512_set1_##i (x);                                  \
  }                                                                           \
                                                                              \
  static_always_inline t##s##x##c t##s##x##c##_load_aligned (void *p)         \
  {                                                                           \
    return (t##s##x##c) _mm512_load_si512 (p);                                \
  }                                                                           \
                                                                              \
  static_always_inline void t##s##x##c##_store_aligned (t##s##x##c v,         \
							void *p)              \
  {                                                                           \
    _mm512_store_si512 ((__m512i *) p, (__m512i) v);                          \
  }                                                                           \
                                                                              \
  static_always_inline t##s##x##c t##s##x##c##_load_unaligned (void *p)       \
  {                                                                           \
    return (t##s##x##c) _mm512_loadu_si512 (p);                               \
  }                                                                           \
                                                                              \
  static_always_inline void t##s##x##c##_store_unaligned (t##s##x##c v,       \
							  void *p)            \
  {                                                                           \
    _mm512_storeu_si512 ((__m512i *) p, (__m512i) v);                         \
  }                                                                           \
                                                                              \
  static_always_inline int t##s##x##c##_is_all_zero (t##s##x##c v)            \
  {                                                                           \
    return (_mm512_test_epi64_mask ((__m512i) v, (__m512i) v) == 0);          \
  }                                                                           \
                                                                              \
  static_always_inline int t##s##x##c##_is_equal (t##s##x##c a, t##s##x##c b) \
  {                                                                           \
    return (_mm512_cmpneq_epi64_mask ((__m512i) a, (__m512i) b) == 0);        \
  }                                                                           \
                                                                              \
  static_always_inline int t##s##x##c##_is_all_equal (t##s##x##c v, t##s x)   \
  {                                                                           \
    return t##s##x##c##_is_equal (v, t##s##x##c##_splat (x));                 \
  }                                                                           \
                                                                              \
  static_always_inline u##c t##s##x##c##_is_zero_mask (t##s##x##c v)          \
  {                                                                           \
    return _mm512_test_##i##_mask ((__m512i) v, (__m512i) v);                 \
  }                                                                           \
                                                                              \
  static_always_inline t##s##x##c t##s##x##c##_interleave_lo (t##s##x##c a,   \
							      t##s##x##c b)   \
  {                                                                           \
    return (t##s##x##c) _mm512_unpacklo_##i ((__m512i) a, (__m512i) b);       \
  }                                                                           \
                                                                              \
  static_always_inline t##s##x##c t##s##x##c##_interleave_hi (t##s##x##c a,   \
							      t##s##x##c b)   \
  {                                                                           \
    return (t##s##x##c) _mm512_unpackhi_##i ((__m512i) a, (__m512i) b);       \
  }

foreach_avx512_vec512i foreach_avx512_vec512u
#undef _
/* *INDENT-ON* */

static_always_inline u32
u16x32_msb_mask (u16x32 v)
{
  return (u32) _mm512_movepi16_mask ((__m512i) v);
}

#define u64x8_i64gather(index, base, scale)                                   \
  (u64x8) _mm512_i64gather_epi64 ((__m512i) index, base, scale)

/* 512-bit packs */
#define _(f, t, fn)                                                           \
  always_inline t t##_pack (f lo, f hi)                                       \
  {                                                                           \
    return (t) fn ((__m512i) lo, (__m512i) hi);                               \
  }

_ (i16x32, i8x64, _mm512_packs_epi16)
_ (i16x32, u8x64, _mm512_packus_epi16)
_ (i32x16, i16x32, _mm512_packs_epi32)
_ (i32x16, u16x32, _mm512_packus_epi32)
#undef _

static_always_inline u64x8
u64x8_byte_swap (u64x8 v)
{
  u8x64 swap = {
    7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8,
    7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8,
    7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8,
    7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8,
  };
  return (u64x8) _mm512_shuffle_epi8 ((__m512i) v, (__m512i) swap);
}

static_always_inline u32x16
u32x16_byte_swap (u32x16 v)
{
  u8x64 swap = {
    3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12,
    3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12,
    3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12,
    3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
  };
  return (u32x16) _mm512_shuffle_epi8 ((__m512i) v, (__m512i) swap);
}

static_always_inline u16x32
u16x32_byte_swap (u16x32 v)
{
  u8x64 swap = {
    1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14,
    1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14,
    1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14,
    1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14
  };
  return (u16x32) _mm512_shuffle_epi8 ((__m512i) v, (__m512i) swap);
}

#define _(f, t)                                                               \
  static_always_inline t f##_extract_lo (f v)                                 \
  {                                                                           \
    return (t) _mm512_extracti64x4_epi64 ((__m512i) v, 0);                    \
  }                                                                           \
  static_always_inline t f##_extract_hi (f v)                                 \
  {                                                                           \
    return (t) _mm512_extracti64x4_epi64 ((__m512i) v, 1);                    \
  }

_ (u64x8, u64x4)
_ (u32x16, u32x8)
_ (u16x32, u16x16)
_ (u8x64, u8x32)
#undef _

static_always_inline u32
u32x16_min_scalar (u32x16 v)
{
  return u32x8_min_scalar (u32x8_min (u32x16_extract_lo (v),
				      u32x16_extract_hi (v)));
}

static_always_inline u32x16
u32x16_insert_lo (u32x16 r, u32x8 v)
{
  return (u32x16) _mm512_inserti64x4 ((__m512i) r, (__m256i) v, 0);
}

static_always_inline u32x16
u32x16_insert_hi (u32x16 r, u32x8 v)
{
  return (u32x16) _mm512_inserti64x4 ((__m512i) r, (__m256i) v, 1);
}

static_always_inline u64x8
u64x8_permute (u64x8 a, u64x8 b, u64x8 mask)
{
  return (u64x8) _mm512_permutex2var_epi64 ((__m512i) a, (__m512i) mask,
					    (__m512i) b);
}


#define u32x16_ternary_logic(a, b, c, d) \
  (u32x16) _mm512_ternarylogic_epi32 ((__m512i) a, (__m512i) b, (__m512i) c, d)

#define u8x64_insert_u8x16(a, b, n) \
  (u8x64) _mm512_inserti64x2 ((__m512i) (a), (__m128i) (b), n)

#define u8x64_extract_u8x16(a, n) \
  (u8x16) _mm512_extracti64x2_epi64 ((__m512i) (a), n)

#define u8x64_word_shift_left(a,n)  (u8x64) _mm512_bslli_epi128((__m512i) a, n)
#define u8x64_word_shift_right(a,n) (u8x64) _mm512_bsrli_epi128((__m512i) a, n)

static_always_inline u8x64
u8x64_xor3 (u8x64 a, u8x64 b, u8x64 c)
{
  return (u8x64) _mm512_ternarylogic_epi32 ((__m512i) a, (__m512i) b,
					    (__m512i) c, 0x96);
}

static_always_inline u64x8
u64x8_xor3 (u64x8 a, u64x8 b, u64x8 c)
{
  return (u64x8) _mm512_ternarylogic_epi32 ((__m512i) a, (__m512i) b,
					    (__m512i) c, 0x96);
}

static_always_inline u8x64
u8x64_reflect_u8x16 (u8x64 x)
{
  static const u8x64 mask = {
    15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
    15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
    15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
    15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
  };
  return (u8x64) _mm512_shuffle_epi8 ((__m512i) x, (__m512i) mask);
}

#define u8x64_align_right(a, b, imm) \
  (u8x64) _mm512_alignr_epi8 ((__m512i) a, (__m512i) b, imm)

#define u64x8_align_right(a, b, imm)                                          \
  (u64x8) _mm512_alignr_epi64 ((__m512i) a, (__m512i) b, imm)

static_always_inline u32
u32x16_sum_elts (u32x16 sum16)
{
  u32x8 sum8;
  sum16 += (u32x16) u8x64_align_right (sum16, sum16, 8);
  sum16 += (u32x16) u8x64_align_right (sum16, sum16, 4);
  sum8 = u32x16_extract_hi (sum16) + u32x16_extract_lo (sum16);
  return sum8[0] + sum8[4];
}

#define _(t, m, p, i, e)                                                      \
  static_always_inline t t##_mask_load (t a, void *p, m mask)                 \
  {                                                                           \
    return (t) p##_mask_loadu_##e ((i) a, mask, p);                           \
  }                                                                           \
  static_always_inline t t##_mask_load_zero (void *p, m mask)                 \
  {                                                                           \
    return (t) p##_maskz_loadu_##e (mask, p);                                 \
  }                                                                           \
  static_always_inline void t##_mask_store (t a, void *p, m mask)             \
  {                                                                           \
    p##_mask_storeu_##e (p, mask, (i) a);                                     \
  }

_ (u8x64, u64, _mm512, __m512i, epi8)
_ (u8x32, u32, _mm256, __m256i, epi8)
_ (u8x16, u16, _mm, __m128i, epi8)
_ (u16x32, u32, _mm512, __m512i, epi16)
_ (u16x16, u16, _mm256, __m256i, epi16)
_ (u16x8, u8, _mm, __m128i, epi16)
_ (u32x16, u16, _mm512, __m512i, epi32)
_ (u32x8, u8, _mm256, __m256i, epi32)
_ (u32x4, u8, _mm, __m128i, epi32)
_ (u64x8, u8, _mm512, __m512i, epi64)
_ (u64x4, u8, _mm256, __m256i, epi64)
_ (u64x2, u8, _mm, __m128i, epi64)
#undef _

#define _(t, m, p, i, e)                                                      \
  static_always_inline t t##_mask_and (t a, t b, m mask)                      \
  {                                                                           \
    return (t) p##_mask_and_##e ((i) a, mask, (i) a, (i) b);                  \
  }                                                                           \
  static_always_inline t t##_mask_andnot (t a, t b, m mask)                   \
  {                                                                           \
    return (t) p##_mask_andnot_##e ((i) a, mask, (i) a, (i) b);               \
  }                                                                           \
  static_always_inline t t##_mask_xor (t a, t b, m mask)                      \
  {                                                                           \
    return (t) p##_mask_xor_##e ((i) a, mask, (i) a, (i) b);                  \
  }                                                                           \
  static_always_inline t t##_mask_or (t a, t b, m mask)                       \
  {                                                                           \
    return (t) p##_mask_or_##e ((i) a, mask, (i) a, (i) b);                   \
  }
_ (u32x16, u16, _mm512, __m512i, epi32)
_ (u32x8, u8, _mm256, __m256i, epi32)
_ (u32x4, u8, _mm, __m128i, epi32)
_ (u64x8, u8, _mm512, __m512i, epi64)
_ (u64x4, u8, _mm256, __m256i, epi64)
_ (u64x2, u8, _mm, __m128i, epi64)
#undef _

#ifdef CLIB_HAVE_VEC512
#define CLIB_HAVE_VEC512_MASK_LOAD_STORE
#define CLIB_HAVE_VEC512_MASK_BITWISE_OPS
#endif
#ifdef CLIB_HAVE_VEC256
#define CLIB_HAVE_VEC256_MASK_LOAD_STORE
#define CLIB_HAVE_VEC256_MASK_BITWISE_OPS
#endif
#ifdef CLIB_HAVE_VEC128
#define CLIB_HAVE_VEC128_MASK_LOAD_STORE
#define CLIB_HAVE_VEC128_MASK_BITWISE_OPS
#endif

static_always_inline u8x64
u8x64_splat_u8x16 (u8x16 a)
{
  return (u8x64) _mm512_broadcast_i64x2 ((__m128i) a);
}

static_always_inline u32x16
u32x16_splat_u32x4 (u32x4 a)
{
  return (u32x16) _mm512_broadcast_i64x2 ((__m128i) a);
}

static_always_inline u32x16
u32x16_mask_blend (u32x16 a, u32x16 b, u16 mask)
{
  return (u32x16) _mm512_mask_blend_epi32 (mask, (__m512i) a, (__m512i) b);
}

static_always_inline u8x64
u8x64_mask_blend (u8x64 a, u8x64 b, u64 mask)
{
  return (u8x64) _mm512_mask_blend_epi8 (mask, (__m512i) a, (__m512i) b);
}

static_always_inline u8x64
u8x64_permute (u8x64 idx, u8x64 a)
{
  return (u8x64) _mm512_permutexvar_epi8 ((__m512i) idx, (__m512i) a);
}

static_always_inline u8x64
u8x64_permute2 (u8x64 idx, u8x64 a, u8x64 b)
{
  return (u8x64) _mm512_permutex2var_epi8 ((__m512i) a, (__m512i) idx,
					   (__m512i) b);
}

#define _(t, m, e, p, it)                                                     \
  static_always_inline m t##_is_equal_mask (t a, t b)                         \
  {                                                                           \
    return p##_cmpeq_##e##_mask ((it) a, (it) b);                             \
  }
_ (u8x16, u16, epu8, _mm, __m128i)
_ (u16x8, u8, epu16, _mm, __m128i)
_ (u32x4, u8, epu32, _mm, __m128i)
_ (u64x2, u8, epu64, _mm, __m128i)

_ (u8x32, u32, epu8, _mm256, __m256i)
_ (u16x16, u16, epu16, _mm256, __m256i)
_ (u32x8, u8, epu32, _mm256, __m256i)
_ (u64x4, u8, epu64, _mm256, __m256i)

_ (u8x64, u64, epu8, _mm512, __m512i)
_ (u16x32, u32, epu16, _mm512, __m512i)
_ (u32x16, u16, epu32, _mm512, __m512i)
_ (u64x8, u8, epu64, _mm512, __m512i)
#undef _

#define _(t, m, e, p, it)                                                     \
  static_always_inline m t##_is_not_equal_mask (t a, t b)                     \
  {                                                                           \
    return p##_cmpneq_##e##_mask ((it) a, (it) b);                            \
  }
_ (u8x16, u16, epu8, _mm, __m128i)
_ (u16x8, u8, epu16, _mm, __m128i)
_ (u32x4, u8, epu32, _mm, __m128i)
_ (u64x2, u8, epu64, _mm, __m128i)

_ (u8x32, u32, epu8, _mm256, __m256i)
_ (u16x16, u16, epu16, _mm256, __m256i)
_ (u32x8, u8, epu32, _mm256, __m256i)
_ (u64x4, u8, epu64, _mm256, __m256i)

_ (u8x64, u64, epu8, _mm512, __m512i)
_ (u16x32, u32, epu16, _mm512, __m512i)
_ (u32x16, u16, epu32, _mm512, __m512i)
_ (u64x8, u8, epu64, _mm512, __m512i)
#undef _

#define _(f, t, fn, it)                                                       \
  static_always_inline t t##_from_##f (f x) { return (t) fn ((it) x); }
_ (u16x16, u32x16, _mm512_cvtepi16_epi32, __m256i)
_ (u32x16, u16x16, _mm512_cvtusepi32_epi16, __m512i)
_ (u32x8, u16x8, _mm256_cvtusepi32_epi16, __m256i)
_ (u32x8, u64x8, _mm512_cvtepu32_epi64, __m256i)
#undef _

#define _(vt, mt, p, it, epi)                                                 \
  static_always_inline vt vt##_compress (vt a, mt mask)                       \
  {                                                                           \
    return (vt) p##_maskz_compress_##epi (mask, (it) a);                      \
  }                                                                           \
  static_always_inline vt vt##_expand (vt a, mt mask)                         \
  {                                                                           \
    return (vt) p##_maskz_expand_##epi (mask, (it) a);                        \
  }                                                                           \
  static_always_inline void vt##_compress_store (vt v, mt mask, void *p)      \
  {                                                                           \
    p##_mask_compressstoreu_##epi (p, mask, (it) v);                          \
  }

_ (u64x8, u8, _mm512, __m512i, epi64)
_ (u32x16, u16, _mm512, __m512i, epi32)
_ (u64x4, u8, _mm256, __m256i, epi64)
_ (u32x8, u8, _mm256, __m256i, epi32)
_ (u64x2, u8, _mm, __m128i, epi64)
_ (u32x4, u8, _mm, __m128i, epi32)
#ifdef __AVX512VBMI2__
_ (u16x32, u32, _mm512, __m512i, epi16)
_ (u8x64, u64, _mm512, __m512i, epi8)
_ (u16x16, u16, _mm256, __m256i, epi16)
_ (u8x32, u32, _mm256, __m256i, epi8)
_ (u16x8, u8, _mm, __m128i, epi16)
_ (u8x16, u16, _mm, __m128i, epi8)
#endif
#undef _

#ifdef CLIB_HAVE_VEC256
#define CLIB_HAVE_VEC256_COMPRESS
#ifdef __AVX512VBMI2__
#define CLIB_HAVE_VEC256_COMPRESS_U8_U16
#endif

#endif
#ifdef CLIB_HAVE_VEC512
#define CLIB_HAVE_VEC512_COMPRESS
#ifdef __AVX512VBMI2__
#define CLIB_HAVE_VEC512_COMPRESS_U8_U16
#endif

#endif

#ifndef __AVX512VBMI2__
static_always_inline u16x16
u16x16_compress (u16x16 v, u16 mask)
{
  return u16x16_from_u32x16 (u32x16_compress (u32x16_from_u16x16 (v), mask));
}

static_always_inline u16x8
u16x8_compress (u16x8 v, u8 mask)
{
  return u16x8_from_u32x8 (u32x8_compress (u32x8_from_u16x8 (v), mask));
}
#endif

static_always_inline u64
u64x8_hxor (u64x8 v)
{
  v ^= u64x8_align_right (v, v, 4);
  v ^= u64x8_align_right (v, v, 2);
  return v[0] ^ v[1];
}

static_always_inline void
u32x16_transpose (u32x16 m[16])
{
  __m512i r[16], a, b, c, d, x, y;

  /* *INDENT-OFF* */
  __m512i pm1 = (__m512i) (u64x8) { 0, 1, 8, 9, 4, 5, 12, 13};
  __m512i pm2 = (__m512i) (u64x8) { 2, 3, 10, 11, 6, 7, 14, 15};
  __m512i pm3 = (__m512i) (u64x8) { 0, 1, 2, 3, 8, 9, 10, 11};
  __m512i pm4 = (__m512i) (u64x8) { 4, 5, 6, 7, 12, 13, 14, 15};
  /* *INDENT-ON* */

  r[0] = _mm512_unpacklo_epi32 ((__m512i) m[0], (__m512i) m[1]);
  r[1] = _mm512_unpacklo_epi32 ((__m512i) m[2], (__m512i) m[3]);
  r[2] = _mm512_unpacklo_epi32 ((__m512i) m[4], (__m512i) m[5]);
  r[3] = _mm512_unpacklo_epi32 ((__m512i) m[6], (__m512i) m[7]);
  r[4] = _mm512_unpacklo_epi32 ((__m512i) m[8], (__m512i) m[9]);
  r[5] = _mm512_unpacklo_epi32 ((__m512i) m[10], (__m512i) m[11]);
  r[6] = _mm512_unpacklo_epi32 ((__m512i) m[12], (__m512i) m[13]);
  r[7] = _mm512_unpacklo_epi32 ((__m512i) m[14], (__m512i) m[15]);

  r[8] = _mm512_unpackhi_epi32 ((__m512i) m[0], (__m512i) m[1]);
  r[9] = _mm512_unpackhi_epi32 ((__m512i) m[2], (__m512i) m[3]);
  r[10] = _mm512_unpackhi_epi32 ((__m512i) m[4], (__m512i) m[5]);
  r[11] = _mm512_unpackhi_epi32 ((__m512i) m[6], (__m512i) m[7]);
  r[12] = _mm512_unpackhi_epi32 ((__m512i) m[8], (__m512i) m[9]);
  r[13] = _mm512_unpackhi_epi32 ((__m512i) m[10], (__m512i) m[11]);
  r[14] = _mm512_unpackhi_epi32 ((__m512i) m[12], (__m512i) m[13]);
  r[15] = _mm512_unpackhi_epi32 ((__m512i) m[14], (__m512i) m[15]);

  a = _mm512_unpacklo_epi64 (r[0], r[1]);
  b = _mm512_unpacklo_epi64 (r[2], r[3]);
  c = _mm512_unpacklo_epi64 (r[4], r[5]);
  d = _mm512_unpacklo_epi64 (r[6], r[7]);
  x = _mm512_permutex2var_epi64 (a, pm1, b);
  y = _mm512_permutex2var_epi64 (c, pm1, d);
  m[0] = (u32x16) _mm512_permutex2var_epi64 (x, pm3, y);
  m[8] = (u32x16) _mm512_permutex2var_epi64 (x, pm4, y);
  x = _mm512_permutex2var_epi64 (a, pm2, b);
  y = _mm512_permutex2var_epi64 (c, pm2, d);
  m[4] = (u32x16) _mm512_permutex2var_epi64 (x, pm3, y);
  m[12] = (u32x16) _mm512_permutex2var_epi64 (x, pm4, y);

  a = _mm512_unpacklo_epi64 (r[8], r[9]);
  b = _mm512_unpacklo_epi64 (r[10], r[11]);
  c = _mm512_unpacklo_epi64 (r[12], r[13]);
  d = _mm512_unpacklo_epi64 (r[14], r[15]);
  x = _mm512_permutex2var_epi64 (a, pm1, b);
  y = _mm512_permutex2var_epi64 (c, pm1, d);
  m[2] = (u32x16) _mm512_permutex2var_epi64 (x, pm3, y);
  m[10] = (u32x16) _mm512_permutex2var_epi64 (x, pm4, y);
  x = _mm512_permutex2var_epi64 (a, pm2, b);
  y = _mm512_permutex2var_epi64 (c, pm2, d);
  m[6] = (u32x16) _mm512_permutex2var_epi64 (x, pm3, y);
  m[14] = (u32x16) _mm512_permutex2var_epi64 (x, pm4, y);

  a = _mm512_unpackhi_epi64 (r[0], r[1]);
  b = _mm512_unpackhi_epi64 (r[2], r[3]);
  c = _mm512_unpackhi_epi64 (r[4], r[5]);
  d = _mm512_unpackhi_epi64 (r[6], r[7]);
  x = _mm512_permutex2var_epi64 (a, pm1, b);
  y = _mm512_permutex2var_epi64 (c, pm1, d);
  m[1] = (u32x16) _mm512_permutex2var_epi64 (x, pm3, y);
  m[9] = (u32x16) _mm512_permutex2var_epi64 (x, pm4, y);
  x = _mm512_permutex2var_epi64 (a, pm2, b);
  y = _mm512_permutex2var_epi64 (c, pm2, d);
  m[5] = (u32x16) _mm512_permutex2var_epi64 (x, pm3, y);
  m[13] = (u32x16) _mm512_permutex2var_epi64 (x, pm4, y);

  a = _mm512_unpackhi_epi64 (r[8], r[9]);
  b = _mm512_unpackhi_epi64 (r[10], r[11]);
  c = _mm512_unpackhi_epi64 (r[12], r[13]);
  d = _mm512_unpackhi_epi64 (r[14], r[15]);
  x = _mm512_permutex2var_epi64 (a, pm1, b);
  y = _mm512_permutex2var_epi64 (c, pm1, d);
  m[3] = (u32x16) _mm512_permutex2var_epi64 (x, pm3, y);
  m[11] = (u32x16) _mm512_permutex2var_epi64 (x, pm4, y);
  x = _mm512_permutex2var_epi64 (a, pm2, b);
  y = _mm512_permutex2var_epi64 (c, pm2, d);
  m[7] = (u32x16) _mm512_permutex2var_epi64 (x, pm3, y);
  m[15] = (u32x16) _mm512_permutex2var_epi64 (x, pm4, y);
}



static_always_inline void
u64x8_transpose (u64x8 m[8])
{
  __m512i r[8], x, y;

  /* *INDENT-OFF* */
  __m512i pm1 = (__m512i) (u64x8) { 0, 1, 8, 9, 4, 5, 12, 13};
  __m512i pm2 = (__m512i) (u64x8) { 2, 3, 10, 11, 6, 7, 14, 15};
  __m512i pm3 = (__m512i) (u64x8) { 0, 1, 2, 3, 8, 9, 10, 11};
  __m512i pm4 = (__m512i) (u64x8) { 4, 5, 6, 7, 12, 13, 14, 15};
  /* *INDENT-ON* */

  r[0] = _mm512_unpacklo_epi64 ((__m512i) m[0], (__m512i) m[1]);
  r[1] = _mm512_unpacklo_epi64 ((__m512i) m[2], (__m512i) m[3]);
  r[2] = _mm512_unpacklo_epi64 ((__m512i) m[4], (__m512i) m[5]);
  r[3] = _mm512_unpacklo_epi64 ((__m512i) m[6], (__m512i) m[7]);
  r[4] = _mm512_unpackhi_epi64 ((__m512i) m[0], (__m512i) m[1]);
  r[5] = _mm512_unpackhi_epi64 ((__m512i) m[2], (__m512i) m[3]);
  r[6] = _mm512_unpackhi_epi64 ((__m512i) m[4], (__m512i) m[5]);
  r[7] = _mm512_unpackhi_epi64 ((__m512i) m[6], (__m512i) m[7]);

  x = _mm512_permutex2var_epi64 (r[0], pm1, r[1]);
  y = _mm512_permutex2var_epi64 (r[2], pm1, r[3]);
  m[0] = (u64x8) _mm512_permutex2var_epi64 (x, pm3, y);
  m[4] = (u64x8) _mm512_permutex2var_epi64 (x, pm4, y);
  x = _mm512_permutex2var_epi64 (r[0], pm2, r[1]);
  y = _mm512_permutex2var_epi64 (r[2], pm2, r[3]);
  m[2] = (u64x8) _mm512_permutex2var_epi64 (x, pm3, y);
  m[6] = (u64x8) _mm512_permutex2var_epi64 (x, pm4, y);

  x = _mm512_permutex2var_epi64 (r[4], pm1, r[5]);
  y = _mm512_permutex2var_epi64 (r[6], pm1, r[7]);
  m[1] = (u64x8) _mm512_permutex2var_epi64 (x, pm3, y);
  m[5] = (u64x8) _mm512_permutex2var_epi64 (x, pm4, y);
  x = _mm512_permutex2var_epi64 (r[4], pm2, r[5]);
  y = _mm512_permutex2var_epi64 (r[6], pm2, r[7]);
  m[3] = (u64x8) _mm512_permutex2var_epi64 (x, pm3, y);
  m[7] = (u64x8) _mm512_permutex2var_epi64 (x, pm4, y);
}

#endif /* included_vector_avx512_h */
/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */