summaryrefslogtreecommitdiffstats
path: root/src/vppinfra/vector_avx2.h
blob: 866c82fcec3854b60dcc747a46bf87df4fa4d7a3 (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
/*
 * Copyright (c) 2018 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_avx2_h
#define included_vector_avx2_h

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

#define foreach_avx2_vec256i \
  _(i,8,32,epi8) _(i,16,16,epi16) _(i,32,8,epi32)  _(i,64,4,epi64)
#define foreach_avx2_vec256u \
  _(u,8,32,epi8) _(u,16,16,epi16) _(u,32,8,epi32)  _(u,64,4,epi64)
#define foreach_avx2_vec256f \
  _(f,32,8,ps) _(f,64,4,pd)

#define _mm256_set1_epi64 _mm256_set1_epi64x

/* splat, load_unaligned, store_unaligned, is_all_zero, is_equal,
   is_all_equal */
#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) _mm256_set1_##i (x); }				\
\
static_always_inline t##s##x##c						\
t##s##x##c##_load_unaligned (void *p)					\
{ return (t##s##x##c) _mm256_loadu_si256 (p); }				\
\
static_always_inline void						\
t##s##x##c##_store_unaligned (t##s##x##c v, void *p)			\
{ _mm256_storeu_si256 ((__m256i *) p, (__m256i) v); }			\
\
static_always_inline int						\
t##s##x##c##_is_all_zero (t##s##x##c x)					\
{ return _mm256_testz_si256 ((__m256i) x, (__m256i) x); }		\
\
static_always_inline int						\
t##s##x##c##_is_equal (t##s##x##c a, t##s##x##c b)			\
{ return t##s##x##c##_is_all_zero (a ^ b); }				\
\
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 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) _mm256_unpacklo_##i ((__m256i) a, (__m256i) 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) _mm256_unpackhi_##i ((__m256i) a, (__m256i) b); } \


foreach_avx2_vec256i foreach_avx2_vec256u
#undef _

always_inline u32x8
u32x8_permute (u32x8 v, u32x8 idx)
{
  return (u32x8) _mm256_permutevar8x32_epi32 ((__m256i) v, (__m256i) idx);
}

#define u64x4_permute(v, m0, m1, m2, m3)                                      \
  (u64x4) _mm256_permute4x64_epi64 (                                          \
    (__m256i) v, ((m0) | (m1) << 2 | (m2) << 4 | (m3) << 6))

/* _extract_lo, _extract_hi */
#define _(t1,t2) \
always_inline t1							\
t2##_extract_lo (t2 v)							\
{ return (t1) _mm256_extracti128_si256 ((__m256i) v, 0); }		\
\
always_inline t1							\
t2##_extract_hi (t2 v)							\
{ return (t1) _mm256_extracti128_si256 ((__m256i) v, 1); }		\
\
always_inline t2							\
t2##_insert_lo (t2 v1, t1 v2)						\
{ return (t2) _mm256_inserti128_si256 ((__m256i) v1, (__m128i) v2, 0); }\
\
always_inline t2							\
t2##_insert_hi (t2 v1, t1 v2)						\
{ return (t2) _mm256_inserti128_si256 ((__m256i) v1, (__m128i) v2, 1); }\

_(u8x16, u8x32)
_(u16x8, u16x16)
_(u32x4, u32x8)
_(u64x2, u64x4)
#undef _

/* 256 bit packs. */
#define _(f, t, fn)                                                           \
  always_inline t t##_pack (f lo, f hi)                                       \
  {                                                                           \
    return (t) fn ((__m256i) lo, (__m256i) hi);                               \
  }

_ (i16x16, i8x32, _mm256_packs_epi16)
_ (i16x16, u8x32, _mm256_packus_epi16)
_ (i32x8, i16x16, _mm256_packs_epi32)
_ (i32x8, u16x16, _mm256_packus_epi32)

#undef _

static_always_inline u32
u8x32_msb_mask (u8x32 v)
{
  return _mm256_movemask_epi8 ((__m256i) v);
}

static_always_inline u32
i8x32_msb_mask (i8x32 v)
{
  return _mm256_movemask_epi8 ((__m256i) v);
}

/* _from_ */
#define _(f,t,i) \
static_always_inline t							\
t##_from_##f (f x)							\
{ return (t) _mm256_cvt##i ((__m128i) x); }

_(u16x8, u32x8, epu16_epi32)
_(u16x8, u64x4, epu16_epi64)
_(u32x4, u64x4, epu32_epi64)
_ (u8x16, u16x16, epu8_epi16)
_(u8x16, u32x8, epu8_epi32)
_(u8x16, u64x4, epu8_epi64)
_(i16x8, i32x8, epi16_epi32)
_(i16x8, i64x4, epi16_epi64)
_(i32x4, i64x4, epi32_epi64)
_ (i8x16, i16x16, epi8_epi16)
_(i8x16, i32x8, epi8_epi32)
_(i8x16, i64x4, epi8_epi64)
#undef _

static_always_inline u64x4
u64x4_byte_swap (u64x4 v)
{
  u8x32 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,
  };
  return (u64x4) _mm256_shuffle_epi8 ((__m256i) v, (__m256i) swap);
}

static_always_inline u32x8
u32x8_byte_swap (u32x8 v)
{
  u8x32 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
  };
  return (u32x8) _mm256_shuffle_epi8 ((__m256i) v, (__m256i) swap);
}

static_always_inline u16x16
u16x16_byte_swap (u16x16 v)
{
  u8x32 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
  };
  return (u16x16) _mm256_shuffle_epi8 ((__m256i) v, (__m256i) swap);
}

#define u8x32_align_right(a, b, imm) \
  (u8x32) _mm256_alignr_epi8 ((__m256i) a, (__m256i) b, imm)

#define u64x4_align_right(a, b, imm)                                          \
  (u64x4) _mm256_alignr_epi64 ((__m256i) a, (__m256i) b, imm)

static_always_inline u32
u32x8_sum_elts (u32x8 sum8)
{
  sum8 += (u32x8) u8x32_align_right (sum8, sum8, 8);
  sum8 += (u32x8) u8x32_align_right (sum8, sum8, 4);
  return sum8[0] + sum8[4];
}

static_always_inline u32x8
u32x8_hadd (u32x8 v1, u32x8 v2)
{
  return (u32x8) _mm256_hadd_epi32 ((__m256i) v1, (__m256i) v2);
}

static_always_inline u32
u32x8_hxor (u32x8 v)
{
  u32x4 v4;
  v4 = u32x8_extract_lo (v) ^ u32x8_extract_hi (v);
  v4 ^= (u32x4) u8x16_align_right (v4, v4, 8);
  v4 ^= (u32x4) u8x16_align_right (v4, v4, 4);
  return v4[0];
}

static_always_inline u8x32
u8x32_xor3 (u8x32 a, u8x32 b, u8x32 c)
{
#if __AVX512F__
  return (u8x32) _mm256_ternarylogic_epi32 ((__m256i) a, (__m256i) b,
					    (__m256i) c, 0x96);
#endif
  return a ^ b ^ c;
}

static_always_inline u8x32
u8x32_reflect_u8x16 (u8x32 x)
{
  static const u8x32 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,
  };
  return (u8x32) _mm256_shuffle_epi8 ((__m256i) x, (__m256i) mask);
}

static_always_inline u16x16
u16x16_mask_last (u16x16 v, u8 n_last)
{
  const u16x16 masks[17] = {
    {0},
    {-1},
    {-1, -1},
    {-1, -1, -1},
    {-1, -1, -1, -1},
    {-1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1},
    {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1},
  };

  ASSERT (n_last < 17);

  return v & masks[16 - n_last];
}

static_always_inline f32x8
f32x8_from_u32x8 (u32x8 v)
{
  return (f32x8) _mm256_cvtepi32_ps ((__m256i) v);
}

static_always_inline u32x8
u32x8_from_f32x8 (f32x8 v)
{
  return (u32x8) _mm256_cvttps_epi32 ((__m256) v);
}

#define u32x8_blend(a,b,m) \
  (u32x8) _mm256_blend_epi32 ((__m256i) a, (__m256i) b, m)

#define u16x16_blend(v1, v2, mask) \
  (u16x16) _mm256_blend_epi16 ((__m256i) (v1), (__m256i) (v2), mask)

static_always_inline u64x4
u64x4_gather (void *p0, void *p1, void *p2, void *p3)
{
  u64x4 r = {
    *(u64 *) p0, *(u64 *) p1, *(u64 *) p2, *(u64 *) p3
  };
  return r;
}

static_always_inline u32x8
u32x8_gather (void *p0, void *p1, void *p2, void *p3, void *p4, void *p5,
	      void *p6, void *p7)
{
  u32x8 r = {
    *(u32 *) p0, *(u32 *) p1, *(u32 *) p2, *(u32 *) p3,
    *(u32 *) p4, *(u32 *) p5, *(u32 *) p6, *(u32 *) p7,
  };
  return r;
}


static_always_inline void
u64x4_scatter (u64x4 r, void *p0, void *p1, void *p2, void *p3)
{
  *(u64 *) p0 = r[0];
  *(u64 *) p1 = r[1];
  *(u64 *) p2 = r[2];
  *(u64 *) p3 = r[3];
}

static_always_inline void
u32x8_scatter (u32x8 r, void *p0, void *p1, void *p2, void *p3, void *p4,
	       void *p5, void *p6, void *p7)
{
  *(u32 *) p0 = r[0];
  *(u32 *) p1 = r[1];
  *(u32 *) p2 = r[2];
  *(u32 *) p3 = r[3];
  *(u32 *) p4 = r[4];
  *(u32 *) p5 = r[5];
  *(u32 *) p6 = r[6];
  *(u32 *) p7 = r[7];
}

static_always_inline void
u64x4_scatter_one (u64x4 r, int index, void *p)
{
  *(u64 *) p = r[index];
}

static_always_inline void
u32x8_scatter_one (u32x8 r, int index, void *p)
{
  *(u32 *) p = r[index];
}

#define u32x8_gather_u32(base, indices, scale)                                \
  (u32x8) _mm256_i32gather_epi32 ((const int *) base, (__m256i) indices, scale)

#ifdef __AVX512F__
#define u32x8_scatter_u32(base, indices, v, scale)                            \
  _mm256_i32scatter_epi32 (base, (__m256i) indices, (__m256i) v, scale)
#else
#define u32x8_scatter_u32(base, indices, v, scale)                            \
  for (u32 i = 0; i < 8; i++)                                                 \
    *((u32u *) ((u8 *) base + (scale) * (indices)[i])) = (v)[i];
#endif

static_always_inline u8x32
u8x32_blend (u8x32 v1, u8x32 v2, u8x32 mask)
{
  return (u8x32) _mm256_blendv_epi8 ((__m256i) v1, (__m256i) v2,
				     (__m256i) mask);
}

#define u8x32_word_shift_left(a, n)                                           \
  (u8x32) _mm256_bslli_epi128 ((__m256i) a, n)
#define u8x32_word_shift_right(a, n)                                          \
  (u8x32) _mm256_bsrli_epi128 ((__m256i) a, n)

#define u32x8_permute_lanes(a, b, m) \
  (u32x8) _mm256_permute2x128_si256 ((__m256i) a, (__m256i) b, m)
#define u64x4_permute_lanes(a, b, m) \
  (u64x4) _mm256_permute2x128_si256 ((__m256i) a, (__m256i) b, m)

static_always_inline u32x8
u32x8_min (u32x8 a, u32x8 b)
{
  return (u32x8) _mm256_min_epu32 ((__m256i) a, (__m256i) b);
}

static_always_inline u32
u32x8_min_scalar (u32x8 v)
{
  return u32x4_min_scalar (u32x4_min (u32x8_extract_lo (v),
				      u32x8_extract_hi (v)));
}

static_always_inline void
u32x8_transpose (u32x8 a[8])
{
  u64x4 r[8], x, y;

  r[0] = (u64x4) u32x8_interleave_lo (a[0], a[1]);
  r[1] = (u64x4) u32x8_interleave_hi (a[0], a[1]);
  r[2] = (u64x4) u32x8_interleave_lo (a[2], a[3]);
  r[3] = (u64x4) u32x8_interleave_hi (a[2], a[3]);
  r[4] = (u64x4) u32x8_interleave_lo (a[4], a[5]);
  r[5] = (u64x4) u32x8_interleave_hi (a[4], a[5]);
  r[6] = (u64x4) u32x8_interleave_lo (a[6], a[7]);
  r[7] = (u64x4) u32x8_interleave_hi (a[6], a[7]);

  x = u64x4_interleave_lo (r[0], r[2]);
  y = u64x4_interleave_lo (r[4], r[6]);
  a[0] = u32x8_permute_lanes (x, y, 0x20);
  a[4] = u32x8_permute_lanes (x, y, 0x31);

  x = u64x4_interleave_hi (r[0], r[2]);
  y = u64x4_interleave_hi (r[4], r[6]);
  a[1] = u32x8_permute_lanes (x, y, 0x20);
  a[5] = u32x8_permute_lanes (x, y, 0x31);

  x = u64x4_interleave_lo (r[1], r[3]);
  y = u64x4_interleave_lo (r[5], r[7]);
  a[2] = u32x8_permute_lanes (x, y, 0x20);
  a[6] = u32x8_permute_lanes (x, y, 0x31);

  x = u64x4_interleave_hi (r[1], r[3]);
  y = u64x4_interleave_hi (r[5], r[7]);
  a[3] = u32x8_permute_lanes (x, y, 0x20);
  a[7] = u32x8_permute_lanes (x, y, 0x31);
}

static_always_inline void
u64x4_transpose (u64x4 a[8])
{
  u64x4 r[4];

  r[0] = u64x4_interleave_lo (a[0], a[1]);
  r[1] = u64x4_interleave_hi (a[0], a[1]);
  r[2] = u64x4_interleave_lo (a[2], a[3]);
  r[3] = u64x4_interleave_hi (a[2], a[3]);

  a[0] = u64x4_permute_lanes (r[0], r[2], 0x20);
  a[1] = u64x4_permute_lanes (r[1], r[3], 0x20);
  a[2] = u64x4_permute_lanes (r[0], r[2], 0x31);
  a[3] = u64x4_permute_lanes (r[1], r[3], 0x31);
}

static_always_inline u8x32
u8x32_splat_u8x16 (u8x16 a)
{
  return (u8x32) _mm256_broadcastsi128_si256 ((__m128i) a);
}

static_always_inline u32x8
u32x8_splat_u32x4 (u32x4 a)
{
  return (u32x8) _mm256_broadcastsi128_si256 ((__m128i) a);
}

static_always_inline u64x4
u64x4_splat_u64x2 (u64x2 a)
{
  return (u64x4) _mm256_broadcastsi128_si256 ((__m128i) a);
}

static_always_inline u8x32
u8x32_load_partial (u8 *data, uword n)
{
#if defined(CLIB_HAVE_VEC256_MASK_LOAD_STORE)
  return u8x32_mask_load_zero (data, pow2_mask (n));
#else
  u8x32 r = {};
  if (n > 16)
    {
      r = u8x32_insert_lo (r, *(u8x16u *) data);
      r = u8x32_insert_hi (r, u8x16_load_partial (data + 16, n - 16));
    }
  else
    r = u8x32_insert_lo (r, u8x16_load_partial (data, n));
  return r;
#endif
}

static_always_inline void
u8x32_store_partial (u8x32 r, u8 *data, uword n)
{
#if defined(CLIB_HAVE_VEC256_MASK_LOAD_STORE)
  u8x32_mask_store (r, data, pow2_mask (n));
#else
  if (n > 16)
    {
      *(u8x16u *) data = u8x32_extract_lo (r);
      u8x16_store_partial (u8x32_extract_hi (r), data + 16, n - 16);
    }
  else
    u8x16_store_partial (u8x32_extract_lo (r), data, n);
#endif
}

#endif /* included_vector_avx2_h */

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