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
|
/****************************************************************************
* Copyright (C) 2017 Intel Corporation. All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
****************************************************************************/
#if !defined(__SIMD_LIB_AVX512_HPP__)
#error Do not include this file directly, use "simdlib.hpp" instead.
#endif
//============================================================================
// SIMD128 AVX (512) implementation
//
// Since this implementation inherits from the AVX (2) implementation,
// the only operations below ones that replace AVX (2) operations.
// These use native AVX512 instructions with masking to enable a larger
// register set.
//============================================================================
private:
static SIMDINLINE __m512 __conv(Float r) { return _mm512_castps128_ps512(r.v); }
static SIMDINLINE __m512d __conv(Double r) { return _mm512_castpd128_pd512(r.v); }
static SIMDINLINE __m512i __conv(Integer r) { return _mm512_castsi128_si512(r.v); }
static SIMDINLINE Float __conv(__m512 r) { return _mm512_castps512_ps128(r); }
static SIMDINLINE Double __conv(__m512d r) { return _mm512_castpd512_pd128(r); }
static SIMDINLINE Integer __conv(__m512i r) { return _mm512_castsi512_si128(r); }
public:
#define SIMD_WRAPPER_1_(op, intrin, mask) \
static SIMDINLINE Float SIMDCALL op(Float a) \
{\
return __conv(_mm512_maskz_##intrin((mask), __conv(a)));\
}
#define SIMD_WRAPPER_1(op) SIMD_WRAPPER_1_(op, op, __mmask16(0xf))
#define SIMD_WRAPPER_1I_(op, intrin, mask) \
template<int ImmT> \
static SIMDINLINE Float SIMDCALL op(Float a) \
{\
return __conv(_mm512_maskz_##intrin((mask), __conv(a), ImmT));\
}
#define SIMD_WRAPPER_1I(op) SIMD_WRAPPER_1I_(op, op, __mmask16(0xf))
#define SIMD_WRAPPER_2_(op, intrin, mask) \
static SIMDINLINE Float SIMDCALL op(Float a, Float b) \
{\
return __conv(_mm512_maskz_##intrin((mask), __conv(a), __conv(b)));\
}
#define SIMD_WRAPPER_2(op) SIMD_WRAPPER_2_(op, op, __mmask16(0xf))
#define SIMD_WRAPPER_2I(op) \
template<int ImmT>\
static SIMDINLINE Float SIMDCALL op(Float a, Float b) \
{\
return __conv(_mm512_maskz_##op(0xf, __conv(a), __conv(b), ImmT));\
}
#define SIMD_WRAPPER_3_(op, intrin, mask) \
static SIMDINLINE Float SIMDCALL op(Float a, Float b, Float c) \
{\
return __conv(_mm512_maskz_##intrin((mask), __conv(a), __conv(b), __conv(c)));\
}
#define SIMD_WRAPPER_3(op) SIMD_WRAPPER_3_(op, op, __mmask16(0xf))
#define SIMD_DWRAPPER_2I(op) \
template<int ImmT>\
static SIMDINLINE Double SIMDCALL op(Double a, Double b) \
{\
return __conv(_mm512_maskz_##op(0x3, __conv(a), __conv(b), ImmT));\
}
#define SIMD_IWRAPPER_1_(op, intrin, mask) \
static SIMDINLINE Integer SIMDCALL op(Integer a) \
{\
return __conv(_mm512_maskz_##intrin((mask), __conv(a)));\
}
#define SIMD_IWRAPPER_1_32(op) SIMD_IWRAPPER_1_(op, op, __mmask16(0xf))
#define SIMD_IWRAPPER_1I_(op, intrin, mask) \
template<int ImmT> \
static SIMDINLINE Integer SIMDCALL op(Integer a) \
{\
return __conv(_mm512_maskz_##intrin((mask), __conv(a), ImmT));\
}
#define SIMD_IWRAPPER_1I_32(op) SIMD_IWRAPPER_1I_(op, op, __mmask16(0xf))
#define SIMD_IWRAPPER_2_(op, intrin, mask) \
static SIMDINLINE Integer SIMDCALL op(Integer a, Integer b) \
{\
return __conv(_mm512_maskz_##intrin((mask), __conv(a), __conv(b)));\
}
#define SIMD_IWRAPPER_2_32(op) SIMD_IWRAPPER_2_(op, op, __mmask16(0xf))
#define SIMD_IWRAPPER_2I(op) \
template<int ImmT>\
static SIMDINLINE Integer SIMDCALL op(Integer a, Integer b) \
{\
return __conv(_mm512_maskz_##op(0xf, __conv(a), __conv(b), ImmT));\
}
//-----------------------------------------------------------------------
// Single precision floating point arithmetic operations
//-----------------------------------------------------------------------
SIMD_WRAPPER_2(add_ps); // return a + b
SIMD_WRAPPER_2(div_ps); // return a / b
SIMD_WRAPPER_3(fmadd_ps); // return (a * b) + c
SIMD_WRAPPER_3(fmsub_ps); // return (a * b) - c
SIMD_WRAPPER_2(max_ps); // return (a > b) ? a : b
SIMD_WRAPPER_2(min_ps); // return (a < b) ? a : b
SIMD_WRAPPER_2(mul_ps); // return a * b
SIMD_WRAPPER_1_(rcp_ps, rcp14_ps, __mmask16(0xf)); // return 1.0f / a
SIMD_WRAPPER_1_(rsqrt_ps, rsqrt14_ps, __mmask16(0xf)); // return 1.0f / sqrt(a)
SIMD_WRAPPER_2(sub_ps); // return a - b
//-----------------------------------------------------------------------
// Integer (various width) arithmetic operations
//-----------------------------------------------------------------------
SIMD_IWRAPPER_1_32(abs_epi32); // return absolute_value(a) (int32)
SIMD_IWRAPPER_2_32(add_epi32); // return a + b (int32)
SIMD_IWRAPPER_2_32(max_epi32); // return (a > b) ? a : b (int32)
SIMD_IWRAPPER_2_32(max_epu32); // return (a > b) ? a : b (uint32)
SIMD_IWRAPPER_2_32(min_epi32); // return (a < b) ? a : b (int32)
SIMD_IWRAPPER_2_32(min_epu32); // return (a < b) ? a : b (uint32)
SIMD_IWRAPPER_2_32(mul_epi32); // return a * b (int32)
// SIMD_IWRAPPER_2_8(add_epi8); // return a + b (int8)
// SIMD_IWRAPPER_2_8(adds_epu8); // return ((a + b) > 0xff) ? 0xff : (a + b) (uint8)
// return (a * b) & 0xFFFFFFFF
//
// Multiply the packed 32-bit integers in a and b, producing intermediate 64-bit integers,
// and store the low 32 bits of the intermediate integers in dst.
SIMD_IWRAPPER_2_32(mullo_epi32);
SIMD_IWRAPPER_2_32(sub_epi32); // return a - b (int32)
// SIMD_IWRAPPER_2_64(sub_epi64); // return a - b (int64)
// SIMD_IWRAPPER_2_8(subs_epu8); // return (b > a) ? 0 : (a - b) (uint8)
//-----------------------------------------------------------------------
// Logical operations
//-----------------------------------------------------------------------
SIMD_IWRAPPER_2_(and_si, and_epi32, __mmask16(0xf)); // return a & b (int)
SIMD_IWRAPPER_2_(andnot_si, andnot_epi32, __mmask16(0xf)); // return (~a) & b (int)
SIMD_IWRAPPER_2_(or_si, or_epi32, __mmask16(0xf)); // return a | b (int)
SIMD_IWRAPPER_2_(xor_si, xor_epi32, __mmask16(0xf)); // return a ^ b (int)
//-----------------------------------------------------------------------
// Shift operations
//-----------------------------------------------------------------------
SIMD_IWRAPPER_1I_32(slli_epi32); // return a << ImmT
SIMD_IWRAPPER_2_32(sllv_epi32); // return a << b (uint32)
SIMD_IWRAPPER_1I_32(srai_epi32); // return a >> ImmT (int32)
SIMD_IWRAPPER_1I_32(srli_epi32); // return a >> ImmT (uint32)
SIMD_IWRAPPER_2_32(srlv_epi32); // return a >> b (uint32)
// use AVX2 version
//SIMD_IWRAPPER_1I_(srli_si, srli_si256); // return a >> (ImmT*8) (uint)
//-----------------------------------------------------------------------
// Conversion operations (Use AVX2 versions)
//-----------------------------------------------------------------------
// SIMD_IWRAPPER_1L(cvtepu8_epi16, 0xffff); // return (int16)a (uint8 --> int16)
// SIMD_IWRAPPER_1L(cvtepu8_epi32, 0xff); // return (int32)a (uint8 --> int32)
// SIMD_IWRAPPER_1L(cvtepu16_epi32, 0xff); // return (int32)a (uint16 --> int32)
// SIMD_IWRAPPER_1L(cvtepu16_epi64, 0xf); // return (int64)a (uint16 --> int64)
// SIMD_IWRAPPER_1L(cvtepu32_epi64, 0xf); // return (int64)a (uint32 --> int64)
//-----------------------------------------------------------------------
// Comparison operations (Use AVX2 versions
//-----------------------------------------------------------------------
//SIMD_IWRAPPER_2_CMP(cmpeq_epi8); // return a == b (int8)
//SIMD_IWRAPPER_2_CMP(cmpeq_epi16); // return a == b (int16)
//SIMD_IWRAPPER_2_CMP(cmpeq_epi32); // return a == b (int32)
//SIMD_IWRAPPER_2_CMP(cmpeq_epi64); // return a == b (int64)
//SIMD_IWRAPPER_2_CMP(cmpgt_epi8,); // return a > b (int8)
//SIMD_IWRAPPER_2_CMP(cmpgt_epi16); // return a > b (int16)
//SIMD_IWRAPPER_2_CMP(cmpgt_epi32); // return a > b (int32)
//SIMD_IWRAPPER_2_CMP(cmpgt_epi64); // return a > b (int64)
//
//static SIMDINLINE Integer SIMDCALL cmplt_epi32(Integer a, Integer b) // return a < b (int32)
//{
// return cmpgt_epi32(b, a);
//}
//-----------------------------------------------------------------------
// Blend / shuffle / permute operations
//-----------------------------------------------------------------------
// SIMD_IWRAPPER_2_8(packs_epi16); // int16 --> int8 See documentation for _mm256_packs_epi16 and _mm512_packs_epi16
// SIMD_IWRAPPER_2_16(packs_epi32); // int32 --> int16 See documentation for _mm256_packs_epi32 and _mm512_packs_epi32
// SIMD_IWRAPPER_2_8(packus_epi16); // uint16 --> uint8 See documentation for _mm256_packus_epi16 and _mm512_packus_epi16
// SIMD_IWRAPPER_2_16(packus_epi32); // uint32 --> uint16 See documentation for _mm256_packus_epi32 and _mm512_packus_epi32
// SIMD_IWRAPPER_2_(permute_epi32, permutevar8x32_epi32);
//static SIMDINLINE Float SIMDCALL permute_ps(Float a, Integer swiz) // return a[swiz[i]] for each 32-bit lane i (float)
//{
// return _mm256_permutevar8x32_ps(a, swiz);
//}
SIMD_IWRAPPER_1I_32(shuffle_epi32);
//template<int ImmT>
//static SIMDINLINE Integer SIMDCALL shuffle_epi64(Integer a, Integer b)
//{
// return castpd_si(shuffle_pd<ImmT>(castsi_pd(a), castsi_pd(b)));
//}
//SIMD_IWRAPPER_2(shuffle_epi8);
SIMD_IWRAPPER_2_32(unpackhi_epi32);
SIMD_IWRAPPER_2_32(unpacklo_epi32);
// SIMD_IWRAPPER_2_16(unpackhi_epi16);
// SIMD_IWRAPPER_2_64(unpackhi_epi64);
// SIMD_IWRAPPER_2_8(unpackhi_epi8);
// SIMD_IWRAPPER_2_16(unpacklo_epi16);
// SIMD_IWRAPPER_2_64(unpacklo_epi64);
// SIMD_IWRAPPER_2_8(unpacklo_epi8);
//-----------------------------------------------------------------------
// Load / store operations
//-----------------------------------------------------------------------
static SIMDINLINE Float SIMDCALL load_ps(float const *p) // return *p (loads SIMD width elements from memory)
{
return __conv(_mm512_maskz_loadu_ps(__mmask16(0xf), p));
}
static SIMDINLINE Integer SIMDCALL load_si(Integer const *p) // return *p
{
return __conv(_mm512_maskz_loadu_epi32(__mmask16(0xf), p));
}
static SIMDINLINE Float SIMDCALL loadu_ps(float const *p) // return *p (same as load_ps but allows for unaligned mem)
{
return __conv(_mm512_maskz_loadu_ps(__mmask16(0xf), p));
}
static SIMDINLINE Integer SIMDCALL loadu_si(Integer const *p) // return *p (same as load_si but allows for unaligned mem)
{
return __conv(_mm512_maskz_loadu_epi32(__mmask16(0xf), p));
}
template<ScaleFactor ScaleT>
static SIMDINLINE Float SIMDCALL i32gather_ps(float const* p, Integer idx) // return *(float*)(((int8*)p) + (idx * ScaleT))
{
return __conv(_mm512_mask_i32gather_ps(
_mm512_setzero_ps(),
__mmask16(0xf),
__conv(idx),
p,
static_cast<int>(ScaleT)));
}
// for each element: (mask & (1 << 31)) ? (i32gather_ps<ScaleT>(p, idx), mask = 0) : old
template<ScaleFactor ScaleT>
static SIMDINLINE Float SIMDCALL mask_i32gather_ps(Float old, float const* p, Integer idx, Float mask)
{
__mmask16 m = 0xf;
m = _mm512_mask_test_epi32_mask(m, _mm512_castps_si512(__conv(mask)),
_mm512_set1_epi32(0x80000000));
return __conv(_mm512_mask_i32gather_ps(
__conv(old),
m,
__conv(idx),
p,
static_cast<int>(ScaleT)));
}
// static SIMDINLINE uint32_t SIMDCALL movemask_epi8(Integer a)
// {
// __mmask64 m = 0xffffull;
// return static_cast<uint32_t>(
// _mm512_mask_test_epi8_mask(m, __conv(a), _mm512_set1_epi8(0x80)));
// }
static SIMDINLINE void SIMDCALL maskstore_ps(float *p, Integer mask, Float src)
{
__mmask16 m = 0xf;
m = _mm512_mask_test_epi32_mask(m, __conv(mask), _mm512_set1_epi32(0x80000000));
_mm512_mask_storeu_ps(p, m, __conv(src));
}
static SIMDINLINE void SIMDCALL store_ps(float *p, Float a) // *p = a (stores all elements contiguously in memory)
{
_mm512_mask_storeu_ps(p, __mmask16(0xf), __conv(a));
}
static SIMDINLINE void SIMDCALL store_si(Integer *p, Integer a) // *p = a
{
_mm512_mask_storeu_epi32(p, __mmask16(0xf), __conv(a));
}
static SIMDINLINE Float SIMDCALL vmask_ps(int32_t mask)
{
return castsi_ps(__conv(_mm512_maskz_set1_epi32(__mmask16(mask & 0xf), -1)));
}
//=======================================================================
// Legacy interface (available only in SIMD256 width)
//=======================================================================
#undef SIMD_WRAPPER_1_
#undef SIMD_WRAPPER_1
#undef SIMD_WRAPPER_1I_
#undef SIMD_WRAPPER_1I
#undef SIMD_WRAPPER_2_
#undef SIMD_WRAPPER_2
#undef SIMD_WRAPPER_2I
#undef SIMD_WRAPPER_3_
#undef SIMD_WRAPPER_3
#undef SIMD_DWRAPPER_1_
#undef SIMD_DWRAPPER_1
#undef SIMD_DWRAPPER_1I_
#undef SIMD_DWRAPPER_1I
#undef SIMD_DWRAPPER_2_
#undef SIMD_DWRAPPER_2
#undef SIMD_DWRAPPER_2I
#undef SIMD_IWRAPPER_1_
#undef SIMD_IWRAPPER_1_8
#undef SIMD_IWRAPPER_1_16
#undef SIMD_IWRAPPER_1_32
#undef SIMD_IWRAPPER_1_64
#undef SIMD_IWRAPPER_1I_
#undef SIMD_IWRAPPER_1I_8
#undef SIMD_IWRAPPER_1I_16
#undef SIMD_IWRAPPER_1I_32
#undef SIMD_IWRAPPER_1I_64
#undef SIMD_IWRAPPER_2_
#undef SIMD_IWRAPPER_2_8
#undef SIMD_IWRAPPER_2_16
#undef SIMD_IWRAPPER_2_32
#undef SIMD_IWRAPPER_2_64
#undef SIMD_IWRAPPER_2I
//#undef SIMD_IWRAPPER_2I_8
//#undef SIMD_IWRAPPER_2I_16
//#undef SIMD_IWRAPPER_2I_32
//#undef SIMD_IWRAPPER_2I_64
|
