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
|
/****************************************************************************
* 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.
****************************************************************************/
#pragma once
#if 0
//===========================================================================
// Placeholder name representing either SIMD4, SIMD256, or SIMD16 structures.
//===========================================================================
struct SIMD256 // or SIMD4 or SIMD16
{
//=======================================================================
// SIMD Types
//
// These typedefs are examples. The SIMD256 and SIMD16 implementations will
// use different base types with this same naming.
using Float = __m256; // Packed single-precision float vector
using Double = __m256d; // Packed double-precision float vector
using Integer = __m256i; // Packed integer vector (mutable element widths)
using Mask = uint8_t; // Integer representing mask bits
//=======================================================================
// Standard interface
// (available in both SIMD256 and SIMD16 widths)
//=======================================================================
//-----------------------------------------------------------------------
// Single precision floating point arithmetic operations
//-----------------------------------------------------------------------
static Float add_ps(Float a, Float b); // return a + b
static Float div_ps(Float a, Float b); // return a / b
static Float fmadd_ps(Float a, Float b, Float c); // return (a * b) + c
static Float fmsub_ps(Float a, Float b, Float c); // return (a * b) - c
static Float max_ps(Float a, Float b); // return (a > b) ? a : b
static Float min_ps(Float a, Float b); // return (a < b) ? a : b
static Float mul_ps(Float a, Float b); // return a * b
static Float rcp_ps(Float a); // return 1.0f / a
static Float rsqrt_ps(Float a); // return 1.0f / sqrt(a)
static Float sub_ps(Float a, Float b); // return a - b
enum class RoundMode
{
TO_NEAREST_INT = 0x00, // Round to nearest integer == TRUNCATE(value + (signof(value))0.5)
TO_NEG_INF = 0x01, // Round to negative infinity
TO_POS_INF = 0x02, // Round to positive infinity
TO_ZERO = 0x03, // Round to 0 a.k.a. truncate
CUR_DIRECTION = 0x04, // Round in direction set in MXCSR register
RAISE_EXC = 0x00, // Raise exception on overflow
NO_EXC = 0x08, // Suppress exceptions
NINT = static_cast<int>(TO_NEAREST_INT) | static_cast<int>(RAISE_EXC),
NINT_NOEXC = static_cast<int>(TO_NEAREST_INT) | static_cast<int>(NO_EXC),
FLOOR = static_cast<int>(TO_NEG_INF) | static_cast<int>(RAISE_EXC),
FLOOR_NOEXC = static_cast<int>(TO_NEG_INF) | static_cast<int>(NO_EXC),
CEIL = static_cast<int>(TO_POS_INF) | static_cast<int>(RAISE_EXC),
CEIL_NOEXC = static_cast<int>(TO_POS_INF) | static_cast<int>(NO_EXC),
TRUNC = static_cast<int>(TO_ZERO) | static_cast<int>(RAISE_EXC),
TRUNC_NOEXC = static_cast<int>(TO_ZERO) | static_cast<int>(NO_EXC),
RINT = static_cast<int>(CUR_DIRECTION) | static_cast<int>(RAISE_EXC),
NEARBYINT = static_cast<int>(CUR_DIRECTION) | static_cast<int>(NO_EXC),
};
// return round_func(a)
//
// round_func is chosen on the RMT template parameter. See the documentation
// for the RoundMode enumeration above.
template <RoundMode RMT>
static Float round_ps(Float a); // return round(a)
//-----------------------------------------------------------------------
// Integer (various width) arithmetic operations
//-----------------------------------------------------------------------
static Integer abs_epi32(Integer a); // return absolute_value(a) (int32)
static Integer add_epi32(Integer a, Integer b); // return a + b (int32)
static Integer add_epi8(Integer a, Integer b); // return a + b (int8)
static Integer adds_epu8(Integer a, Integer b); // return ((a + b) > 0xff) ? 0xff : (a + b) (uint8)
static Integer max_epi32(Integer a, Integer b); // return (a > b) ? a : b (int32)
static Integer max_epu32(Integer a, Integer b); // return (a > b) ? a : b (uint32)
static Integer min_epi32(Integer a, Integer b); // return (a < b) ? a : b (int32)
static Integer min_epu32(Integer a, Integer b); // return (a < b) ? a : b (uint32)
static Integer mul_epi32(Integer a, Integer b); // return a * b (int32)
// 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.
static Float mullo_epi32(Integer a, Integer b);
static Integer sub_epi32(Integer a, Integer b); // return a - b (int32)
static Integer sub_epi64(Integer a, Integer b); // return a - b (int64)
static Integer subs_epu8(Integer a, Integer b); // return (b > a) ? 0 : (a - b) (uint8)
//-----------------------------------------------------------------------
// Logical operations
//-----------------------------------------------------------------------
static Float and_ps(Float a, Float b); // return a & b (float treated as int)
static Integer and_si(Integer a, Integer b); // return a & b (int)
static Float andnot_ps(Float a, Float b); // return (~a) & b (float treated as int)
static Integer andnot_si(Integer a, Integer b); // return (~a) & b (int)
static Float or_ps(Float a, Float b); // return a | b (float treated as int)
static Float or_si(Integer a, Integer b); // return a | b (int)
static Float xor_ps(Float a, Float b); // return a ^ b (float treated as int)
static Integer xor_si(Integer a, Integer b); // return a ^ b (int)
//-----------------------------------------------------------------------
// Shift operations
//-----------------------------------------------------------------------
template<int ImmT>
static Integer slli_epi32(Integer a); // return a << ImmT
static Integer sllv_epi32(Integer a, Integer b); // return a << b
template<int ImmT>
static Integer srai_epi32(Integer a); // return a >> ImmT (int32)
template<int ImmT>
static Integer srli_epi32(Integer a); // return a >> ImmT (uint32)
template<int ImmT> // for each 128-bit lane:
static Integer srli_si(Integer a); // return a >> (ImmT*8) (uint)
template<int ImmT>
static Float srlisi_ps(Float a); // same as srli_si, but with Float cast to int
static Integer srlv_epi32(Integer a, Integer b); // return a >> b (uint32)
//-----------------------------------------------------------------------
// Conversion operations
//-----------------------------------------------------------------------
static Float castpd_ps(Double a); // return *(Float*)(&a)
static Integer castps_si(Float a); // return *(Integer*)(&a)
static Double castsi_pd(Integer a); // return *(Double*)(&a)
static Double castps_pd(Float a); // return *(Double*)(&a)
static Float castsi_ps(Integer a); // return *(Float*)(&a)
static Float cvtepi32_ps(Integer a); // return (float)a (int32 --> float)
static Integer cvtepu8_epi16(Integer a); // return (int16)a (uint8 --> int16)
static Integer cvtepu8_epi32(Integer a); // return (int32)a (uint8 --> int32)
static Integer cvtepu16_epi32(Integer a); // return (int32)a (uint16 --> int32)
static Integer cvtepu16_epi64(Integer a); // return (int64)a (uint16 --> int64)
static Integer cvtepu32_epi64(Integer a); // return (int64)a (uint32 --> int64)
static Integer cvtps_epi32(Float a); // return (int32)a (float --> int32)
static Integer cvttps_epi32(Float a); // return (int32)a (rnd_to_zero(float) --> int32)
//-----------------------------------------------------------------------
// Comparison operations
//-----------------------------------------------------------------------
// Comparison types used with cmp_ps:
// - ordered comparisons are always false if either operand is NaN
// - unordered comparisons are always true if either operand is NaN
// - signaling comparisons raise an exception if either operand is NaN
// - non-signaling comparisons will never raise an exception
//
// Ordered: return (a != NaN) && (b != NaN) && (a cmp b)
// Unordered: return (a == NaN) || (b == NaN) || (a cmp b)
enum class CompareType
{
EQ_OQ = 0x00, // Equal (ordered, nonsignaling)
LT_OS = 0x01, // Less-than (ordered, signaling)
LE_OS = 0x02, // Less-than-or-equal (ordered, signaling)
UNORD_Q = 0x03, // Unordered (nonsignaling)
NEQ_UQ = 0x04, // Not-equal (unordered, nonsignaling)
NLT_US = 0x05, // Not-less-than (unordered, signaling)
NLE_US = 0x06, // Not-less-than-or-equal (unordered, signaling)
ORD_Q = 0x07, // Ordered (nonsignaling)
EQ_UQ = 0x08, // Equal (unordered, non-signaling)
NGE_US = 0x09, // Not-greater-than-or-equal (unordered, signaling)
NGT_US = 0x0A, // Not-greater-than (unordered, signaling)
FALSE_OQ = 0x0B, // False (ordered, nonsignaling)
NEQ_OQ = 0x0C, // Not-equal (ordered, non-signaling)
GE_OS = 0x0D, // Greater-than-or-equal (ordered, signaling)
GT_OS = 0x0E, // Greater-than (ordered, signaling)
TRUE_UQ = 0x0F, // True (unordered, non-signaling)
EQ_OS = 0x10, // Equal (ordered, signaling)
LT_OQ = 0x11, // Less-than (ordered, nonsignaling)
LE_OQ = 0x12, // Less-than-or-equal (ordered, nonsignaling)
UNORD_S = 0x13, // Unordered (signaling)
NEQ_US = 0x14, // Not-equal (unordered, signaling)
NLT_UQ = 0x15, // Not-less-than (unordered, nonsignaling)
NLE_UQ = 0x16, // Not-less-than-or-equal (unordered, nonsignaling)
ORD_S = 0x17, // Ordered (signaling)
EQ_US = 0x18, // Equal (unordered, signaling)
NGE_UQ = 0x19, // Not-greater-than-or-equal (unordered, nonsignaling)
NGT_UQ = 0x1A, // Not-greater-than (unordered, nonsignaling)
FALSE_OS = 0x1B, // False (ordered, signaling)
NEQ_OS = 0x1C, // Not-equal (ordered, signaling)
GE_OQ = 0x1D, // Greater-than-or-equal (ordered, nonsignaling)
GT_OQ = 0x1E, // Greater-than (ordered, nonsignaling)
TRUE_US = 0x1F, // True (unordered, signaling)
};
// return a (CmpTypeT) b (float)
//
// See documentation for CompareType above for valid values for CmpTypeT.
template<CompareType CmpTypeT>
static Float cmp_ps(Float a, Float b); // return a (CmtTypeT) b (see above)
static Float cmpgt_ps(Float a, Float b); // return cmp_ps<CompareType::GT_OQ>(a, b)
static Float cmple_ps(Float a, Float b); // return cmp_ps<CompareType::LE_OQ>(a, b)
static Float cmplt_ps(Float a, Float b); // return cmp_ps<CompareType::LT_OQ>(a, b)
static Float cmpneq_ps(Float a, Float b); // return cmp_ps<CompareType::NEQ_OQ>(a, b)
static Float cmpeq_ps(Float a, Float b); // return cmp_ps<CompareType::EQ_OQ>(a, b)
static Float cmpge_ps(Float a, Float b); // return cmp_ps<CompareType::GE_OQ>(a, b)
static Integer cmpeq_epi8(Integer a, Integer b); // return a == b (int8)
static Integer cmpeq_epi16(Integer a, Integer b); // return a == b (int16)
static Integer cmpeq_epi32(Integer a, Integer b); // return a == b (int32)
static Integer cmpeq_epi64(Integer a, Integer b); // return a == b (int64)
static Integer cmpgt_epi8(Integer a, Integer b); // return a > b (int8)
static Integer cmpgt_epi16(Integer a, Integer b); // return a > b (int16)
static Integer cmpgt_epi32(Integer a, Integer b); // return a > b (int32)
static Integer cmpgt_epi64(Integer a, Integer b); // return a > b (int64)
static Integer cmplt_epi32(Integer a, Integer b); // return a < b (int32)
static bool testz_ps(Float a, Float b); // return all_lanes_zero(a & b) ? 1 : 0 (float)
static bool testz_si(Integer a, Integer b); // return all_lanes_zero(a & b) ? 1 : 0 (int)
//-----------------------------------------------------------------------
// Blend / shuffle / permute operations
//-----------------------------------------------------------------------
template<int ImmT>
static Float blend_ps(Float a, Float b); // return ImmT ? b : a (float)
static Integer blendv_epi32(Integer a, Integer b, Float mask); // return mask ? b : a (int)
static Float blendv_ps(Float a, Float b, Float mask); // return mask ? b : a (float)
static Float broadcast_ss(float const *p); // return *p (all elements in vector get same value)
static Integer packs_epi16(Integer a, Integer b); // See documentation for _mm256_packs_epi16 and _mm512_packs_epi16
static Integer packs_epi32(Integer a, Integer b); // See documentation for _mm256_packs_epi32 and _mm512_packs_epi32
static Integer packus_epi16(Integer a, Integer b); // See documentation for _mm256_packus_epi16 and _mm512_packus_epi16
static Integer packus_epi32(Integer a, Integer b); // See documentation for _mm256_packus_epi32 and _mm512_packus_epi32
static Float permute_epi32(Integer a, Integer swiz); // return a[swiz[i]] for each 32-bit lane i (int32)
static Float permute_ps(Float a, Integer swiz); // return a[swiz[i]] for each 32-bit lane i (float)
template<int SwizT>
static Integer shuffle_epi32(Integer a, Integer b);
template<int SwizT>
static Integer shuffle_epi64(Integer a, Integer b);
static Integer shuffle_epi8(Integer a, Integer b);
template<int SwizT>
static Float shuffle_pd(Double a, Double b);
template<int SwizT>
static Float shuffle_ps(Float a, Float b);
static Integer unpackhi_epi16(Integer a, Integer b);
static Integer unpackhi_epi32(Integer a, Integer b);
static Integer unpackhi_epi64(Integer a, Integer b);
static Integer unpackhi_epi8(Integer a, Integer b);
static Float unpackhi_pd(Double a, Double b);
static Float unpackhi_ps(Float a, Float b);
static Integer unpacklo_epi16(Integer a, Integer b);
static Integer unpacklo_epi32(Integer a, Integer b);
static Integer unpacklo_epi64(Integer a, Integer b);
static Integer unpacklo_epi8(Integer a, Integer b);
static Float unpacklo_pd(Double a, Double b);
static Float unpacklo_ps(Float a, Float b);
//-----------------------------------------------------------------------
// Load / store operations
//-----------------------------------------------------------------------
enum class ScaleFactor
{
SF_1, // No scaling
SF_2, // Scale offset by 2
SF_4, // Scale offset by 4
SF_8, // Scale offset by 8
};
template<ScaleFactor ScaleT = ScaleFactor::SF_1>
static Float i32gather_ps(float const* p, Integer idx); // return *(float*)(((int8*)p) + (idx * ScaleT))
static Float load1_ps(float const *p); // return *p (broadcast 1 value to all elements)
static Float load_ps(float const *p); // return *p (loads SIMD width elements from memory)
static Integer load_si(Integer const *p); // return *p
static Float loadu_ps(float const *p); // return *p (same as load_ps but allows for unaligned mem)
static Integer loadu_si(Integer const *p); // return *p (same as load_si but allows for unaligned mem)
// for each element: (mask & (1 << 31)) ? (i32gather_ps<ScaleT>(p, idx), mask = 0) : old
template<int ScaleT>
static Float mask_i32gather_ps(Float old, float const* p, Integer idx, Float mask);
static void maskstore_ps(float *p, Integer mask, Float src);
static int movemask_epi8(Integer a);
static int movemask_pd(Double a);
static int movemask_ps(Float a);
static Integer set1_epi32(int i); // return i (all elements are same value)
static Integer set1_epi8(char i); // return i (all elements are same value)
static Float set1_ps(float f); // return f (all elements are same value)
static Float setzero_ps(); // return 0 (float)
static Integer setzero_si(); // return 0 (integer)
static void store_ps(float *p, Float a); // *p = a (stores all elements contiguously in memory)
static void store_si(Integer *p, Integer a); // *p = a
static void stream_ps(float *p, Float a); // *p = a (same as store_ps, but doesn't keep memory in cache)
//=======================================================================
// Legacy interface (available only in SIMD256 width)
//=======================================================================
static Float broadcast_ps(__m128 const *p);
template<int ImmT>
static __m128d extractf128_pd(Double a);
template<int ImmT>
static __m128 extractf128_ps(Float a);
template<int ImmT>
static __m128i extractf128_si(Integer a);
template<int ImmT>
static Double insertf128_pd(Double a, __m128d b);
template<int ImmT>
static Float insertf128_ps(Float a, __m128 b);
template<int ImmT>
static Integer insertf128_si(Integer a, __m128i b);
static Integer loadu2_si(__m128 const* phi, __m128 const* plo);
template<int ImmT>
static Double permute2f128_pd(Double a, Double b);
template<int ImmT>
static Float permute2f128_ps(Float a, Float b);
template<int ImmT>
static Integer permute2f128_si(Integer a, Integer b);
static Integer set_epi32(int i7, int i6, int i5, int i4, int i3, int i2, int i1, int i0);
static void storeu2_si(__m128i *phi, __m128i *plo, Integer src);
//=======================================================================
// Advanced masking interface (currently available only in SIMD16 width)
//=======================================================================
};
#endif // #if 0
|
