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
|
/*
==============================================================================
This file is part of the JUCE library.
Copyright (c) 2017 - ROLI Ltd.
JUCE is an open source library subject to commercial or open-source
licensing.
By using JUCE, you agree to the terms of both the JUCE 5 End-User License
Agreement and JUCE 5 Privacy Policy (both updated and effective as of the
27th April 2017).
End User License Agreement: www.juce.com/juce-5-licence
Privacy Policy: www.juce.com/juce-5-privacy-policy
Or: You may also use this code under the terms of the GPL v3 (see
www.gnu.org/licenses).
JUCE IS PROVIDED "AS IS" WITHOUT ANY WARRANTY, AND ALL WARRANTIES, WHETHER
EXPRESSED OR IMPLIED, INCLUDING MERCHANTABILITY AND FITNESS FOR PURPOSE, ARE
DISCLAIMED.
==============================================================================
*/
namespace juce
{
namespace dsp
{
#ifndef DOXYGEN
#if JUCE_GCC && (__GNUC__ >= 6)
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wignored-attributes"
#endif
#ifdef _MSC_VER
#define DECLARE_NEON_SIMD_CONST(type, name) \
static __declspec(align(16)) const type name [16 / sizeof (type)]
#define DEFINE_NEON_SIMD_CONST(type, class_type, name) \
__declspec(align(16)) const type SIMDNativeOps<class_type>:: name [16 / sizeof (type)]
#else
#define DECLARE_NEON_SIMD_CONST(type, name) \
static const type name [16 / sizeof (type)] __attribute__((aligned(16)))
#define DEFINE_NEON_SIMD_CONST(type, class_type, name) \
const type SIMDNativeOps<class_type>:: name [16 / sizeof (type)] __attribute__((aligned(16)))
#endif
template <typename type>
struct SIMDNativeOps;
//==============================================================================
/** Unsigned 32-bit integer NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<uint32_t>
{
//==============================================================================
using vSIMDType = uint32x4_t;
using fb = SIMDFallbackOps<uint32_t, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (uint32_t, kAllBitsSet);
//==============================================================================
static forcedinline vSIMDType expand (uint32_t s) noexcept { return vdupq_n_u32 (s); }
static forcedinline vSIMDType load (const uint32_t* a) noexcept { return vld1q_u32 (a); }
static forcedinline void store (vSIMDType value, uint32_t* a) noexcept { vst1q_u32 (a, value); }
static forcedinline uint32_t get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, uint32_t s) noexcept { v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_u32 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_u32 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return vmulq_u32 (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return vandq_u32 (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return vorrq_u32 (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return veorq_u32 (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return vbicq_u32 (b, a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_u32 ((uint32_t*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return vminq_u32 (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return vmaxq_u32 (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vceqq_u32 (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (sum (notEqual (a, b)) == 0); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return bit_not (equal (a, b)); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgtq_u32 (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgeq_u32 (a, b); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return vmlaq_u32 (a, b, c); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return a; }
static forcedinline uint32_t sum (vSIMDType a) noexcept
{
auto rr = vadd_u32 (vget_high_u32 (a), vget_low_u32 (a));
return vget_lane_u32 (vpadd_u32 (rr, rr), 0);
}
};
//==============================================================================
/** Signed 32-bit integer NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<int32_t>
{
//==============================================================================
using vSIMDType = int32x4_t;
using fb = SIMDFallbackOps<int32_t, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (int32_t, kAllBitsSet);
//==============================================================================
static forcedinline vSIMDType expand (int32_t s) noexcept { return vdupq_n_s32 (s); }
static forcedinline vSIMDType load (const int32_t* a) noexcept { return vld1q_s32 (a); }
static forcedinline void store (vSIMDType value, int32_t* a) noexcept { vst1q_s32 (a, value); }
static forcedinline int32_t get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, int32_t s) noexcept { v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_s32 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_s32 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return vmulq_s32 (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return vandq_s32 (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return vorrq_s32 (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return veorq_s32 (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return vbicq_s32 (b, a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_s32 ((int32_t*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return vminq_s32 (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return vmaxq_s32 (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vceqq_s32 (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (sum (notEqual (a, b)) == 0); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return bit_not (equal (a, b)); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgtq_s32 (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgeq_s32 (a, b); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return vmlaq_s32 (a, b, c); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return a; }
static forcedinline int32_t sum (vSIMDType a) noexcept
{
auto rr = vadd_s32 (vget_high_s32 (a), vget_low_s32 (a));
rr = vpadd_s32 (rr, rr);
return vget_lane_s32 (rr, 0);
}
};
//==============================================================================
/** Signed 8-bit integer NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<int8_t>
{
//==============================================================================
using vSIMDType = int8x16_t;
using fb = SIMDFallbackOps<int8_t, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (int8_t, kAllBitsSet);
//==============================================================================
static forcedinline vSIMDType expand (int8_t s) noexcept { return vdupq_n_s8 (s); }
static forcedinline vSIMDType load (const int8_t* a) noexcept { return vld1q_s8 (a); }
static forcedinline void store (vSIMDType value, int8_t* a) noexcept { vst1q_s8 (a, value); }
static forcedinline int8_t get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, int8_t s) noexcept { v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_s8 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_s8 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return vmulq_s8 (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return vandq_s8 (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return vorrq_s8 (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return veorq_s8 (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return vbicq_s8 (b, a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_s8 ((int8_t*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return vminq_s8 (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return vmaxq_s8 (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vceqq_s8 (a, b); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return bit_not (equal (a, b)); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgtq_s8 (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgeq_s8 (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (SIMDNativeOps<int32_t>::sum ((SIMDNativeOps<int32_t>::vSIMDType) notEqual (a, b)) == 0); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return vmlaq_s8 (a, b, c); }
static forcedinline int8_t sum (vSIMDType a) noexcept { return fb::sum (a); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return a; }
};
//==============================================================================
/** Unsigned 8-bit integer NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<uint8_t>
{
//==============================================================================
using vSIMDType = uint8x16_t;
using fb = SIMDFallbackOps<uint8_t, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (uint8_t, kAllBitsSet);
//==============================================================================
static forcedinline vSIMDType expand (uint8_t s) noexcept { return vdupq_n_u8 (s); }
static forcedinline vSIMDType load (const uint8_t* a) noexcept { return vld1q_u8 (a); }
static forcedinline void store (vSIMDType value, uint8_t* a) noexcept { vst1q_u8 (a, value); }
static forcedinline uint8_t get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, uint8_t s) noexcept { v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_u8 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_u8 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return vmulq_u8 (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return vandq_u8 (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return vorrq_u8 (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return veorq_u8 (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return vbicq_u8 (b, a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_u8 ((uint8_t*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return vminq_u8 (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return vmaxq_u8 (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vceqq_u8 (a, b); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return bit_not (equal (a, b)); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgtq_u8 (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgeq_u8 (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (SIMDNativeOps<uint32_t>::sum ((SIMDNativeOps<uint32_t>::vSIMDType) notEqual (a, b)) == 0); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return vmlaq_u8 (a, b, c); }
static forcedinline uint8_t sum (vSIMDType a) noexcept { return fb::sum (a); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return a; }
};
//==============================================================================
/** Signed 16-bit integer NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<int16_t>
{
//==============================================================================
using vSIMDType = int16x8_t;
using fb = SIMDFallbackOps<int16_t, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (int16_t, kAllBitsSet);
//==============================================================================
static forcedinline vSIMDType expand (int16_t s) noexcept { return vdupq_n_s16 (s); }
static forcedinline vSIMDType load (const int16_t* a) noexcept { return vld1q_s16 (a); }
static forcedinline void store (vSIMDType value, int16_t* a) noexcept { vst1q_s16 (a, value); }
static forcedinline int16_t get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, int16_t s) noexcept { v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_s16 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_s16 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return vmulq_s16 (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return vandq_s16 (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return vorrq_s16 (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return veorq_s16 (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return vbicq_s16 (b, a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_s16 ((int16_t*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return vminq_s16 (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return vmaxq_s16 (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vceqq_s16 (a, b); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return bit_not (equal (a, b)); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgtq_s16 (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgeq_s16 (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (SIMDNativeOps<int32_t>::sum ((SIMDNativeOps<int32_t>::vSIMDType) notEqual (a, b)) == 0); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return vmlaq_s16 (a, b, c); }
static forcedinline int16_t sum (vSIMDType a) noexcept { return fb::sum (a); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return a; }
};
//==============================================================================
/** Unsigned 16-bit integer NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<uint16_t>
{
//==============================================================================
using vSIMDType = uint16x8_t;
using fb = SIMDFallbackOps<uint16_t, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (uint16_t, kAllBitsSet);
//==============================================================================
static forcedinline vSIMDType expand (uint16_t s) noexcept { return vdupq_n_u16 (s); }
static forcedinline vSIMDType load (const uint16_t* a) noexcept { return vld1q_u16 (a); }
static forcedinline void store (vSIMDType value, uint16_t* a) noexcept { vst1q_u16 (a, value); }
static forcedinline uint16_t get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, uint16_t s) noexcept { v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_u16 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_u16 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return vmulq_u16 (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return vandq_u16 (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return vorrq_u16 (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return veorq_u16 (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return vbicq_u16 (b, a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_u16 ((uint16_t*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return vminq_u16 (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return vmaxq_u16 (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vceqq_u16 (a, b); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return bit_not (equal (a, b)); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgtq_u16 (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgeq_u16 (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (SIMDNativeOps<uint32_t>::sum ((SIMDNativeOps<uint32_t>::vSIMDType) notEqual (a, b)) == 0); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return vmlaq_u16 (a, b, c); }
static forcedinline uint16_t sum (vSIMDType a) noexcept { return fb::sum (a); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return a; }
};
//==============================================================================
/** Signed 64-bit integer NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<int64_t>
{
//==============================================================================
using vSIMDType = int64x2_t;
using fb = SIMDFallbackOps<int64_t, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (int64_t, kAllBitsSet);
//==============================================================================
static forcedinline vSIMDType expand (int64_t s) noexcept { return vdupq_n_s64 (s); }
static forcedinline vSIMDType load (const int64_t* a) noexcept { return vld1q_s64 (a); }
static forcedinline void store (vSIMDType value, int64_t* a) noexcept { vst1q_s64 (a, value); }
static forcedinline int64_t get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, int64_t s) noexcept { v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_s64 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_s64 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return fb::mul (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return vandq_s64 (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return vorrq_s64 (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return veorq_s64 (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return vbicq_s64 (b, a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_s64 ((int64_t*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return fb::min (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return fb::max (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return fb::equal (a, b); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return fb::notEqual (a, b); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return fb::greaterThan (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return fb::greaterThanOrEqual (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (SIMDNativeOps<int32_t>::sum ((SIMDNativeOps<int32_t>::vSIMDType) notEqual (a, b)) == 0); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return fb::multiplyAdd (a, b, c); }
static forcedinline int64_t sum (vSIMDType a) noexcept { return fb::sum (a); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return a; }
};
//==============================================================================
/** Unsigned 64-bit integer NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<uint64_t>
{
//==============================================================================
using vSIMDType = uint64x2_t;
using fb = SIMDFallbackOps<uint64_t, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (uint64_t, kAllBitsSet);
//==============================================================================
static forcedinline vSIMDType expand (uint64_t s) noexcept { return vdupq_n_u64 (s); }
static forcedinline vSIMDType load (const uint64_t* a) noexcept { return vld1q_u64 (a); }
static forcedinline void store (vSIMDType value, uint64_t* a) noexcept { vst1q_u64 (a, value); }
static forcedinline uint64_t get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, uint64_t s) noexcept { v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_u64 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_u64 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return fb::mul (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return vandq_u64 (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return vorrq_u64 (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return veorq_u64 (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return vbicq_u64 (b, a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_u64 ((uint64_t*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return fb::min (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return fb::max (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return fb::equal (a, b); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return fb::notEqual (a, b); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return fb::greaterThan (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return fb::greaterThanOrEqual (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (SIMDNativeOps<uint32_t>::sum ((SIMDNativeOps<uint32_t>::vSIMDType) notEqual (a, b)) == 0); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return fb::multiplyAdd (a, b, c); }
static forcedinline uint64_t sum (vSIMDType a) noexcept { return fb::sum (a); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return a; }
};
//==============================================================================
/** Single-precision floating point NEON intrinsics.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<float>
{
//==============================================================================
using vSIMDType = float32x4_t;
using vMaskType = uint32x4_t;
using fb = SIMDFallbackOps<float, vSIMDType>;
//==============================================================================
DECLARE_NEON_SIMD_CONST (int32_t, kAllBitsSet);
DECLARE_NEON_SIMD_CONST (int32_t, kEvenHighBit);
DECLARE_NEON_SIMD_CONST (float, kOne);
//==============================================================================
static forcedinline vSIMDType expand (float s) noexcept { return vdupq_n_f32 (s); }
static forcedinline vSIMDType load (const float* a) noexcept { return vld1q_f32 (a); }
static forcedinline float get (vSIMDType v, size_t i) noexcept { return v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, float s) noexcept { v[i] = s; return v; }
static forcedinline void store (vSIMDType value, float* a) noexcept { vst1q_f32 (a, value); }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return vaddq_f32 (a, b); }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return vsubq_f32 (a, b); }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return vmulq_f32 (a, b); }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vandq_u32 ((vMaskType) a, (vMaskType) b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vorrq_u32 ((vMaskType) a, (vMaskType) b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) veorq_u32 ((vMaskType) a, (vMaskType) b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vbicq_u32 ((vMaskType) b, (vMaskType) a); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return bit_notand (a, vld1q_f32 ((float*) kAllBitsSet)); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return vminq_f32 (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return vmaxq_f32 (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vceqq_f32 (a, b); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return bit_not (equal (a, b)); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgtq_f32 (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return (vSIMDType) vcgeq_f32 (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return (SIMDNativeOps<uint32_t>::sum ((SIMDNativeOps<uint32_t>::vSIMDType) notEqual (a, b)) == 0); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return vmlaq_f32 (a, b, c); }
static forcedinline vSIMDType dupeven (vSIMDType a) noexcept { return fb::shuffle<(0 << 0) | (0 << 2) | (2 << 4) | (2 << 6)> (a); }
static forcedinline vSIMDType dupodd (vSIMDType a) noexcept { return fb::shuffle<(1 << 0) | (1 << 2) | (3 << 4) | (3 << 6)> (a); }
static forcedinline vSIMDType swapevenodd (vSIMDType a) noexcept { return fb::shuffle<(1 << 0) | (0 << 2) | (3 << 4) | (2 << 6)> (a); }
static forcedinline vSIMDType oddevensum (vSIMDType a) noexcept { return add (fb::shuffle<(2 << 0) | (3 << 2) | (0 << 4) | (1 << 6)> (a), a); }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return vcvtq_f32_s32 (vcvtq_s32_f32 (a)); }
//==============================================================================
static forcedinline vSIMDType cmplxmul (vSIMDType a, vSIMDType b) noexcept
{
vSIMDType rr_ir = mul (a, dupeven (b));
vSIMDType ii_ri = mul (swapevenodd (a), dupodd (b));
return add (rr_ir, bit_xor (ii_ri, vld1q_f32 ((float*) kEvenHighBit)));
}
static forcedinline float sum (vSIMDType a) noexcept
{
auto rr = vadd_f32 (vget_high_f32 (a), vget_low_f32 (a));
return vget_lane_f32 (vpadd_f32 (rr, rr), 0);
}
};
//==============================================================================
/** Double-precision floating point NEON intrinsics does not exist in NEON
so we need to emulate this.
@tags{DSP}
*/
template <>
struct SIMDNativeOps<double>
{
//==============================================================================
using vSIMDType = struct { double v[2]; };
using fb = SIMDFallbackOps<double, vSIMDType>;
static forcedinline vSIMDType expand (double s) noexcept { return {{s, s}}; }
static forcedinline vSIMDType load (const double* a) noexcept { return {{a[0], a[1]}}; }
static forcedinline void store (vSIMDType v, double* a) noexcept { a[0] = v.v[0]; a[1] = v.v[1]; }
static forcedinline double get (vSIMDType v, size_t i) noexcept { return v.v[i]; }
static forcedinline vSIMDType set (vSIMDType v, size_t i, double s) noexcept { v.v[i] = s; return v; }
static forcedinline vSIMDType add (vSIMDType a, vSIMDType b) noexcept { return {{a.v[0] + b.v[0], a.v[1] + b.v[1]}}; }
static forcedinline vSIMDType sub (vSIMDType a, vSIMDType b) noexcept { return {{a.v[0] - b.v[0], a.v[1] - b.v[1]}}; }
static forcedinline vSIMDType mul (vSIMDType a, vSIMDType b) noexcept { return {{a.v[0] * b.v[0], a.v[1] * b.v[1]}}; }
static forcedinline vSIMDType bit_and (vSIMDType a, vSIMDType b) noexcept { return fb::bit_and (a, b); }
static forcedinline vSIMDType bit_or (vSIMDType a, vSIMDType b) noexcept { return fb::bit_or (a, b); }
static forcedinline vSIMDType bit_xor (vSIMDType a, vSIMDType b) noexcept { return fb::bit_xor (a, b); }
static forcedinline vSIMDType bit_notand (vSIMDType a, vSIMDType b) noexcept { return fb::bit_notand (a, b); }
static forcedinline vSIMDType bit_not (vSIMDType a) noexcept { return fb::bit_not (a); }
static forcedinline vSIMDType min (vSIMDType a, vSIMDType b) noexcept { return fb::min (a, b); }
static forcedinline vSIMDType max (vSIMDType a, vSIMDType b) noexcept { return fb::max (a, b); }
static forcedinline vSIMDType equal (vSIMDType a, vSIMDType b) noexcept { return fb::equal (a, b); }
static forcedinline vSIMDType notEqual (vSIMDType a, vSIMDType b) noexcept { return fb::notEqual (a, b); }
static forcedinline vSIMDType greaterThan (vSIMDType a, vSIMDType b) noexcept { return fb::greaterThan (a, b); }
static forcedinline vSIMDType greaterThanOrEqual (vSIMDType a, vSIMDType b) noexcept { return fb::greaterThanOrEqual (a, b); }
static forcedinline bool allEqual (vSIMDType a, vSIMDType b) noexcept { return fb::allEqual (a, b); }
static forcedinline vSIMDType multiplyAdd (vSIMDType a, vSIMDType b, vSIMDType c) noexcept { return fb::multiplyAdd (a, b, c); }
static forcedinline vSIMDType cmplxmul (vSIMDType a, vSIMDType b) noexcept { return fb::cmplxmul (a, b); }
static forcedinline double sum (vSIMDType a) noexcept { return fb::sum (a); }
static forcedinline vSIMDType oddevensum (vSIMDType a) noexcept { return a; }
static forcedinline vSIMDType truncate (vSIMDType a) noexcept { return fb::truncate (a); }
};
#endif
#if JUCE_GCC && (__GNUC__ >= 6)
#pragma GCC diagnostic pop
#endif
} // namespace dsp
} // namespace juce
|
