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
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
|
/*
* Copyright (C) 2013 The Android Open Source Project
*
* 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 ANDROID_AUDIO_RESAMPLER_FIR_PROCESS_NEON_H
#define ANDROID_AUDIO_RESAMPLER_FIR_PROCESS_NEON_H
namespace android {
// depends on AudioResamplerFirOps.h, AudioResamplerFirProcess.h
#if USE_NEON
//
// NEON specializations are enabled for Process() and ProcessL()
// Macros to save a mono/stereo accumulator sample in q0 (and q4) as stereo out.
#define ASSEMBLY_ACCUMULATE_MONO \
"vld1.s32 {d2}, [%[vLR]:64] \n"/* (1) load volumes */\
"vld1.s32 {d3}, %[out] \n"/* (2) unaligned load the output */\
"vpadd.s32 d0, d0, d1 \n"/* (1) add all 4 partial sums */\
"vpadd.s32 d0, d0, d0 \n"/* (1+4d) and replicate L/R */\
"vqrdmulh.s32 d0, d0, d2 \n"/* (2+3d) apply volume */\
"vqadd.s32 d3, d3, d0 \n"/* (1+4d) accumulate result (saturating) */\
"vst1.s32 {d3}, %[out] \n"/* (2+2d) store result */
#define ASSEMBLY_ACCUMULATE_STEREO \
"vld1.s32 {d2}, [%[vLR]:64] \n"/* (1) load volumes*/\
"vld1.s32 {d3}, %[out] \n"/* (2) unaligned load the output*/\
"vpadd.s32 d0, d0, d1 \n"/* (1) add all 4 partial sums from q0*/\
"vpadd.s32 d8, d8, d9 \n"/* (1) add all 4 partial sums from q4*/\
"vpadd.s32 d0, d0, d8 \n"/* (1+4d) combine into L/R*/\
"vqrdmulh.s32 d0, d0, d2 \n"/* (2+3d) apply volume*/\
"vqadd.s32 d3, d3, d0 \n"/* (1+4d) accumulate result (saturating)*/\
"vst1.s32 {d3}, %[out] \n"/* (2+2d)store result*/
template <>
inline void ProcessL<1, 16>(int32_t* const out,
int count,
const int16_t* coefsP,
const int16_t* coefsN,
const int16_t* sP,
const int16_t* sN,
const int32_t* const volumeLR)
{
const int CHANNELS = 1; // template specialization does not preserve params
const int STRIDE = 16;
sP -= CHANNELS*((STRIDE>>1)-1);
asm (
"veor q0, q0, q0 \n"// (0 - combines+) accumulator = 0
"1: \n"
"vld1.16 {q2}, [%[sP]] \n"// (2+0d) load 8 16-bits mono samples
"vld1.16 {q3}, [%[sN]]! \n"// (2) load 8 16-bits mono samples
"vld1.16 {q8}, [%[coefsP0]:128]! \n"// (1) load 8 16-bits coefs
"vld1.16 {q10}, [%[coefsN0]:128]! \n"// (1) load 8 16-bits coefs
"vrev64.16 q2, q2 \n"// (1) reverse s3, s2, s1, s0, s7, s6, s5, s4
// reordering the vmal to do d6, d7 before d4, d5 is slower(?)
"vmlal.s16 q0, d4, d17 \n"// (1+0d) multiply (reversed)samples by coef
"vmlal.s16 q0, d5, d16 \n"// (1) multiply (reversed)samples by coef
"vmlal.s16 q0, d6, d20 \n"// (1) multiply neg samples
"vmlal.s16 q0, d7, d21 \n"// (1) multiply neg samples
// moving these ARM instructions before neon above seems to be slower
"subs %[count], %[count], #8 \n"// (1) update loop counter
"sub %[sP], %[sP], #16 \n"// (0) move pointer to next set of samples
// sP used after branch (warning)
"bne 1b \n"// loop
ASSEMBLY_ACCUMULATE_MONO
: [out] "=Uv" (out[0]),
[count] "+r" (count),
[coefsP0] "+r" (coefsP),
[coefsN0] "+r" (coefsN),
[sP] "+r" (sP),
[sN] "+r" (sN)
: [vLR] "r" (volumeLR)
: "cc", "memory",
"q0", "q1", "q2", "q3",
"q8", "q10"
);
}
template <>
inline void ProcessL<2, 16>(int32_t* const out,
int count,
const int16_t* coefsP,
const int16_t* coefsN,
const int16_t* sP,
const int16_t* sN,
const int32_t* const volumeLR)
{
const int CHANNELS = 2; // template specialization does not preserve params
const int STRIDE = 16;
sP -= CHANNELS*((STRIDE>>1)-1);
asm (
"veor q0, q0, q0 \n"// (1) acc_L = 0
"veor q4, q4, q4 \n"// (0 combines+) acc_R = 0
"1: \n"
"vld2.16 {q2, q3}, [%[sP]] \n"// (3+0d) load 8 16-bits stereo frames
"vld2.16 {q5, q6}, [%[sN]]! \n"// (3) load 8 16-bits stereo frames
"vld1.16 {q8}, [%[coefsP0]:128]! \n"// (1) load 8 16-bits coefs
"vld1.16 {q10}, [%[coefsN0]:128]! \n"// (1) load 8 16-bits coefs
"vrev64.16 q2, q2 \n"// (1) reverse 8 samples of positive left
"vrev64.16 q3, q3 \n"// (0 combines+) reverse positive right
"vmlal.s16 q0, d4, d17 \n"// (1) multiply (reversed) samples left
"vmlal.s16 q0, d5, d16 \n"// (1) multiply (reversed) samples left
"vmlal.s16 q4, d6, d17 \n"// (1) multiply (reversed) samples right
"vmlal.s16 q4, d7, d16 \n"// (1) multiply (reversed) samples right
"vmlal.s16 q0, d10, d20 \n"// (1) multiply samples left
"vmlal.s16 q0, d11, d21 \n"// (1) multiply samples left
"vmlal.s16 q4, d12, d20 \n"// (1) multiply samples right
"vmlal.s16 q4, d13, d21 \n"// (1) multiply samples right
// moving these ARM before neon seems to be slower
"subs %[count], %[count], #8 \n"// (1) update loop counter
"sub %[sP], %[sP], #32 \n"// (0) move pointer to next set of samples
// sP used after branch (warning)
"bne 1b \n"// loop
ASSEMBLY_ACCUMULATE_STEREO
: [out] "=Uv" (out[0]),
[count] "+r" (count),
[coefsP0] "+r" (coefsP),
[coefsN0] "+r" (coefsN),
[sP] "+r" (sP),
[sN] "+r" (sN)
: [vLR] "r" (volumeLR)
: "cc", "memory",
"q0", "q1", "q2", "q3",
"q4", "q5", "q6",
"q8", "q10"
);
}
template <>
inline void Process<1, 16>(int32_t* const out,
int count,
const int16_t* coefsP,
const int16_t* coefsN,
const int16_t* coefsP1,
const int16_t* coefsN1,
const int16_t* sP,
const int16_t* sN,
uint32_t lerpP,
const int32_t* const volumeLR)
{
const int CHANNELS = 1; // template specialization does not preserve params
const int STRIDE = 16;
sP -= CHANNELS*((STRIDE>>1)-1);
asm (
"vmov.32 d2[0], %[lerpP] \n"// load the positive phase S32 Q15
"veor q0, q0, q0 \n"// (0 - combines+) accumulator = 0
"1: \n"
"vld1.16 {q2}, [%[sP]] \n"// (2+0d) load 8 16-bits mono samples
"vld1.16 {q3}, [%[sN]]! \n"// (2) load 8 16-bits mono samples
"vld1.16 {q8}, [%[coefsP0]:128]! \n"// (1) load 8 16-bits coefs
"vld1.16 {q9}, [%[coefsP1]:128]! \n"// (1) load 8 16-bits coefs for interpolation
"vld1.16 {q10}, [%[coefsN1]:128]! \n"// (1) load 8 16-bits coefs
"vld1.16 {q11}, [%[coefsN0]:128]! \n"// (1) load 8 16-bits coefs for interpolation
"vsub.s16 q9, q9, q8 \n"// (1) interpolate (step1) 1st set of coefs
"vsub.s16 q11, q11, q10 \n"// (1) interpolate (step1) 2nd set of coets
"vqrdmulh.s16 q9, q9, d2[0] \n"// (2) interpolate (step2) 1st set of coefs
"vqrdmulh.s16 q11, q11, d2[0] \n"// (2) interpolate (step2) 2nd set of coefs
"vrev64.16 q2, q2 \n"// (1) reverse s3, s2, s1, s0, s7, s6, s5, s4
"vadd.s16 q8, q8, q9 \n"// (1+2d) interpolate (step3) 1st set
"vadd.s16 q10, q10, q11 \n"// (1+1d) interpolate (step3) 2nd set
// reordering the vmal to do d6, d7 before d4, d5 is slower(?)
"vmlal.s16 q0, d4, d17 \n"// (1+0d) multiply reversed samples by coef
"vmlal.s16 q0, d5, d16 \n"// (1) multiply reversed samples by coef
"vmlal.s16 q0, d6, d20 \n"// (1) multiply neg samples
"vmlal.s16 q0, d7, d21 \n"// (1) multiply neg samples
// moving these ARM instructions before neon above seems to be slower
"subs %[count], %[count], #8 \n"// (1) update loop counter
"sub %[sP], %[sP], #16 \n"// (0) move pointer to next set of samples
// sP used after branch (warning)
"bne 1b \n"// loop
ASSEMBLY_ACCUMULATE_MONO
: [out] "=Uv" (out[0]),
[count] "+r" (count),
[coefsP0] "+r" (coefsP),
[coefsN0] "+r" (coefsN),
[coefsP1] "+r" (coefsP1),
[coefsN1] "+r" (coefsN1),
[sP] "+r" (sP),
[sN] "+r" (sN)
: [lerpP] "r" (lerpP),
[vLR] "r" (volumeLR)
: "cc", "memory",
"q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11"
);
}
template <>
inline void Process<2, 16>(int32_t* const out,
int count,
const int16_t* coefsP,
const int16_t* coefsN,
const int16_t* coefsP1,
const int16_t* coefsN1,
const int16_t* sP,
const int16_t* sN,
uint32_t lerpP,
const int32_t* const volumeLR)
{
const int CHANNELS = 2; // template specialization does not preserve params
const int STRIDE = 16;
sP -= CHANNELS*((STRIDE>>1)-1);
asm (
"vmov.32 d2[0], %[lerpP] \n"// load the positive phase
"veor q0, q0, q0 \n"// (1) acc_L = 0
"veor q4, q4, q4 \n"// (0 combines+) acc_R = 0
"1: \n"
"vld2.16 {q2, q3}, [%[sP]] \n"// (3+0d) load 8 16-bits stereo frames
"vld2.16 {q5, q6}, [%[sN]]! \n"// (3) load 8 16-bits stereo frames
"vld1.16 {q8}, [%[coefsP0]:128]! \n"// (1) load 8 16-bits coefs
"vld1.16 {q9}, [%[coefsP1]:128]! \n"// (1) load 8 16-bits coefs for interpolation
"vld1.16 {q10}, [%[coefsN1]:128]! \n"// (1) load 8 16-bits coefs
"vld1.16 {q11}, [%[coefsN0]:128]! \n"// (1) load 8 16-bits coefs for interpolation
"vsub.s16 q9, q9, q8 \n"// (1) interpolate (step1) 1st set of coefs
"vsub.s16 q11, q11, q10 \n"// (1) interpolate (step1) 2nd set of coets
"vqrdmulh.s16 q9, q9, d2[0] \n"// (2) interpolate (step2) 1st set of coefs
"vqrdmulh.s16 q11, q11, d2[0] \n"// (2) interpolate (step2) 2nd set of coefs
"vrev64.16 q2, q2 \n"// (1) reverse 8 samples of positive left
"vrev64.16 q3, q3 \n"// (1) reverse 8 samples of positive right
"vadd.s16 q8, q8, q9 \n"// (1+1d) interpolate (step3) 1st set
"vadd.s16 q10, q10, q11 \n"// (1+1d) interpolate (step3) 2nd set
"vmlal.s16 q0, d4, d17 \n"// (1) multiply reversed samples left
"vmlal.s16 q0, d5, d16 \n"// (1) multiply reversed samples left
"vmlal.s16 q4, d6, d17 \n"// (1) multiply reversed samples right
"vmlal.s16 q4, d7, d16 \n"// (1) multiply reversed samples right
"vmlal.s16 q0, d10, d20 \n"// (1) multiply samples left
"vmlal.s16 q0, d11, d21 \n"// (1) multiply samples left
"vmlal.s16 q4, d12, d20 \n"// (1) multiply samples right
"vmlal.s16 q4, d13, d21 \n"// (1) multiply samples right
// moving these ARM before neon seems to be slower
"subs %[count], %[count], #8 \n"// (1) update loop counter
"sub %[sP], %[sP], #32 \n"// (0) move pointer to next set of samples
// sP used after branch (warning)
"bne 1b \n"// loop
ASSEMBLY_ACCUMULATE_STEREO
: [out] "=Uv" (out[0]),
[count] "+r" (count),
[coefsP0] "+r" (coefsP),
[coefsN0] "+r" (coefsN),
[coefsP1] "+r" (coefsP1),
[coefsN1] "+r" (coefsN1),
[sP] "+r" (sP),
[sN] "+r" (sN)
: [lerpP] "r" (lerpP),
[vLR] "r" (volumeLR)
: "cc", "memory",
"q0", "q1", "q2", "q3",
"q4", "q5", "q6",
"q8", "q9", "q10", "q11"
);
}
template <>
inline void ProcessL<1, 16>(int32_t* const out,
int count,
const int32_t* coefsP,
const int32_t* coefsN,
const int16_t* sP,
const int16_t* sN,
const int32_t* const volumeLR)
{
const int CHANNELS = 1; // template specialization does not preserve params
const int STRIDE = 16;
sP -= CHANNELS*((STRIDE>>1)-1);
asm (
"veor q0, q0, q0 \n"// result, initialize to 0
"1: \n"
"vld1.16 {q2}, [%[sP]] \n"// load 8 16-bits mono samples
"vld1.16 {q3}, [%[sN]]! \n"// load 8 16-bits mono samples
"vld1.32 {q8, q9}, [%[coefsP0]:128]! \n"// load 8 32-bits coefs
"vld1.32 {q10, q11}, [%[coefsN0]:128]! \n"// load 8 32-bits coefs
"vrev64.16 q2, q2 \n"// reverse 8 samples of the positive side
"vshll.s16 q12, d4, #15 \n"// extend samples to 31 bits
"vshll.s16 q13, d5, #15 \n"// extend samples to 31 bits
"vshll.s16 q14, d6, #15 \n"// extend samples to 31 bits
"vshll.s16 q15, d7, #15 \n"// extend samples to 31 bits
"vqrdmulh.s32 q12, q12, q9 \n"// multiply samples
"vqrdmulh.s32 q13, q13, q8 \n"// multiply samples
"vqrdmulh.s32 q14, q14, q10 \n"// multiply samples
"vqrdmulh.s32 q15, q15, q11 \n"// multiply samples
"vadd.s32 q0, q0, q12 \n"// accumulate result
"vadd.s32 q13, q13, q14 \n"// accumulate result
"vadd.s32 q0, q0, q15 \n"// accumulate result
"vadd.s32 q0, q0, q13 \n"// accumulate result
"sub %[sP], %[sP], #16 \n"// move pointer to next set of samples
"subs %[count], %[count], #8 \n"// update loop counter
"bne 1b \n"// loop
ASSEMBLY_ACCUMULATE_MONO
: [out] "=Uv" (out[0]),
[count] "+r" (count),
[coefsP0] "+r" (coefsP),
[coefsN0] "+r" (coefsN),
[sP] "+r" (sP),
[sN] "+r" (sN)
: [vLR] "r" (volumeLR)
: "cc", "memory",
"q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11",
"q12", "q13", "q14", "q15"
);
}
template <>
inline void ProcessL<2, 16>(int32_t* const out,
int count,
const int32_t* coefsP,
const int32_t* coefsN,
const int16_t* sP,
const int16_t* sN,
const int32_t* const volumeLR)
{
const int CHANNELS = 2; // template specialization does not preserve params
const int STRIDE = 16;
sP -= CHANNELS*((STRIDE>>1)-1);
asm (
"veor q0, q0, q0 \n"// result, initialize to 0
"veor q4, q4, q4 \n"// result, initialize to 0
"1: \n"
"vld2.16 {q2, q3}, [%[sP]] \n"// load 8 16-bits stereo frames
"vld2.16 {q5, q6}, [%[sN]]! \n"// load 8 16-bits stereo frames
"vld1.32 {q8, q9}, [%[coefsP0]:128]! \n"// load 8 32-bits coefs
"vld1.32 {q10, q11}, [%[coefsN0]:128]! \n"// load 8 32-bits coefs
"vrev64.16 q2, q2 \n"// reverse 8 samples of positive left
"vrev64.16 q3, q3 \n"// reverse 8 samples of positive right
"vshll.s16 q12, d4, #15 \n"// extend samples to 31 bits
"vshll.s16 q13, d5, #15 \n"// extend samples to 31 bits
"vshll.s16 q14, d10, #15 \n"// extend samples to 31 bits
"vshll.s16 q15, d11, #15 \n"// extend samples to 31 bits
"vqrdmulh.s32 q12, q12, q9 \n"// multiply samples by coef
"vqrdmulh.s32 q13, q13, q8 \n"// multiply samples by coef
"vqrdmulh.s32 q14, q14, q10 \n"// multiply samples by coef
"vqrdmulh.s32 q15, q15, q11 \n"// multiply samples by coef
"vadd.s32 q0, q0, q12 \n"// accumulate result
"vadd.s32 q13, q13, q14 \n"// accumulate result
"vadd.s32 q0, q0, q15 \n"// accumulate result
"vadd.s32 q0, q0, q13 \n"// accumulate result
"vshll.s16 q12, d6, #15 \n"// extend samples to 31 bits
"vshll.s16 q13, d7, #15 \n"// extend samples to 31 bits
"vshll.s16 q14, d12, #15 \n"// extend samples to 31 bits
"vshll.s16 q15, d13, #15 \n"// extend samples to 31 bits
"vqrdmulh.s32 q12, q12, q9 \n"// multiply samples by coef
"vqrdmulh.s32 q13, q13, q8 \n"// multiply samples by coef
"vqrdmulh.s32 q14, q14, q10 \n"// multiply samples by coef
"vqrdmulh.s32 q15, q15, q11 \n"// multiply samples by coef
"vadd.s32 q4, q4, q12 \n"// accumulate result
"vadd.s32 q13, q13, q14 \n"// accumulate result
"vadd.s32 q4, q4, q15 \n"// accumulate result
"vadd.s32 q4, q4, q13 \n"// accumulate result
"subs %[count], %[count], #8 \n"// update loop counter
"sub %[sP], %[sP], #32 \n"// move pointer to next set of samples
"bne 1b \n"// loop
ASSEMBLY_ACCUMULATE_STEREO
: [out] "=Uv" (out[0]),
[count] "+r" (count),
[coefsP0] "+r" (coefsP),
[coefsN0] "+r" (coefsN),
[sP] "+r" (sP),
[sN] "+r" (sN)
: [vLR] "r" (volumeLR)
: "cc", "memory",
"q0", "q1", "q2", "q3",
"q4", "q5", "q6",
"q8", "q9", "q10", "q11",
"q12", "q13", "q14", "q15"
);
}
template <>
inline void Process<1, 16>(int32_t* const out,
int count,
const int32_t* coefsP,
const int32_t* coefsN,
const int32_t* coefsP1,
const int32_t* coefsN1,
const int16_t* sP,
const int16_t* sN,
uint32_t lerpP,
const int32_t* const volumeLR)
{
const int CHANNELS = 1; // template specialization does not preserve params
const int STRIDE = 16;
sP -= CHANNELS*((STRIDE>>1)-1);
asm (
"vmov.32 d2[0], %[lerpP] \n"// load the positive phase
"veor q0, q0, q0 \n"// result, initialize to 0
"1: \n"
"vld1.16 {q2}, [%[sP]] \n"// load 8 16-bits mono samples
"vld1.16 {q3}, [%[sN]]! \n"// load 8 16-bits mono samples
"vld1.32 {q8, q9}, [%[coefsP0]:128]! \n"// load 8 32-bits coefs
"vld1.32 {q12, q13}, [%[coefsP1]:128]! \n"// load 8 32-bits coefs
"vld1.32 {q10, q11}, [%[coefsN1]:128]! \n"// load 8 32-bits coefs
"vld1.32 {q14, q15}, [%[coefsN0]:128]! \n"// load 8 32-bits coefs
"vsub.s32 q12, q12, q8 \n"// interpolate (step1)
"vsub.s32 q13, q13, q9 \n"// interpolate (step1)
"vsub.s32 q14, q14, q10 \n"// interpolate (step1)
"vsub.s32 q15, q15, q11 \n"// interpolate (step1)
"vqrdmulh.s32 q12, q12, d2[0] \n"// interpolate (step2)
"vqrdmulh.s32 q13, q13, d2[0] \n"// interpolate (step2)
"vqrdmulh.s32 q14, q14, d2[0] \n"// interpolate (step2)
"vqrdmulh.s32 q15, q15, d2[0] \n"// interpolate (step2)
"vadd.s32 q8, q8, q12 \n"// interpolate (step3)
"vadd.s32 q9, q9, q13 \n"// interpolate (step3)
"vadd.s32 q10, q10, q14 \n"// interpolate (step3)
"vadd.s32 q11, q11, q15 \n"// interpolate (step3)
"vrev64.16 q2, q2 \n"// reverse 8 samples of the positive side
"vshll.s16 q12, d4, #15 \n"// extend samples to 31 bits
"vshll.s16 q13, d5, #15 \n"// extend samples to 31 bits
"vshll.s16 q14, d6, #15 \n"// extend samples to 31 bits
"vshll.s16 q15, d7, #15 \n"// extend samples to 31 bits
"vqrdmulh.s32 q12, q12, q9 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q13, q13, q8 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q14, q14, q10 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q15, q15, q11 \n"// multiply samples by interpolated coef
"vadd.s32 q0, q0, q12 \n"// accumulate result
"vadd.s32 q13, q13, q14 \n"// accumulate result
"vadd.s32 q0, q0, q15 \n"// accumulate result
"vadd.s32 q0, q0, q13 \n"// accumulate result
"sub %[sP], %[sP], #16 \n"// move pointer to next set of samples
"subs %[count], %[count], #8 \n"// update loop counter
"bne 1b \n"// loop
ASSEMBLY_ACCUMULATE_MONO
: [out] "=Uv" (out[0]),
[count] "+r" (count),
[coefsP0] "+r" (coefsP),
[coefsN0] "+r" (coefsN),
[coefsP1] "+r" (coefsP1),
[coefsN1] "+r" (coefsN1),
[sP] "+r" (sP),
[sN] "+r" (sN)
: [lerpP] "r" (lerpP),
[vLR] "r" (volumeLR)
: "cc", "memory",
"q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11",
"q12", "q13", "q14", "q15"
);
}
template <>
inline void Process<2, 16>(int32_t* const out,
int count,
const int32_t* coefsP,
const int32_t* coefsN,
const int32_t* coefsP1,
const int32_t* coefsN1,
const int16_t* sP,
const int16_t* sN,
uint32_t lerpP,
const int32_t* const volumeLR)
{
const int CHANNELS = 2; // template specialization does not preserve params
const int STRIDE = 16;
sP -= CHANNELS*((STRIDE>>1)-1);
asm (
"vmov.32 d2[0], %[lerpP] \n"// load the positive phase
"veor q0, q0, q0 \n"// result, initialize to 0
"veor q4, q4, q4 \n"// result, initialize to 0
"1: \n"
"vld2.16 {q2, q3}, [%[sP]] \n"// load 8 16-bits stereo frames
"vld2.16 {q5, q6}, [%[sN]]! \n"// load 8 16-bits stereo frames
"vld1.32 {q8, q9}, [%[coefsP0]:128]! \n"// load 8 32-bits coefs
"vld1.32 {q12, q13}, [%[coefsP1]:128]! \n"// load 8 32-bits coefs
"vld1.32 {q10, q11}, [%[coefsN1]:128]! \n"// load 8 32-bits coefs
"vld1.32 {q14, q15}, [%[coefsN0]:128]! \n"// load 8 32-bits coefs
"vsub.s32 q12, q12, q8 \n"// interpolate (step1)
"vsub.s32 q13, q13, q9 \n"// interpolate (step1)
"vsub.s32 q14, q14, q10 \n"// interpolate (step1)
"vsub.s32 q15, q15, q11 \n"// interpolate (step1)
"vqrdmulh.s32 q12, q12, d2[0] \n"// interpolate (step2)
"vqrdmulh.s32 q13, q13, d2[0] \n"// interpolate (step2)
"vqrdmulh.s32 q14, q14, d2[0] \n"// interpolate (step2)
"vqrdmulh.s32 q15, q15, d2[0] \n"// interpolate (step2)
"vadd.s32 q8, q8, q12 \n"// interpolate (step3)
"vadd.s32 q9, q9, q13 \n"// interpolate (step3)
"vadd.s32 q10, q10, q14 \n"// interpolate (step3)
"vadd.s32 q11, q11, q15 \n"// interpolate (step3)
"vrev64.16 q2, q2 \n"// reverse 8 samples of positive left
"vrev64.16 q3, q3 \n"// reverse 8 samples of positive right
"vshll.s16 q12, d4, #15 \n"// extend samples to 31 bits
"vshll.s16 q13, d5, #15 \n"// extend samples to 31 bits
"vshll.s16 q14, d10, #15 \n"// extend samples to 31 bits
"vshll.s16 q15, d11, #15 \n"// extend samples to 31 bits
"vqrdmulh.s32 q12, q12, q9 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q13, q13, q8 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q14, q14, q10 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q15, q15, q11 \n"// multiply samples by interpolated coef
"vadd.s32 q0, q0, q12 \n"// accumulate result
"vadd.s32 q13, q13, q14 \n"// accumulate result
"vadd.s32 q0, q0, q15 \n"// accumulate result
"vadd.s32 q0, q0, q13 \n"// accumulate result
"vshll.s16 q12, d6, #15 \n"// extend samples to 31 bits
"vshll.s16 q13, d7, #15 \n"// extend samples to 31 bits
"vshll.s16 q14, d12, #15 \n"// extend samples to 31 bits
"vshll.s16 q15, d13, #15 \n"// extend samples to 31 bits
"vqrdmulh.s32 q12, q12, q9 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q13, q13, q8 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q14, q14, q10 \n"// multiply samples by interpolated coef
"vqrdmulh.s32 q15, q15, q11 \n"// multiply samples by interpolated coef
"vadd.s32 q4, q4, q12 \n"// accumulate result
"vadd.s32 q13, q13, q14 \n"// accumulate result
"vadd.s32 q4, q4, q15 \n"// accumulate result
"vadd.s32 q4, q4, q13 \n"// accumulate result
"subs %[count], %[count], #8 \n"// update loop counter
"sub %[sP], %[sP], #32 \n"// move pointer to next set of samples
"bne 1b \n"// loop
ASSEMBLY_ACCUMULATE_STEREO
: [out] "=Uv" (out[0]),
[count] "+r" (count),
[coefsP0] "+r" (coefsP),
[coefsN0] "+r" (coefsN),
[coefsP1] "+r" (coefsP1),
[coefsN1] "+r" (coefsN1),
[sP] "+r" (sP),
[sN] "+r" (sN)
: [lerpP] "r" (lerpP),
[vLR] "r" (volumeLR)
: "cc", "memory",
"q0", "q1", "q2", "q3",
"q4", "q5", "q6",
"q8", "q9", "q10", "q11",
"q12", "q13", "q14", "q15"
);
}
#endif //USE_NEON
}; // namespace android
#endif /*ANDROID_AUDIO_RESAMPLER_FIR_PROCESS_NEON_H*/
|
