Repository navigation
Expand file tree
/
Copy pathSDL_audiotypecvt.c
More file actions
1466 lines (1219 loc) · 53.9 KB
/
Copy pathSDL_audiotypecvt.c
File metadata and controls
1466 lines (1219 loc) · 53.9 KB
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
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
Simple DirectMedia Layer
Copyright (C) 1997-2026 Sam Lantinga <slouken@libsdl.org>
This software is provided 'as-is', without any express or implied
warranty. In no event will the authors be held liable for any damages
arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it
freely, subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not
claim that you wrote the original software. If you use this software
in a product, an acknowledgment in the product documentation would be
appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be
misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
*/
#include "../SDL_internal.h"
#include "SDL_audio.h"
#include "SDL_audio_c.h"
#include "SDL_cpuinfo.h"
#ifdef __ARM_NEON
#define HAVE_NEON_INTRINSICS 1
#endif
#ifdef __SSE2__
#define HAVE_SSE2_INTRINSICS
#endif
#if defined(__x86_64__) && defined(HAVE_SSE2_INTRINSICS)
#define NEED_SCALAR_CONVERTER_FALLBACKS 0 /* x86_64 guarantees SSE2. */
#elif defined(__MACOSX__) && defined(HAVE_SSE2_INTRINSICS)
#define NEED_SCALAR_CONVERTER_FALLBACKS 0 /* Mac OS X/Intel guarantees SSE2. */
#elif defined(__ARM_ARCH) && (__ARM_ARCH >= 8) && defined(HAVE_NEON_INTRINSICS)
#define NEED_SCALAR_CONVERTER_FALLBACKS 0 /* ARMv8+ promise NEON. */
#elif defined(__APPLE__) && defined(__ARM_ARCH) && (__ARM_ARCH >= 7) && defined(HAVE_NEON_INTRINSICS)
#define NEED_SCALAR_CONVERTER_FALLBACKS 0 /* All Apple ARMv7 chips promise NEON support. */
#endif
/* Set to zero if platform is guaranteed to use a SIMD codepath here. */
#ifndef NEED_SCALAR_CONVERTER_FALLBACKS
#define NEED_SCALAR_CONVERTER_FALLBACKS 1
#endif
/* Function pointers set to a CPU-specific implementation. */
SDL_AudioFilter SDL_Convert_S8_to_F32 = NULL;
SDL_AudioFilter SDL_Convert_U8_to_F32 = NULL;
SDL_AudioFilter SDL_Convert_S16_to_F32 = NULL;
SDL_AudioFilter SDL_Convert_U16_to_F32 = NULL;
SDL_AudioFilter SDL_Convert_S32_to_F32 = NULL;
SDL_AudioFilter SDL_Convert_F32_to_S8 = NULL;
SDL_AudioFilter SDL_Convert_F32_to_U8 = NULL;
SDL_AudioFilter SDL_Convert_F32_to_S16 = NULL;
SDL_AudioFilter SDL_Convert_F32_to_U16 = NULL;
SDL_AudioFilter SDL_Convert_F32_to_S32 = NULL;
#define DIVBY128 0.0078125f
#define DIVBY32768 0.000030517578125f
#define DIVBY8388607 0.00000011920930376163766f
#define DIVBY2147483648 0.0000000004656612873077392578125f /* 0x1p-31f */
#if NEED_SCALAR_CONVERTER_FALLBACKS
/* This code requires that floats are in the IEEE-754 binary32 format */
SDL_COMPILE_TIME_ASSERT(float_bits, sizeof(float) == sizeof(Uint32));
union float_bits {
Uint32 u32;
float f32;
};
/* Create a bit-mask based on the sign-bit. Should optimize to a single arithmetic-shift-right */
#define SIGNMASK(x) (Uint32)(0u - ((Uint32)(x) >> 31))
static void SDLCALL SDL_Convert_S8_to_F32_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const int num_samples = cvt->len_cvt;
const Sint8 *src = (const Sint8 *)cvt->buf;
float *dst = (float *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_S8", "AUDIO_F32");
for (i = num_samples - 1; i >= 0; --i) {
/* 1) Construct a float in the range [65536.0, 65538.0)
* 2) Shift the float range to [-1.0, 1.0) */
union float_bits x;
x.u32 = (Uint8)src[i] ^ 0x47800080u;
dst[i] = x.f32 - 65537.0f;
}
cvt->len_cvt *= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_U8_to_F32_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const int num_samples = cvt->len_cvt;
const Uint8 *src = (const Uint8 *)cvt->buf;
float *dst = (float *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_U8", "AUDIO_F32");
for (i = num_samples - 1; i >= 0; --i) {
/* 1) Construct a float in the range [65536.0, 65538.0)
* 2) Shift the float range to [-1.0, 1.0) */
union float_bits x;
x.u32 = src[i] ^ 0x47800000u;
dst[i] = x.f32 - 65537.0f;
}
cvt->len_cvt *= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_S16_to_F32_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const int num_samples = cvt->len_cvt / sizeof(Sint16);
const Sint16 *src = (const Sint16 *)cvt->buf;
float *dst = (float *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_S16", "AUDIO_F32");
for (i = num_samples - 1; i >= 0; --i) {
/* 1) Construct a float in the range [256.0, 258.0)
* 2) Shift the float range to [-1.0, 1.0) */
union float_bits x;
x.u32 = (Uint16)src[i] ^ 0x43808000u;
dst[i] = x.f32 - 257.0f;
}
cvt->len_cvt *= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_U16_to_F32_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Uint16 *src = ((const Uint16 *)(cvt->buf + cvt->len_cvt)) - 1;
float *dst = ((float *)(cvt->buf + cvt->len_cvt * 2)) - 1;
int i;
LOG_DEBUG_CONVERT("AUDIO_U16", "AUDIO_F32");
for (i = cvt->len_cvt / sizeof(Uint16); i; --i, --src, --dst) {
*dst = (((float)*src) * DIVBY32768) - 1.0f;
}
cvt->len_cvt *= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_S32_to_F32_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Sint32 *src = (const Sint32 *)cvt->buf;
float *dst = (float *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_S32", "AUDIO_F32");
for (i = cvt->len_cvt / sizeof(Sint32); i; --i, ++src, ++dst) {
*dst = ((float)(*src >> 8)) * DIVBY8388607;
}
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_F32_to_S8_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const int num_samples = cvt->len_cvt / sizeof (float);
const float *src = (const float *)cvt->buf;
Sint8 *dst = (Sint8 *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S8");
for (i = 0; i < num_samples; ++i) {
/* 1) Shift the float range from [-1.0, 1.0] to [98303.0, 98305.0]
* 2) Shift the integer range from [0x47BFFF80, 0x47C00080] to [-128, 128]
* 3) Clamp the value to [-128, 127] */
union float_bits x;
Uint32 y, z;
x.f32 = src[i] + 98304.0f;
y = x.u32 - 0x47C00000u;
z = 0x7Fu - (y ^ SIGNMASK(y));
y = y ^ (z & SIGNMASK(z));
dst[i] = (Sint8)(y & 0xFF);
}
cvt->len_cvt /= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_S8);
}
}
static void SDLCALL SDL_Convert_F32_to_U8_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const int num_samples = cvt->len_cvt / sizeof (float);
const float *src = (const float *)cvt->buf;
Uint8 *dst = (Uint8 *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_U8");
for (i = 0; i < num_samples; ++i) {
/* 1) Shift the float range from [-1.0, 1.0] to [98303.0, 98305.0]
* 2) Shift the integer range from [0x47BFFF80, 0x47C00080] to [-128, 128]
* 3) Clamp the value to [-128, 127]
* 4) Shift the integer range from [-128, 127] to [0, 255] */
union float_bits x;
Uint32 y, z;
x.f32 = src[i] + 98304.0f;
y = x.u32 - 0x47C00000u;
z = 0x7Fu - (y ^ SIGNMASK(y));
y = (y ^ 0x80u) ^ (z & SIGNMASK(z));
dst[i] = (Uint8)(y & 0xFF);
}
cvt->len_cvt /= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_U8);
}
}
static void SDLCALL SDL_Convert_F32_to_S16_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const int num_samples = cvt->len_cvt / sizeof (float);
const float *src = (const float *)cvt->buf;
Sint16 *dst = (Sint16 *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S16");
for (i = 0; i < num_samples; ++i) {
/* 1) Shift the float range from [-1.0, 1.0] to [383.0, 385.0]
* 2) Shift the integer range from [0x43BF8000, 0x43C08000] to [-32768, 32768]
* 3) Clamp values outside the [-32768, 32767] range */
union float_bits x;
Uint32 y, z;
x.f32 = src[i] + 384.0f;
y = x.u32 - 0x43C00000u;
z = 0x7FFFu - (y ^ SIGNMASK(y));
y = y ^ (z & SIGNMASK(z));
dst[i] = (Sint16)(y & 0xFFFF);
}
cvt->len_cvt /= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_S16SYS);
}
}
static void SDLCALL SDL_Convert_F32_to_U16_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const float *src = (const float *)cvt->buf;
Uint16 *dst = (Uint16 *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_U16");
for (i = cvt->len_cvt / sizeof(float); i; --i, ++src, ++dst) {
const float sample = *src;
if (sample >= 1.0f) {
*dst = 65535;
} else if (sample <= -1.0f) {
*dst = 0;
} else {
*dst = (Uint16)((sample + 1.0f) * 32767.0f);
}
}
cvt->len_cvt /= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_U16SYS);
}
}
static void SDLCALL SDL_Convert_F32_to_S32_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const int num_samples = cvt->len_cvt / sizeof (float);
const float *src = (const float *)cvt->buf;
Sint32 *dst = (Sint32 *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S32");
for (i = 0; i < num_samples; ++i) {
/* 1) Shift the float range from [-1.0, 1.0] to [-2147483648.0, 2147483648.0]
* 2) Set values outside the [-2147483648.0, 2147483647.0] range to -2147483648.0
* 3) Convert the float to an integer, and fixup values outside the valid range */
union float_bits x;
Uint32 y, z;
x.f32 = src[i];
y = x.u32 + 0x0F800000u;
z = y - 0xCF000000u;
z &= SIGNMASK(y ^ z);
x.u32 = y - z;
dst[i] = (Sint32)x.f32 ^ (Sint32)SIGNMASK(z);
}
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_S32SYS);
}
}
#endif
#ifdef HAVE_SSE2_INTRINSICS
static void SDLCALL SDL_Convert_S8_to_F32_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Sint8 *src = (const Sint8 *)cvt->buf;
float *dst = (float *)cvt->buf;
int i = cvt->len_cvt;
/* 1) Flip the sign bit to convert from S8 to U8 format
* 2) Construct a float in the range [65536.0, 65538.0)
* 3) Shift the float range to [-1.0, 1.0)
* dst[i] = i2f((src[i] ^ 0x80) | 0x47800000) - 65537.0 */
const __m128i zero = _mm_setzero_si128();
const __m128i flipper = _mm_set1_epi8(-0x80);
const __m128i caster = _mm_set1_epi16(0x4780 /* 0x47800000 = f2i(65536.0) */);
const __m128 offset = _mm_set1_ps(-65537.0);
LOG_DEBUG_CONVERT("AUDIO_S8", "AUDIO_F32 (using SSE2)");
while (i >= 16) {
i -= 16;
{
const __m128i bytes = _mm_xor_si128(_mm_loadu_si128((const __m128i *)&src[i]), flipper);
const __m128i shorts1 = _mm_unpacklo_epi8(bytes, zero);
const __m128i shorts2 = _mm_unpackhi_epi8(bytes, zero);
const __m128 floats1 = _mm_add_ps(_mm_castsi128_ps(_mm_unpacklo_epi16(shorts1, caster)), offset);
const __m128 floats2 = _mm_add_ps(_mm_castsi128_ps(_mm_unpackhi_epi16(shorts1, caster)), offset);
const __m128 floats3 = _mm_add_ps(_mm_castsi128_ps(_mm_unpacklo_epi16(shorts2, caster)), offset);
const __m128 floats4 = _mm_add_ps(_mm_castsi128_ps(_mm_unpackhi_epi16(shorts2, caster)), offset);
_mm_storeu_ps(&dst[i], floats1);
_mm_storeu_ps(&dst[i + 4], floats2);
_mm_storeu_ps(&dst[i + 8], floats3);
_mm_storeu_ps(&dst[i + 12], floats4);
}
}
while (i) {
--i;
_mm_store_ss(&dst[i], _mm_add_ss(_mm_castsi128_ps(_mm_cvtsi32_si128((Uint8)src[i] ^ 0x47800080u)), offset));
}
cvt->len_cvt *= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_U8_to_F32_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Sint8 *src = (const Sint8 *)cvt->buf;
float *dst = (float *)cvt->buf;
int i = cvt->len_cvt;
/* 1) Construct a float in the range [65536.0, 65538.0)
* 2) Shift the float range to [-1.0, 1.0)
* dst[i] = i2f(src[i] | 0x47800000) - 65537.0 */
const __m128i zero = _mm_setzero_si128();
const __m128i caster = _mm_set1_epi16(0x4780 /* 0x47800000 = f2i(65536.0) */);
const __m128 offset = _mm_set1_ps(-65537.0);
LOG_DEBUG_CONVERT("AUDIO_U8", "AUDIO_F32 (using SSE2)");
while (i >= 16) {
i -= 16;
{
const __m128i bytes = _mm_loadu_si128((const __m128i *)&src[i]);
const __m128i shorts1 = _mm_unpacklo_epi8(bytes, zero);
const __m128i shorts2 = _mm_unpackhi_epi8(bytes, zero);
const __m128 floats1 = _mm_add_ps(_mm_castsi128_ps(_mm_unpacklo_epi16(shorts1, caster)), offset);
const __m128 floats2 = _mm_add_ps(_mm_castsi128_ps(_mm_unpackhi_epi16(shorts1, caster)), offset);
const __m128 floats3 = _mm_add_ps(_mm_castsi128_ps(_mm_unpacklo_epi16(shorts2, caster)), offset);
const __m128 floats4 = _mm_add_ps(_mm_castsi128_ps(_mm_unpackhi_epi16(shorts2, caster)), offset);
_mm_storeu_ps(&dst[i], floats1);
_mm_storeu_ps(&dst[i + 4], floats2);
_mm_storeu_ps(&dst[i + 8], floats3);
_mm_storeu_ps(&dst[i + 12], floats4);
}
}
while (i) {
--i;
_mm_store_ss(&dst[i], _mm_add_ss(_mm_castsi128_ps(_mm_cvtsi32_si128((Uint8)src[i] ^ 0x47800000u)), offset));
}
cvt->len_cvt *= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_S16_to_F32_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Sint16 *src = (const Sint16 *)cvt->buf;
float *dst = (float *)cvt->buf;
int i = cvt->len_cvt / 2;
/* 1) Flip the sign bit to convert from S16 to U16 format
* 2) Construct a float in the range [256.0, 258.0)
* 3) Shift the float range to [-1.0, 1.0)
* dst[i] = i2f((src[i] ^ 0x8000) | 0x43800000) - 257.0 */
const __m128i flipper = _mm_set1_epi16(-0x8000);
const __m128i caster = _mm_set1_epi16(0x4380 /* 0x43800000 = f2i(256.0) */);
const __m128 offset = _mm_set1_ps(-257.0f);
LOG_DEBUG_CONVERT("AUDIO_S16", "AUDIO_F32 (using SSE2)");
while (i >= 16) {
i -= 16;
{
const __m128i shorts1 = _mm_xor_si128(_mm_loadu_si128((const __m128i *)&src[i]), flipper);
const __m128i shorts2 = _mm_xor_si128(_mm_loadu_si128((const __m128i *)&src[i + 8]), flipper);
const __m128 floats1 = _mm_add_ps(_mm_castsi128_ps(_mm_unpacklo_epi16(shorts1, caster)), offset);
const __m128 floats2 = _mm_add_ps(_mm_castsi128_ps(_mm_unpackhi_epi16(shorts1, caster)), offset);
const __m128 floats3 = _mm_add_ps(_mm_castsi128_ps(_mm_unpacklo_epi16(shorts2, caster)), offset);
const __m128 floats4 = _mm_add_ps(_mm_castsi128_ps(_mm_unpackhi_epi16(shorts2, caster)), offset);
_mm_storeu_ps(&dst[i], floats1);
_mm_storeu_ps(&dst[i + 4], floats2);
_mm_storeu_ps(&dst[i + 8], floats3);
_mm_storeu_ps(&dst[i + 12], floats4);
}
}
while (i) {
--i;
_mm_store_ss(&dst[i], _mm_add_ss(_mm_castsi128_ps(_mm_cvtsi32_si128((Uint16)src[i] ^ 0x43808000u)), offset));
}
cvt->len_cvt *= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_U16_to_F32_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Uint16 *src = ((const Uint16 *)(cvt->buf + cvt->len_cvt)) - 1;
float *dst = ((float *)(cvt->buf + cvt->len_cvt * 2)) - 1;
int i;
LOG_DEBUG_CONVERT("AUDIO_U16", "AUDIO_F32 (using SSE2)");
/* Get dst aligned to 16 bytes (since buffer is growing, we don't have to worry about overreading from src) */
for (i = cvt->len_cvt / sizeof(Sint16); i && (((size_t)(dst - 7)) & 15); --i, --src, --dst) {
*dst = (((float)*src) * DIVBY32768) - 1.0f;
}
src -= 7;
dst -= 7; /* adjust to read SSE blocks from the start. */
SDL_assert(!i || !(((size_t)dst) & 15));
/* Make sure src is aligned too. */
if (!(((size_t)src) & 15)) {
/* Aligned! Do SSE blocks as long as we have 16 bytes available. */
const __m128 divby32768 = _mm_set1_ps(DIVBY32768);
const __m128 minus1 = _mm_set1_ps(-1.0f);
while (i >= 8) { /* 8 * 16-bit */
const __m128i ints = _mm_load_si128((__m128i const *)src); /* get 8 sint16 into an XMM register. */
/* treat as int32, shift left to clear every other sint16, then back right with zero-extend. Now sint32. */
const __m128i a = _mm_srli_epi32(_mm_slli_epi32(ints, 16), 16);
/* right-shift-sign-extend gets us sint32 with the other set of values. */
const __m128i b = _mm_srli_epi32(ints, 16);
/* Interleave these back into the right order, convert to float, multiply, store. */
_mm_store_ps(dst, _mm_add_ps(_mm_mul_ps(_mm_cvtepi32_ps(_mm_unpacklo_epi32(a, b)), divby32768), minus1));
_mm_store_ps(dst + 4, _mm_add_ps(_mm_mul_ps(_mm_cvtepi32_ps(_mm_unpackhi_epi32(a, b)), divby32768), minus1));
i -= 8;
src -= 8;
dst -= 8;
}
}
src += 7;
dst += 7; /* adjust for any scalar finishing. */
/* Finish off any leftovers with scalar operations. */
while (i) {
*dst = (((float)*src) * DIVBY32768) - 1.0f;
i--;
src--;
dst--;
}
cvt->len_cvt *= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_S32_to_F32_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Sint32 *src = (const Sint32 *)cvt->buf;
float *dst = (float *)cvt->buf;
int i = cvt->len_cvt / 4;
/* dst[i] = f32(src[i]) / f32(0x80000000) */
const __m128 scaler = _mm_set1_ps(DIVBY2147483648);
LOG_DEBUG_CONVERT("AUDIO_S32", "AUDIO_F32 (using SSE2)");
while (i >= 16) {
i -= 16;
{
const __m128i ints1 = _mm_loadu_si128((const __m128i *)&src[i]);
const __m128i ints2 = _mm_loadu_si128((const __m128i *)&src[i + 4]);
const __m128i ints3 = _mm_loadu_si128((const __m128i *)&src[i + 8]);
const __m128i ints4 = _mm_loadu_si128((const __m128i *)&src[i + 12]);
const __m128 floats1 = _mm_mul_ps(_mm_cvtepi32_ps(ints1), scaler);
const __m128 floats2 = _mm_mul_ps(_mm_cvtepi32_ps(ints2), scaler);
const __m128 floats3 = _mm_mul_ps(_mm_cvtepi32_ps(ints3), scaler);
const __m128 floats4 = _mm_mul_ps(_mm_cvtepi32_ps(ints4), scaler);
_mm_storeu_ps(&dst[i], floats1);
_mm_storeu_ps(&dst[i + 4], floats2);
_mm_storeu_ps(&dst[i + 8], floats3);
_mm_storeu_ps(&dst[i + 12], floats4);
}
}
while (i) {
--i;
_mm_store_ss(&dst[i], _mm_mul_ss(_mm_cvt_si2ss(_mm_setzero_ps(), src[i]), scaler));
}
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_F32_to_S8_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const float *src = (const float *)cvt->buf;
Sint8 *dst = (Sint8 *)cvt->buf;
int i = cvt->len_cvt / 4;
/* 1) Shift the float range from [-1.0, 1.0] to [98303.0, 98305.0]
* 2) Extract the lowest 16 bits and clamp to [-128, 127]
* Overflow is correctly handled for inputs between roughly [-255.0, 255.0]
* dst[i] = clamp(i16(f2i(src[i] + 98304.0) & 0xFFFF), -128, 127) */
const __m128 offset = _mm_set1_ps(98304.0f);
const __m128i mask = _mm_set1_epi16(0xFF);
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S8 (using SSE2)");
while (i >= 16) {
const __m128 floats1 = _mm_loadu_ps(&src[0]);
const __m128 floats2 = _mm_loadu_ps(&src[4]);
const __m128 floats3 = _mm_loadu_ps(&src[8]);
const __m128 floats4 = _mm_loadu_ps(&src[12]);
const __m128i ints1 = _mm_castps_si128(_mm_add_ps(floats1, offset));
const __m128i ints2 = _mm_castps_si128(_mm_add_ps(floats2, offset));
const __m128i ints3 = _mm_castps_si128(_mm_add_ps(floats3, offset));
const __m128i ints4 = _mm_castps_si128(_mm_add_ps(floats4, offset));
const __m128i shorts1 = _mm_and_si128(_mm_packs_epi16(ints1, ints2), mask);
const __m128i shorts2 = _mm_and_si128(_mm_packs_epi16(ints3, ints4), mask);
const __m128i bytes = _mm_packus_epi16(shorts1, shorts2);
_mm_storeu_si128((__m128i*)dst, bytes);
i -= 16;
src += 16;
dst += 16;
}
while (i) {
const __m128i ints = _mm_castps_si128(_mm_add_ss(_mm_load_ss(src), offset));
*dst = (Sint8)(_mm_cvtsi128_si32(_mm_packs_epi16(ints, ints)) & 0xFF);
--i;
++src;
++dst;
}
cvt->len_cvt /= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_S8);
}
}
static void SDLCALL SDL_Convert_F32_to_U8_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const float *src = (const float *)cvt->buf;
Uint8 *dst = cvt->buf;
int i = cvt->len_cvt / 4;
/* 1) Shift the float range from [-1.0, 1.0] to [98304.0, 98306.0]
* 2) Extract the lowest 16 bits and clamp to [0, 255]
* Overflow is correctly handled for inputs between roughly [-254.0, 254.0]
* dst[i] = clamp(i16(f2i(src[i] + 98305.0) & 0xFFFF), 0, 255) */
const __m128 offset = _mm_set1_ps(98305.0f);
const __m128i mask = _mm_set1_epi16(0xFF);
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_U8 (using SSE2)");
while (i >= 16) {
const __m128 floats1 = _mm_loadu_ps(&src[0]);
const __m128 floats2 = _mm_loadu_ps(&src[4]);
const __m128 floats3 = _mm_loadu_ps(&src[8]);
const __m128 floats4 = _mm_loadu_ps(&src[12]);
const __m128i ints1 = _mm_castps_si128(_mm_add_ps(floats1, offset));
const __m128i ints2 = _mm_castps_si128(_mm_add_ps(floats2, offset));
const __m128i ints3 = _mm_castps_si128(_mm_add_ps(floats3, offset));
const __m128i ints4 = _mm_castps_si128(_mm_add_ps(floats4, offset));
const __m128i shorts1 = _mm_and_si128(_mm_packus_epi16(ints1, ints2), mask);
const __m128i shorts2 = _mm_and_si128(_mm_packus_epi16(ints3, ints4), mask);
const __m128i bytes = _mm_packus_epi16(shorts1, shorts2);
_mm_storeu_si128((__m128i*)dst, bytes);
i -= 16;
src += 16;
dst += 16;
}
while (i) {
const __m128i ints = _mm_castps_si128(_mm_add_ss(_mm_load_ss(src), offset));
*dst = (Uint8)(_mm_cvtsi128_si32(_mm_packus_epi16(ints, ints)) & 0xFF);
--i;
++src;
++dst;
}
cvt->len_cvt /= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_U8);
}
}
static void SDLCALL SDL_Convert_F32_to_S16_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const float *src = (const float *)cvt->buf;
Sint16 *dst = (Sint16 *)cvt->buf;
int i = cvt->len_cvt / 4;
/* 1) Shift the float range from [-1.0, 1.0] to [256.0, 258.0]
* 2) Shift the int range from [0x43800000, 0x43810000] to [-32768,32768]
* 3) Clamp to range [-32768,32767]
* Overflow is correctly handled for inputs between roughly [-257.0, +inf)
* dst[i] = clamp(f2i(src[i] + 257.0) - 0x43808000, -32768, 32767) */
const __m128 offset = _mm_set1_ps(257.0f);
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S16 (using SSE2)");
while (i >= 16) {
const __m128 floats1 = _mm_loadu_ps(&src[0]);
const __m128 floats2 = _mm_loadu_ps(&src[4]);
const __m128 floats3 = _mm_loadu_ps(&src[8]);
const __m128 floats4 = _mm_loadu_ps(&src[12]);
const __m128i ints1 = _mm_sub_epi32(_mm_castps_si128(_mm_add_ps(floats1, offset)), _mm_castps_si128(offset));
const __m128i ints2 = _mm_sub_epi32(_mm_castps_si128(_mm_add_ps(floats2, offset)), _mm_castps_si128(offset));
const __m128i ints3 = _mm_sub_epi32(_mm_castps_si128(_mm_add_ps(floats3, offset)), _mm_castps_si128(offset));
const __m128i ints4 = _mm_sub_epi32(_mm_castps_si128(_mm_add_ps(floats4, offset)), _mm_castps_si128(offset));
const __m128i shorts1 = _mm_packs_epi32(ints1, ints2);
const __m128i shorts2 = _mm_packs_epi32(ints3, ints4);
_mm_storeu_si128((__m128i*)&dst[0], shorts1);
_mm_storeu_si128((__m128i*)&dst[8], shorts2);
i -= 16;
src += 16;
dst += 16;
}
while (i) {
const __m128i ints = _mm_sub_epi32(_mm_castps_si128(_mm_add_ss(_mm_load_ss(src), offset)), _mm_castps_si128(offset));
*dst = (Sint16)(_mm_cvtsi128_si32(_mm_packs_epi32(ints, ints)) & 0xFFFF);
--i;
++src;
++dst;
}
cvt->len_cvt /= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_S16SYS);
}
}
static void SDLCALL SDL_Convert_F32_to_U16_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const float *src = (const float *)cvt->buf;
Uint16 *dst = (Uint16 *)cvt->buf;
int i;
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_U16 (using SSE2)");
/* Get dst aligned to 16 bytes */
for (i = cvt->len_cvt / sizeof(float); i && (((size_t)dst) & 15); --i, ++src, ++dst) {
const float sample = *src;
if (sample >= 1.0f) {
*dst = 65535;
} else if (sample <= -1.0f) {
*dst = 0;
} else {
*dst = (Uint16)((sample + 1.0f) * 32767.0f);
}
}
SDL_assert(!i || !(((size_t)dst) & 15));
/* Make sure src is aligned too. */
if (!(((size_t)src) & 15)) {
/* Aligned! Do SSE blocks as long as we have 16 bytes available. */
/* This calculates differently than the scalar path because SSE2 can't
pack int32 data down to unsigned int16. _mm_packs_epi32 does signed
saturation, so that would corrupt our data. _mm_packus_epi32 exists,
but not before SSE 4.1. So we convert from float to sint16, packing
that down with legit signed saturation, and then xor the top bit
against 1. This results in the correct unsigned 16-bit value, even
though it looks like dark magic. */
const __m128 mulby32767 = _mm_set1_ps(32767.0f);
const __m128i topbit = _mm_set1_epi16(-32768);
const __m128 >
const __m128 negone = _mm_set1_ps(-1.0f);
__m128i *mmdst = (__m128i *)dst;
while (i >= 8) { /* 8 * float32 */
const __m128i ints1 = _mm_cvtps_epi32(_mm_mul_ps(_mm_min_ps(_mm_max_ps(negone, _mm_load_ps(src)), one), mulby32767)); /* load 4 floats, clamp, convert to sint32 */
const __m128i ints2 = _mm_cvtps_epi32(_mm_mul_ps(_mm_min_ps(_mm_max_ps(negone, _mm_load_ps(src + 4)), one), mulby32767)); /* load 4 floats, clamp, convert to sint32 */
_mm_store_si128(mmdst, _mm_xor_si128(_mm_packs_epi32(ints1, ints2), topbit)); /* pack to sint16, xor top bit, store out. */
i -= 8;
src += 8;
mmdst++;
}
dst = (Uint16 *)mmdst;
}
/* Finish off any leftovers with scalar operations. */
while (i) {
const float sample = *src;
if (sample >= 1.0f) {
*dst = 65535;
} else if (sample <= -1.0f) {
*dst = 0;
} else {
*dst = (Uint16)((sample + 1.0f) * 32767.0f);
}
i--;
src++;
dst++;
}
cvt->len_cvt /= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_U16SYS);
}
}
static void SDLCALL SDL_Convert_F32_to_S32_SSE2(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const float *src = (const float *)cvt->buf;
Sint32 *dst = (Sint32 *)cvt->buf;
int i = cvt->len_cvt / 4;
/* 1) Scale the float range from [-1.0, 1.0] to [-2147483648.0, 2147483648.0]
* 2) Convert to integer (values too small/large become 0x80000000 = -2147483648)
* 3) Fixup values which were too large (0x80000000 ^ 0xFFFFFFFF = 2147483647)
* dst[i] = i32(src[i] * 2147483648.0) ^ ((src[i] >= 2147483648.0) ? 0xFFFFFFFF : 0x00000000) */
const __m128 limit = _mm_set1_ps(2147483648.0f);
LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S32 (using SSE2)");
while (i >= 16) {
const __m128 floats1 = _mm_loadu_ps(&src[0]);
const __m128 floats2 = _mm_loadu_ps(&src[4]);
const __m128 floats3 = _mm_loadu_ps(&src[8]);
const __m128 floats4 = _mm_loadu_ps(&src[12]);
const __m128 values1 = _mm_mul_ps(floats1, limit);
const __m128 values2 = _mm_mul_ps(floats2, limit);
const __m128 values3 = _mm_mul_ps(floats3, limit);
const __m128 values4 = _mm_mul_ps(floats4, limit);
const __m128i ints1 = _mm_xor_si128(_mm_cvttps_epi32(values1), _mm_castps_si128(_mm_cmpge_ps(values1, limit)));
const __m128i ints2 = _mm_xor_si128(_mm_cvttps_epi32(values2), _mm_castps_si128(_mm_cmpge_ps(values2, limit)));
const __m128i ints3 = _mm_xor_si128(_mm_cvttps_epi32(values3), _mm_castps_si128(_mm_cmpge_ps(values3, limit)));
const __m128i ints4 = _mm_xor_si128(_mm_cvttps_epi32(values4), _mm_castps_si128(_mm_cmpge_ps(values4, limit)));
_mm_storeu_si128((__m128i*)&dst[0], ints1);
_mm_storeu_si128((__m128i*)&dst[4], ints2);
_mm_storeu_si128((__m128i*)&dst[8], ints3);
_mm_storeu_si128((__m128i*)&dst[12], ints4);
i -= 16;
src += 16;
dst += 16;
}
while (i) {
const __m128 floats = _mm_load_ss(src);
const __m128 values = _mm_mul_ss(floats, limit);
const __m128i ints = _mm_xor_si128(_mm_cvttps_epi32(values), _mm_castps_si128(_mm_cmpge_ss(values, limit)));
*dst = (Sint32)_mm_cvtsi128_si32(ints);
--i;
++src;
++dst;
}
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_S32SYS);
}
}
#endif
#ifdef HAVE_NEON_INTRINSICS
static void SDLCALL SDL_Convert_S8_to_F32_NEON(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Sint8 *src = ((const Sint8 *)(cvt->buf + cvt->len_cvt)) - 1;
float *dst = ((float *)(cvt->buf + cvt->len_cvt * 4)) - 1;
int i;
LOG_DEBUG_CONVERT("AUDIO_S8", "AUDIO_F32 (using NEON)");
/* Get dst aligned to 16 bytes (since buffer is growing, we don't have to worry about overreading from src) */
for (i = cvt->len_cvt; i && (((size_t)(dst - 15)) & 15); --i, --src, --dst) {
*dst = ((float)*src) * DIVBY128;
}
src -= 15;
dst -= 15; /* adjust to read NEON blocks from the start. */
SDL_assert(!i || !(((size_t)dst) & 15));
/* Make sure src is aligned too. */
if (!(((size_t)src) & 15)) {
/* Aligned! Do NEON blocks as long as we have 16 bytes available. */
const int8_t *mmsrc = (const int8_t *)src;
const float32x4_t divby128 = vdupq_n_f32(DIVBY128);
while (i >= 16) { /* 16 * 8-bit */
const int8x16_t bytes = vld1q_s8(mmsrc); /* get 16 sint8 into a NEON register. */
const int16x8_t int16hi = vmovl_s8(vget_high_s8(bytes)); /* convert top 8 bytes to 8 int16 */
const int16x8_t int16lo = vmovl_s8(vget_low_s8(bytes)); /* convert bottom 8 bytes to 8 int16 */
/* split int16 to two int32, then convert to float, then multiply to normalize, store. */
vst1q_f32(dst, vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(int16lo))), divby128));
vst1q_f32(dst + 4, vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(int16lo))), divby128));
vst1q_f32(dst + 8, vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(int16hi))), divby128));
vst1q_f32(dst + 12, vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(int16hi))), divby128));
i -= 16;
mmsrc -= 16;
dst -= 16;
}
src = (const Sint8 *)mmsrc;
}
src += 15;
dst += 15; /* adjust for any scalar finishing. */
/* Finish off any leftovers with scalar operations. */
while (i) {
*dst = ((float)*src) * DIVBY128;
i--;
src--;
dst--;
}
cvt->len_cvt *= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_U8_to_F32_NEON(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Uint8 *src = ((const Uint8 *)(cvt->buf + cvt->len_cvt)) - 1;
float *dst = ((float *)(cvt->buf + cvt->len_cvt * 4)) - 1;
int i;
LOG_DEBUG_CONVERT("AUDIO_U8", "AUDIO_F32 (using NEON)");
/* Get dst aligned to 16 bytes (since buffer is growing, we don't have to worry about overreading from src) */
for (i = cvt->len_cvt; i && (((size_t)(dst - 15)) & 15); --i, --src, --dst) {
*dst = (((float)*src) * DIVBY128) - 1.0f;
}
src -= 15;
dst -= 15; /* adjust to read NEON blocks from the start. */
SDL_assert(!i || !(((size_t)dst) & 15));
/* Make sure src is aligned too. */
if (!(((size_t)src) & 15)) {
/* Aligned! Do NEON blocks as long as we have 16 bytes available. */
const uint8_t *mmsrc = (const uint8_t *)src;
const float32x4_t divby128 = vdupq_n_f32(DIVBY128);
const float32x4_t negone = vdupq_n_f32(-1.0f);
while (i >= 16) { /* 16 * 8-bit */
const uint8x16_t bytes = vld1q_u8(mmsrc); /* get 16 uint8 into a NEON register. */
const uint16x8_t uint16hi = vmovl_u8(vget_high_u8(bytes)); /* convert top 8 bytes to 8 uint16 */
const uint16x8_t uint16lo = vmovl_u8(vget_low_u8(bytes)); /* convert bottom 8 bytes to 8 uint16 */
/* split uint16 to two uint32, then convert to float, then multiply to normalize, subtract to adjust for sign, store. */
vst1q_f32(dst, vmlaq_f32(negone, vcvtq_f32_u32(vmovl_u16(vget_low_u16(uint16lo))), divby128));
vst1q_f32(dst + 4, vmlaq_f32(negone, vcvtq_f32_u32(vmovl_u16(vget_high_u16(uint16lo))), divby128));
vst1q_f32(dst + 8, vmlaq_f32(negone, vcvtq_f32_u32(vmovl_u16(vget_low_u16(uint16hi))), divby128));
vst1q_f32(dst + 12, vmlaq_f32(negone, vcvtq_f32_u32(vmovl_u16(vget_high_u16(uint16hi))), divby128));
i -= 16;
mmsrc -= 16;
dst -= 16;
}
src = (const Uint8 *)mmsrc;
}
src += 15;
dst += 15; /* adjust for any scalar finishing. */
/* Finish off any leftovers with scalar operations. */
while (i) {
*dst = (((float)*src) * DIVBY128) - 1.0f;
i--;
src--;
dst--;
}
cvt->len_cvt *= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, AUDIO_F32SYS);
}
}
static void SDLCALL SDL_Convert_S16_to_F32_NEON(SDL_AudioCVT *cvt, SDL_AudioFormat format)
{
const Sint16 *src = ((const Sint16 *)(cvt->buf + cvt->len_cvt)) - 1;
float *dst = ((float *)(cvt->buf + cvt->len_cvt * 2)) - 1;
int i;
LOG_DEBUG_CONVERT("AUDIO_S16", "AUDIO_F32 (using NEON)");
/* Get dst aligned to 16 bytes (since buffer is growing, we don't have to worry about overreading from src) */
for (i = cvt->len_cvt / sizeof(Sint16); i && (((size_t)(dst - 7)) & 15); --i, --src, --dst) {
*dst = ((float)*src) * DIVBY32768;
}
src -= 7;
dst -= 7; /* adjust to read NEON blocks from the start. */
SDL_assert(!i || !(((size_t)dst) & 15));
/* Make sure src is aligned too. */
if (!(((size_t)src) & 15)) {
/* Aligned! Do NEON blocks as long as we have 16 bytes available. */
const float32x4_t divby32768 = vdupq_n_f32(DIVBY32768);
while (i >= 8) { /* 8 * 16-bit */
const int16x8_t ints = vld1q_s16((int16_t const *)src); /* get 8 sint16 into a NEON register. */
/* split int16 to two int32, then convert to float, then multiply to normalize, store. */
vst1q_f32(dst, vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(ints))), divby32768));
vst1q_f32(dst + 4, vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(ints))), divby32768));
i -= 8;
src -= 8;
dst -= 8;
}
}
src += 7;