memops.c 32.3 KB
Newer Older
sletz's avatar
sletz committed
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
/*
    Copyright (C) 2000 Paul Davis 

    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation; either version 2 of the License, or
    (at your option) any later version.

    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program; if not, write to the Free Software
    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

*/

#define _ISOC9X_SOURCE  1
#define _ISOC99_SOURCE  1

#define __USE_ISOC9X    1
#define __USE_ISOC99    1

#include <stdio.h>
#include <string.h>
#include <math.h>
#include <memory.h>
#include <stdlib.h>
31
#include <stdint.h>
sletz's avatar
sletz committed
32
#include <limits.h>
sletz's avatar
sletz committed
33
#ifdef __linux__
sletz's avatar
sletz committed
34
#include <endian.h>
sletz's avatar
sletz committed
35
#endif
sletz's avatar
sletz committed
36
37
#include "memops.h"

sletz's avatar
sletz committed
38
39
#if defined (__SSE2__) && !defined (__sun__)
#include <emmintrin.h>
40
41
42
#ifdef __SSE4_1__
#include <smmintrin.h>
#endif
sletz's avatar
sletz committed
43
44
#endif

45
46
47
48
#ifdef __ARM_NEON__
#include <arm_neon.h>
#endif

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
/* Notes about these *_SCALING values.

   the MAX_<N>BIT values are floating point. when multiplied by
   a full-scale normalized floating point sample value (-1.0..+1.0)
   they should give the maxium value representable with an integer
   sample type of N bits. Note that this is asymmetric. Sample ranges 
   for signed integer, 2's complement values are -(2^(N-1) to +(2^(N-1)-1)

   Complications
   -------------
   If we use +2^(N-1) for the scaling factors, we run into a problem:

   if we start with a normalized float value of -1.0, scaling
   to 24 bits would give -8388608 (-2^23), which is ideal.
   But with +1.0, we get +8388608, which is technically out of range.

   We never multiply a full range normalized value by this constant,
   but we could multiply it by a positive value that is close enough to +1.0
   to produce a value > +(2^(N-1)-1.

   There is no way around this paradox without wasting CPU cycles to determine
   which scaling factor to use (i.e. determine if its negative or not,
   use the right factor).

   So, for now (October 2008) we use 2^(N-1)-1 as the scaling factor.
*/

#define SAMPLE_24BIT_SCALING  8388607.0f
#define SAMPLE_16BIT_SCALING  32767.0f

/* these are just values to use if the floating point value was out of range
   
   advice from Fons Adriaensen: make the limits symmetrical
 */

#define SAMPLE_24BIT_MAX  8388607  
#define SAMPLE_24BIT_MIN  -8388607 
#define SAMPLE_24BIT_MAX_F  8388607.0f  
#define SAMPLE_24BIT_MIN_F  -8388607.0f 

#define SAMPLE_16BIT_MAX  32767
#define SAMPLE_16BIT_MIN  -32767
#define SAMPLE_16BIT_MAX_F  32767.0f
#define SAMPLE_16BIT_MIN_F  -32767.0f

/* these mark the outer edges of the range considered "within" range
   for a floating point sample value. values outside (and on the boundaries) 
   of this range will be clipped before conversion; values within this 
   range will be scaled to appropriate values for the target sample
   type.
*/

#define NORMALIZED_FLOAT_MIN -1.0f
#define NORMALIZED_FLOAT_MAX  1.0f
sletz's avatar
sletz committed
103

104
105
106
/* define this in case we end up on a platform that is missing
   the real lrintf functions
*/
sletz's avatar
sletz committed
107
108
109

#define f_round(f) lrintf(f)

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
#define float_16(s, d)\
	if ((s) <= NORMALIZED_FLOAT_MIN) {\
		(d) = SAMPLE_16BIT_MIN;\
	} else if ((s) >= NORMALIZED_FLOAT_MAX) {\
		(d) = SAMPLE_16BIT_MAX;\
	} else {\
		(d) = f_round ((s) * SAMPLE_16BIT_SCALING);\
	}

/* call this when "s" has already been scaled (e.g. when dithering)
 */

#define float_16_scaled(s, d)\
        if ((s) <= SAMPLE_16BIT_MIN_F) {\
		(d) = SAMPLE_16BIT_MIN_F;\
	} else if ((s) >= SAMPLE_16BIT_MAX_F) {	\
		(d) = SAMPLE_16BIT_MAX;\
	} else {\
	        (d) = f_round ((s));\
	}

#define float_24u32(s, d) \
	if ((s) <= NORMALIZED_FLOAT_MIN) {\
		(d) = SAMPLE_24BIT_MIN << 8;\
	} else if ((s) >= NORMALIZED_FLOAT_MAX) {\
		(d) = SAMPLE_24BIT_MAX << 8;\
	} else {\
		(d) = f_round ((s) * SAMPLE_24BIT_SCALING) << 8;\
	}

/* call this when "s" has already been scaled (e.g. when dithering)
 */

#define float_24u32_scaled(s, d)\
        if ((s) <= SAMPLE_24BIT_MIN_F) {\
		(d) = SAMPLE_24BIT_MIN << 8;\
	} else if ((s) >= SAMPLE_24BIT_MAX_F) {	\
		(d) = SAMPLE_24BIT_MAX << 8;		\
	} else {\
		(d) = f_round ((s)) << 8; \
	}

#define float_24(s, d) \
	if ((s) <= NORMALIZED_FLOAT_MIN) {\
		(d) = SAMPLE_24BIT_MIN;\
	} else if ((s) >= NORMALIZED_FLOAT_MAX) {\
		(d) = SAMPLE_24BIT_MAX;\
	} else {\
		(d) = f_round ((s) * SAMPLE_24BIT_SCALING);\
	}

/* call this when "s" has already been scaled (e.g. when dithering)
 */

#define float_24_scaled(s, d)\
        if ((s) <= SAMPLE_24BIT_MIN_F) {\
		(d) = SAMPLE_24BIT_MIN;\
	} else if ((s) >= SAMPLE_24BIT_MAX_F) {	\
		(d) = SAMPLE_24BIT_MAX;		\
	} else {\
		(d) = f_round ((s)); \
	}


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
#if defined (__SSE2__) && !defined (__sun__)

/* generates same as _mm_set_ps(1.f, 1.f, 1f., 1f) but faster  */
static inline __m128 gen_one(void)
{
    volatile __m128i x;
    __m128i ones = _mm_cmpeq_epi32(x, x);
    return (__m128)_mm_slli_epi32 (_mm_srli_epi32(ones, 25), 23);
}

static inline __m128 clip(__m128 s, __m128 min, __m128 max)
{
    return _mm_min_ps(max, _mm_max_ps(s, min));
}

static inline __m128i float_24_sse(__m128 s)
{
    const __m128 upper_bound = gen_one(); /* NORMALIZED_FLOAT_MAX */
    const __m128 lower_bound = _mm_sub_ps(_mm_setzero_ps(), upper_bound);

    __m128 clipped = clip(s, lower_bound, upper_bound);
    __m128 scaled = _mm_mul_ps(clipped, _mm_set1_ps(SAMPLE_24BIT_SCALING));
    return _mm_cvtps_epi32(scaled);
}
#endif

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

#ifdef __ARM_NEON__

static inline float32x4_t clip(float32x4_t s, float32x4_t min, float32x4_t max)
{
	return vminq_f32(max, vmaxq_f32(s, min));
}

static inline int32x4_t float_24_neon(float32x4_t s)
{
	const float32x4_t upper_bound = vdupq_n_f32(NORMALIZED_FLOAT_MAX);
	const float32x4_t lower_bound = vdupq_n_f32(NORMALIZED_FLOAT_MIN);

	float32x4_t clipped = clip(s, lower_bound, upper_bound);
	float32x4_t scaled = vmulq_f32(clipped, vdupq_n_f32(SAMPLE_24BIT_SCALING));
	return vcvtq_s32_f32(scaled);
}

static inline int16x4_t float_16_neon(float32x4_t s)
{
	const float32x4_t upper_bound = vdupq_n_f32(NORMALIZED_FLOAT_MAX);
	const float32x4_t lower_bound = vdupq_n_f32(NORMALIZED_FLOAT_MIN);

	float32x4_t clipped = clip(s, lower_bound, upper_bound);
	float32x4_t scaled = vmulq_f32(clipped, vdupq_n_f32(SAMPLE_16BIT_SCALING));
	return vmovn_s32(vcvtq_s32_f32(scaled));
}
#endif

sletz's avatar
sletz committed
229
/* Linear Congruential noise generator. From the music-dsp list
230
231
 * less random than rand(), but good enough and 10x faster 
 */
sletz's avatar
sletz committed
232
static unsigned int seed = 22222;
sletz's avatar
sletz committed
233

Adrian Knoth's avatar
Adrian Knoth committed
234
static inline unsigned int fast_rand() {
235
	seed = (seed * 196314165) + 907633515;
sletz's avatar
sletz committed
236
237
238
	return seed;
}

239
240
241
/* functions for native float sample data */

void sample_move_floatLE_sSs (jack_default_audio_sample_t *dst, char *src, unsigned long nsamples, unsigned long src_skip) {
242
	while (nsamples--) {
243
244
		*dst = *((float *) src);
		dst++;
245
246
		src += src_skip;
	}
247
248
249
}

void sample_move_dS_floatLE (char *dst, jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state) {
250
251
252
253
254
	while (nsamples--) {
		*((float *) dst) = *src;
		dst += dst_skip;
		src++;
	}
255
256
}

257
/* NOTES on function naming:
258

259
   foo_bar_d<TYPE>_s<TYPE>
260

261
262
   the "d<TYPE>" component defines the destination type for the operation
   the "s<TYPE>" component defines the source type for the operation
263

264
265
266
267
268
269
270
271
272
273
   TYPE can be one of:
   
   S      - sample is a jack_default_audio_sample_t, currently (October 2008) a 32 bit floating point value
   Ss     - like S but reverse endian from the host CPU
   32u24  - sample is an signed 32 bit integer value, but data is in upper 24 bits only
   32u24s - like 32u24 but reverse endian from the host CPU
   24     - sample is an signed 24 bit integer value
   24s    - like 24 but reverse endian from the host CPU
   16     - sample is an signed 16 bit integer value
   16s    - like 16 but reverse endian from the host CPU
sletz's avatar
sletz committed
274

275
276
277
278
279
280
281
282
   For obvious reasons, the reverse endian versions only show as source types.

   This covers all known sample formats at 16 bits or larger.
*/   

/* functions for native integer sample data */

void sample_move_d32u24_sSs (char *dst, jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)
sletz's avatar
sletz committed
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
#ifdef __ARM_NEON__
	unsigned long unrolled = nsamples / 4;
	nsamples = nsamples & 3;

	while (unrolled--) {
		float32x4_t samples = vld1q_f32(src);
		int32x4_t converted = float_24_neon(samples);
		int32x4_t shifted = vshlq_n_s32(converted, 8);
		shifted = vreinterpretq_s32_u8(vrev32q_u8(vreinterpretq_u8_s32(shifted)));

		switch(dst_skip) {
			case 4:
				vst1q_s32((int32_t*)dst, shifted);
				break;
			default:
				vst1q_lane_s32((int32_t*)(dst),            shifted, 0);
				vst1q_lane_s32((int32_t*)(dst+dst_skip),   shifted, 1);
				vst1q_lane_s32((int32_t*)(dst+2*dst_skip), shifted, 2);
                vst1q_lane_s32((int32_t*)(dst+3*dst_skip), shifted, 3);
				break;
		}
		dst += 4*dst_skip;
		src+= 4;
	}
#endif

310
	int32_t z;
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326

	while (nsamples--) {

		float_24u32 (*src, z);

#if __BYTE_ORDER == __LITTLE_ENDIAN
		dst[0]=(char)(z>>24);
		dst[1]=(char)(z>>16);
		dst[2]=(char)(z>>8);
		dst[3]=(char)(z);
#elif __BYTE_ORDER == __BIG_ENDIAN
		dst[0]=(char)(z);
		dst[1]=(char)(z>>8);
		dst[2]=(char)(z>>16);
		dst[3]=(char)(z>>24);
#endif
sletz's avatar
sletz committed
327
328
329
330
331
		dst += dst_skip;
		src++;
	}
}	

sletz's avatar
sletz committed
332
333
void sample_move_d32u24_sS (char *dst, jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)
{
sletz's avatar
sletz committed
334
335
336
337
338
339
340
341
342
343
344
#if defined (__SSE2__) && !defined (__sun__)
	__m128 int_max = _mm_set1_ps(SAMPLE_24BIT_MAX_F);
	__m128 int_min = _mm_sub_ps(_mm_setzero_ps(), int_max);
	__m128 factor = int_max;

	unsigned long unrolled = nsamples / 4;
	nsamples = nsamples & 3;

	while (unrolled--) {
		__m128 in = _mm_load_ps(src);
		__m128 scaled = _mm_mul_ps(in, factor);
345
		__m128 clipped = clip(scaled, int_min, int_max);
sletz's avatar
sletz committed
346
347
348
349

		__m128i y = _mm_cvttps_epi32(clipped);
		__m128i shifted = _mm_slli_epi32(y, 8);

350
351
352
353
354
355
#ifdef __SSE4_1__
		*(int32_t*)dst              = _mm_extract_epi32(shifted, 0);
		*(int32_t*)(dst+dst_skip)   = _mm_extract_epi32(shifted, 1);
		*(int32_t*)(dst+2*dst_skip) = _mm_extract_epi32(shifted, 2);
		*(int32_t*)(dst+3*dst_skip) = _mm_extract_epi32(shifted, 3);
#else
sletz's avatar
sletz committed
356
357
358
359
360
		__m128i shuffled1 = _mm_shuffle_epi32(shifted, _MM_SHUFFLE(0, 3, 2, 1));
		__m128i shuffled2 = _mm_shuffle_epi32(shifted, _MM_SHUFFLE(1, 0, 3, 2));
		__m128i shuffled3 = _mm_shuffle_epi32(shifted, _MM_SHUFFLE(2, 1, 0, 3));

		_mm_store_ss((float*)dst, (__m128)shifted);
361
362
363
364

		_mm_store_ss((float*)(dst+dst_skip), (__m128)shuffled1);
		_mm_store_ss((float*)(dst+2*dst_skip), (__m128)shuffled2);
		_mm_store_ss((float*)(dst+3*dst_skip), (__m128)shuffled3);
365
#endif
366
		dst += 4*dst_skip;
sletz's avatar
sletz committed
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382

		src+= 4;
	}

	while (nsamples--) {
		__m128 in = _mm_load_ss(src);
		__m128 scaled = _mm_mul_ss(in, factor);
		__m128 clipped = _mm_min_ss(int_max, _mm_max_ss(scaled, int_min));

		int y = _mm_cvttss_si32(clipped);
		*((int *) dst) = y<<8;

		dst += dst_skip;
		src++;
	}

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
#elif defined(__ARM_NEON__)
	unsigned long unrolled = nsamples / 4;
	nsamples = nsamples & 3;

	while (unrolled--) {
		float32x4_t samples = vld1q_f32(src);
		int32x4_t converted = float_24_neon(samples);
		int32x4_t shifted = vshlq_n_s32(converted, 8);

		switch(dst_skip) {
			case 4:
				vst1q_s32((int32_t*)dst, shifted);
				break;
			default:
				vst1q_lane_s32((int32_t*)(dst),            shifted, 0);
				vst1q_lane_s32((int32_t*)(dst+dst_skip),   shifted, 1);
				vst1q_lane_s32((int32_t*)(dst+2*dst_skip), shifted, 2);
                vst1q_lane_s32((int32_t*)(dst+3*dst_skip), shifted, 3);
				break;
		}
		dst += 4*dst_skip;

		src+= 4;
	}
#endif

#if !defined (__SSE2__)
sletz's avatar
sletz committed
410
	while (nsamples--) {
411
		float_24u32 (*src, *((int32_t*) dst));
sletz's avatar
sletz committed
412
413
414
		dst += dst_skip;
		src++;
	}
sletz's avatar
sletz committed
415
#endif
sletz's avatar
sletz committed
416
417
}	

sletz's avatar
sletz committed
418
419
void sample_move_dS_s32u24s (jack_default_audio_sample_t *dst, char *src, unsigned long nsamples, unsigned long src_skip)
{
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
#ifdef __ARM_NEON__
	float32x4_t factor = vdupq_n_f32(1.0 / SAMPLE_24BIT_SCALING);
	unsigned long unrolled = nsamples / 4;
	while (unrolled--) {
		int32x4_t src128;
		switch(src_skip)
		{
			case 4:
				src128 = vld1q_s32((int32_t*)src);
				break;
			case 8:
				src128 = vld2q_s32((int32_t*)src).val[0];
				break;
			default:
				src128 = vld1q_lane_s32((int32_t*)src,              src128, 0);
				src128 = vld1q_lane_s32((int32_t*)(src+src_skip),   src128, 1);
				src128 = vld1q_lane_s32((int32_t*)(src+2*src_skip), src128, 2);
				src128 = vld1q_lane_s32((int32_t*)(src+3*src_skip), src128, 3);
				break;
		}
		src128 = vreinterpretq_s32_u8(vrev32q_u8(vreinterpretq_u8_s32(src128)));
		int32x4_t shifted = vshrq_n_s32(src128, 8);
		float32x4_t as_float = vcvtq_f32_s32(shifted);
		float32x4_t divided = vmulq_f32(as_float, factor);
		vst1q_f32(dst, divided);

		src += 4*src_skip;
		dst += 4;
	}
	nsamples = nsamples & 3;
#endif

sletz's avatar
sletz committed
452
453
	/* ALERT: signed sign-extension portability !!! */

454
455
	const jack_default_audio_sample_t scaling = 1.0/SAMPLE_24BIT_SCALING;

sletz's avatar
sletz committed
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
	while (nsamples--) {
		int x;
#if __BYTE_ORDER == __LITTLE_ENDIAN
		x = (unsigned char)(src[0]);
		x <<= 8;
		x |= (unsigned char)(src[1]);
		x <<= 8;
		x |= (unsigned char)(src[2]);
		x <<= 8;
		x |= (unsigned char)(src[3]);
#elif __BYTE_ORDER == __BIG_ENDIAN
		x = (unsigned char)(src[3]);
		x <<= 8;
		x |= (unsigned char)(src[2]);
		x <<= 8;
		x |= (unsigned char)(src[1]);
		x <<= 8;
		x |= (unsigned char)(src[0]);
#endif
475
		*dst = (x >> 8) * scaling;
sletz's avatar
sletz committed
476
477
478
479
480
		dst++;
		src += src_skip;
	}
}	

sletz's avatar
sletz committed
481
482
void sample_move_dS_s32u24 (jack_default_audio_sample_t *dst, char *src, unsigned long nsamples, unsigned long src_skip)
{
sletz's avatar
sletz committed
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
#if defined (__SSE2__) && !defined (__sun__)
	unsigned long unrolled = nsamples / 4;
	static float inv_sample_max_24bit = 1.0 / SAMPLE_24BIT_SCALING;
	__m128 factor = _mm_set1_ps(inv_sample_max_24bit);
	while (unrolled--)
	{
		int i1 = *((int *) src);
		src+= src_skip;
		int i2 = *((int *) src);
		src+= src_skip;
		int i3 = *((int *) src);
		src+= src_skip;
		int i4 = *((int *) src);
		src+= src_skip;

498
		__m128i src = _mm_set_epi32(i4, i3, i2, i1);
sletz's avatar
sletz committed
499
500
501
502
503
504
505
506
507
508
		__m128i shifted = _mm_srai_epi32(src, 8);

		__m128 as_float = _mm_cvtepi32_ps(shifted);
		__m128 divided = _mm_mul_ps(as_float, factor);

		_mm_storeu_ps(dst, divided);

		dst += 4;
	}
	nsamples = nsamples & 3;
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
#elif defined(__ARM_NEON__)
	unsigned long unrolled = nsamples / 4;
	float32x4_t factor = vdupq_n_f32(1.0 / SAMPLE_24BIT_SCALING);
	while (unrolled--) {
		int32x4_t src128;
		switch(src_skip) {
			case 4:
				src128 = vld1q_s32((int32_t*)src);
				break;
			case 8:
				src128 = vld2q_s32((int32_t*)src).val[0];
				break;
			default:
				src128 = vld1q_lane_s32((int32_t*)src,              src128, 0);
				src128 = vld1q_lane_s32((int32_t*)(src+src_skip),   src128, 1);
				src128 = vld1q_lane_s32((int32_t*)(src+2*src_skip), src128, 2);
				src128 = vld1q_lane_s32((int32_t*)(src+3*src_skip), src128, 3);
				break;
		}
		int32x4_t shifted = vshrq_n_s32(src128, 8);
		float32x4_t as_float = vcvtq_f32_s32(shifted);
		float32x4_t divided = vmulq_f32(as_float, factor);
		vst1q_f32(dst, divided);

		src += 4*src_skip;
		dst += 4;
	}
	nsamples = nsamples & 3;
sletz's avatar
sletz committed
537
538
#endif

sletz's avatar
sletz committed
539
540
	/* ALERT: signed sign-extension portability !!! */

541
	const jack_default_audio_sample_t scaling = 1.0/SAMPLE_24BIT_SCALING;
sletz's avatar
sletz committed
542
	while (nsamples--) {
543
		*dst = (*((int *) src) >> 8) * scaling;
sletz's avatar
sletz committed
544
545
546
547
548
		dst++;
		src += src_skip;
	}
}	

sletz's avatar
sletz committed
549
550
void sample_move_d24_sSs (char *dst, jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)
{
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
#ifdef __ARM_NEON__
	unsigned long unrolled = nsamples / 4;
	while (unrolled--) {
		int i;
		int32_t z[4];
		float32x4_t samples = vld1q_f32(src);
		int32x4_t converted = float_24_neon(samples);
		converted = vreinterpretq_s32_u8(vrev32q_u8(vreinterpretq_u8_s32(converted)));
		vst1q_s32(z, converted);

		for (i = 0; i != 4; ++i) {
			memcpy (dst, ((char*)(z+i))+1, 3);
			dst += dst_skip;
		}
		src += 4;
	}
	nsamples = nsamples & 3;
#endif

570
	int32_t z;
sletz's avatar
sletz committed
571
572

	while (nsamples--) {
573
		float_24 (*src, z);
sletz's avatar
sletz committed
574
#if __BYTE_ORDER == __LITTLE_ENDIAN
575
576
577
		dst[0]=(char)(z>>16);
		dst[1]=(char)(z>>8);
		dst[2]=(char)(z);
578
579
580
581
#elif __BYTE_ORDER == __BIG_ENDIAN
		dst[0]=(char)(z);
		dst[1]=(char)(z>>8);
		dst[2]=(char)(z>>16);
sletz's avatar
sletz committed
582
583
584
585
586
587
#endif
		dst += dst_skip;
		src++;
	}
}	

sletz's avatar
sletz committed
588
589
void sample_move_d24_sS (char *dst, jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)
{
590
591
592
593
594
595
596
597
#if defined (__SSE2__) && !defined (__sun__)
	_MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);
	while (nsamples >= 4) {
		int i;
		int32_t z[4];
		__m128 samples = _mm_loadu_ps(src);
		__m128i converted = float_24_sse(samples);

598
599
600
601
602
603
#ifdef __SSE4_1__
		z[0] = _mm_extract_epi32(converted, 0);
		z[1] = _mm_extract_epi32(converted, 1);
		z[2] = _mm_extract_epi32(converted, 2);
		z[3] = _mm_extract_epi32(converted, 3);
#else
604
605
606
607
608
609
610
611
		__m128i shuffled1 = _mm_shuffle_epi32(converted, _MM_SHUFFLE(0, 3, 2, 1));
		__m128i shuffled2 = _mm_shuffle_epi32(converted, _MM_SHUFFLE(1, 0, 3, 2));
		__m128i shuffled3 = _mm_shuffle_epi32(converted, _MM_SHUFFLE(2, 1, 0, 3));

		_mm_store_ss((float*)z, (__m128)converted);
		_mm_store_ss((float*)z+1, (__m128)shuffled1);
		_mm_store_ss((float*)z+2, (__m128)shuffled2);
		_mm_store_ss((float*)z+3, (__m128)shuffled3);
612
#endif
613
614
615
616
617

		for (i = 0; i != 4; ++i) {
			memcpy (dst, z+i, 3);
			dst += dst_skip;
		}
618

619
620
621
		nsamples -= 4;
		src += 4;
	}
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
#elif defined(__ARM_NEON__)
	unsigned long unrolled = nsamples / 4;
	while (unrolled--) {
		int i;
		int32_t z[4];
		float32x4_t samples = vld1q_f32(src);
		int32x4_t converted = float_24_neon(samples);
		vst1q_s32(z, converted);

		for (i = 0; i != 4; ++i) {
			memcpy (dst, z+i, 3);
			dst += dst_skip;
		}
		src += 4;
	}
	nsamples = nsamples & 3;
638
639
640
641
#endif

    int32_t z;

sletz's avatar
sletz committed
642
	while (nsamples--) {
643
		float_24 (*src, z);
sletz's avatar
sletz committed
644
#if __BYTE_ORDER == __LITTLE_ENDIAN
645
		memcpy (dst, &z, 3);
sletz's avatar
sletz committed
646
#elif __BYTE_ORDER == __BIG_ENDIAN
647
		memcpy (dst, (char *)&z + 1, 3);
sletz's avatar
sletz committed
648
649
650
651
#endif
		dst += dst_skip;
		src++;
	}
652
}
sletz's avatar
sletz committed
653

sletz's avatar
sletz committed
654
655
void sample_move_dS_s24s (jack_default_audio_sample_t *dst, char *src, unsigned long nsamples, unsigned long src_skip)
{
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
	const jack_default_audio_sample_t scaling = 1.0/SAMPLE_24BIT_SCALING;

#ifdef __ARM_NEON__
	// we shift 8 to the right by dividing by 256.0 -> no sign extra handling
	const float32x4_t vscaling = vdupq_n_f32(scaling/256.0);
	int32_t x[4];
	memset(x, 0, sizeof(x));
	unsigned long unrolled = nsamples / 4;
	while (unrolled--) {
#if __BYTE_ORDER == __BIG_ENDIAN	 /* ARM big endian?? */
		// right aligned / inverse sequence below -> *256
		memcpy(((char*)&x[0])+1, src, 3);
		memcpy(((char*)&x[1])+1, src+src_skip, 3);
		memcpy(((char*)&x[2])+1, src+2*src_skip, 3);
		memcpy(((char*)&x[3])+1, src+3*src_skip, 3);
#else
		memcpy(&x[0], src, 3);
		memcpy(&x[1], src+src_skip, 3);
		memcpy(&x[2], src+2*src_skip, 3);
		memcpy(&x[3], src+3*src_skip, 3);
#endif
		src += 4 * src_skip;

		int32x4_t source = vld1q_s32(x);
		source = vreinterpretq_s32_u8(vrev32q_u8(vreinterpretq_u8_s32(source)));
		float32x4_t converted = vcvtq_f32_s32(source);
		float32x4_t scaled = vmulq_f32(converted, vscaling);
		vst1q_f32(dst, scaled);
		dst += 4;
	}
	nsamples = nsamples & 3;
#endif

sletz's avatar
sletz committed
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
	/* ALERT: signed sign-extension portability !!! */

	while (nsamples--) {
		int x;
#if __BYTE_ORDER == __LITTLE_ENDIAN
		x = (unsigned char)(src[0]);
		x <<= 8;
		x |= (unsigned char)(src[1]);
		x <<= 8;
		x |= (unsigned char)(src[2]);
		/* correct sign bit and the rest of the top byte */
		if (src[0] & 0x80) {
			x |= 0xff << 24;
		}
#elif __BYTE_ORDER == __BIG_ENDIAN
		x = (unsigned char)(src[2]);
		x <<= 8;
		x |= (unsigned char)(src[1]);
		x <<= 8;
		x |= (unsigned char)(src[0]);
		/* correct sign bit and the rest of the top byte */
sletz's avatar
sletz committed
710
		if (src[2] & 0x80) {
sletz's avatar
sletz committed
711
712
713
			x |= 0xff << 24;
		}
#endif
714
		*dst = x * scaling;
sletz's avatar
sletz committed
715
716
717
		dst++;
		src += src_skip;
	}
718
}
sletz's avatar
sletz committed
719

sletz's avatar
sletz committed
720
721
void sample_move_dS_s24 (jack_default_audio_sample_t *dst, char *src, unsigned long nsamples, unsigned long src_skip)
{
722
723
724
725
	const jack_default_audio_sample_t scaling = 1.f/SAMPLE_24BIT_SCALING;

#if defined (__SSE2__) && !defined (__sun__)
	const __m128 scaling_block = _mm_set_ps1(scaling);
726
	while (nsamples >= 4) {
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
		int x0, x1, x2, x3;

		memcpy((char*)&x0 + 1, src, 3);
		memcpy((char*)&x1 + 1, src+src_skip, 3);
		memcpy((char*)&x2 + 1, src+2*src_skip, 3);
		memcpy((char*)&x3 + 1, src+3*src_skip, 3);
		src += 4 * src_skip;

		const __m128i block_i = _mm_set_epi32(x3, x2, x1, x0);
		const __m128i shifted = _mm_srai_epi32(block_i, 8);
		const __m128 converted = _mm_cvtepi32_ps (shifted);
		const __m128 scaled = _mm_mul_ps(converted, scaling_block);
		_mm_storeu_ps(dst, scaled);
		dst += 4;
		nsamples -= 4;
	}
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
#elif defined(__ARM_NEON__)
	// we shift 8 to the right by dividing by 256.0 -> no sign extra handling
	const float32x4_t vscaling = vdupq_n_f32(scaling/256.0);
	int32_t x[4];
	memset(x, 0, sizeof(x));
	unsigned long unrolled = nsamples / 4;
	while (unrolled--) {
#if __BYTE_ORDER == __BIG_ENDIAN	/* ARM big endian?? */
		// left aligned -> *256
		memcpy(&x[0], src, 3);
		memcpy(&x[1], src+src_skip, 3);
		memcpy(&x[2], src+2*src_skip, 3);
		memcpy(&x[3], src+3*src_skip, 3);
#else
		memcpy(((char*)&x[0])+1, src, 3);
		memcpy(((char*)&x[1])+1, src+src_skip, 3);
		memcpy(((char*)&x[2])+1, src+2*src_skip, 3);
		memcpy(((char*)&x[3])+1, src+3*src_skip, 3);
#endif
		src += 4 * src_skip;

		int32x4_t source = vld1q_s32(x);
		float32x4_t converted = vcvtq_f32_s32(source);
		float32x4_t scaled = vmulq_f32(converted, vscaling);
		vst1q_f32(dst, scaled);
		dst += 4;
	}
	nsamples = nsamples & 3;
sletz's avatar
sletz committed
771
#endif
sletz's avatar
sletz committed
772
773
774
775
776
777
778
779
780

	while (nsamples--) {
		int x;
#if __BYTE_ORDER == __LITTLE_ENDIAN
		memcpy((char*)&x + 1, src, 3);
#elif __BYTE_ORDER == __BIG_ENDIAN
		memcpy(&x, src, 3);
#endif
		x >>= 8;
781
		*dst = x * scaling;
sletz's avatar
sletz committed
782
783
784
		dst++;
		src += src_skip;
	}
785
}
sletz's avatar
sletz committed
786

sletz's avatar
sletz committed
787

788
void sample_move_d16_sSs (char *dst,  jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)	
sletz's avatar
sletz committed
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
#ifdef __ARM_NEON__
	unsigned long unrolled = nsamples / 4;
	nsamples = nsamples & 3;

	while (unrolled--) {
		float32x4_t samples = vld1q_f32(src);
		int16x4_t converted = float_16_neon(samples);
		converted = vreinterpret_s16_u8(vrev16_u8(vreinterpret_u8_s16(converted)));

		switch(dst_skip) {
			case 2:
				vst1_s16((int16_t*)dst, converted);
				break;
			default:
				vst1_lane_s16((int16_t*)(dst),            converted, 0);
				vst1_lane_s16((int16_t*)(dst+dst_skip),   converted, 1);
				vst1_lane_s16((int16_t*)(dst+2*dst_skip), converted, 2);
				vst1_lane_s16((int16_t*)(dst+3*dst_skip), converted, 3);
				break;
		}
		dst += 4*dst_skip;
		src+= 4;
	}
#endif
814
	int16_t tmp;
sletz's avatar
sletz committed
815
816

	while (nsamples--) {
817
		// float_16 (*src, tmp);
sletz's avatar
sletz committed
818

819
		if (*src <= NORMALIZED_FLOAT_MIN) {
820
			tmp = SAMPLE_16BIT_MIN;
821
		} else if (*src >= NORMALIZED_FLOAT_MAX) {
822
			tmp = SAMPLE_16BIT_MAX;
sletz's avatar
sletz committed
823
		} else {
824
			tmp = (int16_t) f_round (*src * SAMPLE_16BIT_SCALING);
sletz's avatar
sletz committed
825
		}
sletz's avatar
sletz committed
826

827
828
829
830
831
832
833
#if __BYTE_ORDER == __LITTLE_ENDIAN
		dst[0]=(char)(tmp>>8);
		dst[1]=(char)(tmp);
#elif __BYTE_ORDER == __BIG_ENDIAN
		dst[0]=(char)(tmp);
		dst[1]=(char)(tmp>>8);
#endif
sletz's avatar
sletz committed
834
835
836
837
838
		dst += dst_skip;
		src++;
	}
}

839
void sample_move_d16_sS (char *dst,  jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)	
sletz's avatar
sletz committed
840
{
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
#ifdef __ARM_NEON__
	unsigned long unrolled = nsamples / 4;
	nsamples = nsamples & 3;

	while (unrolled--) {
		float32x4_t samples = vld1q_f32(src);
		int16x4_t converted = float_16_neon(samples);

		switch(dst_skip) {
			case 2:
				vst1_s16((int16_t*)dst, converted);
				break;
			default:
				vst1_lane_s16((int16_t*)(dst),            converted, 0);
				vst1_lane_s16((int16_t*)(dst+dst_skip),   converted, 1);
				vst1_lane_s16((int16_t*)(dst+2*dst_skip), converted, 2);
				vst1_lane_s16((int16_t*)(dst+3*dst_skip), converted, 3);
				break;
		}
		dst += 4*dst_skip;
		src+= 4;
	}
#endif
sletz's avatar
sletz committed
864
	while (nsamples--) {
865
		float_16 (*src, *((int16_t*) dst));
sletz's avatar
sletz committed
866
867
868
869
870
		dst += dst_skip;
		src++;
	}
}

871
void sample_move_dither_rect_d16_sSs (char *dst,  jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)	
sletz's avatar
sletz committed
872
873
{
	jack_default_audio_sample_t val;
874
	int16_t      tmp;
sletz's avatar
sletz committed
875
876

	while (nsamples--) {
877
		val = (*src * SAMPLE_16BIT_SCALING) + fast_rand() / (float) UINT_MAX - 0.5f;
878
		float_16_scaled (val, tmp);
879
880
881
882
883
884
885
#if __BYTE_ORDER == __LITTLE_ENDIAN
		dst[0]=(char)(tmp>>8);
		dst[1]=(char)(tmp);
#elif __BYTE_ORDER == __BIG_ENDIAN
		dst[0]=(char)(tmp);
		dst[1]=(char)(tmp>>8);
#endif
sletz's avatar
sletz committed
886
887
888
889
890
		dst += dst_skip;
		src++;
	}
}

891
void sample_move_dither_rect_d16_sS (char *dst,  jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)	
sletz's avatar
sletz committed
892
893
894
895
{
	jack_default_audio_sample_t val;

	while (nsamples--) {
896
897
		val = (*src * SAMPLE_16BIT_SCALING) + fast_rand() / (float)UINT_MAX - 0.5f;
		float_16_scaled (val, *((int16_t*) dst));
sletz's avatar
sletz committed
898
899
900
901
902
		dst += dst_skip;
		src++;
	}
}

903
void sample_move_dither_tri_d16_sSs (char *dst,  jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)	
sletz's avatar
sletz committed
904
{
905
906
	jack_default_audio_sample_t val;
	int16_t      tmp;
sletz's avatar
sletz committed
907
908

	while (nsamples--) {
909
		val = (*src * SAMPLE_16BIT_SCALING) + ((float)fast_rand() + (float)fast_rand()) / (float)UINT_MAX - 1.0f;
910
		float_16_scaled (val, tmp);
911
912
913
914
915
916
917
918

#if __BYTE_ORDER == __LITTLE_ENDIAN
		dst[0]=(char)(tmp>>8);
		dst[1]=(char)(tmp);
#elif __BYTE_ORDER == __BIG_ENDIAN
		dst[0]=(char)(tmp);
		dst[1]=(char)(tmp>>8);
#endif
sletz's avatar
sletz committed
919
920
921
922
923
		dst += dst_skip;
		src++;
	}
}

924
void sample_move_dither_tri_d16_sS (char *dst,  jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)	
sletz's avatar
sletz committed
925
{
926
	jack_default_audio_sample_t val;
sletz's avatar
sletz committed
927
928

	while (nsamples--) {
929
930
		val = (*src * SAMPLE_16BIT_SCALING) + ((float)fast_rand() + (float)fast_rand()) / (float)UINT_MAX - 1.0f;
		float_16_scaled (val, *((int16_t*) dst));
sletz's avatar
sletz committed
931
932
933
934
935
		dst += dst_skip;
		src++;
	}
}

936
void sample_move_dither_shaped_d16_sSs (char *dst,  jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)	
sletz's avatar
sletz committed
937
938
939
940
941
942
943
{
	jack_default_audio_sample_t     x;
	jack_default_audio_sample_t     xe; /* the innput sample - filtered error */
	jack_default_audio_sample_t     xp; /* x' */
	float        r;
	float        rm1 = state->rm1;
	unsigned int idx = state->idx;
944
	int16_t      tmp;
sletz's avatar
sletz committed
945
946

	while (nsamples--) {
947
		x = *src * SAMPLE_16BIT_SCALING;
948
		r = ((float)fast_rand() + (float)fast_rand())  / (float)UINT_MAX - 1.0f;
sletz's avatar
sletz committed
949
950
951
952
953
954
955
956
957
958
959
		/* Filter the error with Lipshitz's minimally audible FIR:
		   [2.033 -2.165 1.959 -1.590 0.6149] */
		xe = x
		     - state->e[idx] * 2.033f
		     + state->e[(idx - 1) & DITHER_BUF_MASK] * 2.165f
		     - state->e[(idx - 2) & DITHER_BUF_MASK] * 1.959f
		     + state->e[(idx - 3) & DITHER_BUF_MASK] * 1.590f
		     - state->e[(idx - 4) & DITHER_BUF_MASK] * 0.6149f;
		xp = xe + r - rm1;
		rm1 = r;

960
		float_16_scaled (xp, tmp);
sletz's avatar
sletz committed
961
962
963

		/* Intrinsic z^-1 delay */
		idx = (idx + 1) & DITHER_BUF_MASK;
964
		state->e[idx] = xp - xe;
965
966
967
968
969
970
971
972

#if __BYTE_ORDER == __LITTLE_ENDIAN
		dst[0]=(char)(tmp>>8);
		dst[1]=(char)(tmp);
#elif __BYTE_ORDER == __BIG_ENDIAN
		dst[0]=(char)(tmp);
		dst[1]=(char)(tmp>>8);
#endif
sletz's avatar
sletz committed
973
974
975
976
977
978
979
		dst += dst_skip;
		src++;
	}
	state->rm1 = rm1;
	state->idx = idx;
}

980
void sample_move_dither_shaped_d16_sS (char *dst,  jack_default_audio_sample_t *src, unsigned long nsamples, unsigned long dst_skip, dither_state_t *state)	
sletz's avatar
sletz committed
981
982
983
984
985
986
987
988
989
{
	jack_default_audio_sample_t     x;
	jack_default_audio_sample_t     xe; /* the innput sample - filtered error */
	jack_default_audio_sample_t     xp; /* x' */
	float        r;
	float        rm1 = state->rm1;
	unsigned int idx = state->idx;

	while (nsamples--) {
990
991
		x = *src * SAMPLE_16BIT_SCALING;
		r = ((float)fast_rand() + (float)fast_rand()) / (float)UINT_MAX - 1.0f;
sletz's avatar
sletz committed
992
993
994
995
996
997
998
999
1000
		/* Filter the error with Lipshitz's minimally audible FIR:
		   [2.033 -2.165 1.959 -1.590 0.6149] */
		xe = x
		     - state->e[idx] * 2.033f
		     + state->e[(idx - 1) & DITHER_BUF_MASK] * 2.165f
		     - state->e[(idx - 2) & DITHER_BUF_MASK] * 1.959f
		     + state->e[(idx - 3) & DITHER_BUF_MASK] * 1.590f
		     - state->e[(idx - 4) & DITHER_BUF_MASK] * 0.6149f;
		xp = xe + r - rm1;