1 /***********************************************************************
2 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
3 Redistribution and use in source and binary forms, with or without
4 modification, are permitted provided that the following conditions
5 are met:
6 - Redistributions of source code must retain the above copyright notice,
7 this list of conditions and the following disclaimer.
8 - Redistributions in binary form must reproduce the above copyright
9 notice, this list of conditions and the following disclaimer in the
10 documentation and/or other materials provided with the distribution.
11 - Neither the name of Internet Society, IETF or IETF Trust, nor the
12 names of specific contributors, may be used to endorse or promote
13 products derived from this software without specific prior written
14 permission.
15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
16 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
19 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25 POSSIBILITY OF SUCH DAMAGE.
26 ***********************************************************************/
27
28 #ifndef SILK_SIGPROC_FIX_H
29 #define SILK_SIGPROC_FIX_H
30
31 #ifdef __cplusplus
32 extern "C"
33 {
34 #endif
35
36 /*#define silk_MACRO_COUNT */ /* Used to enable WMOPS counting */
37
38 #define SILK_MAX_ORDER_LPC 24 /* max order of the LPC analysis in schur() and k2a() */
39
40 #include <string.h> /* for memset(), memcpy(), memmove() */
41 #include "typedef.h"
42 #include "resampler_structs.h"
43 #include "macros.h"
44 #include "cpu_support.h"
45
46 #if defined(OPUS_X86_MAY_HAVE_SSE4_1)
47 #include "x86/SigProc_FIX_sse.h"
48 #endif
49
50 #if (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR))
51 #include "arm/biquad_alt_arm.h"
52 #include "arm/LPC_inv_pred_gain_arm.h"
53 #endif
54
55 /********************************************************************/
56 /* SIGNAL PROCESSING FUNCTIONS */
57 /********************************************************************/
58
59 /*!
60 * Initialize/reset the resampler state for a given pair of input/output sampling rates
61 */
62 opus_int silk_resampler_init(
63 silk_resampler_state_struct *S, /* I/O Resampler state */
64 opus_int32 Fs_Hz_in, /* I Input sampling rate (Hz) */
65 opus_int32 Fs_Hz_out, /* I Output sampling rate (Hz) */
66 opus_int forEnc /* I If 1: encoder; if 0: decoder */
67 );
68
69 /*!
70 * Resampler: convert from one sampling rate to another
71 */
72 opus_int silk_resampler(
73 silk_resampler_state_struct *S, /* I/O Resampler state */
74 opus_int16 out[], /* O Output signal */
75 const opus_int16 in[], /* I Input signal */
76 opus_int32 inLen /* I Number of input samples */
77 );
78
79 /*!
80 * Downsample 2x, mediocre quality
81 */
82 void silk_resampler_down2(
83 opus_int32 *S, /* I/O State vector [ 2 ] */
84 opus_int16 *out, /* O Output signal [ len ] */
85 const opus_int16 *in, /* I Input signal [ floor(len/2) ] */
86 opus_int32 inLen /* I Number of input samples */
87 );
88
89 /*!
90 * Downsample by a factor 2/3, low quality
91 */
92 void silk_resampler_down2_3(
93 opus_int32 *S, /* I/O State vector [ 6 ] */
94 opus_int16 *out, /* O Output signal [ floor(2*inLen/3) ] */
95 const opus_int16 *in, /* I Input signal [ inLen ] */
96 opus_int32 inLen /* I Number of input samples */
97 );
98
99 /*!
100 * second order ARMA filter;
101 * slower than biquad() but uses more precise coefficients
102 * can handle (slowly) varying coefficients
103 */
104 void silk_biquad_alt_stride1(
105 const opus_int16 *in, /* I input signal */
106 const opus_int32 *B_Q28, /* I MA coefficients [3] */
107 const opus_int32 *A_Q28, /* I AR coefficients [2] */
108 opus_int32 *S, /* I/O State vector [2] */
109 opus_int16 *out, /* O output signal */
110 const opus_int32 len /* I signal length (must be even) */
111 );
112
113 void silk_biquad_alt_stride2_c(
114 const opus_int16 *in, /* I input signal */
115 const opus_int32 *B_Q28, /* I MA coefficients [3] */
116 const opus_int32 *A_Q28, /* I AR coefficients [2] */
117 opus_int32 *S, /* I/O State vector [4] */
118 opus_int16 *out, /* O output signal */
119 const opus_int32 len /* I signal length (must be even) */
120 );
121
122 /* Variable order MA prediction error filter. */
123 void silk_LPC_analysis_filter(
124 opus_int16 *out, /* O Output signal */
125 const opus_int16 *in, /* I Input signal */
126 const opus_int16 *B, /* I MA prediction coefficients, Q12 [order] */
127 const opus_int32 len, /* I Signal length */
128 const opus_int32 d, /* I Filter order */
129 int arch /* I Run-time architecture */
130 );
131
132 /* Chirp (bandwidth expand) LP AR filter */
133 void silk_bwexpander(
134 opus_int16 *ar, /* I/O AR filter to be expanded (without leading 1) */
135 const opus_int d, /* I Length of ar */
136 opus_int32 chirp_Q16 /* I Chirp factor (typically in the range 0 to 1) */
137 );
138
139 /* Chirp (bandwidth expand) LP AR filter */
140 void silk_bwexpander_32(
141 opus_int32 *ar, /* I/O AR filter to be expanded (without leading 1) */
142 const opus_int d, /* I Length of ar */
143 opus_int32 chirp_Q16 /* I Chirp factor in Q16 */
144 );
145
146 /* Compute inverse of LPC prediction gain, and */
147 /* test if LPC coefficients are stable (all poles within unit circle) */
148 opus_int32 silk_LPC_inverse_pred_gain_c( /* O Returns inverse prediction gain in energy domain, Q30 */
149 const opus_int16 *A_Q12, /* I Prediction coefficients, Q12 [order] */
150 const opus_int order /* I Prediction order */
151 );
152
153 /* Split signal in two decimated bands using first-order allpass filters */
154 void silk_ana_filt_bank_1(
155 const opus_int16 *in, /* I Input signal [N] */
156 opus_int32 *S, /* I/O State vector [2] */
157 opus_int16 *outL, /* O Low band [N/2] */
158 opus_int16 *outH, /* O High band [N/2] */
159 const opus_int32 N /* I Number of input samples */
160 );
161
162 #if !defined(OVERRIDE_silk_biquad_alt_stride2)
163 #define silk_biquad_alt_stride2(in, B_Q28, A_Q28, S, out, len, arch) ((void)(arch), silk_biquad_alt_stride2_c(in, B_Q28, A_Q28, S, out, len))
164 #endif
165
166 #if !defined(OVERRIDE_silk_LPC_inverse_pred_gain)
167 #define silk_LPC_inverse_pred_gain(A_Q12, order, arch) ((void)(arch), silk_LPC_inverse_pred_gain_c(A_Q12, order))
168 #endif
169
170 /********************************************************************/
171 /* SCALAR FUNCTIONS */
172 /********************************************************************/
173
174 /* Approximation of 128 * log2() (exact inverse of approx 2^() below) */
175 /* Convert input to a log scale */
176 opus_int32 silk_lin2log(
177 const opus_int32 inLin /* I input in linear scale */
178 );
179
180 /* Approximation of a sigmoid function */
181 opus_int silk_sigm_Q15(
182 opus_int in_Q5 /* I */
183 );
184
185 /* Approximation of 2^() (exact inverse of approx log2() above) */
186 /* Convert input to a linear scale */
187 opus_int32 silk_log2lin(
188 const opus_int32 inLog_Q7 /* I input on log scale */
189 );
190
191 /* Compute number of bits to right shift the sum of squares of a vector */
192 /* of int16s to make it fit in an int32 */
193 void silk_sum_sqr_shift(
194 opus_int32 *energy, /* O Energy of x, after shifting to the right */
195 opus_int *shift, /* O Number of bits right shift applied to energy */
196 const opus_int16 *x, /* I Input vector */
197 opus_int len /* I Length of input vector */
198 );
199
200 /* Calculates the reflection coefficients from the correlation sequence */
201 /* Faster than schur64(), but much less accurate. */
202 /* uses SMLAWB(), requiring armv5E and higher. */
203 opus_int32 silk_schur( /* O Returns residual energy */
204 opus_int16 *rc_Q15, /* O reflection coefficients [order] Q15 */
205 const opus_int32 *c, /* I correlations [order+1] */
206 const opus_int32 order /* I prediction order */
207 );
208
209 /* Calculates the reflection coefficients from the correlation sequence */
210 /* Slower than schur(), but more accurate. */
211 /* Uses SMULL(), available on armv4 */
212 opus_int32 silk_schur64( /* O returns residual energy */
213 opus_int32 rc_Q16[], /* O Reflection coefficients [order] Q16 */
214 const opus_int32 c[], /* I Correlations [order+1] */
215 opus_int32 order /* I Prediction order */
216 );
217
218 /* Step up function, converts reflection coefficients to prediction coefficients */
219 void silk_k2a(
220 opus_int32 *A_Q24, /* O Prediction coefficients [order] Q24 */
221 const opus_int16 *rc_Q15, /* I Reflection coefficients [order] Q15 */
222 const opus_int32 order /* I Prediction order */
223 );
224
225 /* Step up function, converts reflection coefficients to prediction coefficients */
226 void silk_k2a_Q16(
227 opus_int32 *A_Q24, /* O Prediction coefficients [order] Q24 */
228 const opus_int32 *rc_Q16, /* I Reflection coefficients [order] Q16 */
229 const opus_int32 order /* I Prediction order */
230 );
231
232 /* Apply sine window to signal vector. */
233 /* Window types: */
234 /* 1 -> sine window from 0 to pi/2 */
235 /* 2 -> sine window from pi/2 to pi */
236 /* every other sample of window is linearly interpolated, for speed */
237 void silk_apply_sine_window(
238 opus_int16 px_win[], /* O Pointer to windowed signal */
239 const opus_int16 px[], /* I Pointer to input signal */
240 const opus_int win_type, /* I Selects a window type */
241 const opus_int length /* I Window length, multiple of 4 */
242 );
243
244 /* Compute autocorrelation */
245 void silk_autocorr(
246 opus_int32 *results, /* O Result (length correlationCount) */
247 opus_int *scale, /* O Scaling of the correlation vector */
248 const opus_int16 *inputData, /* I Input data to correlate */
249 const opus_int inputDataSize, /* I Length of input */
250 const opus_int correlationCount, /* I Number of correlation taps to compute */
251 int arch /* I Run-time architecture */
252 );
253
254 void silk_decode_pitch(
255 opus_int16 lagIndex, /* I */
256 opus_int8 contourIndex, /* O */
257 opus_int pitch_lags[], /* O 4 pitch values */
258 const opus_int Fs_kHz, /* I sampling frequency (kHz) */
259 const opus_int nb_subfr /* I number of sub frames */
260 );
261
262 opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 voiced, 1 unvoiced */
263 const opus_int16 *frame, /* I Signal of length PE_FRAME_LENGTH_MS*Fs_kHz */
264 opus_int *pitch_out, /* O 4 pitch lag values */
265 opus_int16 *lagIndex, /* O Lag Index */
266 opus_int8 *contourIndex, /* O Pitch contour Index */
267 opus_int *LTPCorr_Q15, /* I/O Normalized correlation; input: value from previous frame */
268 opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */
269 const opus_int32 search_thres1_Q16, /* I First stage threshold for lag candidates 0 - 1 */
270 const opus_int search_thres2_Q13, /* I Final threshold for lag candidates 0 - 1 */
271 const opus_int Fs_kHz, /* I Sample frequency (kHz) */
272 const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */
273 const opus_int nb_subfr, /* I number of 5 ms subframes */
274 int arch /* I Run-time architecture */
275 );
276
277 /* Compute Normalized Line Spectral Frequencies (NLSFs) from whitening filter coefficients */
278 /* If not all roots are found, the a_Q16 coefficients are bandwidth expanded until convergence. */
279 void silk_A2NLSF(
280 opus_int16 *NLSF, /* O Normalized Line Spectral Frequencies in Q15 (0..2^15-1) [d] */
281 opus_int32 *a_Q16, /* I/O Monic whitening filter coefficients in Q16 [d] */
282 const opus_int d /* I Filter order (must be even) */
283 );
284
285 /* compute whitening filter coefficients from normalized line spectral frequencies */
286 void silk_NLSF2A(
287 opus_int16 *a_Q12, /* O monic whitening filter coefficients in Q12, [ d ] */
288 const opus_int16 *NLSF, /* I normalized line spectral frequencies in Q15, [ d ] */
289 const opus_int d, /* I filter order (should be even) */
290 int arch /* I Run-time architecture */
291 );
292
293 /* Convert int32 coefficients to int16 coefs and make sure there's no wrap-around */
294 void silk_LPC_fit(
295 opus_int16 *a_QOUT, /* O Output signal */
296 opus_int32 *a_QIN, /* I/O Input signal */
297 const opus_int QOUT, /* I Input Q domain */
298 const opus_int QIN, /* I Input Q domain */
299 const opus_int d /* I Filter order */
300 );
301
302 void silk_insertion_sort_increasing(
303 opus_int32 *a, /* I/O Unsorted / Sorted vector */
304 opus_int *idx, /* O Index vector for the sorted elements */
305 const opus_int L, /* I Vector length */
306 const opus_int K /* I Number of correctly sorted positions */
307 );
308
309 void silk_insertion_sort_decreasing_int16(
310 opus_int16 *a, /* I/O Unsorted / Sorted vector */
311 opus_int *idx, /* O Index vector for the sorted elements */
312 const opus_int L, /* I Vector length */
313 const opus_int K /* I Number of correctly sorted positions */
314 );
315
316 void silk_insertion_sort_increasing_all_values_int16(
317 opus_int16 *a, /* I/O Unsorted / Sorted vector */
318 const opus_int L /* I Vector length */
319 );
320
321 /* NLSF stabilizer, for a single input data vector */
322 void silk_NLSF_stabilize(
323 opus_int16 *NLSF_Q15, /* I/O Unstable/stabilized normalized LSF vector in Q15 [L] */
324 const opus_int16 *NDeltaMin_Q15, /* I Min distance vector, NDeltaMin_Q15[L] must be >= 1 [L+1] */
325 const opus_int L /* I Number of NLSF parameters in the input vector */
326 );
327
328 /* Laroia low complexity NLSF weights */
329 void silk_NLSF_VQ_weights_laroia(
330 opus_int16 *pNLSFW_Q_OUT, /* O Pointer to input vector weights [D] */
331 const opus_int16 *pNLSF_Q15, /* I Pointer to input vector [D] */
332 const opus_int D /* I Input vector dimension (even) */
333 );
334
335 /* Compute reflection coefficients from input signal */
336 void silk_burg_modified_c(
337 opus_int32 *res_nrg, /* O Residual energy */
338 opus_int *res_nrg_Q, /* O Residual energy Q value */
339 opus_int32 A_Q16[], /* O Prediction coefficients (length order) */
340 const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */
341 const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */
342 const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */
343 const opus_int nb_subfr, /* I Number of subframes stacked in x */
344 const opus_int D, /* I Order */
345 int arch /* I Run-time architecture */
346 );
347
348 /* Copy and multiply a vector by a constant */
349 void silk_scale_copy_vector16(
350 opus_int16 *data_out,
351 const opus_int16 *data_in,
352 opus_int32 gain_Q16, /* I Gain in Q16 */
353 const opus_int dataSize /* I Length */
354 );
355
356 /* Some for the LTP related function requires Q26 to work.*/
357 void silk_scale_vector32_Q26_lshift_18(
358 opus_int32 *data1, /* I/O Q0/Q18 */
359 opus_int32 gain_Q26, /* I Q26 */
360 opus_int dataSize /* I length */
361 );
362
363 /********************************************************************/
364 /* INLINE ARM MATH */
365 /********************************************************************/
366
367 /* return sum( inVec1[i] * inVec2[i] ) */
368
369 opus_int32 silk_inner_prod_aligned(
370 const opus_int16 *const inVec1, /* I input vector 1 */
371 const opus_int16 *const inVec2, /* I input vector 2 */
372 const opus_int len, /* I vector lengths */
373 int arch /* I Run-time architecture */
374 );
375
376
377 opus_int32 silk_inner_prod_aligned_scale(
378 const opus_int16 *const inVec1, /* I input vector 1 */
379 const opus_int16 *const inVec2, /* I input vector 2 */
380 const opus_int scale, /* I number of bits to shift */
381 const opus_int len /* I vector lengths */
382 );
383
384 opus_int64 silk_inner_prod16_aligned_64_c(
385 const opus_int16 *inVec1, /* I input vector 1 */
386 const opus_int16 *inVec2, /* I input vector 2 */
387 const opus_int len /* I vector lengths */
388 );
389
390 /********************************************************************/
391 /* MACROS */
392 /********************************************************************/
393
394 /* Rotate a32 right by 'rot' bits. Negative rot values result in rotating
395 left. Output is 32bit int.
396 Note: contemporary compilers recognize the C expression below and
397 compile it into a 'ror' instruction if available. No need for OPUS_INLINE ASM! */
silk_ROR32(opus_int32 a32,opus_int rot)398 static OPUS_INLINE opus_int32 silk_ROR32( opus_int32 a32, opus_int rot )
399 {
400 opus_uint32 x = (opus_uint32) a32;
401 opus_uint32 r = (opus_uint32) rot;
402 opus_uint32 m = (opus_uint32) -rot;
403 if( rot == 0 ) {
404 return a32;
405 } else if( rot < 0 ) {
406 return (opus_int32) ((x << m) | (x >> (32 - m)));
407 } else {
408 return (opus_int32) ((x << (32 - r)) | (x >> r));
409 }
410 }
411
412 /* Allocate opus_int16 aligned to 4-byte memory address */
413 #if EMBEDDED_ARM
414 #define silk_DWORD_ALIGN __attribute__((aligned(4)))
415 #else
416 #define silk_DWORD_ALIGN
417 #endif
418
419 /* Useful Macros that can be adjusted to other platforms */
420 #define silk_memcpy(dest, src, size) memcpy((dest), (src), (size))
421 #define silk_memset(dest, src, size) memset((dest), (src), (size))
422 #define silk_memmove(dest, src, size) memmove((dest), (src), (size))
423
424 /* Fixed point macros */
425
426 /* (a32 * b32) output have to be 32bit int */
427 #define silk_MUL(a32, b32) ((a32) * (b32))
428
429 /* (a32 * b32) output have to be 32bit uint */
430 #define silk_MUL_uint(a32, b32) silk_MUL(a32, b32)
431
432 /* a32 + (b32 * c32) output have to be 32bit int */
433 #define silk_MLA(a32, b32, c32) silk_ADD32((a32),((b32) * (c32)))
434
435 /* a32 + (b32 * c32) output have to be 32bit uint */
436 #define silk_MLA_uint(a32, b32, c32) silk_MLA(a32, b32, c32)
437
438 /* ((a32 >> 16) * (b32 >> 16)) output have to be 32bit int */
439 #define silk_SMULTT(a32, b32) (((a32) >> 16) * ((b32) >> 16))
440
441 /* a32 + ((a32 >> 16) * (b32 >> 16)) output have to be 32bit int */
442 #define silk_SMLATT(a32, b32, c32) silk_ADD32((a32),((b32) >> 16) * ((c32) >> 16))
443
444 #define silk_SMLALBB(a64, b16, c16) silk_ADD64((a64),(opus_int64)((opus_int32)(b16) * (opus_int32)(c16)))
445
446 /* (a32 * b32) */
447 #define silk_SMULL(a32, b32) ((opus_int64)(a32) * /*(opus_int64)*/(b32))
448
449 /* Adds two signed 32-bit values in a way that can overflow, while not relying on undefined behaviour
450 (just standard two's complement implementation-specific behaviour) */
silk_ADD32_ovflw(opus_int32 a,opus_int32 b)451 static OPUS_INLINE opus_int32 silk_ADD32_ovflw(opus_int32 a, opus_int32 b) {
452 opus_int32 _c;
453 __builtin_add_overflow(a, b, &_c);
454 return _c;
455 }
456
457 /* Subtractss two signed 32-bit values in a way that can overflow, while not relying on undefined behaviour
458 (just standard two's complement implementation-specific behaviour) */
silk_SUB32_ovflw(opus_int32 a,opus_int32 b)459 static OPUS_INLINE opus_int32 silk_SUB32_ovflw(opus_int32 a, opus_int32 b) {
460 opus_int32 _c;
461 __builtin_sub_overflow(a, b, &_c);
462 return _c;
463 }
464
465 /* Multiply-accumulate macros that allow overflow in the addition (ie, no asserts in debug mode) */
466 /* .. also ignoring multiply overflows; caller has comment about this happening occasionally */
silk_MLA_ovflw(opus_int32 a,opus_int32 b,opus_int32 c)467 static OPUS_INLINE opus_int32 silk_MLA_ovflw(opus_int32 a, opus_int32 b, opus_int32 c) {
468 opus_int32 _d, _e;
469 __builtin_mul_overflow(b, c, &_d);
470 __builtin_add_overflow(a, _d, &_e);
471 return _e;
472 }
473
474 #define silk_SMLABB_ovflw(a32, b32, c32) (silk_ADD32_ovflw((a32) , ((opus_int32)((opus_int16)(b32))) * (opus_int32)((opus_int16)(c32))))
475
476 #define silk_DIV32_16(a32, b16) ((opus_int32)((a32) / (b16)))
477 #define silk_DIV32(a32, b32) ((opus_int32)((a32) / (b32)))
478
479 /* These macros enables checking for overflow in silk_API_Debug.h*/
480 #define silk_ADD16(a, b) ((a) + (b))
481 #define silk_ADD32(a, b) ((a) + (b))
482 #define silk_ADD64(a, b) ((a) + (b))
483
484 #define silk_SUB16(a, b) ((a) - (b))
485 #define silk_SUB32(a, b) ((a) - (b))
486 #define silk_SUB64(a, b) ((a) - (b))
487
488 #define silk_SAT8(a) ((a) > silk_int8_MAX ? silk_int8_MAX : \
489 ((a) < silk_int8_MIN ? silk_int8_MIN : (a)))
490 #define silk_SAT16(a) ((a) > silk_int16_MAX ? silk_int16_MAX : \
491 ((a) < silk_int16_MIN ? silk_int16_MIN : (a)))
492 #define silk_SAT32(a) ((a) > silk_int32_MAX ? silk_int32_MAX : \
493 ((a) < silk_int32_MIN ? silk_int32_MIN : (a)))
494
495 #define silk_CHECK_FIT8(a) (a)
496 #define silk_CHECK_FIT16(a) (a)
497 #define silk_CHECK_FIT32(a) (a)
498
499 #define silk_ADD_SAT16(a, b) (opus_int16)silk_SAT16( silk_ADD32( (opus_int32)(a), (b) ) )
500 #define silk_ADD_SAT64(a, b) ((((a) + (b)) & 0x8000000000000000LL) == 0 ? \
501 ((((a) & (b)) & 0x8000000000000000LL) != 0 ? silk_int64_MIN : (a)+(b)) : \
502 ((((a) | (b)) & 0x8000000000000000LL) == 0 ? silk_int64_MAX : (a)+(b)) )
503
504 #define silk_SUB_SAT16(a, b) (opus_int16)silk_SAT16( silk_SUB32( (opus_int32)(a), (b) ) )
505 #define silk_SUB_SAT64(a, b) ((((a)-(b)) & 0x8000000000000000LL) == 0 ? \
506 (( (a) & ((b)^0x8000000000000000LL) & 0x8000000000000000LL) ? silk_int64_MIN : (a)-(b)) : \
507 ((((a)^0x8000000000000000LL) & (b) & 0x8000000000000000LL) ? silk_int64_MAX : (a)-(b)) )
508
509 /* Saturation for positive input values */
510 #define silk_POS_SAT32(a) ((a) > silk_int32_MAX ? silk_int32_MAX : (a))
511
512 /* Add with saturation for positive input values */
513 #define silk_ADD_POS_SAT8(a, b) ((((a)+(b)) & 0x80) ? silk_int8_MAX : ((a)+(b)))
514 #define silk_ADD_POS_SAT16(a, b) ((((a)+(b)) & 0x8000) ? silk_int16_MAX : ((a)+(b)))
silk_ADD_POS_SAT32(opus_int32 a,opus_int32 b)515 static OPUS_INLINE opus_int32 silk_ADD_POS_SAT32(opus_int32 a, opus_int32 b) {
516 opus_int32 _c;
517 if (__builtin_add_overflow(a, b, &_c))
518 return silk_int32_MAX;
519 return _c;
520 }
521
522 #define silk_LSHIFT8(a, shift) ((opus_int8)((opus_uint8)(a)<<(shift))) /* shift >= 0, shift < 8 */
523 #define silk_LSHIFT16(a, shift) ((opus_int16)((opus_uint16)(a)<<(shift))) /* shift >= 0, shift < 16 */
524 #define silk_LSHIFT32(a, shift) ((opus_int32)((opus_uint32)(a)<<(shift))) /* shift >= 0, shift < 32 */
525 #define silk_LSHIFT64(a, shift) ((opus_int64)((opus_uint64)(a)<<(shift))) /* shift >= 0, shift < 64 */
526 #define silk_LSHIFT(a, shift) silk_LSHIFT32(a, shift) /* shift >= 0, shift < 32 */
527
528 #define silk_RSHIFT8(a, shift) ((a)>>(shift)) /* shift >= 0, shift < 8 */
529 #define silk_RSHIFT16(a, shift) ((a)>>(shift)) /* shift >= 0, shift < 16 */
530 #define silk_RSHIFT32(a, shift) ((a)>>(shift)) /* shift >= 0, shift < 32 */
531 #define silk_RSHIFT64(a, shift) ((a)>>(shift)) /* shift >= 0, shift < 64 */
532 #define silk_RSHIFT(a, shift) silk_RSHIFT32(a, shift) /* shift >= 0, shift < 32 */
533
534 /* saturates before shifting */
535 #define silk_LSHIFT_SAT32(a, shift) (silk_LSHIFT32( silk_LIMIT( (a), silk_RSHIFT32( silk_int32_MIN, (shift) ), \
536 silk_RSHIFT32( silk_int32_MAX, (shift) ) ), (shift) ))
537
538 #define silk_LSHIFT_ovflw(a, shift) ((opus_int32)((opus_uint32)(a) << (shift))) /* shift >= 0, allowed to overflow */
539 #define silk_LSHIFT_uint(a, shift) ((a) << (shift)) /* shift >= 0 */
540 #define silk_RSHIFT_uint(a, shift) ((a) >> (shift)) /* shift >= 0 */
541
542 #define silk_ADD_LSHIFT(a, b, shift) ((a) + silk_LSHIFT((b), (shift))) /* shift >= 0 */
543 #define silk_ADD_LSHIFT32(a, b, shift) silk_ADD32((a), silk_LSHIFT32((b), (shift))) /* shift >= 0 */
544 #define silk_ADD_LSHIFT_uint(a, b, shift) ((a) + silk_LSHIFT_uint((b), (shift))) /* shift >= 0 */
545 #define silk_ADD_RSHIFT(a, b, shift) ((a) + silk_RSHIFT((b), (shift))) /* shift >= 0 */
546 #define silk_ADD_RSHIFT32(a, b, shift) silk_ADD32((a), silk_RSHIFT32((b), (shift))) /* shift >= 0 */
547 #define silk_ADD_RSHIFT_uint(a, b, shift) ((a) + silk_RSHIFT_uint((b), (shift))) /* shift >= 0 */
548 #define silk_SUB_LSHIFT32(a, b, shift) silk_SUB32((a), silk_LSHIFT32((b), (shift))) /* shift >= 0 */
549 #define silk_SUB_RSHIFT32(a, b, shift) silk_SUB32((a), silk_RSHIFT32((b), (shift))) /* shift >= 0 */
550
551 /* Requires that shift > 0 */
552 #define silk_RSHIFT_ROUND(a, shift) ((shift) == 1 ? ((a) >> 1) + ((a) & 1) : (((a) >> ((shift) - 1)) + 1) >> 1)
553 #define silk_RSHIFT_ROUND64(a, shift) ((shift) == 1 ? ((a) >> 1) + ((a) & 1) : (((a) >> ((shift) - 1)) + 1) >> 1)
554
555 /* Number of rightshift required to fit the multiplication */
556 #define silk_NSHIFT_MUL_32_32(a, b) ( -(31- (32-silk_CLZ32(silk_abs(a)) + (32-silk_CLZ32(silk_abs(b))))) )
557 #define silk_NSHIFT_MUL_16_16(a, b) ( -(15- (16-silk_CLZ16(silk_abs(a)) + (16-silk_CLZ16(silk_abs(b))))) )
558
559
560 #define silk_min(a, b) (((a) < (b)) ? (a) : (b))
561 #define silk_max(a, b) (((a) > (b)) ? (a) : (b))
562
563 /* Macro to convert floating-point constants to fixed-point */
564 #define SILK_FIX_CONST( C, Q ) ((opus_int32)((C) * ((opus_int64)1 << (Q)) + 0.5))
565
566 /* silk_min() versions with typecast in the function call */
silk_min_int(opus_int a,opus_int b)567 static OPUS_INLINE opus_int silk_min_int(opus_int a, opus_int b)
568 {
569 return (((a) < (b)) ? (a) : (b));
570 }
silk_min_16(opus_int16 a,opus_int16 b)571 static OPUS_INLINE opus_int16 silk_min_16(opus_int16 a, opus_int16 b)
572 {
573 return (((a) < (b)) ? (a) : (b));
574 }
silk_min_32(opus_int32 a,opus_int32 b)575 static OPUS_INLINE opus_int32 silk_min_32(opus_int32 a, opus_int32 b)
576 {
577 return (((a) < (b)) ? (a) : (b));
578 }
silk_min_64(opus_int64 a,opus_int64 b)579 static OPUS_INLINE opus_int64 silk_min_64(opus_int64 a, opus_int64 b)
580 {
581 return (((a) < (b)) ? (a) : (b));
582 }
583
584 /* silk_min() versions with typecast in the function call */
silk_max_int(opus_int a,opus_int b)585 static OPUS_INLINE opus_int silk_max_int(opus_int a, opus_int b)
586 {
587 return (((a) > (b)) ? (a) : (b));
588 }
silk_max_16(opus_int16 a,opus_int16 b)589 static OPUS_INLINE opus_int16 silk_max_16(opus_int16 a, opus_int16 b)
590 {
591 return (((a) > (b)) ? (a) : (b));
592 }
silk_max_32(opus_int32 a,opus_int32 b)593 static OPUS_INLINE opus_int32 silk_max_32(opus_int32 a, opus_int32 b)
594 {
595 return (((a) > (b)) ? (a) : (b));
596 }
silk_max_64(opus_int64 a,opus_int64 b)597 static OPUS_INLINE opus_int64 silk_max_64(opus_int64 a, opus_int64 b)
598 {
599 return (((a) > (b)) ? (a) : (b));
600 }
601
602 #define silk_LIMIT( a, limit1, limit2) ((limit1) > (limit2) ? ((a) > (limit1) ? (limit1) : ((a) < (limit2) ? (limit2) : (a))) \
603 : ((a) > (limit2) ? (limit2) : ((a) < (limit1) ? (limit1) : (a))))
604
605 #define silk_LIMIT_int silk_LIMIT
606 #define silk_LIMIT_16 silk_LIMIT
607 #define silk_LIMIT_32 silk_LIMIT
608
609 #define silk_abs(a) (((a) > 0) ? (a) : -(a)) /* Be careful, silk_abs returns wrong when input equals to silk_intXX_MIN */
610 #define silk_abs_int(a) (((a) ^ ((a) >> (8 * sizeof(a) - 1))) - ((a) >> (8 * sizeof(a) - 1)))
611 #define silk_abs_int32(a) (((a) ^ ((a) >> 31)) - ((a) >> 31))
612 #define silk_abs_int64(a) (((a) > 0) ? (a) : -(a))
613
614 #define silk_sign(a) ((a) > 0 ? 1 : ( (a) < 0 ? -1 : 0 ))
615
616 /* PSEUDO-RANDOM GENERATOR */
617 /* Make sure to store the result as the seed for the next call (also in between */
618 /* frames), otherwise result won't be random at all. When only using some of the */
619 /* bits, take the most significant bits by right-shifting. */
620 #define RAND_MULTIPLIER 196314165
621 #define RAND_INCREMENT 907633515
622 #define silk_RAND(seed) (silk_MLA_ovflw((RAND_INCREMENT), (seed), (RAND_MULTIPLIER)))
623
624 /* Add some multiplication functions that can be easily mapped to ARM. */
625
626 /* silk_SMMUL: Signed top word multiply.
627 ARMv6 2 instruction cycles.
628 ARMv3M+ 3 instruction cycles. use SMULL and ignore LSB registers.(except xM)*/
629 /*#define silk_SMMUL(a32, b32) (opus_int32)silk_RSHIFT(silk_SMLAL(silk_SMULWB((a32), (b32)), (a32), silk_RSHIFT_ROUND((b32), 16)), 16)*/
630 /* the following seems faster on x86 */
631 #define silk_SMMUL(a32, b32) (opus_int32)silk_RSHIFT64(silk_SMULL((a32), (b32)), 32)
632
633 #if !defined(OPUS_X86_MAY_HAVE_SSE4_1)
634 #define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \
635 ((void)(arch), silk_burg_modified_c(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch))
636
637 #define silk_inner_prod16_aligned_64(inVec1, inVec2, len, arch) \
638 ((void)(arch),silk_inner_prod16_aligned_64_c(inVec1, inVec2, len))
639 #endif
640
641 #include "Inlines.h"
642 #include "MacroCount.h"
643 #include "MacroDebug.h"
644
645 #ifdef OPUS_ARM_INLINE_ASM
646 #include "arm/SigProc_FIX_armv4.h"
647 #endif
648
649 #ifdef OPUS_ARM_INLINE_EDSP
650 #include "arm/SigProc_FIX_armv5e.h"
651 #endif
652
653 #if defined(MIPSr1_ASM)
654 #include "mips/sigproc_fix_mipsr1.h"
655 #endif
656
657
658 #ifdef __cplusplus
659 }
660 #endif
661
662 #endif /* SILK_SIGPROC_FIX_H */
663